DPDK patches and discussions
 help / color / mirror / Atom feed
* [dpdk-dev] [PATCH 0/7] Introduce event vectorization
@ 2021-02-20 22:09 pbhagavatula
  2021-02-20 22:09 ` [dpdk-dev] [PATCH 1/7] eventdev: introduce event vector capability pbhagavatula
                   ` (8 more replies)
  0 siblings, 9 replies; 153+ messages in thread
From: pbhagavatula @ 2021-02-20 22:09 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

In traditional event programming model, events are identified by a
flow-id and a uintptr_t. The flow-id uniquely identifies a given event
and determines the order of scheduling based on schedule type, the
uintptr_t holds a single object.

Event devices also support burst mode with configurable dequeue depth,
i.e. each dequeue call would return multiple events and each event
might be at a different stage of the pipeline.
Having a burst of events belonging to different stages in a dequeue
burst is not only difficult to vectorize but also increases the scheduler
overhead and application overhead of pipelining events further.
Using event vectors we see a performance gain of ~150% as shown in [1].

By introducing event vectorization, each event will be capable of holding
multiple uintptr_t of the same flow thereby allowing applications
to vectorize their pipeline and reduce the complexity of pipelining
events across multiple stages. This also reduces the complexity of handling
enqueue and dequeue on an event device.

Since event devices are transparent to the events they are scheduling
so the event producers such as eth_rx_adapter, crypto_adapter , etc..
are responsible for vectorizing the buffers of the same flow into a single
event.

The series also breaks ABI in [2/7] patch which we fix in [7/7]. The patch
[7/7] can be changed in the next major release i.e. v21.11.

The dpdk-test-eventdev application has been updated with options to test
multiple vector sizes and timeouts.

[1]
As for performance improvement, with a ARM Cortex-A72 equivalent processer,
software event device (--vdev=event_sw0), single worker core, single stage
and using one service core for Rx adapter, Tx adapter, Scheduling.

Without event vectorization:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
         --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
         --stlist=a --wlcores=20
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    4.728 mpps avg 4.728 mpps

With event vectorization:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
        --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
        --stlist=a --wlcores=20 --enable_vector --nb_eth_queues 1
        --vector_size 256
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    34.383 mpps avg 34.383 mpps

Having dedicated service cores for each Rx queues and tweaking the vector,
dequeue burst size would further improve performance.

API usage is shown below:

Configuration:

	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

	vector_pool = rte_event_vector_pool_create("vector_pool",
			nb_elem, 0, vector_size, socket_id);

	rte_event_eth_rx_adapter_create(id, event_id, &adptr_conf);
	rte_event_eth_rx_adapter_queue_add(id, eth_id, -1, &queue_conf);
	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
		vec_conf.vector_sz = vector_size;
		vec_conf.vector_timeout_ns = vector_tmo_nsec;
		vec_conf.vector_mp = vector_pool;
		rte_event_eth_rx_adapter_queue_event_vector_config(id,
				eth_id, -1, &vec_conf);
	}

Fastpath:

	num = rte_event_dequeue_burst(event_id, port_id, &ev, 1, 0);
	if (!num)
		continue;

	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
		switch (ev.event_type) {
		case RTE_EVENT_TYPE_ETHDEV_VECTOR:
		case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
			struct rte_mbuf **mbufs;

			mbufs = ev.vector_ev->mbufs;
			for (i = 0; i < ev.vector_ev->nb_elem; i++)
				//Process mbufs.
			break;
		case ...
		}
	}
	...

Pavan Nikhilesh (7):
  eventdev: introduce event vector capability
  eventdev: introduce event vector Rx capability
  eventdev: introduce event vector Tx capability
  eventdev: add Rx adapter event vector support
  eventdev: add Tx adapter event vector support
  app/eventdev: add event vector mode in pipeline test
  eventdev: fix ABI breakage due to event vector

 app/test-eventdev/evt_common.h                |   4 +
 app/test-eventdev/evt_options.c               |  52 +++
 app/test-eventdev/evt_options.h               |   4 +
 app/test-eventdev/test_pipeline_atq.c         | 310 +++++++++++++--
 app/test-eventdev/test_pipeline_common.c      |  77 +++-
 app/test-eventdev/test_pipeline_common.h      |  18 +
 app/test-eventdev/test_pipeline_queue.c       | 320 +++++++++++++--
 .../prog_guide/event_ethernet_rx_adapter.rst  |  38 ++
 .../prog_guide/event_ethernet_tx_adapter.rst  |  12 +
 doc/guides/prog_guide/eventdev.rst            |  36 +-
 doc/guides/tools/testeventdev.rst             |  28 ++
 lib/librte_eventdev/eventdev_pmd.h            |  60 ++-
 .../rte_event_eth_rx_adapter.c                | 367 +++++++++++++++++-
 .../rte_event_eth_rx_adapter.h                |  93 +++++
 .../rte_event_eth_tx_adapter.c                |  66 +++-
 lib/librte_eventdev/rte_eventdev.c            |  11 +-
 lib/librte_eventdev/rte_eventdev.h            | 145 ++++++-
 lib/librte_eventdev/version.map               |   5 +
 18 files changed, 1560 insertions(+), 86 deletions(-)

--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH 1/7] eventdev: introduce event vector capability
  2021-02-20 22:09 [dpdk-dev] [PATCH 0/7] Introduce event vectorization pbhagavatula
@ 2021-02-20 22:09 ` pbhagavatula
  2021-03-08 16:49   ` Jerin Jacob
  2021-02-20 22:09 ` [dpdk-dev] [PATCH 2/7] eventdev: introduce event vector Rx capability pbhagavatula
                   ` (7 subsequent siblings)
  8 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-02-20 22:09 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce rte_event_vector datastructure which is capable of holding
multiple uintptr_t of the same flow thereby allowing applications
to vectorize their pipeline and reducing the complexity of pipelining
the events across multiple stages.
This approach also reduces the scheduling overhead on a event device.

Add a event vector mempool create handler to create mempools based on
the best mempool ops available on a given platform.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 doc/guides/prog_guide/eventdev.rst |  36 ++++++++-
 lib/librte_eventdev/rte_eventdev.h | 113 ++++++++++++++++++++++++++++-
 lib/librte_eventdev/version.map    |   3 +
 3 files changed, 149 insertions(+), 3 deletions(-)

diff --git a/doc/guides/prog_guide/eventdev.rst b/doc/guides/prog_guide/eventdev.rst
index ccde086f6..d19c91ab0 100644
--- a/doc/guides/prog_guide/eventdev.rst
+++ b/doc/guides/prog_guide/eventdev.rst
@@ -63,13 +63,45 @@ the actual event being scheduled is. The payload is a union of the following:
 * ``uint64_t u64``
 * ``void *event_ptr``
 * ``struct rte_mbuf *mbuf``
+* ``struct rte_event_vector *vec``
 
-These three items in a union occupy the same 64 bits at the end of the rte_event
+These four items in a union occupy the same 64 bits at the end of the rte_event
 structure. The application can utilize the 64 bits directly by accessing the
-u64 variable, while the event_ptr and mbuf are provided as convenience
+u64 variable, while the event_ptr, mbuf, vec are provided as convenience
 variables.  For example the mbuf pointer in the union can used to schedule a
 DPDK packet.
 
+Event Vector
+~~~~~~~~~~~~
+
+The rte_event_vector struct contains a vector of elements defined by the event
+type specified in the ``rte_event``. The event_vector structure contains the
+following data:
+
+* ``nb_elem`` - The number of elements held within the vector.
+
+Similar to ``rte_event`` the payload of event vector is also a union, allowing
+flexibility in what the actual vector is.
+
+* ``struct rte_mbuf *mbufs[0]`` - An array of mbufs.
+* ``void *ptrs[0]`` - An array of pointers.
+* ``uint64_t *u64s[0]`` - An array of uint64_t elements.
+
+The size of the event vector is related to the total number of elements it is
+configured to hold, this is achieved by making `rte_event_vector` a variable
+length structure.
+A helper function is provided to create a mempool that holds event vector, which
+takes name of the pool, total number of required ``rte_event_vector``,
+cache size, number of elements in each ``rte_event_vector`` and socket id.
+
+.. code-block:: c
+
+        rte_event_vector_pool_create("vector_pool", nb_event_vectors, cache_sz,
+                                     nb_elements_per_vector, socket_id);
+
+The function ``rte_event_vector_pool_create`` creates mempool with the best
+platform mempool ops.
+
 Queues
 ~~~~~~
 
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index ce1fc2ce0..ff6cb3e6a 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -212,8 +212,10 @@ extern "C" {
 
 #include <rte_common.h>
 #include <rte_config.h>
-#include <rte_memory.h>
 #include <rte_errno.h>
+#include <rte_mbuf_pool_ops.h>
+#include <rte_memory.h>
+#include <rte_mempool.h>
 
 #include "rte_eventdev_trace_fp.h"
 
@@ -913,6 +915,25 @@ rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
 int
 rte_event_dev_close(uint8_t dev_id);
 
+/**
+ * Event vector structure.
+ */
+struct rte_event_vector {
+	uint64_t nb_elem : 16;
+	/**< Number of elements in this event vector. */
+	uint64_t rsvd : 48;
+	uint64_t impl_opaque;
+	union {
+		struct rte_mbuf *mbufs[0];
+		void *ptrs[0];
+		uint64_t *u64s[0];
+	} __rte_aligned(16);
+	/**< Start of the vector array union. Depending upon the event type the
+	 * vector array can be an array of mbufs or pointers or opaque u64
+	 * values.
+	 */
+};
+
 /* Scheduler type definitions */
 #define RTE_SCHED_TYPE_ORDERED          0
 /**< Ordered scheduling
@@ -986,6 +1007,21 @@ rte_event_dev_close(uint8_t dev_id);
  */
 #define RTE_EVENT_TYPE_ETH_RX_ADAPTER   0x4
 /**< The event generated from event eth Rx adapter */
+#define RTE_EVENT_TYPE_VECTOR           0x8
+/**< Indicates that event is a vector.
+ * All vector event types should be an logical OR of EVENT_TYPE_VECTOR.
+ * This simplifies the pipeline design as we can split processing the events
+ * between vector events and normal event across event types.
+ * Example:
+ *	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
+ *		// Classify and handle vector event.
+ *	} else {
+ *		// Classify and handle event.
+ *	}
+ */
+#define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
+/**< The event vector generated from cpu for pipelining. */
+
 #define RTE_EVENT_TYPE_MAX              0x10
 /**< Maximum number of event types */
 
@@ -1108,6 +1144,8 @@ struct rte_event {
 		/**< Opaque event pointer */
 		struct rte_mbuf *mbuf;
 		/**< mbuf pointer if dequeued event is associated with mbuf */
+		struct rte_event_vector *vec;
+		/**< Event vector pointer. */
 	};
 };
 
@@ -2023,6 +2061,79 @@ rte_event_dev_xstats_reset(uint8_t dev_id,
  */
 int rte_event_dev_selftest(uint8_t dev_id);
 
+/**
+ * Get the memory required per event vector based on number of elements per
+ * vector.
+ * This should be used to create the mempool that holds the event vectors.
+ *
+ * @param name
+ *   The name of the vector pool.
+ * @param n
+ *   The number of elements in the mbuf pool.
+ * @param cache_size
+ *   Size of the per-core object cache. See rte_mempool_create() for
+ *   details.
+ * @param nb_elem
+ *   The number of elements then a single event vector should be able to hold.
+ * @param socket_id
+ *   The socket identifier where the memory should be allocated. The
+ *   value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
+ *   reserved zone
+ *
+ * @return
+ *   The pointer to the new allocated mempool, on success. NULL on error
+ *   with rte_errno set appropriately. Possible rte_errno values include:
+ *    - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
+ *    - E_RTE_SECONDARY - function was called from a secondary process instance
+ *    - EINVAL - cache size provided is too large, or priv_size is not aligned.
+ *    - ENOSPC - the maximum number of memzones has already been allocated
+ *    - EEXIST - a memzone with the same name already exists
+ *    - ENOMEM - no appropriate memory area found in which to create memzone
+ */
+__rte_experimental
+static inline struct rte_mempool *
+rte_event_vector_pool_create(const char *name, unsigned int n,
+			     unsigned int cache_size, uint16_t nb_elem,
+			     int socket_id)
+{
+	const char *mp_ops_name;
+	struct rte_mempool *mp;
+	unsigned int elt_sz;
+	int ret;
+
+	if (!nb_elem) {
+		RTE_LOG(ERR, EVENTDEV,
+			"Invalid number of elements=%d requested\n", nb_elem);
+		rte_errno = -EINVAL;
+		return NULL;
+	}
+
+	elt_sz =
+		sizeof(struct rte_event_vector) + (nb_elem * sizeof(uintptr_t));
+	mp = rte_mempool_create_empty(name, n, elt_sz, cache_size, 0, socket_id,
+				      0);
+	if (mp == NULL)
+		return NULL;
+
+	mp_ops_name = rte_mbuf_best_mempool_ops();
+	ret = rte_mempool_set_ops_byname(mp, mp_ops_name, NULL);
+	if (ret != 0) {
+		RTE_LOG(ERR, EVENTDEV, "error setting mempool handler\n");
+		rte_mempool_free(mp);
+		rte_errno = -ret;
+		return NULL;
+	}
+
+	ret = rte_mempool_populate_default(mp);
+	if (ret < 0) {
+		rte_mempool_free(mp);
+		rte_errno = -ret;
+		return NULL;
+	}
+
+	return mp;
+}
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index 3e5c09cfd..a070ef56e 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -138,6 +138,9 @@ EXPERIMENTAL {
 	__rte_eventdev_trace_port_setup;
 	# added in 20.11
 	rte_event_pmd_pci_probe_named;
+
+	#added in 21.05
+	rte_event_vector_pool_create;
 };
 
 INTERNAL {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH 2/7] eventdev: introduce event vector Rx capability
  2021-02-20 22:09 [dpdk-dev] [PATCH 0/7] Introduce event vectorization pbhagavatula
  2021-02-20 22:09 ` [dpdk-dev] [PATCH 1/7] eventdev: introduce event vector capability pbhagavatula
@ 2021-02-20 22:09 ` pbhagavatula
  2021-03-08 17:07   ` Jerin Jacob
  2021-02-20 22:09 ` [dpdk-dev] [PATCH 3/7] eventdev: introduce event vector Tx capability pbhagavatula
                   ` (6 subsequent siblings)
  8 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-02-20 22:09 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce event ethernet Rx adapter event vector capability.

If an event eth Rx adapter has the capability of
RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR then a given Rx queue
can be configured to enable event vectorization by passing the
flag RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR to
rte_event_eth_rx_adapter_queue_conf::rx_queue_flags while configuring
Rx adapter through rte_event_eth_rx_adapter_queue_add.

The max vector size, vector timeout define the vector size and
mempool used for allocating vector event are configured through
rte_event_eth_rx_adapter_queue_add. The element size of the element
in the vector pool should be equal to
    sizeof(struct rte_event_vector) + (vector_sz * sizeof(uintptr_t))

Application can use `rte_event_get_event_vector_memory_footprint`
to get the element size before creating the vector mempool
rte_event_eth_rx_adapter_queue_conf::vector_mp.

The Rx adapter would be responsible for vectorizing the mbufs
based on the flow, the vector limits configured by the application
and add the vector event of mbufs to the event queue set via
rte_event_eth_rx_adapter_queue_conf::ev::queue_id.
It should also mark rte_event_vector::union_valid and fill
rte_event_vector::port, rte_event_vector::queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 .../prog_guide/event_ethernet_rx_adapter.rst  | 38 +++++++++++
 .../rte_event_eth_rx_adapter.h                | 66 +++++++++++++++++++
 lib/librte_eventdev/rte_eventdev.h            | 30 ++++++++-
 lib/librte_eventdev/version.map               |  1 +
 4 files changed, 133 insertions(+), 2 deletions(-)

diff --git a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
index cb44ce0e4..735ea2439 100644
--- a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
@@ -186,3 +186,41 @@ the event buffer fill level is low. The
 ``rte_event_eth_rx_adapter_cb_register()`` function allow the application
 to register a callback that selects which packets to enqueue to the event
 device.
+
+Rx event vectorization
+~~~~~~~~~~~~~~~~~~~~~~
+
+The event devices, ethernet device pairs which support the capability
+``RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR`` can aggregate packets based on
+flow characteristics and generate a ``rte_event`` containing ``rte_event_vector``
+whose event type is either ``RTE_EVENT_TYPE_ETHDEV_VECTOR`` or
+``RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR``.
+The aggregation size and timeout are configurable at a queue level and the
+maximum, minimum vector sizes and timeouts vary based on the device capability
+and can be queried using ``rte_event_eth_rx_adapter_vector_limits_get``.
+The Rx adapter additionally might include useful data such as ethernet device
+port and queue identifier in the ``rte_event_vector::port`` and
+``rte_event_vector::queue`` and mark ``rte_event_vector::attr_valid`` as true.
+
+A loop processing ``rte_event_vector`` containing mbufs is shown below.
+
+.. code-block:: c
+
+        event = rte_event_dequeue_burst(event_dev, event_port, &event,
+                                        1, 0);
+        if (!event)
+                continue;
+
+        switch (ev.event_type) {
+        case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
+        case RTE_EVENT_TYPE_ETHDEV_VECTOR:
+                struct rte_mbufs **mbufs;
+
+                mbufs = (struct rte_mbufs **)ev[i].vec->mbufs;
+                for (i = 0; i < ev.vec->nb_elem; i++) {
+                        // Process each mbuf.
+                }
+        break;
+        case ...
+        ...
+        }
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
index 21bb1e54c..4bdb38f08 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
@@ -92,6 +92,10 @@ extern "C" {
 /**< This flag indicates the flow identifier is valid
  * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
  */
+#define RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR	0x2
+/**< This flag indicates that mbufs arriving on the queue need to be vectorized
+ * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+ */
 
 /**
  * Adapter configuration structure that the adapter configuration callback
@@ -167,6 +171,33 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
 	 * enqueued event.
 	 */
+	uint16_t vector_sz;
+	/**<
+	 * Indicates the maximum number for mbufs to combine and form a vector.
+	 * Should be within
+	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_sz
+	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_sz
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+	 */
+	uint64_t vector_timeout_ns;
+	/**<
+	 * Indicates the maximum number of nanoseconds to wait for receiving
+	 * mbufs. Should be within vectorization limits of the
+	 * adapter
+	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_ns
+	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_ns
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+	 */
+	struct rte_mempool *vector_mp;
+	/**<
+	 * Indicates the mempool that should be used for allocating
+	 * rte_event_vector container.
+	 * Should be created by using `rte_event_vector_pool_create`.
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags.
+	 */
 };
 
 /**
@@ -199,6 +230,20 @@ struct rte_event_eth_rx_adapter_stats {
 	/**< Received packet count for interrupt mode Rx queues */
 };
 
+/**
+ * A structure used to retrieve eth rx adapter vector limits.
+ */
+struct rte_event_eth_rx_adapter_vector_limits {
+	uint16_t min_sz;
+	/**< Minimum vector limit configurable. */
+	uint16_t max_sz;
+	/**< Maximum vector limit configurable. */
+	uint64_t min_timeout_ns;
+	/**< Minimum vector timeout configurable. */
+	uint64_t max_timeout_ns;
+	/**< Maximum vector timeout configurable. */
+};
+
 /**
  *
  * Callback function invoked by the SW adapter before it continues
@@ -467,6 +512,27 @@ int rte_event_eth_rx_adapter_cb_register(uint8_t id, uint16_t eth_dev_id,
 					 rte_event_eth_rx_adapter_cb_fn cb_fn,
 					 void *cb_arg);
 
+/**
+ * Retrieve vector limits for a given event dev and eth dev pair.
+ * @see rte_event_eth_rx_adapter_vector_limits
+ *
+ * @param dev_id
+ *  Event device identifier.
+ * @param eth_port_id
+ *  Port identifier of the ethernet device.
+ * @param [out] limits
+ *  A pointer to rte_event_eth_rx_adapter_vector_limits structure that has to
+ * be filled.
+ *
+ * @return
+ *  - 0: Success.
+ *  - <0: Error code on failure.
+ */
+__rte_experimental
+int rte_event_eth_rx_adapter_vector_limits_get(
+	uint8_t dev_id, uint16_t eth_port_id,
+	struct rte_event_eth_rx_adapter_vector_limits *limits);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index ff6cb3e6a..1cf3efa2d 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -919,9 +919,27 @@ rte_event_dev_close(uint8_t dev_id);
  * Event vector structure.
  */
 struct rte_event_vector {
-	uint64_t nb_elem : 16;
+	uint16_t nb_elem;
 	/**< Number of elements in this event vector. */
-	uint64_t rsvd : 48;
+	uint16_t rsvd : 15;
+	uint16_t attr_valid : 1;
+	/**< Indicates that the below union attributes have valid information.
+	 */
+	union {
+		/* Used by Rx adapter.
+		 * Indicats that all the elements in this vector belong to same
+		 * port and queue pair when originating from Rx adapter, valid
+		 * only when event type is ETHDEV_VECTOR or
+		 * ETH_RX_ADAPTER_VECTOR.
+		 */
+		struct {
+			uint16_t port;
+			/* Ethernet device port id. */
+			uint16_t queue;
+			/* Ethernet device queue id. */
+		};
+	};
+	/**< Union to hold common attributes of the vector array. */
 	uint64_t impl_opaque;
 	union {
 		struct rte_mbuf *mbufs[0];
@@ -1019,8 +1037,14 @@ struct rte_event_vector {
  *		// Classify and handle event.
  *	}
  */
+#define RTE_EVENT_TYPE_ETHDEV_VECTOR                                           \
+	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETHDEV)
+/**< The event vector generated from ethdev subsystem */
 #define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
 /**< The event vector generated from cpu for pipelining. */
+#define RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR                                   \
+	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETH_RX_ADAPTER)
+/**< The event vector generated from eth Rx adapter. */
 
 #define RTE_EVENT_TYPE_MAX              0x10
 /**< Maximum number of event types */
@@ -1165,6 +1189,8 @@ struct rte_event {
  * @see struct rte_event_eth_rx_adapter_queue_conf::ev
  * @see struct rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
  */
+#define RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR	0x8
+/**< Adapter supports event vectorization per ethdev. */
 
 /**
  * Retrieve the event device's ethdev Rx adapter capabilities for the
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index a070ef56e..34c1c830e 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -141,6 +141,7 @@ EXPERIMENTAL {
 
 	#added in 21.05
 	rte_event_vector_pool_create;
+	rte_event_eth_rx_adapter_vector_limits_get;
 };
 
 INTERNAL {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH 3/7] eventdev: introduce event vector Tx capability
  2021-02-20 22:09 [dpdk-dev] [PATCH 0/7] Introduce event vectorization pbhagavatula
  2021-02-20 22:09 ` [dpdk-dev] [PATCH 1/7] eventdev: introduce event vector capability pbhagavatula
  2021-02-20 22:09 ` [dpdk-dev] [PATCH 2/7] eventdev: introduce event vector Rx capability pbhagavatula
@ 2021-02-20 22:09 ` pbhagavatula
  2021-03-08 17:09   ` Jerin Jacob
  2021-02-20 22:09 ` [dpdk-dev] [PATCH 4/7] eventdev: add Rx adapter event vector support pbhagavatula
                   ` (5 subsequent siblings)
  8 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-02-20 22:09 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce event vector transmit capability for event eth
tx adapter.

The capability indicates that the Tx adapter is capable of
transmitting event vectors.
When rte_event_vector::union_valid is set, the Tx adapter should
transmit all the packets to the rte_event_vector::port using the
rte_event_vector::queue.
If rte_event_vector::union_valid is not set then the Tx adapter
should peek into each mbuf to get the destination port and queue
pair.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 doc/guides/prog_guide/event_ethernet_tx_adapter.rst | 12 ++++++++++++
 lib/librte_eventdev/rte_eventdev.h                  |  8 +++++++-
 2 files changed, 19 insertions(+), 1 deletion(-)

diff --git a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
index a8c13e136..87277dcaf 100644
--- a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
@@ -164,3 +164,15 @@ The  ``rte_event_eth_tx_adapter_stats_get()`` function reports counters defined
 in struct ``rte_event_eth_tx_adapter_stats``. The counter values are the sum of
 the counts from the eventdev PMD callback if the callback is supported, and
 the counts maintained by the service function, if one exists.
+
+Tx event vectorization
+~~~~~~~~~~~~~~~~~~~~~~
+
+The event device, ethernet device paris which support the capability
+``RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR`` can process event vector of mbufs.
+Additionally, application can provide a hint to that Tx adapter that all the
+mbufs are destined to the same ethernet port and queue by setting the bit
+``rte_event_vector::attr_valid`` and filling `rte_event_vector::port`` and
+``rte_event_vector::queue``.
+If ``rte_event_vector::attr_valid`` is not set then the Tx adapter should peek
+into each mbuf and transmit them to the requested ethernet port and queue pair.
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index 1cf3efa2d..c817c29e6 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -926,11 +926,13 @@ struct rte_event_vector {
 	/**< Indicates that the below union attributes have valid information.
 	 */
 	union {
-		/* Used by Rx adapter.
+		/* Used by Rx/Tx adapter.
 		 * Indicats that all the elements in this vector belong to same
 		 * port and queue pair when originating from Rx adapter, valid
 		 * only when event type is ETHDEV_VECTOR or
 		 * ETH_RX_ADAPTER_VECTOR.
+		 * Can also be used to indicate the Tx adapter the destination
+		 * port and queue of the mbufs in the vector
 		 */
 		struct {
 			uint16_t port;
@@ -1287,6 +1289,10 @@ rte_event_crypto_adapter_caps_get(uint8_t dev_id, uint8_t cdev_id,
 #define RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT	0x1
 /**< This flag is sent when the PMD supports a packet transmit callback
  */
+#define RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR	0x2
+/**< Indicates that the Tx adapter is capable of handling event vector of
+ * mbufs.
+ */
 
 /**
  * Retrieve the event device's eth Tx adapter capabilities
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH 4/7] eventdev: add Rx adapter event vector support
  2021-02-20 22:09 [dpdk-dev] [PATCH 0/7] Introduce event vectorization pbhagavatula
                   ` (2 preceding siblings ...)
  2021-02-20 22:09 ` [dpdk-dev] [PATCH 3/7] eventdev: introduce event vector Tx capability pbhagavatula
@ 2021-02-20 22:09 ` pbhagavatula
  2021-03-08 17:27   ` Jerin Jacob
  2021-02-20 22:09 ` [dpdk-dev] [PATCH 5/7] eventdev: add Tx " pbhagavatula
                   ` (4 subsequent siblings)
  8 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-02-20 22:09 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support for event eth Rx adapter, the implementation
creates vector flows based on port and queue identifier of the received
mbufs.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 lib/librte_eventdev/eventdev_pmd.h            |  31 +-
 .../rte_event_eth_rx_adapter.c                | 305 +++++++++++++++++-
 lib/librte_eventdev/rte_eventdev.c            |   6 +-
 3 files changed, 324 insertions(+), 18 deletions(-)

diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 7eb9a7739..60bfaebc0 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -69,9 +69,10 @@ extern "C" {
 	} \
 } while (0)
 
-#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP \
-		((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) | \
-			(RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ))
+#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP                                        \
+	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |                     \
+	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) |                         \
+	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR))
 
 #define RTE_EVENT_CRYPTO_ADAPTER_SW_CAP \
 		RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA
@@ -645,6 +646,27 @@ typedef int (*eventdev_eth_rx_adapter_stats_reset)
  */
 typedef int (*eventdev_selftest)(void);
 
+struct rte_event_eth_rx_adapter_vector_limits;
+/**
+ * Get event vector limits for a given event, etherner device pair.
+ *
+ * @param dev
+ *   Event device pointer
+ *
+ * @param eth_dev
+ *   Ethernet device pointer
+ *
+ * @param[out] limits
+ *   Pointer to the limits structure to be filled.
+ *
+ * @return
+ *   - 0: Success.
+ *   - <0: Error code returned by the driver function.
+ */
+typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
+	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
+	struct rte_event_eth_rx_adapter_vector_limits *limits);
+
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;
 
@@ -1067,6 +1089,9 @@ struct rte_eventdev_ops {
 	/**< Get ethernet Rx stats */
 	eventdev_eth_rx_adapter_stats_reset eth_rx_adapter_stats_reset;
 	/**< Reset ethernet Rx stats */
+	eventdev_eth_rx_adapter_vector_limits_get_t
+		eth_rx_adapter_vector_limits_get;
+	/**< Get event vector limits for the Rx adapter */
 
 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index d8c635e99..a1990637f 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -26,6 +26,10 @@
 #define BATCH_SIZE		32
 #define BLOCK_CNT_THRESHOLD	10
 #define ETH_EVENT_BUFFER_SIZE	(4*BATCH_SIZE)
+#define MAX_VECTOR_SIZE		1024
+#define MIN_VECTOR_SIZE		4
+#define MAX_VECTOR_NS		1E9
+#define MIN_VECTOR_NS		1E5
 
 #define ETH_RX_ADAPTER_SERVICE_NAME_LEN	32
 #define ETH_RX_ADAPTER_MEM_NAME_LEN	32
@@ -59,6 +63,20 @@ struct eth_rx_poll_entry {
 	uint16_t eth_rx_qid;
 };
 
+struct eth_rx_vector_data {
+	TAILQ_ENTRY(eth_rx_vector_data) next;
+	uint16_t port;
+	uint16_t queue;
+	uint16_t max_vector_count;
+	uint64_t event;
+	uint64_t ts;
+	uint64_t vector_timeout_ticks;
+	struct rte_mempool *vector_pool;
+	struct rte_event_vector *vector_ev;
+} __rte_cache_aligned;
+
+TAILQ_HEAD(eth_rx_vector_data_list, eth_rx_vector_data);
+
 /* Instance per adapter */
 struct rte_eth_event_enqueue_buffer {
 	/* Count of events in this buffer */
@@ -92,6 +110,14 @@ struct rte_event_eth_rx_adapter {
 	uint32_t wrr_pos;
 	/* Event burst buffer */
 	struct rte_eth_event_enqueue_buffer event_enqueue_buffer;
+	/* Vector enable flag */
+	uint8_t ena_vector;
+	/* Timestamp of previous vector expiry list traversal */
+	uint64_t prev_expiry_ts;
+	/* Minimum ticks to wait before traversing expiry list */
+	uint64_t vector_tmo_ticks;
+	/* vector list */
+	struct eth_rx_vector_data_list vector_list;
 	/* Per adapter stats */
 	struct rte_event_eth_rx_adapter_stats stats;
 	/* Block count, counts up to BLOCK_CNT_THRESHOLD */
@@ -198,9 +224,11 @@ struct eth_device_info {
 struct eth_rx_queue_info {
 	int queue_enabled;	/* True if added */
 	int intr_enabled;
+	uint8_t ena_vector;
 	uint16_t wt;		/* Polling weight */
 	uint32_t flow_id_mask;	/* Set to ~0 if app provides flow id else 0 */
 	uint64_t event;
+	struct eth_rx_vector_data vector_data;
 };
 
 static struct rte_event_eth_rx_adapter **event_eth_rx_adapter;
@@ -722,6 +750,9 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
 	    &rx_adapter->event_enqueue_buffer;
 	struct rte_event_eth_rx_adapter_stats *stats = &rx_adapter->stats;
 
+	if (!buf->count)
+		return 0;
+
 	uint16_t n = rte_event_enqueue_new_burst(rx_adapter->eventdev_id,
 					rx_adapter->event_port_id,
 					buf->events,
@@ -742,6 +773,72 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
 	return n;
 }
 
+static inline uint16_t
+rxa_create_event_vector(struct rte_event_eth_rx_adapter *rx_adapter,
+			struct eth_rx_queue_info *queue_info,
+			struct rte_eth_event_enqueue_buffer *buf,
+			struct rte_mbuf **mbufs, uint16_t num)
+{
+	struct rte_event *ev = &buf->events[buf->count];
+	struct eth_rx_vector_data *vec;
+	uint16_t filled, space, sz;
+
+	filled = 0;
+	vec = &queue_info->vector_data;
+	while (num) {
+		if (vec->vector_ev == NULL) {
+			if (rte_mempool_get(vec->vector_pool,
+					    (void **)&vec->vector_ev) < 0) {
+				rte_pktmbuf_free_bulk(mbufs, num);
+				return 0;
+			}
+			vec->vector_ev->nb_elem = 0;
+			vec->vector_ev->port = vec->port;
+			vec->vector_ev->queue = vec->queue;
+			vec->vector_ev->attr_valid = true;
+			TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
+		} else if (vec->vector_ev->nb_elem == vec->max_vector_count) {
+			/* Event ready. */
+			ev->event = vec->event;
+			ev->vec = vec->vector_ev;
+			ev++;
+			filled++;
+			vec->vector_ev = NULL;
+			TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+			if (rte_mempool_get(vec->vector_pool,
+					    (void **)&vec->vector_ev) < 0) {
+				rte_pktmbuf_free_bulk(mbufs, num);
+				return 0;
+			}
+			vec->vector_ev->nb_elem = 0;
+			vec->vector_ev->port = vec->port;
+			vec->vector_ev->queue = vec->queue;
+			vec->vector_ev->attr_valid = true;
+			TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
+		}
+
+		space = vec->max_vector_count - vec->vector_ev->nb_elem;
+		sz = num > space ? space : num;
+		memcpy(vec->vector_ev->mbufs + vec->vector_ev->nb_elem, mbufs,
+		       sizeof(void *) * sz);
+		vec->vector_ev->nb_elem += sz;
+		num -= sz;
+		mbufs += sz;
+		vec->ts = rte_rdtsc();
+	}
+
+	if (vec->vector_ev->nb_elem == vec->max_vector_count) {
+		ev->event = vec->event;
+		ev->vec = vec->vector_ev;
+		ev++;
+		filled++;
+		vec->vector_ev = NULL;
+		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+	}
+
+	return filled;
+}
+
 static inline void
 rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
@@ -770,25 +867,30 @@ rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
 	rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
 	do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
 
-	for (i = 0; i < num; i++) {
-		m = mbufs[i];
-
-		rss = do_rss ?
-			rxa_do_softrss(m, rx_adapter->rss_key_be) :
-			m->hash.rss;
-		ev->event = event;
-		ev->flow_id = (rss & ~flow_id_mask) |
-				(ev->flow_id & flow_id_mask);
-		ev->mbuf = m;
-		ev++;
+	if (!eth_rx_queue_info->ena_vector) {
+		for (i = 0; i < num; i++) {
+			m = mbufs[i];
+
+			rss = do_rss ? rxa_do_softrss(m, rx_adapter->rss_key_be)
+				     : m->hash.rss;
+			ev->event = event;
+			ev->flow_id = (rss & ~flow_id_mask) |
+				      (ev->flow_id & flow_id_mask);
+			ev->mbuf = m;
+			ev++;
+		}
+	} else {
+		num = rxa_create_event_vector(rx_adapter, eth_rx_queue_info,
+					      buf, mbufs, num);
 	}
 
-	if (dev_info->cb_fn) {
+	if (num && dev_info->cb_fn) {
 
 		dropped = 0;
 		nb_cb = dev_info->cb_fn(eth_dev_id, rx_queue_id,
-					ETH_EVENT_BUFFER_SIZE, buf->count, ev,
-					num, dev_info->cb_arg, &dropped);
+					ETH_EVENT_BUFFER_SIZE, buf->count,
+					&buf->events[buf->count], num,
+					dev_info->cb_arg, &dropped);
 		if (unlikely(nb_cb > num))
 			RTE_EDEV_LOG_ERR("Rx CB returned %d (> %d) events",
 				nb_cb, num);
@@ -1124,6 +1226,30 @@ rxa_poll(struct rte_event_eth_rx_adapter *rx_adapter)
 	return nb_rx;
 }
 
+static void
+rxa_vector_expire(struct eth_rx_vector_data *vec, void *arg)
+{
+	struct rte_event_eth_rx_adapter *rx_adapter = arg;
+	struct rte_eth_event_enqueue_buffer *buf =
+		&rx_adapter->event_enqueue_buffer;
+	struct rte_event *ev;
+
+	if (buf->count)
+		rxa_flush_event_buffer(rx_adapter);
+
+	if (vec->vector_ev->nb_elem == 0)
+		return;
+	ev = &buf->events[buf->count];
+
+	/* Event ready. */
+	ev->event = vec->event;
+	ev->vec = vec->vector_ev;
+	buf->count++;
+
+	vec->vector_ev = NULL;
+	vec->ts = 0;
+}
+
 static int
 rxa_service_func(void *args)
 {
@@ -1137,6 +1263,24 @@ rxa_service_func(void *args)
 		return 0;
 	}
 
+	if (rx_adapter->ena_vector) {
+		if ((rte_rdtsc() - rx_adapter->prev_expiry_ts) >=
+		    rx_adapter->vector_tmo_ticks) {
+			struct eth_rx_vector_data *vec;
+
+			TAILQ_FOREACH(vec, &rx_adapter->vector_list, next) {
+				uint64_t elapsed_time = rte_rdtsc() - vec->ts;
+
+				if (elapsed_time >= vec->vector_timeout_ticks) {
+					rxa_vector_expire(vec, rx_adapter);
+					TAILQ_REMOVE(&rx_adapter->vector_list,
+						     vec, next);
+				}
+			}
+			rx_adapter->prev_expiry_ts = rte_rdtsc();
+		}
+	}
+
 	stats = &rx_adapter->stats;
 	stats->rx_packets += rxa_intr_ring_dequeue(rx_adapter);
 	stats->rx_packets += rxa_poll(rx_adapter);
@@ -1640,6 +1784,28 @@ rxa_update_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }
 
+static void
+rxa_set_vector_data(struct eth_rx_queue_info *queue_info, uint16_t vector_count,
+		    uint64_t vector_ns, struct rte_mempool *mp, int32_t qid,
+		    uint16_t port_id)
+{
+#define NSEC2TICK(__ns, __freq) (((__ns) * (__freq)) / 1E9)
+	struct eth_rx_vector_data *vector_data;
+	uint32_t flow_id;
+
+	vector_data = &queue_info->vector_data;
+	vector_data->max_vector_count = vector_count;
+	vector_data->port = port_id;
+	vector_data->queue = qid;
+	vector_data->vector_pool = mp;
+	vector_data->vector_timeout_ticks =
+		NSEC2TICK(vector_ns, rte_get_timer_hz());
+	vector_data->ts = 0;
+	flow_id = queue_info->event & 0xFFFFF;
+	flow_id = flow_id == 0 ? (qid & 0xFF) | (port_id & 0xFFFF) : flow_id;
+	vector_data->event = (queue_info->event & ~0xFFFFF) | flow_id;
+}
+
 static void
 rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
 	struct eth_device_info *dev_info,
@@ -1716,6 +1882,25 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	} else
 		qi_ev->flow_id = 0;
 
+	if (conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+		queue_info->ena_vector = 1;
+		qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+		rxa_set_vector_data(queue_info, conf->vector_sz,
+				    conf->vector_timeout_ns, conf->vector_mp,
+				    rx_queue_id, dev_info->dev->data->port_id);
+		rx_adapter->ena_vector = 1;
+		rx_adapter->vector_tmo_ticks =
+			rx_adapter->vector_tmo_ticks
+				? RTE_MIN(queue_info->vector_data
+						  .vector_timeout_ticks,
+					  rx_adapter->vector_tmo_ticks)
+				: queue_info->vector_data.vector_timeout_ticks;
+		rx_adapter->vector_tmo_ticks <<= 1;
+		TAILQ_INIT(&rx_adapter->vector_list);
+		rx_adapter->prev_expiry_ts = 0;
+	}
+
 	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
 	if (rxa_polled_queue(dev_info, rx_queue_id)) {
 		rx_adapter->num_rx_polled += !pollq;
@@ -2054,6 +2239,7 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	struct rte_event_eth_rx_adapter *rx_adapter;
 	struct rte_eventdev *dev;
 	struct eth_device_info *dev_info;
+	struct rte_event_eth_rx_adapter_vector_limits limits;
 
 	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
 	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
@@ -2081,6 +2267,48 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}
 
+	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
+	    (queue_conf->rx_queue_flags &
+	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
+		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
+	if (queue_conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+		ret = rte_event_eth_rx_adapter_vector_limits_get(
+			rx_adapter->eventdev_id, eth_dev_id, &limits);
+		if (ret < 0) {
+			RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_sz < limits.min_sz ||
+		    queue_conf->vector_sz > limits.max_sz ||
+		    queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
+		    queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
+		    queue_conf->vector_mp == NULL) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_mp->elt_size <
+		    (sizeof(struct rte_event_vector) +
+		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+	}
+
 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
 		(rx_queue_id != -1)) {
 		RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
@@ -2143,6 +2371,17 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	return 0;
 }
 
+static int
+rxa_sw_vector_limits(struct rte_event_eth_rx_adapter_vector_limits *limits)
+{
+	limits->max_sz = MAX_VECTOR_SIZE;
+	limits->min_sz = MIN_VECTOR_SIZE;
+	limits->max_timeout_ns = MAX_VECTOR_NS;
+	limits->min_timeout_ns = MIN_VECTOR_NS;
+
+	return 0;
+}
+
 int
 rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 				int32_t rx_queue_id)
@@ -2263,6 +2502,44 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }
 
+int
+rte_event_eth_rx_adapter_vector_limits_get(
+	uint8_t dev_id, uint16_t eth_port_id,
+	struct rte_event_eth_rx_adapter_vector_limits *limits)
+{
+	struct rte_eventdev *dev;
+	uint32_t cap;
+	int ret;
+
+	RTE_EVENTDEV_VALID_DEVID_OR_ERR_RET(dev_id, -EINVAL);
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_port_id, -EINVAL);
+
+	if (limits == NULL)
+		return -EINVAL;
+
+	dev = &rte_eventdevs[dev_id];
+
+	ret = rte_event_eth_rx_adapter_caps_get(dev_id, eth_port_id, &cap);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
+				 "eth port %" PRIu16,
+				 dev_id, eth_port_id);
+		return ret;
+	}
+
+	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
+		RTE_FUNC_PTR_OR_ERR_RET(
+			*dev->dev_ops->eth_rx_adapter_vector_limits_get,
+			-ENOTSUP);
+		ret = dev->dev_ops->eth_rx_adapter_vector_limits_get(
+			dev, &rte_eth_devices[eth_port_id], limits);
+	} else {
+		ret = rxa_sw_vector_limits(limits);
+	}
+
+	return ret;
+}
+
 int
 rte_event_eth_rx_adapter_start(uint8_t id)
 {
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index b57363f80..2e6e367e0 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -122,7 +122,11 @@ rte_event_eth_rx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
 
 	if (caps == NULL)
 		return -EINVAL;
-	*caps = 0;
+
+	if (dev->dev_ops->eth_rx_adapter_caps_get == NULL)
+		*caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
+	else
+		*caps = 0;
 
 	return dev->dev_ops->eth_rx_adapter_caps_get ?
 				(*dev->dev_ops->eth_rx_adapter_caps_get)(dev,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH 5/7] eventdev: add Tx adapter event vector support
  2021-02-20 22:09 [dpdk-dev] [PATCH 0/7] Introduce event vectorization pbhagavatula
                   ` (3 preceding siblings ...)
  2021-02-20 22:09 ` [dpdk-dev] [PATCH 4/7] eventdev: add Rx adapter event vector support pbhagavatula
@ 2021-02-20 22:09 ` pbhagavatula
  2021-02-20 22:09 ` [dpdk-dev] [PATCH 6/7] app/eventdev: add event vector mode in pipeline test pbhagavatula
                   ` (3 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-02-20 22:09 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support for event eth Tx adapter, the implementation
receives events from the single linked queue and based on
rte_event_vector::union_valid transmits the vector of mbufs to a given
port, queue pair.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 .../rte_event_eth_tx_adapter.c                | 66 ++++++++++++++++---
 lib/librte_eventdev/rte_eventdev.c            |  5 +-
 2 files changed, 60 insertions(+), 11 deletions(-)

diff --git a/lib/librte_eventdev/rte_event_eth_tx_adapter.c b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
index 5b4c42dcf..b50be74a8 100644
--- a/lib/librte_eventdev/rte_event_eth_tx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
@@ -510,6 +510,47 @@ txa_service_buffer_retry(struct rte_mbuf **pkts, uint16_t unsent,
 	stats->tx_dropped += unsent - sent;
 }
 
+static uint16_t
+txa_process_event_vector(struct txa_service_data *txa,
+			 struct rte_event_vector *vec)
+{
+	struct txa_service_queue_info *tqi;
+	uint16_t port, queue, nb_tx;
+	struct rte_mbuf **mbufs;
+	int i;
+
+	mbufs = (struct rte_mbuf **)vec->mbufs;
+	if (vec->attr_valid) {
+		port = vec->port;
+		queue = vec->queue;
+		tqi = txa_service_queue(txa, port, queue);
+		if (unlikely(tqi == NULL || !tqi->added)) {
+			rte_pktmbuf_free_bulk(mbufs, vec->nb_elem);
+			rte_mempool_put(rte_mempool_from_obj(vec), vec);
+			return 0;
+		}
+		for (i = 0; i < vec->nb_elem; i++) {
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
+						   mbufs[i]);
+		}
+	} else {
+		for (i = 0; i < vec->nb_elem; i++) {
+			port = mbufs[i]->port;
+			queue = rte_event_eth_tx_adapter_txq_get(mbufs[i]);
+			tqi = txa_service_queue(txa, port, queue);
+			if (unlikely(tqi == NULL || !tqi->added)) {
+				rte_pktmbuf_free(mbufs[i]);
+				continue;
+			}
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
+						   mbufs[i]);
+		}
+	}
+	rte_mempool_put(rte_mempool_from_obj(vec), vec);
+
+	return nb_tx;
+}
+
 static void
 txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
 	uint32_t n)
@@ -522,22 +563,27 @@ txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
 
 	nb_tx = 0;
 	for (i = 0; i < n; i++) {
-		struct rte_mbuf *m;
 		uint16_t port;
 		uint16_t queue;
 		struct txa_service_queue_info *tqi;
 
-		m = ev[i].mbuf;
-		port = m->port;
-		queue = rte_event_eth_tx_adapter_txq_get(m);
+		if (!(ev[i].event_type & RTE_EVENT_TYPE_VECTOR)) {
+			struct rte_mbuf *m;
 
-		tqi = txa_service_queue(txa, port, queue);
-		if (unlikely(tqi == NULL || !tqi->added)) {
-			rte_pktmbuf_free(m);
-			continue;
-		}
+			m = ev[i].mbuf;
+			port = m->port;
+			queue = rte_event_eth_tx_adapter_txq_get(m);
 
-		nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
+			tqi = txa_service_queue(txa, port, queue);
+			if (unlikely(tqi == NULL || !tqi->added)) {
+				rte_pktmbuf_free(m);
+				continue;
+			}
+
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
+		} else {
+			nb_tx += txa_process_event_vector(txa, ev[i].vec);
+		}
 	}
 
 	stats->tx_packets += nb_tx;
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index 2e6e367e0..b2e85ce67 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -196,7 +196,10 @@ rte_event_eth_tx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
 	if (caps == NULL)
 		return -EINVAL;
 
-	*caps = 0;
+	if (dev->dev_ops->eth_tx_adapter_caps_get == NULL)
+		*caps = RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR;
+	else
+		*caps = 0;
 
 	return dev->dev_ops->eth_tx_adapter_caps_get ?
 			(*dev->dev_ops->eth_tx_adapter_caps_get)(dev,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH 6/7] app/eventdev: add event vector mode in pipeline test
  2021-02-20 22:09 [dpdk-dev] [PATCH 0/7] Introduce event vectorization pbhagavatula
                   ` (4 preceding siblings ...)
  2021-02-20 22:09 ` [dpdk-dev] [PATCH 5/7] eventdev: add Tx " pbhagavatula
@ 2021-02-20 22:09 ` pbhagavatula
  2021-02-20 22:09 ` [dpdk-dev] [PATCH 7/7] eventdev: fix ABI breakage due to event vector pbhagavatula
                   ` (2 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-02-20 22:09 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support in pipeline tests. By default this mode
is disabled, it can be enabled by using the option --enable_vector.
example:
	dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
	--nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
	--wlcores=20-23  --enable_vector

Addtional options to configure vector size and vector timeout are
also implemented and can be used by specifying --vector_size and
--vector_tmo_ns

This patch also adds a new option to set the number of Rx queues
configured per event eth rx adapter.
example:
	dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
	--nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
	--wlcores=20-23  --nb_eth_queues 4

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 app/test-eventdev/evt_common.h           |   4 +
 app/test-eventdev/evt_options.c          |  52 ++++
 app/test-eventdev/evt_options.h          |   4 +
 app/test-eventdev/test_pipeline_atq.c    | 310 ++++++++++++++++++++--
 app/test-eventdev/test_pipeline_common.c |  69 ++++-
 app/test-eventdev/test_pipeline_common.h |  18 ++
 app/test-eventdev/test_pipeline_queue.c  | 320 +++++++++++++++++++++--
 doc/guides/tools/testeventdev.rst        |  28 ++
 8 files changed, 751 insertions(+), 54 deletions(-)

diff --git a/app/test-eventdev/evt_common.h b/app/test-eventdev/evt_common.h
index a1da1cf11..0e228258e 100644
--- a/app/test-eventdev/evt_common.h
+++ b/app/test-eventdev/evt_common.h
@@ -58,16 +58,20 @@ struct evt_options {
 	uint8_t sched_type_list[EVT_MAX_STAGES];
 	uint16_t mbuf_sz;
 	uint16_t wkr_deq_dep;
+	uint16_t vector_size;
+	uint16_t eth_queues;
 	uint32_t nb_flows;
 	uint32_t tx_first;
 	uint32_t max_pkt_sz;
 	uint32_t deq_tmo_nsec;
 	uint32_t q_priority:1;
 	uint32_t fwd_latency:1;
+	uint32_t ena_vector : 1;
 	uint64_t nb_pkts;
 	uint64_t nb_timers;
 	uint64_t expiry_nsec;
 	uint64_t max_tmo_nsec;
+	uint64_t vector_tmo_nsec;
 	uint64_t timer_tick_nsec;
 	uint64_t optm_timer_tick_nsec;
 	enum evt_prod_type prod_type;
diff --git a/app/test-eventdev/evt_options.c b/app/test-eventdev/evt_options.c
index 0d04ea9f8..0d5540574 100644
--- a/app/test-eventdev/evt_options.c
+++ b/app/test-eventdev/evt_options.c
@@ -34,6 +34,9 @@ evt_options_default(struct evt_options *opt)
 	opt->max_tmo_nsec = 1E5;  /* 100000ns ~100us */
 	opt->expiry_nsec = 1E4;   /* 10000ns ~10us */
 	opt->prod_type = EVT_PROD_TYPE_SYNT;
+	opt->eth_queues = 1;
+	opt->vector_size = 64;
+	opt->vector_tmo_nsec = 100E3;
 }
 
 typedef int (*option_parser_t)(struct evt_options *opt,
@@ -257,6 +260,43 @@ evt_parse_max_pkt_sz(struct evt_options *opt, const char *arg)
 	return ret;
 }
 
+static int
+evt_parse_ena_vector(struct evt_options *opt, const char *arg __rte_unused)
+{
+	opt->ena_vector = 1;
+	return 0;
+}
+
+static int
+evt_parse_vector_size(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint16(&(opt->vector_size), arg);
+
+	return ret;
+}
+
+static int
+evt_parse_vector_tmo_ns(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint64(&(opt->vector_tmo_nsec), arg);
+
+	return ret;
+}
+
+static int
+evt_parse_eth_queues(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint16(&(opt->eth_queues), arg);
+
+	return ret;
+}
+
 static void
 usage(char *program)
 {
@@ -289,6 +329,10 @@ usage(char *program)
 		"\t--expiry_nsec      : event timer expiry ns.\n"
 		"\t--mbuf_sz          : packet mbuf size.\n"
 		"\t--max_pkt_sz       : max packet size.\n"
+		"\t--nb_eth_queues    : number of ethernet Rx queues.\n"
+		"\t--enable_vector    : enable event vectorization.\n"
+		"\t--vector_size      : Max vector size.\n"
+		"\t--vector_tmo_ns    : Max vector timeout in nanoseconds\n"
 		);
 	printf("available tests:\n");
 	evt_test_dump_names();
@@ -360,6 +404,10 @@ static struct option lgopts[] = {
 	{ EVT_EXPIRY_NSEC,         1, 0, 0 },
 	{ EVT_MBUF_SZ,             1, 0, 0 },
 	{ EVT_MAX_PKT_SZ,          1, 0, 0 },
+	{ EVT_NB_ETH_QUEUES,       1, 0, 0 },
+	{ EVT_ENA_VECTOR,          0, 0, 0 },
+	{ EVT_VECTOR_SZ,           1, 0, 0 },
+	{ EVT_VECTOR_TMO,          1, 0, 0 },
 	{ EVT_HELP,                0, 0, 0 },
 	{ NULL,                    0, 0, 0 }
 };
@@ -394,6 +442,10 @@ evt_opts_parse_long(int opt_idx, struct evt_options *opt)
 		{ EVT_EXPIRY_NSEC, evt_parse_expiry_nsec},
 		{ EVT_MBUF_SZ, evt_parse_mbuf_sz},
 		{ EVT_MAX_PKT_SZ, evt_parse_max_pkt_sz},
+		{ EVT_NB_ETH_QUEUES, evt_parse_eth_queues},
+		{ EVT_ENA_VECTOR, evt_parse_ena_vector},
+		{ EVT_VECTOR_SZ, evt_parse_vector_size},
+		{ EVT_VECTOR_TMO, evt_parse_vector_tmo_ns},
 	};
 
 	for (i = 0; i < RTE_DIM(parsermap); i++) {
diff --git a/app/test-eventdev/evt_options.h b/app/test-eventdev/evt_options.h
index 748e54fae..1cea2a3e1 100644
--- a/app/test-eventdev/evt_options.h
+++ b/app/test-eventdev/evt_options.h
@@ -42,6 +42,10 @@
 #define EVT_EXPIRY_NSEC          ("expiry_nsec")
 #define EVT_MBUF_SZ              ("mbuf_sz")
 #define EVT_MAX_PKT_SZ           ("max_pkt_sz")
+#define EVT_NB_ETH_QUEUES        ("nb_eth_queues")
+#define EVT_ENA_VECTOR           ("enable_vector")
+#define EVT_VECTOR_SZ            ("vector_size")
+#define EVT_VECTOR_TMO           ("vector_tmo_ns")
 #define EVT_HELP                 ("help")
 
 void evt_options_default(struct evt_options *opt);
diff --git a/app/test-eventdev/test_pipeline_atq.c b/app/test-eventdev/test_pipeline_atq.c
index 0872b25b5..84dd4f44e 100644
--- a/app/test-eventdev/test_pipeline_atq.c
+++ b/app/test-eventdev/test_pipeline_atq.c
@@ -15,6 +15,8 @@ pipeline_atq_nb_event_queues(struct evt_options *opt)
 	return rte_eth_dev_count_avail();
 }
 
+typedef int (*pipeline_atq_worker_t)(void *arg);
+
 static __rte_noinline int
 pipeline_atq_worker_single_stage_tx(void *arg)
 {
@@ -113,6 +115,112 @@ pipeline_atq_worker_single_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_atq_worker_single_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+		vector_sz = ev.vec->nb_elem;
+		pipeline_event_tx_vector(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = ev.vec->nb_elem;
+		ev.queue_id = tx_queue[ev.vec->port];
+		ev.vec->queue = 0;
+		pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			vector_sz += ev[i].vec->nb_elem;
+			ev[i].vec->queue = 0;
+		}
+
+		pipeline_event_tx_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			ev[i].queue_id = tx_queue[ev[i].vec->port];
+			ev[i].vec->queue = 0;
+			vector_sz += ev[i].vec->nb_elem;
+			pipeline_fwd_event_vector(&ev[i],
+						  RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
 static __rte_noinline int
 pipeline_atq_worker_multi_stage_tx(void *arg)
 {
@@ -245,6 +353,147 @@ pipeline_atq_worker_multi_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.sub_event_type % nb_stages;
+
+		if (cq_id == last_queue) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+			continue;
+		}
+
+		ev.sub_event_type++;
+		pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.sub_event_type % nb_stages;
+
+		if (cq_id == last_queue) {
+			ev.queue_id = tx_queue[ev.vec->port];
+			ev.vec->queue = 0;
+			vector_sz = ev.vec->nb_elem;
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			pipeline_event_enqueue(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.sub_event_type++;
+			pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+			pipeline_event_enqueue(dev, port, &ev);
+		}
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].sub_event_type % nb_stages;
+
+			if (cq_id == last_queue) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+				continue;
+			}
+
+			ev[i].sub_event_type++;
+			pipeline_fwd_event_vector(&ev[i],
+						  sched_type_list[cq_id]);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].sub_event_type % nb_stages;
+
+			if (cq_id == last_queue) {
+				vector_sz = ev[i].vec->nb_elem;
+				ev[i].queue_id = tx_queue[ev[i].vec->port];
+				ev[i].vec->queue = 0;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].sub_event_type++;
+				pipeline_fwd_event_vector(
+					&ev[i], sched_type_list[cq_id]);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
 static int
 worker_wrapper(void *arg)
 {
@@ -253,27 +502,36 @@ worker_wrapper(void *arg)
 	const bool burst = evt_has_burst_mode(w->dev_id);
 	const bool internal_port = w->t->internal_port;
 	const uint8_t nb_stages = opt->nb_stages;
-	RTE_SET_USED(opt);
+	/*vector/burst/internal_port*/
+	const pipeline_atq_worker_t
+	pipeline_atq_worker_single_stage[2][2][2] = {
+		[0][0][0] = pipeline_atq_worker_single_stage_fwd,
+		[0][0][1] = pipeline_atq_worker_single_stage_tx,
+		[0][1][0] = pipeline_atq_worker_single_stage_burst_fwd,
+		[0][1][1] = pipeline_atq_worker_single_stage_burst_tx,
+		[1][0][0] = pipeline_atq_worker_single_stage_fwd_vector,
+		[1][0][1] = pipeline_atq_worker_single_stage_tx_vector,
+		[1][1][0] = pipeline_atq_worker_single_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_atq_worker_single_stage_burst_tx_vector,
+	};
+	const pipeline_atq_worker_t
+	pipeline_atq_worker_multi_stage[2][2][2] = {
+		[0][0][0] = pipeline_atq_worker_multi_stage_fwd,
+		[0][0][1] = pipeline_atq_worker_multi_stage_tx,
+		[0][1][0] = pipeline_atq_worker_multi_stage_burst_fwd,
+		[0][1][1] = pipeline_atq_worker_multi_stage_burst_tx,
+		[1][0][0] = pipeline_atq_worker_multi_stage_fwd_vector,
+		[1][0][1] = pipeline_atq_worker_multi_stage_tx_vector,
+		[1][1][0] = pipeline_atq_worker_multi_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_atq_worker_multi_stage_burst_tx_vector,
+	};
 
-	if (nb_stages == 1) {
-		if (!burst && internal_port)
-			return pipeline_atq_worker_single_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_atq_worker_single_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_atq_worker_single_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_atq_worker_single_stage_burst_fwd(arg);
-	} else {
-		if (!burst && internal_port)
-			return pipeline_atq_worker_multi_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_atq_worker_multi_stage_fwd(arg);
-		if (burst && internal_port)
-			return pipeline_atq_worker_multi_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_atq_worker_multi_stage_burst_fwd(arg);
-	}
+	if (nb_stages == 1)
+		return (pipeline_atq_worker_single_stage[opt->ena_vector][burst]
+							[internal_port])(arg);
+	else
+		return (pipeline_atq_worker_multi_stage[opt->ena_vector][burst]
+						       [internal_port])(arg);
 
 	rte_panic("invalid worker\n");
 }
@@ -290,7 +548,7 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
 	int ret;
 	int nb_ports;
 	int nb_queues;
-	uint8_t queue;
+	uint8_t queue, is_prod;
 	uint8_t tx_evqueue_id[RTE_MAX_ETHPORTS];
 	uint8_t queue_arr[RTE_EVENT_MAX_QUEUES_PER_DEV];
 	uint8_t nb_worker_queues = 0;
@@ -330,15 +588,19 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
 		q_conf.event_queue_cfg = RTE_EVENT_QUEUE_CFG_ALL_TYPES;
 
 		if (!t->internal_port) {
+			is_prod = false;
 			RTE_ETH_FOREACH_DEV(prod) {
 				if (queue == tx_evqueue_id[prod]) {
 					q_conf.event_queue_cfg =
 						RTE_EVENT_QUEUE_CFG_SINGLE_LINK;
-				} else {
-					queue_arr[nb_worker_queues] = queue;
-					nb_worker_queues++;
+					is_prod = true;
+					break;
 				}
 			}
+			if (!is_prod) {
+				queue_arr[nb_worker_queues] = queue;
+				nb_worker_queues++;
+			}
 		}
 
 		ret = rte_event_queue_setup(opt->dev_id, queue, &q_conf);
diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index b47d76743..89f73be86 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -36,6 +36,12 @@ pipeline_opt_dump(struct evt_options *opt, uint8_t nb_queues)
 	evt_dump_queue_priority(opt);
 	evt_dump_sched_type_list(opt);
 	evt_dump_producer_type(opt);
+	evt_dump("nb_eth_rx_queues", "%d", opt->eth_queues);
+	evt_dump("event_vector", "%d", opt->ena_vector);
+	if (opt->ena_vector) {
+		evt_dump("vector_size", "%d", opt->vector_size);
+		evt_dump("vector_tmo_ns", "%ld", opt->vector_tmo_nsec);
+	}
 }
 
 static inline uint64_t
@@ -163,7 +169,7 @@ pipeline_opt_check(struct evt_options *opt, uint64_t nb_queues)
 int
 pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 {
-	uint16_t i;
+	uint16_t i, j;
 	int ret;
 	uint8_t nb_queues = 1;
 	struct test_pipeline *t = evt_test_priv(test);
@@ -210,6 +216,16 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 		if (!(caps & RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT))
 			t->internal_port = 0;
 
+		ret = rte_event_eth_rx_adapter_caps_get(opt->dev_id, i, &caps);
+		if (ret != 0) {
+			evt_err("failed to get event tx adapter[%d] caps", i);
+			return ret;
+		}
+
+		if (!(caps & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT))
+			local_port_conf.rxmode.offloads |=
+				DEV_RX_OFFLOAD_RSS_HASH;
+
 		ret = rte_eth_dev_info_get(i, &dev_info);
 		if (ret != 0) {
 			evt_err("Error during getting device (port %u) info: %s\n",
@@ -236,19 +252,22 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 				local_port_conf.rx_adv_conf.rss_conf.rss_hf);
 		}
 
-		if (rte_eth_dev_configure(i, nb_queues, nb_queues,
-					&local_port_conf)
-				< 0) {
+		if (rte_eth_dev_configure(i, opt->eth_queues, nb_queues,
+					  &local_port_conf) < 0) {
 			evt_err("Failed to configure eth port [%d]", i);
 			return -EINVAL;
 		}
 
-		if (rte_eth_rx_queue_setup(i, 0, NB_RX_DESC,
-				rte_socket_id(), &rx_conf, t->pool) < 0) {
-			evt_err("Failed to setup eth port [%d] rx_queue: %d.",
+		for (j = 0; j < opt->eth_queues; j++) {
+			if (rte_eth_rx_queue_setup(i, j, NB_RX_DESC,
+						   rte_socket_id(), &rx_conf,
+						   t->pool) < 0) {
+				evt_err("Failed to setup eth port [%d] rx_queue: %d.",
 					i, 0);
-			return -EINVAL;
+				return -EINVAL;
+			}
 		}
+
 		if (rte_eth_tx_queue_setup(i, 0, NB_TX_DESC,
 					rte_socket_id(), NULL) < 0) {
 			evt_err("Failed to setup eth port [%d] tx_queue: %d.",
@@ -310,11 +329,24 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 {
 	int ret = 0;
 	uint16_t prod;
+	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
 
 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
 	queue_conf.ev.sched_type = opt->sched_type_list[0];
+	if (opt->ena_vector) {
+		unsigned int nb_elem = (opt->pool_sz / opt->vector_size) << 1;
+
+		nb_elem = nb_elem ? nb_elem : 1;
+		vector_pool = rte_event_vector_pool_create(
+			"vector_pool", nb_elem, 0, opt->vector_size,
+			opt->socket_id);
+		if (vector_pool == NULL) {
+			evt_err("failed to create event vector pool");
+			return -ENOMEM;
+		}
+	}
 	RTE_ETH_FOREACH_DEV(prod) {
 		uint32_t cap;
 
@@ -326,6 +358,19 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 					opt->dev_id);
 			return ret;
 		}
+		if (opt->ena_vector) {
+			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.vector_sz = opt->vector_size;
+				queue_conf.vector_timeout_ns =
+					opt->vector_tmo_nsec;
+				queue_conf.rx_queue_flags |=
+				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+				queue_conf.vector_mp = vector_pool;
+			} else {
+				evt_err("Rx adapter doesn't support event vector");
+				return -EINVAL;
+			}
+		}
 		queue_conf.ev.queue_id = prod * stride;
 		ret = rte_event_eth_rx_adapter_create(prod, opt->dev_id,
 				&prod_conf);
@@ -378,6 +423,14 @@ pipeline_event_tx_adapter_setup(struct evt_options *opt,
 			return ret;
 		}
 
+		if (opt->ena_vector) {
+			if (!(cap &
+			      RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR)) {
+				evt_err("Tx adapter doesn't support event vector");
+				return -EINVAL;
+			}
+		}
+
 		ret = rte_event_eth_tx_adapter_create(consm, opt->dev_id,
 				&port_conf);
 		if (ret) {
diff --git a/app/test-eventdev/test_pipeline_common.h b/app/test-eventdev/test_pipeline_common.h
index 6e73c6ab2..800a90616 100644
--- a/app/test-eventdev/test_pipeline_common.h
+++ b/app/test-eventdev/test_pipeline_common.h
@@ -101,6 +101,14 @@ pipeline_fwd_event(struct rte_event *ev, uint8_t sched)
 	ev->sched_type = sched;
 }
 
+static __rte_always_inline void
+pipeline_fwd_event_vector(struct rte_event *ev, uint8_t sched)
+{
+	ev->event_type = RTE_EVENT_TYPE_CPU_VECTOR;
+	ev->op = RTE_EVENT_OP_FORWARD;
+	ev->sched_type = sched;
+}
+
 static __rte_always_inline void
 pipeline_event_tx(const uint8_t dev, const uint8_t port,
 		struct rte_event * const ev)
@@ -110,6 +118,16 @@ pipeline_event_tx(const uint8_t dev, const uint8_t port,
 		rte_pause();
 }
 
+static __rte_always_inline void
+pipeline_event_tx_vector(const uint8_t dev, const uint8_t port,
+			 struct rte_event *const ev)
+{
+	ev->vec->queue = 0;
+
+	while (!rte_event_eth_tx_adapter_enqueue(dev, port, ev, 1, 0))
+		rte_pause();
+}
+
 static __rte_always_inline void
 pipeline_event_tx_burst(const uint8_t dev, const uint8_t port,
 		struct rte_event *ev, const uint16_t nb_rx)
diff --git a/app/test-eventdev/test_pipeline_queue.c b/app/test-eventdev/test_pipeline_queue.c
index 9a9febb19..f6cc3e358 100644
--- a/app/test-eventdev/test_pipeline_queue.c
+++ b/app/test-eventdev/test_pipeline_queue.c
@@ -15,6 +15,8 @@ pipeline_queue_nb_event_queues(struct evt_options *opt)
 	return (eth_count * opt->nb_stages) + eth_count;
 }
 
+typedef int (*pipeline_queue_worker_t)(void *arg);
+
 static __rte_noinline int
 pipeline_queue_worker_single_stage_tx(void *arg)
 {
@@ -126,6 +128,125 @@ pipeline_queue_worker_single_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_queue_worker_single_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		if (ev.sched_type == RTE_SCHED_TYPE_ATOMIC) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.queue_id++;
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			pipeline_event_enqueue(dev, port, &ev);
+		}
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		ev.queue_id = tx_queue[ev.vec->port];
+		ev.vec->queue = 0;
+		vector_sz = ev.vec->nb_elem;
+		pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			if (ev[i].sched_type == RTE_SCHED_TYPE_ATOMIC) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].queue_id++;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			ev[i].queue_id = tx_queue[ev[i].vec->port];
+			ev[i].vec->queue = 0;
+			vector_sz += ev[i].vec->nb_elem;
+			pipeline_fwd_event_vector(&ev[i],
+						  RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
 
 static __rte_noinline int
 pipeline_queue_worker_multi_stage_tx(void *arg)
@@ -267,6 +388,151 @@ pipeline_queue_worker_multi_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.queue_id % nb_stages;
+
+		if (ev.queue_id == tx_queue[ev.vec->port]) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+			continue;
+		}
+
+		ev.queue_id++;
+		pipeline_fwd_event_vector(&ev, cq_id != last_queue
+						       ? sched_type_list[cq_id]
+						       : RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.queue_id % nb_stages;
+
+		if (cq_id == last_queue) {
+			vector_sz = ev.vec->nb_elem;
+			ev.queue_id = tx_queue[ev.vec->port];
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.queue_id++;
+			pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+		}
+
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].queue_id % nb_stages;
+
+			if (ev[i].queue_id == tx_queue[ev[i].vec->port]) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+				continue;
+			}
+
+			ev[i].queue_id++;
+			pipeline_fwd_event_vector(
+				&ev[i], cq_id != last_queue
+						? sched_type_list[cq_id]
+						: RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].queue_id % nb_stages;
+
+			if (cq_id == last_queue) {
+				ev[i].queue_id = tx_queue[ev[i].vec->port];
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].queue_id++;
+				pipeline_fwd_event_vector(
+					&ev[i], sched_type_list[cq_id]);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
 static int
 worker_wrapper(void *arg)
 {
@@ -275,29 +541,39 @@ worker_wrapper(void *arg)
 	const bool burst = evt_has_burst_mode(w->dev_id);
 	const bool internal_port = w->t->internal_port;
 	const uint8_t nb_stages = opt->nb_stages;
-	RTE_SET_USED(opt);
-
-	if (nb_stages == 1) {
-		if (!burst && internal_port)
-			return pipeline_queue_worker_single_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_queue_worker_single_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_queue_worker_single_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_queue_worker_single_stage_burst_fwd(
-					arg);
-	} else {
-		if (!burst && internal_port)
-			return pipeline_queue_worker_multi_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_queue_worker_multi_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_queue_worker_multi_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_queue_worker_multi_stage_burst_fwd(arg);
+	/*vector/burst/internal_port*/
+	const pipeline_queue_worker_t
+	pipeline_queue_worker_single_stage[2][2][2] = {
+		[0][0][0] = pipeline_queue_worker_single_stage_fwd,
+		[0][0][1] = pipeline_queue_worker_single_stage_tx,
+		[0][1][0] = pipeline_queue_worker_single_stage_burst_fwd,
+		[0][1][1] = pipeline_queue_worker_single_stage_burst_tx,
+		[1][0][0] = pipeline_queue_worker_single_stage_fwd_vector,
+		[1][0][1] = pipeline_queue_worker_single_stage_tx_vector,
+		[1][1][0] = pipeline_queue_worker_single_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_queue_worker_single_stage_burst_tx_vector,
+	};
+	const pipeline_queue_worker_t
+	pipeline_queue_worker_multi_stage[2][2][2] = {
+		[0][0][0] = pipeline_queue_worker_multi_stage_fwd,
+		[0][0][1] = pipeline_queue_worker_multi_stage_tx,
+		[0][1][0] = pipeline_queue_worker_multi_stage_burst_fwd,
+		[0][1][1] = pipeline_queue_worker_multi_stage_burst_tx,
+		[1][0][0] = pipeline_queue_worker_multi_stage_fwd_vector,
+		[1][0][1] = pipeline_queue_worker_multi_stage_tx_vector,
+		[1][1][0] = pipeline_queue_worker_multi_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_queue_worker_multi_stage_burst_tx_vector,
+	};
+
+	if (nb_stages == 1)
+		return (pipeline_queue_worker_single_stage[opt->ena_vector]
+							  [burst]
+							  [internal_port])(arg);
+	else
+		return (pipeline_queue_worker_multi_stage[opt->ena_vector]
+							 [burst]
+							 [internal_port])(arg);
 
-	}
 	rte_panic("invalid worker\n");
 }
 
diff --git a/doc/guides/tools/testeventdev.rst b/doc/guides/tools/testeventdev.rst
index ad1788a3d..691cf706e 100644
--- a/doc/guides/tools/testeventdev.rst
+++ b/doc/guides/tools/testeventdev.rst
@@ -158,6 +158,26 @@ The following are the application command-line options:
        Set max packet mbuf size. Can be used configure Rx/Tx scatter gather.
        Only applicable for `pipeline_atq` and `pipeline_queue` tests.
 
+* ``--nb_eth_queues``
+
+       Configure multiple Rx queues per each ethernet port.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--enable_vector``
+
+       Enable event vector for Rx/Tx adapters.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--vector_size``
+
+       Vector size to configure for the Rx adapter.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--vector_tmo_ns``
+
+       Vector timeout nanoseconds to be configured for the Rx adapter.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
 
 Eventdev Tests
 --------------
@@ -607,6 +627,10 @@ Supported application command line options are following::
         --worker_deq_depth
         --prod_type_ethdev
         --deq_tmo_nsec
+        --nb_eth_queues
+        --enable_vector
+        --vector_size
+        --vector_tmo_ns
 
 
 .. Note::
@@ -699,6 +723,10 @@ Supported application command line options are following::
         --worker_deq_depth
         --prod_type_ethdev
         --deq_tmo_nsec
+        --nb_eth_queues
+        --enable_vector
+        --vector_size
+        --vector_tmo_ns
 
 
 .. Note::
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH 7/7] eventdev: fix ABI breakage due to event vector
  2021-02-20 22:09 [dpdk-dev] [PATCH 0/7] Introduce event vectorization pbhagavatula
                   ` (5 preceding siblings ...)
  2021-02-20 22:09 ` [dpdk-dev] [PATCH 6/7] app/eventdev: add event vector mode in pipeline test pbhagavatula
@ 2021-02-20 22:09 ` pbhagavatula
  2021-03-08 18:44   ` Jerin Jacob
  2021-03-08 16:41 ` [dpdk-dev] [PATCH 0/7] Introduce event vectorization Jerin Jacob
  2021-03-16 15:48 ` [dpdk-dev] [PATCH v2 0/8] " pbhagavatula
  8 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-02-20 22:09 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Fix ABI breakage due to event vector configuration by moving
the vector configuration into a new structure and having a separate
function for enabling the vector config on a given ethernet device and
queue pair.
This vector config and function can be merged to queue config in
v21.11.

Fixes: 44c81670cf0a ("eventdev: introduce event vector Rx capability")

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 app/test-eventdev/test_pipeline_common.c      |  16 +-
 lib/librte_eventdev/eventdev_pmd.h            |  29 +++
 .../rte_event_eth_rx_adapter.c                | 168 ++++++++++++------
 .../rte_event_eth_rx_adapter.h                |  27 +++
 lib/librte_eventdev/version.map               |   1 +
 5 files changed, 184 insertions(+), 57 deletions(-)

diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index 89f73be86..9aeefdd5f 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -331,6 +331,7 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 	uint16_t prod;
 	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
+	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
@@ -360,12 +361,8 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 		}
 		if (opt->ena_vector) {
 			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
-				queue_conf.vector_sz = opt->vector_size;
-				queue_conf.vector_timeout_ns =
-					opt->vector_tmo_nsec;
 				queue_conf.rx_queue_flags |=
 				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
-				queue_conf.vector_mp = vector_pool;
 			} else {
 				evt_err("Rx adapter doesn't support event vector");
 				return -EINVAL;
@@ -385,6 +382,17 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}

+		if (opt->ena_vector) {
+			vec_conf.vector_sz = opt->vector_size;
+			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
+			vec_conf.vector_mp = vector_pool;
+			if (rte_event_eth_rx_adapter_queue_event_vector_config(
+				    prod, prod, -1, &vec_conf) < 0) {
+				evt_err("Failed to configure event vectorization for Rx adapter");
+				return -EINVAL;
+			}
+		}
+
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;

diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 60bfaebc0..d79dfd612 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -667,6 +667,32 @@ typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
 	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

+struct rte_event_eth_rx_adapter_event_vector_config;
+/**
+ * Enable event vector on an given Rx queue of a ethernet devices belonging to
+ * the Rx adapter.
+ *
+ * @param dev
+ *   Event device pointer
+ *
+ * @param eth_dev
+ *   Ethernet device pointer
+ *
+ * @param rx_queue_id
+ *   The Rx queue identifier
+ *
+ * @param config
+ *   Pointer to the event vector configuration structure.
+ *
+ * @return
+ *   - 0: Success.
+ *   - <0: Error code returned by the driver function.
+ */
+typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
+	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
+	int32_t rx_queue_id,
+	const struct rte_event_eth_rx_adapter_event_vector_config *config);
+
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;

@@ -1092,6 +1118,9 @@ struct rte_eventdev_ops {
 	eventdev_eth_rx_adapter_vector_limits_get_t
 		eth_rx_adapter_vector_limits_get;
 	/**< Get event vector limits for the Rx adapter */
+	eventdev_eth_rx_adapter_event_vector_config_t
+		eth_rx_adapter_event_vector_config;
+	/**< Configure Rx adapter with event vector */

 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index a1990637f..c71990078 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -1882,25 +1882,6 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	} else
 		qi_ev->flow_id = 0;

-	if (conf->rx_queue_flags &
-	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
-		queue_info->ena_vector = 1;
-		qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
-		rxa_set_vector_data(queue_info, conf->vector_sz,
-				    conf->vector_timeout_ns, conf->vector_mp,
-				    rx_queue_id, dev_info->dev->data->port_id);
-		rx_adapter->ena_vector = 1;
-		rx_adapter->vector_tmo_ticks =
-			rx_adapter->vector_tmo_ticks
-				? RTE_MIN(queue_info->vector_data
-						  .vector_timeout_ticks,
-					  rx_adapter->vector_tmo_ticks)
-				: queue_info->vector_data.vector_timeout_ticks;
-		rx_adapter->vector_tmo_ticks <<= 1;
-		TAILQ_INIT(&rx_adapter->vector_list);
-		rx_adapter->prev_expiry_ts = 0;
-	}
-
 	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
 	if (rxa_polled_queue(dev_info, rx_queue_id)) {
 		rx_adapter->num_rx_polled += !pollq;
@@ -1926,6 +1907,44 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }

+static void
+rxa_sw_event_vector_configure(
+	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
+	int rx_queue_id,
+	const struct rte_event_eth_rx_adapter_event_vector_config *config)
+{
+	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
+	struct eth_rx_queue_info *queue_info;
+	struct rte_event *qi_ev;
+
+	if (rx_queue_id == -1) {
+		uint16_t nb_rx_queues;
+		uint16_t i;
+
+		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
+		for (i = 0; i < nb_rx_queues; i++)
+			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
+						      config);
+		return;
+	}
+
+	queue_info = &dev_info->rx_queue[rx_queue_id];
+	qi_ev = (struct rte_event *)&queue_info->event;
+	queue_info->ena_vector = 1;
+	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+	rxa_set_vector_data(queue_info, config->vector_sz,
+			    config->vector_timeout_ns, config->vector_mp,
+			    rx_queue_id, dev_info->dev->data->port_id);
+	rx_adapter->ena_vector = 1;
+	rx_adapter->vector_tmo_ticks =
+		rx_adapter->vector_tmo_ticks ?
+			      RTE_MIN(config->vector_timeout_ns << 1,
+				      rx_adapter->vector_tmo_ticks) :
+			      config->vector_timeout_ns << 1;
+	rx_adapter->prev_expiry_ts = 0;
+	TAILQ_INIT(&rx_adapter->vector_list);
+}
+
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -2239,7 +2258,6 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	struct rte_event_eth_rx_adapter *rx_adapter;
 	struct rte_eventdev *dev;
 	struct eth_device_info *dev_info;
-	struct rte_event_eth_rx_adapter_vector_limits limits;

 	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
 	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
@@ -2276,39 +2294,6 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}

-	if (queue_conf->rx_queue_flags &
-	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
-		ret = rte_event_eth_rx_adapter_vector_limits_get(
-			rx_adapter->eventdev_id, eth_dev_id, &limits);
-		if (ret < 0) {
-			RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
-					 " eth port: %" PRIu16
-					 " adapter id: %" PRIu8,
-					 eth_dev_id, id);
-			return -EINVAL;
-		}
-		if (queue_conf->vector_sz < limits.min_sz ||
-		    queue_conf->vector_sz > limits.max_sz ||
-		    queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
-		    queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
-		    queue_conf->vector_mp == NULL) {
-			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-					 " eth port: %" PRIu16
-					 " adapter id: %" PRIu8,
-					 eth_dev_id, id);
-			return -EINVAL;
-		}
-		if (queue_conf->vector_mp->elt_size <
-		    (sizeof(struct rte_event_vector) +
-		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
-			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-					 " eth port: %" PRIu16
-					 " adapter id: %" PRIu8,
-					 eth_dev_id, id);
-			return -EINVAL;
-		}
-	}
-
 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
 		(rx_queue_id != -1)) {
 		RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
@@ -2502,6 +2487,83 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }

+int
+rte_event_eth_rx_adapter_queue_event_vector_config(
+	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
+	struct rte_event_eth_rx_adapter_event_vector_config *config)
+{
+	struct rte_event_eth_rx_adapter_vector_limits limits;
+	struct rte_event_eth_rx_adapter *rx_adapter;
+	struct rte_eventdev *dev;
+	uint32_t cap;
+	int ret;
+
+	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
+
+	rx_adapter = rxa_id_to_adapter(id);
+	if ((rx_adapter == NULL) || (config == NULL))
+		return -EINVAL;
+
+	dev = &rte_eventdevs[rx_adapter->eventdev_id];
+	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
+						eth_dev_id, &cap);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
+				 "eth port %" PRIu16,
+				 id, eth_dev_id);
+		return ret;
+	}
+
+	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
+		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
+	ret = rte_event_eth_rx_adapter_vector_limits_get(
+		rx_adapter->eventdev_id, eth_dev_id, &limits);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
+				 "eth port %" PRIu16,
+				 rx_adapter->eventdev_id, eth_dev_id);
+		return ret;
+	}
+
+	if (config->vector_sz < limits.min_sz ||
+	    config->vector_sz > limits.max_sz ||
+	    config->vector_timeout_ns < limits.min_timeout_ns ||
+	    config->vector_timeout_ns > limits.max_timeout_ns ||
+	    config->vector_mp == NULL) {
+		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+	if (config->vector_mp->elt_size <
+	    (sizeof(struct rte_event_vector) +
+	     (sizeof(uintptr_t) * config->vector_sz))) {
+		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
+	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
+		RTE_FUNC_PTR_OR_ERR_RET(
+			*dev->dev_ops->eth_rx_adapter_event_vector_config,
+			-ENOTSUP);
+		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
+			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
+	} else {
+		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
+					      rx_queue_id, config);
+	}
+
+	return ret;
+}
+
 int
 rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
index 4bdb38f08..ceef6d565 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
@@ -171,6 +171,9 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
 	 * enqueued event.
 	 */
+};
+
+struct rte_event_eth_rx_adapter_event_vector_config {
 	uint16_t vector_sz;
 	/**<
 	 * Indicates the maximum number for mbufs to combine and form a vector.
@@ -418,6 +421,30 @@ int rte_event_eth_rx_adapter_queue_add(uint8_t id,
 int rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 				       int32_t rx_queue_id);

+/**
+ * Configure event vectorization for a given ethernet device queue, that has
+ * been added to a event eth Rx adapter.
+ *
+ * @param id
+ *  The identifier of the ethernet Rx event adapter.
+ *
+ * @param eth_dev_id
+ *  The identifier of the ethernet device.
+ *
+ * @param rx_queue_id
+ *  Ethernet device receive queue index.
+ *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
+ *  are configured with event vectorization.
+ *
+ * @return
+ *  - 0: Success, Receive queue configured correctly.
+ *  - <0: Error code on failure.
+ */
+__rte_experimental
+int rte_event_eth_rx_adapter_queue_event_vector_config(
+	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
+	struct rte_event_eth_rx_adapter_event_vector_config *config);
+
 /**
  * Start ethernet Rx event adapter
  *
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index 34c1c830e..902df0ae3 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -142,6 +142,7 @@ EXPERIMENTAL {
 	#added in 21.05
 	rte_event_vector_pool_create;
 	rte_event_eth_rx_adapter_vector_limits_get;
+	rte_event_eth_rx_adapter_queue_event_vector_config;
 };

 INTERNAL {
--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH 0/7] Introduce event vectorization
  2021-02-20 22:09 [dpdk-dev] [PATCH 0/7] Introduce event vectorization pbhagavatula
                   ` (6 preceding siblings ...)
  2021-02-20 22:09 ` [dpdk-dev] [PATCH 7/7] eventdev: fix ABI breakage due to event vector pbhagavatula
@ 2021-03-08 16:41 ` Jerin Jacob
  2021-03-16 15:48 ` [dpdk-dev] [PATCH v2 0/8] " pbhagavatula
  8 siblings, 0 replies; 153+ messages in thread
From: Jerin Jacob @ 2021-03-08 16:41 UTC (permalink / raw)
  To: Pavan Nikhilesh
  Cc: Jerin Jacob, Jayatheerthan, Jay, Erik Gabriel Carrillo, Gujjar,
	Abhinandan S, McDaniel, Timothy, Hemant Agrawal, Van Haaren,
	Harry, Mattias Rönnblom, Liang Ma, dpdk-dev

On Sun, Feb 21, 2021 at 3:40 AM <pbhagavatula@marvell.com> wrote:
>
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>
> In traditional event programming model, events are identified by a
> flow-id and a uintptr_t. The flow-id uniquely identifies a given event
> and determines the order of scheduling based on schedule type, the
> uintptr_t holds a single object.
>
> Event devices also support burst mode with configurable dequeue depth,
> i.e. each dequeue call would return multiple events and each event
> might be at a different stage of the pipeline.
> Having a burst of events belonging to different stages in a dequeue
> burst is not only difficult to vectorize but also increases the scheduler
> overhead and application overhead of pipelining events further.
> Using event vectors we see a performance gain of ~150% as shown in [1].

Since it is from  4.72 to 34.383, it is 628%.

We had an internal review of this series before it is posting to ML.
So my comments will be limited on this series.

Is anyone planning to review this patch series or any objection to
introducing this feature?
CC list already has all eventdev stakeholders.



>
> By introducing event vectorization, each event will be capable of holding
> multiple uintptr_t of the same flow thereby allowing applications
> to vectorize their pipeline and reduce the complexity of pipelining
> events across multiple stages. This also reduces the complexity of handling
> enqueue and dequeue on an event device.
>
> Since event devices are transparent to the events they are scheduling
> so the event producers such as eth_rx_adapter, crypto_adapter , etc..
> are responsible for vectorizing the buffers of the same flow into a single
> event.
>
> The series also breaks ABI in [2/7] patch which we fix in [7/7]. The patch
> [7/7] can be changed in the next major release i.e. v21.11.
>
> The dpdk-test-eventdev application has been updated with options to test
> multiple vector sizes and timeouts.
>
> [1]
> As for performance improvement, with a ARM Cortex-A72 equivalent processer,
> software event device (--vdev=event_sw0), single worker core, single stage
> and using one service core for Rx adapter, Tx adapter, Scheduling.
>
> Without event vectorization:
>     ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
>          --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
>          --stlist=a --wlcores=20
>     Port[0] using Rx adapter[0] configured
>     Port[0] using Tx adapter[0] Configured
>     4.728 mpps avg 4.728 mpps
>
> With event vectorization:
>     ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
>         --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
>         --stlist=a --wlcores=20 --enable_vector --nb_eth_queues 1
>         --vector_size 256
>     Port[0] using Rx adapter[0] configured
>     Port[0] using Tx adapter[0] Configured
>     34.383 mpps avg 34.383 mpps
>
> Having dedicated service cores for each Rx queues and tweaking the vector,
> dequeue burst size would further improve performance.
>
> API usage is shown below:
>
> Configuration:
>
>         struct rte_event_eth_rx_adapter_event_vector_config vec_conf;
>
>         vector_pool = rte_event_vector_pool_create("vector_pool",
>                         nb_elem, 0, vector_size, socket_id);
>
>         rte_event_eth_rx_adapter_create(id, event_id, &adptr_conf);
>         rte_event_eth_rx_adapter_queue_add(id, eth_id, -1, &queue_conf);
>         if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
>                 vec_conf.vector_sz = vector_size;
>                 vec_conf.vector_timeout_ns = vector_tmo_nsec;
>                 vec_conf.vector_mp = vector_pool;
>                 rte_event_eth_rx_adapter_queue_event_vector_config(id,
>                                 eth_id, -1, &vec_conf);
>         }
>
> Fastpath:
>
>         num = rte_event_dequeue_burst(event_id, port_id, &ev, 1, 0);
>         if (!num)
>                 continue;
>
>         if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
>                 switch (ev.event_type) {
>                 case RTE_EVENT_TYPE_ETHDEV_VECTOR:
>                 case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
>                         struct rte_mbuf **mbufs;
>
>                         mbufs = ev.vector_ev->mbufs;
>                         for (i = 0; i < ev.vector_ev->nb_elem; i++)
>                                 //Process mbufs.
>                         break;
>                 case ...
>                 }
>         }
>         ...
>
> Pavan Nikhilesh (7):
>   eventdev: introduce event vector capability
>   eventdev: introduce event vector Rx capability
>   eventdev: introduce event vector Tx capability
>   eventdev: add Rx adapter event vector support
>   eventdev: add Tx adapter event vector support
>   app/eventdev: add event vector mode in pipeline test
>   eventdev: fix ABI breakage due to event vector
>
>  app/test-eventdev/evt_common.h                |   4 +
>  app/test-eventdev/evt_options.c               |  52 +++
>  app/test-eventdev/evt_options.h               |   4 +
>  app/test-eventdev/test_pipeline_atq.c         | 310 +++++++++++++--
>  app/test-eventdev/test_pipeline_common.c      |  77 +++-
>  app/test-eventdev/test_pipeline_common.h      |  18 +
>  app/test-eventdev/test_pipeline_queue.c       | 320 +++++++++++++--
>  .../prog_guide/event_ethernet_rx_adapter.rst  |  38 ++
>  .../prog_guide/event_ethernet_tx_adapter.rst  |  12 +
>  doc/guides/prog_guide/eventdev.rst            |  36 +-
>  doc/guides/tools/testeventdev.rst             |  28 ++
>  lib/librte_eventdev/eventdev_pmd.h            |  60 ++-
>  .../rte_event_eth_rx_adapter.c                | 367 +++++++++++++++++-
>  .../rte_event_eth_rx_adapter.h                |  93 +++++
>  .../rte_event_eth_tx_adapter.c                |  66 +++-
>  lib/librte_eventdev/rte_eventdev.c            |  11 +-
>  lib/librte_eventdev/rte_eventdev.h            | 145 ++++++-
>  lib/librte_eventdev/version.map               |   5 +
>  18 files changed, 1560 insertions(+), 86 deletions(-)
>
> --
> 2.17.1
>

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH 1/7] eventdev: introduce event vector capability
  2021-02-20 22:09 ` [dpdk-dev] [PATCH 1/7] eventdev: introduce event vector capability pbhagavatula
@ 2021-03-08 16:49   ` Jerin Jacob
  0 siblings, 0 replies; 153+ messages in thread
From: Jerin Jacob @ 2021-03-08 16:49 UTC (permalink / raw)
  To: Pavan Nikhilesh
  Cc: Jerin Jacob, Jayatheerthan, Jay, Erik Gabriel Carrillo, Gujjar,
	Abhinandan S, McDaniel, Timothy, Hemant Agrawal, Van Haaren,
	Harry, Mattias Rönnblom, Liang Ma, Ray Kinsella,
	Neil Horman, dpdk-dev

On Sun, Feb 21, 2021 at 3:40 AM <pbhagavatula@marvell.com> wrote:
>
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>
> Introduce rte_event_vector datastructure which is capable of holding
> multiple uintptr_t of the same flow thereby allowing applications
> to vectorize their pipeline and reducing the complexity of pipelining
> the events across multiple stages.
> This approach also reduces the scheduling overhead on a event device.
>
> Add a event vector mempool create handler to create mempools based on
> the best mempool ops available on a given platform.
>
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> ---
>  doc/guides/prog_guide/eventdev.rst |  36 ++++++++-
>  lib/librte_eventdev/rte_eventdev.h | 113 ++++++++++++++++++++++++++++-
>  lib/librte_eventdev/version.map    |   3 +
>  3 files changed, 149 insertions(+), 3 deletions(-)
>
> diff --git a/doc/guides/prog_guide/eventdev.rst b/doc/guides/prog_guide/eventdev.rst
> index ccde086f6..d19c91ab0 100644
> --- a/doc/guides/prog_guide/eventdev.rst
> +++ b/doc/guides/prog_guide/eventdev.rst
> @@ -63,13 +63,45 @@ the actual event being scheduled is. The payload is a union of the following:
>  * ``uint64_t u64``
>  * ``void *event_ptr``
>  * ``struct rte_mbuf *mbuf``
> +* ``struct rte_event_vector *vec``
>
> -These three items in a union occupy the same 64 bits at the end of the rte_event
> +These four items in a union occupy the same 64 bits at the end of the rte_event
>  structure. The application can utilize the 64 bits directly by accessing the
> -u64 variable, while the event_ptr and mbuf are provided as convenience

as a convenience

> +u64 variable, while the event_ptr, mbuf, vec are provided as convenience
>  variables.  For example the mbuf pointer in the union can used to schedule a
>  DPDK packet.
>
> +Event Vector
> +~~~~~~~~~~~~
> +
> +The rte_event_vector struct contains a vector of elements defined by the event
> +type specified in the ``rte_event``. The event_vector structure contains the
> +following data:
> +
> +* ``nb_elem`` - The number of elements held within the vector.
> +
> +Similar to ``rte_event`` the payload of event vector is also a union, allowing
> +flexibility in what the actual vector is.
> +
> +* ``struct rte_mbuf *mbufs[0]`` - An array of mbufs.
> +* ``void *ptrs[0]`` - An array of pointers.
> +* ``uint64_t *u64s[0]`` - An array of uint64_t elements.
> +
> +The size of the event vector is related to the total number of elements it is
> +configured to hold, this is achieved by making `rte_event_vector` a variable
> +length structure.
> +A helper function is provided to create a mempool that holds event vector, which
> +takes name of the pool, total number of required ``rte_event_vector``,
> +cache size, number of elements in each ``rte_event_vector`` and socket id.
> +
> +.. code-block:: c
> +
> +        rte_event_vector_pool_create("vector_pool", nb_event_vectors, cache_sz,
> +                                     nb_elements_per_vector, socket_id);
> +
> +The function ``rte_event_vector_pool_create`` creates mempool with the best
> +platform mempool ops.
> +
>  Queues
>  ~~~~~~
>
> diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
> index ce1fc2ce0..ff6cb3e6a 100644
> --- a/lib/librte_eventdev/rte_eventdev.h
> +++ b/lib/librte_eventdev/rte_eventdev.h
> @@ -212,8 +212,10 @@ extern "C" {
>
>  #include <rte_common.h>
>  #include <rte_config.h>
> -#include <rte_memory.h>
>  #include <rte_errno.h>
> +#include <rte_mbuf_pool_ops.h>
> +#include <rte_memory.h>
> +#include <rte_mempool.h>
>
>  #include "rte_eventdev_trace_fp.h"
>
> @@ -913,6 +915,25 @@ rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
>  int
>  rte_event_dev_close(uint8_t dev_id);
>
> +/**
> + * Event vector structure.
> + */
> +struct rte_event_vector {
> +       uint64_t nb_elem : 16;
> +       /**< Number of elements in this event vector. */
> +       uint64_t rsvd : 48;
> +       uint64_t impl_opaque;
> +       union {
> +               struct rte_mbuf *mbufs[0];
> +               void *ptrs[0];
> +               uint64_t *u64s[0];
> +       } __rte_aligned(16);
> +       /**< Start of the vector array union. Depending upon the event type the
> +        * vector array can be an array of mbufs or pointers or opaque u64
> +        * values.
> +        */
> +};
> +
>  /* Scheduler type definitions */
>  #define RTE_SCHED_TYPE_ORDERED          0
>  /**< Ordered scheduling
> @@ -986,6 +1007,21 @@ rte_event_dev_close(uint8_t dev_id);
>   */
>  #define RTE_EVENT_TYPE_ETH_RX_ADAPTER   0x4
>  /**< The event generated from event eth Rx adapter */
> +#define RTE_EVENT_TYPE_VECTOR           0x8
> +/**< Indicates that event is a vector.
> + * All vector event types should be an logical OR of EVENT_TYPE_VECTOR.
> + * This simplifies the pipeline design as we can split processing the events
> + * between vector events and normal event across event types.
> + * Example:
> + *     if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
> + *             // Classify and handle vector event.


I think, we can remove C++-style comments from the documentation. ie
change from // to /* */


> + *     } else {
> + *             // Classify and handle event.
> + *     }
> + */
> +#define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
> +/**< The event vector generated from cpu for pipelining. */
> +
>  #define RTE_EVENT_TYPE_MAX              0x10
>  /**< Maximum number of event types */
>
> @@ -1108,6 +1144,8 @@ struct rte_event {
>                 /**< Opaque event pointer */
>                 struct rte_mbuf *mbuf;
>                 /**< mbuf pointer if dequeued event is associated with mbuf */
> +               struct rte_event_vector *vec;
> +               /**< Event vector pointer. */
>         };
>  };
>
> @@ -2023,6 +2061,79 @@ rte_event_dev_xstats_reset(uint8_t dev_id,
>   */
>  int rte_event_dev_selftest(uint8_t dev_id);
>
> +/**
> + * Get the memory required per event vector based on number of elements per

the number of

> + * vector.
> + * This should be used to create the mempool that holds the event vectors.
> + *
> + * @param name
> + *   The name of the vector pool.
> + * @param n
> + *   The number of elements in the mbuf pool.
> + * @param cache_size
> + *   Size of the per-core object cache. See rte_mempool_create() for
> + *   details.
> + * @param nb_elem
> + *   The number of elements then a single event vector should be able to hold.
> + * @param socket_id
> + *   The socket identifier where the memory should be allocated. The
> + *   value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
> + *   reserved zone
> + *
> + * @return
> + *   The pointer to the new allocated mempool, on success. NULL on error

s/new/newly

> + *   with rte_errno set appropriately. Possible rte_errno values include:
> + *    - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
> + *    - E_RTE_SECONDARY - function was called from a secondary process instance
> + *    - EINVAL - cache size provided is too large, or priv_size is not aligned.
> + *    - ENOSPC - the maximum number of memzones has already been allocated
> + *    - EEXIST - a memzone with the same name already exists
> + *    - ENOMEM - no appropriate memory area found in which to create memzone
> + */
> +__rte_experimental
> +static inline struct rte_mempool *
> +rte_event_vector_pool_create(const char *name, unsigned int n,
> +                            unsigned int cache_size, uint16_t nb_elem,
> +                            int socket_id)
> +{
> +       const char *mp_ops_name;
> +       struct rte_mempool *mp;
> +       unsigned int elt_sz;
> +       int ret;
> +
> +       if (!nb_elem) {
> +               RTE_LOG(ERR, EVENTDEV,
> +                       "Invalid number of elements=%d requested\n", nb_elem);
> +               rte_errno = -EINVAL;
> +               return NULL;
> +       }
> +
> +       elt_sz =
> +               sizeof(struct rte_event_vector) + (nb_elem * sizeof(uintptr_t));
> +       mp = rte_mempool_create_empty(name, n, elt_sz, cache_size, 0, socket_id,
> +                                     0);
> +       if (mp == NULL)
> +               return NULL;
> +
> +       mp_ops_name = rte_mbuf_best_mempool_ops();
> +       ret = rte_mempool_set_ops_byname(mp, mp_ops_name, NULL);
> +       if (ret != 0) {
> +               RTE_LOG(ERR, EVENTDEV, "error setting mempool handler\n");
> +               rte_mempool_free(mp);
> +               rte_errno = -ret;

See below

> +               return NULL;
> +       }
> +
> +       ret = rte_mempool_populate_default(mp);
> +       if (ret < 0) {
> +               rte_mempool_free(mp);
> +               rte_errno = -ret;
> +               return NULL;

make it as goto err: kind of structure to avoid code duplication.(See above)


> +       }
> +
> +       return mp;
> +}
> +
>  #ifdef __cplusplus
>  }
>  #endif
> diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
> index 3e5c09cfd..a070ef56e 100644
> --- a/lib/librte_eventdev/version.map
> +++ b/lib/librte_eventdev/version.map
> @@ -138,6 +138,9 @@ EXPERIMENTAL {
>         __rte_eventdev_trace_port_setup;
>         # added in 20.11
>         rte_event_pmd_pci_probe_named;
> +
> +       #added in 21.05
> +       rte_event_vector_pool_create;
>  };
>
>  INTERNAL {
> --
> 2.17.1
>

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH 2/7] eventdev: introduce event vector Rx capability
  2021-02-20 22:09 ` [dpdk-dev] [PATCH 2/7] eventdev: introduce event vector Rx capability pbhagavatula
@ 2021-03-08 17:07   ` Jerin Jacob
  0 siblings, 0 replies; 153+ messages in thread
From: Jerin Jacob @ 2021-03-08 17:07 UTC (permalink / raw)
  To: Pavan Nikhilesh
  Cc: Jerin Jacob, Jayatheerthan, Jay, Erik Gabriel Carrillo, Gujjar,
	Abhinandan S, McDaniel, Timothy, Hemant Agrawal, Van Haaren,
	Harry, Mattias Rönnblom, Liang Ma, Ray Kinsella,
	Neil Horman, dpdk-dev

On Sun, Feb 21, 2021 at 3:40 AM <pbhagavatula@marvell.com> wrote:
>
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>
> Introduce event ethernet Rx adapter event vector capability.
>
> If an event eth Rx adapter has the capability of
> RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR then a given Rx queue
> can be configured to enable event vectorization by passing the
> flag RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR to
> rte_event_eth_rx_adapter_queue_conf::rx_queue_flags while configuring
> Rx adapter through rte_event_eth_rx_adapter_queue_add.
>
> The max vector size, vector timeout define the vector size and
> mempool used for allocating vector event are configured through
> rte_event_eth_rx_adapter_queue_add. The element size of the element
> in the vector pool should be equal to
>     sizeof(struct rte_event_vector) + (vector_sz * sizeof(uintptr_t))
>
> Application can use `rte_event_get_event_vector_memory_footprint`

This comment is outdated. There is no
rte_event_get_event_vector_memory_footprint() in series.

> to get the element size before creating the vector mempool
> rte_event_eth_rx_adapter_queue_conf::vector_mp.
>
> The Rx adapter would be responsible for vectorizing the mbufs
> based on the flow, the vector limits configured by the application
> and add the vector event of mbufs to the event queue set via
> rte_event_eth_rx_adapter_queue_conf::ev::queue_id.
> It should also mark rte_event_vector::union_valid and fill
> rte_event_vector::port, rte_event_vector::queue.
>
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> ---
>  .../prog_guide/event_ethernet_rx_adapter.rst  | 38 +++++++++++
>  .../rte_event_eth_rx_adapter.h                | 66 +++++++++++++++++++
>  lib/librte_eventdev/rte_eventdev.h            | 30 ++++++++-
>  lib/librte_eventdev/version.map               |  1 +
>  4 files changed, 133 insertions(+), 2 deletions(-)
>
> diff --git a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
> index cb44ce0e4..735ea2439 100644
> --- a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
> +++ b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
> @@ -186,3 +186,41 @@ the event buffer fill level is low. The
>  ``rte_event_eth_rx_adapter_cb_register()`` function allow the application
>  to register a callback that selects which packets to enqueue to the event
>  device.
> +
> +Rx event vectorization
> +~~~~~~~~~~~~~~~~~~~~~~
> +
> +The event devices, ethernet device pairs which support the capability
> +``RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR`` can aggregate packets based on
> +flow characteristics and generate a ``rte_event`` containing ``rte_event_vector``
> +whose event type is either ``RTE_EVENT_TYPE_ETHDEV_VECTOR`` or
> +``RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR``.
> +The aggregation size and timeout are configurable at a queue level and the
> +maximum, minimum vector sizes and timeouts vary based on the device capability
> +and can be queried using ``rte_event_eth_rx_adapter_vector_limits_get``.
> +The Rx adapter additionally might include useful data such as ethernet device
> +port and queue identifier in the ``rte_event_vector::port`` and
> +``rte_event_vector::queue`` and mark ``rte_event_vector::attr_valid`` as true.
> +
> +A loop processing ``rte_event_vector`` containing mbufs is shown below.
> +
> +.. code-block:: c
> +
> +        event = rte_event_dequeue_burst(event_dev, event_port, &event,
> +                                        1, 0);
> +        if (!event)
> +                continue;
> +
> +        switch (ev.event_type) {
> +        case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
> +        case RTE_EVENT_TYPE_ETHDEV_VECTOR:
> +                struct rte_mbufs **mbufs;
> +
> +                mbufs = (struct rte_mbufs **)ev[i].vec->mbufs;
> +                for (i = 0; i < ev.vec->nb_elem; i++) {
> +                        // Process each mbuf.

Change to C style /* */ comments.

> +                }
> +        break;
> +        case ...
> +        ...
> +        }
> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> index 21bb1e54c..4bdb38f08 100644
> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> @@ -92,6 +92,10 @@ extern "C" {
>  /**< This flag indicates the flow identifier is valid
>   * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
>   */
> +#define RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR    0x2
> +/**< This flag indicates that mbufs arriving on the queue need to be vectorized
> + * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
> + */

>  /**
> @@ -199,6 +230,20 @@ struct rte_event_eth_rx_adapter_stats {
>         /**< Received packet count for interrupt mode Rx queues */
>  };
>
> +/**
> + * A structure used to retrieve eth rx adapter vector limits.
> + */
> +struct rte_event_eth_rx_adapter_vector_limits {
> +       uint16_t min_sz;
> +       /**< Minimum vector limit configurable. */
> +       uint16_t max_sz;

I think, it will common to have constraints such as size needs to be log2.
Please add a  bool type to express that limit/contrain.

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH 3/7] eventdev: introduce event vector Tx capability
  2021-02-20 22:09 ` [dpdk-dev] [PATCH 3/7] eventdev: introduce event vector Tx capability pbhagavatula
@ 2021-03-08 17:09   ` Jerin Jacob
  0 siblings, 0 replies; 153+ messages in thread
From: Jerin Jacob @ 2021-03-08 17:09 UTC (permalink / raw)
  To: Pavan Nikhilesh
  Cc: Jerin Jacob, Jayatheerthan, Jay, Erik Gabriel Carrillo, Gujjar,
	Abhinandan S, McDaniel, Timothy, Hemant Agrawal, Van Haaren,
	Harry, Mattias Rönnblom, Liang Ma, dpdk-dev

On Sun, Feb 21, 2021 at 3:40 AM <pbhagavatula@marvell.com> wrote:
>
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>
> Introduce event vector transmit capability for event eth
> tx adapter.
>
> The capability indicates that the Tx adapter is capable of
> transmitting event vectors.
> When rte_event_vector::union_valid is set, the Tx adapter should
> transmit all the packets to the rte_event_vector::port using the
> rte_event_vector::queue.
> If rte_event_vector::union_valid is not set then the Tx adapter
> should peek into each mbuf to get the destination port and queue
> pair.
>
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>


Acked-by: Jerin Jacob <jerinj@marvell.com>


> ---
>  doc/guides/prog_guide/event_ethernet_tx_adapter.rst | 12 ++++++++++++
>  lib/librte_eventdev/rte_eventdev.h                  |  8 +++++++-
>  2 files changed, 19 insertions(+), 1 deletion(-)
>
> diff --git a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
> index a8c13e136..87277dcaf 100644
> --- a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
> +++ b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
> @@ -164,3 +164,15 @@ The  ``rte_event_eth_tx_adapter_stats_get()`` function reports counters defined
>  in struct ``rte_event_eth_tx_adapter_stats``. The counter values are the sum of
>  the counts from the eventdev PMD callback if the callback is supported, and
>  the counts maintained by the service function, if one exists.
> +
> +Tx event vectorization
> +~~~~~~~~~~~~~~~~~~~~~~
> +
> +The event device, ethernet device paris which support the capability
> +``RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR`` can process event vector of mbufs.
> +Additionally, application can provide a hint to that Tx adapter that all the
> +mbufs are destined to the same ethernet port and queue by setting the bit
> +``rte_event_vector::attr_valid`` and filling `rte_event_vector::port`` and
> +``rte_event_vector::queue``.
> +If ``rte_event_vector::attr_valid`` is not set then the Tx adapter should peek
> +into each mbuf and transmit them to the requested ethernet port and queue pair.
> diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
> index 1cf3efa2d..c817c29e6 100644
> --- a/lib/librte_eventdev/rte_eventdev.h
> +++ b/lib/librte_eventdev/rte_eventdev.h
> @@ -926,11 +926,13 @@ struct rte_event_vector {
>         /**< Indicates that the below union attributes have valid information.
>          */
>         union {
> -               /* Used by Rx adapter.
> +               /* Used by Rx/Tx adapter.
>                  * Indicats that all the elements in this vector belong to same
>                  * port and queue pair when originating from Rx adapter, valid
>                  * only when event type is ETHDEV_VECTOR or
>                  * ETH_RX_ADAPTER_VECTOR.
> +                * Can also be used to indicate the Tx adapter the destination
> +                * port and queue of the mbufs in the vector
>                  */
>                 struct {
>                         uint16_t port;
> @@ -1287,6 +1289,10 @@ rte_event_crypto_adapter_caps_get(uint8_t dev_id, uint8_t cdev_id,
>  #define RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT     0x1
>  /**< This flag is sent when the PMD supports a packet transmit callback
>   */
> +#define RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR      0x2
> +/**< Indicates that the Tx adapter is capable of handling event vector of
> + * mbufs.
> + */
>
>  /**
>   * Retrieve the event device's eth Tx adapter capabilities
> --
> 2.17.1
>

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH 4/7] eventdev: add Rx adapter event vector support
  2021-02-20 22:09 ` [dpdk-dev] [PATCH 4/7] eventdev: add Rx adapter event vector support pbhagavatula
@ 2021-03-08 17:27   ` Jerin Jacob
  2021-03-16 10:41     ` Jayatheerthan, Jay
  0 siblings, 1 reply; 153+ messages in thread
From: Jerin Jacob @ 2021-03-08 17:27 UTC (permalink / raw)
  To: Pavan Nikhilesh, Nikhil Rao
  Cc: Jerin Jacob, Jayatheerthan, Jay, Erik Gabriel Carrillo, Gujjar,
	Abhinandan S, McDaniel, Timothy, Hemant Agrawal, Van Haaren,
	Harry, Mattias Rönnblom, Liang Ma, dpdk-dev

On Sun, Feb 21, 2021 at 3:40 AM <pbhagavatula@marvell.com> wrote:
>
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>
> Add event vector support for event eth Rx adapter, the implementation
> creates vector flows based on port and queue identifier of the received
> mbufs.
>
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>


Hi @Jayatheerthan, Jay

Could you review 4/7 and 5/7 SW adapter-related patches?
Cc: @Nikhil Rao



> ---
>  lib/librte_eventdev/eventdev_pmd.h            |  31 +-
>  .../rte_event_eth_rx_adapter.c                | 305 +++++++++++++++++-
>  lib/librte_eventdev/rte_eventdev.c            |   6 +-
>  3 files changed, 324 insertions(+), 18 deletions(-)

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH 7/7] eventdev: fix ABI breakage due to event vector
  2021-02-20 22:09 ` [dpdk-dev] [PATCH 7/7] eventdev: fix ABI breakage due to event vector pbhagavatula
@ 2021-03-08 18:44   ` Jerin Jacob
  2021-03-12 14:28     ` David Marchand
  2021-03-15 10:01     ` [dpdk-dev] " Kinsella, Ray
  0 siblings, 2 replies; 153+ messages in thread
From: Jerin Jacob @ 2021-03-08 18:44 UTC (permalink / raw)
  To: Pavan Nikhilesh
  Cc: Jerin Jacob, Jayatheerthan, Jay, Erik Gabriel Carrillo, Gujjar,
	Abhinandan S, McDaniel, Timothy, Hemant Agrawal, Van Haaren,
	Harry, Mattias Rönnblom, Liang Ma, Ray Kinsella,
	Neil Horman, dpdk-dev, Thomas Monjalon, David Marchand

On Sun, Feb 21, 2021 at 3:41 AM <pbhagavatula@marvell.com> wrote:
>
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>
> Fix ABI breakage due to event vector configuration by moving
> the vector configuration into a new structure and having a separate
> function for enabling the vector config on a given ethernet device and
> queue pair.
> This vector config and function can be merged to queue config in
> v21.11.
>
> Fixes: 44c81670cf0a ("eventdev: introduce event vector Rx capability")

Hi @Ray Kinsella @Neil Horman @Thomas Monjalon @David Marchand

Is the ABI breakage contract between release to release. Right? i.e it
is not between each patch. Right?

Summary:
1)  Ideal way of adding this feature is to add elements in the
existing structure as mentioned
in  ("eventdev: introduce event vector Rx capability") in this series.
2) Since this breaking ABI, Introducing a new structure to fix this. I
think, we can remove this
limitation in 21.11 as that time we can change ABI as required.

So, Is this patch needs to be squashed to  ("eventdev: introduce event
vector Rx capability") to avoid
ABI compatibility between patches? Or Is it OK to break the ABI
compatibility in a patch in the series
and later fix it in the same series?(This is for more readability as
we can revert this patch in 21.11).



>
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> ---
>  app/test-eventdev/test_pipeline_common.c      |  16 +-
>  lib/librte_eventdev/eventdev_pmd.h            |  29 +++
>  .../rte_event_eth_rx_adapter.c                | 168 ++++++++++++------
>  .../rte_event_eth_rx_adapter.h                |  27 +++
>  lib/librte_eventdev/version.map               |   1 +
>  5 files changed, 184 insertions(+), 57 deletions(-)
>
> diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
> index 89f73be86..9aeefdd5f 100644
> --- a/app/test-eventdev/test_pipeline_common.c
> +++ b/app/test-eventdev/test_pipeline_common.c
> @@ -331,6 +331,7 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
>         uint16_t prod;
>         struct rte_mempool *vector_pool = NULL;
>         struct rte_event_eth_rx_adapter_queue_conf queue_conf;
> +       struct rte_event_eth_rx_adapter_event_vector_config vec_conf;
>
>         memset(&queue_conf, 0,
>                         sizeof(struct rte_event_eth_rx_adapter_queue_conf));
> @@ -360,12 +361,8 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
>                 }
>                 if (opt->ena_vector) {
>                         if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
> -                               queue_conf.vector_sz = opt->vector_size;
> -                               queue_conf.vector_timeout_ns =
> -                                       opt->vector_tmo_nsec;
>                                 queue_conf.rx_queue_flags |=
>                                 RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
> -                               queue_conf.vector_mp = vector_pool;
>                         } else {
>                                 evt_err("Rx adapter doesn't support event vector");
>                                 return -EINVAL;
> @@ -385,6 +382,17 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
>                         return ret;
>                 }
>
> +               if (opt->ena_vector) {
> +                       vec_conf.vector_sz = opt->vector_size;
> +                       vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
> +                       vec_conf.vector_mp = vector_pool;
> +                       if (rte_event_eth_rx_adapter_queue_event_vector_config(
> +                                   prod, prod, -1, &vec_conf) < 0) {
> +                               evt_err("Failed to configure event vectorization for Rx adapter");
> +                               return -EINVAL;
> +                       }
> +               }
> +
>                 if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
>                         uint32_t service_id = -1U;
>
> diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
> index 60bfaebc0..d79dfd612 100644
> --- a/lib/librte_eventdev/eventdev_pmd.h
> +++ b/lib/librte_eventdev/eventdev_pmd.h
> @@ -667,6 +667,32 @@ typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
>         const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
>         struct rte_event_eth_rx_adapter_vector_limits *limits);
>
> +struct rte_event_eth_rx_adapter_event_vector_config;
> +/**
> + * Enable event vector on an given Rx queue of a ethernet devices belonging to
> + * the Rx adapter.
> + *
> + * @param dev
> + *   Event device pointer
> + *
> + * @param eth_dev
> + *   Ethernet device pointer
> + *
> + * @param rx_queue_id
> + *   The Rx queue identifier
> + *
> + * @param config
> + *   Pointer to the event vector configuration structure.
> + *
> + * @return
> + *   - 0: Success.
> + *   - <0: Error code returned by the driver function.
> + */
> +typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
> +       const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
> +       int32_t rx_queue_id,
> +       const struct rte_event_eth_rx_adapter_event_vector_config *config);
> +
>  typedef uint32_t rte_event_pmd_selftest_seqn_t;
>  extern int rte_event_pmd_selftest_seqn_dynfield_offset;
>
> @@ -1092,6 +1118,9 @@ struct rte_eventdev_ops {
>         eventdev_eth_rx_adapter_vector_limits_get_t
>                 eth_rx_adapter_vector_limits_get;
>         /**< Get event vector limits for the Rx adapter */
> +       eventdev_eth_rx_adapter_event_vector_config_t
> +               eth_rx_adapter_event_vector_config;
> +       /**< Configure Rx adapter with event vector */
>
>         eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
>         /**< Get timer adapter capabilities */
> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> index a1990637f..c71990078 100644
> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> @@ -1882,25 +1882,6 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
>         } else
>                 qi_ev->flow_id = 0;
>
> -       if (conf->rx_queue_flags &
> -           RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
> -               queue_info->ena_vector = 1;
> -               qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
> -               rxa_set_vector_data(queue_info, conf->vector_sz,
> -                                   conf->vector_timeout_ns, conf->vector_mp,
> -                                   rx_queue_id, dev_info->dev->data->port_id);
> -               rx_adapter->ena_vector = 1;
> -               rx_adapter->vector_tmo_ticks =
> -                       rx_adapter->vector_tmo_ticks
> -                               ? RTE_MIN(queue_info->vector_data
> -                                                 .vector_timeout_ticks,
> -                                         rx_adapter->vector_tmo_ticks)
> -                               : queue_info->vector_data.vector_timeout_ticks;
> -               rx_adapter->vector_tmo_ticks <<= 1;
> -               TAILQ_INIT(&rx_adapter->vector_list);
> -               rx_adapter->prev_expiry_ts = 0;
> -       }
> -
>         rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
>         if (rxa_polled_queue(dev_info, rx_queue_id)) {
>                 rx_adapter->num_rx_polled += !pollq;
> @@ -1926,6 +1907,44 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
>         }
>  }
>
> +static void
> +rxa_sw_event_vector_configure(
> +       struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
> +       int rx_queue_id,
> +       const struct rte_event_eth_rx_adapter_event_vector_config *config)
> +{
> +       struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
> +       struct eth_rx_queue_info *queue_info;
> +       struct rte_event *qi_ev;
> +
> +       if (rx_queue_id == -1) {
> +               uint16_t nb_rx_queues;
> +               uint16_t i;
> +
> +               nb_rx_queues = dev_info->dev->data->nb_rx_queues;
> +               for (i = 0; i < nb_rx_queues; i++)
> +                       rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
> +                                                     config);
> +               return;
> +       }
> +
> +       queue_info = &dev_info->rx_queue[rx_queue_id];
> +       qi_ev = (struct rte_event *)&queue_info->event;
> +       queue_info->ena_vector = 1;
> +       qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
> +       rxa_set_vector_data(queue_info, config->vector_sz,
> +                           config->vector_timeout_ns, config->vector_mp,
> +                           rx_queue_id, dev_info->dev->data->port_id);
> +       rx_adapter->ena_vector = 1;
> +       rx_adapter->vector_tmo_ticks =
> +               rx_adapter->vector_tmo_ticks ?
> +                             RTE_MIN(config->vector_timeout_ns << 1,
> +                                     rx_adapter->vector_tmo_ticks) :
> +                             config->vector_timeout_ns << 1;
> +       rx_adapter->prev_expiry_ts = 0;
> +       TAILQ_INIT(&rx_adapter->vector_list);
> +}
> +
>  static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
>                 uint16_t eth_dev_id,
>                 int rx_queue_id,
> @@ -2239,7 +2258,6 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
>         struct rte_event_eth_rx_adapter *rx_adapter;
>         struct rte_eventdev *dev;
>         struct eth_device_info *dev_info;
> -       struct rte_event_eth_rx_adapter_vector_limits limits;
>
>         RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
>         RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
> @@ -2276,39 +2294,6 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
>                 return -EINVAL;
>         }
>
> -       if (queue_conf->rx_queue_flags &
> -           RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
> -               ret = rte_event_eth_rx_adapter_vector_limits_get(
> -                       rx_adapter->eventdev_id, eth_dev_id, &limits);
> -               if (ret < 0) {
> -                       RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
> -                                        " eth port: %" PRIu16
> -                                        " adapter id: %" PRIu8,
> -                                        eth_dev_id, id);
> -                       return -EINVAL;
> -               }
> -               if (queue_conf->vector_sz < limits.min_sz ||
> -                   queue_conf->vector_sz > limits.max_sz ||
> -                   queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
> -                   queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
> -                   queue_conf->vector_mp == NULL) {
> -                       RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> -                                        " eth port: %" PRIu16
> -                                        " adapter id: %" PRIu8,
> -                                        eth_dev_id, id);
> -                       return -EINVAL;
> -               }
> -               if (queue_conf->vector_mp->elt_size <
> -                   (sizeof(struct rte_event_vector) +
> -                    (sizeof(uintptr_t) * queue_conf->vector_sz))) {
> -                       RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> -                                        " eth port: %" PRIu16
> -                                        " adapter id: %" PRIu8,
> -                                        eth_dev_id, id);
> -                       return -EINVAL;
> -               }
> -       }
> -
>         if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
>                 (rx_queue_id != -1)) {
>                 RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
> @@ -2502,6 +2487,83 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
>         return ret;
>  }
>
> +int
> +rte_event_eth_rx_adapter_queue_event_vector_config(
> +       uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
> +       struct rte_event_eth_rx_adapter_event_vector_config *config)
> +{
> +       struct rte_event_eth_rx_adapter_vector_limits limits;
> +       struct rte_event_eth_rx_adapter *rx_adapter;
> +       struct rte_eventdev *dev;
> +       uint32_t cap;
> +       int ret;
> +
> +       RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
> +       RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
> +
> +       rx_adapter = rxa_id_to_adapter(id);
> +       if ((rx_adapter == NULL) || (config == NULL))
> +               return -EINVAL;
> +
> +       dev = &rte_eventdevs[rx_adapter->eventdev_id];
> +       ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
> +                                               eth_dev_id, &cap);
> +       if (ret) {
> +               RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
> +                                "eth port %" PRIu16,
> +                                id, eth_dev_id);
> +               return ret;
> +       }
> +
> +       if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
> +               RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
> +                                " eth port: %" PRIu16 " adapter id: %" PRIu8,
> +                                eth_dev_id, id);
> +               return -EINVAL;
> +       }
> +
> +       ret = rte_event_eth_rx_adapter_vector_limits_get(
> +               rx_adapter->eventdev_id, eth_dev_id, &limits);
> +       if (ret) {
> +               RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
> +                                "eth port %" PRIu16,
> +                                rx_adapter->eventdev_id, eth_dev_id);
> +               return ret;
> +       }
> +
> +       if (config->vector_sz < limits.min_sz ||
> +           config->vector_sz > limits.max_sz ||
> +           config->vector_timeout_ns < limits.min_timeout_ns ||
> +           config->vector_timeout_ns > limits.max_timeout_ns ||
> +           config->vector_mp == NULL) {
> +               RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> +                                " eth port: %" PRIu16 " adapter id: %" PRIu8,
> +                                eth_dev_id, id);
> +               return -EINVAL;
> +       }
> +       if (config->vector_mp->elt_size <
> +           (sizeof(struct rte_event_vector) +
> +            (sizeof(uintptr_t) * config->vector_sz))) {
> +               RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> +                                " eth port: %" PRIu16 " adapter id: %" PRIu8,
> +                                eth_dev_id, id);
> +               return -EINVAL;
> +       }
> +
> +       if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
> +               RTE_FUNC_PTR_OR_ERR_RET(
> +                       *dev->dev_ops->eth_rx_adapter_event_vector_config,
> +                       -ENOTSUP);
> +               ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
> +                       dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
> +       } else {
> +               rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
> +                                             rx_queue_id, config);
> +       }
> +
> +       return ret;
> +}
> +
>  int
>  rte_event_eth_rx_adapter_vector_limits_get(
>         uint8_t dev_id, uint16_t eth_port_id,
> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> index 4bdb38f08..ceef6d565 100644
> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> @@ -171,6 +171,9 @@ struct rte_event_eth_rx_adapter_queue_conf {
>          * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
>          * enqueued event.
>          */
> +};
> +
> +struct rte_event_eth_rx_adapter_event_vector_config {
>         uint16_t vector_sz;
>         /**<
>          * Indicates the maximum number for mbufs to combine and form a vector.
> @@ -418,6 +421,30 @@ int rte_event_eth_rx_adapter_queue_add(uint8_t id,
>  int rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
>                                        int32_t rx_queue_id);
>
> +/**
> + * Configure event vectorization for a given ethernet device queue, that has
> + * been added to a event eth Rx adapter.
> + *
> + * @param id
> + *  The identifier of the ethernet Rx event adapter.
> + *
> + * @param eth_dev_id
> + *  The identifier of the ethernet device.
> + *
> + * @param rx_queue_id
> + *  Ethernet device receive queue index.
> + *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
> + *  are configured with event vectorization.
> + *
> + * @return
> + *  - 0: Success, Receive queue configured correctly.
> + *  - <0: Error code on failure.
> + */
> +__rte_experimental
> +int rte_event_eth_rx_adapter_queue_event_vector_config(
> +       uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
> +       struct rte_event_eth_rx_adapter_event_vector_config *config);
> +
>  /**
>   * Start ethernet Rx event adapter
>   *
> diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
> index 34c1c830e..902df0ae3 100644
> --- a/lib/librte_eventdev/version.map
> +++ b/lib/librte_eventdev/version.map
> @@ -142,6 +142,7 @@ EXPERIMENTAL {
>         #added in 21.05
>         rte_event_vector_pool_create;
>         rte_event_eth_rx_adapter_vector_limits_get;
> +       rte_event_eth_rx_adapter_queue_event_vector_config;
>  };
>
>  INTERNAL {
> --
> 2.17.1
>

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH 7/7] eventdev: fix ABI breakage due to event vector
  2021-03-08 18:44   ` Jerin Jacob
@ 2021-03-12 14:28     ` David Marchand
  2021-03-16  5:54       ` [dpdk-dev] [EXT] " Pavan Nikhilesh Bhagavatula
  2021-03-15 10:01     ` [dpdk-dev] " Kinsella, Ray
  1 sibling, 1 reply; 153+ messages in thread
From: David Marchand @ 2021-03-12 14:28 UTC (permalink / raw)
  To: Jerin Jacob
  Cc: Pavan Nikhilesh, Jerin Jacob, Jayatheerthan, Jay,
	Erik Gabriel Carrillo, Gujjar, Abhinandan S, McDaniel, Timothy,
	Hemant Agrawal, Van Haaren, Harry, Mattias Rönnblom,
	Liang Ma, Ray Kinsella, Neil Horman, dpdk-dev, Thomas Monjalon

On Mon, Mar 8, 2021 at 7:44 PM Jerin Jacob <jerinjacobk@gmail.com> wrote:
> Summary:
> 1)  Ideal way of adding this feature is to add elements in the
> existing structure as mentioned
> in  ("eventdev: introduce event vector Rx capability") in this series.
> 2) Since this breaking ABI, Introducing a new structure to fix this. I
> think, we can remove this
> limitation in 21.11 as that time we can change ABI as required.
>
> So, Is this patch needs to be squashed to  ("eventdev: introduce event
> vector Rx capability") to avoid
> ABI compatibility between patches? Or Is it OK to break the ABI
> compatibility in a patch in the series
> and later fix it in the same series?(This is for more readability as
> we can revert this patch in 21.11).

What matters is not to break compilation between patches, so that
bisecting is always possible.
For ABI checks... I don't see the need to enforce such a requirement.


Yet, it is more straightforward to not break the 20.11 ABI at all.
You can announce the intended ABI change in the release notes /
deprecation notices, not sure I saw it in this series.

If you still want to share the final state intended for v21.11, you
can send the patch at the end of the series with something in the
title like "for v21.11" and mark it deferred in patchwork.


-- 
David Marchand


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH 7/7] eventdev: fix ABI breakage due to event vector
  2021-03-08 18:44   ` Jerin Jacob
  2021-03-12 14:28     ` David Marchand
@ 2021-03-15 10:01     ` Kinsella, Ray
  1 sibling, 0 replies; 153+ messages in thread
From: Kinsella, Ray @ 2021-03-15 10:01 UTC (permalink / raw)
  To: Jerin Jacob, Pavan Nikhilesh
  Cc: Jerin Jacob, Jayatheerthan, Jay, Erik Gabriel Carrillo, Gujjar,
	Abhinandan S, McDaniel, Timothy, Hemant Agrawal, Van Haaren,
	Harry, Mattias Rönnblom, Liang Ma, Neil Horman, dpdk-dev,
	Thomas Monjalon, David Marchand



On 08/03/2021 18:44, Jerin Jacob wrote:
> On Sun, Feb 21, 2021 at 3:41 AM <pbhagavatula@marvell.com> wrote:
>>
>> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>>
>> Fix ABI breakage due to event vector configuration by moving
>> the vector configuration into a new structure and having a separate
>> function for enabling the vector config on a given ethernet device and
>> queue pair.
>> This vector config and function can be merged to queue config in
>> v21.11.
>>
>> Fixes: 44c81670cf0a ("eventdev: introduce event vector Rx capability")
> 
> Hi @Ray Kinsella @Neil Horman @Thomas Monjalon @David Marchand
> 
> Is the ABI breakage contract between release to release. Right? i.e it
> is not between each patch. Right?
> 
> Summary:
> 1)  Ideal way of adding this feature is to add elements in the
> existing structure as mentioned
> in  ("eventdev: introduce event vector Rx capability") in this series.
> 2) Since this breaking ABI, Introducing a new structure to fix this. I
> think, we can remove this
> limitation in 21.11 as that time we can change ABI as required.
> 
> So, Is this patch needs to be squashed to  ("eventdev: introduce event
> vector Rx capability") to avoid
> ABI compatibility between patches? Or Is it OK to break the ABI
> compatibility in a patch in the series
> and later fix it in the same series?(This is for more readability as
> we can revert this patch in 21.11).

You are essentially writing it as you want it to appear in 21.11, 
you then add one patch at the end to fix ABI compability until then.
You then only have one patch to revert in the 21.11 cycle. 

Agree with David, I like the approach. 

+1 from me. 

> 
> 
> 
>>
>> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
>> ---
>>  app/test-eventdev/test_pipeline_common.c      |  16 +-
>>  lib/librte_eventdev/eventdev_pmd.h            |  29 +++
>>  .../rte_event_eth_rx_adapter.c                | 168 ++++++++++++------
>>  .../rte_event_eth_rx_adapter.h                |  27 +++
>>  lib/librte_eventdev/version.map               |   1 +
>>  5 files changed, 184 insertions(+), 57 deletions(-)
>>
>> diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
>> index 89f73be86..9aeefdd5f 100644
>> --- a/app/test-eventdev/test_pipeline_common.c
>> +++ b/app/test-eventdev/test_pipeline_common.c
>> @@ -331,6 +331,7 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
>>         uint16_t prod;
>>         struct rte_mempool *vector_pool = NULL;
>>         struct rte_event_eth_rx_adapter_queue_conf queue_conf;
>> +       struct rte_event_eth_rx_adapter_event_vector_config vec_conf;
>>
>>         memset(&queue_conf, 0,
>>                         sizeof(struct rte_event_eth_rx_adapter_queue_conf));
>> @@ -360,12 +361,8 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
>>                 }
>>                 if (opt->ena_vector) {
>>                         if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
>> -                               queue_conf.vector_sz = opt->vector_size;
>> -                               queue_conf.vector_timeout_ns =
>> -                                       opt->vector_tmo_nsec;
>>                                 queue_conf.rx_queue_flags |=
>>                                 RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
>> -                               queue_conf.vector_mp = vector_pool;
>>                         } else {
>>                                 evt_err("Rx adapter doesn't support event vector");
>>                                 return -EINVAL;
>> @@ -385,6 +382,17 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
>>                         return ret;
>>                 }
>>
>> +               if (opt->ena_vector) {
>> +                       vec_conf.vector_sz = opt->vector_size;
>> +                       vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
>> +                       vec_conf.vector_mp = vector_pool;
>> +                       if (rte_event_eth_rx_adapter_queue_event_vector_config(
>> +                                   prod, prod, -1, &vec_conf) < 0) {
>> +                               evt_err("Failed to configure event vectorization for Rx adapter");
>> +                               return -EINVAL;
>> +                       }
>> +               }
>> +
>>                 if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
>>                         uint32_t service_id = -1U;
>>
>> diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
>> index 60bfaebc0..d79dfd612 100644
>> --- a/lib/librte_eventdev/eventdev_pmd.h
>> +++ b/lib/librte_eventdev/eventdev_pmd.h
>> @@ -667,6 +667,32 @@ typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
>>         const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
>>         struct rte_event_eth_rx_adapter_vector_limits *limits);
>>
>> +struct rte_event_eth_rx_adapter_event_vector_config;
>> +/**
>> + * Enable event vector on an given Rx queue of a ethernet devices belonging to
>> + * the Rx adapter.
>> + *
>> + * @param dev
>> + *   Event device pointer
>> + *
>> + * @param eth_dev
>> + *   Ethernet device pointer
>> + *
>> + * @param rx_queue_id
>> + *   The Rx queue identifier
>> + *
>> + * @param config
>> + *   Pointer to the event vector configuration structure.
>> + *
>> + * @return
>> + *   - 0: Success.
>> + *   - <0: Error code returned by the driver function.
>> + */
>> +typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
>> +       const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
>> +       int32_t rx_queue_id,
>> +       const struct rte_event_eth_rx_adapter_event_vector_config *config);
>> +
>>  typedef uint32_t rte_event_pmd_selftest_seqn_t;
>>  extern int rte_event_pmd_selftest_seqn_dynfield_offset;
>>
>> @@ -1092,6 +1118,9 @@ struct rte_eventdev_ops {
>>         eventdev_eth_rx_adapter_vector_limits_get_t
>>                 eth_rx_adapter_vector_limits_get;
>>         /**< Get event vector limits for the Rx adapter */
>> +       eventdev_eth_rx_adapter_event_vector_config_t
>> +               eth_rx_adapter_event_vector_config;
>> +       /**< Configure Rx adapter with event vector */
>>
>>         eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
>>         /**< Get timer adapter capabilities */
>> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> index a1990637f..c71990078 100644
>> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> @@ -1882,25 +1882,6 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
>>         } else
>>                 qi_ev->flow_id = 0;
>>
>> -       if (conf->rx_queue_flags &
>> -           RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
>> -               queue_info->ena_vector = 1;
>> -               qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
>> -               rxa_set_vector_data(queue_info, conf->vector_sz,
>> -                                   conf->vector_timeout_ns, conf->vector_mp,
>> -                                   rx_queue_id, dev_info->dev->data->port_id);
>> -               rx_adapter->ena_vector = 1;
>> -               rx_adapter->vector_tmo_ticks =
>> -                       rx_adapter->vector_tmo_ticks
>> -                               ? RTE_MIN(queue_info->vector_data
>> -                                                 .vector_timeout_ticks,
>> -                                         rx_adapter->vector_tmo_ticks)
>> -                               : queue_info->vector_data.vector_timeout_ticks;
>> -               rx_adapter->vector_tmo_ticks <<= 1;
>> -               TAILQ_INIT(&rx_adapter->vector_list);
>> -               rx_adapter->prev_expiry_ts = 0;
>> -       }
>> -
>>         rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
>>         if (rxa_polled_queue(dev_info, rx_queue_id)) {
>>                 rx_adapter->num_rx_polled += !pollq;
>> @@ -1926,6 +1907,44 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
>>         }
>>  }
>>
>> +static void
>> +rxa_sw_event_vector_configure(
>> +       struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
>> +       int rx_queue_id,
>> +       const struct rte_event_eth_rx_adapter_event_vector_config *config)
>> +{
>> +       struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
>> +       struct eth_rx_queue_info *queue_info;
>> +       struct rte_event *qi_ev;
>> +
>> +       if (rx_queue_id == -1) {
>> +               uint16_t nb_rx_queues;
>> +               uint16_t i;
>> +
>> +               nb_rx_queues = dev_info->dev->data->nb_rx_queues;
>> +               for (i = 0; i < nb_rx_queues; i++)
>> +                       rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
>> +                                                     config);
>> +               return;
>> +       }
>> +
>> +       queue_info = &dev_info->rx_queue[rx_queue_id];
>> +       qi_ev = (struct rte_event *)&queue_info->event;
>> +       queue_info->ena_vector = 1;
>> +       qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
>> +       rxa_set_vector_data(queue_info, config->vector_sz,
>> +                           config->vector_timeout_ns, config->vector_mp,
>> +                           rx_queue_id, dev_info->dev->data->port_id);
>> +       rx_adapter->ena_vector = 1;
>> +       rx_adapter->vector_tmo_ticks =
>> +               rx_adapter->vector_tmo_ticks ?
>> +                             RTE_MIN(config->vector_timeout_ns << 1,
>> +                                     rx_adapter->vector_tmo_ticks) :
>> +                             config->vector_timeout_ns << 1;
>> +       rx_adapter->prev_expiry_ts = 0;
>> +       TAILQ_INIT(&rx_adapter->vector_list);
>> +}
>> +
>>  static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
>>                 uint16_t eth_dev_id,
>>                 int rx_queue_id,
>> @@ -2239,7 +2258,6 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
>>         struct rte_event_eth_rx_adapter *rx_adapter;
>>         struct rte_eventdev *dev;
>>         struct eth_device_info *dev_info;
>> -       struct rte_event_eth_rx_adapter_vector_limits limits;
>>
>>         RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
>>         RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
>> @@ -2276,39 +2294,6 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
>>                 return -EINVAL;
>>         }
>>
>> -       if (queue_conf->rx_queue_flags &
>> -           RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
>> -               ret = rte_event_eth_rx_adapter_vector_limits_get(
>> -                       rx_adapter->eventdev_id, eth_dev_id, &limits);
>> -               if (ret < 0) {
>> -                       RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
>> -                                        " eth port: %" PRIu16
>> -                                        " adapter id: %" PRIu8,
>> -                                        eth_dev_id, id);
>> -                       return -EINVAL;
>> -               }
>> -               if (queue_conf->vector_sz < limits.min_sz ||
>> -                   queue_conf->vector_sz > limits.max_sz ||
>> -                   queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
>> -                   queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
>> -                   queue_conf->vector_mp == NULL) {
>> -                       RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
>> -                                        " eth port: %" PRIu16
>> -                                        " adapter id: %" PRIu8,
>> -                                        eth_dev_id, id);
>> -                       return -EINVAL;
>> -               }
>> -               if (queue_conf->vector_mp->elt_size <
>> -                   (sizeof(struct rte_event_vector) +
>> -                    (sizeof(uintptr_t) * queue_conf->vector_sz))) {
>> -                       RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
>> -                                        " eth port: %" PRIu16
>> -                                        " adapter id: %" PRIu8,
>> -                                        eth_dev_id, id);
>> -                       return -EINVAL;
>> -               }
>> -       }
>> -
>>         if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
>>                 (rx_queue_id != -1)) {
>>                 RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
>> @@ -2502,6 +2487,83 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
>>         return ret;
>>  }
>>
>> +int
>> +rte_event_eth_rx_adapter_queue_event_vector_config(
>> +       uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
>> +       struct rte_event_eth_rx_adapter_event_vector_config *config)
>> +{
>> +       struct rte_event_eth_rx_adapter_vector_limits limits;
>> +       struct rte_event_eth_rx_adapter *rx_adapter;
>> +       struct rte_eventdev *dev;
>> +       uint32_t cap;
>> +       int ret;
>> +
>> +       RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
>> +       RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
>> +
>> +       rx_adapter = rxa_id_to_adapter(id);
>> +       if ((rx_adapter == NULL) || (config == NULL))
>> +               return -EINVAL;
>> +
>> +       dev = &rte_eventdevs[rx_adapter->eventdev_id];
>> +       ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
>> +                                               eth_dev_id, &cap);
>> +       if (ret) {
>> +               RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
>> +                                "eth port %" PRIu16,
>> +                                id, eth_dev_id);
>> +               return ret;
>> +       }
>> +
>> +       if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
>> +               RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
>> +                                " eth port: %" PRIu16 " adapter id: %" PRIu8,
>> +                                eth_dev_id, id);
>> +               return -EINVAL;
>> +       }
>> +
>> +       ret = rte_event_eth_rx_adapter_vector_limits_get(
>> +               rx_adapter->eventdev_id, eth_dev_id, &limits);
>> +       if (ret) {
>> +               RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
>> +                                "eth port %" PRIu16,
>> +                                rx_adapter->eventdev_id, eth_dev_id);
>> +               return ret;
>> +       }
>> +
>> +       if (config->vector_sz < limits.min_sz ||
>> +           config->vector_sz > limits.max_sz ||
>> +           config->vector_timeout_ns < limits.min_timeout_ns ||
>> +           config->vector_timeout_ns > limits.max_timeout_ns ||
>> +           config->vector_mp == NULL) {
>> +               RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
>> +                                " eth port: %" PRIu16 " adapter id: %" PRIu8,
>> +                                eth_dev_id, id);
>> +               return -EINVAL;
>> +       }
>> +       if (config->vector_mp->elt_size <
>> +           (sizeof(struct rte_event_vector) +
>> +            (sizeof(uintptr_t) * config->vector_sz))) {
>> +               RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
>> +                                " eth port: %" PRIu16 " adapter id: %" PRIu8,
>> +                                eth_dev_id, id);
>> +               return -EINVAL;
>> +       }
>> +
>> +       if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
>> +               RTE_FUNC_PTR_OR_ERR_RET(
>> +                       *dev->dev_ops->eth_rx_adapter_event_vector_config,
>> +                       -ENOTSUP);
>> +               ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
>> +                       dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
>> +       } else {
>> +               rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
>> +                                             rx_queue_id, config);
>> +       }
>> +
>> +       return ret;
>> +}
>> +
>>  int
>>  rte_event_eth_rx_adapter_vector_limits_get(
>>         uint8_t dev_id, uint16_t eth_port_id,
>> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
>> index 4bdb38f08..ceef6d565 100644
>> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
>> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
>> @@ -171,6 +171,9 @@ struct rte_event_eth_rx_adapter_queue_conf {
>>          * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
>>          * enqueued event.
>>          */
>> +};
>> +
>> +struct rte_event_eth_rx_adapter_event_vector_config {
>>         uint16_t vector_sz;
>>         /**<
>>          * Indicates the maximum number for mbufs to combine and form a vector.
>> @@ -418,6 +421,30 @@ int rte_event_eth_rx_adapter_queue_add(uint8_t id,
>>  int rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
>>                                        int32_t rx_queue_id);
>>
>> +/**
>> + * Configure event vectorization for a given ethernet device queue, that has
>> + * been added to a event eth Rx adapter.
>> + *
>> + * @param id
>> + *  The identifier of the ethernet Rx event adapter.
>> + *
>> + * @param eth_dev_id
>> + *  The identifier of the ethernet device.
>> + *
>> + * @param rx_queue_id
>> + *  Ethernet device receive queue index.
>> + *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
>> + *  are configured with event vectorization.
>> + *
>> + * @return
>> + *  - 0: Success, Receive queue configured correctly.
>> + *  - <0: Error code on failure.
>> + */
>> +__rte_experimental
>> +int rte_event_eth_rx_adapter_queue_event_vector_config(
>> +       uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
>> +       struct rte_event_eth_rx_adapter_event_vector_config *config);
>> +
>>  /**
>>   * Start ethernet Rx event adapter
>>   *
>> diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
>> index 34c1c830e..902df0ae3 100644
>> --- a/lib/librte_eventdev/version.map
>> +++ b/lib/librte_eventdev/version.map
>> @@ -142,6 +142,7 @@ EXPERIMENTAL {
>>         #added in 21.05
>>         rte_event_vector_pool_create;
>>         rte_event_eth_rx_adapter_vector_limits_get;
>> +       rte_event_eth_rx_adapter_queue_event_vector_config;
>>  };
>>
>>  INTERNAL {
>> --
>> 2.17.1
>>

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [EXT] Re: [PATCH 7/7] eventdev: fix ABI breakage due to event vector
  2021-03-12 14:28     ` David Marchand
@ 2021-03-16  5:54       ` Pavan Nikhilesh Bhagavatula
  0 siblings, 0 replies; 153+ messages in thread
From: Pavan Nikhilesh Bhagavatula @ 2021-03-16  5:54 UTC (permalink / raw)
  To: David Marchand, Jerin Jacob
  Cc: Jerin Jacob Kollanukkaran, Jayatheerthan, Jay,
	Erik Gabriel Carrillo, Gujjar, Abhinandan S, McDaniel, Timothy,
	Hemant Agrawal, Van Haaren, Harry, Mattias Rönnblom,
	Liang Ma, Ray Kinsella, Neil Horman, dpdk-dev, Thomas Monjalon

>On Mon, Mar 8, 2021 at 7:44 PM Jerin Jacob <jerinjacobk@gmail.com>
>wrote:
>> Summary:
>> 1)  Ideal way of adding this feature is to add elements in the
>> existing structure as mentioned
>> in  ("eventdev: introduce event vector Rx capability") in this series.
>> 2) Since this breaking ABI, Introducing a new structure to fix this. I
>> think, we can remove this
>> limitation in 21.11 as that time we can change ABI as required.
>>
>> So, Is this patch needs to be squashed to  ("eventdev: introduce event
>> vector Rx capability") to avoid
>> ABI compatibility between patches? Or Is it OK to break the ABI
>> compatibility in a patch in the series
>> and later fix it in the same series?(This is for more readability as
>> we can revert this patch in 21.11).
>
>What matters is not to break compilation between patches, so that
>bisecting is always possible.
>For ABI checks... I don't see the need to enforce such a requirement.
>
>
>Yet, it is more straightforward to not break the 20.11 ABI at all.
>You can announce the intended ABI change in the release notes /
>deprecation notices, not sure I saw it in this series.
>
>If you still want to share the final state intended for v21.11, you
>can send the patch at the end of the series with something in the
>title like "for v21.11" and mark it deferred in patchwork.
>

Ack, I will rework the series without breaking the ABI till the last patch.

Pavan.

>
>--
>David Marchand


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH 4/7] eventdev: add Rx adapter event vector support
  2021-03-08 17:27   ` Jerin Jacob
@ 2021-03-16 10:41     ` Jayatheerthan, Jay
  0 siblings, 0 replies; 153+ messages in thread
From: Jayatheerthan, Jay @ 2021-03-16 10:41 UTC (permalink / raw)
  To: Jerin Jacob, Pavan Nikhilesh, Rao, Nikhil
  Cc: Jerin Jacob, Carrillo, Erik G, Gujjar, Abhinandan S, McDaniel,
	Timothy, Hemant Agrawal, Van Haaren, Harry, mattias.ronnblom, Ma,
	Liang J, dpdk-dev

> -----Original Message-----
> From: Jerin Jacob <jerinjacobk@gmail.com>
> Sent: Monday, March 8, 2021 10:57 PM
> To: Pavan Nikhilesh <pbhagavatula@marvell.com>; Rao, Nikhil <nikhil.rao@intel.com>
> Cc: Jerin Jacob <jerinj@marvell.com>; Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Carrillo, Erik G <erik.g.carrillo@intel.com>;
> Gujjar, Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy <timothy.mcdaniel@intel.com>; Hemant Agrawal
> <hemant.agrawal@nxp.com>; Van Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
> <mattias.ronnblom@ericsson.com>; Ma, Liang J <liang.j.ma@intel.com>; dpdk-dev <dev@dpdk.org>
> Subject: Re: [dpdk-dev] [PATCH 4/7] eventdev: add Rx adapter event vector support
> 
> On Sun, Feb 21, 2021 at 3:40 AM <pbhagavatula@marvell.com> wrote:
> >
> > From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> >
> > Add event vector support for event eth Rx adapter, the implementation
> > creates vector flows based on port and queue identifier of the received
> > mbufs.
> >
> > Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> 
> Hi @Jayatheerthan, Jay
> 
> Could you review 4/7 and 5/7 SW adapter-related patches?
> Cc: @Nikhil Rao
> 
My apologies for delayed response. Planning to review and get back.
> 
> 
> > ---
> >  lib/librte_eventdev/eventdev_pmd.h            |  31 +-
> >  .../rte_event_eth_rx_adapter.c                | 305 +++++++++++++++++-
> >  lib/librte_eventdev/rte_eventdev.c            |   6 +-
> >  3 files changed, 324 insertions(+), 18 deletions(-)

^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v2 0/8] Introduce event vectorization
  2021-02-20 22:09 [dpdk-dev] [PATCH 0/7] Introduce event vectorization pbhagavatula
                   ` (7 preceding siblings ...)
  2021-03-08 16:41 ` [dpdk-dev] [PATCH 0/7] Introduce event vectorization Jerin Jacob
@ 2021-03-16 15:48 ` pbhagavatula
  2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 1/8] eventdev: introduce event vector capability pbhagavatula
                     ` (8 more replies)
  8 siblings, 9 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-16 15:48 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

In traditional event programming model, events are identified by a
flow-id and a uintptr_t. The flow-id uniquely identifies a given event
and determines the order of scheduling based on schedule type, the
uintptr_t holds a single object.

Event devices also support burst mode with configurable dequeue depth,
i.e. each dequeue call would return multiple events and each event
might be at a different stage of the pipeline.
Having a burst of events belonging to different stages in a dequeue
burst is not only difficult to vectorize but also increases the scheduler
overhead and application overhead of pipelining events further.
Using event vectors we see a performance gain of ~628% as shown in [1].

By introducing event vectorization, each event will be capable of holding
multiple uintptr_t of the same flow thereby allowing applications
to vectorize their pipeline and reduce the complexity of pipelining
events across multiple stages. This also reduces the complexity of handling
enqueue and dequeue on an event device.

Since event devices are transparent to the events they are scheduling
so the event producers such as eth_rx_adapter, crypto_adapter , etc..
are responsible for vectorizing the buffers of the same flow into a single
event.

The series also breaks ABI in the patch [8/8] which is targetted to the
v21.11 release.

The dpdk-test-eventdev application has been updated with options to test
multiple vector sizes and timeouts.

[1]
As for performance improvement, with a ARM Cortex-A72 equivalent processer,
software event device (--vdev=event_sw0), single worker core, single stage
and using one service core for Rx adapter, Tx adapter, Scheduling.

Without event vectorization:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
         --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
         --stlist=a --wlcores=20
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    4.728 mpps avg 4.728 mpps

With event vectorization:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
        --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
        --stlist=a --wlcores=20 --enable_vector --nb_eth_queues 1
        --vector_size 256
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    34.383 mpps avg 34.383 mpps

Having dedicated service cores for each Rx queues and tweaking the vector,
dequeue burst size would further improve performance.

API usage is shown below:

Configuration:

	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

	vector_pool = rte_event_vector_pool_create("vector_pool",
			nb_elem, 0, vector_size, socket_id);

	rte_event_eth_rx_adapter_create(id, event_id, &adptr_conf);
	rte_event_eth_rx_adapter_queue_add(id, eth_id, -1, &queue_conf);
	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
		vec_conf.vector_sz = vector_size;
		vec_conf.vector_timeout_ns = vector_tmo_nsec;
		vec_conf.vector_mp = vector_pool;
		rte_event_eth_rx_adapter_queue_event_vector_config(id,
				eth_id, -1, &vec_conf);
	}

Fastpath:

	num = rte_event_dequeue_burst(event_id, port_id, &ev, 1, 0);
	if (!num)
		continue;

	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
		switch (ev.event_type) {
		case RTE_EVENT_TYPE_ETHDEV_VECTOR:
		case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
			struct rte_mbuf **mbufs;

			mbufs = ev.vector_ev->mbufs;
			for (i = 0; i < ev.vector_ev->nb_elem; i++)
				//Process mbufs.
			break;
		case ...
		}
	}
	...

v2 Changes:
- Multiple gramatical and style fixes.(Jerin)
- Add parameter to define vector size in power of 2. (Jerin)
- Redo patch series w/o breaking ABI till the last patch.(David)
- Add deprication notice to announce ABI break in 21.11.(David)
- Add vector limits validation to app/test-eventdev.

Pavan Nikhilesh (8):
  eventdev: introduce event vector capability
  eventdev: introduce event vector Rx capability
  eventdev: introduce event vector Tx capability
  eventdev: add Rx adapter event vector support
  eventdev: add Tx adapter event vector support
  app/eventdev: add event vector mode in pipeline test
  doc: announce event Rx adapter config changes
  eventdev: simplify Rx adapter event vector config

 app/test-eventdev/evt_common.h                |   4 +
 app/test-eventdev/evt_options.c               |  52 +++
 app/test-eventdev/evt_options.h               |   4 +
 app/test-eventdev/test_pipeline_atq.c         | 310 ++++++++++++-
 app/test-eventdev/test_pipeline_common.c      |  69 ++-
 app/test-eventdev/test_pipeline_common.h      |  18 +
 app/test-eventdev/test_pipeline_queue.c       | 320 ++++++++++++-
 .../prog_guide/event_ethernet_rx_adapter.rst  |  38 ++
 .../prog_guide/event_ethernet_tx_adapter.rst  |  12 +
 doc/guides/prog_guide/eventdev.rst            |  36 +-
 doc/guides/rel_notes/deprecation.rst          |   9 +
 doc/guides/tools/testeventdev.rst             |  28 ++
 lib/librte_eventdev/eventdev_pmd.h            |  31 +-
 .../rte_event_eth_rx_adapter.c                | 305 ++++++++++++-
 .../rte_event_eth_rx_adapter.h                |  68 +++
 .../rte_event_eth_tx_adapter.c                |  66 ++-
 lib/librte_eventdev/rte_eventdev.c            |  11 +-
 lib/librte_eventdev/rte_eventdev.h            | 428 +++++++++++-------
 lib/librte_eventdev/version.map               |   4 +
 19 files changed, 1568 insertions(+), 245 deletions(-)

--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v2 1/8] eventdev: introduce event vector capability
  2021-03-16 15:48 ` [dpdk-dev] [PATCH v2 0/8] " pbhagavatula
@ 2021-03-16 15:48   ` pbhagavatula
  2021-03-16 17:48     ` Jerin Jacob
  2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 2/8] eventdev: introduce event vector Rx capability pbhagavatula
                     ` (7 subsequent siblings)
  8 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-03-16 15:48 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce rte_event_vector datastructure which is capable of holding
multiple uintptr_t of the same flow thereby allowing applications
to vectorize their pipeline and reducing the complexity of pipelining
the events across multiple stages.
This approach also reduces the scheduling overhead on a event device.

Add a event vector mempool create handler to create mempools based on
the best mempool ops available on a given platform.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 doc/guides/prog_guide/eventdev.rst |  36 ++-
 lib/librte_eventdev/rte_eventdev.h | 396 +++++++++++++++++------------
 lib/librte_eventdev/version.map    |   3 +
 3 files changed, 273 insertions(+), 162 deletions(-)

diff --git a/doc/guides/prog_guide/eventdev.rst b/doc/guides/prog_guide/eventdev.rst
index ccde086f6..fda9c3743 100644
--- a/doc/guides/prog_guide/eventdev.rst
+++ b/doc/guides/prog_guide/eventdev.rst
@@ -63,13 +63,45 @@ the actual event being scheduled is. The payload is a union of the following:
 * ``uint64_t u64``
 * ``void *event_ptr``
 * ``struct rte_mbuf *mbuf``
+* ``struct rte_event_vector *vec``
 
-These three items in a union occupy the same 64 bits at the end of the rte_event
+These four items in a union occupy the same 64 bits at the end of the rte_event
 structure. The application can utilize the 64 bits directly by accessing the
-u64 variable, while the event_ptr and mbuf are provided as convenience
+u64 variable, while the event_ptr, mbuf, vec are provided as a convenience
 variables.  For example the mbuf pointer in the union can used to schedule a
 DPDK packet.
 
+Event Vector
+~~~~~~~~~~~~
+
+The rte_event_vector struct contains a vector of elements defined by the event
+type specified in the ``rte_event``. The event_vector structure contains the
+following data:
+
+* ``nb_elem`` - The number of elements held within the vector.
+
+Similar to ``rte_event`` the payload of event vector is also a union, allowing
+flexibility in what the actual vector is.
+
+* ``struct rte_mbuf *mbufs[0]`` - An array of mbufs.
+* ``void *ptrs[0]`` - An array of pointers.
+* ``uint64_t *u64s[0]`` - An array of uint64_t elements.
+
+The size of the event vector is related to the total number of elements it is
+configured to hold, this is achieved by making `rte_event_vector` a variable
+length structure.
+A helper function is provided to create a mempool that holds event vector, which
+takes name of the pool, total number of required ``rte_event_vector``,
+cache size, number of elements in each ``rte_event_vector`` and socket id.
+
+.. code-block:: c
+
+        rte_event_vector_pool_create("vector_pool", nb_event_vectors, cache_sz,
+                                     nb_elements_per_vector, socket_id);
+
+The function ``rte_event_vector_pool_create`` creates mempool with the best
+platform mempool ops.
+
 Queues
 ~~~~~~
 
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index ce1fc2ce0..dda8d0387 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -212,8 +212,10 @@ extern "C" {
 
 #include <rte_common.h>
 #include <rte_config.h>
-#include <rte_memory.h>
 #include <rte_errno.h>
+#include <rte_mbuf_pool_ops.h>
+#include <rte_memory.h>
+#include <rte_mempool.h>
 
 #include "rte_eventdev_trace_fp.h"
 
@@ -221,20 +223,20 @@ struct rte_mbuf; /* we just use mbuf pointers; no need to include rte_mbuf.h */
 struct rte_event;
 
 /* Event device capability bitmap flags */
-#define RTE_EVENT_DEV_CAP_QUEUE_QOS           (1ULL << 0)
+#define RTE_EVENT_DEV_CAP_QUEUE_QOS (1ULL << 0)
 /**< Event scheduling prioritization is based on the priority associated with
  *  each event queue.
  *
  *  @see rte_event_queue_setup()
  */
-#define RTE_EVENT_DEV_CAP_EVENT_QOS           (1ULL << 1)
+#define RTE_EVENT_DEV_CAP_EVENT_QOS (1ULL << 1)
 /**< Event scheduling prioritization is based on the priority associated with
  *  each event. Priority of each event is supplied in *rte_event* structure
  *  on each enqueue operation.
  *
  *  @see rte_event_enqueue_burst()
  */
-#define RTE_EVENT_DEV_CAP_DISTRIBUTED_SCHED   (1ULL << 2)
+#define RTE_EVENT_DEV_CAP_DISTRIBUTED_SCHED (1ULL << 2)
 /**< Event device operates in distributed scheduling mode.
  * In distributed scheduling mode, event scheduling happens in HW or
  * rte_event_dequeue_burst() or the combination of these two.
@@ -243,14 +245,14 @@ struct rte_event;
  *
  * @see rte_event_dequeue_burst()
  */
-#define RTE_EVENT_DEV_CAP_QUEUE_ALL_TYPES     (1ULL << 3)
+#define RTE_EVENT_DEV_CAP_QUEUE_ALL_TYPES (1ULL << 3)
 /**< Event device is capable of enqueuing events of any type to any queue.
  * If this capability is not set, the queue only supports events of the
  *  *RTE_SCHED_TYPE_* type that it was created with.
  *
  * @see RTE_SCHED_TYPE_* values
  */
-#define RTE_EVENT_DEV_CAP_BURST_MODE          (1ULL << 4)
+#define RTE_EVENT_DEV_CAP_BURST_MODE (1ULL << 4)
 /**< Event device is capable of operating in burst mode for enqueue(forward,
  * release) and dequeue operation. If this capability is not set, application
  * still uses the rte_event_dequeue_burst() and rte_event_enqueue_burst() but
@@ -258,7 +260,7 @@ struct rte_event;
  *
  * @see rte_event_dequeue_burst() rte_event_enqueue_burst()
  */
-#define RTE_EVENT_DEV_CAP_IMPLICIT_RELEASE_DISABLE    (1ULL << 5)
+#define RTE_EVENT_DEV_CAP_IMPLICIT_RELEASE_DISABLE (1ULL << 5)
 /**< Event device ports support disabling the implicit release feature, in
  * which the port will release all unreleased events in its dequeue operation.
  * If this capability is set and the port is configured with implicit release
@@ -269,7 +271,7 @@ struct rte_event;
  * @see rte_event_dequeue_burst() rte_event_enqueue_burst()
  */
 
-#define RTE_EVENT_DEV_CAP_NONSEQ_MODE         (1ULL << 6)
+#define RTE_EVENT_DEV_CAP_NONSEQ_MODE (1ULL << 6)
 /**< Event device is capable of operating in none sequential mode. The path
  * of the event is not necessary to be sequential. Application can change
  * the path of event at runtime. If the flag is not set, then event each event
@@ -279,7 +281,7 @@ struct rte_event;
  * qid which is not the next in the sequence.
  */
 
-#define RTE_EVENT_DEV_CAP_RUNTIME_PORT_LINK   (1ULL << 7)
+#define RTE_EVENT_DEV_CAP_RUNTIME_PORT_LINK (1ULL << 7)
 /**< Event device is capable of configuring the queue/port link at runtime.
  * If the flag is not set, the eventdev queue/port link is only can be
  * configured during  initialization.
@@ -298,17 +300,17 @@ struct rte_event;
  */
 
 /* Event device priority levels */
-#define RTE_EVENT_DEV_PRIORITY_HIGHEST   0
+#define RTE_EVENT_DEV_PRIORITY_HIGHEST 0
 /**< Highest priority expressed across eventdev subsystem
  * @see rte_event_queue_setup(), rte_event_enqueue_burst()
  * @see rte_event_port_link()
  */
-#define RTE_EVENT_DEV_PRIORITY_NORMAL    128
+#define RTE_EVENT_DEV_PRIORITY_NORMAL 128
 /**< Normal priority expressed across eventdev subsystem
  * @see rte_event_queue_setup(), rte_event_enqueue_burst()
  * @see rte_event_port_link()
  */
-#define RTE_EVENT_DEV_PRIORITY_LOWEST    255
+#define RTE_EVENT_DEV_PRIORITY_LOWEST 255
 /**< Lowest priority expressed across eventdev subsystem
  * @see rte_event_queue_setup(), rte_event_enqueue_burst()
  * @see rte_event_port_link()
@@ -321,8 +323,7 @@ struct rte_event;
  * @return
  *   The total number of usable event devices.
  */
-uint8_t
-rte_event_dev_count(void);
+uint8_t rte_event_dev_count(void);
 
 /**
  * Get the device identifier for the named event device.
@@ -334,8 +335,7 @@ rte_event_dev_count(void);
  *   Returns event device identifier on success.
  *   - <0: Failure to find named event device.
  */
-int
-rte_event_dev_get_dev_id(const char *name);
+int rte_event_dev_get_dev_id(const char *name);
 
 /**
  * Return the NUMA socket to which a device is connected.
@@ -347,15 +347,14 @@ rte_event_dev_get_dev_id(const char *name);
  *   a default of zero if the socket could not be determined.
  *   -(-EINVAL)  dev_id value is out of range.
  */
-int
-rte_event_dev_socket_id(uint8_t dev_id);
+int rte_event_dev_socket_id(uint8_t dev_id);
 
 /**
  * Event device information
  */
 struct rte_event_dev_info {
-	const char *driver_name;	/**< Event driver name */
-	struct rte_device *dev;	/**< Device information */
+	const char *driver_name; /**< Event driver name */
+	struct rte_device *dev;	 /**< Device information */
 	uint32_t min_dequeue_timeout_ns;
 	/**< Minimum supported global dequeue timeout(ns) by this device */
 	uint32_t max_dequeue_timeout_ns;
@@ -420,8 +419,7 @@ struct rte_event_dev_info {
  *   - <0: Error code returned by the driver info get function.
  *
  */
-int
-rte_event_dev_info_get(uint8_t dev_id, struct rte_event_dev_info *dev_info);
+int rte_event_dev_info_get(uint8_t dev_id, struct rte_event_dev_info *dev_info);
 
 /**
  * The count of ports.
@@ -448,10 +446,8 @@ rte_event_dev_info_get(uint8_t dev_id, struct rte_event_dev_info *dev_info);
  *   - 0: Successfully retrieved attribute value
  *   - -EINVAL: Invalid device or  *attr_id* provided, or *attr_value* is NULL
  */
-int
-rte_event_dev_attr_get(uint8_t dev_id, uint32_t attr_id,
-		       uint32_t *attr_value);
-
+int rte_event_dev_attr_get(uint8_t dev_id, uint32_t attr_id,
+			   uint32_t *attr_value);
 
 /* Event device configuration bitmap flags */
 #define RTE_EVENT_DEV_CFG_PER_DEQUEUE_TIMEOUT (1ULL << 0)
@@ -539,20 +535,19 @@ struct rte_event_dev_config {
  *   - 0: Success, device configured.
  *   - <0: Error code returned by the driver configuration function.
  */
-int
-rte_event_dev_configure(uint8_t dev_id,
-			const struct rte_event_dev_config *dev_conf);
+int rte_event_dev_configure(uint8_t dev_id,
+			    const struct rte_event_dev_config *dev_conf);
 
 /* Event queue specific APIs */
 
 /* Event queue configuration bitmap flags */
-#define RTE_EVENT_QUEUE_CFG_ALL_TYPES          (1ULL << 0)
+#define RTE_EVENT_QUEUE_CFG_ALL_TYPES (1ULL << 0)
 /**< Allow ATOMIC,ORDERED,PARALLEL schedule type enqueue
  *
  * @see RTE_SCHED_TYPE_ORDERED, RTE_SCHED_TYPE_ATOMIC, RTE_SCHED_TYPE_PARALLEL
  * @see rte_event_enqueue_burst()
  */
-#define RTE_EVENT_QUEUE_CFG_SINGLE_LINK        (1ULL << 1)
+#define RTE_EVENT_QUEUE_CFG_SINGLE_LINK (1ULL << 1)
 /**< This event queue links only to a single event port.
  *
  *  @see rte_event_port_setup(), rte_event_port_link()
@@ -621,9 +616,8 @@ struct rte_event_queue_conf {
  * @see rte_event_queue_setup()
  *
  */
-int
-rte_event_queue_default_conf_get(uint8_t dev_id, uint8_t queue_id,
-				 struct rte_event_queue_conf *queue_conf);
+int rte_event_queue_default_conf_get(uint8_t dev_id, uint8_t queue_id,
+				     struct rte_event_queue_conf *queue_conf);
 
 /**
  * Allocate and set up an event queue for an event device.
@@ -643,9 +637,8 @@ rte_event_queue_default_conf_get(uint8_t dev_id, uint8_t queue_id,
  *   - 0: Success, event queue correctly set up.
  *   - <0: event queue configuration failed
  */
-int
-rte_event_queue_setup(uint8_t dev_id, uint8_t queue_id,
-		      const struct rte_event_queue_conf *queue_conf);
+int rte_event_queue_setup(uint8_t dev_id, uint8_t queue_id,
+			  const struct rte_event_queue_conf *queue_conf);
 
 /**
  * The priority of the queue.
@@ -688,21 +681,20 @@ rte_event_queue_setup(uint8_t dev_id, uint8_t queue_id,
  *   RTE_EVENT_QUEUE_ATTR_SCHEDULE_TYPE and event_queue_cfg is set to
  *   RTE_EVENT_QUEUE_CFG_ALL_TYPES
  */
-int
-rte_event_queue_attr_get(uint8_t dev_id, uint8_t queue_id, uint32_t attr_id,
-			uint32_t *attr_value);
+int rte_event_queue_attr_get(uint8_t dev_id, uint8_t queue_id, uint32_t attr_id,
+			     uint32_t *attr_value);
 
 /* Event port specific APIs */
 
 /* Event port configuration bitmap flags */
-#define RTE_EVENT_PORT_CFG_DISABLE_IMPL_REL    (1ULL << 0)
+#define RTE_EVENT_PORT_CFG_DISABLE_IMPL_REL (1ULL << 0)
 /**< Configure the port not to release outstanding events in
  * rte_event_dev_dequeue_burst(). If set, all events received through
  * the port must be explicitly released with RTE_EVENT_OP_RELEASE or
  * RTE_EVENT_OP_FORWARD. Must be unset if the device is not
  * RTE_EVENT_DEV_CAP_IMPLICIT_RELEASE_DISABLE capable.
  */
-#define RTE_EVENT_PORT_CFG_SINGLE_LINK         (1ULL << 1)
+#define RTE_EVENT_PORT_CFG_SINGLE_LINK (1ULL << 1)
 /**< This event port links only to a single event queue.
  *
  *  @see rte_event_port_setup(), rte_event_port_link()
@@ -760,9 +752,8 @@ struct rte_event_port_conf {
  * @see rte_event_port_setup()
  *
  */
-int
-rte_event_port_default_conf_get(uint8_t dev_id, uint8_t port_id,
-				struct rte_event_port_conf *port_conf);
+int rte_event_port_default_conf_get(uint8_t dev_id, uint8_t port_id,
+				    struct rte_event_port_conf *port_conf);
 
 /**
  * Allocate and set up an event port for an event device.
@@ -784,9 +775,8 @@ rte_event_port_default_conf_get(uint8_t dev_id, uint8_t port_id,
  *   - (-EDQUOT) Quota exceeded(Application tried to link the queue configured
  *   with RTE_EVENT_QUEUE_CFG_SINGLE_LINK to more than one event ports)
  */
-int
-rte_event_port_setup(uint8_t dev_id, uint8_t port_id,
-		     const struct rte_event_port_conf *port_conf);
+int rte_event_port_setup(uint8_t dev_id, uint8_t port_id,
+			 const struct rte_event_port_conf *port_conf);
 
 /**
  * The queue depth of the port on the enqueue side
@@ -821,9 +811,8 @@ rte_event_port_setup(uint8_t dev_id, uint8_t port_id,
  *   - 0: Successfully returned value
  *   - (-EINVAL) Invalid device, port or attr_id, or attr_value was NULL
  */
-int
-rte_event_port_attr_get(uint8_t dev_id, uint8_t port_id, uint32_t attr_id,
-			uint32_t *attr_value);
+int rte_event_port_attr_get(uint8_t dev_id, uint8_t port_id, uint32_t attr_id,
+			    uint32_t *attr_value);
 
 /**
  * Start an event device.
@@ -841,8 +830,7 @@ rte_event_port_attr_get(uint8_t dev_id, uint8_t port_id, uint32_t attr_id,
  *   - -ESTALE : Not all ports of the device are configured
  *   - -ENOLINK: Not all queues are linked, which could lead to deadlock.
  */
-int
-rte_event_dev_start(uint8_t dev_id);
+int rte_event_dev_start(uint8_t dev_id);
 
 /**
  * Stop an event device.
@@ -862,11 +850,10 @@ rte_event_dev_start(uint8_t dev_id);
  *
  * @see rte_event_dev_stop_flush_callback_register()
  */
-void
-rte_event_dev_stop(uint8_t dev_id);
+void rte_event_dev_stop(uint8_t dev_id);
 
 typedef void (*eventdev_stop_flush_t)(uint8_t dev_id, struct rte_event event,
-		void *arg);
+				      void *arg);
 /**< Callback function called during rte_event_dev_stop(), invoked once per
  * flushed event.
  */
@@ -895,9 +882,9 @@ typedef void (*eventdev_stop_flush_t)(uint8_t dev_id, struct rte_event event,
  *
  * @see rte_event_dev_stop()
  */
-int
-rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
-		eventdev_stop_flush_t callback, void *userdata);
+int rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
+					       eventdev_stop_flush_t callback,
+					       void *userdata);
 
 /**
  * Close an event device. The device cannot be restarted!
@@ -910,11 +897,29 @@ rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
  *  - <0 on failure to close device
  *  - (-EAGAIN) if device is busy
  */
-int
-rte_event_dev_close(uint8_t dev_id);
+int rte_event_dev_close(uint8_t dev_id);
+
+/**
+ * Event vector structure.
+ */
+struct rte_event_vector {
+	uint64_t nb_elem : 16;
+	/**< Number of elements in this event vector. */
+	uint64_t rsvd : 48;
+	uint64_t impl_opaque;
+	union {
+		struct rte_mbuf *mbufs[0];
+		void *ptrs[0];
+		uint64_t *u64s[0];
+	} __rte_aligned(16);
+	/**< Start of the vector array union. Depending upon the event type the
+	 * vector array can be an array of mbufs or pointers or opaque u64
+	 * values.
+	 */
+};
 
 /* Scheduler type definitions */
-#define RTE_SCHED_TYPE_ORDERED          0
+#define RTE_SCHED_TYPE_ORDERED 0
 /**< Ordered scheduling
  *
  * Events from an ordered flow of an event queue can be scheduled to multiple
@@ -941,7 +946,7 @@ rte_event_dev_close(uint8_t dev_id);
  * @see rte_event_queue_setup(), rte_event_dequeue_burst(), RTE_EVENT_OP_RELEASE
  */
 
-#define RTE_SCHED_TYPE_ATOMIC           1
+#define RTE_SCHED_TYPE_ATOMIC 1
 /**< Atomic scheduling
  *
  * Events from an atomic flow of an event queue can be scheduled only to a
@@ -960,7 +965,7 @@ rte_event_dev_close(uint8_t dev_id);
  * @see rte_event_queue_setup(), rte_event_dequeue_burst(), RTE_EVENT_OP_RELEASE
  */
 
-#define RTE_SCHED_TYPE_PARALLEL         2
+#define RTE_SCHED_TYPE_PARALLEL 2
 /**< Parallel scheduling
  *
  * The scheduler performs priority scheduling, load balancing, etc. functions
@@ -974,27 +979,42 @@ rte_event_dev_close(uint8_t dev_id);
  */
 
 /* Event types to classify the event source */
-#define RTE_EVENT_TYPE_ETHDEV           0x0
+#define RTE_EVENT_TYPE_ETHDEV 0x0
 /**< The event generated from ethdev subsystem */
-#define RTE_EVENT_TYPE_CRYPTODEV        0x1
+#define RTE_EVENT_TYPE_CRYPTODEV 0x1
 /**< The event generated from crypodev subsystem */
-#define RTE_EVENT_TYPE_TIMER		0x2
+#define RTE_EVENT_TYPE_TIMER 0x2
 /**< The event generated from event timer adapter */
-#define RTE_EVENT_TYPE_CPU              0x3
+#define RTE_EVENT_TYPE_CPU 0x3
 /**< The event generated from cpu for pipelining.
  * Application may use *sub_event_type* to further classify the event
  */
-#define RTE_EVENT_TYPE_ETH_RX_ADAPTER   0x4
+#define RTE_EVENT_TYPE_ETH_RX_ADAPTER 0x4
 /**< The event generated from event eth Rx adapter */
-#define RTE_EVENT_TYPE_MAX              0x10
+#define RTE_EVENT_TYPE_VECTOR 0x8
+/**< Indicates that event is a vector.
+ * All vector event types should be an logical OR of EVENT_TYPE_VECTOR.
+ * This simplifies the pipeline design as we can split processing the events
+ * between vector events and normal event across event types.
+ * Example:
+ *	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
+ *		// Classify and handle vector event.
+ *	} else {
+ *		// Classify and handle event.
+ *	}
+ */
+#define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
+/**< The event vector generated from cpu for pipelining. */
+
+#define RTE_EVENT_TYPE_MAX 0x10
 /**< Maximum number of event types */
 
 /* Event enqueue operations */
-#define RTE_EVENT_OP_NEW                0
+#define RTE_EVENT_OP_NEW 0
 /**< The event producers use this operation to inject a new event to the
  * event device.
  */
-#define RTE_EVENT_OP_FORWARD            1
+#define RTE_EVENT_OP_FORWARD 1
 /**< The CPU use this operation to forward the event to different event queue or
  * change to new application specific flow or schedule type to enable
  * pipelining.
@@ -1002,7 +1022,7 @@ rte_event_dev_close(uint8_t dev_id);
  * This operation must only be enqueued to the same port that the
  * event to be forwarded was dequeued from.
  */
-#define RTE_EVENT_OP_RELEASE            2
+#define RTE_EVENT_OP_RELEASE 2
 /**< Release the flow context associated with the schedule type.
  *
  * If current flow's scheduler type method is *RTE_SCHED_TYPE_ATOMIC*
@@ -1046,30 +1066,30 @@ struct rte_event {
 		uint64_t event;
 		/** Event attributes for dequeue or enqueue operation */
 		struct {
-			uint32_t flow_id:20;
+			uint32_t flow_id : 20;
 			/**< Targeted flow identifier for the enqueue and
 			 * dequeue operation.
 			 * The value must be in the range of
 			 * [0, nb_event_queue_flows - 1] which
 			 * previously supplied to rte_event_dev_configure().
 			 */
-			uint32_t sub_event_type:8;
+			uint32_t sub_event_type : 8;
 			/**< Sub-event types based on the event source.
 			 * @see RTE_EVENT_TYPE_CPU
 			 */
-			uint32_t event_type:4;
+			uint32_t event_type : 4;
 			/**< Event type to classify the event source.
 			 * @see RTE_EVENT_TYPE_ETHDEV, (RTE_EVENT_TYPE_*)
 			 */
-			uint8_t op:2;
+			uint8_t op : 2;
 			/**< The type of event enqueue operation - new/forward/
 			 * etc.This field is not preserved across an instance
 			 * and is undefined on dequeue.
 			 * @see RTE_EVENT_OP_NEW, (RTE_EVENT_OP_*)
 			 */
-			uint8_t rsvd:4;
+			uint8_t rsvd : 4;
 			/**< Reserved for future use */
-			uint8_t sched_type:2;
+			uint8_t sched_type : 2;
 			/**< Scheduler synchronization type (RTE_SCHED_TYPE_*)
 			 * associated with flow id on a given event queue
 			 * for the enqueue and dequeue operation.
@@ -1108,19 +1128,21 @@ struct rte_event {
 		/**< Opaque event pointer */
 		struct rte_mbuf *mbuf;
 		/**< mbuf pointer if dequeued event is associated with mbuf */
+		struct rte_event_vector *vec;
+		/**< Event vector pointer. */
 	};
 };
 
 /* Ethdev Rx adapter capability bitmap flags */
-#define RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT	0x1
+#define RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT 0x1
 /**< This flag is sent when the packet transfer mechanism is in HW.
  * Ethdev can send packets to the event device using internal event port.
  */
-#define RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ	0x2
+#define RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ 0x2
 /**< Adapter supports multiple event queues per ethdev. Every ethdev
  * Rx queue can be connected to a unique event queue.
  */
-#define RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID	0x4
+#define RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID 0x4
 /**< The application can override the adapter generated flow ID in the
  * event. This flow ID can be specified when adding an ethdev Rx queue
  * to the adapter using the ev member of struct rte_event_eth_rx_adapter
@@ -1147,9 +1169,8 @@ struct rte_event {
  *   - <0: Error code returned by the driver function.
  *
  */
-int
-rte_event_eth_rx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
-				uint32_t *caps);
+int rte_event_eth_rx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
+				      uint32_t *caps);
 
 #define RTE_EVENT_TIMER_ADAPTER_CAP_INTERNAL_PORT (1ULL << 0)
 /**< This flag is set when the timer mechanism is in HW. */
@@ -1167,30 +1188,29 @@ rte_event_eth_rx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
  *   - 0: Success, driver provided event timer adapter capabilities.
  *   - <0: Error code returned by the driver function.
  */
-int
-rte_event_timer_adapter_caps_get(uint8_t dev_id, uint32_t *caps);
+int rte_event_timer_adapter_caps_get(uint8_t dev_id, uint32_t *caps);
 
 /* Crypto adapter capability bitmap flag */
-#define RTE_EVENT_CRYPTO_ADAPTER_CAP_INTERNAL_PORT_OP_NEW   0x1
+#define RTE_EVENT_CRYPTO_ADAPTER_CAP_INTERNAL_PORT_OP_NEW 0x1
 /**< Flag indicates HW is capable of generating events in
  * RTE_EVENT_OP_NEW enqueue operation. Cryptodev will send
  * packets to the event device as new events using an internal
  * event port.
  */
 
-#define RTE_EVENT_CRYPTO_ADAPTER_CAP_INTERNAL_PORT_OP_FWD   0x2
+#define RTE_EVENT_CRYPTO_ADAPTER_CAP_INTERNAL_PORT_OP_FWD 0x2
 /**< Flag indicates HW is capable of generating events in
  * RTE_EVENT_OP_FORWARD enqueue operation. Cryptodev will send
  * packets to the event device as forwarded event using an
  * internal event port.
  */
 
-#define RTE_EVENT_CRYPTO_ADAPTER_CAP_INTERNAL_PORT_QP_EV_BIND  0x4
+#define RTE_EVENT_CRYPTO_ADAPTER_CAP_INTERNAL_PORT_QP_EV_BIND 0x4
 /**< Flag indicates HW is capable of mapping crypto queue pair to
  * event queue.
  */
 
-#define RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA   0x8
+#define RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA 0x8
 /**< Flag indicates HW/SW supports a mechanism to store and retrieve
  * the private data information along with the crypto session.
  */
@@ -1215,12 +1235,11 @@ rte_event_timer_adapter_caps_get(uint8_t dev_id, uint32_t *caps);
  *   - <0: Error code returned by the driver function.
  *
  */
-int
-rte_event_crypto_adapter_caps_get(uint8_t dev_id, uint8_t cdev_id,
-				  uint32_t *caps);
+int rte_event_crypto_adapter_caps_get(uint8_t dev_id, uint8_t cdev_id,
+				      uint32_t *caps);
 
 /* Ethdev Tx adapter capability bitmap flags */
-#define RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT	0x1
+#define RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT 0x1
 /**< This flag is sent when the PMD supports a packet transmit callback
  */
 
@@ -1241,9 +1260,8 @@ rte_event_crypto_adapter_caps_get(uint8_t dev_id, uint8_t cdev_id,
  *   - <0: Error code returned by the driver function.
  *
  */
-int
-rte_event_eth_tx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
-				uint32_t *caps);
+int rte_event_eth_tx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
+				      uint32_t *caps);
 
 struct rte_eventdev_ops;
 struct rte_eventdev;
@@ -1252,28 +1270,31 @@ typedef uint16_t (*event_enqueue_t)(void *port, const struct rte_event *ev);
 /**< @internal Enqueue event on port of a device */
 
 typedef uint16_t (*event_enqueue_burst_t)(void *port,
-			const struct rte_event ev[], uint16_t nb_events);
+					  const struct rte_event ev[],
+					  uint16_t nb_events);
 /**< @internal Enqueue burst of events on port of a device */
 
 typedef uint16_t (*event_dequeue_t)(void *port, struct rte_event *ev,
-		uint64_t timeout_ticks);
+				    uint64_t timeout_ticks);
 /**< @internal Dequeue event from port of a device */
 
 typedef uint16_t (*event_dequeue_burst_t)(void *port, struct rte_event ev[],
-		uint16_t nb_events, uint64_t timeout_ticks);
+					  uint16_t nb_events,
+					  uint64_t timeout_ticks);
 /**< @internal Dequeue burst of events from port of a device */
 
-typedef uint16_t (*event_tx_adapter_enqueue)(void *port,
-				struct rte_event ev[], uint16_t nb_events);
+typedef uint16_t (*event_tx_adapter_enqueue)(void *port, struct rte_event ev[],
+					     uint16_t nb_events);
 /**< @internal Enqueue burst of events on port of a device */
 
 typedef uint16_t (*event_tx_adapter_enqueue_same_dest)(void *port,
-		struct rte_event ev[], uint16_t nb_events);
+						       struct rte_event ev[],
+						       uint16_t nb_events);
 /**< @internal Enqueue burst of events on port of a device supporting
  * burst having same destination Ethernet port & Tx queue.
  */
 
-#define RTE_EVENTDEV_NAME_MAX_LEN	(64)
+#define RTE_EVENTDEV_NAME_MAX_LEN (64)
 /**< @internal Max length of name of event PMD */
 
 /**
@@ -1321,7 +1342,7 @@ struct rte_eventdev_data {
 	/**< Unique identifier name */
 
 	uint64_t reserved_64s[4]; /**< Reserved for future fields */
-	void *reserved_ptrs[4];   /**< Reserved for future fields */
+	void *reserved_ptrs[4];	  /**< Reserved for future fields */
 } __rte_cache_aligned;
 
 /** @internal The data structure associated with each event device. */
@@ -1356,7 +1377,7 @@ struct rte_eventdev {
 	/**< Flag indicating the device is attached */
 
 	uint64_t reserved_64s[4]; /**< Reserved for future fields */
-	void *reserved_ptrs[4];   /**< Reserved for future fields */
+	void *reserved_ptrs[4];	  /**< Reserved for future fields */
 } __rte_cache_aligned;
 
 extern struct rte_eventdev *rte_eventdevs;
@@ -1364,8 +1385,8 @@ extern struct rte_eventdev *rte_eventdevs;
 
 static __rte_always_inline uint16_t
 __rte_event_enqueue_burst(uint8_t dev_id, uint8_t port_id,
-			const struct rte_event ev[], uint16_t nb_events,
-			const event_enqueue_burst_t fn)
+			  const struct rte_event ev[], uint16_t nb_events,
+			  const event_enqueue_burst_t fn)
 {
 	const struct rte_eventdev *dev = &rte_eventdevs[dev_id];
 
@@ -1441,7 +1462,7 @@ rte_event_enqueue_burst(uint8_t dev_id, uint8_t port_id,
 	const struct rte_eventdev *dev = &rte_eventdevs[dev_id];
 
 	return __rte_event_enqueue_burst(dev_id, port_id, ev, nb_events,
-			dev->enqueue_burst);
+					 dev->enqueue_burst);
 }
 
 /**
@@ -1487,12 +1508,12 @@ rte_event_enqueue_burst(uint8_t dev_id, uint8_t port_id,
  */
 static inline uint16_t
 rte_event_enqueue_new_burst(uint8_t dev_id, uint8_t port_id,
-			const struct rte_event ev[], uint16_t nb_events)
+			    const struct rte_event ev[], uint16_t nb_events)
 {
 	const struct rte_eventdev *dev = &rte_eventdevs[dev_id];
 
 	return __rte_event_enqueue_burst(dev_id, port_id, ev, nb_events,
-			dev->enqueue_new_burst);
+					 dev->enqueue_new_burst);
 }
 
 /**
@@ -1538,12 +1559,12 @@ rte_event_enqueue_new_burst(uint8_t dev_id, uint8_t port_id,
  */
 static inline uint16_t
 rte_event_enqueue_forward_burst(uint8_t dev_id, uint8_t port_id,
-			const struct rte_event ev[], uint16_t nb_events)
+				const struct rte_event ev[], uint16_t nb_events)
 {
 	const struct rte_eventdev *dev = &rte_eventdevs[dev_id];
 
 	return __rte_event_enqueue_burst(dev_id, port_id, ev, nb_events,
-			dev->enqueue_forward_burst);
+					 dev->enqueue_forward_burst);
 }
 
 /**
@@ -1571,9 +1592,8 @@ rte_event_enqueue_forward_burst(uint8_t dev_id, uint8_t port_id,
  * @see rte_event_dev_configure()
  *
  */
-int
-rte_event_dequeue_timeout_ticks(uint8_t dev_id, uint64_t ns,
-					uint64_t *timeout_ticks);
+int rte_event_dequeue_timeout_ticks(uint8_t dev_id, uint64_t ns,
+				    uint64_t *timeout_ticks);
 
 /**
  * Dequeue a burst of events objects or an event object from the event port
@@ -1664,12 +1684,11 @@ rte_event_dequeue_burst(uint8_t dev_id, uint8_t port_id, struct rte_event ev[],
 	 * requests nb_events as const one
 	 */
 	if (nb_events == 1)
-		return (*dev->dequeue)(
-			dev->data->ports[port_id], ev, timeout_ticks);
+		return (*dev->dequeue)(dev->data->ports[port_id], ev,
+				       timeout_ticks);
 	else
-		return (*dev->dequeue_burst)(
-			dev->data->ports[port_id], ev, nb_events,
-				timeout_ticks);
+		return (*dev->dequeue_burst)(dev->data->ports[port_id], ev,
+					     nb_events, timeout_ticks);
 }
 
 /**
@@ -1732,10 +1751,8 @@ rte_event_dequeue_burst(uint8_t dev_id, uint8_t port_id, struct rte_event ev[],
  * (EINVAL) Invalid parameter
  *
  */
-int
-rte_event_port_link(uint8_t dev_id, uint8_t port_id,
-		    const uint8_t queues[], const uint8_t priorities[],
-		    uint16_t nb_links);
+int rte_event_port_link(uint8_t dev_id, uint8_t port_id, const uint8_t queues[],
+			const uint8_t priorities[], uint16_t nb_links);
 
 /**
  * Unlink multiple source event queues supplied in *queues* from the destination
@@ -1776,9 +1793,8 @@ rte_event_port_link(uint8_t dev_id, uint8_t port_id,
  * the rte_errno accordingly, Possible rte_errno values are
  * (EINVAL) Invalid parameter
  */
-int
-rte_event_port_unlink(uint8_t dev_id, uint8_t port_id,
-		      uint8_t queues[], uint16_t nb_unlinks);
+int rte_event_port_unlink(uint8_t dev_id, uint8_t port_id, uint8_t queues[],
+			  uint16_t nb_unlinks);
 
 /**
  * Returns the number of unlinks in progress.
@@ -1801,8 +1817,7 @@ rte_event_port_unlink(uint8_t dev_id, uint8_t port_id,
  * A negative return value indicates an error, -EINVAL indicates an invalid
  * parameter passed for *dev_id* or *port_id*.
  */
-int
-rte_event_port_unlinks_in_progress(uint8_t dev_id, uint8_t port_id);
+int rte_event_port_unlinks_in_progress(uint8_t dev_id, uint8_t port_id);
 
 /**
  * Retrieve the list of source event queues and its associated service priority
@@ -1831,9 +1846,8 @@ rte_event_port_unlinks_in_progress(uint8_t dev_id, uint8_t port_id);
  * - <0 on failure.
  *
  */
-int
-rte_event_port_links_get(uint8_t dev_id, uint8_t port_id,
-			 uint8_t queues[], uint8_t priorities[]);
+int rte_event_port_links_get(uint8_t dev_id, uint8_t port_id, uint8_t queues[],
+			     uint8_t priorities[]);
 
 /**
  * Retrieve the service ID of the event dev. If the adapter doesn't use
@@ -1850,8 +1864,7 @@ rte_event_port_links_get(uint8_t dev_id, uint8_t port_id,
  *   - <0: Error code on failure, if the event dev doesn't use a rte_service
  *   function, this function returns -ESRCH.
  */
-int
-rte_event_dev_service_id_get(uint8_t dev_id, uint32_t *service_id);
+int rte_event_dev_service_id_get(uint8_t dev_id, uint32_t *service_id);
 
 /**
  * Dump internal information about *dev_id* to the FILE* provided in *f*.
@@ -1866,8 +1879,7 @@ rte_event_dev_service_id_get(uint8_t dev_id, uint32_t *service_id);
  *   - 0: on success
  *   - <0: on failure.
  */
-int
-rte_event_dev_dump(uint8_t dev_id, FILE *f);
+int rte_event_dev_dump(uint8_t dev_id, FILE *f);
 
 /** Maximum name length for extended statistics counters */
 #define RTE_EVENT_DEV_XSTATS_NAME_SIZE 64
@@ -1923,13 +1935,10 @@ struct rte_event_dev_xstats_name {
  *        -EINVAL for invalid mode, queue port or id parameters
  *        -ENOTSUP if the device doesn't support this function.
  */
-int
-rte_event_dev_xstats_names_get(uint8_t dev_id,
-			       enum rte_event_dev_xstats_mode mode,
-			       uint8_t queue_port_id,
-			       struct rte_event_dev_xstats_name *xstats_names,
-			       unsigned int *ids,
-			       unsigned int size);
+int rte_event_dev_xstats_names_get(
+	uint8_t dev_id, enum rte_event_dev_xstats_mode mode,
+	uint8_t queue_port_id, struct rte_event_dev_xstats_name *xstats_names,
+	unsigned int *ids, unsigned int size);
 
 /**
  * Retrieve extended statistics of an event device.
@@ -1957,12 +1966,10 @@ rte_event_dev_xstats_names_get(uint8_t dev_id,
  *        -EINVAL for invalid mode, queue port or id parameters
  *        -ENOTSUP if the device doesn't support this function.
  */
-int
-rte_event_dev_xstats_get(uint8_t dev_id,
-			 enum rte_event_dev_xstats_mode mode,
-			 uint8_t queue_port_id,
-			 const unsigned int ids[],
-			 uint64_t values[], unsigned int n);
+int rte_event_dev_xstats_get(uint8_t dev_id,
+			     enum rte_event_dev_xstats_mode mode,
+			     uint8_t queue_port_id, const unsigned int ids[],
+			     uint64_t values[], unsigned int n);
 
 /**
  * Retrieve the value of a single stat by requesting it by name.
@@ -1980,9 +1987,8 @@ rte_event_dev_xstats_get(uint8_t dev_id,
  *   - positive value or zero: the stat value
  *   - negative value: -EINVAL if stat not found, -ENOTSUP if not supported.
  */
-uint64_t
-rte_event_dev_xstats_by_name_get(uint8_t dev_id, const char *name,
-				 unsigned int *id);
+uint64_t rte_event_dev_xstats_by_name_get(uint8_t dev_id, const char *name,
+					  unsigned int *id);
 
 /**
  * Reset the values of the xstats of the selected component in the device.
@@ -2004,12 +2010,10 @@ rte_event_dev_xstats_by_name_get(uint8_t dev_id, const char *name,
  *   - zero: successfully reset the statistics to zero
  *   - negative value: -EINVAL invalid parameters, -ENOTSUP if not supported.
  */
-int
-rte_event_dev_xstats_reset(uint8_t dev_id,
-			   enum rte_event_dev_xstats_mode mode,
-			   int16_t queue_port_id,
-			   const uint32_t ids[],
-			   uint32_t nb_ids);
+int rte_event_dev_xstats_reset(uint8_t dev_id,
+			       enum rte_event_dev_xstats_mode mode,
+			       int16_t queue_port_id, const uint32_t ids[],
+			       uint32_t nb_ids);
 
 /**
  * Trigger the eventdev self test.
@@ -2023,6 +2027,78 @@ rte_event_dev_xstats_reset(uint8_t dev_id,
  */
 int rte_event_dev_selftest(uint8_t dev_id);
 
+/**
+ * Get the memory required per event vector based on the number of elements per
+ * vector.
+ * This should be used to create the mempool that holds the event vectors.
+ *
+ * @param name
+ *   The name of the vector pool.
+ * @param n
+ *   The number of elements in the mbuf pool.
+ * @param cache_size
+ *   Size of the per-core object cache. See rte_mempool_create() for
+ *   details.
+ * @param nb_elem
+ *   The number of elements then a single event vector should be able to hold.
+ * @param socket_id
+ *   The socket identifier where the memory should be allocated. The
+ *   value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
+ *   reserved zone
+ *
+ * @return
+ *   The pointer to the newly allocated mempool, on success. NULL on error
+ *   with rte_errno set appropriately. Possible rte_errno values include:
+ *    - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
+ *    - E_RTE_SECONDARY - function was called from a secondary process instance
+ *    - EINVAL - cache size provided is too large, or priv_size is not aligned.
+ *    - ENOSPC - the maximum number of memzones has already been allocated
+ *    - EEXIST - a memzone with the same name already exists
+ *    - ENOMEM - no appropriate memory area found in which to create memzone
+ */
+__rte_experimental
+static inline struct rte_mempool *
+rte_event_vector_pool_create(const char *name, unsigned int n,
+			     unsigned int cache_size, uint16_t nb_elem,
+			     int socket_id)
+{
+	const char *mp_ops_name;
+	struct rte_mempool *mp;
+	unsigned int elt_sz;
+	int ret;
+
+	if (!nb_elem) {
+		RTE_LOG(ERR, EVENTDEV,
+			"Invalid number of elements=%d requested\n", nb_elem);
+		rte_errno = -EINVAL;
+		return NULL;
+	}
+
+	elt_sz =
+		sizeof(struct rte_event_vector) + (nb_elem * sizeof(uintptr_t));
+	mp = rte_mempool_create_empty(name, n, elt_sz, cache_size, 0, socket_id,
+				      0);
+	if (mp == NULL)
+		return NULL;
+
+	mp_ops_name = rte_mbuf_best_mempool_ops();
+	ret = rte_mempool_set_ops_byname(mp, mp_ops_name, NULL);
+	if (ret != 0) {
+		RTE_LOG(ERR, EVENTDEV, "error setting mempool handler\n");
+		goto err;
+	}
+
+	ret = rte_mempool_populate_default(mp);
+	if (ret < 0)
+		goto err;
+
+	return mp;
+err:
+	rte_mempool_free(mp);
+	rte_errno = -ret;
+	return NULL;
+}
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index 3e5c09cfd..a070ef56e 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -138,6 +138,9 @@ EXPERIMENTAL {
 	__rte_eventdev_trace_port_setup;
 	# added in 20.11
 	rte_event_pmd_pci_probe_named;
+
+	#added in 21.05
+	rte_event_vector_pool_create;
 };
 
 INTERNAL {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v2 2/8] eventdev: introduce event vector Rx capability
  2021-03-16 15:48 ` [dpdk-dev] [PATCH v2 0/8] " pbhagavatula
  2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 1/8] eventdev: introduce event vector capability pbhagavatula
@ 2021-03-16 15:48   ` pbhagavatula
  2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 3/8] eventdev: introduce event vector Tx capability pbhagavatula
                     ` (6 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-16 15:48 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce event ethernet Rx adapter event vector capability.

If an event eth Rx adapter has the capability of
RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR then a given Rx queue
can be configured to enable event vectorization by passing the
flag RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR to
rte_event_eth_rx_adapter_queue_conf::rx_queue_flags while configuring
Rx adapter through rte_event_eth_rx_adapter_queue_add.

The max vector size, vector timeout define the vector size and
mempool used for allocating vector event are configured through
rte_event_eth_rx_adapter_queue_add. The element size of the element
in the vector pool should be equal to
    sizeof(struct rte_event_vector) + (vector_sz * sizeof(uintptr_t))

Application can use `rte_event_vector_pool_create` to create the
vector mempool used for
rte_event_eth_rx_adapter_queue_conf::vector_mp.

The Rx adapter would be responsible for vectorizing the mbufs
based on the flow, the vector limits configured by the application
and add the vector event of mbufs to the event queue set via
rte_event_eth_rx_adapter_queue_conf::ev::queue_id.
It should also mark rte_event_vector::union_valid and fill
rte_event_vector::port, rte_event_vector::queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 .../prog_guide/event_ethernet_rx_adapter.rst  | 38 ++++++++
 .../rte_event_eth_rx_adapter.h                | 95 +++++++++++++++++++
 lib/librte_eventdev/rte_eventdev.h            | 30 +++++-
 lib/librte_eventdev/version.map               |  2 +
 4 files changed, 163 insertions(+), 2 deletions(-)

diff --git a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
index cb44ce0e4..5eefef355 100644
--- a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
@@ -186,3 +186,41 @@ the event buffer fill level is low. The
 ``rte_event_eth_rx_adapter_cb_register()`` function allow the application
 to register a callback that selects which packets to enqueue to the event
 device.
+
+Rx event vectorization
+~~~~~~~~~~~~~~~~~~~~~~
+
+The event devices, ethernet device pairs which support the capability
+``RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR`` can aggregate packets based on
+flow characteristics and generate a ``rte_event`` containing ``rte_event_vector``
+whose event type is either ``RTE_EVENT_TYPE_ETHDEV_VECTOR`` or
+``RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR``.
+The aggregation size and timeout are configurable at a queue level and the
+maximum, minimum vector sizes and timeouts vary based on the device capability
+and can be queried using ``rte_event_eth_rx_adapter_vector_limits_get``.
+The Rx adapter additionally might include useful data such as ethernet device
+port and queue identifier in the ``rte_event_vector::port`` and
+``rte_event_vector::queue`` and mark ``rte_event_vector::attr_valid`` as true.
+
+A loop processing ``rte_event_vector`` containing mbufs is shown below.
+
+.. code-block:: c
+
+        event = rte_event_dequeue_burst(event_dev, event_port, &event,
+                                        1, 0);
+        if (!event)
+                continue;
+
+        switch (ev.event_type) {
+        case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
+        case RTE_EVENT_TYPE_ETHDEV_VECTOR:
+                struct rte_mbufs **mbufs;
+
+                mbufs = (struct rte_mbufs **)ev[i].vec->mbufs;
+                for (i = 0; i < ev.vec->nb_elem; i++) {
+                        /* Process each mbuf. */
+                }
+        break;
+        case ...
+        ...
+        }
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
index 21bb1e54c..dda846f57 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
@@ -92,6 +92,10 @@ extern "C" {
 /**< This flag indicates the flow identifier is valid
  * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
  */
+#define RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR	0x2
+/**< This flag indicates that mbufs arriving on the queue need to be vectorized
+ * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+ */
 
 /**
  * Adapter configuration structure that the adapter configuration callback
@@ -169,6 +173,36 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 */
 };
 
+struct rte_event_eth_rx_adapter_event_vector_config {
+	uint16_t vector_sz;
+	/**<
+	 * Indicates the maximum number for mbufs to combine and form a vector.
+	 * Should be within
+	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_sz
+	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_sz
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+	 */
+	uint64_t vector_timeout_ns;
+	/**<
+	 * Indicates the maximum number of nanoseconds to wait for receiving
+	 * mbufs. Should be within vectorization limits of the
+	 * adapter
+	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_ns
+	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_ns
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+	 */
+	struct rte_mempool *vector_mp;
+	/**<
+	 * Indicates the mempool that should be used for allocating
+	 * rte_event_vector container.
+	 * Should be created by using `rte_event_vector_pool_create`.
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags.
+	 */
+};
+
 /**
  * A structure used to retrieve statistics for an eth rx adapter instance.
  */
@@ -199,6 +233,22 @@ struct rte_event_eth_rx_adapter_stats {
 	/**< Received packet count for interrupt mode Rx queues */
 };
 
+/**
+ * A structure used to retrieve eth rx adapter vector limits.
+ */
+struct rte_event_eth_rx_adapter_vector_limits {
+	uint16_t min_sz;
+	/**< Minimum vector limit configurable. */
+	uint16_t max_sz;
+	/**< Maximum vector limit configurable. */
+	uint8_t log2_sz;
+	/**< The size configured should be in log2. */
+	uint64_t min_timeout_ns;
+	/**< Minimum vector timeout configurable. */
+	uint64_t max_timeout_ns;
+	/**< Maximum vector timeout configurable. */
+};
+
 /**
  *
  * Callback function invoked by the SW adapter before it continues
@@ -467,6 +517,51 @@ int rte_event_eth_rx_adapter_cb_register(uint8_t id, uint16_t eth_dev_id,
 					 rte_event_eth_rx_adapter_cb_fn cb_fn,
 					 void *cb_arg);
 
+/**
+ * Retrieve vector limits for a given event dev and eth dev pair.
+ * @see rte_event_eth_rx_adapter_vector_limits
+ *
+ * @param dev_id
+ *  Event device identifier.
+ * @param eth_port_id
+ *  Port identifier of the ethernet device.
+ * @param [out] limits
+ *  A pointer to rte_event_eth_rx_adapter_vector_limits structure that has to
+ * be filled.
+ *
+ * @return
+ *  - 0: Success.
+ *  - <0: Error code on failure.
+ */
+__rte_experimental
+int rte_event_eth_rx_adapter_vector_limits_get(
+	uint8_t dev_id, uint16_t eth_port_id,
+	struct rte_event_eth_rx_adapter_vector_limits *limits);
+
+/**
+ * Configure event vectorization for a given ethernet device queue, that has
+ * been added to a event eth Rx adapter.
+ *
+ * @param id
+ *  The identifier of the ethernet Rx event adapter.
+ *
+ * @param eth_dev_id
+ *  The identifier of the ethernet device.
+ *
+ * @param rx_queue_id
+ *  Ethernet device receive queue index.
+ *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
+ *  are configured with event vectorization.
+ *
+ * @return
+ *  - 0: Success, Receive queue configured correctly.
+ *  - <0: Error code on failure.
+ */
+__rte_experimental
+int rte_event_eth_rx_adapter_queue_event_vector_config(
+	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
+	struct rte_event_eth_rx_adapter_event_vector_config *config);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index dda8d0387..963231df4 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -903,9 +903,27 @@ int rte_event_dev_close(uint8_t dev_id);
  * Event vector structure.
  */
 struct rte_event_vector {
-	uint64_t nb_elem : 16;
+	uint16_t nb_elem;
 	/**< Number of elements in this event vector. */
-	uint64_t rsvd : 48;
+	uint16_t rsvd : 15;
+	uint16_t attr_valid : 1;
+	/**< Indicates that the below union attributes have valid information.
+	 */
+	union {
+		/* Used by Rx adapter.
+		 * Indicats that all the elements in this vector belong to same
+		 * port and queue pair when originating from Rx adapter, valid
+		 * only when event type is ETHDEV_VECTOR or
+		 * ETH_RX_ADAPTER_VECTOR.
+		 */
+		struct {
+			uint16_t port;
+			/* Ethernet device port id. */
+			uint16_t queue;
+			/* Ethernet device queue id. */
+		};
+	};
+	/**< Union to hold common attributes of the vector array. */
 	uint64_t impl_opaque;
 	union {
 		struct rte_mbuf *mbufs[0];
@@ -1003,8 +1021,14 @@ struct rte_event_vector {
  *		// Classify and handle event.
  *	}
  */
+#define RTE_EVENT_TYPE_ETHDEV_VECTOR                                           \
+	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETHDEV)
+/**< The event vector generated from ethdev subsystem */
 #define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
 /**< The event vector generated from cpu for pipelining. */
+#define RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR                                   \
+	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETH_RX_ADAPTER)
+/**< The event vector generated from eth Rx adapter. */
 
 #define RTE_EVENT_TYPE_MAX 0x10
 /**< Maximum number of event types */
@@ -1149,6 +1173,8 @@ struct rte_event {
  * @see struct rte_event_eth_rx_adapter_queue_conf::ev
  * @see struct rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
  */
+#define RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR	0x8
+/**< Adapter supports event vectorization per ethdev. */
 
 /**
  * Retrieve the event device's ethdev Rx adapter capabilities for the
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index a070ef56e..902df0ae3 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -141,6 +141,8 @@ EXPERIMENTAL {
 
 	#added in 21.05
 	rte_event_vector_pool_create;
+	rte_event_eth_rx_adapter_vector_limits_get;
+	rte_event_eth_rx_adapter_queue_event_vector_config;
 };
 
 INTERNAL {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v2 3/8] eventdev: introduce event vector Tx capability
  2021-03-16 15:48 ` [dpdk-dev] [PATCH v2 0/8] " pbhagavatula
  2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 1/8] eventdev: introduce event vector capability pbhagavatula
  2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 2/8] eventdev: introduce event vector Rx capability pbhagavatula
@ 2021-03-16 15:48   ` pbhagavatula
  2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 4/8] eventdev: add Rx adapter event vector support pbhagavatula
                     ` (5 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-16 15:48 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce event vector transmit capability for event eth
tx adapter.

The capability indicates that the Tx adapter is capable of
transmitting event vectors.
When rte_event_vector::union_valid is set, the Tx adapter should
transmit all the packets to the rte_event_vector::port using the
rte_event_vector::queue.
If rte_event_vector::union_valid is not set then the Tx adapter
should peek into each mbuf to get the destination port and queue
pair.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
---
 doc/guides/prog_guide/event_ethernet_tx_adapter.rst | 12 ++++++++++++
 lib/librte_eventdev/rte_eventdev.h                  |  8 +++++++-
 2 files changed, 19 insertions(+), 1 deletion(-)

diff --git a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
index a8c13e136..87277dcaf 100644
--- a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
@@ -164,3 +164,15 @@ The  ``rte_event_eth_tx_adapter_stats_get()`` function reports counters defined
 in struct ``rte_event_eth_tx_adapter_stats``. The counter values are the sum of
 the counts from the eventdev PMD callback if the callback is supported, and
 the counts maintained by the service function, if one exists.
+
+Tx event vectorization
+~~~~~~~~~~~~~~~~~~~~~~
+
+The event device, ethernet device paris which support the capability
+``RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR`` can process event vector of mbufs.
+Additionally, application can provide a hint to that Tx adapter that all the
+mbufs are destined to the same ethernet port and queue by setting the bit
+``rte_event_vector::attr_valid`` and filling `rte_event_vector::port`` and
+``rte_event_vector::queue``.
+If ``rte_event_vector::attr_valid`` is not set then the Tx adapter should peek
+into each mbuf and transmit them to the requested ethernet port and queue pair.
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index 963231df4..006b6957a 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -910,11 +910,13 @@ struct rte_event_vector {
 	/**< Indicates that the below union attributes have valid information.
 	 */
 	union {
-		/* Used by Rx adapter.
+		/* Used by Rx/Tx adapter.
 		 * Indicats that all the elements in this vector belong to same
 		 * port and queue pair when originating from Rx adapter, valid
 		 * only when event type is ETHDEV_VECTOR or
 		 * ETH_RX_ADAPTER_VECTOR.
+		 * Can also be used to indicate the Tx adapter the destination
+		 * port and queue of the mbufs in the vector
 		 */
 		struct {
 			uint16_t port;
@@ -1268,6 +1270,10 @@ int rte_event_crypto_adapter_caps_get(uint8_t dev_id, uint8_t cdev_id,
 #define RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT 0x1
 /**< This flag is sent when the PMD supports a packet transmit callback
  */
+#define RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR	0x2
+/**< Indicates that the Tx adapter is capable of handling event vector of
+ * mbufs.
+ */
 
 /**
  * Retrieve the event device's eth Tx adapter capabilities
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v2 4/8] eventdev: add Rx adapter event vector support
  2021-03-16 15:48 ` [dpdk-dev] [PATCH v2 0/8] " pbhagavatula
                     ` (2 preceding siblings ...)
  2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 3/8] eventdev: introduce event vector Tx capability pbhagavatula
@ 2021-03-16 15:48   ` pbhagavatula
  2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 5/8] eventdev: add Tx " pbhagavatula
                     ` (4 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-16 15:48 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support for event eth Rx adapter, the implementation
creates vector flows based on port and queue identifier of the received
mbufs.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 lib/librte_eventdev/eventdev_pmd.h            |  60 ++-
 .../rte_event_eth_rx_adapter.c                | 367 +++++++++++++++++-
 lib/librte_eventdev/rte_eventdev.c            |   6 +-
 3 files changed, 415 insertions(+), 18 deletions(-)

diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 7eb9a7739..d79dfd612 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -69,9 +69,10 @@ extern "C" {
 	} \
 } while (0)
 
-#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP \
-		((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) | \
-			(RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ))
+#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP                                        \
+	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |                     \
+	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) |                         \
+	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR))
 
 #define RTE_EVENT_CRYPTO_ADAPTER_SW_CAP \
 		RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA
@@ -645,6 +646,53 @@ typedef int (*eventdev_eth_rx_adapter_stats_reset)
  */
 typedef int (*eventdev_selftest)(void);
 
+struct rte_event_eth_rx_adapter_vector_limits;
+/**
+ * Get event vector limits for a given event, etherner device pair.
+ *
+ * @param dev
+ *   Event device pointer
+ *
+ * @param eth_dev
+ *   Ethernet device pointer
+ *
+ * @param[out] limits
+ *   Pointer to the limits structure to be filled.
+ *
+ * @return
+ *   - 0: Success.
+ *   - <0: Error code returned by the driver function.
+ */
+typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
+	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
+	struct rte_event_eth_rx_adapter_vector_limits *limits);
+
+struct rte_event_eth_rx_adapter_event_vector_config;
+/**
+ * Enable event vector on an given Rx queue of a ethernet devices belonging to
+ * the Rx adapter.
+ *
+ * @param dev
+ *   Event device pointer
+ *
+ * @param eth_dev
+ *   Ethernet device pointer
+ *
+ * @param rx_queue_id
+ *   The Rx queue identifier
+ *
+ * @param config
+ *   Pointer to the event vector configuration structure.
+ *
+ * @return
+ *   - 0: Success.
+ *   - <0: Error code returned by the driver function.
+ */
+typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
+	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
+	int32_t rx_queue_id,
+	const struct rte_event_eth_rx_adapter_event_vector_config *config);
+
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;
 
@@ -1067,6 +1115,12 @@ struct rte_eventdev_ops {
 	/**< Get ethernet Rx stats */
 	eventdev_eth_rx_adapter_stats_reset eth_rx_adapter_stats_reset;
 	/**< Reset ethernet Rx stats */
+	eventdev_eth_rx_adapter_vector_limits_get_t
+		eth_rx_adapter_vector_limits_get;
+	/**< Get event vector limits for the Rx adapter */
+	eventdev_eth_rx_adapter_event_vector_config_t
+		eth_rx_adapter_event_vector_config;
+	/**< Configure Rx adapter with event vector */
 
 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index d8c635e99..c71990078 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -26,6 +26,10 @@
 #define BATCH_SIZE		32
 #define BLOCK_CNT_THRESHOLD	10
 #define ETH_EVENT_BUFFER_SIZE	(4*BATCH_SIZE)
+#define MAX_VECTOR_SIZE		1024
+#define MIN_VECTOR_SIZE		4
+#define MAX_VECTOR_NS		1E9
+#define MIN_VECTOR_NS		1E5
 
 #define ETH_RX_ADAPTER_SERVICE_NAME_LEN	32
 #define ETH_RX_ADAPTER_MEM_NAME_LEN	32
@@ -59,6 +63,20 @@ struct eth_rx_poll_entry {
 	uint16_t eth_rx_qid;
 };
 
+struct eth_rx_vector_data {
+	TAILQ_ENTRY(eth_rx_vector_data) next;
+	uint16_t port;
+	uint16_t queue;
+	uint16_t max_vector_count;
+	uint64_t event;
+	uint64_t ts;
+	uint64_t vector_timeout_ticks;
+	struct rte_mempool *vector_pool;
+	struct rte_event_vector *vector_ev;
+} __rte_cache_aligned;
+
+TAILQ_HEAD(eth_rx_vector_data_list, eth_rx_vector_data);
+
 /* Instance per adapter */
 struct rte_eth_event_enqueue_buffer {
 	/* Count of events in this buffer */
@@ -92,6 +110,14 @@ struct rte_event_eth_rx_adapter {
 	uint32_t wrr_pos;
 	/* Event burst buffer */
 	struct rte_eth_event_enqueue_buffer event_enqueue_buffer;
+	/* Vector enable flag */
+	uint8_t ena_vector;
+	/* Timestamp of previous vector expiry list traversal */
+	uint64_t prev_expiry_ts;
+	/* Minimum ticks to wait before traversing expiry list */
+	uint64_t vector_tmo_ticks;
+	/* vector list */
+	struct eth_rx_vector_data_list vector_list;
 	/* Per adapter stats */
 	struct rte_event_eth_rx_adapter_stats stats;
 	/* Block count, counts up to BLOCK_CNT_THRESHOLD */
@@ -198,9 +224,11 @@ struct eth_device_info {
 struct eth_rx_queue_info {
 	int queue_enabled;	/* True if added */
 	int intr_enabled;
+	uint8_t ena_vector;
 	uint16_t wt;		/* Polling weight */
 	uint32_t flow_id_mask;	/* Set to ~0 if app provides flow id else 0 */
 	uint64_t event;
+	struct eth_rx_vector_data vector_data;
 };
 
 static struct rte_event_eth_rx_adapter **event_eth_rx_adapter;
@@ -722,6 +750,9 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
 	    &rx_adapter->event_enqueue_buffer;
 	struct rte_event_eth_rx_adapter_stats *stats = &rx_adapter->stats;
 
+	if (!buf->count)
+		return 0;
+
 	uint16_t n = rte_event_enqueue_new_burst(rx_adapter->eventdev_id,
 					rx_adapter->event_port_id,
 					buf->events,
@@ -742,6 +773,72 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
 	return n;
 }
 
+static inline uint16_t
+rxa_create_event_vector(struct rte_event_eth_rx_adapter *rx_adapter,
+			struct eth_rx_queue_info *queue_info,
+			struct rte_eth_event_enqueue_buffer *buf,
+			struct rte_mbuf **mbufs, uint16_t num)
+{
+	struct rte_event *ev = &buf->events[buf->count];
+	struct eth_rx_vector_data *vec;
+	uint16_t filled, space, sz;
+
+	filled = 0;
+	vec = &queue_info->vector_data;
+	while (num) {
+		if (vec->vector_ev == NULL) {
+			if (rte_mempool_get(vec->vector_pool,
+					    (void **)&vec->vector_ev) < 0) {
+				rte_pktmbuf_free_bulk(mbufs, num);
+				return 0;
+			}
+			vec->vector_ev->nb_elem = 0;
+			vec->vector_ev->port = vec->port;
+			vec->vector_ev->queue = vec->queue;
+			vec->vector_ev->attr_valid = true;
+			TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
+		} else if (vec->vector_ev->nb_elem == vec->max_vector_count) {
+			/* Event ready. */
+			ev->event = vec->event;
+			ev->vec = vec->vector_ev;
+			ev++;
+			filled++;
+			vec->vector_ev = NULL;
+			TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+			if (rte_mempool_get(vec->vector_pool,
+					    (void **)&vec->vector_ev) < 0) {
+				rte_pktmbuf_free_bulk(mbufs, num);
+				return 0;
+			}
+			vec->vector_ev->nb_elem = 0;
+			vec->vector_ev->port = vec->port;
+			vec->vector_ev->queue = vec->queue;
+			vec->vector_ev->attr_valid = true;
+			TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
+		}
+
+		space = vec->max_vector_count - vec->vector_ev->nb_elem;
+		sz = num > space ? space : num;
+		memcpy(vec->vector_ev->mbufs + vec->vector_ev->nb_elem, mbufs,
+		       sizeof(void *) * sz);
+		vec->vector_ev->nb_elem += sz;
+		num -= sz;
+		mbufs += sz;
+		vec->ts = rte_rdtsc();
+	}
+
+	if (vec->vector_ev->nb_elem == vec->max_vector_count) {
+		ev->event = vec->event;
+		ev->vec = vec->vector_ev;
+		ev++;
+		filled++;
+		vec->vector_ev = NULL;
+		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+	}
+
+	return filled;
+}
+
 static inline void
 rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
@@ -770,25 +867,30 @@ rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
 	rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
 	do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
 
-	for (i = 0; i < num; i++) {
-		m = mbufs[i];
-
-		rss = do_rss ?
-			rxa_do_softrss(m, rx_adapter->rss_key_be) :
-			m->hash.rss;
-		ev->event = event;
-		ev->flow_id = (rss & ~flow_id_mask) |
-				(ev->flow_id & flow_id_mask);
-		ev->mbuf = m;
-		ev++;
+	if (!eth_rx_queue_info->ena_vector) {
+		for (i = 0; i < num; i++) {
+			m = mbufs[i];
+
+			rss = do_rss ? rxa_do_softrss(m, rx_adapter->rss_key_be)
+				     : m->hash.rss;
+			ev->event = event;
+			ev->flow_id = (rss & ~flow_id_mask) |
+				      (ev->flow_id & flow_id_mask);
+			ev->mbuf = m;
+			ev++;
+		}
+	} else {
+		num = rxa_create_event_vector(rx_adapter, eth_rx_queue_info,
+					      buf, mbufs, num);
 	}
 
-	if (dev_info->cb_fn) {
+	if (num && dev_info->cb_fn) {
 
 		dropped = 0;
 		nb_cb = dev_info->cb_fn(eth_dev_id, rx_queue_id,
-					ETH_EVENT_BUFFER_SIZE, buf->count, ev,
-					num, dev_info->cb_arg, &dropped);
+					ETH_EVENT_BUFFER_SIZE, buf->count,
+					&buf->events[buf->count], num,
+					dev_info->cb_arg, &dropped);
 		if (unlikely(nb_cb > num))
 			RTE_EDEV_LOG_ERR("Rx CB returned %d (> %d) events",
 				nb_cb, num);
@@ -1124,6 +1226,30 @@ rxa_poll(struct rte_event_eth_rx_adapter *rx_adapter)
 	return nb_rx;
 }
 
+static void
+rxa_vector_expire(struct eth_rx_vector_data *vec, void *arg)
+{
+	struct rte_event_eth_rx_adapter *rx_adapter = arg;
+	struct rte_eth_event_enqueue_buffer *buf =
+		&rx_adapter->event_enqueue_buffer;
+	struct rte_event *ev;
+
+	if (buf->count)
+		rxa_flush_event_buffer(rx_adapter);
+
+	if (vec->vector_ev->nb_elem == 0)
+		return;
+	ev = &buf->events[buf->count];
+
+	/* Event ready. */
+	ev->event = vec->event;
+	ev->vec = vec->vector_ev;
+	buf->count++;
+
+	vec->vector_ev = NULL;
+	vec->ts = 0;
+}
+
 static int
 rxa_service_func(void *args)
 {
@@ -1137,6 +1263,24 @@ rxa_service_func(void *args)
 		return 0;
 	}
 
+	if (rx_adapter->ena_vector) {
+		if ((rte_rdtsc() - rx_adapter->prev_expiry_ts) >=
+		    rx_adapter->vector_tmo_ticks) {
+			struct eth_rx_vector_data *vec;
+
+			TAILQ_FOREACH(vec, &rx_adapter->vector_list, next) {
+				uint64_t elapsed_time = rte_rdtsc() - vec->ts;
+
+				if (elapsed_time >= vec->vector_timeout_ticks) {
+					rxa_vector_expire(vec, rx_adapter);
+					TAILQ_REMOVE(&rx_adapter->vector_list,
+						     vec, next);
+				}
+			}
+			rx_adapter->prev_expiry_ts = rte_rdtsc();
+		}
+	}
+
 	stats = &rx_adapter->stats;
 	stats->rx_packets += rxa_intr_ring_dequeue(rx_adapter);
 	stats->rx_packets += rxa_poll(rx_adapter);
@@ -1640,6 +1784,28 @@ rxa_update_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }
 
+static void
+rxa_set_vector_data(struct eth_rx_queue_info *queue_info, uint16_t vector_count,
+		    uint64_t vector_ns, struct rte_mempool *mp, int32_t qid,
+		    uint16_t port_id)
+{
+#define NSEC2TICK(__ns, __freq) (((__ns) * (__freq)) / 1E9)
+	struct eth_rx_vector_data *vector_data;
+	uint32_t flow_id;
+
+	vector_data = &queue_info->vector_data;
+	vector_data->max_vector_count = vector_count;
+	vector_data->port = port_id;
+	vector_data->queue = qid;
+	vector_data->vector_pool = mp;
+	vector_data->vector_timeout_ticks =
+		NSEC2TICK(vector_ns, rte_get_timer_hz());
+	vector_data->ts = 0;
+	flow_id = queue_info->event & 0xFFFFF;
+	flow_id = flow_id == 0 ? (qid & 0xFF) | (port_id & 0xFFFF) : flow_id;
+	vector_data->event = (queue_info->event & ~0xFFFFF) | flow_id;
+}
+
 static void
 rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
 	struct eth_device_info *dev_info,
@@ -1741,6 +1907,44 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }
 
+static void
+rxa_sw_event_vector_configure(
+	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
+	int rx_queue_id,
+	const struct rte_event_eth_rx_adapter_event_vector_config *config)
+{
+	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
+	struct eth_rx_queue_info *queue_info;
+	struct rte_event *qi_ev;
+
+	if (rx_queue_id == -1) {
+		uint16_t nb_rx_queues;
+		uint16_t i;
+
+		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
+		for (i = 0; i < nb_rx_queues; i++)
+			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
+						      config);
+		return;
+	}
+
+	queue_info = &dev_info->rx_queue[rx_queue_id];
+	qi_ev = (struct rte_event *)&queue_info->event;
+	queue_info->ena_vector = 1;
+	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+	rxa_set_vector_data(queue_info, config->vector_sz,
+			    config->vector_timeout_ns, config->vector_mp,
+			    rx_queue_id, dev_info->dev->data->port_id);
+	rx_adapter->ena_vector = 1;
+	rx_adapter->vector_tmo_ticks =
+		rx_adapter->vector_tmo_ticks ?
+			      RTE_MIN(config->vector_timeout_ns << 1,
+				      rx_adapter->vector_tmo_ticks) :
+			      config->vector_timeout_ns << 1;
+	rx_adapter->prev_expiry_ts = 0;
+	TAILQ_INIT(&rx_adapter->vector_list);
+}
+
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -2081,6 +2285,15 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}
 
+	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
+	    (queue_conf->rx_queue_flags &
+	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
+		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
 		(rx_queue_id != -1)) {
 		RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
@@ -2143,6 +2356,17 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	return 0;
 }
 
+static int
+rxa_sw_vector_limits(struct rte_event_eth_rx_adapter_vector_limits *limits)
+{
+	limits->max_sz = MAX_VECTOR_SIZE;
+	limits->min_sz = MIN_VECTOR_SIZE;
+	limits->max_timeout_ns = MAX_VECTOR_NS;
+	limits->min_timeout_ns = MIN_VECTOR_NS;
+
+	return 0;
+}
+
 int
 rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 				int32_t rx_queue_id)
@@ -2263,6 +2487,121 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }
 
+int
+rte_event_eth_rx_adapter_queue_event_vector_config(
+	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
+	struct rte_event_eth_rx_adapter_event_vector_config *config)
+{
+	struct rte_event_eth_rx_adapter_vector_limits limits;
+	struct rte_event_eth_rx_adapter *rx_adapter;
+	struct rte_eventdev *dev;
+	uint32_t cap;
+	int ret;
+
+	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
+
+	rx_adapter = rxa_id_to_adapter(id);
+	if ((rx_adapter == NULL) || (config == NULL))
+		return -EINVAL;
+
+	dev = &rte_eventdevs[rx_adapter->eventdev_id];
+	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
+						eth_dev_id, &cap);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
+				 "eth port %" PRIu16,
+				 id, eth_dev_id);
+		return ret;
+	}
+
+	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
+		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
+	ret = rte_event_eth_rx_adapter_vector_limits_get(
+		rx_adapter->eventdev_id, eth_dev_id, &limits);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
+				 "eth port %" PRIu16,
+				 rx_adapter->eventdev_id, eth_dev_id);
+		return ret;
+	}
+
+	if (config->vector_sz < limits.min_sz ||
+	    config->vector_sz > limits.max_sz ||
+	    config->vector_timeout_ns < limits.min_timeout_ns ||
+	    config->vector_timeout_ns > limits.max_timeout_ns ||
+	    config->vector_mp == NULL) {
+		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+	if (config->vector_mp->elt_size <
+	    (sizeof(struct rte_event_vector) +
+	     (sizeof(uintptr_t) * config->vector_sz))) {
+		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
+	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
+		RTE_FUNC_PTR_OR_ERR_RET(
+			*dev->dev_ops->eth_rx_adapter_event_vector_config,
+			-ENOTSUP);
+		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
+			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
+	} else {
+		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
+					      rx_queue_id, config);
+	}
+
+	return ret;
+}
+
+int
+rte_event_eth_rx_adapter_vector_limits_get(
+	uint8_t dev_id, uint16_t eth_port_id,
+	struct rte_event_eth_rx_adapter_vector_limits *limits)
+{
+	struct rte_eventdev *dev;
+	uint32_t cap;
+	int ret;
+
+	RTE_EVENTDEV_VALID_DEVID_OR_ERR_RET(dev_id, -EINVAL);
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_port_id, -EINVAL);
+
+	if (limits == NULL)
+		return -EINVAL;
+
+	dev = &rte_eventdevs[dev_id];
+
+	ret = rte_event_eth_rx_adapter_caps_get(dev_id, eth_port_id, &cap);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
+				 "eth port %" PRIu16,
+				 dev_id, eth_port_id);
+		return ret;
+	}
+
+	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
+		RTE_FUNC_PTR_OR_ERR_RET(
+			*dev->dev_ops->eth_rx_adapter_vector_limits_get,
+			-ENOTSUP);
+		ret = dev->dev_ops->eth_rx_adapter_vector_limits_get(
+			dev, &rte_eth_devices[eth_port_id], limits);
+	} else {
+		ret = rxa_sw_vector_limits(limits);
+	}
+
+	return ret;
+}
+
 int
 rte_event_eth_rx_adapter_start(uint8_t id)
 {
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index b57363f80..2e6e367e0 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -122,7 +122,11 @@ rte_event_eth_rx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
 
 	if (caps == NULL)
 		return -EINVAL;
-	*caps = 0;
+
+	if (dev->dev_ops->eth_rx_adapter_caps_get == NULL)
+		*caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
+	else
+		*caps = 0;
 
 	return dev->dev_ops->eth_rx_adapter_caps_get ?
 				(*dev->dev_ops->eth_rx_adapter_caps_get)(dev,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v2 5/8] eventdev: add Tx adapter event vector support
  2021-03-16 15:48 ` [dpdk-dev] [PATCH v2 0/8] " pbhagavatula
                     ` (3 preceding siblings ...)
  2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 4/8] eventdev: add Rx adapter event vector support pbhagavatula
@ 2021-03-16 15:48   ` pbhagavatula
  2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
                     ` (3 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-16 15:48 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support for event eth Tx adapter, the implementation
receives events from the single linked queue and based on
rte_event_vector::union_valid transmits the vector of mbufs to a given
port, queue pair.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 .../rte_event_eth_tx_adapter.c                | 66 ++++++++++++++++---
 lib/librte_eventdev/rte_eventdev.c            |  5 +-
 2 files changed, 60 insertions(+), 11 deletions(-)

diff --git a/lib/librte_eventdev/rte_event_eth_tx_adapter.c b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
index 5b4c42dcf..db260bfb6 100644
--- a/lib/librte_eventdev/rte_event_eth_tx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
@@ -510,6 +510,47 @@ txa_service_buffer_retry(struct rte_mbuf **pkts, uint16_t unsent,
 	stats->tx_dropped += unsent - sent;
 }
 
+static uint16_t
+txa_process_event_vector(struct txa_service_data *txa,
+			 struct rte_event_vector *vec)
+{
+	struct txa_service_queue_info *tqi;
+	uint16_t port, queue, nb_tx = 0;
+	struct rte_mbuf **mbufs;
+	int i;
+
+	mbufs = (struct rte_mbuf **)vec->mbufs;
+	if (vec->attr_valid) {
+		port = vec->port;
+		queue = vec->queue;
+		tqi = txa_service_queue(txa, port, queue);
+		if (unlikely(tqi == NULL || !tqi->added)) {
+			rte_pktmbuf_free_bulk(mbufs, vec->nb_elem);
+			rte_mempool_put(rte_mempool_from_obj(vec), vec);
+			return 0;
+		}
+		for (i = 0; i < vec->nb_elem; i++) {
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
+						   mbufs[i]);
+		}
+	} else {
+		for (i = 0; i < vec->nb_elem; i++) {
+			port = mbufs[i]->port;
+			queue = rte_event_eth_tx_adapter_txq_get(mbufs[i]);
+			tqi = txa_service_queue(txa, port, queue);
+			if (unlikely(tqi == NULL || !tqi->added)) {
+				rte_pktmbuf_free(mbufs[i]);
+				continue;
+			}
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
+						   mbufs[i]);
+		}
+	}
+	rte_mempool_put(rte_mempool_from_obj(vec), vec);
+
+	return nb_tx;
+}
+
 static void
 txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
 	uint32_t n)
@@ -522,22 +563,27 @@ txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
 
 	nb_tx = 0;
 	for (i = 0; i < n; i++) {
-		struct rte_mbuf *m;
 		uint16_t port;
 		uint16_t queue;
 		struct txa_service_queue_info *tqi;
 
-		m = ev[i].mbuf;
-		port = m->port;
-		queue = rte_event_eth_tx_adapter_txq_get(m);
+		if (!(ev[i].event_type & RTE_EVENT_TYPE_VECTOR)) {
+			struct rte_mbuf *m;
 
-		tqi = txa_service_queue(txa, port, queue);
-		if (unlikely(tqi == NULL || !tqi->added)) {
-			rte_pktmbuf_free(m);
-			continue;
-		}
+			m = ev[i].mbuf;
+			port = m->port;
+			queue = rte_event_eth_tx_adapter_txq_get(m);
 
-		nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
+			tqi = txa_service_queue(txa, port, queue);
+			if (unlikely(tqi == NULL || !tqi->added)) {
+				rte_pktmbuf_free(m);
+				continue;
+			}
+
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
+		} else {
+			nb_tx += txa_process_event_vector(txa, ev[i].vec);
+		}
 	}
 
 	stats->tx_packets += nb_tx;
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index 2e6e367e0..b2e85ce67 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -196,7 +196,10 @@ rte_event_eth_tx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
 	if (caps == NULL)
 		return -EINVAL;
 
-	*caps = 0;
+	if (dev->dev_ops->eth_tx_adapter_caps_get == NULL)
+		*caps = RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR;
+	else
+		*caps = 0;
 
 	return dev->dev_ops->eth_tx_adapter_caps_get ?
 			(*dev->dev_ops->eth_tx_adapter_caps_get)(dev,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v2 6/8] app/eventdev: add event vector mode in pipeline test
  2021-03-16 15:48 ` [dpdk-dev] [PATCH v2 0/8] " pbhagavatula
                     ` (4 preceding siblings ...)
  2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 5/8] eventdev: add Tx " pbhagavatula
@ 2021-03-16 15:48   ` pbhagavatula
  2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 7/8] doc: announce event Rx adapter config changes pbhagavatula
                     ` (2 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-16 15:48 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support in pipeline tests. By default this mode
is disabled, it can be enabled by using the option --enable_vector.
example:
	dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
	--nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
	--wlcores=20-23  --enable_vector

Addtional options to configure vector size and vector timeout are
also implemented and can be used by specifying --vector_size and
--vector_tmo_ns

This patch also adds a new option to set the number of Rx queues
configured per event eth rx adapter.
example:
	dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
	--nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
	--wlcores=20-23  --nb_eth_queues 4

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 app/test-eventdev/evt_common.h           |   4 +
 app/test-eventdev/evt_options.c          |  52 ++++
 app/test-eventdev/evt_options.h          |   4 +
 app/test-eventdev/test_pipeline_atq.c    | 310 ++++++++++++++++++++--
 app/test-eventdev/test_pipeline_common.c | 113 +++++++-
 app/test-eventdev/test_pipeline_common.h |  18 ++
 app/test-eventdev/test_pipeline_queue.c  | 320 +++++++++++++++++++++--
 doc/guides/tools/testeventdev.rst        |  28 ++
 8 files changed, 795 insertions(+), 54 deletions(-)

diff --git a/app/test-eventdev/evt_common.h b/app/test-eventdev/evt_common.h
index a1da1cf11..0e228258e 100644
--- a/app/test-eventdev/evt_common.h
+++ b/app/test-eventdev/evt_common.h
@@ -58,16 +58,20 @@ struct evt_options {
 	uint8_t sched_type_list[EVT_MAX_STAGES];
 	uint16_t mbuf_sz;
 	uint16_t wkr_deq_dep;
+	uint16_t vector_size;
+	uint16_t eth_queues;
 	uint32_t nb_flows;
 	uint32_t tx_first;
 	uint32_t max_pkt_sz;
 	uint32_t deq_tmo_nsec;
 	uint32_t q_priority:1;
 	uint32_t fwd_latency:1;
+	uint32_t ena_vector : 1;
 	uint64_t nb_pkts;
 	uint64_t nb_timers;
 	uint64_t expiry_nsec;
 	uint64_t max_tmo_nsec;
+	uint64_t vector_tmo_nsec;
 	uint64_t timer_tick_nsec;
 	uint64_t optm_timer_tick_nsec;
 	enum evt_prod_type prod_type;
diff --git a/app/test-eventdev/evt_options.c b/app/test-eventdev/evt_options.c
index 0d04ea9f8..0d5540574 100644
--- a/app/test-eventdev/evt_options.c
+++ b/app/test-eventdev/evt_options.c
@@ -34,6 +34,9 @@ evt_options_default(struct evt_options *opt)
 	opt->max_tmo_nsec = 1E5;  /* 100000ns ~100us */
 	opt->expiry_nsec = 1E4;   /* 10000ns ~10us */
 	opt->prod_type = EVT_PROD_TYPE_SYNT;
+	opt->eth_queues = 1;
+	opt->vector_size = 64;
+	opt->vector_tmo_nsec = 100E3;
 }

 typedef int (*option_parser_t)(struct evt_options *opt,
@@ -257,6 +260,43 @@ evt_parse_max_pkt_sz(struct evt_options *opt, const char *arg)
 	return ret;
 }

+static int
+evt_parse_ena_vector(struct evt_options *opt, const char *arg __rte_unused)
+{
+	opt->ena_vector = 1;
+	return 0;
+}
+
+static int
+evt_parse_vector_size(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint16(&(opt->vector_size), arg);
+
+	return ret;
+}
+
+static int
+evt_parse_vector_tmo_ns(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint64(&(opt->vector_tmo_nsec), arg);
+
+	return ret;
+}
+
+static int
+evt_parse_eth_queues(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint16(&(opt->eth_queues), arg);
+
+	return ret;
+}
+
 static void
 usage(char *program)
 {
@@ -289,6 +329,10 @@ usage(char *program)
 		"\t--expiry_nsec      : event timer expiry ns.\n"
 		"\t--mbuf_sz          : packet mbuf size.\n"
 		"\t--max_pkt_sz       : max packet size.\n"
+		"\t--nb_eth_queues    : number of ethernet Rx queues.\n"
+		"\t--enable_vector    : enable event vectorization.\n"
+		"\t--vector_size      : Max vector size.\n"
+		"\t--vector_tmo_ns    : Max vector timeout in nanoseconds\n"
 		);
 	printf("available tests:\n");
 	evt_test_dump_names();
@@ -360,6 +404,10 @@ static struct option lgopts[] = {
 	{ EVT_EXPIRY_NSEC,         1, 0, 0 },
 	{ EVT_MBUF_SZ,             1, 0, 0 },
 	{ EVT_MAX_PKT_SZ,          1, 0, 0 },
+	{ EVT_NB_ETH_QUEUES,       1, 0, 0 },
+	{ EVT_ENA_VECTOR,          0, 0, 0 },
+	{ EVT_VECTOR_SZ,           1, 0, 0 },
+	{ EVT_VECTOR_TMO,          1, 0, 0 },
 	{ EVT_HELP,                0, 0, 0 },
 	{ NULL,                    0, 0, 0 }
 };
@@ -394,6 +442,10 @@ evt_opts_parse_long(int opt_idx, struct evt_options *opt)
 		{ EVT_EXPIRY_NSEC, evt_parse_expiry_nsec},
 		{ EVT_MBUF_SZ, evt_parse_mbuf_sz},
 		{ EVT_MAX_PKT_SZ, evt_parse_max_pkt_sz},
+		{ EVT_NB_ETH_QUEUES, evt_parse_eth_queues},
+		{ EVT_ENA_VECTOR, evt_parse_ena_vector},
+		{ EVT_VECTOR_SZ, evt_parse_vector_size},
+		{ EVT_VECTOR_TMO, evt_parse_vector_tmo_ns},
 	};

 	for (i = 0; i < RTE_DIM(parsermap); i++) {
diff --git a/app/test-eventdev/evt_options.h b/app/test-eventdev/evt_options.h
index 748e54fae..1cea2a3e1 100644
--- a/app/test-eventdev/evt_options.h
+++ b/app/test-eventdev/evt_options.h
@@ -42,6 +42,10 @@
 #define EVT_EXPIRY_NSEC          ("expiry_nsec")
 #define EVT_MBUF_SZ              ("mbuf_sz")
 #define EVT_MAX_PKT_SZ           ("max_pkt_sz")
+#define EVT_NB_ETH_QUEUES        ("nb_eth_queues")
+#define EVT_ENA_VECTOR           ("enable_vector")
+#define EVT_VECTOR_SZ            ("vector_size")
+#define EVT_VECTOR_TMO           ("vector_tmo_ns")
 #define EVT_HELP                 ("help")

 void evt_options_default(struct evt_options *opt);
diff --git a/app/test-eventdev/test_pipeline_atq.c b/app/test-eventdev/test_pipeline_atq.c
index 0872b25b5..84dd4f44e 100644
--- a/app/test-eventdev/test_pipeline_atq.c
+++ b/app/test-eventdev/test_pipeline_atq.c
@@ -15,6 +15,8 @@ pipeline_atq_nb_event_queues(struct evt_options *opt)
 	return rte_eth_dev_count_avail();
 }

+typedef int (*pipeline_atq_worker_t)(void *arg);
+
 static __rte_noinline int
 pipeline_atq_worker_single_stage_tx(void *arg)
 {
@@ -113,6 +115,112 @@ pipeline_atq_worker_single_stage_burst_fwd(void *arg)
 	return 0;
 }

+static __rte_noinline int
+pipeline_atq_worker_single_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+		vector_sz = ev.vec->nb_elem;
+		pipeline_event_tx_vector(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = ev.vec->nb_elem;
+		ev.queue_id = tx_queue[ev.vec->port];
+		ev.vec->queue = 0;
+		pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			vector_sz += ev[i].vec->nb_elem;
+			ev[i].vec->queue = 0;
+		}
+
+		pipeline_event_tx_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			ev[i].queue_id = tx_queue[ev[i].vec->port];
+			ev[i].vec->queue = 0;
+			vector_sz += ev[i].vec->nb_elem;
+			pipeline_fwd_event_vector(&ev[i],
+						  RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
 static __rte_noinline int
 pipeline_atq_worker_multi_stage_tx(void *arg)
 {
@@ -245,6 +353,147 @@ pipeline_atq_worker_multi_stage_burst_fwd(void *arg)
 	return 0;
 }

+static __rte_noinline int
+pipeline_atq_worker_multi_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.sub_event_type % nb_stages;
+
+		if (cq_id == last_queue) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+			continue;
+		}
+
+		ev.sub_event_type++;
+		pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.sub_event_type % nb_stages;
+
+		if (cq_id == last_queue) {
+			ev.queue_id = tx_queue[ev.vec->port];
+			ev.vec->queue = 0;
+			vector_sz = ev.vec->nb_elem;
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			pipeline_event_enqueue(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.sub_event_type++;
+			pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+			pipeline_event_enqueue(dev, port, &ev);
+		}
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].sub_event_type % nb_stages;
+
+			if (cq_id == last_queue) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+				continue;
+			}
+
+			ev[i].sub_event_type++;
+			pipeline_fwd_event_vector(&ev[i],
+						  sched_type_list[cq_id]);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].sub_event_type % nb_stages;
+
+			if (cq_id == last_queue) {
+				vector_sz = ev[i].vec->nb_elem;
+				ev[i].queue_id = tx_queue[ev[i].vec->port];
+				ev[i].vec->queue = 0;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].sub_event_type++;
+				pipeline_fwd_event_vector(
+					&ev[i], sched_type_list[cq_id]);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
 static int
 worker_wrapper(void *arg)
 {
@@ -253,27 +502,36 @@ worker_wrapper(void *arg)
 	const bool burst = evt_has_burst_mode(w->dev_id);
 	const bool internal_port = w->t->internal_port;
 	const uint8_t nb_stages = opt->nb_stages;
-	RTE_SET_USED(opt);
+	/*vector/burst/internal_port*/
+	const pipeline_atq_worker_t
+	pipeline_atq_worker_single_stage[2][2][2] = {
+		[0][0][0] = pipeline_atq_worker_single_stage_fwd,
+		[0][0][1] = pipeline_atq_worker_single_stage_tx,
+		[0][1][0] = pipeline_atq_worker_single_stage_burst_fwd,
+		[0][1][1] = pipeline_atq_worker_single_stage_burst_tx,
+		[1][0][0] = pipeline_atq_worker_single_stage_fwd_vector,
+		[1][0][1] = pipeline_atq_worker_single_stage_tx_vector,
+		[1][1][0] = pipeline_atq_worker_single_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_atq_worker_single_stage_burst_tx_vector,
+	};
+	const pipeline_atq_worker_t
+	pipeline_atq_worker_multi_stage[2][2][2] = {
+		[0][0][0] = pipeline_atq_worker_multi_stage_fwd,
+		[0][0][1] = pipeline_atq_worker_multi_stage_tx,
+		[0][1][0] = pipeline_atq_worker_multi_stage_burst_fwd,
+		[0][1][1] = pipeline_atq_worker_multi_stage_burst_tx,
+		[1][0][0] = pipeline_atq_worker_multi_stage_fwd_vector,
+		[1][0][1] = pipeline_atq_worker_multi_stage_tx_vector,
+		[1][1][0] = pipeline_atq_worker_multi_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_atq_worker_multi_stage_burst_tx_vector,
+	};

-	if (nb_stages == 1) {
-		if (!burst && internal_port)
-			return pipeline_atq_worker_single_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_atq_worker_single_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_atq_worker_single_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_atq_worker_single_stage_burst_fwd(arg);
-	} else {
-		if (!burst && internal_port)
-			return pipeline_atq_worker_multi_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_atq_worker_multi_stage_fwd(arg);
-		if (burst && internal_port)
-			return pipeline_atq_worker_multi_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_atq_worker_multi_stage_burst_fwd(arg);
-	}
+	if (nb_stages == 1)
+		return (pipeline_atq_worker_single_stage[opt->ena_vector][burst]
+							[internal_port])(arg);
+	else
+		return (pipeline_atq_worker_multi_stage[opt->ena_vector][burst]
+						       [internal_port])(arg);

 	rte_panic("invalid worker\n");
 }
@@ -290,7 +548,7 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
 	int ret;
 	int nb_ports;
 	int nb_queues;
-	uint8_t queue;
+	uint8_t queue, is_prod;
 	uint8_t tx_evqueue_id[RTE_MAX_ETHPORTS];
 	uint8_t queue_arr[RTE_EVENT_MAX_QUEUES_PER_DEV];
 	uint8_t nb_worker_queues = 0;
@@ -330,15 +588,19 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
 		q_conf.event_queue_cfg = RTE_EVENT_QUEUE_CFG_ALL_TYPES;

 		if (!t->internal_port) {
+			is_prod = false;
 			RTE_ETH_FOREACH_DEV(prod) {
 				if (queue == tx_evqueue_id[prod]) {
 					q_conf.event_queue_cfg =
 						RTE_EVENT_QUEUE_CFG_SINGLE_LINK;
-				} else {
-					queue_arr[nb_worker_queues] = queue;
-					nb_worker_queues++;
+					is_prod = true;
+					break;
 				}
 			}
+			if (!is_prod) {
+				queue_arr[nb_worker_queues] = queue;
+				nb_worker_queues++;
+			}
 		}

 		ret = rte_event_queue_setup(opt->dev_id, queue, &q_conf);
diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index b47d76743..d5ef90500 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -36,6 +36,12 @@ pipeline_opt_dump(struct evt_options *opt, uint8_t nb_queues)
 	evt_dump_queue_priority(opt);
 	evt_dump_sched_type_list(opt);
 	evt_dump_producer_type(opt);
+	evt_dump("nb_eth_rx_queues", "%d", opt->eth_queues);
+	evt_dump("event_vector", "%d", opt->ena_vector);
+	if (opt->ena_vector) {
+		evt_dump("vector_size", "%d", opt->vector_size);
+		evt_dump("vector_tmo_ns", "%" PRIu64 "", opt->vector_tmo_nsec);
+	}
 }

 static inline uint64_t
@@ -163,7 +169,7 @@ pipeline_opt_check(struct evt_options *opt, uint64_t nb_queues)
 int
 pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 {
-	uint16_t i;
+	uint16_t i, j;
 	int ret;
 	uint8_t nb_queues = 1;
 	struct test_pipeline *t = evt_test_priv(test);
@@ -210,6 +216,16 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 		if (!(caps & RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT))
 			t->internal_port = 0;

+		ret = rte_event_eth_rx_adapter_caps_get(opt->dev_id, i, &caps);
+		if (ret != 0) {
+			evt_err("failed to get event tx adapter[%d] caps", i);
+			return ret;
+		}
+
+		if (!(caps & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT))
+			local_port_conf.rxmode.offloads |=
+				DEV_RX_OFFLOAD_RSS_HASH;
+
 		ret = rte_eth_dev_info_get(i, &dev_info);
 		if (ret != 0) {
 			evt_err("Error during getting device (port %u) info: %s\n",
@@ -236,19 +252,22 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 				local_port_conf.rx_adv_conf.rss_conf.rss_hf);
 		}

-		if (rte_eth_dev_configure(i, nb_queues, nb_queues,
-					&local_port_conf)
-				< 0) {
+		if (rte_eth_dev_configure(i, opt->eth_queues, nb_queues,
+					  &local_port_conf) < 0) {
 			evt_err("Failed to configure eth port [%d]", i);
 			return -EINVAL;
 		}

-		if (rte_eth_rx_queue_setup(i, 0, NB_RX_DESC,
-				rte_socket_id(), &rx_conf, t->pool) < 0) {
-			evt_err("Failed to setup eth port [%d] rx_queue: %d.",
+		for (j = 0; j < opt->eth_queues; j++) {
+			if (rte_eth_rx_queue_setup(i, j, NB_RX_DESC,
+						   rte_socket_id(), &rx_conf,
+						   t->pool) < 0) {
+				evt_err("Failed to setup eth port [%d] rx_queue: %d.",
 					i, 0);
-			return -EINVAL;
+				return -EINVAL;
+			}
 		}
+
 		if (rte_eth_tx_queue_setup(i, 0, NB_TX_DESC,
 					rte_socket_id(), NULL) < 0) {
 			evt_err("Failed to setup eth port [%d] tx_queue: %d.",
@@ -310,12 +329,27 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 {
 	int ret = 0;
 	uint16_t prod;
+	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
+	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
 	queue_conf.ev.sched_type = opt->sched_type_list[0];
+	if (opt->ena_vector) {
+		unsigned int nb_elem = (opt->pool_sz / opt->vector_size) << 1;
+
+		nb_elem = nb_elem ? nb_elem : 1;
+		vector_pool = rte_event_vector_pool_create(
+			"vector_pool", nb_elem, 0, opt->vector_size,
+			opt->socket_id);
+		if (vector_pool == NULL) {
+			evt_err("failed to create event vector pool");
+			return -ENOMEM;
+		}
+	}
 	RTE_ETH_FOREACH_DEV(prod) {
+		struct rte_event_eth_rx_adapter_vector_limits limits;
 		uint32_t cap;

 		ret = rte_event_eth_rx_adapter_caps_get(opt->dev_id,
@@ -326,6 +360,50 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 					opt->dev_id);
 			return ret;
 		}
+
+		if (opt->ena_vector) {
+			memset(&limits, 0, sizeof(limits));
+			ret = rte_event_eth_rx_adapter_vector_limits_get(
+				opt->dev_id, prod, &limits);
+			if (ret) {
+				evt_err("failed to get vector limits");
+				return ret;
+			}
+
+			if (opt->vector_size < limits.min_sz ||
+			    opt->vector_size > limits.max_sz) {
+				evt_err("Vector size [%d] not within limits min[%d] max[%d]",
+					opt->vector_size, limits.min_sz,
+					limits.max_sz);
+				return -EINVAL;
+			}
+
+			if (limits.log2_sz &&
+			    !rte_is_power_of_2(opt->vector_size)) {
+				evt_err("Vector size [%d] not power of 2",
+					opt->vector_size);
+				return -EINVAL;
+			}
+
+			if (opt->vector_tmo_nsec > limits.max_timeout_ns ||
+			    opt->vector_tmo_nsec < limits.min_timeout_ns) {
+				evt_err("Vector timeout [%" PRIu64
+					"] not within limits min[%" PRIu64
+					"] max[%" PRIu64 "]",
+					opt->vector_tmo_nsec,
+					limits.min_timeout_ns,
+					limits.max_timeout_ns);
+				return -EINVAL;
+			}
+
+			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.rx_queue_flags |=
+				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+			} else {
+				evt_err("Rx adapter doesn't support event vector");
+				return -EINVAL;
+			}
+		}
 		queue_conf.ev.queue_id = prod * stride;
 		ret = rte_event_eth_rx_adapter_create(prod, opt->dev_id,
 				&prod_conf);
@@ -340,6 +418,17 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}

+		if (opt->ena_vector) {
+			vec_conf.vector_sz = opt->vector_size;
+			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
+			vec_conf.vector_mp = vector_pool;
+			if (rte_event_eth_rx_adapter_queue_event_vector_config(
+				    prod, prod, -1, &vec_conf) < 0) {
+				evt_err("Failed to configure event vectorization for Rx adapter");
+				return -EINVAL;
+			}
+		}
+
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;

@@ -378,6 +467,14 @@ pipeline_event_tx_adapter_setup(struct evt_options *opt,
 			return ret;
 		}

+		if (opt->ena_vector) {
+			if (!(cap &
+			      RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR)) {
+				evt_err("Tx adapter doesn't support event vector");
+				return -EINVAL;
+			}
+		}
+
 		ret = rte_event_eth_tx_adapter_create(consm, opt->dev_id,
 				&port_conf);
 		if (ret) {
diff --git a/app/test-eventdev/test_pipeline_common.h b/app/test-eventdev/test_pipeline_common.h
index 6e73c6ab2..800a90616 100644
--- a/app/test-eventdev/test_pipeline_common.h
+++ b/app/test-eventdev/test_pipeline_common.h
@@ -101,6 +101,14 @@ pipeline_fwd_event(struct rte_event *ev, uint8_t sched)
 	ev->sched_type = sched;
 }

+static __rte_always_inline void
+pipeline_fwd_event_vector(struct rte_event *ev, uint8_t sched)
+{
+	ev->event_type = RTE_EVENT_TYPE_CPU_VECTOR;
+	ev->op = RTE_EVENT_OP_FORWARD;
+	ev->sched_type = sched;
+}
+
 static __rte_always_inline void
 pipeline_event_tx(const uint8_t dev, const uint8_t port,
 		struct rte_event * const ev)
@@ -110,6 +118,16 @@ pipeline_event_tx(const uint8_t dev, const uint8_t port,
 		rte_pause();
 }

+static __rte_always_inline void
+pipeline_event_tx_vector(const uint8_t dev, const uint8_t port,
+			 struct rte_event *const ev)
+{
+	ev->vec->queue = 0;
+
+	while (!rte_event_eth_tx_adapter_enqueue(dev, port, ev, 1, 0))
+		rte_pause();
+}
+
 static __rte_always_inline void
 pipeline_event_tx_burst(const uint8_t dev, const uint8_t port,
 		struct rte_event *ev, const uint16_t nb_rx)
diff --git a/app/test-eventdev/test_pipeline_queue.c b/app/test-eventdev/test_pipeline_queue.c
index 9a9febb19..f6cc3e358 100644
--- a/app/test-eventdev/test_pipeline_queue.c
+++ b/app/test-eventdev/test_pipeline_queue.c
@@ -15,6 +15,8 @@ pipeline_queue_nb_event_queues(struct evt_options *opt)
 	return (eth_count * opt->nb_stages) + eth_count;
 }

+typedef int (*pipeline_queue_worker_t)(void *arg);
+
 static __rte_noinline int
 pipeline_queue_worker_single_stage_tx(void *arg)
 {
@@ -126,6 +128,125 @@ pipeline_queue_worker_single_stage_burst_fwd(void *arg)
 	return 0;
 }

+static __rte_noinline int
+pipeline_queue_worker_single_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		if (ev.sched_type == RTE_SCHED_TYPE_ATOMIC) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.queue_id++;
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			pipeline_event_enqueue(dev, port, &ev);
+		}
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		ev.queue_id = tx_queue[ev.vec->port];
+		ev.vec->queue = 0;
+		vector_sz = ev.vec->nb_elem;
+		pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			if (ev[i].sched_type == RTE_SCHED_TYPE_ATOMIC) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].queue_id++;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			ev[i].queue_id = tx_queue[ev[i].vec->port];
+			ev[i].vec->queue = 0;
+			vector_sz += ev[i].vec->nb_elem;
+			pipeline_fwd_event_vector(&ev[i],
+						  RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}

 static __rte_noinline int
 pipeline_queue_worker_multi_stage_tx(void *arg)
@@ -267,6 +388,151 @@ pipeline_queue_worker_multi_stage_burst_fwd(void *arg)
 	return 0;
 }

+static __rte_noinline int
+pipeline_queue_worker_multi_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.queue_id % nb_stages;
+
+		if (ev.queue_id == tx_queue[ev.vec->port]) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+			continue;
+		}
+
+		ev.queue_id++;
+		pipeline_fwd_event_vector(&ev, cq_id != last_queue
+						       ? sched_type_list[cq_id]
+						       : RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.queue_id % nb_stages;
+
+		if (cq_id == last_queue) {
+			vector_sz = ev.vec->nb_elem;
+			ev.queue_id = tx_queue[ev.vec->port];
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.queue_id++;
+			pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+		}
+
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].queue_id % nb_stages;
+
+			if (ev[i].queue_id == tx_queue[ev[i].vec->port]) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+				continue;
+			}
+
+			ev[i].queue_id++;
+			pipeline_fwd_event_vector(
+				&ev[i], cq_id != last_queue
+						? sched_type_list[cq_id]
+						: RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].queue_id % nb_stages;
+
+			if (cq_id == last_queue) {
+				ev[i].queue_id = tx_queue[ev[i].vec->port];
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].queue_id++;
+				pipeline_fwd_event_vector(
+					&ev[i], sched_type_list[cq_id]);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
 static int
 worker_wrapper(void *arg)
 {
@@ -275,29 +541,39 @@ worker_wrapper(void *arg)
 	const bool burst = evt_has_burst_mode(w->dev_id);
 	const bool internal_port = w->t->internal_port;
 	const uint8_t nb_stages = opt->nb_stages;
-	RTE_SET_USED(opt);
-
-	if (nb_stages == 1) {
-		if (!burst && internal_port)
-			return pipeline_queue_worker_single_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_queue_worker_single_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_queue_worker_single_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_queue_worker_single_stage_burst_fwd(
-					arg);
-	} else {
-		if (!burst && internal_port)
-			return pipeline_queue_worker_multi_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_queue_worker_multi_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_queue_worker_multi_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_queue_worker_multi_stage_burst_fwd(arg);
+	/*vector/burst/internal_port*/
+	const pipeline_queue_worker_t
+	pipeline_queue_worker_single_stage[2][2][2] = {
+		[0][0][0] = pipeline_queue_worker_single_stage_fwd,
+		[0][0][1] = pipeline_queue_worker_single_stage_tx,
+		[0][1][0] = pipeline_queue_worker_single_stage_burst_fwd,
+		[0][1][1] = pipeline_queue_worker_single_stage_burst_tx,
+		[1][0][0] = pipeline_queue_worker_single_stage_fwd_vector,
+		[1][0][1] = pipeline_queue_worker_single_stage_tx_vector,
+		[1][1][0] = pipeline_queue_worker_single_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_queue_worker_single_stage_burst_tx_vector,
+	};
+	const pipeline_queue_worker_t
+	pipeline_queue_worker_multi_stage[2][2][2] = {
+		[0][0][0] = pipeline_queue_worker_multi_stage_fwd,
+		[0][0][1] = pipeline_queue_worker_multi_stage_tx,
+		[0][1][0] = pipeline_queue_worker_multi_stage_burst_fwd,
+		[0][1][1] = pipeline_queue_worker_multi_stage_burst_tx,
+		[1][0][0] = pipeline_queue_worker_multi_stage_fwd_vector,
+		[1][0][1] = pipeline_queue_worker_multi_stage_tx_vector,
+		[1][1][0] = pipeline_queue_worker_multi_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_queue_worker_multi_stage_burst_tx_vector,
+	};
+
+	if (nb_stages == 1)
+		return (pipeline_queue_worker_single_stage[opt->ena_vector]
+							  [burst]
+							  [internal_port])(arg);
+	else
+		return (pipeline_queue_worker_multi_stage[opt->ena_vector]
+							 [burst]
+							 [internal_port])(arg);

-	}
 	rte_panic("invalid worker\n");
 }

diff --git a/doc/guides/tools/testeventdev.rst b/doc/guides/tools/testeventdev.rst
index ad1788a3d..691cf706e 100644
--- a/doc/guides/tools/testeventdev.rst
+++ b/doc/guides/tools/testeventdev.rst
@@ -158,6 +158,26 @@ The following are the application command-line options:
        Set max packet mbuf size. Can be used configure Rx/Tx scatter gather.
        Only applicable for `pipeline_atq` and `pipeline_queue` tests.

+* ``--nb_eth_queues``
+
+       Configure multiple Rx queues per each ethernet port.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--enable_vector``
+
+       Enable event vector for Rx/Tx adapters.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--vector_size``
+
+       Vector size to configure for the Rx adapter.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--vector_tmo_ns``
+
+       Vector timeout nanoseconds to be configured for the Rx adapter.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+

 Eventdev Tests
 --------------
@@ -607,6 +627,10 @@ Supported application command line options are following::
         --worker_deq_depth
         --prod_type_ethdev
         --deq_tmo_nsec
+        --nb_eth_queues
+        --enable_vector
+        --vector_size
+        --vector_tmo_ns


 .. Note::
@@ -699,6 +723,10 @@ Supported application command line options are following::
         --worker_deq_depth
         --prod_type_ethdev
         --deq_tmo_nsec
+        --nb_eth_queues
+        --enable_vector
+        --vector_size
+        --vector_tmo_ns


 .. Note::
--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v2 7/8] doc: announce event Rx adapter config changes
  2021-03-16 15:48 ` [dpdk-dev] [PATCH v2 0/8] " pbhagavatula
                     ` (5 preceding siblings ...)
  2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
@ 2021-03-16 15:48   ` pbhagavatula
  2021-03-16 15:48   ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v2 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
  2021-03-16 20:01   ` [dpdk-dev] [PATCH v3 0/8] Introduce event vectorization pbhagavatula
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-16 15:48 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

The Rx adapter event vector configuration will be merged into
Rx adapter queue configuration to simplify enabling event
vectorization.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 doc/guides/rel_notes/deprecation.rst | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst
index 64629e064..ef6d1bb18 100644
--- a/doc/guides/rel_notes/deprecation.rst
+++ b/doc/guides/rel_notes/deprecation.rst
@@ -130,3 +130,12 @@ Deprecation Notices
 * cmdline: ``cmdline`` structure will be made opaque to hide platform-specific
   content. On Linux and FreeBSD, supported prior to DPDK 20.11,
   original structure will be kept until DPDK 21.11.
+
+* eventdev: The structure ``rte_event_eth_rx_adapter_queue_conf`` will be
+  extended to include ``rte_event_eth_rx_adapter_event_vector_config`` elements
+  and the function ``rte_event_eth_rx_adapter_queue_event_vector_config`` will
+  be removed in DPDK 21.11.
+
+  An application can enable event vectorization by passing the desired vector
+  values to the function ``rte_event_eth_rx_adapter_queue_add`` using
+  the structure ``rte_event_eth_rx_adapter_queue_add``.
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [dpdk-dev v21.11] [PATCH v2 8/8] eventdev: simplify Rx adapter event vector config
  2021-03-16 15:48 ` [dpdk-dev] [PATCH v2 0/8] " pbhagavatula
                     ` (6 preceding siblings ...)
  2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 7/8] doc: announce event Rx adapter config changes pbhagavatula
@ 2021-03-16 15:48   ` pbhagavatula
  2021-03-16 20:01   ` [dpdk-dev] [PATCH v3 0/8] Introduce event vectorization pbhagavatula
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-16 15:48 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Include vector configuration into the structure
``rte_event_eth_rx_adapter_queue_conf`` used when configuring rest
of the Rx adapter ethernet device Rx queue parameters.
This simplifies event vector configuration as it avoids splitting
configuration per Rx queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 app/test-eventdev/test_pipeline_common.c      |  16 +-
 lib/librte_eventdev/eventdev_pmd.h            |  29 ---
 .../rte_event_eth_rx_adapter.c                | 168 ++++++------------
 .../rte_event_eth_rx_adapter.h                |  27 ---
 lib/librte_eventdev/version.map               |   1 -
 5 files changed, 57 insertions(+), 184 deletions(-)

diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index d5ef90500..76aee254b 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -331,7 +331,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 	uint16_t prod;
 	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
-	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
@@ -397,8 +396,12 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			}

 			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.vector_sz = opt->vector_size;
+				queue_conf.vector_timeout_ns =
+					opt->vector_tmo_nsec;
 				queue_conf.rx_queue_flags |=
 				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+				queue_conf.vector_mp = vector_pool;
 			} else {
 				evt_err("Rx adapter doesn't support event vector");
 				return -EINVAL;
@@ -418,17 +421,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}

-		if (opt->ena_vector) {
-			vec_conf.vector_sz = opt->vector_size;
-			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
-			vec_conf.vector_mp = vector_pool;
-			if (rte_event_eth_rx_adapter_queue_event_vector_config(
-				    prod, prod, -1, &vec_conf) < 0) {
-				evt_err("Failed to configure event vectorization for Rx adapter");
-				return -EINVAL;
-			}
-		}
-
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;

diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index d79dfd612..60bfaebc0 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -667,32 +667,6 @@ typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
 	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-struct rte_event_eth_rx_adapter_event_vector_config;
-/**
- * Enable event vector on an given Rx queue of a ethernet devices belonging to
- * the Rx adapter.
- *
- * @param dev
- *   Event device pointer
- *
- * @param eth_dev
- *   Ethernet device pointer
- *
- * @param rx_queue_id
- *   The Rx queue identifier
- *
- * @param config
- *   Pointer to the event vector configuration structure.
- *
- * @return
- *   - 0: Success.
- *   - <0: Error code returned by the driver function.
- */
-typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
-	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
-	int32_t rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;

@@ -1118,9 +1092,6 @@ struct rte_eventdev_ops {
 	eventdev_eth_rx_adapter_vector_limits_get_t
 		eth_rx_adapter_vector_limits_get;
 	/**< Get event vector limits for the Rx adapter */
-	eventdev_eth_rx_adapter_event_vector_config_t
-		eth_rx_adapter_event_vector_config;
-	/**< Configure Rx adapter with event vector */

 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index c71990078..a1990637f 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -1882,6 +1882,25 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	} else
 		qi_ev->flow_id = 0;

+	if (conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+		queue_info->ena_vector = 1;
+		qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+		rxa_set_vector_data(queue_info, conf->vector_sz,
+				    conf->vector_timeout_ns, conf->vector_mp,
+				    rx_queue_id, dev_info->dev->data->port_id);
+		rx_adapter->ena_vector = 1;
+		rx_adapter->vector_tmo_ticks =
+			rx_adapter->vector_tmo_ticks
+				? RTE_MIN(queue_info->vector_data
+						  .vector_timeout_ticks,
+					  rx_adapter->vector_tmo_ticks)
+				: queue_info->vector_data.vector_timeout_ticks;
+		rx_adapter->vector_tmo_ticks <<= 1;
+		TAILQ_INIT(&rx_adapter->vector_list);
+		rx_adapter->prev_expiry_ts = 0;
+	}
+
 	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
 	if (rxa_polled_queue(dev_info, rx_queue_id)) {
 		rx_adapter->num_rx_polled += !pollq;
@@ -1907,44 +1926,6 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }

-static void
-rxa_sw_event_vector_configure(
-	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
-	int rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
-	struct eth_rx_queue_info *queue_info;
-	struct rte_event *qi_ev;
-
-	if (rx_queue_id == -1) {
-		uint16_t nb_rx_queues;
-		uint16_t i;
-
-		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
-		for (i = 0; i < nb_rx_queues; i++)
-			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
-						      config);
-		return;
-	}
-
-	queue_info = &dev_info->rx_queue[rx_queue_id];
-	qi_ev = (struct rte_event *)&queue_info->event;
-	queue_info->ena_vector = 1;
-	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
-	rxa_set_vector_data(queue_info, config->vector_sz,
-			    config->vector_timeout_ns, config->vector_mp,
-			    rx_queue_id, dev_info->dev->data->port_id);
-	rx_adapter->ena_vector = 1;
-	rx_adapter->vector_tmo_ticks =
-		rx_adapter->vector_tmo_ticks ?
-			      RTE_MIN(config->vector_timeout_ns << 1,
-				      rx_adapter->vector_tmo_ticks) :
-			      config->vector_timeout_ns << 1;
-	rx_adapter->prev_expiry_ts = 0;
-	TAILQ_INIT(&rx_adapter->vector_list);
-}
-
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -2258,6 +2239,7 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	struct rte_event_eth_rx_adapter *rx_adapter;
 	struct rte_eventdev *dev;
 	struct eth_device_info *dev_info;
+	struct rte_event_eth_rx_adapter_vector_limits limits;

 	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
 	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
@@ -2294,6 +2276,39 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}

+	if (queue_conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+		ret = rte_event_eth_rx_adapter_vector_limits_get(
+			rx_adapter->eventdev_id, eth_dev_id, &limits);
+		if (ret < 0) {
+			RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_sz < limits.min_sz ||
+		    queue_conf->vector_sz > limits.max_sz ||
+		    queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
+		    queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
+		    queue_conf->vector_mp == NULL) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_mp->elt_size <
+		    (sizeof(struct rte_event_vector) +
+		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+	}
+
 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
 		(rx_queue_id != -1)) {
 		RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
@@ -2487,83 +2502,6 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }

-int
-rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct rte_event_eth_rx_adapter_vector_limits limits;
-	struct rte_event_eth_rx_adapter *rx_adapter;
-	struct rte_eventdev *dev;
-	uint32_t cap;
-	int ret;
-
-	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
-	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
-
-	rx_adapter = rxa_id_to_adapter(id);
-	if ((rx_adapter == NULL) || (config == NULL))
-		return -EINVAL;
-
-	dev = &rte_eventdevs[rx_adapter->eventdev_id];
-	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
-						eth_dev_id, &cap);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
-				 "eth port %" PRIu16,
-				 id, eth_dev_id);
-		return ret;
-	}
-
-	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
-		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	ret = rte_event_eth_rx_adapter_vector_limits_get(
-		rx_adapter->eventdev_id, eth_dev_id, &limits);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
-				 "eth port %" PRIu16,
-				 rx_adapter->eventdev_id, eth_dev_id);
-		return ret;
-	}
-
-	if (config->vector_sz < limits.min_sz ||
-	    config->vector_sz > limits.max_sz ||
-	    config->vector_timeout_ns < limits.min_timeout_ns ||
-	    config->vector_timeout_ns > limits.max_timeout_ns ||
-	    config->vector_mp == NULL) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-	if (config->vector_mp->elt_size <
-	    (sizeof(struct rte_event_vector) +
-	     (sizeof(uintptr_t) * config->vector_sz))) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
-		RTE_FUNC_PTR_OR_ERR_RET(
-			*dev->dev_ops->eth_rx_adapter_event_vector_config,
-			-ENOTSUP);
-		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
-			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
-	} else {
-		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
-					      rx_queue_id, config);
-	}
-
-	return ret;
-}
-
 int
 rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
index dda846f57..3f9c749e9 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
@@ -171,9 +171,6 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
 	 * enqueued event.
 	 */
-};
-
-struct rte_event_eth_rx_adapter_event_vector_config {
 	uint16_t vector_sz;
 	/**<
 	 * Indicates the maximum number for mbufs to combine and form a vector.
@@ -538,30 +535,6 @@ int rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-/**
- * Configure event vectorization for a given ethernet device queue, that has
- * been added to a event eth Rx adapter.
- *
- * @param id
- *  The identifier of the ethernet Rx event adapter.
- *
- * @param eth_dev_id
- *  The identifier of the ethernet device.
- *
- * @param rx_queue_id
- *  Ethernet device receive queue index.
- *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
- *  are configured with event vectorization.
- *
- * @return
- *  - 0: Success, Receive queue configured correctly.
- *  - <0: Error code on failure.
- */
-__rte_experimental
-int rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index 902df0ae3..34c1c830e 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -142,7 +142,6 @@ EXPERIMENTAL {
 	#added in 21.05
 	rte_event_vector_pool_create;
 	rte_event_eth_rx_adapter_vector_limits_get;
-	rte_event_eth_rx_adapter_queue_event_vector_config;
 };

 INTERNAL {
--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v2 1/8] eventdev: introduce event vector capability
  2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 1/8] eventdev: introduce event vector capability pbhagavatula
@ 2021-03-16 17:48     ` Jerin Jacob
  2021-03-16 19:17       ` [dpdk-dev] [EXT] " Pavan Nikhilesh Bhagavatula
  0 siblings, 1 reply; 153+ messages in thread
From: Jerin Jacob @ 2021-03-16 17:48 UTC (permalink / raw)
  To: Pavan Nikhilesh
  Cc: Jerin Jacob, Jayatheerthan, Jay, Erik Gabriel Carrillo, Gujjar,
	Abhinandan S, McDaniel, Timothy, Hemant Agrawal, Van Haaren,
	Harry, Mattias Rönnblom, Liang Ma, Ray Kinsella,
	Neil Horman, dpdk-dev

On Tue, Mar 16, 2021 at 9:19 PM <pbhagavatula@marvell.com> wrote:
>
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>
> Introduce rte_event_vector datastructure which is capable of holding
> multiple uintptr_t of the same flow thereby allowing applications
> to vectorize their pipeline and reducing the complexity of pipelining
> the events across multiple stages.
> This approach also reduces the scheduling overhead on a event device.
>
> Add a event vector mempool create handler to create mempools based on
> the best mempool ops available on a given platform.
>
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> -#define RTE_EVENT_DEV_CAP_EVENT_QOS           (1ULL << 1)
> +#define RTE_EVENT_DEV_CAP_EVENT_QOS (1ULL << 1)

A lot of the style changes are accidentally added to this patch.
Please resend.

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [EXT] Re: [PATCH v2 1/8] eventdev: introduce event vector capability
  2021-03-16 17:48     ` Jerin Jacob
@ 2021-03-16 19:17       ` Pavan Nikhilesh Bhagavatula
  0 siblings, 0 replies; 153+ messages in thread
From: Pavan Nikhilesh Bhagavatula @ 2021-03-16 19:17 UTC (permalink / raw)
  To: Jerin Jacob
  Cc: Jerin Jacob Kollanukkaran, Jayatheerthan, Jay,
	Erik Gabriel Carrillo, Gujjar, Abhinandan S, McDaniel, Timothy,
	Hemant Agrawal, Van Haaren, Harry, Mattias Rönnblom,
	Liang Ma, Ray Kinsella, Neil Horman, dpdk-dev

>On Tue, Mar 16, 2021 at 9:19 PM <pbhagavatula@marvell.com> wrote:
>>
>> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>>
>> Introduce rte_event_vector datastructure which is capable of holding
>> multiple uintptr_t of the same flow thereby allowing applications
>> to vectorize their pipeline and reducing the complexity of pipelining
>> the events across multiple stages.
>> This approach also reduces the scheduling overhead on a event
>device.
>>
>> Add a event vector mempool create handler to create mempools
>based on
>> the best mempool ops available on a given platform.
>>
>> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
>> -#define RTE_EVENT_DEV_CAP_EVENT_QOS           (1ULL << 1)
>> +#define RTE_EVENT_DEV_CAP_EVENT_QOS (1ULL << 1)
>
>A lot of the style changes are accidentally added to this patch.
>Please resend.

My bad, I will send v3.

^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v3 0/8] Introduce event vectorization
  2021-03-16 15:48 ` [dpdk-dev] [PATCH v2 0/8] " pbhagavatula
                     ` (7 preceding siblings ...)
  2021-03-16 15:48   ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v2 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
@ 2021-03-16 20:01   ` pbhagavatula
  2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 1/8] eventdev: introduce event vector capability pbhagavatula
                       ` (8 more replies)
  8 siblings, 9 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-16 20:01 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

In traditional event programming model, events are identified by a
flow-id and a uintptr_t. The flow-id uniquely identifies a given event
and determines the order of scheduling based on schedule type, the
uintptr_t holds a single object.

Event devices also support burst mode with configurable dequeue depth,
i.e. each dequeue call would return multiple events and each event
might be at a different stage of the pipeline.
Having a burst of events belonging to different stages in a dequeue
burst is not only difficult to vectorize but also increases the scheduler
overhead and application overhead of pipelining events further.
Using event vectors we see a performance gain of ~628% as shown in [1].

By introducing event vectorization, each event will be capable of holding
multiple uintptr_t of the same flow thereby allowing applications
to vectorize their pipeline and reduce the complexity of pipelining
events across multiple stages. This also reduces the complexity of handling
enqueue and dequeue on an event device.

Since event devices are transparent to the events they are scheduling
so the event producers such as eth_rx_adapter, crypto_adapter , etc..
are responsible for vectorizing the buffers of the same flow into a single
event.

The series also breaks ABI in the patch [8/8] which is targetted to the
v21.11 release.

The dpdk-test-eventdev application has been updated with options to test
multiple vector sizes and timeouts.

[1]
As for performance improvement, with a ARM Cortex-A72 equivalent processer,
software event device (--vdev=event_sw0), single worker core, single stage
and using one service core for Rx adapter, Tx adapter, Scheduling.

Without event vectorization:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
         --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
         --stlist=a --wlcores=20
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    4.728 mpps avg 4.728 mpps

With event vectorization:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
        --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
        --stlist=a --wlcores=20 --enable_vector --nb_eth_queues 1
        --vector_size 256
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    34.383 mpps avg 34.383 mpps

Having dedicated service cores for each Rx queues and tweaking the vector,
dequeue burst size would further improve performance.

API usage is shown below:

Configuration:

	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

	vector_pool = rte_event_vector_pool_create("vector_pool",
			nb_elem, 0, vector_size, socket_id);

	rte_event_eth_rx_adapter_create(id, event_id, &adptr_conf);
	rte_event_eth_rx_adapter_queue_add(id, eth_id, -1, &queue_conf);
	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
		vec_conf.vector_sz = vector_size;
		vec_conf.vector_timeout_ns = vector_tmo_nsec;
		vec_conf.vector_mp = vector_pool;
		rte_event_eth_rx_adapter_queue_event_vector_config(id,
				eth_id, -1, &vec_conf);
	}

Fastpath:

	num = rte_event_dequeue_burst(event_id, port_id, &ev, 1, 0);
	if (!num)
		continue;

	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
		switch (ev.event_type) {
		case RTE_EVENT_TYPE_ETHDEV_VECTOR:
		case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
			struct rte_mbuf **mbufs;

			mbufs = ev.vector_ev->mbufs;
			for (i = 0; i < ev.vector_ev->nb_elem; i++)
				//Process mbufs.
			break;
		case ...
		}
	}
	...

v3 Changes:
- Fix unintended formatting changes.

v2 Changes:
- Multiple gramatical and style fixes.(Jerin)
- Add parameter to define vector size in power of 2. (Jerin)
- Redo patch series w/o breaking ABI till the last patch.(David)
- Add deprication notice to announce ABI break in 21.11.(David)
- Add vector limits validation to app/test-eventdev.

Pavan Nikhilesh (8):
  eventdev: introduce event vector capability
  eventdev: introduce event vector Rx capability
  eventdev: introduce event vector Tx capability
  eventdev: add Rx adapter event vector support
  eventdev: add Tx adapter event vector support
  app/eventdev: add event vector mode in pipeline test
  doc: announce event Rx adapter config changes
  eventdev: simplify Rx adapter event vector config

 app/test-eventdev/evt_common.h                |   4 +
 app/test-eventdev/evt_options.c               |  52 +++
 app/test-eventdev/evt_options.h               |   4 +
 app/test-eventdev/test_pipeline_atq.c         | 310 +++++++++++++++--
 app/test-eventdev/test_pipeline_common.c      | 105 +++++-
 app/test-eventdev/test_pipeline_common.h      |  18 +
 app/test-eventdev/test_pipeline_queue.c       | 320 ++++++++++++++++--
 .../prog_guide/event_ethernet_rx_adapter.rst  |  38 +++
 .../prog_guide/event_ethernet_tx_adapter.rst  |  12 +
 doc/guides/prog_guide/eventdev.rst            |  36 +-
 doc/guides/rel_notes/deprecation.rst          |   9 +
 doc/guides/tools/testeventdev.rst             |  28 ++
 lib/librte_eventdev/eventdev_pmd.h            |  31 +-
 .../rte_event_eth_rx_adapter.c                | 305 ++++++++++++++++-
 .../rte_event_eth_rx_adapter.h                |  68 ++++
 .../rte_event_eth_tx_adapter.c                |  66 +++-
 lib/librte_eventdev/rte_eventdev.c            |  11 +-
 lib/librte_eventdev/rte_eventdev.h            | 142 +++++++-
 lib/librte_eventdev/version.map               |   4 +
 19 files changed, 1477 insertions(+), 86 deletions(-)

--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v3 1/8] eventdev: introduce event vector capability
  2021-03-16 20:01   ` [dpdk-dev] [PATCH v3 0/8] Introduce event vectorization pbhagavatula
@ 2021-03-16 20:01     ` pbhagavatula
  2021-03-18  6:19       ` Jayatheerthan, Jay
  2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 2/8] eventdev: introduce event vector Rx capability pbhagavatula
                       ` (7 subsequent siblings)
  8 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-03-16 20:01 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce rte_event_vector datastructure which is capable of holding
multiple uintptr_t of the same flow thereby allowing applications
to vectorize their pipeline and reducing the complexity of pipelining
the events across multiple stages.
This approach also reduces the scheduling overhead on a event device.

Add a event vector mempool create handler to create mempools based on
the best mempool ops available on a given platform.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 doc/guides/prog_guide/eventdev.rst |  36 +++++++++-
 lib/librte_eventdev/rte_eventdev.h | 110 ++++++++++++++++++++++++++++-
 lib/librte_eventdev/version.map    |   3 +
 3 files changed, 146 insertions(+), 3 deletions(-)

diff --git a/doc/guides/prog_guide/eventdev.rst b/doc/guides/prog_guide/eventdev.rst
index ccde086f6..fda9c3743 100644
--- a/doc/guides/prog_guide/eventdev.rst
+++ b/doc/guides/prog_guide/eventdev.rst
@@ -63,13 +63,45 @@ the actual event being scheduled is. The payload is a union of the following:
 * ``uint64_t u64``
 * ``void *event_ptr``
 * ``struct rte_mbuf *mbuf``
+* ``struct rte_event_vector *vec``
 
-These three items in a union occupy the same 64 bits at the end of the rte_event
+These four items in a union occupy the same 64 bits at the end of the rte_event
 structure. The application can utilize the 64 bits directly by accessing the
-u64 variable, while the event_ptr and mbuf are provided as convenience
+u64 variable, while the event_ptr, mbuf, vec are provided as a convenience
 variables.  For example the mbuf pointer in the union can used to schedule a
 DPDK packet.
 
+Event Vector
+~~~~~~~~~~~~
+
+The rte_event_vector struct contains a vector of elements defined by the event
+type specified in the ``rte_event``. The event_vector structure contains the
+following data:
+
+* ``nb_elem`` - The number of elements held within the vector.
+
+Similar to ``rte_event`` the payload of event vector is also a union, allowing
+flexibility in what the actual vector is.
+
+* ``struct rte_mbuf *mbufs[0]`` - An array of mbufs.
+* ``void *ptrs[0]`` - An array of pointers.
+* ``uint64_t *u64s[0]`` - An array of uint64_t elements.
+
+The size of the event vector is related to the total number of elements it is
+configured to hold, this is achieved by making `rte_event_vector` a variable
+length structure.
+A helper function is provided to create a mempool that holds event vector, which
+takes name of the pool, total number of required ``rte_event_vector``,
+cache size, number of elements in each ``rte_event_vector`` and socket id.
+
+.. code-block:: c
+
+        rte_event_vector_pool_create("vector_pool", nb_event_vectors, cache_sz,
+                                     nb_elements_per_vector, socket_id);
+
+The function ``rte_event_vector_pool_create`` creates mempool with the best
+platform mempool ops.
+
 Queues
 ~~~~~~
 
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index ce1fc2ce0..c0d01c873 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -212,8 +212,10 @@ extern "C" {
 
 #include <rte_common.h>
 #include <rte_config.h>
-#include <rte_memory.h>
 #include <rte_errno.h>
+#include <rte_mbuf_pool_ops.h>
+#include <rte_memory.h>
+#include <rte_mempool.h>
 
 #include "rte_eventdev_trace_fp.h"
 
@@ -913,6 +915,25 @@ rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
 int
 rte_event_dev_close(uint8_t dev_id);
 
+/**
+ * Event vector structure.
+ */
+struct rte_event_vector {
+	uint64_t nb_elem : 16;
+	/**< Number of elements in this event vector. */
+	uint64_t rsvd : 48;
+	uint64_t impl_opaque;
+	union {
+		struct rte_mbuf *mbufs[0];
+		void *ptrs[0];
+		uint64_t *u64s[0];
+	} __rte_aligned(16);
+	/**< Start of the vector array union. Depending upon the event type the
+	 * vector array can be an array of mbufs or pointers or opaque u64
+	 * values.
+	 */
+};
+
 /* Scheduler type definitions */
 #define RTE_SCHED_TYPE_ORDERED          0
 /**< Ordered scheduling
@@ -986,6 +1007,21 @@ rte_event_dev_close(uint8_t dev_id);
  */
 #define RTE_EVENT_TYPE_ETH_RX_ADAPTER   0x4
 /**< The event generated from event eth Rx adapter */
+#define RTE_EVENT_TYPE_VECTOR           0x8
+/**< Indicates that event is a vector.
+ * All vector event types should be an logical OR of EVENT_TYPE_VECTOR.
+ * This simplifies the pipeline design as we can split processing the events
+ * between vector events and normal event across event types.
+ * Example:
+ *	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
+ *		// Classify and handle vector event.
+ *	} else {
+ *		// Classify and handle event.
+ *	}
+ */
+#define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
+/**< The event vector generated from cpu for pipelining. */
+
 #define RTE_EVENT_TYPE_MAX              0x10
 /**< Maximum number of event types */
 
@@ -2023,6 +2059,78 @@ rte_event_dev_xstats_reset(uint8_t dev_id,
  */
 int rte_event_dev_selftest(uint8_t dev_id);
 
+/**
+ * Get the memory required per event vector based on the number of elements per
+ * vector.
+ * This should be used to create the mempool that holds the event vectors.
+ *
+ * @param name
+ *   The name of the vector pool.
+ * @param n
+ *   The number of elements in the mbuf pool.
+ * @param cache_size
+ *   Size of the per-core object cache. See rte_mempool_create() for
+ *   details.
+ * @param nb_elem
+ *   The number of elements then a single event vector should be able to hold.
+ * @param socket_id
+ *   The socket identifier where the memory should be allocated. The
+ *   value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
+ *   reserved zone
+ *
+ * @return
+ *   The pointer to the newly allocated mempool, on success. NULL on error
+ *   with rte_errno set appropriately. Possible rte_errno values include:
+ *    - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
+ *    - E_RTE_SECONDARY - function was called from a secondary process instance
+ *    - EINVAL - cache size provided is too large, or priv_size is not aligned.
+ *    - ENOSPC - the maximum number of memzones has already been allocated
+ *    - EEXIST - a memzone with the same name already exists
+ *    - ENOMEM - no appropriate memory area found in which to create memzone
+ */
+__rte_experimental
+static inline struct rte_mempool *
+rte_event_vector_pool_create(const char *name, unsigned int n,
+			     unsigned int cache_size, uint16_t nb_elem,
+			     int socket_id)
+{
+	const char *mp_ops_name;
+	struct rte_mempool *mp;
+	unsigned int elt_sz;
+	int ret;
+
+	if (!nb_elem) {
+		RTE_LOG(ERR, EVENTDEV,
+			"Invalid number of elements=%d requested\n", nb_elem);
+		rte_errno = -EINVAL;
+		return NULL;
+	}
+
+	elt_sz =
+		sizeof(struct rte_event_vector) + (nb_elem * sizeof(uintptr_t));
+	mp = rte_mempool_create_empty(name, n, elt_sz, cache_size, 0, socket_id,
+				      0);
+	if (mp == NULL)
+		return NULL;
+
+	mp_ops_name = rte_mbuf_best_mempool_ops();
+	ret = rte_mempool_set_ops_byname(mp, mp_ops_name, NULL);
+	if (ret != 0) {
+		RTE_LOG(ERR, EVENTDEV, "error setting mempool handler\n");
+		goto err;
+	}
+
+	ret = rte_mempool_populate_default(mp);
+	if (ret < 0)
+		goto err;
+
+	return mp;
+err:
+	rte_mempool_free(mp);
+	rte_errno = -ret;
+	return NULL;
+}
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index 3e5c09cfd..a070ef56e 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -138,6 +138,9 @@ EXPERIMENTAL {
 	__rte_eventdev_trace_port_setup;
 	# added in 20.11
 	rte_event_pmd_pci_probe_named;
+
+	#added in 21.05
+	rte_event_vector_pool_create;
 };
 
 INTERNAL {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v3 2/8] eventdev: introduce event vector Rx capability
  2021-03-16 20:01   ` [dpdk-dev] [PATCH v3 0/8] Introduce event vectorization pbhagavatula
  2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 1/8] eventdev: introduce event vector capability pbhagavatula
@ 2021-03-16 20:01     ` pbhagavatula
  2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 3/8] eventdev: introduce event vector Tx capability pbhagavatula
                       ` (6 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-16 20:01 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce event ethernet Rx adapter event vector capability.

If an event eth Rx adapter has the capability of
RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR then a given Rx queue
can be configured to enable event vectorization by passing the
flag RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR to
rte_event_eth_rx_adapter_queue_conf::rx_queue_flags while configuring
Rx adapter through rte_event_eth_rx_adapter_queue_add.

The max vector size, vector timeout define the vector size and
mempool used for allocating vector event are configured through
rte_event_eth_rx_adapter_queue_add. The element size of the element
in the vector pool should be equal to
    sizeof(struct rte_event_vector) + (vector_sz * sizeof(uintptr_t))

Application can use `rte_event_vector_pool_create` to create the
vector mempool used for
rte_event_eth_rx_adapter_queue_conf::vector_mp.

The Rx adapter would be responsible for vectorizing the mbufs
based on the flow, the vector limits configured by the application
and add the vector event of mbufs to the event queue set via
rte_event_eth_rx_adapter_queue_conf::ev::queue_id.
It should also mark rte_event_vector::union_valid and fill
rte_event_vector::port, rte_event_vector::queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 .../prog_guide/event_ethernet_rx_adapter.rst  | 38 ++++++++
 .../rte_event_eth_rx_adapter.h                | 95 +++++++++++++++++++
 lib/librte_eventdev/rte_eventdev.h            | 30 +++++-
 lib/librte_eventdev/version.map               |  2 +
 4 files changed, 163 insertions(+), 2 deletions(-)

diff --git a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
index cb44ce0e4..5eefef355 100644
--- a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
@@ -186,3 +186,41 @@ the event buffer fill level is low. The
 ``rte_event_eth_rx_adapter_cb_register()`` function allow the application
 to register a callback that selects which packets to enqueue to the event
 device.
+
+Rx event vectorization
+~~~~~~~~~~~~~~~~~~~~~~
+
+The event devices, ethernet device pairs which support the capability
+``RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR`` can aggregate packets based on
+flow characteristics and generate a ``rte_event`` containing ``rte_event_vector``
+whose event type is either ``RTE_EVENT_TYPE_ETHDEV_VECTOR`` or
+``RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR``.
+The aggregation size and timeout are configurable at a queue level and the
+maximum, minimum vector sizes and timeouts vary based on the device capability
+and can be queried using ``rte_event_eth_rx_adapter_vector_limits_get``.
+The Rx adapter additionally might include useful data such as ethernet device
+port and queue identifier in the ``rte_event_vector::port`` and
+``rte_event_vector::queue`` and mark ``rte_event_vector::attr_valid`` as true.
+
+A loop processing ``rte_event_vector`` containing mbufs is shown below.
+
+.. code-block:: c
+
+        event = rte_event_dequeue_burst(event_dev, event_port, &event,
+                                        1, 0);
+        if (!event)
+                continue;
+
+        switch (ev.event_type) {
+        case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
+        case RTE_EVENT_TYPE_ETHDEV_VECTOR:
+                struct rte_mbufs **mbufs;
+
+                mbufs = (struct rte_mbufs **)ev[i].vec->mbufs;
+                for (i = 0; i < ev.vec->nb_elem; i++) {
+                        /* Process each mbuf. */
+                }
+        break;
+        case ...
+        ...
+        }
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
index 21bb1e54c..dda846f57 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
@@ -92,6 +92,10 @@ extern "C" {
 /**< This flag indicates the flow identifier is valid
  * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
  */
+#define RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR	0x2
+/**< This flag indicates that mbufs arriving on the queue need to be vectorized
+ * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+ */
 
 /**
  * Adapter configuration structure that the adapter configuration callback
@@ -169,6 +173,36 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 */
 };
 
+struct rte_event_eth_rx_adapter_event_vector_config {
+	uint16_t vector_sz;
+	/**<
+	 * Indicates the maximum number for mbufs to combine and form a vector.
+	 * Should be within
+	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_sz
+	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_sz
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+	 */
+	uint64_t vector_timeout_ns;
+	/**<
+	 * Indicates the maximum number of nanoseconds to wait for receiving
+	 * mbufs. Should be within vectorization limits of the
+	 * adapter
+	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_ns
+	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_ns
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+	 */
+	struct rte_mempool *vector_mp;
+	/**<
+	 * Indicates the mempool that should be used for allocating
+	 * rte_event_vector container.
+	 * Should be created by using `rte_event_vector_pool_create`.
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags.
+	 */
+};
+
 /**
  * A structure used to retrieve statistics for an eth rx adapter instance.
  */
@@ -199,6 +233,22 @@ struct rte_event_eth_rx_adapter_stats {
 	/**< Received packet count for interrupt mode Rx queues */
 };
 
+/**
+ * A structure used to retrieve eth rx adapter vector limits.
+ */
+struct rte_event_eth_rx_adapter_vector_limits {
+	uint16_t min_sz;
+	/**< Minimum vector limit configurable. */
+	uint16_t max_sz;
+	/**< Maximum vector limit configurable. */
+	uint8_t log2_sz;
+	/**< The size configured should be in log2. */
+	uint64_t min_timeout_ns;
+	/**< Minimum vector timeout configurable. */
+	uint64_t max_timeout_ns;
+	/**< Maximum vector timeout configurable. */
+};
+
 /**
  *
  * Callback function invoked by the SW adapter before it continues
@@ -467,6 +517,51 @@ int rte_event_eth_rx_adapter_cb_register(uint8_t id, uint16_t eth_dev_id,
 					 rte_event_eth_rx_adapter_cb_fn cb_fn,
 					 void *cb_arg);
 
+/**
+ * Retrieve vector limits for a given event dev and eth dev pair.
+ * @see rte_event_eth_rx_adapter_vector_limits
+ *
+ * @param dev_id
+ *  Event device identifier.
+ * @param eth_port_id
+ *  Port identifier of the ethernet device.
+ * @param [out] limits
+ *  A pointer to rte_event_eth_rx_adapter_vector_limits structure that has to
+ * be filled.
+ *
+ * @return
+ *  - 0: Success.
+ *  - <0: Error code on failure.
+ */
+__rte_experimental
+int rte_event_eth_rx_adapter_vector_limits_get(
+	uint8_t dev_id, uint16_t eth_port_id,
+	struct rte_event_eth_rx_adapter_vector_limits *limits);
+
+/**
+ * Configure event vectorization for a given ethernet device queue, that has
+ * been added to a event eth Rx adapter.
+ *
+ * @param id
+ *  The identifier of the ethernet Rx event adapter.
+ *
+ * @param eth_dev_id
+ *  The identifier of the ethernet device.
+ *
+ * @param rx_queue_id
+ *  Ethernet device receive queue index.
+ *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
+ *  are configured with event vectorization.
+ *
+ * @return
+ *  - 0: Success, Receive queue configured correctly.
+ *  - <0: Error code on failure.
+ */
+__rte_experimental
+int rte_event_eth_rx_adapter_queue_event_vector_config(
+	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
+	struct rte_event_eth_rx_adapter_event_vector_config *config);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index c0d01c873..e62825e49 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -919,9 +919,27 @@ rte_event_dev_close(uint8_t dev_id);
  * Event vector structure.
  */
 struct rte_event_vector {
-	uint64_t nb_elem : 16;
+	uint16_t nb_elem;
 	/**< Number of elements in this event vector. */
-	uint64_t rsvd : 48;
+	uint16_t rsvd : 15;
+	uint16_t attr_valid : 1;
+	/**< Indicates that the below union attributes have valid information.
+	 */
+	union {
+		/* Used by Rx adapter.
+		 * Indicats that all the elements in this vector belong to same
+		 * port and queue pair when originating from Rx adapter, valid
+		 * only when event type is ETHDEV_VECTOR or
+		 * ETH_RX_ADAPTER_VECTOR.
+		 */
+		struct {
+			uint16_t port;
+			/* Ethernet device port id. */
+			uint16_t queue;
+			/* Ethernet device queue id. */
+		};
+	};
+	/**< Union to hold common attributes of the vector array. */
 	uint64_t impl_opaque;
 	union {
 		struct rte_mbuf *mbufs[0];
@@ -1019,8 +1037,14 @@ struct rte_event_vector {
  *		// Classify and handle event.
  *	}
  */
+#define RTE_EVENT_TYPE_ETHDEV_VECTOR                                           \
+	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETHDEV)
+/**< The event vector generated from ethdev subsystem */
 #define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
 /**< The event vector generated from cpu for pipelining. */
+#define RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR                                   \
+	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETH_RX_ADAPTER)
+/**< The event vector generated from eth Rx adapter. */
 
 #define RTE_EVENT_TYPE_MAX              0x10
 /**< Maximum number of event types */
@@ -1163,6 +1187,8 @@ struct rte_event {
  * @see struct rte_event_eth_rx_adapter_queue_conf::ev
  * @see struct rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
  */
+#define RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR	0x8
+/**< Adapter supports event vectorization per ethdev. */
 
 /**
  * Retrieve the event device's ethdev Rx adapter capabilities for the
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index a070ef56e..902df0ae3 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -141,6 +141,8 @@ EXPERIMENTAL {
 
 	#added in 21.05
 	rte_event_vector_pool_create;
+	rte_event_eth_rx_adapter_vector_limits_get;
+	rte_event_eth_rx_adapter_queue_event_vector_config;
 };
 
 INTERNAL {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v3 3/8] eventdev: introduce event vector Tx capability
  2021-03-16 20:01   ` [dpdk-dev] [PATCH v3 0/8] Introduce event vectorization pbhagavatula
  2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 1/8] eventdev: introduce event vector capability pbhagavatula
  2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 2/8] eventdev: introduce event vector Rx capability pbhagavatula
@ 2021-03-16 20:01     ` pbhagavatula
  2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 4/8] eventdev: add Rx adapter event vector support pbhagavatula
                       ` (5 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-16 20:01 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce event vector transmit capability for event eth
tx adapter.

The capability indicates that the Tx adapter is capable of
transmitting event vectors.
When rte_event_vector::union_valid is set, the Tx adapter should
transmit all the packets to the rte_event_vector::port using the
rte_event_vector::queue.
If rte_event_vector::union_valid is not set then the Tx adapter
should peek into each mbuf to get the destination port and queue
pair.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
---
 doc/guides/prog_guide/event_ethernet_tx_adapter.rst | 12 ++++++++++++
 lib/librte_eventdev/rte_eventdev.h                  |  8 +++++++-
 2 files changed, 19 insertions(+), 1 deletion(-)

diff --git a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
index a8c13e136..87277dcaf 100644
--- a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
@@ -164,3 +164,15 @@ The  ``rte_event_eth_tx_adapter_stats_get()`` function reports counters defined
 in struct ``rte_event_eth_tx_adapter_stats``. The counter values are the sum of
 the counts from the eventdev PMD callback if the callback is supported, and
 the counts maintained by the service function, if one exists.
+
+Tx event vectorization
+~~~~~~~~~~~~~~~~~~~~~~
+
+The event device, ethernet device paris which support the capability
+``RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR`` can process event vector of mbufs.
+Additionally, application can provide a hint to that Tx adapter that all the
+mbufs are destined to the same ethernet port and queue by setting the bit
+``rte_event_vector::attr_valid`` and filling `rte_event_vector::port`` and
+``rte_event_vector::queue``.
+If ``rte_event_vector::attr_valid`` is not set then the Tx adapter should peek
+into each mbuf and transmit them to the requested ethernet port and queue pair.
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index e62825e49..6737d113f 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -926,11 +926,13 @@ struct rte_event_vector {
 	/**< Indicates that the below union attributes have valid information.
 	 */
 	union {
-		/* Used by Rx adapter.
+		/* Used by Rx/Tx adapter.
 		 * Indicats that all the elements in this vector belong to same
 		 * port and queue pair when originating from Rx adapter, valid
 		 * only when event type is ETHDEV_VECTOR or
 		 * ETH_RX_ADAPTER_VECTOR.
+		 * Can also be used to indicate the Tx adapter the destination
+		 * port and queue of the mbufs in the vector
 		 */
 		struct {
 			uint16_t port;
@@ -1285,6 +1287,10 @@ rte_event_crypto_adapter_caps_get(uint8_t dev_id, uint8_t cdev_id,
 #define RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT	0x1
 /**< This flag is sent when the PMD supports a packet transmit callback
  */
+#define RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR	0x2
+/**< Indicates that the Tx adapter is capable of handling event vector of
+ * mbufs.
+ */
 
 /**
  * Retrieve the event device's eth Tx adapter capabilities
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v3 4/8] eventdev: add Rx adapter event vector support
  2021-03-16 20:01   ` [dpdk-dev] [PATCH v3 0/8] Introduce event vectorization pbhagavatula
                       ` (2 preceding siblings ...)
  2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 3/8] eventdev: introduce event vector Tx capability pbhagavatula
@ 2021-03-16 20:01     ` pbhagavatula
  2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 5/8] eventdev: add Tx " pbhagavatula
                       ` (4 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-16 20:01 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support for event eth Rx adapter, the implementation
creates vector flows based on port and queue identifier of the received
mbufs.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 lib/librte_eventdev/eventdev_pmd.h            |  60 ++-
 .../rte_event_eth_rx_adapter.c                | 367 +++++++++++++++++-
 lib/librte_eventdev/rte_eventdev.c            |   6 +-
 3 files changed, 415 insertions(+), 18 deletions(-)

diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 7eb9a7739..d79dfd612 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -69,9 +69,10 @@ extern "C" {
 	} \
 } while (0)
 
-#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP \
-		((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) | \
-			(RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ))
+#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP                                        \
+	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |                     \
+	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) |                         \
+	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR))
 
 #define RTE_EVENT_CRYPTO_ADAPTER_SW_CAP \
 		RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA
@@ -645,6 +646,53 @@ typedef int (*eventdev_eth_rx_adapter_stats_reset)
  */
 typedef int (*eventdev_selftest)(void);
 
+struct rte_event_eth_rx_adapter_vector_limits;
+/**
+ * Get event vector limits for a given event, etherner device pair.
+ *
+ * @param dev
+ *   Event device pointer
+ *
+ * @param eth_dev
+ *   Ethernet device pointer
+ *
+ * @param[out] limits
+ *   Pointer to the limits structure to be filled.
+ *
+ * @return
+ *   - 0: Success.
+ *   - <0: Error code returned by the driver function.
+ */
+typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
+	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
+	struct rte_event_eth_rx_adapter_vector_limits *limits);
+
+struct rte_event_eth_rx_adapter_event_vector_config;
+/**
+ * Enable event vector on an given Rx queue of a ethernet devices belonging to
+ * the Rx adapter.
+ *
+ * @param dev
+ *   Event device pointer
+ *
+ * @param eth_dev
+ *   Ethernet device pointer
+ *
+ * @param rx_queue_id
+ *   The Rx queue identifier
+ *
+ * @param config
+ *   Pointer to the event vector configuration structure.
+ *
+ * @return
+ *   - 0: Success.
+ *   - <0: Error code returned by the driver function.
+ */
+typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
+	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
+	int32_t rx_queue_id,
+	const struct rte_event_eth_rx_adapter_event_vector_config *config);
+
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;
 
@@ -1067,6 +1115,12 @@ struct rte_eventdev_ops {
 	/**< Get ethernet Rx stats */
 	eventdev_eth_rx_adapter_stats_reset eth_rx_adapter_stats_reset;
 	/**< Reset ethernet Rx stats */
+	eventdev_eth_rx_adapter_vector_limits_get_t
+		eth_rx_adapter_vector_limits_get;
+	/**< Get event vector limits for the Rx adapter */
+	eventdev_eth_rx_adapter_event_vector_config_t
+		eth_rx_adapter_event_vector_config;
+	/**< Configure Rx adapter with event vector */
 
 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index d8c635e99..c71990078 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -26,6 +26,10 @@
 #define BATCH_SIZE		32
 #define BLOCK_CNT_THRESHOLD	10
 #define ETH_EVENT_BUFFER_SIZE	(4*BATCH_SIZE)
+#define MAX_VECTOR_SIZE		1024
+#define MIN_VECTOR_SIZE		4
+#define MAX_VECTOR_NS		1E9
+#define MIN_VECTOR_NS		1E5
 
 #define ETH_RX_ADAPTER_SERVICE_NAME_LEN	32
 #define ETH_RX_ADAPTER_MEM_NAME_LEN	32
@@ -59,6 +63,20 @@ struct eth_rx_poll_entry {
 	uint16_t eth_rx_qid;
 };
 
+struct eth_rx_vector_data {
+	TAILQ_ENTRY(eth_rx_vector_data) next;
+	uint16_t port;
+	uint16_t queue;
+	uint16_t max_vector_count;
+	uint64_t event;
+	uint64_t ts;
+	uint64_t vector_timeout_ticks;
+	struct rte_mempool *vector_pool;
+	struct rte_event_vector *vector_ev;
+} __rte_cache_aligned;
+
+TAILQ_HEAD(eth_rx_vector_data_list, eth_rx_vector_data);
+
 /* Instance per adapter */
 struct rte_eth_event_enqueue_buffer {
 	/* Count of events in this buffer */
@@ -92,6 +110,14 @@ struct rte_event_eth_rx_adapter {
 	uint32_t wrr_pos;
 	/* Event burst buffer */
 	struct rte_eth_event_enqueue_buffer event_enqueue_buffer;
+	/* Vector enable flag */
+	uint8_t ena_vector;
+	/* Timestamp of previous vector expiry list traversal */
+	uint64_t prev_expiry_ts;
+	/* Minimum ticks to wait before traversing expiry list */
+	uint64_t vector_tmo_ticks;
+	/* vector list */
+	struct eth_rx_vector_data_list vector_list;
 	/* Per adapter stats */
 	struct rte_event_eth_rx_adapter_stats stats;
 	/* Block count, counts up to BLOCK_CNT_THRESHOLD */
@@ -198,9 +224,11 @@ struct eth_device_info {
 struct eth_rx_queue_info {
 	int queue_enabled;	/* True if added */
 	int intr_enabled;
+	uint8_t ena_vector;
 	uint16_t wt;		/* Polling weight */
 	uint32_t flow_id_mask;	/* Set to ~0 if app provides flow id else 0 */
 	uint64_t event;
+	struct eth_rx_vector_data vector_data;
 };
 
 static struct rte_event_eth_rx_adapter **event_eth_rx_adapter;
@@ -722,6 +750,9 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
 	    &rx_adapter->event_enqueue_buffer;
 	struct rte_event_eth_rx_adapter_stats *stats = &rx_adapter->stats;
 
+	if (!buf->count)
+		return 0;
+
 	uint16_t n = rte_event_enqueue_new_burst(rx_adapter->eventdev_id,
 					rx_adapter->event_port_id,
 					buf->events,
@@ -742,6 +773,72 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
 	return n;
 }
 
+static inline uint16_t
+rxa_create_event_vector(struct rte_event_eth_rx_adapter *rx_adapter,
+			struct eth_rx_queue_info *queue_info,
+			struct rte_eth_event_enqueue_buffer *buf,
+			struct rte_mbuf **mbufs, uint16_t num)
+{
+	struct rte_event *ev = &buf->events[buf->count];
+	struct eth_rx_vector_data *vec;
+	uint16_t filled, space, sz;
+
+	filled = 0;
+	vec = &queue_info->vector_data;
+	while (num) {
+		if (vec->vector_ev == NULL) {
+			if (rte_mempool_get(vec->vector_pool,
+					    (void **)&vec->vector_ev) < 0) {
+				rte_pktmbuf_free_bulk(mbufs, num);
+				return 0;
+			}
+			vec->vector_ev->nb_elem = 0;
+			vec->vector_ev->port = vec->port;
+			vec->vector_ev->queue = vec->queue;
+			vec->vector_ev->attr_valid = true;
+			TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
+		} else if (vec->vector_ev->nb_elem == vec->max_vector_count) {
+			/* Event ready. */
+			ev->event = vec->event;
+			ev->vec = vec->vector_ev;
+			ev++;
+			filled++;
+			vec->vector_ev = NULL;
+			TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+			if (rte_mempool_get(vec->vector_pool,
+					    (void **)&vec->vector_ev) < 0) {
+				rte_pktmbuf_free_bulk(mbufs, num);
+				return 0;
+			}
+			vec->vector_ev->nb_elem = 0;
+			vec->vector_ev->port = vec->port;
+			vec->vector_ev->queue = vec->queue;
+			vec->vector_ev->attr_valid = true;
+			TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
+		}
+
+		space = vec->max_vector_count - vec->vector_ev->nb_elem;
+		sz = num > space ? space : num;
+		memcpy(vec->vector_ev->mbufs + vec->vector_ev->nb_elem, mbufs,
+		       sizeof(void *) * sz);
+		vec->vector_ev->nb_elem += sz;
+		num -= sz;
+		mbufs += sz;
+		vec->ts = rte_rdtsc();
+	}
+
+	if (vec->vector_ev->nb_elem == vec->max_vector_count) {
+		ev->event = vec->event;
+		ev->vec = vec->vector_ev;
+		ev++;
+		filled++;
+		vec->vector_ev = NULL;
+		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+	}
+
+	return filled;
+}
+
 static inline void
 rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
@@ -770,25 +867,30 @@ rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
 	rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
 	do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
 
-	for (i = 0; i < num; i++) {
-		m = mbufs[i];
-
-		rss = do_rss ?
-			rxa_do_softrss(m, rx_adapter->rss_key_be) :
-			m->hash.rss;
-		ev->event = event;
-		ev->flow_id = (rss & ~flow_id_mask) |
-				(ev->flow_id & flow_id_mask);
-		ev->mbuf = m;
-		ev++;
+	if (!eth_rx_queue_info->ena_vector) {
+		for (i = 0; i < num; i++) {
+			m = mbufs[i];
+
+			rss = do_rss ? rxa_do_softrss(m, rx_adapter->rss_key_be)
+				     : m->hash.rss;
+			ev->event = event;
+			ev->flow_id = (rss & ~flow_id_mask) |
+				      (ev->flow_id & flow_id_mask);
+			ev->mbuf = m;
+			ev++;
+		}
+	} else {
+		num = rxa_create_event_vector(rx_adapter, eth_rx_queue_info,
+					      buf, mbufs, num);
 	}
 
-	if (dev_info->cb_fn) {
+	if (num && dev_info->cb_fn) {
 
 		dropped = 0;
 		nb_cb = dev_info->cb_fn(eth_dev_id, rx_queue_id,
-					ETH_EVENT_BUFFER_SIZE, buf->count, ev,
-					num, dev_info->cb_arg, &dropped);
+					ETH_EVENT_BUFFER_SIZE, buf->count,
+					&buf->events[buf->count], num,
+					dev_info->cb_arg, &dropped);
 		if (unlikely(nb_cb > num))
 			RTE_EDEV_LOG_ERR("Rx CB returned %d (> %d) events",
 				nb_cb, num);
@@ -1124,6 +1226,30 @@ rxa_poll(struct rte_event_eth_rx_adapter *rx_adapter)
 	return nb_rx;
 }
 
+static void
+rxa_vector_expire(struct eth_rx_vector_data *vec, void *arg)
+{
+	struct rte_event_eth_rx_adapter *rx_adapter = arg;
+	struct rte_eth_event_enqueue_buffer *buf =
+		&rx_adapter->event_enqueue_buffer;
+	struct rte_event *ev;
+
+	if (buf->count)
+		rxa_flush_event_buffer(rx_adapter);
+
+	if (vec->vector_ev->nb_elem == 0)
+		return;
+	ev = &buf->events[buf->count];
+
+	/* Event ready. */
+	ev->event = vec->event;
+	ev->vec = vec->vector_ev;
+	buf->count++;
+
+	vec->vector_ev = NULL;
+	vec->ts = 0;
+}
+
 static int
 rxa_service_func(void *args)
 {
@@ -1137,6 +1263,24 @@ rxa_service_func(void *args)
 		return 0;
 	}
 
+	if (rx_adapter->ena_vector) {
+		if ((rte_rdtsc() - rx_adapter->prev_expiry_ts) >=
+		    rx_adapter->vector_tmo_ticks) {
+			struct eth_rx_vector_data *vec;
+
+			TAILQ_FOREACH(vec, &rx_adapter->vector_list, next) {
+				uint64_t elapsed_time = rte_rdtsc() - vec->ts;
+
+				if (elapsed_time >= vec->vector_timeout_ticks) {
+					rxa_vector_expire(vec, rx_adapter);
+					TAILQ_REMOVE(&rx_adapter->vector_list,
+						     vec, next);
+				}
+			}
+			rx_adapter->prev_expiry_ts = rte_rdtsc();
+		}
+	}
+
 	stats = &rx_adapter->stats;
 	stats->rx_packets += rxa_intr_ring_dequeue(rx_adapter);
 	stats->rx_packets += rxa_poll(rx_adapter);
@@ -1640,6 +1784,28 @@ rxa_update_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }
 
+static void
+rxa_set_vector_data(struct eth_rx_queue_info *queue_info, uint16_t vector_count,
+		    uint64_t vector_ns, struct rte_mempool *mp, int32_t qid,
+		    uint16_t port_id)
+{
+#define NSEC2TICK(__ns, __freq) (((__ns) * (__freq)) / 1E9)
+	struct eth_rx_vector_data *vector_data;
+	uint32_t flow_id;
+
+	vector_data = &queue_info->vector_data;
+	vector_data->max_vector_count = vector_count;
+	vector_data->port = port_id;
+	vector_data->queue = qid;
+	vector_data->vector_pool = mp;
+	vector_data->vector_timeout_ticks =
+		NSEC2TICK(vector_ns, rte_get_timer_hz());
+	vector_data->ts = 0;
+	flow_id = queue_info->event & 0xFFFFF;
+	flow_id = flow_id == 0 ? (qid & 0xFF) | (port_id & 0xFFFF) : flow_id;
+	vector_data->event = (queue_info->event & ~0xFFFFF) | flow_id;
+}
+
 static void
 rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
 	struct eth_device_info *dev_info,
@@ -1741,6 +1907,44 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }
 
+static void
+rxa_sw_event_vector_configure(
+	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
+	int rx_queue_id,
+	const struct rte_event_eth_rx_adapter_event_vector_config *config)
+{
+	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
+	struct eth_rx_queue_info *queue_info;
+	struct rte_event *qi_ev;
+
+	if (rx_queue_id == -1) {
+		uint16_t nb_rx_queues;
+		uint16_t i;
+
+		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
+		for (i = 0; i < nb_rx_queues; i++)
+			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
+						      config);
+		return;
+	}
+
+	queue_info = &dev_info->rx_queue[rx_queue_id];
+	qi_ev = (struct rte_event *)&queue_info->event;
+	queue_info->ena_vector = 1;
+	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+	rxa_set_vector_data(queue_info, config->vector_sz,
+			    config->vector_timeout_ns, config->vector_mp,
+			    rx_queue_id, dev_info->dev->data->port_id);
+	rx_adapter->ena_vector = 1;
+	rx_adapter->vector_tmo_ticks =
+		rx_adapter->vector_tmo_ticks ?
+			      RTE_MIN(config->vector_timeout_ns << 1,
+				      rx_adapter->vector_tmo_ticks) :
+			      config->vector_timeout_ns << 1;
+	rx_adapter->prev_expiry_ts = 0;
+	TAILQ_INIT(&rx_adapter->vector_list);
+}
+
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -2081,6 +2285,15 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}
 
+	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
+	    (queue_conf->rx_queue_flags &
+	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
+		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
 		(rx_queue_id != -1)) {
 		RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
@@ -2143,6 +2356,17 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	return 0;
 }
 
+static int
+rxa_sw_vector_limits(struct rte_event_eth_rx_adapter_vector_limits *limits)
+{
+	limits->max_sz = MAX_VECTOR_SIZE;
+	limits->min_sz = MIN_VECTOR_SIZE;
+	limits->max_timeout_ns = MAX_VECTOR_NS;
+	limits->min_timeout_ns = MIN_VECTOR_NS;
+
+	return 0;
+}
+
 int
 rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 				int32_t rx_queue_id)
@@ -2263,6 +2487,121 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }
 
+int
+rte_event_eth_rx_adapter_queue_event_vector_config(
+	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
+	struct rte_event_eth_rx_adapter_event_vector_config *config)
+{
+	struct rte_event_eth_rx_adapter_vector_limits limits;
+	struct rte_event_eth_rx_adapter *rx_adapter;
+	struct rte_eventdev *dev;
+	uint32_t cap;
+	int ret;
+
+	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
+
+	rx_adapter = rxa_id_to_adapter(id);
+	if ((rx_adapter == NULL) || (config == NULL))
+		return -EINVAL;
+
+	dev = &rte_eventdevs[rx_adapter->eventdev_id];
+	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
+						eth_dev_id, &cap);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
+				 "eth port %" PRIu16,
+				 id, eth_dev_id);
+		return ret;
+	}
+
+	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
+		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
+	ret = rte_event_eth_rx_adapter_vector_limits_get(
+		rx_adapter->eventdev_id, eth_dev_id, &limits);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
+				 "eth port %" PRIu16,
+				 rx_adapter->eventdev_id, eth_dev_id);
+		return ret;
+	}
+
+	if (config->vector_sz < limits.min_sz ||
+	    config->vector_sz > limits.max_sz ||
+	    config->vector_timeout_ns < limits.min_timeout_ns ||
+	    config->vector_timeout_ns > limits.max_timeout_ns ||
+	    config->vector_mp == NULL) {
+		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+	if (config->vector_mp->elt_size <
+	    (sizeof(struct rte_event_vector) +
+	     (sizeof(uintptr_t) * config->vector_sz))) {
+		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
+	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
+		RTE_FUNC_PTR_OR_ERR_RET(
+			*dev->dev_ops->eth_rx_adapter_event_vector_config,
+			-ENOTSUP);
+		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
+			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
+	} else {
+		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
+					      rx_queue_id, config);
+	}
+
+	return ret;
+}
+
+int
+rte_event_eth_rx_adapter_vector_limits_get(
+	uint8_t dev_id, uint16_t eth_port_id,
+	struct rte_event_eth_rx_adapter_vector_limits *limits)
+{
+	struct rte_eventdev *dev;
+	uint32_t cap;
+	int ret;
+
+	RTE_EVENTDEV_VALID_DEVID_OR_ERR_RET(dev_id, -EINVAL);
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_port_id, -EINVAL);
+
+	if (limits == NULL)
+		return -EINVAL;
+
+	dev = &rte_eventdevs[dev_id];
+
+	ret = rte_event_eth_rx_adapter_caps_get(dev_id, eth_port_id, &cap);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
+				 "eth port %" PRIu16,
+				 dev_id, eth_port_id);
+		return ret;
+	}
+
+	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
+		RTE_FUNC_PTR_OR_ERR_RET(
+			*dev->dev_ops->eth_rx_adapter_vector_limits_get,
+			-ENOTSUP);
+		ret = dev->dev_ops->eth_rx_adapter_vector_limits_get(
+			dev, &rte_eth_devices[eth_port_id], limits);
+	} else {
+		ret = rxa_sw_vector_limits(limits);
+	}
+
+	return ret;
+}
+
 int
 rte_event_eth_rx_adapter_start(uint8_t id)
 {
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index b57363f80..2e6e367e0 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -122,7 +122,11 @@ rte_event_eth_rx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
 
 	if (caps == NULL)
 		return -EINVAL;
-	*caps = 0;
+
+	if (dev->dev_ops->eth_rx_adapter_caps_get == NULL)
+		*caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
+	else
+		*caps = 0;
 
 	return dev->dev_ops->eth_rx_adapter_caps_get ?
 				(*dev->dev_ops->eth_rx_adapter_caps_get)(dev,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v3 5/8] eventdev: add Tx adapter event vector support
  2021-03-16 20:01   ` [dpdk-dev] [PATCH v3 0/8] Introduce event vectorization pbhagavatula
                       ` (3 preceding siblings ...)
  2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 4/8] eventdev: add Rx adapter event vector support pbhagavatula
@ 2021-03-16 20:01     ` pbhagavatula
  2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
                       ` (3 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-16 20:01 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support for event eth Tx adapter, the implementation
receives events from the single linked queue and based on
rte_event_vector::union_valid transmits the vector of mbufs to a given
port, queue pair.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 .../rte_event_eth_tx_adapter.c                | 66 ++++++++++++++++---
 lib/librte_eventdev/rte_eventdev.c            |  5 +-
 2 files changed, 60 insertions(+), 11 deletions(-)

diff --git a/lib/librte_eventdev/rte_event_eth_tx_adapter.c b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
index 5b4c42dcf..db260bfb6 100644
--- a/lib/librte_eventdev/rte_event_eth_tx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
@@ -510,6 +510,47 @@ txa_service_buffer_retry(struct rte_mbuf **pkts, uint16_t unsent,
 	stats->tx_dropped += unsent - sent;
 }
 
+static uint16_t
+txa_process_event_vector(struct txa_service_data *txa,
+			 struct rte_event_vector *vec)
+{
+	struct txa_service_queue_info *tqi;
+	uint16_t port, queue, nb_tx = 0;
+	struct rte_mbuf **mbufs;
+	int i;
+
+	mbufs = (struct rte_mbuf **)vec->mbufs;
+	if (vec->attr_valid) {
+		port = vec->port;
+		queue = vec->queue;
+		tqi = txa_service_queue(txa, port, queue);
+		if (unlikely(tqi == NULL || !tqi->added)) {
+			rte_pktmbuf_free_bulk(mbufs, vec->nb_elem);
+			rte_mempool_put(rte_mempool_from_obj(vec), vec);
+			return 0;
+		}
+		for (i = 0; i < vec->nb_elem; i++) {
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
+						   mbufs[i]);
+		}
+	} else {
+		for (i = 0; i < vec->nb_elem; i++) {
+			port = mbufs[i]->port;
+			queue = rte_event_eth_tx_adapter_txq_get(mbufs[i]);
+			tqi = txa_service_queue(txa, port, queue);
+			if (unlikely(tqi == NULL || !tqi->added)) {
+				rte_pktmbuf_free(mbufs[i]);
+				continue;
+			}
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
+						   mbufs[i]);
+		}
+	}
+	rte_mempool_put(rte_mempool_from_obj(vec), vec);
+
+	return nb_tx;
+}
+
 static void
 txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
 	uint32_t n)
@@ -522,22 +563,27 @@ txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
 
 	nb_tx = 0;
 	for (i = 0; i < n; i++) {
-		struct rte_mbuf *m;
 		uint16_t port;
 		uint16_t queue;
 		struct txa_service_queue_info *tqi;
 
-		m = ev[i].mbuf;
-		port = m->port;
-		queue = rte_event_eth_tx_adapter_txq_get(m);
+		if (!(ev[i].event_type & RTE_EVENT_TYPE_VECTOR)) {
+			struct rte_mbuf *m;
 
-		tqi = txa_service_queue(txa, port, queue);
-		if (unlikely(tqi == NULL || !tqi->added)) {
-			rte_pktmbuf_free(m);
-			continue;
-		}
+			m = ev[i].mbuf;
+			port = m->port;
+			queue = rte_event_eth_tx_adapter_txq_get(m);
 
-		nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
+			tqi = txa_service_queue(txa, port, queue);
+			if (unlikely(tqi == NULL || !tqi->added)) {
+				rte_pktmbuf_free(m);
+				continue;
+			}
+
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
+		} else {
+			nb_tx += txa_process_event_vector(txa, ev[i].vec);
+		}
 	}
 
 	stats->tx_packets += nb_tx;
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index 2e6e367e0..b2e85ce67 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -196,7 +196,10 @@ rte_event_eth_tx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
 	if (caps == NULL)
 		return -EINVAL;
 
-	*caps = 0;
+	if (dev->dev_ops->eth_tx_adapter_caps_get == NULL)
+		*caps = RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR;
+	else
+		*caps = 0;
 
 	return dev->dev_ops->eth_tx_adapter_caps_get ?
 			(*dev->dev_ops->eth_tx_adapter_caps_get)(dev,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v3 6/8] app/eventdev: add event vector mode in pipeline test
  2021-03-16 20:01   ` [dpdk-dev] [PATCH v3 0/8] Introduce event vectorization pbhagavatula
                       ` (4 preceding siblings ...)
  2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 5/8] eventdev: add Tx " pbhagavatula
@ 2021-03-16 20:01     ` pbhagavatula
  2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 7/8] doc: announce event Rx adapter config changes pbhagavatula
                       ` (2 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-16 20:01 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support in pipeline tests. By default this mode
is disabled, it can be enabled by using the option --enable_vector.
example:
	dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
	--nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
	--wlcores=20-23  --enable_vector

Addtional options to configure vector size and vector timeout are
also implemented and can be used by specifying --vector_size and
--vector_tmo_ns

This patch also adds a new option to set the number of Rx queues
configured per event eth rx adapter.
example:
	dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
	--nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
	--wlcores=20-23  --nb_eth_queues 4

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 app/test-eventdev/evt_common.h           |   4 +
 app/test-eventdev/evt_options.c          |  52 ++++
 app/test-eventdev/evt_options.h          |   4 +
 app/test-eventdev/test_pipeline_atq.c    | 310 ++++++++++++++++++++--
 app/test-eventdev/test_pipeline_common.c | 113 +++++++-
 app/test-eventdev/test_pipeline_common.h |  18 ++
 app/test-eventdev/test_pipeline_queue.c  | 320 +++++++++++++++++++++--
 doc/guides/tools/testeventdev.rst        |  28 ++
 8 files changed, 795 insertions(+), 54 deletions(-)

diff --git a/app/test-eventdev/evt_common.h b/app/test-eventdev/evt_common.h
index a1da1cf11..0e228258e 100644
--- a/app/test-eventdev/evt_common.h
+++ b/app/test-eventdev/evt_common.h
@@ -58,16 +58,20 @@ struct evt_options {
 	uint8_t sched_type_list[EVT_MAX_STAGES];
 	uint16_t mbuf_sz;
 	uint16_t wkr_deq_dep;
+	uint16_t vector_size;
+	uint16_t eth_queues;
 	uint32_t nb_flows;
 	uint32_t tx_first;
 	uint32_t max_pkt_sz;
 	uint32_t deq_tmo_nsec;
 	uint32_t q_priority:1;
 	uint32_t fwd_latency:1;
+	uint32_t ena_vector : 1;
 	uint64_t nb_pkts;
 	uint64_t nb_timers;
 	uint64_t expiry_nsec;
 	uint64_t max_tmo_nsec;
+	uint64_t vector_tmo_nsec;
 	uint64_t timer_tick_nsec;
 	uint64_t optm_timer_tick_nsec;
 	enum evt_prod_type prod_type;
diff --git a/app/test-eventdev/evt_options.c b/app/test-eventdev/evt_options.c
index 0d04ea9f8..0d5540574 100644
--- a/app/test-eventdev/evt_options.c
+++ b/app/test-eventdev/evt_options.c
@@ -34,6 +34,9 @@ evt_options_default(struct evt_options *opt)
 	opt->max_tmo_nsec = 1E5;  /* 100000ns ~100us */
 	opt->expiry_nsec = 1E4;   /* 10000ns ~10us */
 	opt->prod_type = EVT_PROD_TYPE_SYNT;
+	opt->eth_queues = 1;
+	opt->vector_size = 64;
+	opt->vector_tmo_nsec = 100E3;
 }
 
 typedef int (*option_parser_t)(struct evt_options *opt,
@@ -257,6 +260,43 @@ evt_parse_max_pkt_sz(struct evt_options *opt, const char *arg)
 	return ret;
 }
 
+static int
+evt_parse_ena_vector(struct evt_options *opt, const char *arg __rte_unused)
+{
+	opt->ena_vector = 1;
+	return 0;
+}
+
+static int
+evt_parse_vector_size(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint16(&(opt->vector_size), arg);
+
+	return ret;
+}
+
+static int
+evt_parse_vector_tmo_ns(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint64(&(opt->vector_tmo_nsec), arg);
+
+	return ret;
+}
+
+static int
+evt_parse_eth_queues(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint16(&(opt->eth_queues), arg);
+
+	return ret;
+}
+
 static void
 usage(char *program)
 {
@@ -289,6 +329,10 @@ usage(char *program)
 		"\t--expiry_nsec      : event timer expiry ns.\n"
 		"\t--mbuf_sz          : packet mbuf size.\n"
 		"\t--max_pkt_sz       : max packet size.\n"
+		"\t--nb_eth_queues    : number of ethernet Rx queues.\n"
+		"\t--enable_vector    : enable event vectorization.\n"
+		"\t--vector_size      : Max vector size.\n"
+		"\t--vector_tmo_ns    : Max vector timeout in nanoseconds\n"
 		);
 	printf("available tests:\n");
 	evt_test_dump_names();
@@ -360,6 +404,10 @@ static struct option lgopts[] = {
 	{ EVT_EXPIRY_NSEC,         1, 0, 0 },
 	{ EVT_MBUF_SZ,             1, 0, 0 },
 	{ EVT_MAX_PKT_SZ,          1, 0, 0 },
+	{ EVT_NB_ETH_QUEUES,       1, 0, 0 },
+	{ EVT_ENA_VECTOR,          0, 0, 0 },
+	{ EVT_VECTOR_SZ,           1, 0, 0 },
+	{ EVT_VECTOR_TMO,          1, 0, 0 },
 	{ EVT_HELP,                0, 0, 0 },
 	{ NULL,                    0, 0, 0 }
 };
@@ -394,6 +442,10 @@ evt_opts_parse_long(int opt_idx, struct evt_options *opt)
 		{ EVT_EXPIRY_NSEC, evt_parse_expiry_nsec},
 		{ EVT_MBUF_SZ, evt_parse_mbuf_sz},
 		{ EVT_MAX_PKT_SZ, evt_parse_max_pkt_sz},
+		{ EVT_NB_ETH_QUEUES, evt_parse_eth_queues},
+		{ EVT_ENA_VECTOR, evt_parse_ena_vector},
+		{ EVT_VECTOR_SZ, evt_parse_vector_size},
+		{ EVT_VECTOR_TMO, evt_parse_vector_tmo_ns},
 	};
 
 	for (i = 0; i < RTE_DIM(parsermap); i++) {
diff --git a/app/test-eventdev/evt_options.h b/app/test-eventdev/evt_options.h
index 748e54fae..1cea2a3e1 100644
--- a/app/test-eventdev/evt_options.h
+++ b/app/test-eventdev/evt_options.h
@@ -42,6 +42,10 @@
 #define EVT_EXPIRY_NSEC          ("expiry_nsec")
 #define EVT_MBUF_SZ              ("mbuf_sz")
 #define EVT_MAX_PKT_SZ           ("max_pkt_sz")
+#define EVT_NB_ETH_QUEUES        ("nb_eth_queues")
+#define EVT_ENA_VECTOR           ("enable_vector")
+#define EVT_VECTOR_SZ            ("vector_size")
+#define EVT_VECTOR_TMO           ("vector_tmo_ns")
 #define EVT_HELP                 ("help")
 
 void evt_options_default(struct evt_options *opt);
diff --git a/app/test-eventdev/test_pipeline_atq.c b/app/test-eventdev/test_pipeline_atq.c
index 0872b25b5..84dd4f44e 100644
--- a/app/test-eventdev/test_pipeline_atq.c
+++ b/app/test-eventdev/test_pipeline_atq.c
@@ -15,6 +15,8 @@ pipeline_atq_nb_event_queues(struct evt_options *opt)
 	return rte_eth_dev_count_avail();
 }
 
+typedef int (*pipeline_atq_worker_t)(void *arg);
+
 static __rte_noinline int
 pipeline_atq_worker_single_stage_tx(void *arg)
 {
@@ -113,6 +115,112 @@ pipeline_atq_worker_single_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_atq_worker_single_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+		vector_sz = ev.vec->nb_elem;
+		pipeline_event_tx_vector(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = ev.vec->nb_elem;
+		ev.queue_id = tx_queue[ev.vec->port];
+		ev.vec->queue = 0;
+		pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			vector_sz += ev[i].vec->nb_elem;
+			ev[i].vec->queue = 0;
+		}
+
+		pipeline_event_tx_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			ev[i].queue_id = tx_queue[ev[i].vec->port];
+			ev[i].vec->queue = 0;
+			vector_sz += ev[i].vec->nb_elem;
+			pipeline_fwd_event_vector(&ev[i],
+						  RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
 static __rte_noinline int
 pipeline_atq_worker_multi_stage_tx(void *arg)
 {
@@ -245,6 +353,147 @@ pipeline_atq_worker_multi_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.sub_event_type % nb_stages;
+
+		if (cq_id == last_queue) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+			continue;
+		}
+
+		ev.sub_event_type++;
+		pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.sub_event_type % nb_stages;
+
+		if (cq_id == last_queue) {
+			ev.queue_id = tx_queue[ev.vec->port];
+			ev.vec->queue = 0;
+			vector_sz = ev.vec->nb_elem;
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			pipeline_event_enqueue(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.sub_event_type++;
+			pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+			pipeline_event_enqueue(dev, port, &ev);
+		}
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].sub_event_type % nb_stages;
+
+			if (cq_id == last_queue) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+				continue;
+			}
+
+			ev[i].sub_event_type++;
+			pipeline_fwd_event_vector(&ev[i],
+						  sched_type_list[cq_id]);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].sub_event_type % nb_stages;
+
+			if (cq_id == last_queue) {
+				vector_sz = ev[i].vec->nb_elem;
+				ev[i].queue_id = tx_queue[ev[i].vec->port];
+				ev[i].vec->queue = 0;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].sub_event_type++;
+				pipeline_fwd_event_vector(
+					&ev[i], sched_type_list[cq_id]);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
 static int
 worker_wrapper(void *arg)
 {
@@ -253,27 +502,36 @@ worker_wrapper(void *arg)
 	const bool burst = evt_has_burst_mode(w->dev_id);
 	const bool internal_port = w->t->internal_port;
 	const uint8_t nb_stages = opt->nb_stages;
-	RTE_SET_USED(opt);
+	/*vector/burst/internal_port*/
+	const pipeline_atq_worker_t
+	pipeline_atq_worker_single_stage[2][2][2] = {
+		[0][0][0] = pipeline_atq_worker_single_stage_fwd,
+		[0][0][1] = pipeline_atq_worker_single_stage_tx,
+		[0][1][0] = pipeline_atq_worker_single_stage_burst_fwd,
+		[0][1][1] = pipeline_atq_worker_single_stage_burst_tx,
+		[1][0][0] = pipeline_atq_worker_single_stage_fwd_vector,
+		[1][0][1] = pipeline_atq_worker_single_stage_tx_vector,
+		[1][1][0] = pipeline_atq_worker_single_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_atq_worker_single_stage_burst_tx_vector,
+	};
+	const pipeline_atq_worker_t
+	pipeline_atq_worker_multi_stage[2][2][2] = {
+		[0][0][0] = pipeline_atq_worker_multi_stage_fwd,
+		[0][0][1] = pipeline_atq_worker_multi_stage_tx,
+		[0][1][0] = pipeline_atq_worker_multi_stage_burst_fwd,
+		[0][1][1] = pipeline_atq_worker_multi_stage_burst_tx,
+		[1][0][0] = pipeline_atq_worker_multi_stage_fwd_vector,
+		[1][0][1] = pipeline_atq_worker_multi_stage_tx_vector,
+		[1][1][0] = pipeline_atq_worker_multi_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_atq_worker_multi_stage_burst_tx_vector,
+	};
 
-	if (nb_stages == 1) {
-		if (!burst && internal_port)
-			return pipeline_atq_worker_single_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_atq_worker_single_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_atq_worker_single_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_atq_worker_single_stage_burst_fwd(arg);
-	} else {
-		if (!burst && internal_port)
-			return pipeline_atq_worker_multi_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_atq_worker_multi_stage_fwd(arg);
-		if (burst && internal_port)
-			return pipeline_atq_worker_multi_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_atq_worker_multi_stage_burst_fwd(arg);
-	}
+	if (nb_stages == 1)
+		return (pipeline_atq_worker_single_stage[opt->ena_vector][burst]
+							[internal_port])(arg);
+	else
+		return (pipeline_atq_worker_multi_stage[opt->ena_vector][burst]
+						       [internal_port])(arg);
 
 	rte_panic("invalid worker\n");
 }
@@ -290,7 +548,7 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
 	int ret;
 	int nb_ports;
 	int nb_queues;
-	uint8_t queue;
+	uint8_t queue, is_prod;
 	uint8_t tx_evqueue_id[RTE_MAX_ETHPORTS];
 	uint8_t queue_arr[RTE_EVENT_MAX_QUEUES_PER_DEV];
 	uint8_t nb_worker_queues = 0;
@@ -330,15 +588,19 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
 		q_conf.event_queue_cfg = RTE_EVENT_QUEUE_CFG_ALL_TYPES;
 
 		if (!t->internal_port) {
+			is_prod = false;
 			RTE_ETH_FOREACH_DEV(prod) {
 				if (queue == tx_evqueue_id[prod]) {
 					q_conf.event_queue_cfg =
 						RTE_EVENT_QUEUE_CFG_SINGLE_LINK;
-				} else {
-					queue_arr[nb_worker_queues] = queue;
-					nb_worker_queues++;
+					is_prod = true;
+					break;
 				}
 			}
+			if (!is_prod) {
+				queue_arr[nb_worker_queues] = queue;
+				nb_worker_queues++;
+			}
 		}
 
 		ret = rte_event_queue_setup(opt->dev_id, queue, &q_conf);
diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index b47d76743..d5ef90500 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -36,6 +36,12 @@ pipeline_opt_dump(struct evt_options *opt, uint8_t nb_queues)
 	evt_dump_queue_priority(opt);
 	evt_dump_sched_type_list(opt);
 	evt_dump_producer_type(opt);
+	evt_dump("nb_eth_rx_queues", "%d", opt->eth_queues);
+	evt_dump("event_vector", "%d", opt->ena_vector);
+	if (opt->ena_vector) {
+		evt_dump("vector_size", "%d", opt->vector_size);
+		evt_dump("vector_tmo_ns", "%" PRIu64 "", opt->vector_tmo_nsec);
+	}
 }
 
 static inline uint64_t
@@ -163,7 +169,7 @@ pipeline_opt_check(struct evt_options *opt, uint64_t nb_queues)
 int
 pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 {
-	uint16_t i;
+	uint16_t i, j;
 	int ret;
 	uint8_t nb_queues = 1;
 	struct test_pipeline *t = evt_test_priv(test);
@@ -210,6 +216,16 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 		if (!(caps & RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT))
 			t->internal_port = 0;
 
+		ret = rte_event_eth_rx_adapter_caps_get(opt->dev_id, i, &caps);
+		if (ret != 0) {
+			evt_err("failed to get event tx adapter[%d] caps", i);
+			return ret;
+		}
+
+		if (!(caps & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT))
+			local_port_conf.rxmode.offloads |=
+				DEV_RX_OFFLOAD_RSS_HASH;
+
 		ret = rte_eth_dev_info_get(i, &dev_info);
 		if (ret != 0) {
 			evt_err("Error during getting device (port %u) info: %s\n",
@@ -236,19 +252,22 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 				local_port_conf.rx_adv_conf.rss_conf.rss_hf);
 		}
 
-		if (rte_eth_dev_configure(i, nb_queues, nb_queues,
-					&local_port_conf)
-				< 0) {
+		if (rte_eth_dev_configure(i, opt->eth_queues, nb_queues,
+					  &local_port_conf) < 0) {
 			evt_err("Failed to configure eth port [%d]", i);
 			return -EINVAL;
 		}
 
-		if (rte_eth_rx_queue_setup(i, 0, NB_RX_DESC,
-				rte_socket_id(), &rx_conf, t->pool) < 0) {
-			evt_err("Failed to setup eth port [%d] rx_queue: %d.",
+		for (j = 0; j < opt->eth_queues; j++) {
+			if (rte_eth_rx_queue_setup(i, j, NB_RX_DESC,
+						   rte_socket_id(), &rx_conf,
+						   t->pool) < 0) {
+				evt_err("Failed to setup eth port [%d] rx_queue: %d.",
 					i, 0);
-			return -EINVAL;
+				return -EINVAL;
+			}
 		}
+
 		if (rte_eth_tx_queue_setup(i, 0, NB_TX_DESC,
 					rte_socket_id(), NULL) < 0) {
 			evt_err("Failed to setup eth port [%d] tx_queue: %d.",
@@ -310,12 +329,27 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 {
 	int ret = 0;
 	uint16_t prod;
+	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
+	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;
 
 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
 	queue_conf.ev.sched_type = opt->sched_type_list[0];
+	if (opt->ena_vector) {
+		unsigned int nb_elem = (opt->pool_sz / opt->vector_size) << 1;
+
+		nb_elem = nb_elem ? nb_elem : 1;
+		vector_pool = rte_event_vector_pool_create(
+			"vector_pool", nb_elem, 0, opt->vector_size,
+			opt->socket_id);
+		if (vector_pool == NULL) {
+			evt_err("failed to create event vector pool");
+			return -ENOMEM;
+		}
+	}
 	RTE_ETH_FOREACH_DEV(prod) {
+		struct rte_event_eth_rx_adapter_vector_limits limits;
 		uint32_t cap;
 
 		ret = rte_event_eth_rx_adapter_caps_get(opt->dev_id,
@@ -326,6 +360,50 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 					opt->dev_id);
 			return ret;
 		}
+
+		if (opt->ena_vector) {
+			memset(&limits, 0, sizeof(limits));
+			ret = rte_event_eth_rx_adapter_vector_limits_get(
+				opt->dev_id, prod, &limits);
+			if (ret) {
+				evt_err("failed to get vector limits");
+				return ret;
+			}
+
+			if (opt->vector_size < limits.min_sz ||
+			    opt->vector_size > limits.max_sz) {
+				evt_err("Vector size [%d] not within limits max[%d] min[%d]",
+					opt->vector_size, limits.min_sz,
+					limits.max_sz);
+				return -EINVAL;
+			}
+
+			if (limits.log2_sz &&
+			    !rte_is_power_of_2(opt->vector_size)) {
+				evt_err("Vector size [%d] not power of 2",
+					opt->vector_size);
+				return -EINVAL;
+			}
+
+			if (opt->vector_tmo_nsec > limits.max_timeout_ns ||
+			    opt->vector_tmo_nsec < limits.min_timeout_ns) {
+				evt_err("Vector timeout [%" PRIu64
+					"] not within limits max[%" PRIu64
+					"] min[%" PRIu64 "]",
+					opt->vector_tmo_nsec,
+					limits.max_timeout_ns,
+					limits.min_timeout_ns);
+				return -EINVAL;
+			}
+
+			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.rx_queue_flags |=
+				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+			} else {
+				evt_err("Rx adapter doesn't support event vector");
+				return -EINVAL;
+			}
+		}
 		queue_conf.ev.queue_id = prod * stride;
 		ret = rte_event_eth_rx_adapter_create(prod, opt->dev_id,
 				&prod_conf);
@@ -340,6 +418,17 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}
 
+		if (opt->ena_vector) {
+			vec_conf.vector_sz = opt->vector_size;
+			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
+			vec_conf.vector_mp = vector_pool;
+			if (rte_event_eth_rx_adapter_queue_event_vector_config(
+				    prod, prod, -1, &vec_conf) < 0) {
+				evt_err("Failed to configure event vectorization for Rx adapter");
+				return -EINVAL;
+			}
+		}
+
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;
 
@@ -378,6 +467,14 @@ pipeline_event_tx_adapter_setup(struct evt_options *opt,
 			return ret;
 		}
 
+		if (opt->ena_vector) {
+			if (!(cap &
+			      RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR)) {
+				evt_err("Tx adapter doesn't support event vector");
+				return -EINVAL;
+			}
+		}
+
 		ret = rte_event_eth_tx_adapter_create(consm, opt->dev_id,
 				&port_conf);
 		if (ret) {
diff --git a/app/test-eventdev/test_pipeline_common.h b/app/test-eventdev/test_pipeline_common.h
index 6e73c6ab2..800a90616 100644
--- a/app/test-eventdev/test_pipeline_common.h
+++ b/app/test-eventdev/test_pipeline_common.h
@@ -101,6 +101,14 @@ pipeline_fwd_event(struct rte_event *ev, uint8_t sched)
 	ev->sched_type = sched;
 }
 
+static __rte_always_inline void
+pipeline_fwd_event_vector(struct rte_event *ev, uint8_t sched)
+{
+	ev->event_type = RTE_EVENT_TYPE_CPU_VECTOR;
+	ev->op = RTE_EVENT_OP_FORWARD;
+	ev->sched_type = sched;
+}
+
 static __rte_always_inline void
 pipeline_event_tx(const uint8_t dev, const uint8_t port,
 		struct rte_event * const ev)
@@ -110,6 +118,16 @@ pipeline_event_tx(const uint8_t dev, const uint8_t port,
 		rte_pause();
 }
 
+static __rte_always_inline void
+pipeline_event_tx_vector(const uint8_t dev, const uint8_t port,
+			 struct rte_event *const ev)
+{
+	ev->vec->queue = 0;
+
+	while (!rte_event_eth_tx_adapter_enqueue(dev, port, ev, 1, 0))
+		rte_pause();
+}
+
 static __rte_always_inline void
 pipeline_event_tx_burst(const uint8_t dev, const uint8_t port,
 		struct rte_event *ev, const uint16_t nb_rx)
diff --git a/app/test-eventdev/test_pipeline_queue.c b/app/test-eventdev/test_pipeline_queue.c
index 9a9febb19..f6cc3e358 100644
--- a/app/test-eventdev/test_pipeline_queue.c
+++ b/app/test-eventdev/test_pipeline_queue.c
@@ -15,6 +15,8 @@ pipeline_queue_nb_event_queues(struct evt_options *opt)
 	return (eth_count * opt->nb_stages) + eth_count;
 }
 
+typedef int (*pipeline_queue_worker_t)(void *arg);
+
 static __rte_noinline int
 pipeline_queue_worker_single_stage_tx(void *arg)
 {
@@ -126,6 +128,125 @@ pipeline_queue_worker_single_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_queue_worker_single_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		if (ev.sched_type == RTE_SCHED_TYPE_ATOMIC) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.queue_id++;
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			pipeline_event_enqueue(dev, port, &ev);
+		}
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		ev.queue_id = tx_queue[ev.vec->port];
+		ev.vec->queue = 0;
+		vector_sz = ev.vec->nb_elem;
+		pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			if (ev[i].sched_type == RTE_SCHED_TYPE_ATOMIC) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].queue_id++;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			ev[i].queue_id = tx_queue[ev[i].vec->port];
+			ev[i].vec->queue = 0;
+			vector_sz += ev[i].vec->nb_elem;
+			pipeline_fwd_event_vector(&ev[i],
+						  RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
 
 static __rte_noinline int
 pipeline_queue_worker_multi_stage_tx(void *arg)
@@ -267,6 +388,151 @@ pipeline_queue_worker_multi_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.queue_id % nb_stages;
+
+		if (ev.queue_id == tx_queue[ev.vec->port]) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+			continue;
+		}
+
+		ev.queue_id++;
+		pipeline_fwd_event_vector(&ev, cq_id != last_queue
+						       ? sched_type_list[cq_id]
+						       : RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.queue_id % nb_stages;
+
+		if (cq_id == last_queue) {
+			vector_sz = ev.vec->nb_elem;
+			ev.queue_id = tx_queue[ev.vec->port];
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.queue_id++;
+			pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+		}
+
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].queue_id % nb_stages;
+
+			if (ev[i].queue_id == tx_queue[ev[i].vec->port]) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+				continue;
+			}
+
+			ev[i].queue_id++;
+			pipeline_fwd_event_vector(
+				&ev[i], cq_id != last_queue
+						? sched_type_list[cq_id]
+						: RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].queue_id % nb_stages;
+
+			if (cq_id == last_queue) {
+				ev[i].queue_id = tx_queue[ev[i].vec->port];
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].queue_id++;
+				pipeline_fwd_event_vector(
+					&ev[i], sched_type_list[cq_id]);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
 static int
 worker_wrapper(void *arg)
 {
@@ -275,29 +541,39 @@ worker_wrapper(void *arg)
 	const bool burst = evt_has_burst_mode(w->dev_id);
 	const bool internal_port = w->t->internal_port;
 	const uint8_t nb_stages = opt->nb_stages;
-	RTE_SET_USED(opt);
-
-	if (nb_stages == 1) {
-		if (!burst && internal_port)
-			return pipeline_queue_worker_single_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_queue_worker_single_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_queue_worker_single_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_queue_worker_single_stage_burst_fwd(
-					arg);
-	} else {
-		if (!burst && internal_port)
-			return pipeline_queue_worker_multi_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_queue_worker_multi_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_queue_worker_multi_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_queue_worker_multi_stage_burst_fwd(arg);
+	/*vector/burst/internal_port*/
+	const pipeline_queue_worker_t
+	pipeline_queue_worker_single_stage[2][2][2] = {
+		[0][0][0] = pipeline_queue_worker_single_stage_fwd,
+		[0][0][1] = pipeline_queue_worker_single_stage_tx,
+		[0][1][0] = pipeline_queue_worker_single_stage_burst_fwd,
+		[0][1][1] = pipeline_queue_worker_single_stage_burst_tx,
+		[1][0][0] = pipeline_queue_worker_single_stage_fwd_vector,
+		[1][0][1] = pipeline_queue_worker_single_stage_tx_vector,
+		[1][1][0] = pipeline_queue_worker_single_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_queue_worker_single_stage_burst_tx_vector,
+	};
+	const pipeline_queue_worker_t
+	pipeline_queue_worker_multi_stage[2][2][2] = {
+		[0][0][0] = pipeline_queue_worker_multi_stage_fwd,
+		[0][0][1] = pipeline_queue_worker_multi_stage_tx,
+		[0][1][0] = pipeline_queue_worker_multi_stage_burst_fwd,
+		[0][1][1] = pipeline_queue_worker_multi_stage_burst_tx,
+		[1][0][0] = pipeline_queue_worker_multi_stage_fwd_vector,
+		[1][0][1] = pipeline_queue_worker_multi_stage_tx_vector,
+		[1][1][0] = pipeline_queue_worker_multi_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_queue_worker_multi_stage_burst_tx_vector,
+	};
+
+	if (nb_stages == 1)
+		return (pipeline_queue_worker_single_stage[opt->ena_vector]
+							  [burst]
+							  [internal_port])(arg);
+	else
+		return (pipeline_queue_worker_multi_stage[opt->ena_vector]
+							 [burst]
+							 [internal_port])(arg);
 
-	}
 	rte_panic("invalid worker\n");
 }
 
diff --git a/doc/guides/tools/testeventdev.rst b/doc/guides/tools/testeventdev.rst
index ad1788a3d..691cf706e 100644
--- a/doc/guides/tools/testeventdev.rst
+++ b/doc/guides/tools/testeventdev.rst
@@ -158,6 +158,26 @@ The following are the application command-line options:
        Set max packet mbuf size. Can be used configure Rx/Tx scatter gather.
        Only applicable for `pipeline_atq` and `pipeline_queue` tests.
 
+* ``--nb_eth_queues``
+
+       Configure multiple Rx queues per each ethernet port.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--enable_vector``
+
+       Enable event vector for Rx/Tx adapters.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--vector_size``
+
+       Vector size to configure for the Rx adapter.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--vector_tmo_ns``
+
+       Vector timeout nanoseconds to be configured for the Rx adapter.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
 
 Eventdev Tests
 --------------
@@ -607,6 +627,10 @@ Supported application command line options are following::
         --worker_deq_depth
         --prod_type_ethdev
         --deq_tmo_nsec
+        --nb_eth_queues
+        --enable_vector
+        --vector_size
+        --vector_tmo_ns
 
 
 .. Note::
@@ -699,6 +723,10 @@ Supported application command line options are following::
         --worker_deq_depth
         --prod_type_ethdev
         --deq_tmo_nsec
+        --nb_eth_queues
+        --enable_vector
+        --vector_size
+        --vector_tmo_ns
 
 
 .. Note::
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v3 7/8] doc: announce event Rx adapter config changes
  2021-03-16 20:01   ` [dpdk-dev] [PATCH v3 0/8] Introduce event vectorization pbhagavatula
                       ` (5 preceding siblings ...)
  2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
@ 2021-03-16 20:01     ` pbhagavatula
  2021-03-16 20:01     ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v3 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
  2021-03-19 20:57     ` [dpdk-dev] [PATCH v4 0/8] Introduce event vectorization pbhagavatula
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-16 20:01 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

The Rx adapter event vector configuration will be merged into
Rx adapter queue configuration to simplify enabling event
vectorization.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 doc/guides/rel_notes/deprecation.rst | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst
index 64629e064..ef6d1bb18 100644
--- a/doc/guides/rel_notes/deprecation.rst
+++ b/doc/guides/rel_notes/deprecation.rst
@@ -130,3 +130,12 @@ Deprecation Notices
 * cmdline: ``cmdline`` structure will be made opaque to hide platform-specific
   content. On Linux and FreeBSD, supported prior to DPDK 20.11,
   original structure will be kept until DPDK 21.11.
+
+* eventdev: The structure ``rte_event_eth_rx_adapter_queue_conf`` will be
+  extended to include ``rte_event_eth_rx_adapter_event_vector_config`` elements
+  and the function ``rte_event_eth_rx_adapter_queue_event_vector_config`` will
+  be removed in DPDK 21.11.
+
+  An application can enable event vectorization by passing the desired vector
+  values to the function ``rte_event_eth_rx_adapter_queue_add`` using
+  the structure ``rte_event_eth_rx_adapter_queue_add``.
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [dpdk-dev v21.11] [PATCH v3 8/8] eventdev: simplify Rx adapter event vector config
  2021-03-16 20:01   ` [dpdk-dev] [PATCH v3 0/8] Introduce event vectorization pbhagavatula
                       ` (6 preceding siblings ...)
  2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 7/8] doc: announce event Rx adapter config changes pbhagavatula
@ 2021-03-16 20:01     ` pbhagavatula
  2021-03-19 20:57     ` [dpdk-dev] [PATCH v4 0/8] Introduce event vectorization pbhagavatula
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-16 20:01 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Include vector configuration into the structure
``rte_event_eth_rx_adapter_queue_conf`` used when configuring rest
of the Rx adapter ethernet device Rx queue parameters.
This simplifies event vector configuration as it avoids splitting
configuration per Rx queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 app/test-eventdev/test_pipeline_common.c      |  16 +-
 lib/librte_eventdev/eventdev_pmd.h            |  29 ---
 .../rte_event_eth_rx_adapter.c                | 168 ++++++------------
 .../rte_event_eth_rx_adapter.h                |  27 ---
 lib/librte_eventdev/version.map               |   1 -
 5 files changed, 57 insertions(+), 184 deletions(-)

diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index d5ef90500..76aee254b 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -331,7 +331,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 	uint16_t prod;
 	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
-	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
@@ -397,8 +396,12 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			}

 			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.vector_sz = opt->vector_size;
+				queue_conf.vector_timeout_ns =
+					opt->vector_tmo_nsec;
 				queue_conf.rx_queue_flags |=
 				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+				queue_conf.vector_mp = vector_pool;
 			} else {
 				evt_err("Rx adapter doesn't support event vector");
 				return -EINVAL;
@@ -418,17 +421,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}

-		if (opt->ena_vector) {
-			vec_conf.vector_sz = opt->vector_size;
-			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
-			vec_conf.vector_mp = vector_pool;
-			if (rte_event_eth_rx_adapter_queue_event_vector_config(
-				    prod, prod, -1, &vec_conf) < 0) {
-				evt_err("Failed to configure event vectorization for Rx adapter");
-				return -EINVAL;
-			}
-		}
-
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;

diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index d79dfd612..60bfaebc0 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -667,32 +667,6 @@ typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
 	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-struct rte_event_eth_rx_adapter_event_vector_config;
-/**
- * Enable event vector on an given Rx queue of a ethernet devices belonging to
- * the Rx adapter.
- *
- * @param dev
- *   Event device pointer
- *
- * @param eth_dev
- *   Ethernet device pointer
- *
- * @param rx_queue_id
- *   The Rx queue identifier
- *
- * @param config
- *   Pointer to the event vector configuration structure.
- *
- * @return
- *   - 0: Success.
- *   - <0: Error code returned by the driver function.
- */
-typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
-	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
-	int32_t rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;

@@ -1118,9 +1092,6 @@ struct rte_eventdev_ops {
 	eventdev_eth_rx_adapter_vector_limits_get_t
 		eth_rx_adapter_vector_limits_get;
 	/**< Get event vector limits for the Rx adapter */
-	eventdev_eth_rx_adapter_event_vector_config_t
-		eth_rx_adapter_event_vector_config;
-	/**< Configure Rx adapter with event vector */

 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index c71990078..a1990637f 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -1882,6 +1882,25 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	} else
 		qi_ev->flow_id = 0;

+	if (conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+		queue_info->ena_vector = 1;
+		qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+		rxa_set_vector_data(queue_info, conf->vector_sz,
+				    conf->vector_timeout_ns, conf->vector_mp,
+				    rx_queue_id, dev_info->dev->data->port_id);
+		rx_adapter->ena_vector = 1;
+		rx_adapter->vector_tmo_ticks =
+			rx_adapter->vector_tmo_ticks
+				? RTE_MIN(queue_info->vector_data
+						  .vector_timeout_ticks,
+					  rx_adapter->vector_tmo_ticks)
+				: queue_info->vector_data.vector_timeout_ticks;
+		rx_adapter->vector_tmo_ticks <<= 1;
+		TAILQ_INIT(&rx_adapter->vector_list);
+		rx_adapter->prev_expiry_ts = 0;
+	}
+
 	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
 	if (rxa_polled_queue(dev_info, rx_queue_id)) {
 		rx_adapter->num_rx_polled += !pollq;
@@ -1907,44 +1926,6 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }

-static void
-rxa_sw_event_vector_configure(
-	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
-	int rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
-	struct eth_rx_queue_info *queue_info;
-	struct rte_event *qi_ev;
-
-	if (rx_queue_id == -1) {
-		uint16_t nb_rx_queues;
-		uint16_t i;
-
-		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
-		for (i = 0; i < nb_rx_queues; i++)
-			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
-						      config);
-		return;
-	}
-
-	queue_info = &dev_info->rx_queue[rx_queue_id];
-	qi_ev = (struct rte_event *)&queue_info->event;
-	queue_info->ena_vector = 1;
-	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
-	rxa_set_vector_data(queue_info, config->vector_sz,
-			    config->vector_timeout_ns, config->vector_mp,
-			    rx_queue_id, dev_info->dev->data->port_id);
-	rx_adapter->ena_vector = 1;
-	rx_adapter->vector_tmo_ticks =
-		rx_adapter->vector_tmo_ticks ?
-			      RTE_MIN(config->vector_timeout_ns << 1,
-				      rx_adapter->vector_tmo_ticks) :
-			      config->vector_timeout_ns << 1;
-	rx_adapter->prev_expiry_ts = 0;
-	TAILQ_INIT(&rx_adapter->vector_list);
-}
-
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -2258,6 +2239,7 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	struct rte_event_eth_rx_adapter *rx_adapter;
 	struct rte_eventdev *dev;
 	struct eth_device_info *dev_info;
+	struct rte_event_eth_rx_adapter_vector_limits limits;

 	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
 	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
@@ -2294,6 +2276,39 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}

+	if (queue_conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+		ret = rte_event_eth_rx_adapter_vector_limits_get(
+			rx_adapter->eventdev_id, eth_dev_id, &limits);
+		if (ret < 0) {
+			RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_sz < limits.min_sz ||
+		    queue_conf->vector_sz > limits.max_sz ||
+		    queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
+		    queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
+		    queue_conf->vector_mp == NULL) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_mp->elt_size <
+		    (sizeof(struct rte_event_vector) +
+		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+	}
+
 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
 		(rx_queue_id != -1)) {
 		RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
@@ -2487,83 +2502,6 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }

-int
-rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct rte_event_eth_rx_adapter_vector_limits limits;
-	struct rte_event_eth_rx_adapter *rx_adapter;
-	struct rte_eventdev *dev;
-	uint32_t cap;
-	int ret;
-
-	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
-	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
-
-	rx_adapter = rxa_id_to_adapter(id);
-	if ((rx_adapter == NULL) || (config == NULL))
-		return -EINVAL;
-
-	dev = &rte_eventdevs[rx_adapter->eventdev_id];
-	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
-						eth_dev_id, &cap);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
-				 "eth port %" PRIu16,
-				 id, eth_dev_id);
-		return ret;
-	}
-
-	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
-		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	ret = rte_event_eth_rx_adapter_vector_limits_get(
-		rx_adapter->eventdev_id, eth_dev_id, &limits);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
-				 "eth port %" PRIu16,
-				 rx_adapter->eventdev_id, eth_dev_id);
-		return ret;
-	}
-
-	if (config->vector_sz < limits.min_sz ||
-	    config->vector_sz > limits.max_sz ||
-	    config->vector_timeout_ns < limits.min_timeout_ns ||
-	    config->vector_timeout_ns > limits.max_timeout_ns ||
-	    config->vector_mp == NULL) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-	if (config->vector_mp->elt_size <
-	    (sizeof(struct rte_event_vector) +
-	     (sizeof(uintptr_t) * config->vector_sz))) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
-		RTE_FUNC_PTR_OR_ERR_RET(
-			*dev->dev_ops->eth_rx_adapter_event_vector_config,
-			-ENOTSUP);
-		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
-			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
-	} else {
-		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
-					      rx_queue_id, config);
-	}
-
-	return ret;
-}
-
 int
 rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
index dda846f57..3f9c749e9 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
@@ -171,9 +171,6 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
 	 * enqueued event.
 	 */
-};
-
-struct rte_event_eth_rx_adapter_event_vector_config {
 	uint16_t vector_sz;
 	/**<
 	 * Indicates the maximum number for mbufs to combine and form a vector.
@@ -538,30 +535,6 @@ int rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-/**
- * Configure event vectorization for a given ethernet device queue, that has
- * been added to a event eth Rx adapter.
- *
- * @param id
- *  The identifier of the ethernet Rx event adapter.
- *
- * @param eth_dev_id
- *  The identifier of the ethernet device.
- *
- * @param rx_queue_id
- *  Ethernet device receive queue index.
- *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
- *  are configured with event vectorization.
- *
- * @return
- *  - 0: Success, Receive queue configured correctly.
- *  - <0: Error code on failure.
- */
-__rte_experimental
-int rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index 902df0ae3..34c1c830e 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -142,7 +142,6 @@ EXPERIMENTAL {
 	#added in 21.05
 	rte_event_vector_pool_create;
 	rte_event_eth_rx_adapter_vector_limits_get;
-	rte_event_eth_rx_adapter_queue_event_vector_config;
 };

 INTERNAL {
--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v3 1/8] eventdev: introduce event vector capability
  2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 1/8] eventdev: introduce event vector capability pbhagavatula
@ 2021-03-18  6:19       ` Jayatheerthan, Jay
  2021-03-18  6:23         ` Pavan Nikhilesh Bhagavatula
  0 siblings, 1 reply; 153+ messages in thread
From: Jayatheerthan, Jay @ 2021-03-18  6:19 UTC (permalink / raw)
  To: pbhagavatula, jerinj, Carrillo, Erik G, Gujjar, Abhinandan S,
	McDaniel, Timothy, hemant.agrawal, Van Haaren, Harry,
	mattias.ronnblom, Ma, Liang J, Ray Kinsella, Neil Horman
  Cc: dev

Posting my comments in beginning. No further comments.

@pbhagavatula@marvell.com, it seems like v3 patch1 drops some changes from v2 patch1 (for e.g. missing event vector pointer in rte_event) causing build failure. Could you have a look?

Thanks!

-Jay


> -----Original Message-----
> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
> Sent: Wednesday, March 17, 2021 1:32 AM
> To: jerinj@marvell.com; Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Carrillo, Erik G <erik.g.carrillo@intel.com>; Gujjar,
> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy <timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom <mattias.ronnblom@ericsson.com>; Ma, Liang J
> <liang.j.ma@intel.com>; Ray Kinsella <mdr@ashroe.eu>; Neil Horman <nhorman@tuxdriver.com>
> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
> Subject: [dpdk-dev] [PATCH v3 1/8] eventdev: introduce event vector capability
> 
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> Introduce rte_event_vector datastructure which is capable of holding
> multiple uintptr_t of the same flow thereby allowing applications
> to vectorize their pipeline and reducing the complexity of pipelining
> the events across multiple stages.
> This approach also reduces the scheduling overhead on a event device.
> 
> Add a event vector mempool create handler to create mempools based on
> the best mempool ops available on a given platform.
> 
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> ---
>  doc/guides/prog_guide/eventdev.rst |  36 +++++++++-
>  lib/librte_eventdev/rte_eventdev.h | 110 ++++++++++++++++++++++++++++-
>  lib/librte_eventdev/version.map    |   3 +
>  3 files changed, 146 insertions(+), 3 deletions(-)
> 
> diff --git a/doc/guides/prog_guide/eventdev.rst b/doc/guides/prog_guide/eventdev.rst
> index ccde086f6..fda9c3743 100644
> --- a/doc/guides/prog_guide/eventdev.rst
> +++ b/doc/guides/prog_guide/eventdev.rst
> @@ -63,13 +63,45 @@ the actual event being scheduled is. The payload is a union of the following:
>  * ``uint64_t u64``
>  * ``void *event_ptr``
>  * ``struct rte_mbuf *mbuf``
> +* ``struct rte_event_vector *vec``
> 
> -These three items in a union occupy the same 64 bits at the end of the rte_event
> +These four items in a union occupy the same 64 bits at the end of the rte_event
>  structure. The application can utilize the 64 bits directly by accessing the
> -u64 variable, while the event_ptr and mbuf are provided as convenience
> +u64 variable, while the event_ptr, mbuf, vec are provided as a convenience
>  variables.  For example the mbuf pointer in the union can used to schedule a
>  DPDK packet.
> 
> +Event Vector
> +~~~~~~~~~~~~
> +
> +The rte_event_vector struct contains a vector of elements defined by the event
> +type specified in the ``rte_event``. The event_vector structure contains the
> +following data:
> +
> +* ``nb_elem`` - The number of elements held within the vector.
> +
> +Similar to ``rte_event`` the payload of event vector is also a union, allowing
> +flexibility in what the actual vector is.
> +
> +* ``struct rte_mbuf *mbufs[0]`` - An array of mbufs.
> +* ``void *ptrs[0]`` - An array of pointers.
> +* ``uint64_t *u64s[0]`` - An array of uint64_t elements.
> +
> +The size of the event vector is related to the total number of elements it is
> +configured to hold, this is achieved by making `rte_event_vector` a variable
> +length structure.
> +A helper function is provided to create a mempool that holds event vector, which
> +takes name of the pool, total number of required ``rte_event_vector``,
> +cache size, number of elements in each ``rte_event_vector`` and socket id.
> +
> +.. code-block:: c
> +
> +        rte_event_vector_pool_create("vector_pool", nb_event_vectors, cache_sz,
> +                                     nb_elements_per_vector, socket_id);
> +
> +The function ``rte_event_vector_pool_create`` creates mempool with the best
> +platform mempool ops.
> +
>  Queues
>  ~~~~~~
> 
> diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
> index ce1fc2ce0..c0d01c873 100644
> --- a/lib/librte_eventdev/rte_eventdev.h
> +++ b/lib/librte_eventdev/rte_eventdev.h
> @@ -212,8 +212,10 @@ extern "C" {
> 
>  #include <rte_common.h>
>  #include <rte_config.h>
> -#include <rte_memory.h>
>  #include <rte_errno.h>
> +#include <rte_mbuf_pool_ops.h>
> +#include <rte_memory.h>
> +#include <rte_mempool.h>
> 
>  #include "rte_eventdev_trace_fp.h"
> 
> @@ -913,6 +915,25 @@ rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
>  int
>  rte_event_dev_close(uint8_t dev_id);
> 
> +/**
> + * Event vector structure.
> + */
> +struct rte_event_vector {
> +	uint64_t nb_elem : 16;
> +	/**< Number of elements in this event vector. */
> +	uint64_t rsvd : 48;
> +	uint64_t impl_opaque;
> +	union {
> +		struct rte_mbuf *mbufs[0];
> +		void *ptrs[0];
> +		uint64_t *u64s[0];
> +	} __rte_aligned(16);
> +	/**< Start of the vector array union. Depending upon the event type the
> +	 * vector array can be an array of mbufs or pointers or opaque u64
> +	 * values.
> +	 */
> +};
> +
>  /* Scheduler type definitions */
>  #define RTE_SCHED_TYPE_ORDERED          0
>  /**< Ordered scheduling
> @@ -986,6 +1007,21 @@ rte_event_dev_close(uint8_t dev_id);
>   */
>  #define RTE_EVENT_TYPE_ETH_RX_ADAPTER   0x4
>  /**< The event generated from event eth Rx adapter */
> +#define RTE_EVENT_TYPE_VECTOR           0x8
> +/**< Indicates that event is a vector.
> + * All vector event types should be an logical OR of EVENT_TYPE_VECTOR.
> + * This simplifies the pipeline design as we can split processing the events
> + * between vector events and normal event across event types.
> + * Example:
> + *	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
> + *		// Classify and handle vector event.
> + *	} else {
> + *		// Classify and handle event.
> + *	}
> + */
> +#define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
> +/**< The event vector generated from cpu for pipelining. */
> +
>  #define RTE_EVENT_TYPE_MAX              0x10
>  /**< Maximum number of event types */
> 
> @@ -2023,6 +2059,78 @@ rte_event_dev_xstats_reset(uint8_t dev_id,
>   */
>  int rte_event_dev_selftest(uint8_t dev_id);
> 
> +/**
> + * Get the memory required per event vector based on the number of elements per
> + * vector.
> + * This should be used to create the mempool that holds the event vectors.
> + *
> + * @param name
> + *   The name of the vector pool.
> + * @param n
> + *   The number of elements in the mbuf pool.
> + * @param cache_size
> + *   Size of the per-core object cache. See rte_mempool_create() for
> + *   details.
> + * @param nb_elem
> + *   The number of elements then a single event vector should be able to hold.
> + * @param socket_id
> + *   The socket identifier where the memory should be allocated. The
> + *   value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
> + *   reserved zone
> + *
> + * @return
> + *   The pointer to the newly allocated mempool, on success. NULL on error
> + *   with rte_errno set appropriately. Possible rte_errno values include:
> + *    - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
> + *    - E_RTE_SECONDARY - function was called from a secondary process instance
> + *    - EINVAL - cache size provided is too large, or priv_size is not aligned.
> + *    - ENOSPC - the maximum number of memzones has already been allocated
> + *    - EEXIST - a memzone with the same name already exists
> + *    - ENOMEM - no appropriate memory area found in which to create memzone
> + */
> +__rte_experimental
> +static inline struct rte_mempool *
> +rte_event_vector_pool_create(const char *name, unsigned int n,
> +			     unsigned int cache_size, uint16_t nb_elem,
> +			     int socket_id)
> +{
> +	const char *mp_ops_name;
> +	struct rte_mempool *mp;
> +	unsigned int elt_sz;
> +	int ret;
> +
> +	if (!nb_elem) {
> +		RTE_LOG(ERR, EVENTDEV,
> +			"Invalid number of elements=%d requested\n", nb_elem);
> +		rte_errno = -EINVAL;
> +		return NULL;
> +	}
> +
> +	elt_sz =
> +		sizeof(struct rte_event_vector) + (nb_elem * sizeof(uintptr_t));
> +	mp = rte_mempool_create_empty(name, n, elt_sz, cache_size, 0, socket_id,
> +				      0);
> +	if (mp == NULL)
> +		return NULL;
> +
> +	mp_ops_name = rte_mbuf_best_mempool_ops();
> +	ret = rte_mempool_set_ops_byname(mp, mp_ops_name, NULL);
> +	if (ret != 0) {
> +		RTE_LOG(ERR, EVENTDEV, "error setting mempool handler\n");
> +		goto err;
> +	}
> +
> +	ret = rte_mempool_populate_default(mp);
> +	if (ret < 0)
> +		goto err;
> +
> +	return mp;
> +err:
> +	rte_mempool_free(mp);
> +	rte_errno = -ret;
> +	return NULL;
> +}
> +
>  #ifdef __cplusplus
>  }
>  #endif
> diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
> index 3e5c09cfd..a070ef56e 100644
> --- a/lib/librte_eventdev/version.map
> +++ b/lib/librte_eventdev/version.map
> @@ -138,6 +138,9 @@ EXPERIMENTAL {
>  	__rte_eventdev_trace_port_setup;
>  	# added in 20.11
>  	rte_event_pmd_pci_probe_named;
> +
> +	#added in 21.05
> +	rte_event_vector_pool_create;
>  };
> 
>  INTERNAL {
> --
> 2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v3 1/8] eventdev: introduce event vector capability
  2021-03-18  6:19       ` Jayatheerthan, Jay
@ 2021-03-18  6:23         ` Pavan Nikhilesh Bhagavatula
  0 siblings, 0 replies; 153+ messages in thread
From: Pavan Nikhilesh Bhagavatula @ 2021-03-18  6:23 UTC (permalink / raw)
  To: Jayatheerthan, Jay, Jerin Jacob Kollanukkaran, Carrillo, Erik G,
	Gujjar, Abhinandan S, McDaniel, Timothy, hemant.agrawal,
	Van Haaren, Harry, mattias.ronnblom, Ma, Liang J, Ray Kinsella,
	Neil Horman
  Cc: dev

>Posting my comments in beginning. No further comments.
>
>@pbhagavatula@marvell.com, it seems like v3 patch1 drops some
>changes from v2 patch1 (for e.g. missing event vector pointer in
>rte_event) causing build failure. Could you have a look?

Ah, my bad, my auto-formatter really messed-up the series.
I will fix it in v4.

>
>Thanks!
>
>-Jay
>

Thanks,
Pavan.

>
>> -----Original Message-----
>> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
>> Sent: Wednesday, March 17, 2021 1:32 AM
>> To: jerinj@marvell.com; Jayatheerthan, Jay
><jay.jayatheerthan@intel.com>; Carrillo, Erik G
><erik.g.carrillo@intel.com>; Gujjar,
>> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy
><timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
>> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
><mattias.ronnblom@ericsson.com>; Ma, Liang J
>> <liang.j.ma@intel.com>; Ray Kinsella <mdr@ashroe.eu>; Neil Horman
><nhorman@tuxdriver.com>
>> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
>> Subject: [dpdk-dev] [PATCH v3 1/8] eventdev: introduce event vector
>capability
>>
>> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>>
>> Introduce rte_event_vector datastructure which is capable of holding
>> multiple uintptr_t of the same flow thereby allowing applications
>> to vectorize their pipeline and reducing the complexity of pipelining
>> the events across multiple stages.
>> This approach also reduces the scheduling overhead on a event
>device.
>>
>> Add a event vector mempool create handler to create mempools
>based on
>> the best mempool ops available on a given platform.
>>
>> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
>> ---
>>  doc/guides/prog_guide/eventdev.rst |  36 +++++++++-
>>  lib/librte_eventdev/rte_eventdev.h | 110
>++++++++++++++++++++++++++++-
>>  lib/librte_eventdev/version.map    |   3 +
>>  3 files changed, 146 insertions(+), 3 deletions(-)
>>
>> diff --git a/doc/guides/prog_guide/eventdev.rst
>b/doc/guides/prog_guide/eventdev.rst
>> index ccde086f6..fda9c3743 100644
>> --- a/doc/guides/prog_guide/eventdev.rst
>> +++ b/doc/guides/prog_guide/eventdev.rst
>> @@ -63,13 +63,45 @@ the actual event being scheduled is. The
>payload is a union of the following:
>>  * ``uint64_t u64``
>>  * ``void *event_ptr``
>>  * ``struct rte_mbuf *mbuf``
>> +* ``struct rte_event_vector *vec``
>>
>> -These three items in a union occupy the same 64 bits at the end of
>the rte_event
>> +These four items in a union occupy the same 64 bits at the end of the
>rte_event
>>  structure. The application can utilize the 64 bits directly by accessing
>the
>> -u64 variable, while the event_ptr and mbuf are provided as
>convenience
>> +u64 variable, while the event_ptr, mbuf, vec are provided as a
>convenience
>>  variables.  For example the mbuf pointer in the union can used to
>schedule a
>>  DPDK packet.
>>
>> +Event Vector
>> +~~~~~~~~~~~~
>> +
>> +The rte_event_vector struct contains a vector of elements defined
>by the event
>> +type specified in the ``rte_event``. The event_vector structure
>contains the
>> +following data:
>> +
>> +* ``nb_elem`` - The number of elements held within the vector.
>> +
>> +Similar to ``rte_event`` the payload of event vector is also a union,
>allowing
>> +flexibility in what the actual vector is.
>> +
>> +* ``struct rte_mbuf *mbufs[0]`` - An array of mbufs.
>> +* ``void *ptrs[0]`` - An array of pointers.
>> +* ``uint64_t *u64s[0]`` - An array of uint64_t elements.
>> +
>> +The size of the event vector is related to the total number of
>elements it is
>> +configured to hold, this is achieved by making `rte_event_vector` a
>variable
>> +length structure.
>> +A helper function is provided to create a mempool that holds event
>vector, which
>> +takes name of the pool, total number of required
>``rte_event_vector``,
>> +cache size, number of elements in each ``rte_event_vector`` and
>socket id.
>> +
>> +.. code-block:: c
>> +
>> +        rte_event_vector_pool_create("vector_pool",
>nb_event_vectors, cache_sz,
>> +                                     nb_elements_per_vector, socket_id);
>> +
>> +The function ``rte_event_vector_pool_create`` creates mempool
>with the best
>> +platform mempool ops.
>> +
>>  Queues
>>  ~~~~~~
>>
>> diff --git a/lib/librte_eventdev/rte_eventdev.h
>b/lib/librte_eventdev/rte_eventdev.h
>> index ce1fc2ce0..c0d01c873 100644
>> --- a/lib/librte_eventdev/rte_eventdev.h
>> +++ b/lib/librte_eventdev/rte_eventdev.h
>> @@ -212,8 +212,10 @@ extern "C" {
>>
>>  #include <rte_common.h>
>>  #include <rte_config.h>
>> -#include <rte_memory.h>
>>  #include <rte_errno.h>
>> +#include <rte_mbuf_pool_ops.h>
>> +#include <rte_memory.h>
>> +#include <rte_mempool.h>
>>
>>  #include "rte_eventdev_trace_fp.h"
>>
>> @@ -913,6 +915,25 @@
>rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
>>  int
>>  rte_event_dev_close(uint8_t dev_id);
>>
>> +/**
>> + * Event vector structure.
>> + */
>> +struct rte_event_vector {
>> +	uint64_t nb_elem : 16;
>> +	/**< Number of elements in this event vector. */
>> +	uint64_t rsvd : 48;
>> +	uint64_t impl_opaque;
>> +	union {
>> +		struct rte_mbuf *mbufs[0];
>> +		void *ptrs[0];
>> +		uint64_t *u64s[0];
>> +	} __rte_aligned(16);
>> +	/**< Start of the vector array union. Depending upon the event
>type the
>> +	 * vector array can be an array of mbufs or pointers or opaque
>u64
>> +	 * values.
>> +	 */
>> +};
>> +
>>  /* Scheduler type definitions */
>>  #define RTE_SCHED_TYPE_ORDERED          0
>>  /**< Ordered scheduling
>> @@ -986,6 +1007,21 @@ rte_event_dev_close(uint8_t dev_id);
>>   */
>>  #define RTE_EVENT_TYPE_ETH_RX_ADAPTER   0x4
>>  /**< The event generated from event eth Rx adapter */
>> +#define RTE_EVENT_TYPE_VECTOR           0x8
>> +/**< Indicates that event is a vector.
>> + * All vector event types should be an logical OR of
>EVENT_TYPE_VECTOR.
>> + * This simplifies the pipeline design as we can split processing the
>events
>> + * between vector events and normal event across event types.
>> + * Example:
>> + *	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
>> + *		// Classify and handle vector event.
>> + *	} else {
>> + *		// Classify and handle event.
>> + *	}
>> + */
>> +#define RTE_EVENT_TYPE_CPU_VECTOR
>(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
>> +/**< The event vector generated from cpu for pipelining. */
>> +
>>  #define RTE_EVENT_TYPE_MAX              0x10
>>  /**< Maximum number of event types */
>>
>> @@ -2023,6 +2059,78 @@ rte_event_dev_xstats_reset(uint8_t
>dev_id,
>>   */
>>  int rte_event_dev_selftest(uint8_t dev_id);
>>
>> +/**
>> + * Get the memory required per event vector based on the number of
>elements per
>> + * vector.
>> + * This should be used to create the mempool that holds the event
>vectors.
>> + *
>> + * @param name
>> + *   The name of the vector pool.
>> + * @param n
>> + *   The number of elements in the mbuf pool.
>> + * @param cache_size
>> + *   Size of the per-core object cache. See rte_mempool_create() for
>> + *   details.
>> + * @param nb_elem
>> + *   The number of elements then a single event vector should be
>able to hold.
>> + * @param socket_id
>> + *   The socket identifier where the memory should be allocated. The
>> + *   value can be *SOCKET_ID_ANY* if there is no NUMA constraint
>for the
>> + *   reserved zone
>> + *
>> + * @return
>> + *   The pointer to the newly allocated mempool, on success. NULL
>on error
>> + *   with rte_errno set appropriately. Possible rte_errno values
>include:
>> + *    - E_RTE_NO_CONFIG - function could not get pointer to
>rte_config structure
>> + *    - E_RTE_SECONDARY - function was called from a secondary
>process instance
>> + *    - EINVAL - cache size provided is too large, or priv_size is not
>aligned.
>> + *    - ENOSPC - the maximum number of memzones has already been
>allocated
>> + *    - EEXIST - a memzone with the same name already exists
>> + *    - ENOMEM - no appropriate memory area found in which to
>create memzone
>> + */
>> +__rte_experimental
>> +static inline struct rte_mempool *
>> +rte_event_vector_pool_create(const char *name, unsigned int n,
>> +			     unsigned int cache_size, uint16_t nb_elem,
>> +			     int socket_id)
>> +{
>> +	const char *mp_ops_name;
>> +	struct rte_mempool *mp;
>> +	unsigned int elt_sz;
>> +	int ret;
>> +
>> +	if (!nb_elem) {
>> +		RTE_LOG(ERR, EVENTDEV,
>> +			"Invalid number of elements=%d requested\n",
>nb_elem);
>> +		rte_errno = -EINVAL;
>> +		return NULL;
>> +	}
>> +
>> +	elt_sz =
>> +		sizeof(struct rte_event_vector) + (nb_elem *
>sizeof(uintptr_t));
>> +	mp = rte_mempool_create_empty(name, n, elt_sz, cache_size,
>0, socket_id,
>> +				      0);
>> +	if (mp == NULL)
>> +		return NULL;
>> +
>> +	mp_ops_name = rte_mbuf_best_mempool_ops();
>> +	ret = rte_mempool_set_ops_byname(mp, mp_ops_name,
>NULL);
>> +	if (ret != 0) {
>> +		RTE_LOG(ERR, EVENTDEV, "error setting mempool
>handler\n");
>> +		goto err;
>> +	}
>> +
>> +	ret = rte_mempool_populate_default(mp);
>> +	if (ret < 0)
>> +		goto err;
>> +
>> +	return mp;
>> +err:
>> +	rte_mempool_free(mp);
>> +	rte_errno = -ret;
>> +	return NULL;
>> +}
>> +
>>  #ifdef __cplusplus
>>  }
>>  #endif
>> diff --git a/lib/librte_eventdev/version.map
>b/lib/librte_eventdev/version.map
>> index 3e5c09cfd..a070ef56e 100644
>> --- a/lib/librte_eventdev/version.map
>> +++ b/lib/librte_eventdev/version.map
>> @@ -138,6 +138,9 @@ EXPERIMENTAL {
>>  	__rte_eventdev_trace_port_setup;
>>  	# added in 20.11
>>  	rte_event_pmd_pci_probe_named;
>> +
>> +	#added in 21.05
>> +	rte_event_vector_pool_create;
>>  };
>>
>>  INTERNAL {
>> --
>> 2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v4 0/8] Introduce event vectorization
  2021-03-16 20:01   ` [dpdk-dev] [PATCH v3 0/8] Introduce event vectorization pbhagavatula
                       ` (7 preceding siblings ...)
  2021-03-16 20:01     ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v3 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
@ 2021-03-19 20:57     ` pbhagavatula
  2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 1/8] eventdev: introduce event vector capability pbhagavatula
                         ` (9 more replies)
  8 siblings, 10 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-19 20:57 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

In traditional event programming model, events are identified by a
flow-id and a uintptr_t. The flow-id uniquely identifies a given event
and determines the order of scheduling based on schedule type, the
uintptr_t holds a single object.

Event devices also support burst mode with configurable dequeue depth,
i.e. each dequeue call would return multiple events and each event
might be at a different stage of the pipeline.
Having a burst of events belonging to different stages in a dequeue
burst is not only difficult to vectorize but also increases the scheduler
overhead and application overhead of pipelining events further.
Using event vectors we see a performance gain of ~628% as shown in [1].

By introducing event vectorization, each event will be capable of holding
multiple uintptr_t of the same flow thereby allowing applications
to vectorize their pipeline and reduce the complexity of pipelining
events across multiple stages. This also reduces the complexity of handling
enqueue and dequeue on an event device.

Since event devices are transparent to the events they are scheduling
so the event producers such as eth_rx_adapter, crypto_adapter , etc..
are responsible for vectorizing the buffers of the same flow into a single
event.

The series also breaks ABI in the patch [8/8] which is targetted to the
v21.11 release.

The dpdk-test-eventdev application has been updated with options to test
multiple vector sizes and timeouts.

[1]
As for performance improvement, with a ARM Cortex-A72 equivalent processer,
software event device (--vdev=event_sw0), single worker core, single stage
and using one service core for Rx adapter, Tx adapter, Scheduling.

Without event vectorization:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
         --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
         --stlist=a --wlcores=20
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    4.728 mpps avg 4.728 mpps

With event vectorization:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
        --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
        --stlist=a --wlcores=20 --enable_vector --nb_eth_queues 1
        --vector_size 256
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    34.383 mpps avg 34.383 mpps

Having dedicated service cores for each Rx queues and tweaking the vector,
dequeue burst size would further improve performance.

API usage is shown below:

Configuration:

	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

	vector_pool = rte_event_vector_pool_create("vector_pool",
			nb_elem, 0, vector_size, socket_id);

	rte_event_eth_rx_adapter_create(id, event_id, &adptr_conf);
	rte_event_eth_rx_adapter_queue_add(id, eth_id, -1, &queue_conf);
	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
		vec_conf.vector_sz = vector_size;
		vec_conf.vector_timeout_ns = vector_tmo_nsec;
		vec_conf.vector_mp = vector_pool;
		rte_event_eth_rx_adapter_queue_event_vector_config(id,
				eth_id, -1, &vec_conf);
	}

Fastpath:

	num = rte_event_dequeue_burst(event_id, port_id, &ev, 1, 0);
	if (!num)
		continue;

	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
		switch (ev.event_type) {
		case RTE_EVENT_TYPE_ETHDEV_VECTOR:
		case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
			struct rte_mbuf **mbufs;

			mbufs = ev.vector_ev->mbufs;
			for (i = 0; i < ev.vector_ev->nb_elem; i++)
				//Process mbufs.
			break;
		case ...
		}
	}
	...

v4 Changes:
- Fix missing event vector structure in event structure.(Jay)

v3 Changes:
- Fix unintended formatting changes.

v2 Changes:
- Multiple gramatical and style fixes.(Jerin)
- Add parameter to define vector size in power of 2. (Jerin)
- Redo patch series w/o breaking ABI till the last patch.(David)
- Add deprication notice to announce ABI break in 21.11.(David)
- Add vector limits validation to app/test-eventdev.

Pavan Nikhilesh (8):
  eventdev: introduce event vector capability
  eventdev: introduce event vector Rx capability
  eventdev: introduce event vector Tx capability
  eventdev: add Rx adapter event vector support
  eventdev: add Tx adapter event vector support
  app/eventdev: add event vector mode in pipeline test
  doc: announce event Rx adapter config changes
  eventdev: simplify Rx adapter event vector config

 app/test-eventdev/evt_common.h                |   4 +
 app/test-eventdev/evt_options.c               |  52 +++
 app/test-eventdev/evt_options.h               |   4 +
 app/test-eventdev/test_pipeline_atq.c         | 310 +++++++++++++++--
 app/test-eventdev/test_pipeline_common.c      | 105 +++++-
 app/test-eventdev/test_pipeline_common.h      |  18 +
 app/test-eventdev/test_pipeline_queue.c       | 320 ++++++++++++++++--
 .../prog_guide/event_ethernet_rx_adapter.rst  |  38 +++
 .../prog_guide/event_ethernet_tx_adapter.rst  |  12 +
 doc/guides/prog_guide/eventdev.rst            |  36 +-
 doc/guides/rel_notes/deprecation.rst          |   9 +
 doc/guides/tools/testeventdev.rst             |  28 ++
 lib/librte_eventdev/eventdev_pmd.h            |  31 +-
 .../rte_event_eth_rx_adapter.c                | 305 ++++++++++++++++-
 .../rte_event_eth_rx_adapter.h                |  68 ++++
 .../rte_event_eth_tx_adapter.c                |  66 +++-
 lib/librte_eventdev/rte_eventdev.c            |  11 +-
 lib/librte_eventdev/rte_eventdev.h            | 144 +++++++-
 lib/librte_eventdev/version.map               |   4 +
 19 files changed, 1479 insertions(+), 86 deletions(-)

--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v4 1/8] eventdev: introduce event vector capability
  2021-03-19 20:57     ` [dpdk-dev] [PATCH v4 0/8] Introduce event vectorization pbhagavatula
@ 2021-03-19 20:57       ` pbhagavatula
  2021-03-22  9:06         ` Kinsella, Ray
  2021-03-23 11:12         ` [dpdk-dev] " Jerin Jacob
  2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 2/8] eventdev: introduce event vector Rx capability pbhagavatula
                         ` (8 subsequent siblings)
  9 siblings, 2 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-19 20:57 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce rte_event_vector datastructure which is capable of holding
multiple uintptr_t of the same flow thereby allowing applications
to vectorize their pipeline and reducing the complexity of pipelining
the events across multiple stages.
This approach also reduces the scheduling overhead on a event device.

Add a event vector mempool create handler to create mempools based on
the best mempool ops available on a given platform.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 doc/guides/prog_guide/eventdev.rst |  36 +++++++++-
 lib/librte_eventdev/rte_eventdev.h | 112 ++++++++++++++++++++++++++++-
 lib/librte_eventdev/version.map    |   3 +
 3 files changed, 148 insertions(+), 3 deletions(-)

diff --git a/doc/guides/prog_guide/eventdev.rst b/doc/guides/prog_guide/eventdev.rst
index ccde086f6..fda9c3743 100644
--- a/doc/guides/prog_guide/eventdev.rst
+++ b/doc/guides/prog_guide/eventdev.rst
@@ -63,13 +63,45 @@ the actual event being scheduled is. The payload is a union of the following:
 * ``uint64_t u64``
 * ``void *event_ptr``
 * ``struct rte_mbuf *mbuf``
+* ``struct rte_event_vector *vec``
 
-These three items in a union occupy the same 64 bits at the end of the rte_event
+These four items in a union occupy the same 64 bits at the end of the rte_event
 structure. The application can utilize the 64 bits directly by accessing the
-u64 variable, while the event_ptr and mbuf are provided as convenience
+u64 variable, while the event_ptr, mbuf, vec are provided as a convenience
 variables.  For example the mbuf pointer in the union can used to schedule a
 DPDK packet.
 
+Event Vector
+~~~~~~~~~~~~
+
+The rte_event_vector struct contains a vector of elements defined by the event
+type specified in the ``rte_event``. The event_vector structure contains the
+following data:
+
+* ``nb_elem`` - The number of elements held within the vector.
+
+Similar to ``rte_event`` the payload of event vector is also a union, allowing
+flexibility in what the actual vector is.
+
+* ``struct rte_mbuf *mbufs[0]`` - An array of mbufs.
+* ``void *ptrs[0]`` - An array of pointers.
+* ``uint64_t *u64s[0]`` - An array of uint64_t elements.
+
+The size of the event vector is related to the total number of elements it is
+configured to hold, this is achieved by making `rte_event_vector` a variable
+length structure.
+A helper function is provided to create a mempool that holds event vector, which
+takes name of the pool, total number of required ``rte_event_vector``,
+cache size, number of elements in each ``rte_event_vector`` and socket id.
+
+.. code-block:: c
+
+        rte_event_vector_pool_create("vector_pool", nb_event_vectors, cache_sz,
+                                     nb_elements_per_vector, socket_id);
+
+The function ``rte_event_vector_pool_create`` creates mempool with the best
+platform mempool ops.
+
 Queues
 ~~~~~~
 
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index ce1fc2ce0..5586a3f15 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -212,8 +212,10 @@ extern "C" {
 
 #include <rte_common.h>
 #include <rte_config.h>
-#include <rte_memory.h>
 #include <rte_errno.h>
+#include <rte_mbuf_pool_ops.h>
+#include <rte_memory.h>
+#include <rte_mempool.h>
 
 #include "rte_eventdev_trace_fp.h"
 
@@ -913,6 +915,25 @@ rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
 int
 rte_event_dev_close(uint8_t dev_id);
 
+/**
+ * Event vector structure.
+ */
+struct rte_event_vector {
+	uint64_t nb_elem : 16;
+	/**< Number of elements in this event vector. */
+	uint64_t rsvd : 48;
+	uint64_t impl_opaque;
+	union {
+		struct rte_mbuf *mbufs[0];
+		void *ptrs[0];
+		uint64_t *u64s[0];
+	} __rte_aligned(16);
+	/**< Start of the vector array union. Depending upon the event type the
+	 * vector array can be an array of mbufs or pointers or opaque u64
+	 * values.
+	 */
+};
+
 /* Scheduler type definitions */
 #define RTE_SCHED_TYPE_ORDERED          0
 /**< Ordered scheduling
@@ -986,6 +1007,21 @@ rte_event_dev_close(uint8_t dev_id);
  */
 #define RTE_EVENT_TYPE_ETH_RX_ADAPTER   0x4
 /**< The event generated from event eth Rx adapter */
+#define RTE_EVENT_TYPE_VECTOR           0x8
+/**< Indicates that event is a vector.
+ * All vector event types should be an logical OR of EVENT_TYPE_VECTOR.
+ * This simplifies the pipeline design as we can split processing the events
+ * between vector events and normal event across event types.
+ * Example:
+ *	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
+ *		// Classify and handle vector event.
+ *	} else {
+ *		// Classify and handle event.
+ *	}
+ */
+#define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
+/**< The event vector generated from cpu for pipelining. */
+
 #define RTE_EVENT_TYPE_MAX              0x10
 /**< Maximum number of event types */
 
@@ -1108,6 +1144,8 @@ struct rte_event {
 		/**< Opaque event pointer */
 		struct rte_mbuf *mbuf;
 		/**< mbuf pointer if dequeued event is associated with mbuf */
+		struct rte_event_vector *vec;
+		/**< Event vector pointer. */
 	};
 };
 
@@ -2023,6 +2061,78 @@ rte_event_dev_xstats_reset(uint8_t dev_id,
  */
 int rte_event_dev_selftest(uint8_t dev_id);
 
+/**
+ * Get the memory required per event vector based on the number of elements per
+ * vector.
+ * This should be used to create the mempool that holds the event vectors.
+ *
+ * @param name
+ *   The name of the vector pool.
+ * @param n
+ *   The number of elements in the mbuf pool.
+ * @param cache_size
+ *   Size of the per-core object cache. See rte_mempool_create() for
+ *   details.
+ * @param nb_elem
+ *   The number of elements then a single event vector should be able to hold.
+ * @param socket_id
+ *   The socket identifier where the memory should be allocated. The
+ *   value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
+ *   reserved zone
+ *
+ * @return
+ *   The pointer to the newly allocated mempool, on success. NULL on error
+ *   with rte_errno set appropriately. Possible rte_errno values include:
+ *    - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
+ *    - E_RTE_SECONDARY - function was called from a secondary process instance
+ *    - EINVAL - cache size provided is too large, or priv_size is not aligned.
+ *    - ENOSPC - the maximum number of memzones has already been allocated
+ *    - EEXIST - a memzone with the same name already exists
+ *    - ENOMEM - no appropriate memory area found in which to create memzone
+ */
+__rte_experimental
+static inline struct rte_mempool *
+rte_event_vector_pool_create(const char *name, unsigned int n,
+			     unsigned int cache_size, uint16_t nb_elem,
+			     int socket_id)
+{
+	const char *mp_ops_name;
+	struct rte_mempool *mp;
+	unsigned int elt_sz;
+	int ret;
+
+	if (!nb_elem) {
+		RTE_LOG(ERR, EVENTDEV,
+			"Invalid number of elements=%d requested\n", nb_elem);
+		rte_errno = -EINVAL;
+		return NULL;
+	}
+
+	elt_sz =
+		sizeof(struct rte_event_vector) + (nb_elem * sizeof(uintptr_t));
+	mp = rte_mempool_create_empty(name, n, elt_sz, cache_size, 0, socket_id,
+				      0);
+	if (mp == NULL)
+		return NULL;
+
+	mp_ops_name = rte_mbuf_best_mempool_ops();
+	ret = rte_mempool_set_ops_byname(mp, mp_ops_name, NULL);
+	if (ret != 0) {
+		RTE_LOG(ERR, EVENTDEV, "error setting mempool handler\n");
+		goto err;
+	}
+
+	ret = rte_mempool_populate_default(mp);
+	if (ret < 0)
+		goto err;
+
+	return mp;
+err:
+	rte_mempool_free(mp);
+	rte_errno = -ret;
+	return NULL;
+}
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index 3e5c09cfd..a070ef56e 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -138,6 +138,9 @@ EXPERIMENTAL {
 	__rte_eventdev_trace_port_setup;
 	# added in 20.11
 	rte_event_pmd_pci_probe_named;
+
+	#added in 21.05
+	rte_event_vector_pool_create;
 };
 
 INTERNAL {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v4 2/8] eventdev: introduce event vector Rx capability
  2021-03-19 20:57     ` [dpdk-dev] [PATCH v4 0/8] Introduce event vectorization pbhagavatula
  2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 1/8] eventdev: introduce event vector capability pbhagavatula
@ 2021-03-19 20:57       ` pbhagavatula
  2021-03-22  9:12         ` Kinsella, Ray
  2021-03-23 16:56         ` [dpdk-dev] " Jerin Jacob
  2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 3/8] eventdev: introduce event vector Tx capability pbhagavatula
                         ` (7 subsequent siblings)
  9 siblings, 2 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-19 20:57 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce event ethernet Rx adapter event vector capability.

If an event eth Rx adapter has the capability of
RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR then a given Rx queue
can be configured to enable event vectorization by passing the
flag RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR to
rte_event_eth_rx_adapter_queue_conf::rx_queue_flags while configuring
Rx adapter through rte_event_eth_rx_adapter_queue_add.

The max vector size, vector timeout define the vector size and
mempool used for allocating vector event are configured through
rte_event_eth_rx_adapter_queue_add. The element size of the element
in the vector pool should be equal to
    sizeof(struct rte_event_vector) + (vector_sz * sizeof(uintptr_t))

Application can use `rte_event_vector_pool_create` to create the
vector mempool used for
rte_event_eth_rx_adapter_queue_conf::vector_mp.

The Rx adapter would be responsible for vectorizing the mbufs
based on the flow, the vector limits configured by the application
and add the vector event of mbufs to the event queue set via
rte_event_eth_rx_adapter_queue_conf::ev::queue_id.
It should also mark rte_event_vector::union_valid and fill
rte_event_vector::port, rte_event_vector::queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 .../prog_guide/event_ethernet_rx_adapter.rst  | 38 ++++++++
 .../rte_event_eth_rx_adapter.h                | 95 +++++++++++++++++++
 lib/librte_eventdev/rte_eventdev.h            | 30 +++++-
 lib/librte_eventdev/version.map               |  2 +
 4 files changed, 163 insertions(+), 2 deletions(-)

diff --git a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
index cb44ce0e4..5eefef355 100644
--- a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
@@ -186,3 +186,41 @@ the event buffer fill level is low. The
 ``rte_event_eth_rx_adapter_cb_register()`` function allow the application
 to register a callback that selects which packets to enqueue to the event
 device.
+
+Rx event vectorization
+~~~~~~~~~~~~~~~~~~~~~~
+
+The event devices, ethernet device pairs which support the capability
+``RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR`` can aggregate packets based on
+flow characteristics and generate a ``rte_event`` containing ``rte_event_vector``
+whose event type is either ``RTE_EVENT_TYPE_ETHDEV_VECTOR`` or
+``RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR``.
+The aggregation size and timeout are configurable at a queue level and the
+maximum, minimum vector sizes and timeouts vary based on the device capability
+and can be queried using ``rte_event_eth_rx_adapter_vector_limits_get``.
+The Rx adapter additionally might include useful data such as ethernet device
+port and queue identifier in the ``rte_event_vector::port`` and
+``rte_event_vector::queue`` and mark ``rte_event_vector::attr_valid`` as true.
+
+A loop processing ``rte_event_vector`` containing mbufs is shown below.
+
+.. code-block:: c
+
+        event = rte_event_dequeue_burst(event_dev, event_port, &event,
+                                        1, 0);
+        if (!event)
+                continue;
+
+        switch (ev.event_type) {
+        case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
+        case RTE_EVENT_TYPE_ETHDEV_VECTOR:
+                struct rte_mbufs **mbufs;
+
+                mbufs = (struct rte_mbufs **)ev[i].vec->mbufs;
+                for (i = 0; i < ev.vec->nb_elem; i++) {
+                        /* Process each mbuf. */
+                }
+        break;
+        case ...
+        ...
+        }
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
index 21bb1e54c..dda846f57 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
@@ -92,6 +92,10 @@ extern "C" {
 /**< This flag indicates the flow identifier is valid
  * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
  */
+#define RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR	0x2
+/**< This flag indicates that mbufs arriving on the queue need to be vectorized
+ * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+ */
 
 /**
  * Adapter configuration structure that the adapter configuration callback
@@ -169,6 +173,36 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 */
 };
 
+struct rte_event_eth_rx_adapter_event_vector_config {
+	uint16_t vector_sz;
+	/**<
+	 * Indicates the maximum number for mbufs to combine and form a vector.
+	 * Should be within
+	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_sz
+	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_sz
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+	 */
+	uint64_t vector_timeout_ns;
+	/**<
+	 * Indicates the maximum number of nanoseconds to wait for receiving
+	 * mbufs. Should be within vectorization limits of the
+	 * adapter
+	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_ns
+	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_ns
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+	 */
+	struct rte_mempool *vector_mp;
+	/**<
+	 * Indicates the mempool that should be used for allocating
+	 * rte_event_vector container.
+	 * Should be created by using `rte_event_vector_pool_create`.
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags.
+	 */
+};
+
 /**
  * A structure used to retrieve statistics for an eth rx adapter instance.
  */
@@ -199,6 +233,22 @@ struct rte_event_eth_rx_adapter_stats {
 	/**< Received packet count for interrupt mode Rx queues */
 };
 
+/**
+ * A structure used to retrieve eth rx adapter vector limits.
+ */
+struct rte_event_eth_rx_adapter_vector_limits {
+	uint16_t min_sz;
+	/**< Minimum vector limit configurable. */
+	uint16_t max_sz;
+	/**< Maximum vector limit configurable. */
+	uint8_t log2_sz;
+	/**< The size configured should be in log2. */
+	uint64_t min_timeout_ns;
+	/**< Minimum vector timeout configurable. */
+	uint64_t max_timeout_ns;
+	/**< Maximum vector timeout configurable. */
+};
+
 /**
  *
  * Callback function invoked by the SW adapter before it continues
@@ -467,6 +517,51 @@ int rte_event_eth_rx_adapter_cb_register(uint8_t id, uint16_t eth_dev_id,
 					 rte_event_eth_rx_adapter_cb_fn cb_fn,
 					 void *cb_arg);
 
+/**
+ * Retrieve vector limits for a given event dev and eth dev pair.
+ * @see rte_event_eth_rx_adapter_vector_limits
+ *
+ * @param dev_id
+ *  Event device identifier.
+ * @param eth_port_id
+ *  Port identifier of the ethernet device.
+ * @param [out] limits
+ *  A pointer to rte_event_eth_rx_adapter_vector_limits structure that has to
+ * be filled.
+ *
+ * @return
+ *  - 0: Success.
+ *  - <0: Error code on failure.
+ */
+__rte_experimental
+int rte_event_eth_rx_adapter_vector_limits_get(
+	uint8_t dev_id, uint16_t eth_port_id,
+	struct rte_event_eth_rx_adapter_vector_limits *limits);
+
+/**
+ * Configure event vectorization for a given ethernet device queue, that has
+ * been added to a event eth Rx adapter.
+ *
+ * @param id
+ *  The identifier of the ethernet Rx event adapter.
+ *
+ * @param eth_dev_id
+ *  The identifier of the ethernet device.
+ *
+ * @param rx_queue_id
+ *  Ethernet device receive queue index.
+ *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
+ *  are configured with event vectorization.
+ *
+ * @return
+ *  - 0: Success, Receive queue configured correctly.
+ *  - <0: Error code on failure.
+ */
+__rte_experimental
+int rte_event_eth_rx_adapter_queue_event_vector_config(
+	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
+	struct rte_event_eth_rx_adapter_event_vector_config *config);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index 5586a3f15..1be79b856 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -919,9 +919,27 @@ rte_event_dev_close(uint8_t dev_id);
  * Event vector structure.
  */
 struct rte_event_vector {
-	uint64_t nb_elem : 16;
+	uint16_t nb_elem;
 	/**< Number of elements in this event vector. */
-	uint64_t rsvd : 48;
+	uint16_t rsvd : 15;
+	uint16_t attr_valid : 1;
+	/**< Indicates that the below union attributes have valid information.
+	 */
+	union {
+		/* Used by Rx adapter.
+		 * Indicats that all the elements in this vector belong to same
+		 * port and queue pair when originating from Rx adapter, valid
+		 * only when event type is ETHDEV_VECTOR or
+		 * ETH_RX_ADAPTER_VECTOR.
+		 */
+		struct {
+			uint16_t port;
+			/* Ethernet device port id. */
+			uint16_t queue;
+			/* Ethernet device queue id. */
+		};
+	};
+	/**< Union to hold common attributes of the vector array. */
 	uint64_t impl_opaque;
 	union {
 		struct rte_mbuf *mbufs[0];
@@ -1019,8 +1037,14 @@ struct rte_event_vector {
  *		// Classify and handle event.
  *	}
  */
+#define RTE_EVENT_TYPE_ETHDEV_VECTOR                                           \
+	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETHDEV)
+/**< The event vector generated from ethdev subsystem */
 #define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
 /**< The event vector generated from cpu for pipelining. */
+#define RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR                                   \
+	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETH_RX_ADAPTER)
+/**< The event vector generated from eth Rx adapter. */
 
 #define RTE_EVENT_TYPE_MAX              0x10
 /**< Maximum number of event types */
@@ -1165,6 +1189,8 @@ struct rte_event {
  * @see struct rte_event_eth_rx_adapter_queue_conf::ev
  * @see struct rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
  */
+#define RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR	0x8
+/**< Adapter supports event vectorization per ethdev. */
 
 /**
  * Retrieve the event device's ethdev Rx adapter capabilities for the
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index a070ef56e..902df0ae3 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -141,6 +141,8 @@ EXPERIMENTAL {
 
 	#added in 21.05
 	rte_event_vector_pool_create;
+	rte_event_eth_rx_adapter_vector_limits_get;
+	rte_event_eth_rx_adapter_queue_event_vector_config;
 };
 
 INTERNAL {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v4 3/8] eventdev: introduce event vector Tx capability
  2021-03-19 20:57     ` [dpdk-dev] [PATCH v4 0/8] Introduce event vectorization pbhagavatula
  2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 1/8] eventdev: introduce event vector capability pbhagavatula
  2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 2/8] eventdev: introduce event vector Rx capability pbhagavatula
@ 2021-03-19 20:57       ` pbhagavatula
  2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 4/8] eventdev: add Rx adapter event vector support pbhagavatula
                         ` (6 subsequent siblings)
  9 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-19 20:57 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce event vector transmit capability for event eth
tx adapter.

The capability indicates that the Tx adapter is capable of
transmitting event vectors.
When rte_event_vector::union_valid is set, the Tx adapter should
transmit all the packets to the rte_event_vector::port using the
rte_event_vector::queue.
If rte_event_vector::union_valid is not set then the Tx adapter
should peek into each mbuf to get the destination port and queue
pair.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
---
 doc/guides/prog_guide/event_ethernet_tx_adapter.rst | 12 ++++++++++++
 lib/librte_eventdev/rte_eventdev.h                  |  8 +++++++-
 2 files changed, 19 insertions(+), 1 deletion(-)

diff --git a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
index a8c13e136..87277dcaf 100644
--- a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
@@ -164,3 +164,15 @@ The  ``rte_event_eth_tx_adapter_stats_get()`` function reports counters defined
 in struct ``rte_event_eth_tx_adapter_stats``. The counter values are the sum of
 the counts from the eventdev PMD callback if the callback is supported, and
 the counts maintained by the service function, if one exists.
+
+Tx event vectorization
+~~~~~~~~~~~~~~~~~~~~~~
+
+The event device, ethernet device paris which support the capability
+``RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR`` can process event vector of mbufs.
+Additionally, application can provide a hint to that Tx adapter that all the
+mbufs are destined to the same ethernet port and queue by setting the bit
+``rte_event_vector::attr_valid`` and filling `rte_event_vector::port`` and
+``rte_event_vector::queue``.
+If ``rte_event_vector::attr_valid`` is not set then the Tx adapter should peek
+into each mbuf and transmit them to the requested ethernet port and queue pair.
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index 1be79b856..6ed0fbc52 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -926,11 +926,13 @@ struct rte_event_vector {
 	/**< Indicates that the below union attributes have valid information.
 	 */
 	union {
-		/* Used by Rx adapter.
+		/* Used by Rx/Tx adapter.
 		 * Indicats that all the elements in this vector belong to same
 		 * port and queue pair when originating from Rx adapter, valid
 		 * only when event type is ETHDEV_VECTOR or
 		 * ETH_RX_ADAPTER_VECTOR.
+		 * Can also be used to indicate the Tx adapter the destination
+		 * port and queue of the mbufs in the vector
 		 */
 		struct {
 			uint16_t port;
@@ -1287,6 +1289,10 @@ rte_event_crypto_adapter_caps_get(uint8_t dev_id, uint8_t cdev_id,
 #define RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT	0x1
 /**< This flag is sent when the PMD supports a packet transmit callback
  */
+#define RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR	0x2
+/**< Indicates that the Tx adapter is capable of handling event vector of
+ * mbufs.
+ */
 
 /**
  * Retrieve the event device's eth Tx adapter capabilities
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v4 4/8] eventdev: add Rx adapter event vector support
  2021-03-19 20:57     ` [dpdk-dev] [PATCH v4 0/8] Introduce event vectorization pbhagavatula
                         ` (2 preceding siblings ...)
  2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 3/8] eventdev: introduce event vector Tx capability pbhagavatula
@ 2021-03-19 20:57       ` pbhagavatula
  2021-03-23 18:30         ` Jerin Jacob
  2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 5/8] eventdev: add Tx " pbhagavatula
                         ` (5 subsequent siblings)
  9 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-03-19 20:57 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support for event eth Rx adapter, the implementation
creates vector flows based on port and queue identifier of the received
mbufs.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 lib/librte_eventdev/eventdev_pmd.h            |  60 ++-
 .../rte_event_eth_rx_adapter.c                | 367 +++++++++++++++++-
 lib/librte_eventdev/rte_eventdev.c            |   6 +-
 3 files changed, 415 insertions(+), 18 deletions(-)

diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 7eb9a7739..d79dfd612 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -69,9 +69,10 @@ extern "C" {
 	} \
 } while (0)
 
-#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP \
-		((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) | \
-			(RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ))
+#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP                                        \
+	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |                     \
+	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) |                         \
+	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR))
 
 #define RTE_EVENT_CRYPTO_ADAPTER_SW_CAP \
 		RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA
@@ -645,6 +646,53 @@ typedef int (*eventdev_eth_rx_adapter_stats_reset)
  */
 typedef int (*eventdev_selftest)(void);
 
+struct rte_event_eth_rx_adapter_vector_limits;
+/**
+ * Get event vector limits for a given event, etherner device pair.
+ *
+ * @param dev
+ *   Event device pointer
+ *
+ * @param eth_dev
+ *   Ethernet device pointer
+ *
+ * @param[out] limits
+ *   Pointer to the limits structure to be filled.
+ *
+ * @return
+ *   - 0: Success.
+ *   - <0: Error code returned by the driver function.
+ */
+typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
+	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
+	struct rte_event_eth_rx_adapter_vector_limits *limits);
+
+struct rte_event_eth_rx_adapter_event_vector_config;
+/**
+ * Enable event vector on an given Rx queue of a ethernet devices belonging to
+ * the Rx adapter.
+ *
+ * @param dev
+ *   Event device pointer
+ *
+ * @param eth_dev
+ *   Ethernet device pointer
+ *
+ * @param rx_queue_id
+ *   The Rx queue identifier
+ *
+ * @param config
+ *   Pointer to the event vector configuration structure.
+ *
+ * @return
+ *   - 0: Success.
+ *   - <0: Error code returned by the driver function.
+ */
+typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
+	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
+	int32_t rx_queue_id,
+	const struct rte_event_eth_rx_adapter_event_vector_config *config);
+
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;
 
@@ -1067,6 +1115,12 @@ struct rte_eventdev_ops {
 	/**< Get ethernet Rx stats */
 	eventdev_eth_rx_adapter_stats_reset eth_rx_adapter_stats_reset;
 	/**< Reset ethernet Rx stats */
+	eventdev_eth_rx_adapter_vector_limits_get_t
+		eth_rx_adapter_vector_limits_get;
+	/**< Get event vector limits for the Rx adapter */
+	eventdev_eth_rx_adapter_event_vector_config_t
+		eth_rx_adapter_event_vector_config;
+	/**< Configure Rx adapter with event vector */
 
 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index d8c635e99..c71990078 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -26,6 +26,10 @@
 #define BATCH_SIZE		32
 #define BLOCK_CNT_THRESHOLD	10
 #define ETH_EVENT_BUFFER_SIZE	(4*BATCH_SIZE)
+#define MAX_VECTOR_SIZE		1024
+#define MIN_VECTOR_SIZE		4
+#define MAX_VECTOR_NS		1E9
+#define MIN_VECTOR_NS		1E5
 
 #define ETH_RX_ADAPTER_SERVICE_NAME_LEN	32
 #define ETH_RX_ADAPTER_MEM_NAME_LEN	32
@@ -59,6 +63,20 @@ struct eth_rx_poll_entry {
 	uint16_t eth_rx_qid;
 };
 
+struct eth_rx_vector_data {
+	TAILQ_ENTRY(eth_rx_vector_data) next;
+	uint16_t port;
+	uint16_t queue;
+	uint16_t max_vector_count;
+	uint64_t event;
+	uint64_t ts;
+	uint64_t vector_timeout_ticks;
+	struct rte_mempool *vector_pool;
+	struct rte_event_vector *vector_ev;
+} __rte_cache_aligned;
+
+TAILQ_HEAD(eth_rx_vector_data_list, eth_rx_vector_data);
+
 /* Instance per adapter */
 struct rte_eth_event_enqueue_buffer {
 	/* Count of events in this buffer */
@@ -92,6 +110,14 @@ struct rte_event_eth_rx_adapter {
 	uint32_t wrr_pos;
 	/* Event burst buffer */
 	struct rte_eth_event_enqueue_buffer event_enqueue_buffer;
+	/* Vector enable flag */
+	uint8_t ena_vector;
+	/* Timestamp of previous vector expiry list traversal */
+	uint64_t prev_expiry_ts;
+	/* Minimum ticks to wait before traversing expiry list */
+	uint64_t vector_tmo_ticks;
+	/* vector list */
+	struct eth_rx_vector_data_list vector_list;
 	/* Per adapter stats */
 	struct rte_event_eth_rx_adapter_stats stats;
 	/* Block count, counts up to BLOCK_CNT_THRESHOLD */
@@ -198,9 +224,11 @@ struct eth_device_info {
 struct eth_rx_queue_info {
 	int queue_enabled;	/* True if added */
 	int intr_enabled;
+	uint8_t ena_vector;
 	uint16_t wt;		/* Polling weight */
 	uint32_t flow_id_mask;	/* Set to ~0 if app provides flow id else 0 */
 	uint64_t event;
+	struct eth_rx_vector_data vector_data;
 };
 
 static struct rte_event_eth_rx_adapter **event_eth_rx_adapter;
@@ -722,6 +750,9 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
 	    &rx_adapter->event_enqueue_buffer;
 	struct rte_event_eth_rx_adapter_stats *stats = &rx_adapter->stats;
 
+	if (!buf->count)
+		return 0;
+
 	uint16_t n = rte_event_enqueue_new_burst(rx_adapter->eventdev_id,
 					rx_adapter->event_port_id,
 					buf->events,
@@ -742,6 +773,72 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
 	return n;
 }
 
+static inline uint16_t
+rxa_create_event_vector(struct rte_event_eth_rx_adapter *rx_adapter,
+			struct eth_rx_queue_info *queue_info,
+			struct rte_eth_event_enqueue_buffer *buf,
+			struct rte_mbuf **mbufs, uint16_t num)
+{
+	struct rte_event *ev = &buf->events[buf->count];
+	struct eth_rx_vector_data *vec;
+	uint16_t filled, space, sz;
+
+	filled = 0;
+	vec = &queue_info->vector_data;
+	while (num) {
+		if (vec->vector_ev == NULL) {
+			if (rte_mempool_get(vec->vector_pool,
+					    (void **)&vec->vector_ev) < 0) {
+				rte_pktmbuf_free_bulk(mbufs, num);
+				return 0;
+			}
+			vec->vector_ev->nb_elem = 0;
+			vec->vector_ev->port = vec->port;
+			vec->vector_ev->queue = vec->queue;
+			vec->vector_ev->attr_valid = true;
+			TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
+		} else if (vec->vector_ev->nb_elem == vec->max_vector_count) {
+			/* Event ready. */
+			ev->event = vec->event;
+			ev->vec = vec->vector_ev;
+			ev++;
+			filled++;
+			vec->vector_ev = NULL;
+			TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+			if (rte_mempool_get(vec->vector_pool,
+					    (void **)&vec->vector_ev) < 0) {
+				rte_pktmbuf_free_bulk(mbufs, num);
+				return 0;
+			}
+			vec->vector_ev->nb_elem = 0;
+			vec->vector_ev->port = vec->port;
+			vec->vector_ev->queue = vec->queue;
+			vec->vector_ev->attr_valid = true;
+			TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
+		}
+
+		space = vec->max_vector_count - vec->vector_ev->nb_elem;
+		sz = num > space ? space : num;
+		memcpy(vec->vector_ev->mbufs + vec->vector_ev->nb_elem, mbufs,
+		       sizeof(void *) * sz);
+		vec->vector_ev->nb_elem += sz;
+		num -= sz;
+		mbufs += sz;
+		vec->ts = rte_rdtsc();
+	}
+
+	if (vec->vector_ev->nb_elem == vec->max_vector_count) {
+		ev->event = vec->event;
+		ev->vec = vec->vector_ev;
+		ev++;
+		filled++;
+		vec->vector_ev = NULL;
+		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+	}
+
+	return filled;
+}
+
 static inline void
 rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
@@ -770,25 +867,30 @@ rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
 	rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
 	do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
 
-	for (i = 0; i < num; i++) {
-		m = mbufs[i];
-
-		rss = do_rss ?
-			rxa_do_softrss(m, rx_adapter->rss_key_be) :
-			m->hash.rss;
-		ev->event = event;
-		ev->flow_id = (rss & ~flow_id_mask) |
-				(ev->flow_id & flow_id_mask);
-		ev->mbuf = m;
-		ev++;
+	if (!eth_rx_queue_info->ena_vector) {
+		for (i = 0; i < num; i++) {
+			m = mbufs[i];
+
+			rss = do_rss ? rxa_do_softrss(m, rx_adapter->rss_key_be)
+				     : m->hash.rss;
+			ev->event = event;
+			ev->flow_id = (rss & ~flow_id_mask) |
+				      (ev->flow_id & flow_id_mask);
+			ev->mbuf = m;
+			ev++;
+		}
+	} else {
+		num = rxa_create_event_vector(rx_adapter, eth_rx_queue_info,
+					      buf, mbufs, num);
 	}
 
-	if (dev_info->cb_fn) {
+	if (num && dev_info->cb_fn) {
 
 		dropped = 0;
 		nb_cb = dev_info->cb_fn(eth_dev_id, rx_queue_id,
-					ETH_EVENT_BUFFER_SIZE, buf->count, ev,
-					num, dev_info->cb_arg, &dropped);
+					ETH_EVENT_BUFFER_SIZE, buf->count,
+					&buf->events[buf->count], num,
+					dev_info->cb_arg, &dropped);
 		if (unlikely(nb_cb > num))
 			RTE_EDEV_LOG_ERR("Rx CB returned %d (> %d) events",
 				nb_cb, num);
@@ -1124,6 +1226,30 @@ rxa_poll(struct rte_event_eth_rx_adapter *rx_adapter)
 	return nb_rx;
 }
 
+static void
+rxa_vector_expire(struct eth_rx_vector_data *vec, void *arg)
+{
+	struct rte_event_eth_rx_adapter *rx_adapter = arg;
+	struct rte_eth_event_enqueue_buffer *buf =
+		&rx_adapter->event_enqueue_buffer;
+	struct rte_event *ev;
+
+	if (buf->count)
+		rxa_flush_event_buffer(rx_adapter);
+
+	if (vec->vector_ev->nb_elem == 0)
+		return;
+	ev = &buf->events[buf->count];
+
+	/* Event ready. */
+	ev->event = vec->event;
+	ev->vec = vec->vector_ev;
+	buf->count++;
+
+	vec->vector_ev = NULL;
+	vec->ts = 0;
+}
+
 static int
 rxa_service_func(void *args)
 {
@@ -1137,6 +1263,24 @@ rxa_service_func(void *args)
 		return 0;
 	}
 
+	if (rx_adapter->ena_vector) {
+		if ((rte_rdtsc() - rx_adapter->prev_expiry_ts) >=
+		    rx_adapter->vector_tmo_ticks) {
+			struct eth_rx_vector_data *vec;
+
+			TAILQ_FOREACH(vec, &rx_adapter->vector_list, next) {
+				uint64_t elapsed_time = rte_rdtsc() - vec->ts;
+
+				if (elapsed_time >= vec->vector_timeout_ticks) {
+					rxa_vector_expire(vec, rx_adapter);
+					TAILQ_REMOVE(&rx_adapter->vector_list,
+						     vec, next);
+				}
+			}
+			rx_adapter->prev_expiry_ts = rte_rdtsc();
+		}
+	}
+
 	stats = &rx_adapter->stats;
 	stats->rx_packets += rxa_intr_ring_dequeue(rx_adapter);
 	stats->rx_packets += rxa_poll(rx_adapter);
@@ -1640,6 +1784,28 @@ rxa_update_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }
 
+static void
+rxa_set_vector_data(struct eth_rx_queue_info *queue_info, uint16_t vector_count,
+		    uint64_t vector_ns, struct rte_mempool *mp, int32_t qid,
+		    uint16_t port_id)
+{
+#define NSEC2TICK(__ns, __freq) (((__ns) * (__freq)) / 1E9)
+	struct eth_rx_vector_data *vector_data;
+	uint32_t flow_id;
+
+	vector_data = &queue_info->vector_data;
+	vector_data->max_vector_count = vector_count;
+	vector_data->port = port_id;
+	vector_data->queue = qid;
+	vector_data->vector_pool = mp;
+	vector_data->vector_timeout_ticks =
+		NSEC2TICK(vector_ns, rte_get_timer_hz());
+	vector_data->ts = 0;
+	flow_id = queue_info->event & 0xFFFFF;
+	flow_id = flow_id == 0 ? (qid & 0xFF) | (port_id & 0xFFFF) : flow_id;
+	vector_data->event = (queue_info->event & ~0xFFFFF) | flow_id;
+}
+
 static void
 rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
 	struct eth_device_info *dev_info,
@@ -1741,6 +1907,44 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }
 
+static void
+rxa_sw_event_vector_configure(
+	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
+	int rx_queue_id,
+	const struct rte_event_eth_rx_adapter_event_vector_config *config)
+{
+	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
+	struct eth_rx_queue_info *queue_info;
+	struct rte_event *qi_ev;
+
+	if (rx_queue_id == -1) {
+		uint16_t nb_rx_queues;
+		uint16_t i;
+
+		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
+		for (i = 0; i < nb_rx_queues; i++)
+			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
+						      config);
+		return;
+	}
+
+	queue_info = &dev_info->rx_queue[rx_queue_id];
+	qi_ev = (struct rte_event *)&queue_info->event;
+	queue_info->ena_vector = 1;
+	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+	rxa_set_vector_data(queue_info, config->vector_sz,
+			    config->vector_timeout_ns, config->vector_mp,
+			    rx_queue_id, dev_info->dev->data->port_id);
+	rx_adapter->ena_vector = 1;
+	rx_adapter->vector_tmo_ticks =
+		rx_adapter->vector_tmo_ticks ?
+			      RTE_MIN(config->vector_timeout_ns << 1,
+				      rx_adapter->vector_tmo_ticks) :
+			      config->vector_timeout_ns << 1;
+	rx_adapter->prev_expiry_ts = 0;
+	TAILQ_INIT(&rx_adapter->vector_list);
+}
+
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -2081,6 +2285,15 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}
 
+	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
+	    (queue_conf->rx_queue_flags &
+	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
+		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
 		(rx_queue_id != -1)) {
 		RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
@@ -2143,6 +2356,17 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	return 0;
 }
 
+static int
+rxa_sw_vector_limits(struct rte_event_eth_rx_adapter_vector_limits *limits)
+{
+	limits->max_sz = MAX_VECTOR_SIZE;
+	limits->min_sz = MIN_VECTOR_SIZE;
+	limits->max_timeout_ns = MAX_VECTOR_NS;
+	limits->min_timeout_ns = MIN_VECTOR_NS;
+
+	return 0;
+}
+
 int
 rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 				int32_t rx_queue_id)
@@ -2263,6 +2487,121 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }
 
+int
+rte_event_eth_rx_adapter_queue_event_vector_config(
+	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
+	struct rte_event_eth_rx_adapter_event_vector_config *config)
+{
+	struct rte_event_eth_rx_adapter_vector_limits limits;
+	struct rte_event_eth_rx_adapter *rx_adapter;
+	struct rte_eventdev *dev;
+	uint32_t cap;
+	int ret;
+
+	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
+
+	rx_adapter = rxa_id_to_adapter(id);
+	if ((rx_adapter == NULL) || (config == NULL))
+		return -EINVAL;
+
+	dev = &rte_eventdevs[rx_adapter->eventdev_id];
+	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
+						eth_dev_id, &cap);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
+				 "eth port %" PRIu16,
+				 id, eth_dev_id);
+		return ret;
+	}
+
+	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
+		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
+	ret = rte_event_eth_rx_adapter_vector_limits_get(
+		rx_adapter->eventdev_id, eth_dev_id, &limits);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
+				 "eth port %" PRIu16,
+				 rx_adapter->eventdev_id, eth_dev_id);
+		return ret;
+	}
+
+	if (config->vector_sz < limits.min_sz ||
+	    config->vector_sz > limits.max_sz ||
+	    config->vector_timeout_ns < limits.min_timeout_ns ||
+	    config->vector_timeout_ns > limits.max_timeout_ns ||
+	    config->vector_mp == NULL) {
+		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+	if (config->vector_mp->elt_size <
+	    (sizeof(struct rte_event_vector) +
+	     (sizeof(uintptr_t) * config->vector_sz))) {
+		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
+	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
+		RTE_FUNC_PTR_OR_ERR_RET(
+			*dev->dev_ops->eth_rx_adapter_event_vector_config,
+			-ENOTSUP);
+		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
+			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
+	} else {
+		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
+					      rx_queue_id, config);
+	}
+
+	return ret;
+}
+
+int
+rte_event_eth_rx_adapter_vector_limits_get(
+	uint8_t dev_id, uint16_t eth_port_id,
+	struct rte_event_eth_rx_adapter_vector_limits *limits)
+{
+	struct rte_eventdev *dev;
+	uint32_t cap;
+	int ret;
+
+	RTE_EVENTDEV_VALID_DEVID_OR_ERR_RET(dev_id, -EINVAL);
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_port_id, -EINVAL);
+
+	if (limits == NULL)
+		return -EINVAL;
+
+	dev = &rte_eventdevs[dev_id];
+
+	ret = rte_event_eth_rx_adapter_caps_get(dev_id, eth_port_id, &cap);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
+				 "eth port %" PRIu16,
+				 dev_id, eth_port_id);
+		return ret;
+	}
+
+	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
+		RTE_FUNC_PTR_OR_ERR_RET(
+			*dev->dev_ops->eth_rx_adapter_vector_limits_get,
+			-ENOTSUP);
+		ret = dev->dev_ops->eth_rx_adapter_vector_limits_get(
+			dev, &rte_eth_devices[eth_port_id], limits);
+	} else {
+		ret = rxa_sw_vector_limits(limits);
+	}
+
+	return ret;
+}
+
 int
 rte_event_eth_rx_adapter_start(uint8_t id)
 {
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index b57363f80..2e6e367e0 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -122,7 +122,11 @@ rte_event_eth_rx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
 
 	if (caps == NULL)
 		return -EINVAL;
-	*caps = 0;
+
+	if (dev->dev_ops->eth_rx_adapter_caps_get == NULL)
+		*caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
+	else
+		*caps = 0;
 
 	return dev->dev_ops->eth_rx_adapter_caps_get ?
 				(*dev->dev_ops->eth_rx_adapter_caps_get)(dev,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v4 5/8] eventdev: add Tx adapter event vector support
  2021-03-19 20:57     ` [dpdk-dev] [PATCH v4 0/8] Introduce event vectorization pbhagavatula
                         ` (3 preceding siblings ...)
  2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 4/8] eventdev: add Rx adapter event vector support pbhagavatula
@ 2021-03-19 20:57       ` pbhagavatula
  2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
                         ` (4 subsequent siblings)
  9 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-19 20:57 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support for event eth Tx adapter, the implementation
receives events from the single linked queue and based on
rte_event_vector::union_valid transmits the vector of mbufs to a given
port, queue pair.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 .../rte_event_eth_tx_adapter.c                | 66 ++++++++++++++++---
 lib/librte_eventdev/rte_eventdev.c            |  5 +-
 2 files changed, 60 insertions(+), 11 deletions(-)

diff --git a/lib/librte_eventdev/rte_event_eth_tx_adapter.c b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
index 5b4c42dcf..db260bfb6 100644
--- a/lib/librte_eventdev/rte_event_eth_tx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
@@ -510,6 +510,47 @@ txa_service_buffer_retry(struct rte_mbuf **pkts, uint16_t unsent,
 	stats->tx_dropped += unsent - sent;
 }
 
+static uint16_t
+txa_process_event_vector(struct txa_service_data *txa,
+			 struct rte_event_vector *vec)
+{
+	struct txa_service_queue_info *tqi;
+	uint16_t port, queue, nb_tx = 0;
+	struct rte_mbuf **mbufs;
+	int i;
+
+	mbufs = (struct rte_mbuf **)vec->mbufs;
+	if (vec->attr_valid) {
+		port = vec->port;
+		queue = vec->queue;
+		tqi = txa_service_queue(txa, port, queue);
+		if (unlikely(tqi == NULL || !tqi->added)) {
+			rte_pktmbuf_free_bulk(mbufs, vec->nb_elem);
+			rte_mempool_put(rte_mempool_from_obj(vec), vec);
+			return 0;
+		}
+		for (i = 0; i < vec->nb_elem; i++) {
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
+						   mbufs[i]);
+		}
+	} else {
+		for (i = 0; i < vec->nb_elem; i++) {
+			port = mbufs[i]->port;
+			queue = rte_event_eth_tx_adapter_txq_get(mbufs[i]);
+			tqi = txa_service_queue(txa, port, queue);
+			if (unlikely(tqi == NULL || !tqi->added)) {
+				rte_pktmbuf_free(mbufs[i]);
+				continue;
+			}
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
+						   mbufs[i]);
+		}
+	}
+	rte_mempool_put(rte_mempool_from_obj(vec), vec);
+
+	return nb_tx;
+}
+
 static void
 txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
 	uint32_t n)
@@ -522,22 +563,27 @@ txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
 
 	nb_tx = 0;
 	for (i = 0; i < n; i++) {
-		struct rte_mbuf *m;
 		uint16_t port;
 		uint16_t queue;
 		struct txa_service_queue_info *tqi;
 
-		m = ev[i].mbuf;
-		port = m->port;
-		queue = rte_event_eth_tx_adapter_txq_get(m);
+		if (!(ev[i].event_type & RTE_EVENT_TYPE_VECTOR)) {
+			struct rte_mbuf *m;
 
-		tqi = txa_service_queue(txa, port, queue);
-		if (unlikely(tqi == NULL || !tqi->added)) {
-			rte_pktmbuf_free(m);
-			continue;
-		}
+			m = ev[i].mbuf;
+			port = m->port;
+			queue = rte_event_eth_tx_adapter_txq_get(m);
 
-		nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
+			tqi = txa_service_queue(txa, port, queue);
+			if (unlikely(tqi == NULL || !tqi->added)) {
+				rte_pktmbuf_free(m);
+				continue;
+			}
+
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
+		} else {
+			nb_tx += txa_process_event_vector(txa, ev[i].vec);
+		}
 	}
 
 	stats->tx_packets += nb_tx;
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index 2e6e367e0..b2e85ce67 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -196,7 +196,10 @@ rte_event_eth_tx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
 	if (caps == NULL)
 		return -EINVAL;
 
-	*caps = 0;
+	if (dev->dev_ops->eth_tx_adapter_caps_get == NULL)
+		*caps = RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR;
+	else
+		*caps = 0;
 
 	return dev->dev_ops->eth_tx_adapter_caps_get ?
 			(*dev->dev_ops->eth_tx_adapter_caps_get)(dev,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v4 6/8] app/eventdev: add event vector mode in pipeline test
  2021-03-19 20:57     ` [dpdk-dev] [PATCH v4 0/8] Introduce event vectorization pbhagavatula
                         ` (4 preceding siblings ...)
  2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 5/8] eventdev: add Tx " pbhagavatula
@ 2021-03-19 20:57       ` pbhagavatula
  2021-03-23 18:39         ` Jerin Jacob
  2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 7/8] doc: announce event Rx adapter config changes pbhagavatula
                         ` (3 subsequent siblings)
  9 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-03-19 20:57 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support in pipeline tests. By default this mode
is disabled, it can be enabled by using the option --enable_vector.
example:
	dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
	--nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
	--wlcores=20-23  --enable_vector

Addtional options to configure vector size and vector timeout are
also implemented and can be used by specifying --vector_size and
--vector_tmo_ns

This patch also adds a new option to set the number of Rx queues
configured per event eth rx adapter.
example:
	dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
	--nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
	--wlcores=20-23  --nb_eth_queues 4

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 app/test-eventdev/evt_common.h           |   4 +
 app/test-eventdev/evt_options.c          |  52 ++++
 app/test-eventdev/evt_options.h          |   4 +
 app/test-eventdev/test_pipeline_atq.c    | 310 ++++++++++++++++++++--
 app/test-eventdev/test_pipeline_common.c | 113 +++++++-
 app/test-eventdev/test_pipeline_common.h |  18 ++
 app/test-eventdev/test_pipeline_queue.c  | 320 +++++++++++++++++++++--
 doc/guides/tools/testeventdev.rst        |  28 ++
 8 files changed, 795 insertions(+), 54 deletions(-)

diff --git a/app/test-eventdev/evt_common.h b/app/test-eventdev/evt_common.h
index a1da1cf11..0e228258e 100644
--- a/app/test-eventdev/evt_common.h
+++ b/app/test-eventdev/evt_common.h
@@ -58,16 +58,20 @@ struct evt_options {
 	uint8_t sched_type_list[EVT_MAX_STAGES];
 	uint16_t mbuf_sz;
 	uint16_t wkr_deq_dep;
+	uint16_t vector_size;
+	uint16_t eth_queues;
 	uint32_t nb_flows;
 	uint32_t tx_first;
 	uint32_t max_pkt_sz;
 	uint32_t deq_tmo_nsec;
 	uint32_t q_priority:1;
 	uint32_t fwd_latency:1;
+	uint32_t ena_vector : 1;
 	uint64_t nb_pkts;
 	uint64_t nb_timers;
 	uint64_t expiry_nsec;
 	uint64_t max_tmo_nsec;
+	uint64_t vector_tmo_nsec;
 	uint64_t timer_tick_nsec;
 	uint64_t optm_timer_tick_nsec;
 	enum evt_prod_type prod_type;
diff --git a/app/test-eventdev/evt_options.c b/app/test-eventdev/evt_options.c
index 0d04ea9f8..0d5540574 100644
--- a/app/test-eventdev/evt_options.c
+++ b/app/test-eventdev/evt_options.c
@@ -34,6 +34,9 @@ evt_options_default(struct evt_options *opt)
 	opt->max_tmo_nsec = 1E5;  /* 100000ns ~100us */
 	opt->expiry_nsec = 1E4;   /* 10000ns ~10us */
 	opt->prod_type = EVT_PROD_TYPE_SYNT;
+	opt->eth_queues = 1;
+	opt->vector_size = 64;
+	opt->vector_tmo_nsec = 100E3;
 }
 
 typedef int (*option_parser_t)(struct evt_options *opt,
@@ -257,6 +260,43 @@ evt_parse_max_pkt_sz(struct evt_options *opt, const char *arg)
 	return ret;
 }
 
+static int
+evt_parse_ena_vector(struct evt_options *opt, const char *arg __rte_unused)
+{
+	opt->ena_vector = 1;
+	return 0;
+}
+
+static int
+evt_parse_vector_size(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint16(&(opt->vector_size), arg);
+
+	return ret;
+}
+
+static int
+evt_parse_vector_tmo_ns(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint64(&(opt->vector_tmo_nsec), arg);
+
+	return ret;
+}
+
+static int
+evt_parse_eth_queues(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint16(&(opt->eth_queues), arg);
+
+	return ret;
+}
+
 static void
 usage(char *program)
 {
@@ -289,6 +329,10 @@ usage(char *program)
 		"\t--expiry_nsec      : event timer expiry ns.\n"
 		"\t--mbuf_sz          : packet mbuf size.\n"
 		"\t--max_pkt_sz       : max packet size.\n"
+		"\t--nb_eth_queues    : number of ethernet Rx queues.\n"
+		"\t--enable_vector    : enable event vectorization.\n"
+		"\t--vector_size      : Max vector size.\n"
+		"\t--vector_tmo_ns    : Max vector timeout in nanoseconds\n"
 		);
 	printf("available tests:\n");
 	evt_test_dump_names();
@@ -360,6 +404,10 @@ static struct option lgopts[] = {
 	{ EVT_EXPIRY_NSEC,         1, 0, 0 },
 	{ EVT_MBUF_SZ,             1, 0, 0 },
 	{ EVT_MAX_PKT_SZ,          1, 0, 0 },
+	{ EVT_NB_ETH_QUEUES,       1, 0, 0 },
+	{ EVT_ENA_VECTOR,          0, 0, 0 },
+	{ EVT_VECTOR_SZ,           1, 0, 0 },
+	{ EVT_VECTOR_TMO,          1, 0, 0 },
 	{ EVT_HELP,                0, 0, 0 },
 	{ NULL,                    0, 0, 0 }
 };
@@ -394,6 +442,10 @@ evt_opts_parse_long(int opt_idx, struct evt_options *opt)
 		{ EVT_EXPIRY_NSEC, evt_parse_expiry_nsec},
 		{ EVT_MBUF_SZ, evt_parse_mbuf_sz},
 		{ EVT_MAX_PKT_SZ, evt_parse_max_pkt_sz},
+		{ EVT_NB_ETH_QUEUES, evt_parse_eth_queues},
+		{ EVT_ENA_VECTOR, evt_parse_ena_vector},
+		{ EVT_VECTOR_SZ, evt_parse_vector_size},
+		{ EVT_VECTOR_TMO, evt_parse_vector_tmo_ns},
 	};
 
 	for (i = 0; i < RTE_DIM(parsermap); i++) {
diff --git a/app/test-eventdev/evt_options.h b/app/test-eventdev/evt_options.h
index 748e54fae..1cea2a3e1 100644
--- a/app/test-eventdev/evt_options.h
+++ b/app/test-eventdev/evt_options.h
@@ -42,6 +42,10 @@
 #define EVT_EXPIRY_NSEC          ("expiry_nsec")
 #define EVT_MBUF_SZ              ("mbuf_sz")
 #define EVT_MAX_PKT_SZ           ("max_pkt_sz")
+#define EVT_NB_ETH_QUEUES        ("nb_eth_queues")
+#define EVT_ENA_VECTOR           ("enable_vector")
+#define EVT_VECTOR_SZ            ("vector_size")
+#define EVT_VECTOR_TMO           ("vector_tmo_ns")
 #define EVT_HELP                 ("help")
 
 void evt_options_default(struct evt_options *opt);
diff --git a/app/test-eventdev/test_pipeline_atq.c b/app/test-eventdev/test_pipeline_atq.c
index 0872b25b5..84dd4f44e 100644
--- a/app/test-eventdev/test_pipeline_atq.c
+++ b/app/test-eventdev/test_pipeline_atq.c
@@ -15,6 +15,8 @@ pipeline_atq_nb_event_queues(struct evt_options *opt)
 	return rte_eth_dev_count_avail();
 }
 
+typedef int (*pipeline_atq_worker_t)(void *arg);
+
 static __rte_noinline int
 pipeline_atq_worker_single_stage_tx(void *arg)
 {
@@ -113,6 +115,112 @@ pipeline_atq_worker_single_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_atq_worker_single_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+		vector_sz = ev.vec->nb_elem;
+		pipeline_event_tx_vector(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = ev.vec->nb_elem;
+		ev.queue_id = tx_queue[ev.vec->port];
+		ev.vec->queue = 0;
+		pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			vector_sz += ev[i].vec->nb_elem;
+			ev[i].vec->queue = 0;
+		}
+
+		pipeline_event_tx_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			ev[i].queue_id = tx_queue[ev[i].vec->port];
+			ev[i].vec->queue = 0;
+			vector_sz += ev[i].vec->nb_elem;
+			pipeline_fwd_event_vector(&ev[i],
+						  RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
 static __rte_noinline int
 pipeline_atq_worker_multi_stage_tx(void *arg)
 {
@@ -245,6 +353,147 @@ pipeline_atq_worker_multi_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.sub_event_type % nb_stages;
+
+		if (cq_id == last_queue) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+			continue;
+		}
+
+		ev.sub_event_type++;
+		pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.sub_event_type % nb_stages;
+
+		if (cq_id == last_queue) {
+			ev.queue_id = tx_queue[ev.vec->port];
+			ev.vec->queue = 0;
+			vector_sz = ev.vec->nb_elem;
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			pipeline_event_enqueue(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.sub_event_type++;
+			pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+			pipeline_event_enqueue(dev, port, &ev);
+		}
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].sub_event_type % nb_stages;
+
+			if (cq_id == last_queue) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+				continue;
+			}
+
+			ev[i].sub_event_type++;
+			pipeline_fwd_event_vector(&ev[i],
+						  sched_type_list[cq_id]);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].sub_event_type % nb_stages;
+
+			if (cq_id == last_queue) {
+				vector_sz = ev[i].vec->nb_elem;
+				ev[i].queue_id = tx_queue[ev[i].vec->port];
+				ev[i].vec->queue = 0;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].sub_event_type++;
+				pipeline_fwd_event_vector(
+					&ev[i], sched_type_list[cq_id]);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
 static int
 worker_wrapper(void *arg)
 {
@@ -253,27 +502,36 @@ worker_wrapper(void *arg)
 	const bool burst = evt_has_burst_mode(w->dev_id);
 	const bool internal_port = w->t->internal_port;
 	const uint8_t nb_stages = opt->nb_stages;
-	RTE_SET_USED(opt);
+	/*vector/burst/internal_port*/
+	const pipeline_atq_worker_t
+	pipeline_atq_worker_single_stage[2][2][2] = {
+		[0][0][0] = pipeline_atq_worker_single_stage_fwd,
+		[0][0][1] = pipeline_atq_worker_single_stage_tx,
+		[0][1][0] = pipeline_atq_worker_single_stage_burst_fwd,
+		[0][1][1] = pipeline_atq_worker_single_stage_burst_tx,
+		[1][0][0] = pipeline_atq_worker_single_stage_fwd_vector,
+		[1][0][1] = pipeline_atq_worker_single_stage_tx_vector,
+		[1][1][0] = pipeline_atq_worker_single_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_atq_worker_single_stage_burst_tx_vector,
+	};
+	const pipeline_atq_worker_t
+	pipeline_atq_worker_multi_stage[2][2][2] = {
+		[0][0][0] = pipeline_atq_worker_multi_stage_fwd,
+		[0][0][1] = pipeline_atq_worker_multi_stage_tx,
+		[0][1][0] = pipeline_atq_worker_multi_stage_burst_fwd,
+		[0][1][1] = pipeline_atq_worker_multi_stage_burst_tx,
+		[1][0][0] = pipeline_atq_worker_multi_stage_fwd_vector,
+		[1][0][1] = pipeline_atq_worker_multi_stage_tx_vector,
+		[1][1][0] = pipeline_atq_worker_multi_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_atq_worker_multi_stage_burst_tx_vector,
+	};
 
-	if (nb_stages == 1) {
-		if (!burst && internal_port)
-			return pipeline_atq_worker_single_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_atq_worker_single_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_atq_worker_single_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_atq_worker_single_stage_burst_fwd(arg);
-	} else {
-		if (!burst && internal_port)
-			return pipeline_atq_worker_multi_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_atq_worker_multi_stage_fwd(arg);
-		if (burst && internal_port)
-			return pipeline_atq_worker_multi_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_atq_worker_multi_stage_burst_fwd(arg);
-	}
+	if (nb_stages == 1)
+		return (pipeline_atq_worker_single_stage[opt->ena_vector][burst]
+							[internal_port])(arg);
+	else
+		return (pipeline_atq_worker_multi_stage[opt->ena_vector][burst]
+						       [internal_port])(arg);
 
 	rte_panic("invalid worker\n");
 }
@@ -290,7 +548,7 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
 	int ret;
 	int nb_ports;
 	int nb_queues;
-	uint8_t queue;
+	uint8_t queue, is_prod;
 	uint8_t tx_evqueue_id[RTE_MAX_ETHPORTS];
 	uint8_t queue_arr[RTE_EVENT_MAX_QUEUES_PER_DEV];
 	uint8_t nb_worker_queues = 0;
@@ -330,15 +588,19 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
 		q_conf.event_queue_cfg = RTE_EVENT_QUEUE_CFG_ALL_TYPES;
 
 		if (!t->internal_port) {
+			is_prod = false;
 			RTE_ETH_FOREACH_DEV(prod) {
 				if (queue == tx_evqueue_id[prod]) {
 					q_conf.event_queue_cfg =
 						RTE_EVENT_QUEUE_CFG_SINGLE_LINK;
-				} else {
-					queue_arr[nb_worker_queues] = queue;
-					nb_worker_queues++;
+					is_prod = true;
+					break;
 				}
 			}
+			if (!is_prod) {
+				queue_arr[nb_worker_queues] = queue;
+				nb_worker_queues++;
+			}
 		}
 
 		ret = rte_event_queue_setup(opt->dev_id, queue, &q_conf);
diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index b47d76743..d5ef90500 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -36,6 +36,12 @@ pipeline_opt_dump(struct evt_options *opt, uint8_t nb_queues)
 	evt_dump_queue_priority(opt);
 	evt_dump_sched_type_list(opt);
 	evt_dump_producer_type(opt);
+	evt_dump("nb_eth_rx_queues", "%d", opt->eth_queues);
+	evt_dump("event_vector", "%d", opt->ena_vector);
+	if (opt->ena_vector) {
+		evt_dump("vector_size", "%d", opt->vector_size);
+		evt_dump("vector_tmo_ns", "%" PRIu64 "", opt->vector_tmo_nsec);
+	}
 }
 
 static inline uint64_t
@@ -163,7 +169,7 @@ pipeline_opt_check(struct evt_options *opt, uint64_t nb_queues)
 int
 pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 {
-	uint16_t i;
+	uint16_t i, j;
 	int ret;
 	uint8_t nb_queues = 1;
 	struct test_pipeline *t = evt_test_priv(test);
@@ -210,6 +216,16 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 		if (!(caps & RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT))
 			t->internal_port = 0;
 
+		ret = rte_event_eth_rx_adapter_caps_get(opt->dev_id, i, &caps);
+		if (ret != 0) {
+			evt_err("failed to get event tx adapter[%d] caps", i);
+			return ret;
+		}
+
+		if (!(caps & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT))
+			local_port_conf.rxmode.offloads |=
+				DEV_RX_OFFLOAD_RSS_HASH;
+
 		ret = rte_eth_dev_info_get(i, &dev_info);
 		if (ret != 0) {
 			evt_err("Error during getting device (port %u) info: %s\n",
@@ -236,19 +252,22 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 				local_port_conf.rx_adv_conf.rss_conf.rss_hf);
 		}
 
-		if (rte_eth_dev_configure(i, nb_queues, nb_queues,
-					&local_port_conf)
-				< 0) {
+		if (rte_eth_dev_configure(i, opt->eth_queues, nb_queues,
+					  &local_port_conf) < 0) {
 			evt_err("Failed to configure eth port [%d]", i);
 			return -EINVAL;
 		}
 
-		if (rte_eth_rx_queue_setup(i, 0, NB_RX_DESC,
-				rte_socket_id(), &rx_conf, t->pool) < 0) {
-			evt_err("Failed to setup eth port [%d] rx_queue: %d.",
+		for (j = 0; j < opt->eth_queues; j++) {
+			if (rte_eth_rx_queue_setup(i, j, NB_RX_DESC,
+						   rte_socket_id(), &rx_conf,
+						   t->pool) < 0) {
+				evt_err("Failed to setup eth port [%d] rx_queue: %d.",
 					i, 0);
-			return -EINVAL;
+				return -EINVAL;
+			}
 		}
+
 		if (rte_eth_tx_queue_setup(i, 0, NB_TX_DESC,
 					rte_socket_id(), NULL) < 0) {
 			evt_err("Failed to setup eth port [%d] tx_queue: %d.",
@@ -310,12 +329,27 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 {
 	int ret = 0;
 	uint16_t prod;
+	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
+	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;
 
 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
 	queue_conf.ev.sched_type = opt->sched_type_list[0];
+	if (opt->ena_vector) {
+		unsigned int nb_elem = (opt->pool_sz / opt->vector_size) << 1;
+
+		nb_elem = nb_elem ? nb_elem : 1;
+		vector_pool = rte_event_vector_pool_create(
+			"vector_pool", nb_elem, 0, opt->vector_size,
+			opt->socket_id);
+		if (vector_pool == NULL) {
+			evt_err("failed to create event vector pool");
+			return -ENOMEM;
+		}
+	}
 	RTE_ETH_FOREACH_DEV(prod) {
+		struct rte_event_eth_rx_adapter_vector_limits limits;
 		uint32_t cap;
 
 		ret = rte_event_eth_rx_adapter_caps_get(opt->dev_id,
@@ -326,6 +360,50 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 					opt->dev_id);
 			return ret;
 		}
+
+		if (opt->ena_vector) {
+			memset(&limits, 0, sizeof(limits));
+			ret = rte_event_eth_rx_adapter_vector_limits_get(
+				opt->dev_id, prod, &limits);
+			if (ret) {
+				evt_err("failed to get vector limits");
+				return ret;
+			}
+
+			if (opt->vector_size < limits.min_sz ||
+			    opt->vector_size > limits.max_sz) {
+				evt_err("Vector size [%d] not within limits max[%d] min[%d]",
+					opt->vector_size, limits.min_sz,
+					limits.max_sz);
+				return -EINVAL;
+			}
+
+			if (limits.log2_sz &&
+			    !rte_is_power_of_2(opt->vector_size)) {
+				evt_err("Vector size [%d] not power of 2",
+					opt->vector_size);
+				return -EINVAL;
+			}
+
+			if (opt->vector_tmo_nsec > limits.max_timeout_ns ||
+			    opt->vector_tmo_nsec < limits.min_timeout_ns) {
+				evt_err("Vector timeout [%" PRIu64
+					"] not within limits max[%" PRIu64
+					"] min[%" PRIu64 "]",
+					opt->vector_tmo_nsec,
+					limits.max_timeout_ns,
+					limits.min_timeout_ns);
+				return -EINVAL;
+			}
+
+			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.rx_queue_flags |=
+				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+			} else {
+				evt_err("Rx adapter doesn't support event vector");
+				return -EINVAL;
+			}
+		}
 		queue_conf.ev.queue_id = prod * stride;
 		ret = rte_event_eth_rx_adapter_create(prod, opt->dev_id,
 				&prod_conf);
@@ -340,6 +418,17 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}
 
+		if (opt->ena_vector) {
+			vec_conf.vector_sz = opt->vector_size;
+			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
+			vec_conf.vector_mp = vector_pool;
+			if (rte_event_eth_rx_adapter_queue_event_vector_config(
+				    prod, prod, -1, &vec_conf) < 0) {
+				evt_err("Failed to configure event vectorization for Rx adapter");
+				return -EINVAL;
+			}
+		}
+
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;
 
@@ -378,6 +467,14 @@ pipeline_event_tx_adapter_setup(struct evt_options *opt,
 			return ret;
 		}
 
+		if (opt->ena_vector) {
+			if (!(cap &
+			      RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR)) {
+				evt_err("Tx adapter doesn't support event vector");
+				return -EINVAL;
+			}
+		}
+
 		ret = rte_event_eth_tx_adapter_create(consm, opt->dev_id,
 				&port_conf);
 		if (ret) {
diff --git a/app/test-eventdev/test_pipeline_common.h b/app/test-eventdev/test_pipeline_common.h
index 6e73c6ab2..800a90616 100644
--- a/app/test-eventdev/test_pipeline_common.h
+++ b/app/test-eventdev/test_pipeline_common.h
@@ -101,6 +101,14 @@ pipeline_fwd_event(struct rte_event *ev, uint8_t sched)
 	ev->sched_type = sched;
 }
 
+static __rte_always_inline void
+pipeline_fwd_event_vector(struct rte_event *ev, uint8_t sched)
+{
+	ev->event_type = RTE_EVENT_TYPE_CPU_VECTOR;
+	ev->op = RTE_EVENT_OP_FORWARD;
+	ev->sched_type = sched;
+}
+
 static __rte_always_inline void
 pipeline_event_tx(const uint8_t dev, const uint8_t port,
 		struct rte_event * const ev)
@@ -110,6 +118,16 @@ pipeline_event_tx(const uint8_t dev, const uint8_t port,
 		rte_pause();
 }
 
+static __rte_always_inline void
+pipeline_event_tx_vector(const uint8_t dev, const uint8_t port,
+			 struct rte_event *const ev)
+{
+	ev->vec->queue = 0;
+
+	while (!rte_event_eth_tx_adapter_enqueue(dev, port, ev, 1, 0))
+		rte_pause();
+}
+
 static __rte_always_inline void
 pipeline_event_tx_burst(const uint8_t dev, const uint8_t port,
 		struct rte_event *ev, const uint16_t nb_rx)
diff --git a/app/test-eventdev/test_pipeline_queue.c b/app/test-eventdev/test_pipeline_queue.c
index 9a9febb19..f6cc3e358 100644
--- a/app/test-eventdev/test_pipeline_queue.c
+++ b/app/test-eventdev/test_pipeline_queue.c
@@ -15,6 +15,8 @@ pipeline_queue_nb_event_queues(struct evt_options *opt)
 	return (eth_count * opt->nb_stages) + eth_count;
 }
 
+typedef int (*pipeline_queue_worker_t)(void *arg);
+
 static __rte_noinline int
 pipeline_queue_worker_single_stage_tx(void *arg)
 {
@@ -126,6 +128,125 @@ pipeline_queue_worker_single_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_queue_worker_single_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		if (ev.sched_type == RTE_SCHED_TYPE_ATOMIC) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.queue_id++;
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			pipeline_event_enqueue(dev, port, &ev);
+		}
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		ev.queue_id = tx_queue[ev.vec->port];
+		ev.vec->queue = 0;
+		vector_sz = ev.vec->nb_elem;
+		pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			if (ev[i].sched_type == RTE_SCHED_TYPE_ATOMIC) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].queue_id++;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			ev[i].queue_id = tx_queue[ev[i].vec->port];
+			ev[i].vec->queue = 0;
+			vector_sz += ev[i].vec->nb_elem;
+			pipeline_fwd_event_vector(&ev[i],
+						  RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
 
 static __rte_noinline int
 pipeline_queue_worker_multi_stage_tx(void *arg)
@@ -267,6 +388,151 @@ pipeline_queue_worker_multi_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.queue_id % nb_stages;
+
+		if (ev.queue_id == tx_queue[ev.vec->port]) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+			continue;
+		}
+
+		ev.queue_id++;
+		pipeline_fwd_event_vector(&ev, cq_id != last_queue
+						       ? sched_type_list[cq_id]
+						       : RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.queue_id % nb_stages;
+
+		if (cq_id == last_queue) {
+			vector_sz = ev.vec->nb_elem;
+			ev.queue_id = tx_queue[ev.vec->port];
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.queue_id++;
+			pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+		}
+
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].queue_id % nb_stages;
+
+			if (ev[i].queue_id == tx_queue[ev[i].vec->port]) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+				continue;
+			}
+
+			ev[i].queue_id++;
+			pipeline_fwd_event_vector(
+				&ev[i], cq_id != last_queue
+						? sched_type_list[cq_id]
+						: RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].queue_id % nb_stages;
+
+			if (cq_id == last_queue) {
+				ev[i].queue_id = tx_queue[ev[i].vec->port];
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].queue_id++;
+				pipeline_fwd_event_vector(
+					&ev[i], sched_type_list[cq_id]);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
 static int
 worker_wrapper(void *arg)
 {
@@ -275,29 +541,39 @@ worker_wrapper(void *arg)
 	const bool burst = evt_has_burst_mode(w->dev_id);
 	const bool internal_port = w->t->internal_port;
 	const uint8_t nb_stages = opt->nb_stages;
-	RTE_SET_USED(opt);
-
-	if (nb_stages == 1) {
-		if (!burst && internal_port)
-			return pipeline_queue_worker_single_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_queue_worker_single_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_queue_worker_single_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_queue_worker_single_stage_burst_fwd(
-					arg);
-	} else {
-		if (!burst && internal_port)
-			return pipeline_queue_worker_multi_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_queue_worker_multi_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_queue_worker_multi_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_queue_worker_multi_stage_burst_fwd(arg);
+	/*vector/burst/internal_port*/
+	const pipeline_queue_worker_t
+	pipeline_queue_worker_single_stage[2][2][2] = {
+		[0][0][0] = pipeline_queue_worker_single_stage_fwd,
+		[0][0][1] = pipeline_queue_worker_single_stage_tx,
+		[0][1][0] = pipeline_queue_worker_single_stage_burst_fwd,
+		[0][1][1] = pipeline_queue_worker_single_stage_burst_tx,
+		[1][0][0] = pipeline_queue_worker_single_stage_fwd_vector,
+		[1][0][1] = pipeline_queue_worker_single_stage_tx_vector,
+		[1][1][0] = pipeline_queue_worker_single_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_queue_worker_single_stage_burst_tx_vector,
+	};
+	const pipeline_queue_worker_t
+	pipeline_queue_worker_multi_stage[2][2][2] = {
+		[0][0][0] = pipeline_queue_worker_multi_stage_fwd,
+		[0][0][1] = pipeline_queue_worker_multi_stage_tx,
+		[0][1][0] = pipeline_queue_worker_multi_stage_burst_fwd,
+		[0][1][1] = pipeline_queue_worker_multi_stage_burst_tx,
+		[1][0][0] = pipeline_queue_worker_multi_stage_fwd_vector,
+		[1][0][1] = pipeline_queue_worker_multi_stage_tx_vector,
+		[1][1][0] = pipeline_queue_worker_multi_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_queue_worker_multi_stage_burst_tx_vector,
+	};
+
+	if (nb_stages == 1)
+		return (pipeline_queue_worker_single_stage[opt->ena_vector]
+							  [burst]
+							  [internal_port])(arg);
+	else
+		return (pipeline_queue_worker_multi_stage[opt->ena_vector]
+							 [burst]
+							 [internal_port])(arg);
 
-	}
 	rte_panic("invalid worker\n");
 }
 
diff --git a/doc/guides/tools/testeventdev.rst b/doc/guides/tools/testeventdev.rst
index ad1788a3d..691cf706e 100644
--- a/doc/guides/tools/testeventdev.rst
+++ b/doc/guides/tools/testeventdev.rst
@@ -158,6 +158,26 @@ The following are the application command-line options:
        Set max packet mbuf size. Can be used configure Rx/Tx scatter gather.
        Only applicable for `pipeline_atq` and `pipeline_queue` tests.
 
+* ``--nb_eth_queues``
+
+       Configure multiple Rx queues per each ethernet port.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--enable_vector``
+
+       Enable event vector for Rx/Tx adapters.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--vector_size``
+
+       Vector size to configure for the Rx adapter.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--vector_tmo_ns``
+
+       Vector timeout nanoseconds to be configured for the Rx adapter.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
 
 Eventdev Tests
 --------------
@@ -607,6 +627,10 @@ Supported application command line options are following::
         --worker_deq_depth
         --prod_type_ethdev
         --deq_tmo_nsec
+        --nb_eth_queues
+        --enable_vector
+        --vector_size
+        --vector_tmo_ns
 
 
 .. Note::
@@ -699,6 +723,10 @@ Supported application command line options are following::
         --worker_deq_depth
         --prod_type_ethdev
         --deq_tmo_nsec
+        --nb_eth_queues
+        --enable_vector
+        --vector_size
+        --vector_tmo_ns
 
 
 .. Note::
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v4 7/8] doc: announce event Rx adapter config changes
  2021-03-19 20:57     ` [dpdk-dev] [PATCH v4 0/8] Introduce event vectorization pbhagavatula
                         ` (5 preceding siblings ...)
  2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
@ 2021-03-19 20:57       ` pbhagavatula
  2021-03-19 20:57       ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v4 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
                         ` (2 subsequent siblings)
  9 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-19 20:57 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

The Rx adapter event vector configuration will be merged into
Rx adapter queue configuration to simplify enabling event
vectorization.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 doc/guides/rel_notes/deprecation.rst | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst
index 64629e064..ef6d1bb18 100644
--- a/doc/guides/rel_notes/deprecation.rst
+++ b/doc/guides/rel_notes/deprecation.rst
@@ -130,3 +130,12 @@ Deprecation Notices
 * cmdline: ``cmdline`` structure will be made opaque to hide platform-specific
   content. On Linux and FreeBSD, supported prior to DPDK 20.11,
   original structure will be kept until DPDK 21.11.
+
+* eventdev: The structure ``rte_event_eth_rx_adapter_queue_conf`` will be
+  extended to include ``rte_event_eth_rx_adapter_event_vector_config`` elements
+  and the function ``rte_event_eth_rx_adapter_queue_event_vector_config`` will
+  be removed in DPDK 21.11.
+
+  An application can enable event vectorization by passing the desired vector
+  values to the function ``rte_event_eth_rx_adapter_queue_add`` using
+  the structure ``rte_event_eth_rx_adapter_queue_add``.
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [dpdk-dev v21.11] [PATCH v4 8/8] eventdev: simplify Rx adapter event vector config
  2021-03-19 20:57     ` [dpdk-dev] [PATCH v4 0/8] Introduce event vectorization pbhagavatula
                         ` (6 preceding siblings ...)
  2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 7/8] doc: announce event Rx adapter config changes pbhagavatula
@ 2021-03-19 20:57       ` pbhagavatula
  2021-03-23 18:44       ` [dpdk-dev] [PATCH v4 0/8] Introduce event vectorization Jerin Jacob
  2021-03-24  5:05       ` [dpdk-dev] [PATCH v5 " pbhagavatula
  9 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-19 20:57 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Include vector configuration into the structure
``rte_event_eth_rx_adapter_queue_conf`` used when configuring rest
of the Rx adapter ethernet device Rx queue parameters.
This simplifies event vector configuration as it avoids splitting
configuration per Rx queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 app/test-eventdev/test_pipeline_common.c      |  16 +-
 lib/librte_eventdev/eventdev_pmd.h            |  29 ---
 .../rte_event_eth_rx_adapter.c                | 168 ++++++------------
 .../rte_event_eth_rx_adapter.h                |  27 ---
 lib/librte_eventdev/version.map               |   1 -
 5 files changed, 57 insertions(+), 184 deletions(-)

diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index d5ef90500..76aee254b 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -331,7 +331,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 	uint16_t prod;
 	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
-	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
@@ -397,8 +396,12 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			}

 			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.vector_sz = opt->vector_size;
+				queue_conf.vector_timeout_ns =
+					opt->vector_tmo_nsec;
 				queue_conf.rx_queue_flags |=
 				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+				queue_conf.vector_mp = vector_pool;
 			} else {
 				evt_err("Rx adapter doesn't support event vector");
 				return -EINVAL;
@@ -418,17 +421,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}

-		if (opt->ena_vector) {
-			vec_conf.vector_sz = opt->vector_size;
-			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
-			vec_conf.vector_mp = vector_pool;
-			if (rte_event_eth_rx_adapter_queue_event_vector_config(
-				    prod, prod, -1, &vec_conf) < 0) {
-				evt_err("Failed to configure event vectorization for Rx adapter");
-				return -EINVAL;
-			}
-		}
-
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;

diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index d79dfd612..60bfaebc0 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -667,32 +667,6 @@ typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
 	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-struct rte_event_eth_rx_adapter_event_vector_config;
-/**
- * Enable event vector on an given Rx queue of a ethernet devices belonging to
- * the Rx adapter.
- *
- * @param dev
- *   Event device pointer
- *
- * @param eth_dev
- *   Ethernet device pointer
- *
- * @param rx_queue_id
- *   The Rx queue identifier
- *
- * @param config
- *   Pointer to the event vector configuration structure.
- *
- * @return
- *   - 0: Success.
- *   - <0: Error code returned by the driver function.
- */
-typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
-	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
-	int32_t rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;

@@ -1118,9 +1092,6 @@ struct rte_eventdev_ops {
 	eventdev_eth_rx_adapter_vector_limits_get_t
 		eth_rx_adapter_vector_limits_get;
 	/**< Get event vector limits for the Rx adapter */
-	eventdev_eth_rx_adapter_event_vector_config_t
-		eth_rx_adapter_event_vector_config;
-	/**< Configure Rx adapter with event vector */

 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index c71990078..a1990637f 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -1882,6 +1882,25 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	} else
 		qi_ev->flow_id = 0;

+	if (conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+		queue_info->ena_vector = 1;
+		qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+		rxa_set_vector_data(queue_info, conf->vector_sz,
+				    conf->vector_timeout_ns, conf->vector_mp,
+				    rx_queue_id, dev_info->dev->data->port_id);
+		rx_adapter->ena_vector = 1;
+		rx_adapter->vector_tmo_ticks =
+			rx_adapter->vector_tmo_ticks
+				? RTE_MIN(queue_info->vector_data
+						  .vector_timeout_ticks,
+					  rx_adapter->vector_tmo_ticks)
+				: queue_info->vector_data.vector_timeout_ticks;
+		rx_adapter->vector_tmo_ticks <<= 1;
+		TAILQ_INIT(&rx_adapter->vector_list);
+		rx_adapter->prev_expiry_ts = 0;
+	}
+
 	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
 	if (rxa_polled_queue(dev_info, rx_queue_id)) {
 		rx_adapter->num_rx_polled += !pollq;
@@ -1907,44 +1926,6 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }

-static void
-rxa_sw_event_vector_configure(
-	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
-	int rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
-	struct eth_rx_queue_info *queue_info;
-	struct rte_event *qi_ev;
-
-	if (rx_queue_id == -1) {
-		uint16_t nb_rx_queues;
-		uint16_t i;
-
-		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
-		for (i = 0; i < nb_rx_queues; i++)
-			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
-						      config);
-		return;
-	}
-
-	queue_info = &dev_info->rx_queue[rx_queue_id];
-	qi_ev = (struct rte_event *)&queue_info->event;
-	queue_info->ena_vector = 1;
-	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
-	rxa_set_vector_data(queue_info, config->vector_sz,
-			    config->vector_timeout_ns, config->vector_mp,
-			    rx_queue_id, dev_info->dev->data->port_id);
-	rx_adapter->ena_vector = 1;
-	rx_adapter->vector_tmo_ticks =
-		rx_adapter->vector_tmo_ticks ?
-			      RTE_MIN(config->vector_timeout_ns << 1,
-				      rx_adapter->vector_tmo_ticks) :
-			      config->vector_timeout_ns << 1;
-	rx_adapter->prev_expiry_ts = 0;
-	TAILQ_INIT(&rx_adapter->vector_list);
-}
-
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -2258,6 +2239,7 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	struct rte_event_eth_rx_adapter *rx_adapter;
 	struct rte_eventdev *dev;
 	struct eth_device_info *dev_info;
+	struct rte_event_eth_rx_adapter_vector_limits limits;

 	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
 	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
@@ -2294,6 +2276,39 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}

+	if (queue_conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+		ret = rte_event_eth_rx_adapter_vector_limits_get(
+			rx_adapter->eventdev_id, eth_dev_id, &limits);
+		if (ret < 0) {
+			RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_sz < limits.min_sz ||
+		    queue_conf->vector_sz > limits.max_sz ||
+		    queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
+		    queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
+		    queue_conf->vector_mp == NULL) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_mp->elt_size <
+		    (sizeof(struct rte_event_vector) +
+		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+	}
+
 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
 		(rx_queue_id != -1)) {
 		RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
@@ -2487,83 +2502,6 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }

-int
-rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct rte_event_eth_rx_adapter_vector_limits limits;
-	struct rte_event_eth_rx_adapter *rx_adapter;
-	struct rte_eventdev *dev;
-	uint32_t cap;
-	int ret;
-
-	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
-	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
-
-	rx_adapter = rxa_id_to_adapter(id);
-	if ((rx_adapter == NULL) || (config == NULL))
-		return -EINVAL;
-
-	dev = &rte_eventdevs[rx_adapter->eventdev_id];
-	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
-						eth_dev_id, &cap);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
-				 "eth port %" PRIu16,
-				 id, eth_dev_id);
-		return ret;
-	}
-
-	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
-		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	ret = rte_event_eth_rx_adapter_vector_limits_get(
-		rx_adapter->eventdev_id, eth_dev_id, &limits);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
-				 "eth port %" PRIu16,
-				 rx_adapter->eventdev_id, eth_dev_id);
-		return ret;
-	}
-
-	if (config->vector_sz < limits.min_sz ||
-	    config->vector_sz > limits.max_sz ||
-	    config->vector_timeout_ns < limits.min_timeout_ns ||
-	    config->vector_timeout_ns > limits.max_timeout_ns ||
-	    config->vector_mp == NULL) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-	if (config->vector_mp->elt_size <
-	    (sizeof(struct rte_event_vector) +
-	     (sizeof(uintptr_t) * config->vector_sz))) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
-		RTE_FUNC_PTR_OR_ERR_RET(
-			*dev->dev_ops->eth_rx_adapter_event_vector_config,
-			-ENOTSUP);
-		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
-			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
-	} else {
-		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
-					      rx_queue_id, config);
-	}
-
-	return ret;
-}
-
 int
 rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
index dda846f57..3f9c749e9 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
@@ -171,9 +171,6 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
 	 * enqueued event.
 	 */
-};
-
-struct rte_event_eth_rx_adapter_event_vector_config {
 	uint16_t vector_sz;
 	/**<
 	 * Indicates the maximum number for mbufs to combine and form a vector.
@@ -538,30 +535,6 @@ int rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-/**
- * Configure event vectorization for a given ethernet device queue, that has
- * been added to a event eth Rx adapter.
- *
- * @param id
- *  The identifier of the ethernet Rx event adapter.
- *
- * @param eth_dev_id
- *  The identifier of the ethernet device.
- *
- * @param rx_queue_id
- *  Ethernet device receive queue index.
- *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
- *  are configured with event vectorization.
- *
- * @return
- *  - 0: Success, Receive queue configured correctly.
- *  - <0: Error code on failure.
- */
-__rte_experimental
-int rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index 902df0ae3..34c1c830e 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -142,7 +142,6 @@ EXPERIMENTAL {
 	#added in 21.05
 	rte_event_vector_pool_create;
 	rte_event_eth_rx_adapter_vector_limits_get;
-	rte_event_eth_rx_adapter_queue_event_vector_config;
 };

 INTERNAL {
--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v4 1/8] eventdev: introduce event vector capability
  2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 1/8] eventdev: introduce event vector capability pbhagavatula
@ 2021-03-22  9:06         ` Kinsella, Ray
  2021-03-22  9:10           ` [dpdk-dev] [EXT] " Pavan Nikhilesh Bhagavatula
  2021-03-23 11:12         ` [dpdk-dev] " Jerin Jacob
  1 sibling, 1 reply; 153+ messages in thread
From: Kinsella, Ray @ 2021-03-22  9:06 UTC (permalink / raw)
  To: pbhagavatula, jerinj, jay.jayatheerthan, erik.g.carrillo,
	abhinandan.gujjar, timothy.mcdaniel, hemant.agrawal,
	harry.van.haaren, mattias.ronnblom, liang.j.ma, Neil Horman
  Cc: dev



On 19/03/2021 20:57, pbhagavatula@marvell.com wrote:
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> Introduce rte_event_vector datastructure which is capable of holding
> multiple uintptr_t of the same flow thereby allowing applications
> to vectorize their pipeline and reducing the complexity of pipelining
> the events across multiple stages.
> This approach also reduces the scheduling overhead on a event device.
> 
> Add a event vector mempool create handler to create mempools based on
> the best mempool ops available on a given platform.
> 
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> ---
>  doc/guides/prog_guide/eventdev.rst |  36 +++++++++-
>  lib/librte_eventdev/rte_eventdev.h | 112 ++++++++++++++++++++++++++++-
>  lib/librte_eventdev/version.map    |   3 +
>  3 files changed, 148 insertions(+), 3 deletions(-)
> 

[SNIP]

>  
> diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
> index ce1fc2ce0..5586a3f15 100644
> --- a/lib/librte_eventdev/rte_eventdev.h
> +++ b/lib/librte_eventdev/rte_eventdev.h
> @@ -212,8 +212,10 @@ extern "C" {
>  
>  #include <rte_common.h>
>  #include <rte_config.h>
> -#include <rte_memory.h>
>  #include <rte_errno.h>
> +#include <rte_mbuf_pool_ops.h>
> +#include <rte_memory.h>
> +#include <rte_mempool.h>
>  
>  #include "rte_eventdev_trace_fp.h"
>  
> @@ -913,6 +915,25 @@ rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
>  int
>  rte_event_dev_close(uint8_t dev_id);
>  
> +/**
> + * Event vector structure.
> + */
> +struct rte_event_vector {
> +	uint64_t nb_elem : 16;
> +	/**< Number of elements in this event vector. */
> +	uint64_t rsvd : 48;
> +	uint64_t impl_opaque;
> +	union {
> +		struct rte_mbuf *mbufs[0];
> +		void *ptrs[0];
> +		uint64_t *u64s[0];
> +	} __rte_aligned(16);
> +	/**< Start of the vector array union. Depending upon the event type the
> +	 * vector array can be an array of mbufs or pointers or opaque u64
> +	 * values.
> +	 */
> +};
> +
>  /* Scheduler type definitions */
>  #define RTE_SCHED_TYPE_ORDERED          0
>  /**< Ordered scheduling
> @@ -986,6 +1007,21 @@ rte_event_dev_close(uint8_t dev_id);
>   */
>  #define RTE_EVENT_TYPE_ETH_RX_ADAPTER   0x4
>  /**< The event generated from event eth Rx adapter */
> +#define RTE_EVENT_TYPE_VECTOR           0x8
> +/**< Indicates that event is a vector.
> + * All vector event types should be an logical OR of EVENT_TYPE_VECTOR.
> + * This simplifies the pipeline design as we can split processing the events
> + * between vector events and normal event across event types.
> + * Example:
> + *	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
> + *		// Classify and handle vector event.
> + *	} else {
> + *		// Classify and handle event.
> + *	}
> + */
> +#define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
> +/**< The event vector generated from cpu for pipelining. */
> +
>  #define RTE_EVENT_TYPE_MAX              0x10
>  /**< Maximum number of event types */
>  
> @@ -1108,6 +1144,8 @@ struct rte_event {
>  		/**< Opaque event pointer */
>  		struct rte_mbuf *mbuf;
>  		/**< mbuf pointer if dequeued event is associated with mbuf */
> +		struct rte_event_vector *vec;
> +		/**< Event vector pointer. */
>  	};
>  };
>  
> @@ -2023,6 +2061,78 @@ rte_event_dev_xstats_reset(uint8_t dev_id,
>   */
>  int rte_event_dev_selftest(uint8_t dev_id);
>  
> +/**
> + * Get the memory required per event vector based on the number of elements per
> + * vector.
> + * This should be used to create the mempool that holds the event vectors.
> + *
> + * @param name
> + *   The name of the vector pool.
> + * @param n
> + *   The number of elements in the mbuf pool.
> + * @param cache_size
> + *   Size of the per-core object cache. See rte_mempool_create() for
> + *   details.
> + * @param nb_elem
> + *   The number of elements then a single event vector should be able to hold.
> + * @param socket_id
> + *   The socket identifier where the memory should be allocated. The
> + *   value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
> + *   reserved zone
> + *
> + * @return
> + *   The pointer to the newly allocated mempool, on success. NULL on error
> + *   with rte_errno set appropriately. Possible rte_errno values include:
> + *    - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
> + *    - E_RTE_SECONDARY - function was called from a secondary process instance
> + *    - EINVAL - cache size provided is too large, or priv_size is not aligned.
> + *    - ENOSPC - the maximum number of memzones has already been allocated
> + *    - EEXIST - a memzone with the same name already exists
> + *    - ENOMEM - no appropriate memory area found in which to create memzone
> + */
> +__rte_experimental
> +static inline struct rte_mempool *
> +rte_event_vector_pool_create(const char *name, unsigned int n,
> +			     unsigned int cache_size, uint16_t nb_elem,
> +			     int socket_id)

Handling in-lined function is tricky at best from an ABI stability PoV. 

Since this function is used at initialization time and I would suggest since performance is not issue here.
There is no need for this function to be an inline. 

> +{
> +	const char *mp_ops_name;
> +	struct rte_mempool *mp;
> +	unsigned int elt_sz;
> +	int ret;
> +
> +	if (!nb_elem) {
> +		RTE_LOG(ERR, EVENTDEV,
> +			"Invalid number of elements=%d requested\n", nb_elem);
> +		rte_errno = -EINVAL;
> +		return NULL;
> +	}
> +
> +	elt_sz =
> +		sizeof(struct rte_event_vector) + (nb_elem * sizeof(uintptr_t));
> +	mp = rte_mempool_create_empty(name, n, elt_sz, cache_size, 0, socket_id,
> +				      0);
> +	if (mp == NULL)
> +		return NULL;
> +
> +	mp_ops_name = rte_mbuf_best_mempool_ops();
> +	ret = rte_mempool_set_ops_byname(mp, mp_ops_name, NULL);
> +	if (ret != 0) {
> +		RTE_LOG(ERR, EVENTDEV, "error setting mempool handler\n");
> +		goto err;
> +	}
> +
> +	ret = rte_mempool_populate_default(mp);
> +	if (ret < 0)
> +		goto err;
> +
> +	return mp;
> +err:
> +	rte_mempool_free(mp);
> +	rte_errno = -ret;
> +	return NULL;
> +}
> +
>  #ifdef __cplusplus
>  }
>  #endif
> diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
> index 3e5c09cfd..a070ef56e 100644
> --- a/lib/librte_eventdev/version.map
> +++ b/lib/librte_eventdev/version.map
> @@ -138,6 +138,9 @@ EXPERIMENTAL {
>  	__rte_eventdev_trace_port_setup;
>  	# added in 20.11
>  	rte_event_pmd_pci_probe_named;
> +
> +	#added in 21.05
> +	rte_event_vector_pool_create;
>  };
>  
>  INTERNAL {
> 

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [EXT] Re: [PATCH v4 1/8] eventdev: introduce event vector capability
  2021-03-22  9:06         ` Kinsella, Ray
@ 2021-03-22  9:10           ` Pavan Nikhilesh Bhagavatula
  0 siblings, 0 replies; 153+ messages in thread
From: Pavan Nikhilesh Bhagavatula @ 2021-03-22  9:10 UTC (permalink / raw)
  To: Kinsella, Ray, Jerin Jacob Kollanukkaran, jay.jayatheerthan,
	erik.g.carrillo, abhinandan.gujjar, timothy.mcdaniel,
	hemant.agrawal, harry.van.haaren, mattias.ronnblom, liang.j.ma,
	Neil Horman
  Cc: dev

>On 19/03/2021 20:57, pbhagavatula@marvell.com wrote:
>> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>>
>> Introduce rte_event_vector datastructure which is capable of holding
>> multiple uintptr_t of the same flow thereby allowing applications
>> to vectorize their pipeline and reducing the complexity of pipelining
>> the events across multiple stages.
>> This approach also reduces the scheduling overhead on a event
>device.
>>
>> Add a event vector mempool create handler to create mempools
>based on
>> the best mempool ops available on a given platform.
>>
>> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
>> ---
>>  doc/guides/prog_guide/eventdev.rst |  36 +++++++++-
>>  lib/librte_eventdev/rte_eventdev.h | 112
>++++++++++++++++++++++++++++-
>>  lib/librte_eventdev/version.map    |   3 +
>>  3 files changed, 148 insertions(+), 3 deletions(-)
>>
>
>[SNIP]
>
>>
>> diff --git a/lib/librte_eventdev/rte_eventdev.h
>b/lib/librte_eventdev/rte_eventdev.h
>> index ce1fc2ce0..5586a3f15 100644
>> --- a/lib/librte_eventdev/rte_eventdev.h
>> +++ b/lib/librte_eventdev/rte_eventdev.h
>> @@ -212,8 +212,10 @@ extern "C" {
>>
>>  #include <rte_common.h>
>>  #include <rte_config.h>
>> -#include <rte_memory.h>
>>  #include <rte_errno.h>
>> +#include <rte_mbuf_pool_ops.h>
>> +#include <rte_memory.h>
>> +#include <rte_mempool.h>
>>
>>  #include "rte_eventdev_trace_fp.h"
>>
>> @@ -913,6 +915,25 @@
>rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
>>  int
>>  rte_event_dev_close(uint8_t dev_id);
>>
>> +/**
>> + * Event vector structure.
>> + */
>> +struct rte_event_vector {
>> +	uint64_t nb_elem : 16;
>> +	/**< Number of elements in this event vector. */
>> +	uint64_t rsvd : 48;
>> +	uint64_t impl_opaque;
>> +	union {
>> +		struct rte_mbuf *mbufs[0];
>> +		void *ptrs[0];
>> +		uint64_t *u64s[0];
>> +	} __rte_aligned(16);
>> +	/**< Start of the vector array union. Depending upon the event
>type the
>> +	 * vector array can be an array of mbufs or pointers or opaque
>u64
>> +	 * values.
>> +	 */
>> +};
>> +
>>  /* Scheduler type definitions */
>>  #define RTE_SCHED_TYPE_ORDERED          0
>>  /**< Ordered scheduling
>> @@ -986,6 +1007,21 @@ rte_event_dev_close(uint8_t dev_id);
>>   */
>>  #define RTE_EVENT_TYPE_ETH_RX_ADAPTER   0x4
>>  /**< The event generated from event eth Rx adapter */
>> +#define RTE_EVENT_TYPE_VECTOR           0x8
>> +/**< Indicates that event is a vector.
>> + * All vector event types should be an logical OR of
>EVENT_TYPE_VECTOR.
>> + * This simplifies the pipeline design as we can split processing the
>events
>> + * between vector events and normal event across event types.
>> + * Example:
>> + *	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
>> + *		// Classify and handle vector event.
>> + *	} else {
>> + *		// Classify and handle event.
>> + *	}
>> + */
>> +#define RTE_EVENT_TYPE_CPU_VECTOR
>(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
>> +/**< The event vector generated from cpu for pipelining. */
>> +
>>  #define RTE_EVENT_TYPE_MAX              0x10
>>  /**< Maximum number of event types */
>>
>> @@ -1108,6 +1144,8 @@ struct rte_event {
>>  		/**< Opaque event pointer */
>>  		struct rte_mbuf *mbuf;
>>  		/**< mbuf pointer if dequeued event is associated with
>mbuf */
>> +		struct rte_event_vector *vec;
>> +		/**< Event vector pointer. */
>>  	};
>>  };
>>
>> @@ -2023,6 +2061,78 @@ rte_event_dev_xstats_reset(uint8_t
>dev_id,
>>   */
>>  int rte_event_dev_selftest(uint8_t dev_id);
>>
>> +/**
>> + * Get the memory required per event vector based on the number of
>elements per
>> + * vector.
>> + * This should be used to create the mempool that holds the event
>vectors.
>> + *
>> + * @param name
>> + *   The name of the vector pool.
>> + * @param n
>> + *   The number of elements in the mbuf pool.
>> + * @param cache_size
>> + *   Size of the per-core object cache. See rte_mempool_create() for
>> + *   details.
>> + * @param nb_elem
>> + *   The number of elements then a single event vector should be
>able to hold.
>> + * @param socket_id
>> + *   The socket identifier where the memory should be allocated. The
>> + *   value can be *SOCKET_ID_ANY* if there is no NUMA constraint
>for the
>> + *   reserved zone
>> + *
>> + * @return
>> + *   The pointer to the newly allocated mempool, on success. NULL
>on error
>> + *   with rte_errno set appropriately. Possible rte_errno values
>include:
>> + *    - E_RTE_NO_CONFIG - function could not get pointer to
>rte_config structure
>> + *    - E_RTE_SECONDARY - function was called from a secondary
>process instance
>> + *    - EINVAL - cache size provided is too large, or priv_size is not
>aligned.
>> + *    - ENOSPC - the maximum number of memzones has already been
>allocated
>> + *    - EEXIST - a memzone with the same name already exists
>> + *    - ENOMEM - no appropriate memory area found in which to
>create memzone
>> + */
>> +__rte_experimental
>> +static inline struct rte_mempool *
>> +rte_event_vector_pool_create(const char *name, unsigned int n,
>> +			     unsigned int cache_size, uint16_t nb_elem,
>> +			     int socket_id)
>
>Handling in-lined function is tricky at best from an ABI stability PoV.
>
>Since this function is used at initialization time and I would suggest since
>performance is not issue here.
>There is no need for this function to be an inline.

Makes sense, I will move it to .c in the next version.

Thanks, 
Pavan.

>
>> +{
>> +	const char *mp_ops_name;
>> +	struct rte_mempool *mp;
>> +	unsigned int elt_sz;
>> +	int ret;
>> +
>> +	if (!nb_elem) {
>> +		RTE_LOG(ERR, EVENTDEV,
>> +			"Invalid number of elements=%d requested\n",
>nb_elem);
>> +		rte_errno = -EINVAL;
>> +		return NULL;
>> +	}
>> +
>> +	elt_sz =
>> +		sizeof(struct rte_event_vector) + (nb_elem *
>sizeof(uintptr_t));
>> +	mp = rte_mempool_create_empty(name, n, elt_sz, cache_size,
>0, socket_id,
>> +				      0);
>> +	if (mp == NULL)
>> +		return NULL;
>> +
>> +	mp_ops_name = rte_mbuf_best_mempool_ops();
>> +	ret = rte_mempool_set_ops_byname(mp, mp_ops_name,
>NULL);
>> +	if (ret != 0) {
>> +		RTE_LOG(ERR, EVENTDEV, "error setting mempool
>handler\n");
>> +		goto err;
>> +	}
>> +
>> +	ret = rte_mempool_populate_default(mp);
>> +	if (ret < 0)
>> +		goto err;
>> +
>> +	return mp;
>> +err:
>> +	rte_mempool_free(mp);
>> +	rte_errno = -ret;
>> +	return NULL;
>> +}
>> +
>>  #ifdef __cplusplus
>>  }
>>  #endif
>> diff --git a/lib/librte_eventdev/version.map
>b/lib/librte_eventdev/version.map
>> index 3e5c09cfd..a070ef56e 100644
>> --- a/lib/librte_eventdev/version.map
>> +++ b/lib/librte_eventdev/version.map
>> @@ -138,6 +138,9 @@ EXPERIMENTAL {
>>  	__rte_eventdev_trace_port_setup;
>>  	# added in 20.11
>>  	rte_event_pmd_pci_probe_named;
>> +
>> +	#added in 21.05
>> +	rte_event_vector_pool_create;
>>  };
>>
>>  INTERNAL {
>>

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v4 2/8] eventdev: introduce event vector Rx capability
  2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 2/8] eventdev: introduce event vector Rx capability pbhagavatula
@ 2021-03-22  9:12         ` Kinsella, Ray
  2021-03-22 10:07           ` [dpdk-dev] [EXT] " Pavan Nikhilesh Bhagavatula
  2021-03-23 16:56         ` [dpdk-dev] " Jerin Jacob
  1 sibling, 1 reply; 153+ messages in thread
From: Kinsella, Ray @ 2021-03-22  9:12 UTC (permalink / raw)
  To: pbhagavatula, jerinj, jay.jayatheerthan, erik.g.carrillo,
	abhinandan.gujjar, timothy.mcdaniel, hemant.agrawal,
	harry.van.haaren, mattias.ronnblom, liang.j.ma, Neil Horman
  Cc: dev



On 19/03/2021 20:57, pbhagavatula@marvell.com wrote:
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> Introduce event ethernet Rx adapter event vector capability.
> 
> If an event eth Rx adapter has the capability of
> RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR then a given Rx queue
> can be configured to enable event vectorization by passing the
> flag RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR to
> rte_event_eth_rx_adapter_queue_conf::rx_queue_flags while configuring
> Rx adapter through rte_event_eth_rx_adapter_queue_add.
> 
> The max vector size, vector timeout define the vector size and
> mempool used for allocating vector event are configured through
> rte_event_eth_rx_adapter_queue_add. The element size of the element
> in the vector pool should be equal to
>     sizeof(struct rte_event_vector) + (vector_sz * sizeof(uintptr_t))
> 
> Application can use `rte_event_vector_pool_create` to create the
> vector mempool used for
> rte_event_eth_rx_adapter_queue_conf::vector_mp.
> 
> The Rx adapter would be responsible for vectorizing the mbufs
> based on the flow, the vector limits configured by the application
> and add the vector event of mbufs to the event queue set via
> rte_event_eth_rx_adapter_queue_conf::ev::queue_id.
> It should also mark rte_event_vector::union_valid and fill
> rte_event_vector::port, rte_event_vector::queue.
> 
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> ---
>  .../prog_guide/event_ethernet_rx_adapter.rst  | 38 ++++++++
>  .../rte_event_eth_rx_adapter.h                | 95 +++++++++++++++++++
>  lib/librte_eventdev/rte_eventdev.h            | 30 +++++-
>  lib/librte_eventdev/version.map               |  2 +
>  4 files changed, 163 insertions(+), 2 deletions(-)
> 

Seems strange that you add the forward declaration of the symbol,
and you also detail the symbol in the map file, indicating how the symbol should be treated. 

But the patch is missing the symbol implementation itself, it should be here right? 




^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [EXT] Re: [PATCH v4 2/8] eventdev: introduce event vector Rx capability
  2021-03-22  9:12         ` Kinsella, Ray
@ 2021-03-22 10:07           ` Pavan Nikhilesh Bhagavatula
  2021-03-22 11:07             ` Kinsella, Ray
  0 siblings, 1 reply; 153+ messages in thread
From: Pavan Nikhilesh Bhagavatula @ 2021-03-22 10:07 UTC (permalink / raw)
  To: Kinsella, Ray, Jerin Jacob Kollanukkaran, jay.jayatheerthan,
	erik.g.carrillo, abhinandan.gujjar, timothy.mcdaniel,
	hemant.agrawal, harry.van.haaren, mattias.ronnblom, liang.j.ma,
	Neil Horman
  Cc: dev


>On 19/03/2021 20:57, pbhagavatula@marvell.com wrote:
>> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>>
>> Introduce event ethernet Rx adapter event vector capability.
>>
>> If an event eth Rx adapter has the capability of
>> RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR then a given Rx
>queue
>> can be configured to enable event vectorization by passing the
>> flag RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR to
>> rte_event_eth_rx_adapter_queue_conf::rx_queue_flags while
>configuring
>> Rx adapter through rte_event_eth_rx_adapter_queue_add.
>>
>> The max vector size, vector timeout define the vector size and
>> mempool used for allocating vector event are configured through
>> rte_event_eth_rx_adapter_queue_add. The element size of the
>element
>> in the vector pool should be equal to
>>     sizeof(struct rte_event_vector) + (vector_sz * sizeof(uintptr_t))
>>
>> Application can use `rte_event_vector_pool_create` to create the
>> vector mempool used for
>> rte_event_eth_rx_adapter_queue_conf::vector_mp.
>>
>> The Rx adapter would be responsible for vectorizing the mbufs
>> based on the flow, the vector limits configured by the application
>> and add the vector event of mbufs to the event queue set via
>> rte_event_eth_rx_adapter_queue_conf::ev::queue_id.
>> It should also mark rte_event_vector::union_valid and fill
>> rte_event_vector::port, rte_event_vector::queue.
>>
>> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
>> ---
>>  .../prog_guide/event_ethernet_rx_adapter.rst  | 38 ++++++++
>>  .../rte_event_eth_rx_adapter.h                | 95 +++++++++++++++++++
>>  lib/librte_eventdev/rte_eventdev.h            | 30 +++++-
>>  lib/librte_eventdev/version.map               |  2 +
>>  4 files changed, 163 insertions(+), 2 deletions(-)
>>
>
>Seems strange that you add the forward declaration of the symbol,
>and you also detail the symbol in the map file, indicating how the
>symbol should be treated.

I wanted to have only spec change here to make it easier for review,
the actual implementation is in [4/8].

>
>But the patch is missing the symbol implementation itself, it should be
>here right?
>

I will add a skeleton implementation here.

>


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [EXT] Re: [PATCH v4 2/8] eventdev: introduce event vector Rx capability
  2021-03-22 10:07           ` [dpdk-dev] [EXT] " Pavan Nikhilesh Bhagavatula
@ 2021-03-22 11:07             ` Kinsella, Ray
  0 siblings, 0 replies; 153+ messages in thread
From: Kinsella, Ray @ 2021-03-22 11:07 UTC (permalink / raw)
  To: Pavan Nikhilesh Bhagavatula, Jerin Jacob Kollanukkaran,
	jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Neil Horman
  Cc: dev



On 22/03/2021 10:07, Pavan Nikhilesh Bhagavatula wrote:
> 
>> On 19/03/2021 20:57, pbhagavatula@marvell.com wrote:
>>> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>>>
>>> Introduce event ethernet Rx adapter event vector capability.
>>>
>>> If an event eth Rx adapter has the capability of
>>> RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR then a given Rx
>> queue
>>> can be configured to enable event vectorization by passing the
>>> flag RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR to
>>> rte_event_eth_rx_adapter_queue_conf::rx_queue_flags while
>> configuring
>>> Rx adapter through rte_event_eth_rx_adapter_queue_add.
>>>
>>> The max vector size, vector timeout define the vector size and
>>> mempool used for allocating vector event are configured through
>>> rte_event_eth_rx_adapter_queue_add. The element size of the
>> element
>>> in the vector pool should be equal to
>>>     sizeof(struct rte_event_vector) + (vector_sz * sizeof(uintptr_t))
>>>
>>> Application can use `rte_event_vector_pool_create` to create the
>>> vector mempool used for
>>> rte_event_eth_rx_adapter_queue_conf::vector_mp.
>>>
>>> The Rx adapter would be responsible for vectorizing the mbufs
>>> based on the flow, the vector limits configured by the application
>>> and add the vector event of mbufs to the event queue set via
>>> rte_event_eth_rx_adapter_queue_conf::ev::queue_id.
>>> It should also mark rte_event_vector::union_valid and fill
>>> rte_event_vector::port, rte_event_vector::queue.
>>>
>>> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
>>> ---
>>>  .../prog_guide/event_ethernet_rx_adapter.rst  | 38 ++++++++
>>>  .../rte_event_eth_rx_adapter.h                | 95 +++++++++++++++++++
>>>  lib/librte_eventdev/rte_eventdev.h            | 30 +++++-
>>>  lib/librte_eventdev/version.map               |  2 +
>>>  4 files changed, 163 insertions(+), 2 deletions(-)
>>>
>>
>> Seems strange that you add the forward declaration of the symbol,
>> and you also detail the symbol in the map file, indicating how the
>> symbol should be treated.
> 
> I wanted to have only spec change here to make it easier for review,
> the actual implementation is in [4/8].
> 
>>
>> But the patch is missing the symbol implementation itself, it should be
>> here right?
>>
> 
> I will add a skeleton implementation here.

perfect.
> 
>>
> 

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v4 1/8] eventdev: introduce event vector capability
  2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 1/8] eventdev: introduce event vector capability pbhagavatula
  2021-03-22  9:06         ` Kinsella, Ray
@ 2021-03-23 11:12         ` Jerin Jacob
  1 sibling, 0 replies; 153+ messages in thread
From: Jerin Jacob @ 2021-03-23 11:12 UTC (permalink / raw)
  To: Pavan Nikhilesh
  Cc: Jerin Jacob, Jayatheerthan, Jay, Erik Gabriel Carrillo, Gujjar,
	Abhinandan S, McDaniel, Timothy, Hemant Agrawal, Van Haaren,
	Harry, Mattias Rönnblom, Liang Ma, Ray Kinsella,
	Neil Horman, dpdk-dev

On Sat, Mar 20, 2021 at 2:27 AM <pbhagavatula@marvell.com> wrote:
>
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>
> Introduce rte_event_vector datastructure which is capable of holding
> multiple uintptr_t of the same flow thereby allowing applications
> to vectorize their pipeline and reducing the complexity of pipelining
> the events across multiple stages.
> This approach also reduces the scheduling overhead on a event device.
>
> Add a event vector mempool create handler to create mempools based on
> the best mempool ops available on a given platform.
>
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>

Some minor comments below.
Feel free to add Acked-by: Jerin Jacob <jerinj@marvell.com> after
fixing the comments and Ray's suggestion.

> ---
>  doc/guides/prog_guide/eventdev.rst |  36 +++++++++-
>  lib/librte_eventdev/rte_eventdev.h | 112 ++++++++++++++++++++++++++++-
>  lib/librte_eventdev/version.map    |   3 +
>  3 files changed, 148 insertions(+), 3 deletions(-)
>
> diff --git a/doc/guides/prog_guide/eventdev.rst b/doc/guides/prog_guide/eventdev.rst
> index ccde086f6..fda9c3743 100644
> --- a/doc/guides/prog_guide/eventdev.rst
> +++ b/doc/guides/prog_guide/eventdev.rst
> @@ -63,13 +63,45 @@ the actual event being scheduled is. The payload is a union of the following:
>  * ``uint64_t u64``
>  * ``void *event_ptr``
>  * ``struct rte_mbuf *mbuf``
> +* ``struct rte_event_vector *vec``
>
> -These three items in a union occupy the same 64 bits at the end of the rte_event
> +These four items in a union occupy the same 64 bits at the end of the rte_event
>  structure. The application can utilize the 64 bits directly by accessing the
> -u64 variable, while the event_ptr and mbuf are provided as convenience
> +u64 variable, while the event_ptr, mbuf, vec are provided as a convenience
>  variables.  For example the mbuf pointer in the union can used to schedule a
>  DPDK packet.
>
> +Event Vector
> +~~~~~~~~~~~~
> +
> +The rte_event_vector struct contains a vector of elements defined by the event
> +type specified in the ``rte_event``. The event_vector structure contains the
> +following data:
> +
> +* ``nb_elem`` - The number of elements held within the vector.
> +
> +Similar to ``rte_event`` the payload of event vector is also a union, allowing
> +flexibility in what the actual vector is.
> +
> +* ``struct rte_mbuf *mbufs[0]`` - An array of mbufs.
> +* ``void *ptrs[0]`` - An array of pointers.
> +* ``uint64_t *u64s[0]`` - An array of uint64_t elements.
> +
> +The size of the event vector is related to the total number of elements it is
> +configured to hold, this is achieved by making `rte_event_vector` a variable
> +length structure.
> +A helper function is provided to create a mempool that holds event vector, which
> +takes name of the pool, total number of required ``rte_event_vector``,
> +cache size, number of elements in each ``rte_event_vector`` and socket id.
> +
> +.. code-block:: c
> +
> +        rte_event_vector_pool_create("vector_pool", nb_event_vectors, cache_sz,
> +                                     nb_elements_per_vector, socket_id);
> +
> +The function ``rte_event_vector_pool_create`` creates mempool with the best
> +platform mempool ops.
> +
>  Queues
>  ~~~~~~
>
> diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
> index ce1fc2ce0..5586a3f15 100644
> --- a/lib/librte_eventdev/rte_eventdev.h
> +++ b/lib/librte_eventdev/rte_eventdev.h
> @@ -212,8 +212,10 @@ extern "C" {
>
>  #include <rte_common.h>
>  #include <rte_config.h>
> -#include <rte_memory.h>
>  #include <rte_errno.h>
> +#include <rte_mbuf_pool_ops.h>
> +#include <rte_memory.h>
> +#include <rte_mempool.h>
>
>  #include "rte_eventdev_trace_fp.h"
>
> @@ -913,6 +915,25 @@ rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
>  int
>  rte_event_dev_close(uint8_t dev_id);
>
> +/**
> + * Event vector structure.
> + */
> +struct rte_event_vector {
> +       uint64_t nb_elem : 16;
> +       /**< Number of elements in this event vector. */
> +       uint64_t rsvd : 48;

Please add comment here to look Doxygen output correctly.


> +       uint64_t impl_opaque;

Please add comment here to look Doxygen output correctly.

> +       union {
> +               struct rte_mbuf *mbufs[0];
> +               void *ptrs[0];
> +               uint64_t *u64s[0];
> +       } __rte_aligned(16);
> +       /**< Start of the vector array union. Depending upon the event type the
> +        * vector array can be an array of mbufs or pointers or opaque u64
> +        * values.
> +        */
> +};
> +
>  /* Scheduler type definitions */
>  #define RTE_SCHED_TYPE_ORDERED          0
>  /**< Ordered scheduling
> @@ -986,6 +1007,21 @@ rte_event_dev_close(uint8_t dev_id);
>   */
>  #define RTE_EVENT_TYPE_ETH_RX_ADAPTER   0x4
>  /**< The event generated from event eth Rx adapter */
> +#define RTE_EVENT_TYPE_VECTOR           0x8
> +/**< Indicates that event is a vector.
> + * All vector event types should be an logical OR of EVENT_TYPE_VECTOR.

an logical -> a logical?

> + * This simplifies the pipeline design as we can split processing the events

we -> one

> + * between vector events and normal event across event types.
> + * Example:
> + *     if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
> + *             // Classify and handle vector event.
> + *     } else {
> + *             // Classify and handle event.
> + *     }
> + */
> +#define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
> +/**< The event vector generated from cpu for pipelining. */
> +
>  #define RTE_EVENT_TYPE_MAX              0x10
>  /**< Maximum number of event types */
>
> @@ -1108,6 +1144,8 @@ struct rte_event {
>                 /**< Opaque event pointer */
>                 struct rte_mbuf *mbuf;
>                 /**< mbuf pointer if dequeued event is associated with mbuf */
> +               struct rte_event_vector *vec;
> +               /**< Event vector pointer. */
>         };
>  };
>
> @@ -2023,6 +2061,78 @@ rte_event_dev_xstats_reset(uint8_t dev_id,
>   */
>  int rte_event_dev_selftest(uint8_t dev_id);
>
> +/**
> + * Get the memory required per event vector based on the number of elements per
> + * vector.
> + * This should be used to create the mempool that holds the event vectors.
> + *
> + * @param name
> + *   The name of the vector pool.
> + * @param n
> + *   The number of elements in the mbuf pool.
> + * @param cache_size
> + *   Size of the per-core object cache. See rte_mempool_create() for
> + *   details.
> + * @param nb_elem
> + *   The number of elements then a single event vector should be able to hold.
> + * @param socket_id
> + *   The socket identifier where the memory should be allocated. The
> + *   value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
> + *   reserved zone
> + *
> + * @return
> + *   The pointer to the newly allocated mempool, on success. NULL on error
> + *   with rte_errno set appropriately. Possible rte_errno values include:
> + *    - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
> + *    - E_RTE_SECONDARY - function was called from a secondary process instance
> + *    - EINVAL - cache size provided is too large, or priv_size is not aligned.
> + *    - ENOSPC - the maximum number of memzones has already been allocated
> + *    - EEXIST - a memzone with the same name already exists
> + *    - ENOMEM - no appropriate memory area found in which to create memzone
> + */
> +__rte_experimental
> +static inline struct rte_mempool *
> +rte_event_vector_pool_create(const char *name, unsigned int n,
> +                            unsigned int cache_size, uint16_t nb_elem,
> +                            int socket_id)
> +{
> +       const char *mp_ops_name;
> +       struct rte_mempool *mp;
> +       unsigned int elt_sz;
> +       int ret;
> +
> +       if (!nb_elem) {
> +               RTE_LOG(ERR, EVENTDEV,
> +                       "Invalid number of elements=%d requested\n", nb_elem);
> +               rte_errno = -EINVAL;
> +               return NULL;
> +       }
> +
> +       elt_sz =
> +               sizeof(struct rte_event_vector) + (nb_elem * sizeof(uintptr_t));
> +       mp = rte_mempool_create_empty(name, n, elt_sz, cache_size, 0, socket_id,
> +                                     0);
> +       if (mp == NULL)
> +               return NULL;
> +
> +       mp_ops_name = rte_mbuf_best_mempool_ops();
> +       ret = rte_mempool_set_ops_byname(mp, mp_ops_name, NULL);
> +       if (ret != 0) {
> +               RTE_LOG(ERR, EVENTDEV, "error setting mempool handler\n");
> +               goto err;
> +       }
> +
> +       ret = rte_mempool_populate_default(mp);
> +       if (ret < 0)
> +               goto err;
> +
> +       return mp;
> +err:
> +       rte_mempool_free(mp);
> +       rte_errno = -ret;
> +       return NULL;
> +}
> +
>  #ifdef __cplusplus
>  }
>  #endif
> diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
> index 3e5c09cfd..a070ef56e 100644
> --- a/lib/librte_eventdev/version.map
> +++ b/lib/librte_eventdev/version.map
> @@ -138,6 +138,9 @@ EXPERIMENTAL {
>         __rte_eventdev_trace_port_setup;
>         # added in 20.11
>         rte_event_pmd_pci_probe_named;
> +
> +       #added in 21.05
> +       rte_event_vector_pool_create;
>  };
>
>  INTERNAL {
> --
> 2.17.1
>

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v4 2/8] eventdev: introduce event vector Rx capability
  2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 2/8] eventdev: introduce event vector Rx capability pbhagavatula
  2021-03-22  9:12         ` Kinsella, Ray
@ 2021-03-23 16:56         ` Jerin Jacob
  1 sibling, 0 replies; 153+ messages in thread
From: Jerin Jacob @ 2021-03-23 16:56 UTC (permalink / raw)
  To: Pavan Nikhilesh
  Cc: Jerin Jacob, Jayatheerthan, Jay, Erik Gabriel Carrillo, Gujjar,
	Abhinandan S, McDaniel, Timothy, Hemant Agrawal, Van Haaren,
	Harry, Mattias Rönnblom, Liang Ma, Ray Kinsella,
	Neil Horman, dpdk-dev

On Sat, Mar 20, 2021 at 2:27 AM <pbhagavatula@marvell.com> wrote:
>
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>
> Introduce event ethernet Rx adapter event vector capability.
>
> If an event eth Rx adapter has the capability of
> RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR then a given Rx queue
> can be configured to enable event vectorization by passing the
> flag RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR to
> rte_event_eth_rx_adapter_queue_conf::rx_queue_flags while configuring
> Rx adapter through rte_event_eth_rx_adapter_queue_add.

through rte_event_eth_rx_adapter_queue_add()

>
> The max vector size, vector timeout define the vector size and
> mempool used for allocating vector event are configured through
> rte_event_eth_rx_adapter_queue_add. The element size of the element
> in the vector pool should be equal to
>     sizeof(struct rte_event_vector) + (vector_sz * sizeof(uintptr_t))
>
> Application can use `rte_event_vector_pool_create` to create the
> vector mempool used for
> rte_event_eth_rx_adapter_queue_conf::vector_mp.
>
> The Rx adapter would be responsible for vectorizing the mbufs
> based on the flow, the vector limits configured by the application
> and add the vector event of mbufs to the event queue set via
> rte_event_eth_rx_adapter_queue_conf::ev::queue_id.
> It should also mark rte_event_vector::union_valid and fill
> rte_event_vector::port, rte_event_vector::queue.
>
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>


Please find below some comments.
Feel free to add Acked-by: Jerin Jacob <jerinj@marvell.com> after
addressing those and Ray's comment.


> ---
>  .../prog_guide/event_ethernet_rx_adapter.rst  | 38 ++++++++
>  .../rte_event_eth_rx_adapter.h                | 95 +++++++++++++++++++
>  lib/librte_eventdev/rte_eventdev.h            | 30 +++++-
>  lib/librte_eventdev/version.map               |  2 +
>  4 files changed, 163 insertions(+), 2 deletions(-)
>

>  /**
>   * A structure used to retrieve statistics for an eth rx adapter instance.
>   */
> @@ -199,6 +233,22 @@ struct rte_event_eth_rx_adapter_stats {
>         /**< Received packet count for interrupt mode Rx queues */
>  };
>
> +/**
> + * A structure used to retrieve eth rx adapter vector limits.
> + */
> +struct rte_event_eth_rx_adapter_vector_limits {
> +       uint16_t min_sz;
> +       /**< Minimum vector limit configurable. */

Please add @see to rte_event_eth_rx_adapter_event_vector_config::vector_sz

> +       uint16_t max_sz;

Please add @see to rte_event_eth_rx_adapter_event_vector_config::vector_sz

> +       /**< Maximum vector limit configurable. */
> +       uint8_t log2_sz;

I think, we can change to bool and comment as "True if  the size
configured should be in log2"

Please add @see to rte_event_eth_rx_adapter_event_vector_config::vector_sz


> +       /**< The size configured should be in log2. */
> +       uint64_t min_timeout_ns;
> +       /**< Minimum vector timeout configurable. */

Please add @see to
rte_event_eth_rx_adapter_event_vector_config::vector_timeout_ns

> +       uint64_t max_timeout_ns;
> +       /**< Maximum vector timeout configurable. */

Please add @see to
rte_event_eth_rx_adapter_event_vector_config::vector_timeout_ns

> +};
> +
>  /**
>   *
>   * Callback function invoked by the SW adapter before it continues
> @@ -467,6 +517,51 @@ int rte_event_eth_rx_adapter_cb_register(uint8_t id, uint16_t eth_dev_id,
>                                          rte_event_eth_rx_adapter_cb_fn cb_fn,
>                                          void *cb_arg);
>
> +/**
> + * Retrieve vector limits for a given event dev and eth dev pair.
> + * @see rte_event_eth_rx_adapter_vector_limits
> + *
> + * @param dev_id
> + *  Event device identifier.
> + * @param eth_port_id
> + *  Port identifier of the ethernet device.
> + * @param [out] limits
> + *  A pointer to rte_event_eth_rx_adapter_vector_limits structure that has to
> + * be filled.
> + *
> + * @return
> + *  - 0: Success.
> + *  - <0: Error code on failure.
> + */
> +__rte_experimental
> +int rte_event_eth_rx_adapter_vector_limits_get(
> +       uint8_t dev_id, uint16_t eth_port_id,
> +       struct rte_event_eth_rx_adapter_vector_limits *limits);
> +
> +/**
> + * Configure event vectorization for a given ethernet device queue, that has
> + * been added to a event eth Rx adapter.
> + *
> + * @param id
> + *  The identifier of the ethernet Rx event adapter.
> + *
> + * @param eth_dev_id
> + *  The identifier of the ethernet device.
> + *
> + * @param rx_queue_id
> + *  Ethernet device receive queue index.
> + *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
> + *  are configured with event vectorization.
> + *
> + * @return
> + *  - 0: Success, Receive queue configured correctly.
> + *  - <0: Error code on failure.
> + */
> +__rte_experimental
> +int rte_event_eth_rx_adapter_queue_event_vector_config(
> +       uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
> +       struct rte_event_eth_rx_adapter_event_vector_config *config);
> +
>  #ifdef __cplusplus
>  }
>  #endif
> diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
> index 5586a3f15..1be79b856 100644
> --- a/lib/librte_eventdev/rte_eventdev.h
> +++ b/lib/librte_eventdev/rte_eventdev.h
> @@ -919,9 +919,27 @@ rte_event_dev_close(uint8_t dev_id);
>   * Event vector structure.
>   */
>  struct rte_event_vector {
> -       uint64_t nb_elem : 16;
> +       uint16_t nb_elem;
>         /**< Number of elements in this event vector. */
> -       uint64_t rsvd : 48;
> +       uint16_t rsvd : 15;
> +       uint16_t attr_valid : 1;
> +       /**< Indicates that the below union attributes have valid information.
> +        */
> +       union {
> +               /* Used by Rx adapter.
> +                * Indicats that all the elements in this vector belong to same

Indicats -> Indicates
to same -> to the same

> +                * port and queue pair when originating from Rx adapter, valid
> +                * only when event type is ETHDEV_VECTOR or
> +                * ETH_RX_ADAPTER_VECTOR.
> +                */
> +               struct {
> +                       uint16_t port;
> +                       /* Ethernet device port id. */
> +                       uint16_t queue;
> +                       /* Ethernet device queue id. */
> +               };
> +       };
> +       /**< Union to hold common attributes of the vector array. */
>         uint64_t impl_opaque;
>         union {
>                 struct rte_mbuf *mbufs[0];
> @@ -1019,8 +1037,14 @@ struct rte_event_vector {
>   *             // Classify and handle event.
>   *     }
>   */
> +#define RTE_EVENT_TYPE_ETHDEV_VECTOR                                           \
> +       (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETHDEV)
> +/**< The event vector generated from ethdev subsystem */
>  #define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
>  /**< The event vector generated from cpu for pipelining. */
> +#define RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR                                   \
> +       (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETH_RX_ADAPTER)
> +/**< The event vector generated from eth Rx adapter. */
>
>  #define RTE_EVENT_TYPE_MAX              0x10
>  /**< Maximum number of event types */
> @@ -1165,6 +1189,8 @@ struct rte_event {
>   * @see struct rte_event_eth_rx_adapter_queue_conf::ev
>   * @see struct rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
>   */
> +#define RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR      0x8
> +/**< Adapter supports event vectorization per ethdev. */
>
>  /**
>   * Retrieve the event device's ethdev Rx adapter capabilities for the
> diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
> index a070ef56e..902df0ae3 100644
> --- a/lib/librte_eventdev/version.map
> +++ b/lib/librte_eventdev/version.map
> @@ -141,6 +141,8 @@ EXPERIMENTAL {
>
>         #added in 21.05
>         rte_event_vector_pool_create;
> +       rte_event_eth_rx_adapter_vector_limits_get;
> +       rte_event_eth_rx_adapter_queue_event_vector_config;
>  };
>
>  INTERNAL {
> --
> 2.17.1
>

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v4 4/8] eventdev: add Rx adapter event vector support
  2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 4/8] eventdev: add Rx adapter event vector support pbhagavatula
@ 2021-03-23 18:30         ` Jerin Jacob
  0 siblings, 0 replies; 153+ messages in thread
From: Jerin Jacob @ 2021-03-23 18:30 UTC (permalink / raw)
  To: Pavan Nikhilesh
  Cc: Jerin Jacob, Jayatheerthan, Jay, Erik Gabriel Carrillo, Gujjar,
	Abhinandan S, McDaniel, Timothy, Hemant Agrawal, Van Haaren,
	Harry, Mattias Rönnblom, Liang Ma, dpdk-dev

On Sat, Mar 20, 2021 at 2:28 AM <pbhagavatula@marvell.com> wrote:
>
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>
> Add event vector support for event eth Rx adapter, the implementation
> creates vector flows based on port and queue identifier of the received
> mbufs.
>
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> ---

@jayatheethan,

Let me know if you have any concerns about this patch. I planning to
merge this for RC1 if there are no comments on this.

> +struct rte_event_eth_rx_adapter_vector_limits;
> +/**
> + * Get event vector limits for a given event, etherner device pair.

etherner -> ethernet

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v4 6/8] app/eventdev: add event vector mode in pipeline test
  2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
@ 2021-03-23 18:39         ` Jerin Jacob
  0 siblings, 0 replies; 153+ messages in thread
From: Jerin Jacob @ 2021-03-23 18:39 UTC (permalink / raw)
  To: Pavan Nikhilesh
  Cc: Jerin Jacob, Jayatheerthan, Jay, Erik Gabriel Carrillo, Gujjar,
	Abhinandan S, McDaniel, Timothy, Hemant Agrawal, Van Haaren,
	Harry, Mattias Rönnblom, Liang Ma, dpdk-dev

On Sat, Mar 20, 2021 at 2:28 AM <pbhagavatula@marvell.com> wrote:
>
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>
> Add event vector support in pipeline tests. By default this mode
> is disabled, it can be enabled by using the option --enable_vector.
> example:
>         dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
>         --nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
>         --wlcores=20-23  --enable_vector
>
> Addtional options to configure vector size and vector timeout are

### app/eventdev: add event vector mode in pipeline test

WARNING:TYPO_SPELLING: 'Addtional' may be misspelled - perhaps 'Additional'?
#13:
Addtional options to configure vector size and vector timeout are
^^^^^^^^^



> also implemented and can be used by specifying --vector_size and
> --vector_tmo_ns
>
> This patch also adds a new option to set the number of Rx queues
> configured per event eth rx adapter.
> example:
>         dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
>         --nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
>         --wlcores=20-23  --nb_eth_queues 4
>
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> ---

Acked-by: Jerin Jacob <jerinj@marvell.com>


> +++ b/doc/guides/tools/testeventdev.rst
> @@ -158,6 +158,26 @@ The following are the application command-line options:
>         Set max packet mbuf size. Can be used configure Rx/Tx scatter gather.
>         Only applicable for `pipeline_atq` and `pipeline_queue` tests.
>
> +* ``--nb_eth_queues``
> +
> +       Configure multiple Rx queues per each ethernet port.
> +       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
> +
> +* ``--enable_vector``
> +
> +       Enable event vector for Rx/Tx adapters.
> +       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
> +
> +* ``--vector_size``
> +
> +       Vector size to configure for the Rx adapter.
> +       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
> +
> +* ``--vector_tmo_ns``
> +
> +       Vector timeout nanoseconds to be configured for the Rx adapter.
> +       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
> +
>

Please add some of the example commands such as following in .rst file
in the relevant section.

        dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
        --nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
        --wlcores=20-23  --nb_eth_queues 4

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v4 0/8] Introduce event vectorization
  2021-03-19 20:57     ` [dpdk-dev] [PATCH v4 0/8] Introduce event vectorization pbhagavatula
                         ` (7 preceding siblings ...)
  2021-03-19 20:57       ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v4 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
@ 2021-03-23 18:44       ` Jerin Jacob
  2021-03-24  5:05       ` [dpdk-dev] [PATCH v5 " pbhagavatula
  9 siblings, 0 replies; 153+ messages in thread
From: Jerin Jacob @ 2021-03-23 18:44 UTC (permalink / raw)
  To: Pavan Nikhilesh
  Cc: Jerin Jacob, Jayatheerthan, Jay, Erik Gabriel Carrillo, Gujjar,
	Abhinandan S, McDaniel, Timothy, Hemant Agrawal, Van Haaren,
	Harry, Mattias Rönnblom, Liang Ma, dpdk-dev

On Sat, Mar 20, 2021 at 2:27 AM <pbhagavatula@marvell.com> wrote:
>
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>
> In traditional event programming model, events are identified by a
> flow-id and a uintptr_t. The flow-id uniquely identifies a given event
> and determines the order of scheduling based on schedule type, the
> uintptr_t holds a single object.
>
> Event devices also support burst mode with configurable dequeue depth,
> i.e. each dequeue call would return multiple events and each event
> might be at a different stage of the pipeline.
> Having a burst of events belonging to different stages in a dequeue
> burst is not only difficult to vectorize but also increases the scheduler
> overhead and application overhead of pipelining events further.
> Using event vectors we see a performance gain of ~628% as shown in [1].
>
> By introducing event vectorization, each event will be capable of holding
> multiple uintptr_t of the same flow thereby allowing applications
> to vectorize their pipeline and reduce the complexity of pipelining
> events across multiple stages. This also reduces the complexity of handling
> enqueue and dequeue on an event device.
>
> Since event devices are transparent to the events they are scheduling
> so the event producers such as eth_rx_adapter, crypto_adapter , etc..
> are responsible for vectorizing the buffers of the same flow into a single
> event.
>
> The series also breaks ABI in the patch [8/8] which is targetted to the
> v21.11 release.
>
> The dpdk-test-eventdev application has been updated with options to test
> multiple vector sizes and timeouts.
>
> [1]
> As for performance improvement, with a ARM Cortex-A72 equivalent processer,
> software event device (--vdev=event_sw0), single worker core, single stage
> and using one service core for Rx adapter, Tx adapter, Scheduling.
>
> Without event vectorization:
>     ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
>          --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
>          --stlist=a --wlcores=20
>     Port[0] using Rx adapter[0] configured
>     Port[0] using Tx adapter[0] Configured
>     4.728 mpps avg 4.728 mpps
>
> With event vectorization:
>     ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
>         --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
>         --stlist=a --wlcores=20 --enable_vector --nb_eth_queues 1
>         --vector_size 256
>     Port[0] using Rx adapter[0] configured
>     Port[0] using Tx adapter[0] Configured
>     34.383 mpps avg 34.383 mpps
>
> Having dedicated service cores for each Rx queues and tweaking the vector,
> dequeue burst size would further improve performance.
>
> API usage is shown below:
>
> Configuration:
>
>         struct rte_event_eth_rx_adapter_event_vector_config vec_conf;
>
>         vector_pool = rte_event_vector_pool_create("vector_pool",
>                         nb_elem, 0, vector_size, socket_id);
>
>         rte_event_eth_rx_adapter_create(id, event_id, &adptr_conf);
>         rte_event_eth_rx_adapter_queue_add(id, eth_id, -1, &queue_conf);
>         if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
>                 vec_conf.vector_sz = vector_size;
>                 vec_conf.vector_timeout_ns = vector_tmo_nsec;
>                 vec_conf.vector_mp = vector_pool;
>                 rte_event_eth_rx_adapter_queue_event_vector_config(id,
>                                 eth_id, -1, &vec_conf);
>         }
>
> Fastpath:
>
>         num = rte_event_dequeue_burst(event_id, port_id, &ev, 1, 0);
>         if (!num)
>                 continue;
>
>         if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
>                 switch (ev.event_type) {
>                 case RTE_EVENT_TYPE_ETHDEV_VECTOR:
>                 case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
>                         struct rte_mbuf **mbufs;
>
>                         mbufs = ev.vector_ev->mbufs;
>                         for (i = 0; i < ev.vector_ev->nb_elem; i++)
>                                 //Process mbufs.
>                         break;
>                 case ...
>                 }
>         }
>         ...
>
> v4 Changes:
> - Fix missing event vector structure in event structure.(Jay)
>
> v3 Changes:
> - Fix unintended formatting changes.
>
> v2 Changes:
> - Multiple gramatical and style fixes.(Jerin)
> - Add parameter to define vector size in power of 2. (Jerin)
> - Redo patch series w/o breaking ABI till the last patch.(David)
> - Add deprication notice to announce ABI break in 21.11.(David)
> - Add vector limits validation to app/test-eventdev.
>
> Pavan Nikhilesh (8):
>   eventdev: introduce event vector capability
>   eventdev: introduce event vector Rx capability
>   eventdev: introduce event vector Tx capability
>   eventdev: add Rx adapter event vector support
>   eventdev: add Tx adapter event vector support
>   app/eventdev: add event vector mode in pipeline test
>   doc: announce event Rx adapter config changes
>   eventdev: simplify Rx adapter event vector config
>
>  app/test-eventdev/evt_common.h                |   4 +
>  app/test-eventdev/evt_options.c               |  52 +++
>  app/test-eventdev/evt_options.h               |   4 +
>  app/test-eventdev/test_pipeline_atq.c         | 310 +++++++++++++++--
>  app/test-eventdev/test_pipeline_common.c      | 105 +++++-
>  app/test-eventdev/test_pipeline_common.h      |  18 +
>  app/test-eventdev/test_pipeline_queue.c       | 320 ++++++++++++++++--
>  .../prog_guide/event_ethernet_rx_adapter.rst  |  38 +++
>  .../prog_guide/event_ethernet_tx_adapter.rst  |  12 +
>  doc/guides/prog_guide/eventdev.rst            |  36 +-
>  doc/guides/rel_notes/deprecation.rst          |   9 +
>  doc/guides/tools/testeventdev.rst             |  28 ++
>  lib/librte_eventdev/eventdev_pmd.h            |  31 +-
>  .../rte_event_eth_rx_adapter.c                | 305 ++++++++++++++++-
>  .../rte_event_eth_rx_adapter.h                |  68 ++++
>  .../rte_event_eth_tx_adapter.c                |  66 +++-
>  lib/librte_eventdev/rte_eventdev.c            |  11 +-
>  lib/librte_eventdev/rte_eventdev.h            | 144 +++++++-
>  lib/librte_eventdev/version.map               |   4 +
>  19 files changed, 1479 insertions(+), 86 deletions(-)

Please update release notes(doc/guides/rel_notes/release_21_05.rst)
for this feature.

If there are no more comments on this series from others. IMO, Good to
merge the next series for RC1.

>
> --
> 2.17.1
>

^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v5 0/8] Introduce event vectorization
  2021-03-19 20:57     ` [dpdk-dev] [PATCH v4 0/8] Introduce event vectorization pbhagavatula
                         ` (8 preceding siblings ...)
  2021-03-23 18:44       ` [dpdk-dev] [PATCH v4 0/8] Introduce event vectorization Jerin Jacob
@ 2021-03-24  5:05       ` pbhagavatula
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 1/8] eventdev: introduce event vector capability pbhagavatula
                           ` (9 more replies)
  9 siblings, 10 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-24  5:05 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

In traditional event programming model, events are identified by a
flow-id and a uintptr_t. The flow-id uniquely identifies a given event
and determines the order of scheduling based on schedule type, the
uintptr_t holds a single object.

Event devices also support burst mode with configurable dequeue depth,
i.e. each dequeue call would return multiple events and each event
might be at a different stage of the pipeline.
Having a burst of events belonging to different stages in a dequeue
burst is not only difficult to vectorize but also increases the scheduler
overhead and application overhead of pipelining events further.
Using event vectors we see a performance gain of ~628% as shown in [1].

By introducing event vectorization, each event will be capable of holding
multiple uintptr_t of the same flow thereby allowing applications
to vectorize their pipeline and reduce the complexity of pipelining
events across multiple stages. This also reduces the complexity of handling
enqueue and dequeue on an event device.

Since event devices are transparent to the events they are scheduling
so the event producers such as eth_rx_adapter, crypto_adapter , etc..
are responsible for vectorizing the buffers of the same flow into a single
event.

The series also breaks ABI in the patch [8/8] which is targetted to the
v21.11 release.

The dpdk-test-eventdev application has been updated with options to test
multiple vector sizes and timeouts.

[1]
As for performance improvement, with a ARM Cortex-A72 equivalent processer,
software event device (--vdev=event_sw0), single worker core, single stage
and using one service core for Rx adapter, Tx adapter, Scheduling.

Without event vectorization:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
         --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
         --stlist=a --wlcores=20
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    4.728 mpps avg 4.728 mpps

With event vectorization:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
        --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
        --stlist=a --wlcores=20 --enable_vector --nb_eth_queues 1
        --vector_size 256
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    34.383 mpps avg 34.383 mpps

Having dedicated service cores for each Rx queues and tweaking the vector,
dequeue burst size would further improve performance.

API usage is shown below:

Configuration:

	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

	vector_pool = rte_event_vector_pool_create("vector_pool",
			nb_elem, 0, vector_size, socket_id);

	rte_event_eth_rx_adapter_create(id, event_id, &adptr_conf);
	rte_event_eth_rx_adapter_queue_add(id, eth_id, -1, &queue_conf);
	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
		vec_conf.vector_sz = vector_size;
		vec_conf.vector_timeout_ns = vector_tmo_nsec;
		vec_conf.vector_mp = vector_pool;
		rte_event_eth_rx_adapter_queue_event_vector_config(id,
				eth_id, -1, &vec_conf);
	}

Fastpath:

	num = rte_event_dequeue_burst(event_id, port_id, &ev, 1, 0);
	if (!num)
		continue;

	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
		switch (ev.event_type) {
		case RTE_EVENT_TYPE_ETHDEV_VECTOR:
		case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
			struct rte_mbuf **mbufs;

			mbufs = ev.vector_ev->mbufs;
			for (i = 0; i < ev.vector_ev->nb_elem; i++)
				//Process mbufs.
			break;
		case ...
		}
	}
	...

v5 Changes:
- Make `rte_event_vector_pool_create non-inline` to ease ABI stability.(Ray)
- Move `rte_event_eth_rx_adapter_queue_event_vector_config` and
  `rte_event_eth_rx_adapter_vector_limits_get` implementation to the patch
  where they are initially defined.(Ray)
- Multiple gramatical and style fixes.(Jerin)
- Add missing release notes.(Jerin)

v4 Changes:
- Fix missing event vector structure in event structure.(Jay)

v3 Changes:
- Fix unintended formatting changes.

v2 Changes:
- Multiple gramatical and style fixes.(Jerin)
- Add parameter to define vector size in power of 2. (Jerin)
- Redo patch series w/o breaking ABI till the last patch.(David)
- Add deprication notice to announce ABI break in 21.11.(David)
- Add vector limits validation to app/test-eventdev.

Pavan Nikhilesh (8):
  eventdev: introduce event vector capability
  eventdev: introduce event vector Rx capability
  eventdev: introduce event vector Tx capability
  eventdev: add Rx adapter event vector support
  eventdev: add Tx adapter event vector support
  app/eventdev: add event vector mode in pipeline test
  doc: announce event Rx adapter config changes
  eventdev: simplify Rx adapter event vector config

 app/test-eventdev/evt_common.h                |   4 +
 app/test-eventdev/evt_options.c               |  52 +++
 app/test-eventdev/evt_options.h               |   4 +
 app/test-eventdev/test_pipeline_atq.c         | 310 +++++++++++++++--
 app/test-eventdev/test_pipeline_common.c      | 105 +++++-
 app/test-eventdev/test_pipeline_common.h      |  18 +
 app/test-eventdev/test_pipeline_queue.c       | 320 ++++++++++++++++--
 .../prog_guide/event_ethernet_rx_adapter.rst  |  38 +++
 .../prog_guide/event_ethernet_tx_adapter.rst  |  12 +
 doc/guides/prog_guide/eventdev.rst            |  36 +-
 doc/guides/rel_notes/deprecation.rst          |   9 +
 doc/guides/rel_notes/release_21_05.rst        |   8 +
 doc/guides/tools/testeventdev.rst             |  45 ++-
 lib/librte_eventdev/eventdev_pmd.h            |  31 +-
 .../rte_event_eth_rx_adapter.c                | 305 ++++++++++++++++-
 .../rte_event_eth_rx_adapter.h                |  78 +++++
 .../rte_event_eth_tx_adapter.c                |  66 +++-
 lib/librte_eventdev/rte_eventdev.c            |  53 ++-
 lib/librte_eventdev/rte_eventdev.h            | 113 ++++++-
 lib/librte_eventdev/version.map               |   4 +
 20 files changed, 1524 insertions(+), 87 deletions(-)

--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v5 1/8] eventdev: introduce event vector capability
  2021-03-24  5:05       ` [dpdk-dev] [PATCH v5 " pbhagavatula
@ 2021-03-24  5:05         ` pbhagavatula
  2021-03-24  6:48           ` Jayatheerthan, Jay
  2021-03-24  9:16           ` Kinsella, Ray
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 2/8] eventdev: introduce event vector Rx capability pbhagavatula
                           ` (8 subsequent siblings)
  9 siblings, 2 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-24  5:05 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce rte_event_vector datastructure which is capable of holding
multiple uintptr_t of the same flow thereby allowing applications
to vectorize their pipeline and reducing the complexity of pipelining
the events across multiple stages.
This approach also reduces the scheduling overhead on a event device.

Add a event vector mempool create handler to create mempools based on
the best mempool ops available on a given platform.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
---
 doc/guides/prog_guide/eventdev.rst     | 36 +++++++++++-
 doc/guides/rel_notes/release_21_05.rst |  8 +++
 lib/librte_eventdev/rte_eventdev.c     | 42 +++++++++++++
 lib/librte_eventdev/rte_eventdev.h     | 81 +++++++++++++++++++++++++-
 lib/librte_eventdev/version.map        |  3 +
 5 files changed, 167 insertions(+), 3 deletions(-)

diff --git a/doc/guides/prog_guide/eventdev.rst b/doc/guides/prog_guide/eventdev.rst
index ccde086f6..fda9c3743 100644
--- a/doc/guides/prog_guide/eventdev.rst
+++ b/doc/guides/prog_guide/eventdev.rst
@@ -63,13 +63,45 @@ the actual event being scheduled is. The payload is a union of the following:
 * ``uint64_t u64``
 * ``void *event_ptr``
 * ``struct rte_mbuf *mbuf``
+* ``struct rte_event_vector *vec``
 
-These three items in a union occupy the same 64 bits at the end of the rte_event
+These four items in a union occupy the same 64 bits at the end of the rte_event
 structure. The application can utilize the 64 bits directly by accessing the
-u64 variable, while the event_ptr and mbuf are provided as convenience
+u64 variable, while the event_ptr, mbuf, vec are provided as a convenience
 variables.  For example the mbuf pointer in the union can used to schedule a
 DPDK packet.
 
+Event Vector
+~~~~~~~~~~~~
+
+The rte_event_vector struct contains a vector of elements defined by the event
+type specified in the ``rte_event``. The event_vector structure contains the
+following data:
+
+* ``nb_elem`` - The number of elements held within the vector.
+
+Similar to ``rte_event`` the payload of event vector is also a union, allowing
+flexibility in what the actual vector is.
+
+* ``struct rte_mbuf *mbufs[0]`` - An array of mbufs.
+* ``void *ptrs[0]`` - An array of pointers.
+* ``uint64_t *u64s[0]`` - An array of uint64_t elements.
+
+The size of the event vector is related to the total number of elements it is
+configured to hold, this is achieved by making `rte_event_vector` a variable
+length structure.
+A helper function is provided to create a mempool that holds event vector, which
+takes name of the pool, total number of required ``rte_event_vector``,
+cache size, number of elements in each ``rte_event_vector`` and socket id.
+
+.. code-block:: c
+
+        rte_event_vector_pool_create("vector_pool", nb_event_vectors, cache_sz,
+                                     nb_elements_per_vector, socket_id);
+
+The function ``rte_event_vector_pool_create`` creates mempool with the best
+platform mempool ops.
+
 Queues
 ~~~~~~
 
diff --git a/doc/guides/rel_notes/release_21_05.rst b/doc/guides/rel_notes/release_21_05.rst
index 8e686cc62..358623f2f 100644
--- a/doc/guides/rel_notes/release_21_05.rst
+++ b/doc/guides/rel_notes/release_21_05.rst
@@ -101,6 +101,14 @@ New Features
   * Added command to display Rx queue used descriptor count.
     ``show port (port_id) rxq (queue_id) desc used count``
 
+* **Add Event device vector capability.**
+
+  * Added ``rte_event_vector`` data structure which is capable of holding
+    multiple ``uintptr_t`` of the same flow thereby allowing applications
+    to vectorize their pipelines and also reduce the complexity of pipelining
+    the events across multiple stages.
+  * This also reduces the scheduling overhead on a event device.
+
 
 Removed Items
 -------------
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index b57363f80..f95edc075 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -1266,6 +1266,48 @@ int rte_event_dev_selftest(uint8_t dev_id)
 	return -ENOTSUP;
 }
 
+struct rte_mempool *
+rte_event_vector_pool_create(const char *name, unsigned int n,
+			     unsigned int cache_size, uint16_t nb_elem,
+			     int socket_id)
+{
+	const char *mp_ops_name;
+	struct rte_mempool *mp;
+	unsigned int elt_sz;
+	int ret;
+
+	if (!nb_elem) {
+		RTE_LOG(ERR, EVENTDEV,
+			"Invalid number of elements=%d requested\n", nb_elem);
+		rte_errno = -EINVAL;
+		return NULL;
+	}
+
+	elt_sz =
+		sizeof(struct rte_event_vector) + (nb_elem * sizeof(uintptr_t));
+	mp = rte_mempool_create_empty(name, n, elt_sz, cache_size, 0, socket_id,
+				      0);
+	if (mp == NULL)
+		return NULL;
+
+	mp_ops_name = rte_mbuf_best_mempool_ops();
+	ret = rte_mempool_set_ops_byname(mp, mp_ops_name, NULL);
+	if (ret != 0) {
+		RTE_LOG(ERR, EVENTDEV, "error setting mempool handler\n");
+		goto err;
+	}
+
+	ret = rte_mempool_populate_default(mp);
+	if (ret < 0)
+		goto err;
+
+	return mp;
+err:
+	rte_mempool_free(mp);
+	rte_errno = -ret;
+	return NULL;
+}
+
 int
 rte_event_dev_start(uint8_t dev_id)
 {
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index ce1fc2ce0..aa4dd3959 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -212,8 +212,10 @@ extern "C" {
 
 #include <rte_common.h>
 #include <rte_config.h>
-#include <rte_memory.h>
 #include <rte_errno.h>
+#include <rte_mbuf_pool_ops.h>
+#include <rte_memory.h>
+#include <rte_mempool.h>
 
 #include "rte_eventdev_trace_fp.h"
 
@@ -913,6 +915,31 @@ rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
 int
 rte_event_dev_close(uint8_t dev_id);
 
+/**
+ * Event vector structure.
+ */
+struct rte_event_vector {
+	uint64_t nb_elem : 16;
+	/**< Number of elements in this event vector. */
+	uint64_t rsvd : 48;
+	/**< Reserved for future use */
+	uint64_t impl_opaque;
+	/**< Implementation specific opaque value.
+	 * An implementation may use this field to hold implementation specific
+	 * value to share between dequeue and enqueue operation.
+	 * The application should not modify this field.
+	 */
+	union {
+		struct rte_mbuf *mbufs[0];
+		void *ptrs[0];
+		uint64_t *u64s[0];
+	} __rte_aligned(16);
+	/**< Start of the vector array union. Depending upon the event type the
+	 * vector array can be an array of mbufs or pointers or opaque u64
+	 * values.
+	 */
+};
+
 /* Scheduler type definitions */
 #define RTE_SCHED_TYPE_ORDERED          0
 /**< Ordered scheduling
@@ -986,6 +1013,21 @@ rte_event_dev_close(uint8_t dev_id);
  */
 #define RTE_EVENT_TYPE_ETH_RX_ADAPTER   0x4
 /**< The event generated from event eth Rx adapter */
+#define RTE_EVENT_TYPE_VECTOR           0x8
+/**< Indicates that event is a vector.
+ * All vector event types should be a logical OR of EVENT_TYPE_VECTOR.
+ * This simplifies the pipeline design as one can split processing the events
+ * between vector events and normal event across event types.
+ * Example:
+ *	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
+ *		// Classify and handle vector event.
+ *	} else {
+ *		// Classify and handle event.
+ *	}
+ */
+#define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
+/**< The event vector generated from cpu for pipelining. */
+
 #define RTE_EVENT_TYPE_MAX              0x10
 /**< Maximum number of event types */
 
@@ -1108,6 +1150,8 @@ struct rte_event {
 		/**< Opaque event pointer */
 		struct rte_mbuf *mbuf;
 		/**< mbuf pointer if dequeued event is associated with mbuf */
+		struct rte_event_vector *vec;
+		/**< Event vector pointer. */
 	};
 };
 
@@ -2023,6 +2067,41 @@ rte_event_dev_xstats_reset(uint8_t dev_id,
  */
 int rte_event_dev_selftest(uint8_t dev_id);
 
+/**
+ * Get the memory required per event vector based on the number of elements per
+ * vector.
+ * This should be used to create the mempool that holds the event vectors.
+ *
+ * @param name
+ *   The name of the vector pool.
+ * @param n
+ *   The number of elements in the mbuf pool.
+ * @param cache_size
+ *   Size of the per-core object cache. See rte_mempool_create() for
+ *   details.
+ * @param nb_elem
+ *   The number of elements then a single event vector should be able to hold.
+ * @param socket_id
+ *   The socket identifier where the memory should be allocated. The
+ *   value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
+ *   reserved zone
+ *
+ * @return
+ *   The pointer to the newly allocated mempool, on success. NULL on error
+ *   with rte_errno set appropriately. Possible rte_errno values include:
+ *    - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
+ *    - E_RTE_SECONDARY - function was called from a secondary process instance
+ *    - EINVAL - cache size provided is too large, or priv_size is not aligned.
+ *    - ENOSPC - the maximum number of memzones has already been allocated
+ *    - EEXIST - a memzone with the same name already exists
+ *    - ENOMEM - no appropriate memory area found in which to create memzone
+ */
+__rte_experimental
+struct rte_mempool *
+rte_event_vector_pool_create(const char *name, unsigned int n,
+			     unsigned int cache_size, uint16_t nb_elem,
+			     int socket_id);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index 3e5c09cfd..a070ef56e 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -138,6 +138,9 @@ EXPERIMENTAL {
 	__rte_eventdev_trace_port_setup;
 	# added in 20.11
 	rte_event_pmd_pci_probe_named;
+
+	#added in 21.05
+	rte_event_vector_pool_create;
 };
 
 INTERNAL {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v5 2/8] eventdev: introduce event vector Rx capability
  2021-03-24  5:05       ` [dpdk-dev] [PATCH v5 " pbhagavatula
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 1/8] eventdev: introduce event vector capability pbhagavatula
@ 2021-03-24  5:05         ` pbhagavatula
  2021-03-24  9:15           ` Kinsella, Ray
  2021-03-25  8:15           ` Jayatheerthan, Jay
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 3/8] eventdev: introduce event vector Tx capability pbhagavatula
                           ` (7 subsequent siblings)
  9 siblings, 2 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-24  5:05 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce event ethernet Rx adapter event vector capability.

If an event eth Rx adapter has the capability of
RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR then a given Rx queue
can be configured to enable event vectorization by passing the
flag RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR to
rte_event_eth_rx_adapter_queue_conf::rx_queue_flags while configuring
Rx adapter through rte_event_eth_rx_adapter_queue_add().

The max vector size, vector timeout define the vector size and
mempool used for allocating vector event are configured through
rte_event_eth_rx_adapter_queue_add. The element size of the element
in the vector pool should be equal to
    sizeof(struct rte_event_vector) + (vector_sz * sizeof(uintptr_t))

Application can use `rte_event_vector_pool_create` to create the
vector mempool used for
rte_event_eth_rx_adapter_queue_conf::vector_mp.

The Rx adapter would be responsible for vectorizing the mbufs
based on the flow, the vector limits configured by the application
and add the vector event of mbufs to the event queue set via
rte_event_eth_rx_adapter_queue_conf::ev::queue_id.
It should also mark rte_event_vector::union_valid and fill
rte_event_vector::port, rte_event_vector::queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
---
 .../prog_guide/event_ethernet_rx_adapter.rst  |  38 ++++++
 lib/librte_eventdev/eventdev_pmd.h            |  53 ++++++++
 .../rte_event_eth_rx_adapter.c                | 114 ++++++++++++++++++
 .../rte_event_eth_rx_adapter.h                | 105 ++++++++++++++++
 lib/librte_eventdev/rte_eventdev.h            |  30 ++++-
 lib/librte_eventdev/version.map               |   2 +
 6 files changed, 340 insertions(+), 2 deletions(-)

diff --git a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
index cb44ce0e4..5eefef355 100644
--- a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
@@ -186,3 +186,41 @@ the event buffer fill level is low. The
 ``rte_event_eth_rx_adapter_cb_register()`` function allow the application
 to register a callback that selects which packets to enqueue to the event
 device.
+
+Rx event vectorization
+~~~~~~~~~~~~~~~~~~~~~~
+
+The event devices, ethernet device pairs which support the capability
+``RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR`` can aggregate packets based on
+flow characteristics and generate a ``rte_event`` containing ``rte_event_vector``
+whose event type is either ``RTE_EVENT_TYPE_ETHDEV_VECTOR`` or
+``RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR``.
+The aggregation size and timeout are configurable at a queue level and the
+maximum, minimum vector sizes and timeouts vary based on the device capability
+and can be queried using ``rte_event_eth_rx_adapter_vector_limits_get``.
+The Rx adapter additionally might include useful data such as ethernet device
+port and queue identifier in the ``rte_event_vector::port`` and
+``rte_event_vector::queue`` and mark ``rte_event_vector::attr_valid`` as true.
+
+A loop processing ``rte_event_vector`` containing mbufs is shown below.
+
+.. code-block:: c
+
+        event = rte_event_dequeue_burst(event_dev, event_port, &event,
+                                        1, 0);
+        if (!event)
+                continue;
+
+        switch (ev.event_type) {
+        case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
+        case RTE_EVENT_TYPE_ETHDEV_VECTOR:
+                struct rte_mbufs **mbufs;
+
+                mbufs = (struct rte_mbufs **)ev[i].vec->mbufs;
+                for (i = 0; i < ev.vec->nb_elem; i++) {
+                        /* Process each mbuf. */
+                }
+        break;
+        case ...
+        ...
+        }
diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 7eb9a7739..9297f1433 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -645,6 +645,53 @@ typedef int (*eventdev_eth_rx_adapter_stats_reset)
  */
 typedef int (*eventdev_selftest)(void);
 
+struct rte_event_eth_rx_adapter_vector_limits;
+/**
+ * Get event vector limits for a given event, ethernet device pair.
+ *
+ * @param dev
+ *   Event device pointer
+ *
+ * @param eth_dev
+ *   Ethernet device pointer
+ *
+ * @param[out] limits
+ *   Pointer to the limits structure to be filled.
+ *
+ * @return
+ *   - 0: Success.
+ *   - <0: Error code returned by the driver function.
+ */
+typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
+	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
+	struct rte_event_eth_rx_adapter_vector_limits *limits);
+
+struct rte_event_eth_rx_adapter_event_vector_config;
+/**
+ * Enable event vector on an given Rx queue of a ethernet devices belonging to
+ * the Rx adapter.
+ *
+ * @param dev
+ *   Event device pointer
+ *
+ * @param eth_dev
+ *   Ethernet device pointer
+ *
+ * @param rx_queue_id
+ *   The Rx queue identifier
+ *
+ * @param config
+ *   Pointer to the event vector configuration structure.
+ *
+ * @return
+ *   - 0: Success.
+ *   - <0: Error code returned by the driver function.
+ */
+typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
+	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
+	int32_t rx_queue_id,
+	const struct rte_event_eth_rx_adapter_event_vector_config *config);
+
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;
 
@@ -1067,6 +1114,12 @@ struct rte_eventdev_ops {
 	/**< Get ethernet Rx stats */
 	eventdev_eth_rx_adapter_stats_reset eth_rx_adapter_stats_reset;
 	/**< Reset ethernet Rx stats */
+	eventdev_eth_rx_adapter_vector_limits_get_t
+		eth_rx_adapter_vector_limits_get;
+	/**< Get event vector limits for the Rx adapter */
+	eventdev_eth_rx_adapter_event_vector_config_t
+		eth_rx_adapter_event_vector_config;
+	/**< Configure Rx adapter with event vector */
 
 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index d8c635e99..ac8ba5bf0 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -2263,6 +2263,120 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }
 
+int
+rte_event_eth_rx_adapter_queue_event_vector_config(
+	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
+	struct rte_event_eth_rx_adapter_event_vector_config *config)
+{
+	struct rte_event_eth_rx_adapter_vector_limits limits;
+	struct rte_event_eth_rx_adapter *rx_adapter;
+	struct rte_eventdev *dev;
+	uint32_t cap;
+	int ret;
+
+	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
+
+	rx_adapter = rxa_id_to_adapter(id);
+	if ((rx_adapter == NULL) || (config == NULL))
+		return -EINVAL;
+
+	dev = &rte_eventdevs[rx_adapter->eventdev_id];
+	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
+						eth_dev_id, &cap);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
+				 "eth port %" PRIu16,
+				 id, eth_dev_id);
+		return ret;
+	}
+
+	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
+		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
+	ret = rte_event_eth_rx_adapter_vector_limits_get(
+		rx_adapter->eventdev_id, eth_dev_id, &limits);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
+				 "eth port %" PRIu16,
+				 rx_adapter->eventdev_id, eth_dev_id);
+		return ret;
+	}
+
+	if (config->vector_sz < limits.min_sz ||
+	    config->vector_sz > limits.max_sz ||
+	    config->vector_timeout_ns < limits.min_timeout_ns ||
+	    config->vector_timeout_ns > limits.max_timeout_ns ||
+	    config->vector_mp == NULL) {
+		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+	if (config->vector_mp->elt_size <
+	    (sizeof(struct rte_event_vector) +
+	     (sizeof(uintptr_t) * config->vector_sz))) {
+		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
+	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
+		RTE_FUNC_PTR_OR_ERR_RET(
+			*dev->dev_ops->eth_rx_adapter_event_vector_config,
+			-ENOTSUP);
+		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
+			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
+	} else {
+		ret = -ENOTSUP;
+	}
+
+	return ret;
+}
+
+int
+rte_event_eth_rx_adapter_vector_limits_get(
+	uint8_t dev_id, uint16_t eth_port_id,
+	struct rte_event_eth_rx_adapter_vector_limits *limits)
+{
+	struct rte_eventdev *dev;
+	uint32_t cap;
+	int ret;
+
+	RTE_EVENTDEV_VALID_DEVID_OR_ERR_RET(dev_id, -EINVAL);
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_port_id, -EINVAL);
+
+	if (limits == NULL)
+		return -EINVAL;
+
+	dev = &rte_eventdevs[dev_id];
+
+	ret = rte_event_eth_rx_adapter_caps_get(dev_id, eth_port_id, &cap);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
+				 "eth port %" PRIu16,
+				 dev_id, eth_port_id);
+		return ret;
+	}
+
+	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
+		RTE_FUNC_PTR_OR_ERR_RET(
+			*dev->dev_ops->eth_rx_adapter_vector_limits_get,
+			-ENOTSUP);
+		ret = dev->dev_ops->eth_rx_adapter_vector_limits_get(
+			dev, &rte_eth_devices[eth_port_id], limits);
+	} else {
+		ret = -ENOTSUP;
+	}
+
+	return ret;
+}
+
 int
 rte_event_eth_rx_adapter_start(uint8_t id)
 {
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
index 21bb1e54c..7407cde00 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
@@ -92,6 +92,10 @@ extern "C" {
 /**< This flag indicates the flow identifier is valid
  * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
  */
+#define RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR	0x2
+/**< This flag indicates that mbufs arriving on the queue need to be vectorized
+ * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+ */
 
 /**
  * Adapter configuration structure that the adapter configuration callback
@@ -169,6 +173,36 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 */
 };
 
+struct rte_event_eth_rx_adapter_event_vector_config {
+	uint16_t vector_sz;
+	/**<
+	 * Indicates the maximum number for mbufs to combine and form a vector.
+	 * Should be within
+	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_sz
+	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_sz
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+	 */
+	uint64_t vector_timeout_ns;
+	/**<
+	 * Indicates the maximum number of nanoseconds to wait for receiving
+	 * mbufs. Should be within vectorization limits of the
+	 * adapter
+	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_ns
+	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_ns
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+	 */
+	struct rte_mempool *vector_mp;
+	/**<
+	 * Indicates the mempool that should be used for allocating
+	 * rte_event_vector container.
+	 * Should be created by using `rte_event_vector_pool_create`.
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags.
+	 */
+};
+
 /**
  * A structure used to retrieve statistics for an eth rx adapter instance.
  */
@@ -199,6 +233,32 @@ struct rte_event_eth_rx_adapter_stats {
 	/**< Received packet count for interrupt mode Rx queues */
 };
 
+/**
+ * A structure used to retrieve eth rx adapter vector limits.
+ */
+struct rte_event_eth_rx_adapter_vector_limits {
+	uint16_t min_sz;
+	/**< Minimum vector limit configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_sz
+	 */
+	uint16_t max_sz;
+	/**< Maximum vector limit configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_sz
+	 */
+	uint8_t log2_sz;
+	/**< True if the size configured should be in log2.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_sz
+	 */
+	uint64_t min_timeout_ns;
+	/**< Minimum vector timeout configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_timeout_ns
+	 */
+	uint64_t max_timeout_ns;
+	/**< Maximum vector timeout configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_timeout_ns
+	 */
+};
+
 /**
  *
  * Callback function invoked by the SW adapter before it continues
@@ -467,6 +527,51 @@ int rte_event_eth_rx_adapter_cb_register(uint8_t id, uint16_t eth_dev_id,
 					 rte_event_eth_rx_adapter_cb_fn cb_fn,
 					 void *cb_arg);
 
+/**
+ * Retrieve vector limits for a given event dev and eth dev pair.
+ * @see rte_event_eth_rx_adapter_vector_limits
+ *
+ * @param dev_id
+ *  Event device identifier.
+ * @param eth_port_id
+ *  Port identifier of the ethernet device.
+ * @param [out] limits
+ *  A pointer to rte_event_eth_rx_adapter_vector_limits structure that has to
+ * be filled.
+ *
+ * @return
+ *  - 0: Success.
+ *  - <0: Error code on failure.
+ */
+__rte_experimental
+int rte_event_eth_rx_adapter_vector_limits_get(
+	uint8_t dev_id, uint16_t eth_port_id,
+	struct rte_event_eth_rx_adapter_vector_limits *limits);
+
+/**
+ * Configure event vectorization for a given ethernet device queue, that has
+ * been added to a event eth Rx adapter.
+ *
+ * @param id
+ *  The identifier of the ethernet Rx event adapter.
+ *
+ * @param eth_dev_id
+ *  The identifier of the ethernet device.
+ *
+ * @param rx_queue_id
+ *  Ethernet device receive queue index.
+ *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
+ *  are configured with event vectorization.
+ *
+ * @return
+ *  - 0: Success, Receive queue configured correctly.
+ *  - <0: Error code on failure.
+ */
+__rte_experimental
+int rte_event_eth_rx_adapter_queue_event_vector_config(
+	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
+	struct rte_event_eth_rx_adapter_event_vector_config *config);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index aa4dd3959..678338247 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -919,10 +919,28 @@ rte_event_dev_close(uint8_t dev_id);
  * Event vector structure.
  */
 struct rte_event_vector {
-	uint64_t nb_elem : 16;
+	uint16_t nb_elem;
 	/**< Number of elements in this event vector. */
-	uint64_t rsvd : 48;
+	uint16_t rsvd : 15;
 	/**< Reserved for future use */
+	uint16_t attr_valid : 1;
+	/**< Indicates that the below union attributes have valid information.
+	 */
+	union {
+		/* Used by Rx adapter.
+		 * Indicates that all the elements in this vector belong to the
+		 * same port and queue pair when originating from Rx adapter,
+		 * valid only when event type is ETHDEV_VECTOR or
+		 * ETH_RX_ADAPTER_VECTOR.
+		 */
+		struct {
+			uint16_t port;
+			/* Ethernet device port id. */
+			uint16_t queue;
+			/* Ethernet device queue id. */
+		};
+	};
+	/**< Union to hold common attributes of the vector array. */
 	uint64_t impl_opaque;
 	/**< Implementation specific opaque value.
 	 * An implementation may use this field to hold implementation specific
@@ -1025,8 +1043,14 @@ struct rte_event_vector {
  *		// Classify and handle event.
  *	}
  */
+#define RTE_EVENT_TYPE_ETHDEV_VECTOR                                           \
+	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETHDEV)
+/**< The event vector generated from ethdev subsystem */
 #define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
 /**< The event vector generated from cpu for pipelining. */
+#define RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR                                   \
+	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETH_RX_ADAPTER)
+/**< The event vector generated from eth Rx adapter. */
 
 #define RTE_EVENT_TYPE_MAX              0x10
 /**< Maximum number of event types */
@@ -1171,6 +1195,8 @@ struct rte_event {
  * @see struct rte_event_eth_rx_adapter_queue_conf::ev
  * @see struct rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
  */
+#define RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR	0x8
+/**< Adapter supports event vectorization per ethdev. */
 
 /**
  * Retrieve the event device's ethdev Rx adapter capabilities for the
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index a070ef56e..902df0ae3 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -141,6 +141,8 @@ EXPERIMENTAL {
 
 	#added in 21.05
 	rte_event_vector_pool_create;
+	rte_event_eth_rx_adapter_vector_limits_get;
+	rte_event_eth_rx_adapter_queue_event_vector_config;
 };
 
 INTERNAL {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v5 3/8] eventdev: introduce event vector Tx capability
  2021-03-24  5:05       ` [dpdk-dev] [PATCH v5 " pbhagavatula
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 1/8] eventdev: introduce event vector capability pbhagavatula
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 2/8] eventdev: introduce event vector Rx capability pbhagavatula
@ 2021-03-24  5:05         ` pbhagavatula
  2021-03-25  8:16           ` Jayatheerthan, Jay
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 4/8] eventdev: add Rx adapter event vector support pbhagavatula
                           ` (6 subsequent siblings)
  9 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-03-24  5:05 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce event vector transmit capability for event eth
tx adapter.

The capability indicates that the Tx adapter is capable of
transmitting event vectors.
When rte_event_vector::union_valid is set, the Tx adapter should
transmit all the packets to the rte_event_vector::port using the
rte_event_vector::queue.
If rte_event_vector::union_valid is not set then the Tx adapter
should peek into each mbuf to get the destination port and queue
pair.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
---
 doc/guides/prog_guide/event_ethernet_tx_adapter.rst | 12 ++++++++++++
 lib/librte_eventdev/rte_eventdev.h                  |  8 +++++++-
 2 files changed, 19 insertions(+), 1 deletion(-)

diff --git a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
index a8c13e136..87277dcaf 100644
--- a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
@@ -164,3 +164,15 @@ The  ``rte_event_eth_tx_adapter_stats_get()`` function reports counters defined
 in struct ``rte_event_eth_tx_adapter_stats``. The counter values are the sum of
 the counts from the eventdev PMD callback if the callback is supported, and
 the counts maintained by the service function, if one exists.
+
+Tx event vectorization
+~~~~~~~~~~~~~~~~~~~~~~
+
+The event device, ethernet device paris which support the capability
+``RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR`` can process event vector of mbufs.
+Additionally, application can provide a hint to that Tx adapter that all the
+mbufs are destined to the same ethernet port and queue by setting the bit
+``rte_event_vector::attr_valid`` and filling `rte_event_vector::port`` and
+``rte_event_vector::queue``.
+If ``rte_event_vector::attr_valid`` is not set then the Tx adapter should peek
+into each mbuf and transmit them to the requested ethernet port and queue pair.
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index 678338247..67b681ef6 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -927,11 +927,13 @@ struct rte_event_vector {
 	/**< Indicates that the below union attributes have valid information.
 	 */
 	union {
-		/* Used by Rx adapter.
+		/* Used by Rx/Tx adapter.
 		 * Indicates that all the elements in this vector belong to the
 		 * same port and queue pair when originating from Rx adapter,
 		 * valid only when event type is ETHDEV_VECTOR or
 		 * ETH_RX_ADAPTER_VECTOR.
+		 * Can also be used to indicate the Tx adapter the destination
+		 * port and queue of the mbufs in the vector
 		 */
 		struct {
 			uint16_t port;
@@ -1293,6 +1295,10 @@ rte_event_crypto_adapter_caps_get(uint8_t dev_id, uint8_t cdev_id,
 #define RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT	0x1
 /**< This flag is sent when the PMD supports a packet transmit callback
  */
+#define RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR	0x2
+/**< Indicates that the Tx adapter is capable of handling event vector of
+ * mbufs.
+ */
 
 /**
  * Retrieve the event device's eth Tx adapter capabilities
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v5 4/8] eventdev: add Rx adapter event vector support
  2021-03-24  5:05       ` [dpdk-dev] [PATCH v5 " pbhagavatula
                           ` (2 preceding siblings ...)
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 3/8] eventdev: introduce event vector Tx capability pbhagavatula
@ 2021-03-24  5:05         ` pbhagavatula
  2021-03-25 10:37           ` Jayatheerthan, Jay
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 5/8] eventdev: add Tx " pbhagavatula
                           ` (5 subsequent siblings)
  9 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-03-24  5:05 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support for event eth Rx adapter, the implementation
creates vector flows based on port and queue identifier of the received
mbufs.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 lib/librte_eventdev/eventdev_pmd.h            |   7 +-
 .../rte_event_eth_rx_adapter.c                | 257 ++++++++++++++++--
 lib/librte_eventdev/rte_eventdev.c            |   6 +-
 3 files changed, 250 insertions(+), 20 deletions(-)

diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 9297f1433..0f724ac85 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -69,9 +69,10 @@ extern "C" {
 	} \
 } while (0)
 
-#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP \
-		((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) | \
-			(RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ))
+#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP                                        \
+	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |                     \
+	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) |                         \
+	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR))
 
 #define RTE_EVENT_CRYPTO_ADAPTER_SW_CAP \
 		RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index ac8ba5bf0..c71990078 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -26,6 +26,10 @@
 #define BATCH_SIZE		32
 #define BLOCK_CNT_THRESHOLD	10
 #define ETH_EVENT_BUFFER_SIZE	(4*BATCH_SIZE)
+#define MAX_VECTOR_SIZE		1024
+#define MIN_VECTOR_SIZE		4
+#define MAX_VECTOR_NS		1E9
+#define MIN_VECTOR_NS		1E5
 
 #define ETH_RX_ADAPTER_SERVICE_NAME_LEN	32
 #define ETH_RX_ADAPTER_MEM_NAME_LEN	32
@@ -59,6 +63,20 @@ struct eth_rx_poll_entry {
 	uint16_t eth_rx_qid;
 };
 
+struct eth_rx_vector_data {
+	TAILQ_ENTRY(eth_rx_vector_data) next;
+	uint16_t port;
+	uint16_t queue;
+	uint16_t max_vector_count;
+	uint64_t event;
+	uint64_t ts;
+	uint64_t vector_timeout_ticks;
+	struct rte_mempool *vector_pool;
+	struct rte_event_vector *vector_ev;
+} __rte_cache_aligned;
+
+TAILQ_HEAD(eth_rx_vector_data_list, eth_rx_vector_data);
+
 /* Instance per adapter */
 struct rte_eth_event_enqueue_buffer {
 	/* Count of events in this buffer */
@@ -92,6 +110,14 @@ struct rte_event_eth_rx_adapter {
 	uint32_t wrr_pos;
 	/* Event burst buffer */
 	struct rte_eth_event_enqueue_buffer event_enqueue_buffer;
+	/* Vector enable flag */
+	uint8_t ena_vector;
+	/* Timestamp of previous vector expiry list traversal */
+	uint64_t prev_expiry_ts;
+	/* Minimum ticks to wait before traversing expiry list */
+	uint64_t vector_tmo_ticks;
+	/* vector list */
+	struct eth_rx_vector_data_list vector_list;
 	/* Per adapter stats */
 	struct rte_event_eth_rx_adapter_stats stats;
 	/* Block count, counts up to BLOCK_CNT_THRESHOLD */
@@ -198,9 +224,11 @@ struct eth_device_info {
 struct eth_rx_queue_info {
 	int queue_enabled;	/* True if added */
 	int intr_enabled;
+	uint8_t ena_vector;
 	uint16_t wt;		/* Polling weight */
 	uint32_t flow_id_mask;	/* Set to ~0 if app provides flow id else 0 */
 	uint64_t event;
+	struct eth_rx_vector_data vector_data;
 };
 
 static struct rte_event_eth_rx_adapter **event_eth_rx_adapter;
@@ -722,6 +750,9 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
 	    &rx_adapter->event_enqueue_buffer;
 	struct rte_event_eth_rx_adapter_stats *stats = &rx_adapter->stats;
 
+	if (!buf->count)
+		return 0;
+
 	uint16_t n = rte_event_enqueue_new_burst(rx_adapter->eventdev_id,
 					rx_adapter->event_port_id,
 					buf->events,
@@ -742,6 +773,72 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
 	return n;
 }
 
+static inline uint16_t
+rxa_create_event_vector(struct rte_event_eth_rx_adapter *rx_adapter,
+			struct eth_rx_queue_info *queue_info,
+			struct rte_eth_event_enqueue_buffer *buf,
+			struct rte_mbuf **mbufs, uint16_t num)
+{
+	struct rte_event *ev = &buf->events[buf->count];
+	struct eth_rx_vector_data *vec;
+	uint16_t filled, space, sz;
+
+	filled = 0;
+	vec = &queue_info->vector_data;
+	while (num) {
+		if (vec->vector_ev == NULL) {
+			if (rte_mempool_get(vec->vector_pool,
+					    (void **)&vec->vector_ev) < 0) {
+				rte_pktmbuf_free_bulk(mbufs, num);
+				return 0;
+			}
+			vec->vector_ev->nb_elem = 0;
+			vec->vector_ev->port = vec->port;
+			vec->vector_ev->queue = vec->queue;
+			vec->vector_ev->attr_valid = true;
+			TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
+		} else if (vec->vector_ev->nb_elem == vec->max_vector_count) {
+			/* Event ready. */
+			ev->event = vec->event;
+			ev->vec = vec->vector_ev;
+			ev++;
+			filled++;
+			vec->vector_ev = NULL;
+			TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+			if (rte_mempool_get(vec->vector_pool,
+					    (void **)&vec->vector_ev) < 0) {
+				rte_pktmbuf_free_bulk(mbufs, num);
+				return 0;
+			}
+			vec->vector_ev->nb_elem = 0;
+			vec->vector_ev->port = vec->port;
+			vec->vector_ev->queue = vec->queue;
+			vec->vector_ev->attr_valid = true;
+			TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
+		}
+
+		space = vec->max_vector_count - vec->vector_ev->nb_elem;
+		sz = num > space ? space : num;
+		memcpy(vec->vector_ev->mbufs + vec->vector_ev->nb_elem, mbufs,
+		       sizeof(void *) * sz);
+		vec->vector_ev->nb_elem += sz;
+		num -= sz;
+		mbufs += sz;
+		vec->ts = rte_rdtsc();
+	}
+
+	if (vec->vector_ev->nb_elem == vec->max_vector_count) {
+		ev->event = vec->event;
+		ev->vec = vec->vector_ev;
+		ev++;
+		filled++;
+		vec->vector_ev = NULL;
+		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+	}
+
+	return filled;
+}
+
 static inline void
 rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
@@ -770,25 +867,30 @@ rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
 	rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
 	do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
 
-	for (i = 0; i < num; i++) {
-		m = mbufs[i];
-
-		rss = do_rss ?
-			rxa_do_softrss(m, rx_adapter->rss_key_be) :
-			m->hash.rss;
-		ev->event = event;
-		ev->flow_id = (rss & ~flow_id_mask) |
-				(ev->flow_id & flow_id_mask);
-		ev->mbuf = m;
-		ev++;
+	if (!eth_rx_queue_info->ena_vector) {
+		for (i = 0; i < num; i++) {
+			m = mbufs[i];
+
+			rss = do_rss ? rxa_do_softrss(m, rx_adapter->rss_key_be)
+				     : m->hash.rss;
+			ev->event = event;
+			ev->flow_id = (rss & ~flow_id_mask) |
+				      (ev->flow_id & flow_id_mask);
+			ev->mbuf = m;
+			ev++;
+		}
+	} else {
+		num = rxa_create_event_vector(rx_adapter, eth_rx_queue_info,
+					      buf, mbufs, num);
 	}
 
-	if (dev_info->cb_fn) {
+	if (num && dev_info->cb_fn) {
 
 		dropped = 0;
 		nb_cb = dev_info->cb_fn(eth_dev_id, rx_queue_id,
-					ETH_EVENT_BUFFER_SIZE, buf->count, ev,
-					num, dev_info->cb_arg, &dropped);
+					ETH_EVENT_BUFFER_SIZE, buf->count,
+					&buf->events[buf->count], num,
+					dev_info->cb_arg, &dropped);
 		if (unlikely(nb_cb > num))
 			RTE_EDEV_LOG_ERR("Rx CB returned %d (> %d) events",
 				nb_cb, num);
@@ -1124,6 +1226,30 @@ rxa_poll(struct rte_event_eth_rx_adapter *rx_adapter)
 	return nb_rx;
 }
 
+static void
+rxa_vector_expire(struct eth_rx_vector_data *vec, void *arg)
+{
+	struct rte_event_eth_rx_adapter *rx_adapter = arg;
+	struct rte_eth_event_enqueue_buffer *buf =
+		&rx_adapter->event_enqueue_buffer;
+	struct rte_event *ev;
+
+	if (buf->count)
+		rxa_flush_event_buffer(rx_adapter);
+
+	if (vec->vector_ev->nb_elem == 0)
+		return;
+	ev = &buf->events[buf->count];
+
+	/* Event ready. */
+	ev->event = vec->event;
+	ev->vec = vec->vector_ev;
+	buf->count++;
+
+	vec->vector_ev = NULL;
+	vec->ts = 0;
+}
+
 static int
 rxa_service_func(void *args)
 {
@@ -1137,6 +1263,24 @@ rxa_service_func(void *args)
 		return 0;
 	}
 
+	if (rx_adapter->ena_vector) {
+		if ((rte_rdtsc() - rx_adapter->prev_expiry_ts) >=
+		    rx_adapter->vector_tmo_ticks) {
+			struct eth_rx_vector_data *vec;
+
+			TAILQ_FOREACH(vec, &rx_adapter->vector_list, next) {
+				uint64_t elapsed_time = rte_rdtsc() - vec->ts;
+
+				if (elapsed_time >= vec->vector_timeout_ticks) {
+					rxa_vector_expire(vec, rx_adapter);
+					TAILQ_REMOVE(&rx_adapter->vector_list,
+						     vec, next);
+				}
+			}
+			rx_adapter->prev_expiry_ts = rte_rdtsc();
+		}
+	}
+
 	stats = &rx_adapter->stats;
 	stats->rx_packets += rxa_intr_ring_dequeue(rx_adapter);
 	stats->rx_packets += rxa_poll(rx_adapter);
@@ -1640,6 +1784,28 @@ rxa_update_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }
 
+static void
+rxa_set_vector_data(struct eth_rx_queue_info *queue_info, uint16_t vector_count,
+		    uint64_t vector_ns, struct rte_mempool *mp, int32_t qid,
+		    uint16_t port_id)
+{
+#define NSEC2TICK(__ns, __freq) (((__ns) * (__freq)) / 1E9)
+	struct eth_rx_vector_data *vector_data;
+	uint32_t flow_id;
+
+	vector_data = &queue_info->vector_data;
+	vector_data->max_vector_count = vector_count;
+	vector_data->port = port_id;
+	vector_data->queue = qid;
+	vector_data->vector_pool = mp;
+	vector_data->vector_timeout_ticks =
+		NSEC2TICK(vector_ns, rte_get_timer_hz());
+	vector_data->ts = 0;
+	flow_id = queue_info->event & 0xFFFFF;
+	flow_id = flow_id == 0 ? (qid & 0xFF) | (port_id & 0xFFFF) : flow_id;
+	vector_data->event = (queue_info->event & ~0xFFFFF) | flow_id;
+}
+
 static void
 rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
 	struct eth_device_info *dev_info,
@@ -1741,6 +1907,44 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }
 
+static void
+rxa_sw_event_vector_configure(
+	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
+	int rx_queue_id,
+	const struct rte_event_eth_rx_adapter_event_vector_config *config)
+{
+	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
+	struct eth_rx_queue_info *queue_info;
+	struct rte_event *qi_ev;
+
+	if (rx_queue_id == -1) {
+		uint16_t nb_rx_queues;
+		uint16_t i;
+
+		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
+		for (i = 0; i < nb_rx_queues; i++)
+			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
+						      config);
+		return;
+	}
+
+	queue_info = &dev_info->rx_queue[rx_queue_id];
+	qi_ev = (struct rte_event *)&queue_info->event;
+	queue_info->ena_vector = 1;
+	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+	rxa_set_vector_data(queue_info, config->vector_sz,
+			    config->vector_timeout_ns, config->vector_mp,
+			    rx_queue_id, dev_info->dev->data->port_id);
+	rx_adapter->ena_vector = 1;
+	rx_adapter->vector_tmo_ticks =
+		rx_adapter->vector_tmo_ticks ?
+			      RTE_MIN(config->vector_timeout_ns << 1,
+				      rx_adapter->vector_tmo_ticks) :
+			      config->vector_timeout_ns << 1;
+	rx_adapter->prev_expiry_ts = 0;
+	TAILQ_INIT(&rx_adapter->vector_list);
+}
+
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -2081,6 +2285,15 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}
 
+	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
+	    (queue_conf->rx_queue_flags &
+	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
+		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
 		(rx_queue_id != -1)) {
 		RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
@@ -2143,6 +2356,17 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	return 0;
 }
 
+static int
+rxa_sw_vector_limits(struct rte_event_eth_rx_adapter_vector_limits *limits)
+{
+	limits->max_sz = MAX_VECTOR_SIZE;
+	limits->min_sz = MIN_VECTOR_SIZE;
+	limits->max_timeout_ns = MAX_VECTOR_NS;
+	limits->min_timeout_ns = MIN_VECTOR_NS;
+
+	return 0;
+}
+
 int
 rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 				int32_t rx_queue_id)
@@ -2333,7 +2557,8 @@ rte_event_eth_rx_adapter_queue_event_vector_config(
 		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
 			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
 	} else {
-		ret = -ENOTSUP;
+		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
+					      rx_queue_id, config);
 	}
 
 	return ret;
@@ -2371,7 +2596,7 @@ rte_event_eth_rx_adapter_vector_limits_get(
 		ret = dev->dev_ops->eth_rx_adapter_vector_limits_get(
 			dev, &rte_eth_devices[eth_port_id], limits);
 	} else {
-		ret = -ENOTSUP;
+		ret = rxa_sw_vector_limits(limits);
 	}
 
 	return ret;
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index f95edc075..254a31b1f 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -122,7 +122,11 @@ rte_event_eth_rx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
 
 	if (caps == NULL)
 		return -EINVAL;
-	*caps = 0;
+
+	if (dev->dev_ops->eth_rx_adapter_caps_get == NULL)
+		*caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
+	else
+		*caps = 0;
 
 	return dev->dev_ops->eth_rx_adapter_caps_get ?
 				(*dev->dev_ops->eth_rx_adapter_caps_get)(dev,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v5 5/8] eventdev: add Tx adapter event vector support
  2021-03-24  5:05       ` [dpdk-dev] [PATCH v5 " pbhagavatula
                           ` (3 preceding siblings ...)
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 4/8] eventdev: add Rx adapter event vector support pbhagavatula
@ 2021-03-24  5:05         ` pbhagavatula
  2021-03-25 11:44           ` Jayatheerthan, Jay
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
                           ` (4 subsequent siblings)
  9 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-03-24  5:05 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support for event eth Tx adapter, the implementation
receives events from the single linked queue and based on
rte_event_vector::union_valid transmits the vector of mbufs to a given
port, queue pair.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 .../rte_event_eth_tx_adapter.c                | 66 ++++++++++++++++---
 lib/librte_eventdev/rte_eventdev.c            |  5 +-
 2 files changed, 60 insertions(+), 11 deletions(-)

diff --git a/lib/librte_eventdev/rte_event_eth_tx_adapter.c b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
index 5b4c42dcf..db260bfb6 100644
--- a/lib/librte_eventdev/rte_event_eth_tx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
@@ -510,6 +510,47 @@ txa_service_buffer_retry(struct rte_mbuf **pkts, uint16_t unsent,
 	stats->tx_dropped += unsent - sent;
 }
 
+static uint16_t
+txa_process_event_vector(struct txa_service_data *txa,
+			 struct rte_event_vector *vec)
+{
+	struct txa_service_queue_info *tqi;
+	uint16_t port, queue, nb_tx = 0;
+	struct rte_mbuf **mbufs;
+	int i;
+
+	mbufs = (struct rte_mbuf **)vec->mbufs;
+	if (vec->attr_valid) {
+		port = vec->port;
+		queue = vec->queue;
+		tqi = txa_service_queue(txa, port, queue);
+		if (unlikely(tqi == NULL || !tqi->added)) {
+			rte_pktmbuf_free_bulk(mbufs, vec->nb_elem);
+			rte_mempool_put(rte_mempool_from_obj(vec), vec);
+			return 0;
+		}
+		for (i = 0; i < vec->nb_elem; i++) {
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
+						   mbufs[i]);
+		}
+	} else {
+		for (i = 0; i < vec->nb_elem; i++) {
+			port = mbufs[i]->port;
+			queue = rte_event_eth_tx_adapter_txq_get(mbufs[i]);
+			tqi = txa_service_queue(txa, port, queue);
+			if (unlikely(tqi == NULL || !tqi->added)) {
+				rte_pktmbuf_free(mbufs[i]);
+				continue;
+			}
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
+						   mbufs[i]);
+		}
+	}
+	rte_mempool_put(rte_mempool_from_obj(vec), vec);
+
+	return nb_tx;
+}
+
 static void
 txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
 	uint32_t n)
@@ -522,22 +563,27 @@ txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
 
 	nb_tx = 0;
 	for (i = 0; i < n; i++) {
-		struct rte_mbuf *m;
 		uint16_t port;
 		uint16_t queue;
 		struct txa_service_queue_info *tqi;
 
-		m = ev[i].mbuf;
-		port = m->port;
-		queue = rte_event_eth_tx_adapter_txq_get(m);
+		if (!(ev[i].event_type & RTE_EVENT_TYPE_VECTOR)) {
+			struct rte_mbuf *m;
 
-		tqi = txa_service_queue(txa, port, queue);
-		if (unlikely(tqi == NULL || !tqi->added)) {
-			rte_pktmbuf_free(m);
-			continue;
-		}
+			m = ev[i].mbuf;
+			port = m->port;
+			queue = rte_event_eth_tx_adapter_txq_get(m);
 
-		nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
+			tqi = txa_service_queue(txa, port, queue);
+			if (unlikely(tqi == NULL || !tqi->added)) {
+				rte_pktmbuf_free(m);
+				continue;
+			}
+
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
+		} else {
+			nb_tx += txa_process_event_vector(txa, ev[i].vec);
+		}
 	}
 
 	stats->tx_packets += nb_tx;
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index 254a31b1f..ed6b5ba59 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -196,7 +196,10 @@ rte_event_eth_tx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
 	if (caps == NULL)
 		return -EINVAL;
 
-	*caps = 0;
+	if (dev->dev_ops->eth_tx_adapter_caps_get == NULL)
+		*caps = RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR;
+	else
+		*caps = 0;
 
 	return dev->dev_ops->eth_tx_adapter_caps_get ?
 			(*dev->dev_ops->eth_tx_adapter_caps_get)(dev,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v5 6/8] app/eventdev: add event vector mode in pipeline test
  2021-03-24  5:05       ` [dpdk-dev] [PATCH v5 " pbhagavatula
                           ` (4 preceding siblings ...)
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 5/8] eventdev: add Tx " pbhagavatula
@ 2021-03-24  5:05         ` pbhagavatula
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 7/8] doc: announce event Rx adapter config changes pbhagavatula
                           ` (3 subsequent siblings)
  9 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-24  5:05 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support in pipeline tests. By default this mode
is disabled, it can be enabled by using the option --enable_vector.
example:
	dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
	--nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
	--wlcores=20-23  --enable_vector

Additional options to configure vector size and vector timeout are
also implemented and can be used by specifying --vector_size and
--vector_tmo_ns

This patch also adds a new option to set the number of Rx queues
configured per event eth rx adapter.
example:
	dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
	--nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
	--wlcores=20-23  --nb_eth_queues 4

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
---
 app/test-eventdev/evt_common.h           |   4 +
 app/test-eventdev/evt_options.c          |  52 ++++
 app/test-eventdev/evt_options.h          |   4 +
 app/test-eventdev/test_pipeline_atq.c    | 310 ++++++++++++++++++++--
 app/test-eventdev/test_pipeline_common.c | 113 +++++++-
 app/test-eventdev/test_pipeline_common.h |  18 ++
 app/test-eventdev/test_pipeline_queue.c  | 320 +++++++++++++++++++++--
 doc/guides/tools/testeventdev.rst        |  45 +++-
 8 files changed, 811 insertions(+), 55 deletions(-)

diff --git a/app/test-eventdev/evt_common.h b/app/test-eventdev/evt_common.h
index a1da1cf11..0e228258e 100644
--- a/app/test-eventdev/evt_common.h
+++ b/app/test-eventdev/evt_common.h
@@ -58,16 +58,20 @@ struct evt_options {
 	uint8_t sched_type_list[EVT_MAX_STAGES];
 	uint16_t mbuf_sz;
 	uint16_t wkr_deq_dep;
+	uint16_t vector_size;
+	uint16_t eth_queues;
 	uint32_t nb_flows;
 	uint32_t tx_first;
 	uint32_t max_pkt_sz;
 	uint32_t deq_tmo_nsec;
 	uint32_t q_priority:1;
 	uint32_t fwd_latency:1;
+	uint32_t ena_vector : 1;
 	uint64_t nb_pkts;
 	uint64_t nb_timers;
 	uint64_t expiry_nsec;
 	uint64_t max_tmo_nsec;
+	uint64_t vector_tmo_nsec;
 	uint64_t timer_tick_nsec;
 	uint64_t optm_timer_tick_nsec;
 	enum evt_prod_type prod_type;
diff --git a/app/test-eventdev/evt_options.c b/app/test-eventdev/evt_options.c
index 0d04ea9f8..0d5540574 100644
--- a/app/test-eventdev/evt_options.c
+++ b/app/test-eventdev/evt_options.c
@@ -34,6 +34,9 @@ evt_options_default(struct evt_options *opt)
 	opt->max_tmo_nsec = 1E5;  /* 100000ns ~100us */
 	opt->expiry_nsec = 1E4;   /* 10000ns ~10us */
 	opt->prod_type = EVT_PROD_TYPE_SYNT;
+	opt->eth_queues = 1;
+	opt->vector_size = 64;
+	opt->vector_tmo_nsec = 100E3;
 }
 
 typedef int (*option_parser_t)(struct evt_options *opt,
@@ -257,6 +260,43 @@ evt_parse_max_pkt_sz(struct evt_options *opt, const char *arg)
 	return ret;
 }
 
+static int
+evt_parse_ena_vector(struct evt_options *opt, const char *arg __rte_unused)
+{
+	opt->ena_vector = 1;
+	return 0;
+}
+
+static int
+evt_parse_vector_size(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint16(&(opt->vector_size), arg);
+
+	return ret;
+}
+
+static int
+evt_parse_vector_tmo_ns(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint64(&(opt->vector_tmo_nsec), arg);
+
+	return ret;
+}
+
+static int
+evt_parse_eth_queues(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint16(&(opt->eth_queues), arg);
+
+	return ret;
+}
+
 static void
 usage(char *program)
 {
@@ -289,6 +329,10 @@ usage(char *program)
 		"\t--expiry_nsec      : event timer expiry ns.\n"
 		"\t--mbuf_sz          : packet mbuf size.\n"
 		"\t--max_pkt_sz       : max packet size.\n"
+		"\t--nb_eth_queues    : number of ethernet Rx queues.\n"
+		"\t--enable_vector    : enable event vectorization.\n"
+		"\t--vector_size      : Max vector size.\n"
+		"\t--vector_tmo_ns    : Max vector timeout in nanoseconds\n"
 		);
 	printf("available tests:\n");
 	evt_test_dump_names();
@@ -360,6 +404,10 @@ static struct option lgopts[] = {
 	{ EVT_EXPIRY_NSEC,         1, 0, 0 },
 	{ EVT_MBUF_SZ,             1, 0, 0 },
 	{ EVT_MAX_PKT_SZ,          1, 0, 0 },
+	{ EVT_NB_ETH_QUEUES,       1, 0, 0 },
+	{ EVT_ENA_VECTOR,          0, 0, 0 },
+	{ EVT_VECTOR_SZ,           1, 0, 0 },
+	{ EVT_VECTOR_TMO,          1, 0, 0 },
 	{ EVT_HELP,                0, 0, 0 },
 	{ NULL,                    0, 0, 0 }
 };
@@ -394,6 +442,10 @@ evt_opts_parse_long(int opt_idx, struct evt_options *opt)
 		{ EVT_EXPIRY_NSEC, evt_parse_expiry_nsec},
 		{ EVT_MBUF_SZ, evt_parse_mbuf_sz},
 		{ EVT_MAX_PKT_SZ, evt_parse_max_pkt_sz},
+		{ EVT_NB_ETH_QUEUES, evt_parse_eth_queues},
+		{ EVT_ENA_VECTOR, evt_parse_ena_vector},
+		{ EVT_VECTOR_SZ, evt_parse_vector_size},
+		{ EVT_VECTOR_TMO, evt_parse_vector_tmo_ns},
 	};
 
 	for (i = 0; i < RTE_DIM(parsermap); i++) {
diff --git a/app/test-eventdev/evt_options.h b/app/test-eventdev/evt_options.h
index 748e54fae..1cea2a3e1 100644
--- a/app/test-eventdev/evt_options.h
+++ b/app/test-eventdev/evt_options.h
@@ -42,6 +42,10 @@
 #define EVT_EXPIRY_NSEC          ("expiry_nsec")
 #define EVT_MBUF_SZ              ("mbuf_sz")
 #define EVT_MAX_PKT_SZ           ("max_pkt_sz")
+#define EVT_NB_ETH_QUEUES        ("nb_eth_queues")
+#define EVT_ENA_VECTOR           ("enable_vector")
+#define EVT_VECTOR_SZ            ("vector_size")
+#define EVT_VECTOR_TMO           ("vector_tmo_ns")
 #define EVT_HELP                 ("help")
 
 void evt_options_default(struct evt_options *opt);
diff --git a/app/test-eventdev/test_pipeline_atq.c b/app/test-eventdev/test_pipeline_atq.c
index 0872b25b5..84dd4f44e 100644
--- a/app/test-eventdev/test_pipeline_atq.c
+++ b/app/test-eventdev/test_pipeline_atq.c
@@ -15,6 +15,8 @@ pipeline_atq_nb_event_queues(struct evt_options *opt)
 	return rte_eth_dev_count_avail();
 }
 
+typedef int (*pipeline_atq_worker_t)(void *arg);
+
 static __rte_noinline int
 pipeline_atq_worker_single_stage_tx(void *arg)
 {
@@ -113,6 +115,112 @@ pipeline_atq_worker_single_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_atq_worker_single_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+		vector_sz = ev.vec->nb_elem;
+		pipeline_event_tx_vector(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = ev.vec->nb_elem;
+		ev.queue_id = tx_queue[ev.vec->port];
+		ev.vec->queue = 0;
+		pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			vector_sz += ev[i].vec->nb_elem;
+			ev[i].vec->queue = 0;
+		}
+
+		pipeline_event_tx_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			ev[i].queue_id = tx_queue[ev[i].vec->port];
+			ev[i].vec->queue = 0;
+			vector_sz += ev[i].vec->nb_elem;
+			pipeline_fwd_event_vector(&ev[i],
+						  RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
 static __rte_noinline int
 pipeline_atq_worker_multi_stage_tx(void *arg)
 {
@@ -245,6 +353,147 @@ pipeline_atq_worker_multi_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.sub_event_type % nb_stages;
+
+		if (cq_id == last_queue) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+			continue;
+		}
+
+		ev.sub_event_type++;
+		pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.sub_event_type % nb_stages;
+
+		if (cq_id == last_queue) {
+			ev.queue_id = tx_queue[ev.vec->port];
+			ev.vec->queue = 0;
+			vector_sz = ev.vec->nb_elem;
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			pipeline_event_enqueue(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.sub_event_type++;
+			pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+			pipeline_event_enqueue(dev, port, &ev);
+		}
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].sub_event_type % nb_stages;
+
+			if (cq_id == last_queue) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+				continue;
+			}
+
+			ev[i].sub_event_type++;
+			pipeline_fwd_event_vector(&ev[i],
+						  sched_type_list[cq_id]);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].sub_event_type % nb_stages;
+
+			if (cq_id == last_queue) {
+				vector_sz = ev[i].vec->nb_elem;
+				ev[i].queue_id = tx_queue[ev[i].vec->port];
+				ev[i].vec->queue = 0;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].sub_event_type++;
+				pipeline_fwd_event_vector(
+					&ev[i], sched_type_list[cq_id]);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
 static int
 worker_wrapper(void *arg)
 {
@@ -253,27 +502,36 @@ worker_wrapper(void *arg)
 	const bool burst = evt_has_burst_mode(w->dev_id);
 	const bool internal_port = w->t->internal_port;
 	const uint8_t nb_stages = opt->nb_stages;
-	RTE_SET_USED(opt);
+	/*vector/burst/internal_port*/
+	const pipeline_atq_worker_t
+	pipeline_atq_worker_single_stage[2][2][2] = {
+		[0][0][0] = pipeline_atq_worker_single_stage_fwd,
+		[0][0][1] = pipeline_atq_worker_single_stage_tx,
+		[0][1][0] = pipeline_atq_worker_single_stage_burst_fwd,
+		[0][1][1] = pipeline_atq_worker_single_stage_burst_tx,
+		[1][0][0] = pipeline_atq_worker_single_stage_fwd_vector,
+		[1][0][1] = pipeline_atq_worker_single_stage_tx_vector,
+		[1][1][0] = pipeline_atq_worker_single_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_atq_worker_single_stage_burst_tx_vector,
+	};
+	const pipeline_atq_worker_t
+	pipeline_atq_worker_multi_stage[2][2][2] = {
+		[0][0][0] = pipeline_atq_worker_multi_stage_fwd,
+		[0][0][1] = pipeline_atq_worker_multi_stage_tx,
+		[0][1][0] = pipeline_atq_worker_multi_stage_burst_fwd,
+		[0][1][1] = pipeline_atq_worker_multi_stage_burst_tx,
+		[1][0][0] = pipeline_atq_worker_multi_stage_fwd_vector,
+		[1][0][1] = pipeline_atq_worker_multi_stage_tx_vector,
+		[1][1][0] = pipeline_atq_worker_multi_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_atq_worker_multi_stage_burst_tx_vector,
+	};
 
-	if (nb_stages == 1) {
-		if (!burst && internal_port)
-			return pipeline_atq_worker_single_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_atq_worker_single_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_atq_worker_single_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_atq_worker_single_stage_burst_fwd(arg);
-	} else {
-		if (!burst && internal_port)
-			return pipeline_atq_worker_multi_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_atq_worker_multi_stage_fwd(arg);
-		if (burst && internal_port)
-			return pipeline_atq_worker_multi_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_atq_worker_multi_stage_burst_fwd(arg);
-	}
+	if (nb_stages == 1)
+		return (pipeline_atq_worker_single_stage[opt->ena_vector][burst]
+							[internal_port])(arg);
+	else
+		return (pipeline_atq_worker_multi_stage[opt->ena_vector][burst]
+						       [internal_port])(arg);
 
 	rte_panic("invalid worker\n");
 }
@@ -290,7 +548,7 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
 	int ret;
 	int nb_ports;
 	int nb_queues;
-	uint8_t queue;
+	uint8_t queue, is_prod;
 	uint8_t tx_evqueue_id[RTE_MAX_ETHPORTS];
 	uint8_t queue_arr[RTE_EVENT_MAX_QUEUES_PER_DEV];
 	uint8_t nb_worker_queues = 0;
@@ -330,15 +588,19 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
 		q_conf.event_queue_cfg = RTE_EVENT_QUEUE_CFG_ALL_TYPES;
 
 		if (!t->internal_port) {
+			is_prod = false;
 			RTE_ETH_FOREACH_DEV(prod) {
 				if (queue == tx_evqueue_id[prod]) {
 					q_conf.event_queue_cfg =
 						RTE_EVENT_QUEUE_CFG_SINGLE_LINK;
-				} else {
-					queue_arr[nb_worker_queues] = queue;
-					nb_worker_queues++;
+					is_prod = true;
+					break;
 				}
 			}
+			if (!is_prod) {
+				queue_arr[nb_worker_queues] = queue;
+				nb_worker_queues++;
+			}
 		}
 
 		ret = rte_event_queue_setup(opt->dev_id, queue, &q_conf);
diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index b47d76743..d5ef90500 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -36,6 +36,12 @@ pipeline_opt_dump(struct evt_options *opt, uint8_t nb_queues)
 	evt_dump_queue_priority(opt);
 	evt_dump_sched_type_list(opt);
 	evt_dump_producer_type(opt);
+	evt_dump("nb_eth_rx_queues", "%d", opt->eth_queues);
+	evt_dump("event_vector", "%d", opt->ena_vector);
+	if (opt->ena_vector) {
+		evt_dump("vector_size", "%d", opt->vector_size);
+		evt_dump("vector_tmo_ns", "%" PRIu64 "", opt->vector_tmo_nsec);
+	}
 }
 
 static inline uint64_t
@@ -163,7 +169,7 @@ pipeline_opt_check(struct evt_options *opt, uint64_t nb_queues)
 int
 pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 {
-	uint16_t i;
+	uint16_t i, j;
 	int ret;
 	uint8_t nb_queues = 1;
 	struct test_pipeline *t = evt_test_priv(test);
@@ -210,6 +216,16 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 		if (!(caps & RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT))
 			t->internal_port = 0;
 
+		ret = rte_event_eth_rx_adapter_caps_get(opt->dev_id, i, &caps);
+		if (ret != 0) {
+			evt_err("failed to get event tx adapter[%d] caps", i);
+			return ret;
+		}
+
+		if (!(caps & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT))
+			local_port_conf.rxmode.offloads |=
+				DEV_RX_OFFLOAD_RSS_HASH;
+
 		ret = rte_eth_dev_info_get(i, &dev_info);
 		if (ret != 0) {
 			evt_err("Error during getting device (port %u) info: %s\n",
@@ -236,19 +252,22 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 				local_port_conf.rx_adv_conf.rss_conf.rss_hf);
 		}
 
-		if (rte_eth_dev_configure(i, nb_queues, nb_queues,
-					&local_port_conf)
-				< 0) {
+		if (rte_eth_dev_configure(i, opt->eth_queues, nb_queues,
+					  &local_port_conf) < 0) {
 			evt_err("Failed to configure eth port [%d]", i);
 			return -EINVAL;
 		}
 
-		if (rte_eth_rx_queue_setup(i, 0, NB_RX_DESC,
-				rte_socket_id(), &rx_conf, t->pool) < 0) {
-			evt_err("Failed to setup eth port [%d] rx_queue: %d.",
+		for (j = 0; j < opt->eth_queues; j++) {
+			if (rte_eth_rx_queue_setup(i, j, NB_RX_DESC,
+						   rte_socket_id(), &rx_conf,
+						   t->pool) < 0) {
+				evt_err("Failed to setup eth port [%d] rx_queue: %d.",
 					i, 0);
-			return -EINVAL;
+				return -EINVAL;
+			}
 		}
+
 		if (rte_eth_tx_queue_setup(i, 0, NB_TX_DESC,
 					rte_socket_id(), NULL) < 0) {
 			evt_err("Failed to setup eth port [%d] tx_queue: %d.",
@@ -310,12 +329,27 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 {
 	int ret = 0;
 	uint16_t prod;
+	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
+	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;
 
 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
 	queue_conf.ev.sched_type = opt->sched_type_list[0];
+	if (opt->ena_vector) {
+		unsigned int nb_elem = (opt->pool_sz / opt->vector_size) << 1;
+
+		nb_elem = nb_elem ? nb_elem : 1;
+		vector_pool = rte_event_vector_pool_create(
+			"vector_pool", nb_elem, 0, opt->vector_size,
+			opt->socket_id);
+		if (vector_pool == NULL) {
+			evt_err("failed to create event vector pool");
+			return -ENOMEM;
+		}
+	}
 	RTE_ETH_FOREACH_DEV(prod) {
+		struct rte_event_eth_rx_adapter_vector_limits limits;
 		uint32_t cap;
 
 		ret = rte_event_eth_rx_adapter_caps_get(opt->dev_id,
@@ -326,6 +360,50 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 					opt->dev_id);
 			return ret;
 		}
+
+		if (opt->ena_vector) {
+			memset(&limits, 0, sizeof(limits));
+			ret = rte_event_eth_rx_adapter_vector_limits_get(
+				opt->dev_id, prod, &limits);
+			if (ret) {
+				evt_err("failed to get vector limits");
+				return ret;
+			}
+
+			if (opt->vector_size < limits.min_sz ||
+			    opt->vector_size > limits.max_sz) {
+				evt_err("Vector size [%d] not within limits max[%d] min[%d]",
+					opt->vector_size, limits.min_sz,
+					limits.max_sz);
+				return -EINVAL;
+			}
+
+			if (limits.log2_sz &&
+			    !rte_is_power_of_2(opt->vector_size)) {
+				evt_err("Vector size [%d] not power of 2",
+					opt->vector_size);
+				return -EINVAL;
+			}
+
+			if (opt->vector_tmo_nsec > limits.max_timeout_ns ||
+			    opt->vector_tmo_nsec < limits.min_timeout_ns) {
+				evt_err("Vector timeout [%" PRIu64
+					"] not within limits max[%" PRIu64
+					"] min[%" PRIu64 "]",
+					opt->vector_tmo_nsec,
+					limits.max_timeout_ns,
+					limits.min_timeout_ns);
+				return -EINVAL;
+			}
+
+			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.rx_queue_flags |=
+				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+			} else {
+				evt_err("Rx adapter doesn't support event vector");
+				return -EINVAL;
+			}
+		}
 		queue_conf.ev.queue_id = prod * stride;
 		ret = rte_event_eth_rx_adapter_create(prod, opt->dev_id,
 				&prod_conf);
@@ -340,6 +418,17 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}
 
+		if (opt->ena_vector) {
+			vec_conf.vector_sz = opt->vector_size;
+			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
+			vec_conf.vector_mp = vector_pool;
+			if (rte_event_eth_rx_adapter_queue_event_vector_config(
+				    prod, prod, -1, &vec_conf) < 0) {
+				evt_err("Failed to configure event vectorization for Rx adapter");
+				return -EINVAL;
+			}
+		}
+
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;
 
@@ -378,6 +467,14 @@ pipeline_event_tx_adapter_setup(struct evt_options *opt,
 			return ret;
 		}
 
+		if (opt->ena_vector) {
+			if (!(cap &
+			      RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR)) {
+				evt_err("Tx adapter doesn't support event vector");
+				return -EINVAL;
+			}
+		}
+
 		ret = rte_event_eth_tx_adapter_create(consm, opt->dev_id,
 				&port_conf);
 		if (ret) {
diff --git a/app/test-eventdev/test_pipeline_common.h b/app/test-eventdev/test_pipeline_common.h
index 6e73c6ab2..800a90616 100644
--- a/app/test-eventdev/test_pipeline_common.h
+++ b/app/test-eventdev/test_pipeline_common.h
@@ -101,6 +101,14 @@ pipeline_fwd_event(struct rte_event *ev, uint8_t sched)
 	ev->sched_type = sched;
 }
 
+static __rte_always_inline void
+pipeline_fwd_event_vector(struct rte_event *ev, uint8_t sched)
+{
+	ev->event_type = RTE_EVENT_TYPE_CPU_VECTOR;
+	ev->op = RTE_EVENT_OP_FORWARD;
+	ev->sched_type = sched;
+}
+
 static __rte_always_inline void
 pipeline_event_tx(const uint8_t dev, const uint8_t port,
 		struct rte_event * const ev)
@@ -110,6 +118,16 @@ pipeline_event_tx(const uint8_t dev, const uint8_t port,
 		rte_pause();
 }
 
+static __rte_always_inline void
+pipeline_event_tx_vector(const uint8_t dev, const uint8_t port,
+			 struct rte_event *const ev)
+{
+	ev->vec->queue = 0;
+
+	while (!rte_event_eth_tx_adapter_enqueue(dev, port, ev, 1, 0))
+		rte_pause();
+}
+
 static __rte_always_inline void
 pipeline_event_tx_burst(const uint8_t dev, const uint8_t port,
 		struct rte_event *ev, const uint16_t nb_rx)
diff --git a/app/test-eventdev/test_pipeline_queue.c b/app/test-eventdev/test_pipeline_queue.c
index 9a9febb19..f6cc3e358 100644
--- a/app/test-eventdev/test_pipeline_queue.c
+++ b/app/test-eventdev/test_pipeline_queue.c
@@ -15,6 +15,8 @@ pipeline_queue_nb_event_queues(struct evt_options *opt)
 	return (eth_count * opt->nb_stages) + eth_count;
 }
 
+typedef int (*pipeline_queue_worker_t)(void *arg);
+
 static __rte_noinline int
 pipeline_queue_worker_single_stage_tx(void *arg)
 {
@@ -126,6 +128,125 @@ pipeline_queue_worker_single_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_queue_worker_single_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		if (ev.sched_type == RTE_SCHED_TYPE_ATOMIC) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.queue_id++;
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			pipeline_event_enqueue(dev, port, &ev);
+		}
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		ev.queue_id = tx_queue[ev.vec->port];
+		ev.vec->queue = 0;
+		vector_sz = ev.vec->nb_elem;
+		pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			if (ev[i].sched_type == RTE_SCHED_TYPE_ATOMIC) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].queue_id++;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			ev[i].queue_id = tx_queue[ev[i].vec->port];
+			ev[i].vec->queue = 0;
+			vector_sz += ev[i].vec->nb_elem;
+			pipeline_fwd_event_vector(&ev[i],
+						  RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
 
 static __rte_noinline int
 pipeline_queue_worker_multi_stage_tx(void *arg)
@@ -267,6 +388,151 @@ pipeline_queue_worker_multi_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.queue_id % nb_stages;
+
+		if (ev.queue_id == tx_queue[ev.vec->port]) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+			continue;
+		}
+
+		ev.queue_id++;
+		pipeline_fwd_event_vector(&ev, cq_id != last_queue
+						       ? sched_type_list[cq_id]
+						       : RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.queue_id % nb_stages;
+
+		if (cq_id == last_queue) {
+			vector_sz = ev.vec->nb_elem;
+			ev.queue_id = tx_queue[ev.vec->port];
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.queue_id++;
+			pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+		}
+
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].queue_id % nb_stages;
+
+			if (ev[i].queue_id == tx_queue[ev[i].vec->port]) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+				continue;
+			}
+
+			ev[i].queue_id++;
+			pipeline_fwd_event_vector(
+				&ev[i], cq_id != last_queue
+						? sched_type_list[cq_id]
+						: RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].queue_id % nb_stages;
+
+			if (cq_id == last_queue) {
+				ev[i].queue_id = tx_queue[ev[i].vec->port];
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].queue_id++;
+				pipeline_fwd_event_vector(
+					&ev[i], sched_type_list[cq_id]);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
 static int
 worker_wrapper(void *arg)
 {
@@ -275,29 +541,39 @@ worker_wrapper(void *arg)
 	const bool burst = evt_has_burst_mode(w->dev_id);
 	const bool internal_port = w->t->internal_port;
 	const uint8_t nb_stages = opt->nb_stages;
-	RTE_SET_USED(opt);
-
-	if (nb_stages == 1) {
-		if (!burst && internal_port)
-			return pipeline_queue_worker_single_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_queue_worker_single_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_queue_worker_single_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_queue_worker_single_stage_burst_fwd(
-					arg);
-	} else {
-		if (!burst && internal_port)
-			return pipeline_queue_worker_multi_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_queue_worker_multi_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_queue_worker_multi_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_queue_worker_multi_stage_burst_fwd(arg);
+	/*vector/burst/internal_port*/
+	const pipeline_queue_worker_t
+	pipeline_queue_worker_single_stage[2][2][2] = {
+		[0][0][0] = pipeline_queue_worker_single_stage_fwd,
+		[0][0][1] = pipeline_queue_worker_single_stage_tx,
+		[0][1][0] = pipeline_queue_worker_single_stage_burst_fwd,
+		[0][1][1] = pipeline_queue_worker_single_stage_burst_tx,
+		[1][0][0] = pipeline_queue_worker_single_stage_fwd_vector,
+		[1][0][1] = pipeline_queue_worker_single_stage_tx_vector,
+		[1][1][0] = pipeline_queue_worker_single_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_queue_worker_single_stage_burst_tx_vector,
+	};
+	const pipeline_queue_worker_t
+	pipeline_queue_worker_multi_stage[2][2][2] = {
+		[0][0][0] = pipeline_queue_worker_multi_stage_fwd,
+		[0][0][1] = pipeline_queue_worker_multi_stage_tx,
+		[0][1][0] = pipeline_queue_worker_multi_stage_burst_fwd,
+		[0][1][1] = pipeline_queue_worker_multi_stage_burst_tx,
+		[1][0][0] = pipeline_queue_worker_multi_stage_fwd_vector,
+		[1][0][1] = pipeline_queue_worker_multi_stage_tx_vector,
+		[1][1][0] = pipeline_queue_worker_multi_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_queue_worker_multi_stage_burst_tx_vector,
+	};
+
+	if (nb_stages == 1)
+		return (pipeline_queue_worker_single_stage[opt->ena_vector]
+							  [burst]
+							  [internal_port])(arg);
+	else
+		return (pipeline_queue_worker_multi_stage[opt->ena_vector]
+							 [burst]
+							 [internal_port])(arg);
 
-	}
 	rte_panic("invalid worker\n");
 }
 
diff --git a/doc/guides/tools/testeventdev.rst b/doc/guides/tools/testeventdev.rst
index ad1788a3d..f252dc2c1 100644
--- a/doc/guides/tools/testeventdev.rst
+++ b/doc/guides/tools/testeventdev.rst
@@ -158,6 +158,26 @@ The following are the application command-line options:
        Set max packet mbuf size. Can be used configure Rx/Tx scatter gather.
        Only applicable for `pipeline_atq` and `pipeline_queue` tests.
 
+* ``--nb_eth_queues``
+
+       Configure multiple Rx queues per each ethernet port.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--enable_vector``
+
+       Enable event vector for Rx/Tx adapters.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--vector_size``
+
+       Vector size to configure for the Rx adapter.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--vector_tmo_ns``
+
+       Vector timeout nanoseconds to be configured for the Rx adapter.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
 
 Eventdev Tests
 --------------
@@ -607,6 +627,10 @@ Supported application command line options are following::
         --worker_deq_depth
         --prod_type_ethdev
         --deq_tmo_nsec
+        --nb_eth_queues
+        --enable_vector
+        --vector_size
+        --vector_tmo_ns
 
 
 .. Note::
@@ -623,6 +647,13 @@ Example command to run pipeline queue test:
     sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
         --test=pipeline_queue --wlcore=1 --prod_type_ethdev --stlist=a
 
+Example command to run pipeline atq test with vector events:
+
+.. code-block:: console
+
+    sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
+        --test=pipeline_queue --wlcore=1 --prod_type_ethdev --stlist=a \
+        --enable_vector  --vector_size 512
 
 PIPELINE_ATQ Test
 ~~~~~~~~~~~~~~~~~~~
@@ -699,6 +730,10 @@ Supported application command line options are following::
         --worker_deq_depth
         --prod_type_ethdev
         --deq_tmo_nsec
+        --nb_eth_queues
+        --enable_vector
+        --vector_size
+        --vector_tmo_ns
 
 
 .. Note::
@@ -708,9 +743,17 @@ Supported application command line options are following::
 Example
 ^^^^^^^
 
-Example command to run pipeline queue test:
+Example command to run pipeline atq test:
 
 .. code-block:: console
 
     sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
         --test=pipeline_atq --wlcore=1 --prod_type_ethdev --stlist=a
+
+Example command to run pipeline atq test with vector events:
+
+.. code-block:: console
+
+    sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
+        --test=pipeline_atq --wlcore=1 --prod_type_ethdev --stlist=a \
+        --enable_vector  --vector_size 512
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v5 7/8] doc: announce event Rx adapter config changes
  2021-03-24  5:05       ` [dpdk-dev] [PATCH v5 " pbhagavatula
                           ` (5 preceding siblings ...)
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
@ 2021-03-24  5:05         ` pbhagavatula
  2021-03-24  9:16           ` Kinsella, Ray
  2021-03-24  5:05         ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v5 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
                           ` (2 subsequent siblings)
  9 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-03-24  5:05 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

The Rx adapter event vector configuration will be merged into
Rx adapter queue configuration to simplify enabling event
vectorization.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 doc/guides/rel_notes/deprecation.rst | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst
index 64629e064..ef6d1bb18 100644
--- a/doc/guides/rel_notes/deprecation.rst
+++ b/doc/guides/rel_notes/deprecation.rst
@@ -130,3 +130,12 @@ Deprecation Notices
 * cmdline: ``cmdline`` structure will be made opaque to hide platform-specific
   content. On Linux and FreeBSD, supported prior to DPDK 20.11,
   original structure will be kept until DPDK 21.11.
+
+* eventdev: The structure ``rte_event_eth_rx_adapter_queue_conf`` will be
+  extended to include ``rte_event_eth_rx_adapter_event_vector_config`` elements
+  and the function ``rte_event_eth_rx_adapter_queue_event_vector_config`` will
+  be removed in DPDK 21.11.
+
+  An application can enable event vectorization by passing the desired vector
+  values to the function ``rte_event_eth_rx_adapter_queue_add`` using
+  the structure ``rte_event_eth_rx_adapter_queue_add``.
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [dpdk-dev v21.11] [PATCH v5 8/8] eventdev: simplify Rx adapter event vector config
  2021-03-24  5:05       ` [dpdk-dev] [PATCH v5 " pbhagavatula
                           ` (6 preceding siblings ...)
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 7/8] doc: announce event Rx adapter config changes pbhagavatula
@ 2021-03-24  5:05         ` pbhagavatula
  2021-03-25 12:27           ` Jayatheerthan, Jay
  2021-03-24  5:39         ` [dpdk-dev] [PATCH v5 0/8] Introduce event vectorization Jayatheerthan, Jay
  2021-03-24 19:28         ` [dpdk-dev] [PATCH v6 " pbhagavatula
  9 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-03-24  5:05 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Include vector configuration into the structure
``rte_event_eth_rx_adapter_queue_conf`` used when configuring rest
of the Rx adapter ethernet device Rx queue parameters.
This simplifies event vector configuration as it avoids splitting
configuration per Rx queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 app/test-eventdev/test_pipeline_common.c      |  16 +-
 lib/librte_eventdev/eventdev_pmd.h            |  29 ---
 .../rte_event_eth_rx_adapter.c                | 168 ++++++------------
 .../rte_event_eth_rx_adapter.h                |  27 ---
 lib/librte_eventdev/version.map               |   1 -
 5 files changed, 57 insertions(+), 184 deletions(-)

diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index d5ef90500..76aee254b 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -331,7 +331,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 	uint16_t prod;
 	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
-	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
@@ -397,8 +396,12 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			}

 			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.vector_sz = opt->vector_size;
+				queue_conf.vector_timeout_ns =
+					opt->vector_tmo_nsec;
 				queue_conf.rx_queue_flags |=
 				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+				queue_conf.vector_mp = vector_pool;
 			} else {
 				evt_err("Rx adapter doesn't support event vector");
 				return -EINVAL;
@@ -418,17 +421,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}

-		if (opt->ena_vector) {
-			vec_conf.vector_sz = opt->vector_size;
-			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
-			vec_conf.vector_mp = vector_pool;
-			if (rte_event_eth_rx_adapter_queue_event_vector_config(
-				    prod, prod, -1, &vec_conf) < 0) {
-				evt_err("Failed to configure event vectorization for Rx adapter");
-				return -EINVAL;
-			}
-		}
-
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;

diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 0f724ac85..63b3bc4b5 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -667,32 +667,6 @@ typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
 	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-struct rte_event_eth_rx_adapter_event_vector_config;
-/**
- * Enable event vector on an given Rx queue of a ethernet devices belonging to
- * the Rx adapter.
- *
- * @param dev
- *   Event device pointer
- *
- * @param eth_dev
- *   Ethernet device pointer
- *
- * @param rx_queue_id
- *   The Rx queue identifier
- *
- * @param config
- *   Pointer to the event vector configuration structure.
- *
- * @return
- *   - 0: Success.
- *   - <0: Error code returned by the driver function.
- */
-typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
-	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
-	int32_t rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;

@@ -1118,9 +1092,6 @@ struct rte_eventdev_ops {
 	eventdev_eth_rx_adapter_vector_limits_get_t
 		eth_rx_adapter_vector_limits_get;
 	/**< Get event vector limits for the Rx adapter */
-	eventdev_eth_rx_adapter_event_vector_config_t
-		eth_rx_adapter_event_vector_config;
-	/**< Configure Rx adapter with event vector */

 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index c71990078..a1990637f 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -1882,6 +1882,25 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	} else
 		qi_ev->flow_id = 0;

+	if (conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+		queue_info->ena_vector = 1;
+		qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+		rxa_set_vector_data(queue_info, conf->vector_sz,
+				    conf->vector_timeout_ns, conf->vector_mp,
+				    rx_queue_id, dev_info->dev->data->port_id);
+		rx_adapter->ena_vector = 1;
+		rx_adapter->vector_tmo_ticks =
+			rx_adapter->vector_tmo_ticks
+				? RTE_MIN(queue_info->vector_data
+						  .vector_timeout_ticks,
+					  rx_adapter->vector_tmo_ticks)
+				: queue_info->vector_data.vector_timeout_ticks;
+		rx_adapter->vector_tmo_ticks <<= 1;
+		TAILQ_INIT(&rx_adapter->vector_list);
+		rx_adapter->prev_expiry_ts = 0;
+	}
+
 	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
 	if (rxa_polled_queue(dev_info, rx_queue_id)) {
 		rx_adapter->num_rx_polled += !pollq;
@@ -1907,44 +1926,6 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }

-static void
-rxa_sw_event_vector_configure(
-	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
-	int rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
-	struct eth_rx_queue_info *queue_info;
-	struct rte_event *qi_ev;
-
-	if (rx_queue_id == -1) {
-		uint16_t nb_rx_queues;
-		uint16_t i;
-
-		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
-		for (i = 0; i < nb_rx_queues; i++)
-			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
-						      config);
-		return;
-	}
-
-	queue_info = &dev_info->rx_queue[rx_queue_id];
-	qi_ev = (struct rte_event *)&queue_info->event;
-	queue_info->ena_vector = 1;
-	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
-	rxa_set_vector_data(queue_info, config->vector_sz,
-			    config->vector_timeout_ns, config->vector_mp,
-			    rx_queue_id, dev_info->dev->data->port_id);
-	rx_adapter->ena_vector = 1;
-	rx_adapter->vector_tmo_ticks =
-		rx_adapter->vector_tmo_ticks ?
-			      RTE_MIN(config->vector_timeout_ns << 1,
-				      rx_adapter->vector_tmo_ticks) :
-			      config->vector_timeout_ns << 1;
-	rx_adapter->prev_expiry_ts = 0;
-	TAILQ_INIT(&rx_adapter->vector_list);
-}
-
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -2258,6 +2239,7 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	struct rte_event_eth_rx_adapter *rx_adapter;
 	struct rte_eventdev *dev;
 	struct eth_device_info *dev_info;
+	struct rte_event_eth_rx_adapter_vector_limits limits;

 	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
 	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
@@ -2294,6 +2276,39 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}

+	if (queue_conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+		ret = rte_event_eth_rx_adapter_vector_limits_get(
+			rx_adapter->eventdev_id, eth_dev_id, &limits);
+		if (ret < 0) {
+			RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_sz < limits.min_sz ||
+		    queue_conf->vector_sz > limits.max_sz ||
+		    queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
+		    queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
+		    queue_conf->vector_mp == NULL) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_mp->elt_size <
+		    (sizeof(struct rte_event_vector) +
+		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+	}
+
 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
 		(rx_queue_id != -1)) {
 		RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
@@ -2487,83 +2502,6 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }

-int
-rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct rte_event_eth_rx_adapter_vector_limits limits;
-	struct rte_event_eth_rx_adapter *rx_adapter;
-	struct rte_eventdev *dev;
-	uint32_t cap;
-	int ret;
-
-	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
-	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
-
-	rx_adapter = rxa_id_to_adapter(id);
-	if ((rx_adapter == NULL) || (config == NULL))
-		return -EINVAL;
-
-	dev = &rte_eventdevs[rx_adapter->eventdev_id];
-	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
-						eth_dev_id, &cap);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
-				 "eth port %" PRIu16,
-				 id, eth_dev_id);
-		return ret;
-	}
-
-	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
-		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	ret = rte_event_eth_rx_adapter_vector_limits_get(
-		rx_adapter->eventdev_id, eth_dev_id, &limits);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
-				 "eth port %" PRIu16,
-				 rx_adapter->eventdev_id, eth_dev_id);
-		return ret;
-	}
-
-	if (config->vector_sz < limits.min_sz ||
-	    config->vector_sz > limits.max_sz ||
-	    config->vector_timeout_ns < limits.min_timeout_ns ||
-	    config->vector_timeout_ns > limits.max_timeout_ns ||
-	    config->vector_mp == NULL) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-	if (config->vector_mp->elt_size <
-	    (sizeof(struct rte_event_vector) +
-	     (sizeof(uintptr_t) * config->vector_sz))) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
-		RTE_FUNC_PTR_OR_ERR_RET(
-			*dev->dev_ops->eth_rx_adapter_event_vector_config,
-			-ENOTSUP);
-		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
-			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
-	} else {
-		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
-					      rx_queue_id, config);
-	}
-
-	return ret;
-}
-
 int
 rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
index 7407cde00..3f8b36229 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
@@ -171,9 +171,6 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
 	 * enqueued event.
 	 */
-};
-
-struct rte_event_eth_rx_adapter_event_vector_config {
 	uint16_t vector_sz;
 	/**<
 	 * Indicates the maximum number for mbufs to combine and form a vector.
@@ -548,30 +545,6 @@ int rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-/**
- * Configure event vectorization for a given ethernet device queue, that has
- * been added to a event eth Rx adapter.
- *
- * @param id
- *  The identifier of the ethernet Rx event adapter.
- *
- * @param eth_dev_id
- *  The identifier of the ethernet device.
- *
- * @param rx_queue_id
- *  Ethernet device receive queue index.
- *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
- *  are configured with event vectorization.
- *
- * @return
- *  - 0: Success, Receive queue configured correctly.
- *  - <0: Error code on failure.
- */
-__rte_experimental
-int rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index 902df0ae3..34c1c830e 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -142,7 +142,6 @@ EXPERIMENTAL {
 	#added in 21.05
 	rte_event_vector_pool_create;
 	rte_event_eth_rx_adapter_vector_limits_get;
-	rte_event_eth_rx_adapter_queue_event_vector_config;
 };

 INTERNAL {
--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v5 0/8] Introduce event vectorization
  2021-03-24  5:05       ` [dpdk-dev] [PATCH v5 " pbhagavatula
                           ` (7 preceding siblings ...)
  2021-03-24  5:05         ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v5 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
@ 2021-03-24  5:39         ` Jayatheerthan, Jay
  2021-03-24  6:44           ` Pavan Nikhilesh Bhagavatula
  2021-03-24 19:28         ` [dpdk-dev] [PATCH v6 " pbhagavatula
  9 siblings, 1 reply; 153+ messages in thread
From: Jayatheerthan, Jay @ 2021-03-24  5:39 UTC (permalink / raw)
  To: pbhagavatula, jerinj, Carrillo, Erik G, Gujjar, Abhinandan S,
	McDaniel, Timothy, hemant.agrawal, Van Haaren, Harry,
	mattias.ronnblom, Ma, Liang J
  Cc: dev

> -----Original Message-----
> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
> Sent: Wednesday, March 24, 2021 10:35 AM
> To: jerinj@marvell.com; Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Carrillo, Erik G <erik.g.carrillo@intel.com>; Gujjar, Abhinandan
> S <abhinandan.gujjar@intel.com>; McDaniel, Timothy <timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van Haaren, Harry
> <harry.van.haaren@intel.com>; mattias.ronnblom <mattias.ronnblom@ericsson.com>; Ma, Liang J <liang.j.ma@intel.com>
> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
> Subject: [dpdk-dev] [PATCH v5 0/8] Introduce event vectorization
> 
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> In traditional event programming model, events are identified by a
> flow-id and a uintptr_t. The flow-id uniquely identifies a given event
> and determines the order of scheduling based on schedule type, the
> uintptr_t holds a single object.
> 
> Event devices also support burst mode with configurable dequeue depth,
> i.e. each dequeue call would return multiple events and each event
> might be at a different stage of the pipeline.
> Having a burst of events belonging to different stages in a dequeue
> burst is not only difficult to vectorize but also increases the scheduler
> overhead and application overhead of pipelining events further.
> Using event vectors we see a performance gain of ~628% as shown in [1].
This is very impressive performance boost. Thanks so much for putting this patchset together! Just curious, was any performance measurement done for existing applications (non-vector)?
> 
> By introducing event vectorization, each event will be capable of holding
> multiple uintptr_t of the same flow thereby allowing applications
> to vectorize their pipeline and reduce the complexity of pipelining
> events across multiple stages. This also reduces the complexity of handling
> enqueue and dequeue on an event device.
> 
> Since event devices are transparent to the events they are scheduling
> so the event producers such as eth_rx_adapter, crypto_adapter , etc..
> are responsible for vectorizing the buffers of the same flow into a single
> event.
> 
> The series also breaks ABI in the patch [8/8] which is targetted to the
> v21.11 release.
> 
> The dpdk-test-eventdev application has been updated with options to test
> multiple vector sizes and timeouts.
> 
> [1]
> As for performance improvement, with a ARM Cortex-A72 equivalent processer,
> software event device (--vdev=event_sw0), single worker core, single stage
> and using one service core for Rx adapter, Tx adapter, Scheduling.
> 
> Without event vectorization:
>     ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
>          --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
>          --stlist=a --wlcores=20
>     Port[0] using Rx adapter[0] configured
>     Port[0] using Tx adapter[0] Configured
>     4.728 mpps avg 4.728 mpps
Is this number before the patchset? If so, it would help put similar number with the patchset but not using vectorization feature.
> 
> With event vectorization:
>     ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
>         --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
>         --stlist=a --wlcores=20 --enable_vector --nb_eth_queues 1
>         --vector_size 256
>     Port[0] using Rx adapter[0] configured
>     Port[0] using Tx adapter[0] Configured
>     34.383 mpps avg 34.383 mpps
> 
> Having dedicated service cores for each Rx queues and tweaking the vector,
> dequeue burst size would further improve performance.
> 
> API usage is shown below:
> 
> Configuration:
> 
> 	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;
> 
> 	vector_pool = rte_event_vector_pool_create("vector_pool",
> 			nb_elem, 0, vector_size, socket_id);
> 
> 	rte_event_eth_rx_adapter_create(id, event_id, &adptr_conf);
> 	rte_event_eth_rx_adapter_queue_add(id, eth_id, -1, &queue_conf);
> 	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
> 		vec_conf.vector_sz = vector_size;
> 		vec_conf.vector_timeout_ns = vector_tmo_nsec;
> 		vec_conf.vector_mp = vector_pool;
> 		rte_event_eth_rx_adapter_queue_event_vector_config(id,
> 				eth_id, -1, &vec_conf);
> 	}
> 
> Fastpath:
> 
> 	num = rte_event_dequeue_burst(event_id, port_id, &ev, 1, 0);
> 	if (!num)
> 		continue;
> 
> 	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
> 		switch (ev.event_type) {
> 		case RTE_EVENT_TYPE_ETHDEV_VECTOR:
> 		case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
> 			struct rte_mbuf **mbufs;
> 
> 			mbufs = ev.vector_ev->mbufs;
> 			for (i = 0; i < ev.vector_ev->nb_elem; i++)
> 				//Process mbufs.
> 			break;
> 		case ...
> 		}
> 	}
> 	...
> 
> v5 Changes:
> - Make `rte_event_vector_pool_create non-inline` to ease ABI stability.(Ray)
> - Move `rte_event_eth_rx_adapter_queue_event_vector_config` and
>   `rte_event_eth_rx_adapter_vector_limits_get` implementation to the patch
>   where they are initially defined.(Ray)
> - Multiple gramatical and style fixes.(Jerin)
> - Add missing release notes.(Jerin)
> 
> v4 Changes:
> - Fix missing event vector structure in event structure.(Jay)
> 
> v3 Changes:
> - Fix unintended formatting changes.
> 
> v2 Changes:
> - Multiple gramatical and style fixes.(Jerin)
> - Add parameter to define vector size in power of 2. (Jerin)
> - Redo patch series w/o breaking ABI till the last patch.(David)
> - Add deprication notice to announce ABI break in 21.11.(David)
> - Add vector limits validation to app/test-eventdev.
> 
> Pavan Nikhilesh (8):
>   eventdev: introduce event vector capability
>   eventdev: introduce event vector Rx capability
>   eventdev: introduce event vector Tx capability
>   eventdev: add Rx adapter event vector support
>   eventdev: add Tx adapter event vector support
>   app/eventdev: add event vector mode in pipeline test
>   doc: announce event Rx adapter config changes
>   eventdev: simplify Rx adapter event vector config
> 
>  app/test-eventdev/evt_common.h                |   4 +
>  app/test-eventdev/evt_options.c               |  52 +++
>  app/test-eventdev/evt_options.h               |   4 +
>  app/test-eventdev/test_pipeline_atq.c         | 310 +++++++++++++++--
>  app/test-eventdev/test_pipeline_common.c      | 105 +++++-
>  app/test-eventdev/test_pipeline_common.h      |  18 +
>  app/test-eventdev/test_pipeline_queue.c       | 320 ++++++++++++++++--
>  .../prog_guide/event_ethernet_rx_adapter.rst  |  38 +++
>  .../prog_guide/event_ethernet_tx_adapter.rst  |  12 +
>  doc/guides/prog_guide/eventdev.rst            |  36 +-
>  doc/guides/rel_notes/deprecation.rst          |   9 +
>  doc/guides/rel_notes/release_21_05.rst        |   8 +
>  doc/guides/tools/testeventdev.rst             |  45 ++-
>  lib/librte_eventdev/eventdev_pmd.h            |  31 +-
>  .../rte_event_eth_rx_adapter.c                | 305 ++++++++++++++++-
>  .../rte_event_eth_rx_adapter.h                |  78 +++++
>  .../rte_event_eth_tx_adapter.c                |  66 +++-
>  lib/librte_eventdev/rte_eventdev.c            |  53 ++-
>  lib/librte_eventdev/rte_eventdev.h            | 113 ++++++-
>  lib/librte_eventdev/version.map               |   4 +
>  20 files changed, 1524 insertions(+), 87 deletions(-)
> 
> --
> 2.17.1

Just a heads up. v5 patchset doesn't apply cleanly on HEAD (5f0849c1155849dfdbf950c91c52cdf9cd301f59). Although, it applies cleanly on app/eventdev: fix timeout accuracy (c33d48387dc8ccf1b432820f6e0cd4992ab486df).

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v5 0/8] Introduce event vectorization
  2021-03-24  5:39         ` [dpdk-dev] [PATCH v5 0/8] Introduce event vectorization Jayatheerthan, Jay
@ 2021-03-24  6:44           ` Pavan Nikhilesh Bhagavatula
  2021-03-24  8:10             ` Jayatheerthan, Jay
  0 siblings, 1 reply; 153+ messages in thread
From: Pavan Nikhilesh Bhagavatula @ 2021-03-24  6:44 UTC (permalink / raw)
  To: Jayatheerthan, Jay, Jerin Jacob Kollanukkaran, Carrillo, Erik G,
	Gujjar, Abhinandan S, McDaniel, Timothy, hemant.agrawal,
	Van Haaren, Harry, mattias.ronnblom, Ma, Liang J
  Cc: dev

>> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
>> Sent: Wednesday, March 24, 2021 10:35 AM
>> To: jerinj@marvell.com; Jayatheerthan, Jay
><jay.jayatheerthan@intel.com>; Carrillo, Erik G
><erik.g.carrillo@intel.com>; Gujjar, Abhinandan
>> S <abhinandan.gujjar@intel.com>; McDaniel, Timothy
><timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
>Haaren, Harry
>> <harry.van.haaren@intel.com>; mattias.ronnblom
><mattias.ronnblom@ericsson.com>; Ma, Liang J
><liang.j.ma@intel.com>
>> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
>> Subject: [dpdk-dev] [PATCH v5 0/8] Introduce event vectorization
>>
>> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>>
>> In traditional event programming model, events are identified by a
>> flow-id and a uintptr_t. The flow-id uniquely identifies a given event
>> and determines the order of scheduling based on schedule type, the
>> uintptr_t holds a single object.
>>
>> Event devices also support burst mode with configurable dequeue
>depth,
>> i.e. each dequeue call would return multiple events and each event
>> might be at a different stage of the pipeline.
>> Having a burst of events belonging to different stages in a dequeue
>> burst is not only difficult to vectorize but also increases the scheduler
>> overhead and application overhead of pipelining events further.
>> Using event vectors we see a performance gain of ~628% as shown in
>[1].
>This is very impressive performance boost. Thanks so much for putting
>this patchset together! Just curious, was any performance
>measurement done for existing applications (non-vector)?
>>
>> By introducing event vectorization, each event will be capable of
>holding
>> multiple uintptr_t of the same flow thereby allowing applications
>> to vectorize their pipeline and reduce the complexity of pipelining
>> events across multiple stages. This also reduces the complexity of
>handling
>> enqueue and dequeue on an event device.
>>
>> Since event devices are transparent to the events they are scheduling
>> so the event producers such as eth_rx_adapter, crypto_adapter , etc..
>> are responsible for vectorizing the buffers of the same flow into a
>single
>> event.
>>
>> The series also breaks ABI in the patch [8/8] which is targetted to the
>> v21.11 release.
>>
>> The dpdk-test-eventdev application has been updated with options to
>test
>> multiple vector sizes and timeouts.
>>
>> [1]
>> As for performance improvement, with a ARM Cortex-A72 equivalent
>processer,
>> software event device (--vdev=event_sw0), single worker core, single
>stage
>> and using one service core for Rx adapter, Tx adapter, Scheduling.
>>
>> Without event vectorization:
>>     ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --
>vdev="event_sw0" --
>>          --prod_type_ethdev --nb_pkts=0 --verbose 2 --
>test=pipeline_queue
>>          --stlist=a --wlcores=20
>>     Port[0] using Rx adapter[0] configured
>>     Port[0] using Tx adapter[0] Configured
>>     4.728 mpps avg 4.728 mpps
>Is this number before the patchset? If so, it would help put similar
>number with the patchset but not using vectorization feature.

I don’t remember the exact clock frequency I was using when I ran 
the above test but with equal clocks:
1. Without the patchset applied
	5.071 mpps
2. With patchset applied w/o enabling vector
	5.123 mpps
3. With patchset applied with enabling vector
	vector_sz@256 42.715 mpps
	vector_sz@512 45.335 mpps
	
>>
>> With event vectorization:
>>     ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --
>vdev="event_sw0" --
>>         --prod_type_ethdev --nb_pkts=0 --verbose 2 --
>test=pipeline_queue
>>         --stlist=a --wlcores=20 --enable_vector --nb_eth_queues 1
>>         --vector_size 256
>>     Port[0] using Rx adapter[0] configured
>>     Port[0] using Tx adapter[0] Configured
>>     34.383 mpps avg 34.383 mpps
>>
>> Having dedicated service cores for each Rx queues and tweaking the
>vector,
>> dequeue burst size would further improve performance.
>>
>> API usage is shown below:
>>
>> Configuration:
>>
>> 	struct rte_event_eth_rx_adapter_event_vector_config
>vec_conf;
>>
>> 	vector_pool = rte_event_vector_pool_create("vector_pool",
>> 			nb_elem, 0, vector_size, socket_id);
>>
>> 	rte_event_eth_rx_adapter_create(id, event_id, &adptr_conf);
>> 	rte_event_eth_rx_adapter_queue_add(id, eth_id, -1,
>&queue_conf);
>> 	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)
>{
>> 		vec_conf.vector_sz = vector_size;
>> 		vec_conf.vector_timeout_ns = vector_tmo_nsec;
>> 		vec_conf.vector_mp = vector_pool;
>>
>	rte_event_eth_rx_adapter_queue_event_vector_config(id,
>> 				eth_id, -1, &vec_conf);
>> 	}
>>
>> Fastpath:
>>
>> 	num = rte_event_dequeue_burst(event_id, port_id, &ev, 1, 0);
>> 	if (!num)
>> 		continue;
>>
>> 	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
>> 		switch (ev.event_type) {
>> 		case RTE_EVENT_TYPE_ETHDEV_VECTOR:
>> 		case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
>> 			struct rte_mbuf **mbufs;
>>
>> 			mbufs = ev.vector_ev->mbufs;
>> 			for (i = 0; i < ev.vector_ev->nb_elem; i++)
>> 				//Process mbufs.
>> 			break;
>> 		case ...
>> 		}
>> 	}
>> 	...
>>
>> v5 Changes:
>> - Make `rte_event_vector_pool_create non-inline` to ease ABI
>stability.(Ray)
>> - Move `rte_event_eth_rx_adapter_queue_event_vector_config` and
>>   `rte_event_eth_rx_adapter_vector_limits_get` implementation to
>the patch
>>   where they are initially defined.(Ray)
>> - Multiple gramatical and style fixes.(Jerin)
>> - Add missing release notes.(Jerin)
>>
>> v4 Changes:
>> - Fix missing event vector structure in event structure.(Jay)
>>
>> v3 Changes:
>> - Fix unintended formatting changes.
>>
>> v2 Changes:
>> - Multiple gramatical and style fixes.(Jerin)
>> - Add parameter to define vector size in power of 2. (Jerin)
>> - Redo patch series w/o breaking ABI till the last patch.(David)
>> - Add deprication notice to announce ABI break in 21.11.(David)
>> - Add vector limits validation to app/test-eventdev.
>>
>> Pavan Nikhilesh (8):
>>   eventdev: introduce event vector capability
>>   eventdev: introduce event vector Rx capability
>>   eventdev: introduce event vector Tx capability
>>   eventdev: add Rx adapter event vector support
>>   eventdev: add Tx adapter event vector support
>>   app/eventdev: add event vector mode in pipeline test
>>   doc: announce event Rx adapter config changes
>>   eventdev: simplify Rx adapter event vector config
>>
>>  app/test-eventdev/evt_common.h                |   4 +
>>  app/test-eventdev/evt_options.c               |  52 +++
>>  app/test-eventdev/evt_options.h               |   4 +
>>  app/test-eventdev/test_pipeline_atq.c         | 310 +++++++++++++++--
>>  app/test-eventdev/test_pipeline_common.c      | 105 +++++-
>>  app/test-eventdev/test_pipeline_common.h      |  18 +
>>  app/test-eventdev/test_pipeline_queue.c       | 320
>++++++++++++++++--
>>  .../prog_guide/event_ethernet_rx_adapter.rst  |  38 +++
>>  .../prog_guide/event_ethernet_tx_adapter.rst  |  12 +
>>  doc/guides/prog_guide/eventdev.rst            |  36 +-
>>  doc/guides/rel_notes/deprecation.rst          |   9 +
>>  doc/guides/rel_notes/release_21_05.rst        |   8 +
>>  doc/guides/tools/testeventdev.rst             |  45 ++-
>>  lib/librte_eventdev/eventdev_pmd.h            |  31 +-
>>  .../rte_event_eth_rx_adapter.c                | 305 ++++++++++++++++-
>>  .../rte_event_eth_rx_adapter.h                |  78 +++++
>>  .../rte_event_eth_tx_adapter.c                |  66 +++-
>>  lib/librte_eventdev/rte_eventdev.c            |  53 ++-
>>  lib/librte_eventdev/rte_eventdev.h            | 113 ++++++-
>>  lib/librte_eventdev/version.map               |   4 +
>>  20 files changed, 1524 insertions(+), 87 deletions(-)
>>
>> --
>> 2.17.1
>
>Just a heads up. v5 patchset doesn't apply cleanly on HEAD
>(5f0849c1155849dfdbf950c91c52cdf9cd301f59). Although, it applies
>cleanly on app/eventdev: fix timeout accuracy
>(c33d48387dc8ccf1b432820f6e0cd4992ab486df).

This patchset is currently rebased on main branch, I will rebase it on 
dpdk-next-event in next version.



^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v5 1/8] eventdev: introduce event vector capability
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 1/8] eventdev: introduce event vector capability pbhagavatula
@ 2021-03-24  6:48           ` Jayatheerthan, Jay
  2021-03-24 18:20             ` Pavan Nikhilesh Bhagavatula
  2021-03-24  9:16           ` Kinsella, Ray
  1 sibling, 1 reply; 153+ messages in thread
From: Jayatheerthan, Jay @ 2021-03-24  6:48 UTC (permalink / raw)
  To: pbhagavatula, jerinj, Carrillo, Erik G, Gujjar, Abhinandan S,
	McDaniel, Timothy, hemant.agrawal, Van Haaren, Harry,
	mattias.ronnblom, Ma, Liang J, Ray Kinsella, Neil Horman
  Cc: dev

> -----Original Message-----
> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
> Sent: Wednesday, March 24, 2021 10:35 AM
> To: jerinj@marvell.com; Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Carrillo, Erik G <erik.g.carrillo@intel.com>; Gujjar,
> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy <timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom <mattias.ronnblom@ericsson.com>; Ma, Liang J
> <liang.j.ma@intel.com>; Ray Kinsella <mdr@ashroe.eu>; Neil Horman <nhorman@tuxdriver.com>
> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
> Subject: [dpdk-dev] [PATCH v5 1/8] eventdev: introduce event vector capability
> 
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> Introduce rte_event_vector datastructure which is capable of holding
> multiple uintptr_t of the same flow thereby allowing applications
> to vectorize their pipeline and reducing the complexity of pipelining
> the events across multiple stages.
> This approach also reduces the scheduling overhead on a event device.
> 
> Add a event vector mempool create handler to create mempools based on
> the best mempool ops available on a given platform.
> 
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> Acked-by: Jerin Jacob <jerinj@marvell.com>
> ---
>  doc/guides/prog_guide/eventdev.rst     | 36 +++++++++++-
>  doc/guides/rel_notes/release_21_05.rst |  8 +++
>  lib/librte_eventdev/rte_eventdev.c     | 42 +++++++++++++
>  lib/librte_eventdev/rte_eventdev.h     | 81 +++++++++++++++++++++++++-
>  lib/librte_eventdev/version.map        |  3 +
>  5 files changed, 167 insertions(+), 3 deletions(-)
> 
> diff --git a/doc/guides/prog_guide/eventdev.rst b/doc/guides/prog_guide/eventdev.rst
> index ccde086f6..fda9c3743 100644
> --- a/doc/guides/prog_guide/eventdev.rst
> +++ b/doc/guides/prog_guide/eventdev.rst
> @@ -63,13 +63,45 @@ the actual event being scheduled is. The payload is a union of the following:
>  * ``uint64_t u64``
>  * ``void *event_ptr``
>  * ``struct rte_mbuf *mbuf``
> +* ``struct rte_event_vector *vec``
> 
> -These three items in a union occupy the same 64 bits at the end of the rte_event
> +These four items in a union occupy the same 64 bits at the end of the rte_event
>  structure. The application can utilize the 64 bits directly by accessing the
> -u64 variable, while the event_ptr and mbuf are provided as convenience
> +u64 variable, while the event_ptr, mbuf, vec are provided as a convenience
>  variables.  For example the mbuf pointer in the union can used to schedule a
>  DPDK packet.
> 
> +Event Vector
> +~~~~~~~~~~~~
> +
> +The rte_event_vector struct contains a vector of elements defined by the event
> +type specified in the ``rte_event``. The event_vector structure contains the
> +following data:
> +
> +* ``nb_elem`` - The number of elements held within the vector.
> +
> +Similar to ``rte_event`` the payload of event vector is also a union, allowing
> +flexibility in what the actual vector is.
> +
> +* ``struct rte_mbuf *mbufs[0]`` - An array of mbufs.
> +* ``void *ptrs[0]`` - An array of pointers.
> +* ``uint64_t *u64s[0]`` - An array of uint64_t elements.
> +
> +The size of the event vector is related to the total number of elements it is
> +configured to hold, this is achieved by making `rte_event_vector` a variable
> +length structure.
> +A helper function is provided to create a mempool that holds event vector, which
> +takes name of the pool, total number of required ``rte_event_vector``,
> +cache size, number of elements in each ``rte_event_vector`` and socket id.
> +
> +.. code-block:: c
> +
> +        rte_event_vector_pool_create("vector_pool", nb_event_vectors, cache_sz,
> +                                     nb_elements_per_vector, socket_id);
> +
> +The function ``rte_event_vector_pool_create`` creates mempool with the best
> +platform mempool ops.
> +
>  Queues
>  ~~~~~~
> 
> diff --git a/doc/guides/rel_notes/release_21_05.rst b/doc/guides/rel_notes/release_21_05.rst
> index 8e686cc62..358623f2f 100644
> --- a/doc/guides/rel_notes/release_21_05.rst
> +++ b/doc/guides/rel_notes/release_21_05.rst
> @@ -101,6 +101,14 @@ New Features
>    * Added command to display Rx queue used descriptor count.
>      ``show port (port_id) rxq (queue_id) desc used count``
> 
> +* **Add Event device vector capability.**
> +
> +  * Added ``rte_event_vector`` data structure which is capable of holding
> +    multiple ``uintptr_t`` of the same flow thereby allowing applications
> +    to vectorize their pipelines and also reduce the complexity of pipelining
> +    the events across multiple stages.
> +  * This also reduces the scheduling overhead on a event device.
> +
> 
>  Removed Items
>  -------------
> diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
> index b57363f80..f95edc075 100644
> --- a/lib/librte_eventdev/rte_eventdev.c
> +++ b/lib/librte_eventdev/rte_eventdev.c
> @@ -1266,6 +1266,48 @@ int rte_event_dev_selftest(uint8_t dev_id)
>  	return -ENOTSUP;
>  }
> 
> +struct rte_mempool *
> +rte_event_vector_pool_create(const char *name, unsigned int n,
> +			     unsigned int cache_size, uint16_t nb_elem,
> +			     int socket_id)
> +{
> +	const char *mp_ops_name;
> +	struct rte_mempool *mp;
> +	unsigned int elt_sz;
> +	int ret;
> +
> +	if (!nb_elem) {
> +		RTE_LOG(ERR, EVENTDEV,
> +			"Invalid number of elements=%d requested\n", nb_elem);
> +		rte_errno = -EINVAL;

rte_mempool_create_empty() call below returns non-negative EINVAL. Should we maintain consistency within same API call?

> +		return NULL;
> +	}
> +
> +	elt_sz =
> +		sizeof(struct rte_event_vector) + (nb_elem * sizeof(uintptr_t));
> +	mp = rte_mempool_create_empty(name, n, elt_sz, cache_size, 0, socket_id,
> +				      0);
> +	if (mp == NULL)
> +		return NULL;
> +
> +	mp_ops_name = rte_mbuf_best_mempool_ops();
> +	ret = rte_mempool_set_ops_byname(mp, mp_ops_name, NULL);
> +	if (ret != 0) {
> +		RTE_LOG(ERR, EVENTDEV, "error setting mempool handler\n");
> +		goto err;
> +	}
> +
> +	ret = rte_mempool_populate_default(mp);
> +	if (ret < 0)
> +		goto err;
> +
> +	return mp;
> +err:
> +	rte_mempool_free(mp);
> +	rte_errno = -ret;

rte_mempool_set_ops_byname() API already returns negative ret and we are making it positive. DPDK has many instances of error/ret being negative and positive. Probably a larger effort to make it consistent would help in general.

> +	return NULL;
> +}
> +
>  int
>  rte_event_dev_start(uint8_t dev_id)
>  {
> diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
> index ce1fc2ce0..aa4dd3959 100644
> --- a/lib/librte_eventdev/rte_eventdev.h
> +++ b/lib/librte_eventdev/rte_eventdev.h
> @@ -212,8 +212,10 @@ extern "C" {
> 
>  #include <rte_common.h>
>  #include <rte_config.h>
> -#include <rte_memory.h>
>  #include <rte_errno.h>
> +#include <rte_mbuf_pool_ops.h>
> +#include <rte_memory.h>
> +#include <rte_mempool.h>
> 
>  #include "rte_eventdev_trace_fp.h"
> 
> @@ -913,6 +915,31 @@ rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
>  int
>  rte_event_dev_close(uint8_t dev_id);
> 
> +/**
> + * Event vector structure.
> + */
> +struct rte_event_vector {
> +	uint64_t nb_elem : 16;
> +	/**< Number of elements in this event vector. */
> +	uint64_t rsvd : 48;
> +	/**< Reserved for future use */
> +	uint64_t impl_opaque;
> +	/**< Implementation specific opaque value.
> +	 * An implementation may use this field to hold implementation specific
> +	 * value to share between dequeue and enqueue operation.
> +	 * The application should not modify this field.
> +	 */
> +	union {
> +		struct rte_mbuf *mbufs[0];
> +		void *ptrs[0];
> +		uint64_t *u64s[0];
> +	} __rte_aligned(16);
> +	/**< Start of the vector array union. Depending upon the event type the
> +	 * vector array can be an array of mbufs or pointers or opaque u64
> +	 * values.
> +	 */
> +};
> +
>  /* Scheduler type definitions */
>  #define RTE_SCHED_TYPE_ORDERED          0
>  /**< Ordered scheduling
> @@ -986,6 +1013,21 @@ rte_event_dev_close(uint8_t dev_id);
>   */
>  #define RTE_EVENT_TYPE_ETH_RX_ADAPTER   0x4
>  /**< The event generated from event eth Rx adapter */
> +#define RTE_EVENT_TYPE_VECTOR           0x8
> +/**< Indicates that event is a vector.
> + * All vector event types should be a logical OR of EVENT_TYPE_VECTOR.
> + * This simplifies the pipeline design as one can split processing the events
> + * between vector events and normal event across event types.
> + * Example:
> + *	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
> + *		// Classify and handle vector event.
> + *	} else {
> + *		// Classify and handle event.
> + *	}
> + */
> +#define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
> +/**< The event vector generated from cpu for pipelining. */
> +
>  #define RTE_EVENT_TYPE_MAX              0x10
>  /**< Maximum number of event types */
> 
> @@ -1108,6 +1150,8 @@ struct rte_event {
>  		/**< Opaque event pointer */
>  		struct rte_mbuf *mbuf;
>  		/**< mbuf pointer if dequeued event is associated with mbuf */
> +		struct rte_event_vector *vec;
> +		/**< Event vector pointer. */
>  	};
>  };
> 
> @@ -2023,6 +2067,41 @@ rte_event_dev_xstats_reset(uint8_t dev_id,
>   */
>  int rte_event_dev_selftest(uint8_t dev_id);
> 
> +/**
> + * Get the memory required per event vector based on the number of elements per
> + * vector.
> + * This should be used to create the mempool that holds the event vectors.
> + *
> + * @param name
> + *   The name of the vector pool.
> + * @param n
> + *   The number of elements in the mbuf pool.
> + * @param cache_size
> + *   Size of the per-core object cache. See rte_mempool_create() for
> + *   details.
> + * @param nb_elem
> + *   The number of elements then a single event vector should be able to hold.

Typo: that instead of then.

> + * @param socket_id
> + *   The socket identifier where the memory should be allocated. The
> + *   value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
> + *   reserved zone
> + *
> + * @return
> + *   The pointer to the newly allocated mempool, on success. NULL on error
> + *   with rte_errno set appropriately. Possible rte_errno values include:
> + *    - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
> + *    - E_RTE_SECONDARY - function was called from a secondary process instance
> + *    - EINVAL - cache size provided is too large, or priv_size is not aligned.
> + *    - ENOSPC - the maximum number of memzones has already been allocated
> + *    - EEXIST - a memzone with the same name already exists
> + *    - ENOMEM - no appropriate memory area found in which to create memzone

rte_mempool_create_empty() can return ENAMETOOLONG if name is too long.

> + */
> +__rte_experimental
> +struct rte_mempool *
> +rte_event_vector_pool_create(const char *name, unsigned int n,
> +			     unsigned int cache_size, uint16_t nb_elem,
> +			     int socket_id);
> +
>  #ifdef __cplusplus
>  }
>  #endif
> diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
> index 3e5c09cfd..a070ef56e 100644
> --- a/lib/librte_eventdev/version.map
> +++ b/lib/librte_eventdev/version.map
> @@ -138,6 +138,9 @@ EXPERIMENTAL {
>  	__rte_eventdev_trace_port_setup;
>  	# added in 20.11
>  	rte_event_pmd_pci_probe_named;
> +
> +	#added in 21.05
> +	rte_event_vector_pool_create;
>  };
> 
>  INTERNAL {
> --
> 2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v5 0/8] Introduce event vectorization
  2021-03-24  6:44           ` Pavan Nikhilesh Bhagavatula
@ 2021-03-24  8:10             ` Jayatheerthan, Jay
  0 siblings, 0 replies; 153+ messages in thread
From: Jayatheerthan, Jay @ 2021-03-24  8:10 UTC (permalink / raw)
  To: Pavan Nikhilesh Bhagavatula, Jerin Jacob Kollanukkaran, Carrillo,
	Erik G, Gujjar, Abhinandan S, McDaniel, Timothy, hemant.agrawal,
	Van Haaren, Harry, mattias.ronnblom, Ma, Liang J
  Cc: dev

> -----Original Message-----
> From: Pavan Nikhilesh Bhagavatula <pbhagavatula@marvell.com>
> Sent: Wednesday, March 24, 2021 12:15 PM
> To: Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Jerin Jacob Kollanukkaran <jerinj@marvell.com>; Carrillo, Erik G
> <erik.g.carrillo@intel.com>; Gujjar, Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy
> <timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
> <mattias.ronnblom@ericsson.com>; Ma, Liang J <liang.j.ma@intel.com>
> Cc: dev@dpdk.org
> Subject: RE: [dpdk-dev] [PATCH v5 0/8] Introduce event vectorization
> 
> >> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
> >> Sent: Wednesday, March 24, 2021 10:35 AM
> >> To: jerinj@marvell.com; Jayatheerthan, Jay
> ><jay.jayatheerthan@intel.com>; Carrillo, Erik G
> ><erik.g.carrillo@intel.com>; Gujjar, Abhinandan
> >> S <abhinandan.gujjar@intel.com>; McDaniel, Timothy
> ><timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
> >Haaren, Harry
> >> <harry.van.haaren@intel.com>; mattias.ronnblom
> ><mattias.ronnblom@ericsson.com>; Ma, Liang J
> ><liang.j.ma@intel.com>
> >> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
> >> Subject: [dpdk-dev] [PATCH v5 0/8] Introduce event vectorization
> >>
> >> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> >>
> >> In traditional event programming model, events are identified by a
> >> flow-id and a uintptr_t. The flow-id uniquely identifies a given event
> >> and determines the order of scheduling based on schedule type, the
> >> uintptr_t holds a single object.
> >>
> >> Event devices also support burst mode with configurable dequeue
> >depth,
> >> i.e. each dequeue call would return multiple events and each event
> >> might be at a different stage of the pipeline.
> >> Having a burst of events belonging to different stages in a dequeue
> >> burst is not only difficult to vectorize but also increases the scheduler
> >> overhead and application overhead of pipelining events further.
> >> Using event vectors we see a performance gain of ~628% as shown in
> >[1].
> >This is very impressive performance boost. Thanks so much for putting
> >this patchset together! Just curious, was any performance
> >measurement done for existing applications (non-vector)?
> >>
> >> By introducing event vectorization, each event will be capable of
> >holding
> >> multiple uintptr_t of the same flow thereby allowing applications
> >> to vectorize their pipeline and reduce the complexity of pipelining
> >> events across multiple stages. This also reduces the complexity of
> >handling
> >> enqueue and dequeue on an event device.
> >>
> >> Since event devices are transparent to the events they are scheduling
> >> so the event producers such as eth_rx_adapter, crypto_adapter , etc..
> >> are responsible for vectorizing the buffers of the same flow into a
> >single
> >> event.
> >>
> >> The series also breaks ABI in the patch [8/8] which is targetted to the
> >> v21.11 release.
> >>
> >> The dpdk-test-eventdev application has been updated with options to
> >test
> >> multiple vector sizes and timeouts.
> >>
> >> [1]
> >> As for performance improvement, with a ARM Cortex-A72 equivalent
> >processer,
> >> software event device (--vdev=event_sw0), single worker core, single
> >stage
> >> and using one service core for Rx adapter, Tx adapter, Scheduling.
> >>
> >> Without event vectorization:
> >>     ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --
> >vdev="event_sw0" --
> >>          --prod_type_ethdev --nb_pkts=0 --verbose 2 --
> >test=pipeline_queue
> >>          --stlist=a --wlcores=20
> >>     Port[0] using Rx adapter[0] configured
> >>     Port[0] using Tx adapter[0] Configured
> >>     4.728 mpps avg 4.728 mpps
> >Is this number before the patchset? If so, it would help put similar
> >number with the patchset but not using vectorization feature.
> 
> I don’t remember the exact clock frequency I was using when I ran
> the above test but with equal clocks:
> 1. Without the patchset applied
> 	5.071 mpps
> 2. With patchset applied w/o enabling vector
> 	5.123 mpps
> 3. With patchset applied with enabling vector
> 	vector_sz@256 42.715 mpps
> 	vector_sz@512 45.335 mpps
> 

Thanks Pavan for the details. It may be useful to include this info in the patchset.

> >>
> >> With event vectorization:
> >>     ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --
> >vdev="event_sw0" --
> >>         --prod_type_ethdev --nb_pkts=0 --verbose 2 --
> >test=pipeline_queue
> >>         --stlist=a --wlcores=20 --enable_vector --nb_eth_queues 1
> >>         --vector_size 256
> >>     Port[0] using Rx adapter[0] configured
> >>     Port[0] using Tx adapter[0] Configured
> >>     34.383 mpps avg 34.383 mpps
> >>
> >> Having dedicated service cores for each Rx queues and tweaking the
> >vector,
> >> dequeue burst size would further improve performance.
> >>
> >> API usage is shown below:
> >>
> >> Configuration:
> >>
> >> 	struct rte_event_eth_rx_adapter_event_vector_config
> >vec_conf;
> >>
> >> 	vector_pool = rte_event_vector_pool_create("vector_pool",
> >> 			nb_elem, 0, vector_size, socket_id);
> >>
> >> 	rte_event_eth_rx_adapter_create(id, event_id, &adptr_conf);
> >> 	rte_event_eth_rx_adapter_queue_add(id, eth_id, -1,
> >&queue_conf);
> >> 	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)
> >{
> >> 		vec_conf.vector_sz = vector_size;
> >> 		vec_conf.vector_timeout_ns = vector_tmo_nsec;
> >> 		vec_conf.vector_mp = vector_pool;
> >>
> >	rte_event_eth_rx_adapter_queue_event_vector_config(id,
> >> 				eth_id, -1, &vec_conf);
> >> 	}
> >>
> >> Fastpath:
> >>
> >> 	num = rte_event_dequeue_burst(event_id, port_id, &ev, 1, 0);
> >> 	if (!num)
> >> 		continue;
> >>
> >> 	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
> >> 		switch (ev.event_type) {
> >> 		case RTE_EVENT_TYPE_ETHDEV_VECTOR:
> >> 		case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
> >> 			struct rte_mbuf **mbufs;
> >>
> >> 			mbufs = ev.vector_ev->mbufs;
> >> 			for (i = 0; i < ev.vector_ev->nb_elem; i++)
> >> 				//Process mbufs.
> >> 			break;
> >> 		case ...
> >> 		}
> >> 	}
> >> 	...
> >>
> >> v5 Changes:
> >> - Make `rte_event_vector_pool_create non-inline` to ease ABI
> >stability.(Ray)
> >> - Move `rte_event_eth_rx_adapter_queue_event_vector_config` and
> >>   `rte_event_eth_rx_adapter_vector_limits_get` implementation to
> >the patch
> >>   where they are initially defined.(Ray)
> >> - Multiple gramatical and style fixes.(Jerin)
> >> - Add missing release notes.(Jerin)
> >>
> >> v4 Changes:
> >> - Fix missing event vector structure in event structure.(Jay)
> >>
> >> v3 Changes:
> >> - Fix unintended formatting changes.
> >>
> >> v2 Changes:
> >> - Multiple gramatical and style fixes.(Jerin)
> >> - Add parameter to define vector size in power of 2. (Jerin)
> >> - Redo patch series w/o breaking ABI till the last patch.(David)
> >> - Add deprication notice to announce ABI break in 21.11.(David)
> >> - Add vector limits validation to app/test-eventdev.
> >>
> >> Pavan Nikhilesh (8):
> >>   eventdev: introduce event vector capability
> >>   eventdev: introduce event vector Rx capability
> >>   eventdev: introduce event vector Tx capability
> >>   eventdev: add Rx adapter event vector support
> >>   eventdev: add Tx adapter event vector support
> >>   app/eventdev: add event vector mode in pipeline test
> >>   doc: announce event Rx adapter config changes
> >>   eventdev: simplify Rx adapter event vector config
> >>
> >>  app/test-eventdev/evt_common.h                |   4 +
> >>  app/test-eventdev/evt_options.c               |  52 +++
> >>  app/test-eventdev/evt_options.h               |   4 +
> >>  app/test-eventdev/test_pipeline_atq.c         | 310 +++++++++++++++--
> >>  app/test-eventdev/test_pipeline_common.c      | 105 +++++-
> >>  app/test-eventdev/test_pipeline_common.h      |  18 +
> >>  app/test-eventdev/test_pipeline_queue.c       | 320
> >++++++++++++++++--
> >>  .../prog_guide/event_ethernet_rx_adapter.rst  |  38 +++
> >>  .../prog_guide/event_ethernet_tx_adapter.rst  |  12 +
> >>  doc/guides/prog_guide/eventdev.rst            |  36 +-
> >>  doc/guides/rel_notes/deprecation.rst          |   9 +
> >>  doc/guides/rel_notes/release_21_05.rst        |   8 +
> >>  doc/guides/tools/testeventdev.rst             |  45 ++-
> >>  lib/librte_eventdev/eventdev_pmd.h            |  31 +-
> >>  .../rte_event_eth_rx_adapter.c                | 305 ++++++++++++++++-
> >>  .../rte_event_eth_rx_adapter.h                |  78 +++++
> >>  .../rte_event_eth_tx_adapter.c                |  66 +++-
> >>  lib/librte_eventdev/rte_eventdev.c            |  53 ++-
> >>  lib/librte_eventdev/rte_eventdev.h            | 113 ++++++-
> >>  lib/librte_eventdev/version.map               |   4 +
> >>  20 files changed, 1524 insertions(+), 87 deletions(-)
> >>
> >> --
> >> 2.17.1
> >
> >Just a heads up. v5 patchset doesn't apply cleanly on HEAD
> >(5f0849c1155849dfdbf950c91c52cdf9cd301f59). Although, it applies
> >cleanly on app/eventdev: fix timeout accuracy
> >(c33d48387dc8ccf1b432820f6e0cd4992ab486df).
> 
> This patchset is currently rebased on main branch, I will rebase it on
> dpdk-next-event in next version.
> 


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v5 2/8] eventdev: introduce event vector Rx capability
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 2/8] eventdev: introduce event vector Rx capability pbhagavatula
@ 2021-03-24  9:15           ` Kinsella, Ray
  2021-03-25  8:15           ` Jayatheerthan, Jay
  1 sibling, 0 replies; 153+ messages in thread
From: Kinsella, Ray @ 2021-03-24  9:15 UTC (permalink / raw)
  To: pbhagavatula, jerinj, jay.jayatheerthan, erik.g.carrillo,
	abhinandan.gujjar, timothy.mcdaniel, hemant.agrawal,
	harry.van.haaren, mattias.ronnblom, liang.j.ma, Neil Horman
  Cc: dev



On 24/03/2021 05:05, pbhagavatula@marvell.com wrote:
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> Introduce event ethernet Rx adapter event vector capability.
> 
> If an event eth Rx adapter has the capability of
> RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR then a given Rx queue
> can be configured to enable event vectorization by passing the
> flag RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR to
> rte_event_eth_rx_adapter_queue_conf::rx_queue_flags while configuring
> Rx adapter through rte_event_eth_rx_adapter_queue_add().
> 
> The max vector size, vector timeout define the vector size and
> mempool used for allocating vector event are configured through
> rte_event_eth_rx_adapter_queue_add. The element size of the element
> in the vector pool should be equal to
>     sizeof(struct rte_event_vector) + (vector_sz * sizeof(uintptr_t))
> 
> Application can use `rte_event_vector_pool_create` to create the
> vector mempool used for
> rte_event_eth_rx_adapter_queue_conf::vector_mp.
> 
> The Rx adapter would be responsible for vectorizing the mbufs
> based on the flow, the vector limits configured by the application
> and add the vector event of mbufs to the event queue set via
> rte_event_eth_rx_adapter_queue_conf::ev::queue_id.
> It should also mark rte_event_vector::union_valid and fill
> rte_event_vector::port, rte_event_vector::queue.
> 
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> Acked-by: Jerin Jacob <jerinj@marvell.com>
> ---
>  .../prog_guide/event_ethernet_rx_adapter.rst  |  38 ++++++
>  lib/librte_eventdev/eventdev_pmd.h            |  53 ++++++++
>  .../rte_event_eth_rx_adapter.c                | 114 ++++++++++++++++++
>  .../rte_event_eth_rx_adapter.h                | 105 ++++++++++++++++
>  lib/librte_eventdev/rte_eventdev.h            |  30 ++++-
>  lib/librte_eventdev/version.map               |   2 +
>  6 files changed, 340 insertions(+), 2 deletions(-)
> 

Acked-by: Ray Kinsella <mdr@ashroe.eu>

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v5 7/8] doc: announce event Rx adapter config changes
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 7/8] doc: announce event Rx adapter config changes pbhagavatula
@ 2021-03-24  9:16           ` Kinsella, Ray
  0 siblings, 0 replies; 153+ messages in thread
From: Kinsella, Ray @ 2021-03-24  9:16 UTC (permalink / raw)
  To: pbhagavatula, jerinj, jay.jayatheerthan, erik.g.carrillo,
	abhinandan.gujjar, timothy.mcdaniel, hemant.agrawal,
	harry.van.haaren, mattias.ronnblom, liang.j.ma, Neil Horman
  Cc: dev



On 24/03/2021 05:05, pbhagavatula@marvell.com wrote:
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> The Rx adapter event vector configuration will be merged into
> Rx adapter queue configuration to simplify enabling event
> vectorization.
> 
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> ---
>  doc/guides/rel_notes/deprecation.rst | 9 +++++++++
>  1 file changed, 9 insertions(+)
> 
> diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst
> index 64629e064..ef6d1bb18 100644
> --- a/doc/guides/rel_notes/deprecation.rst
> +++ b/doc/guides/rel_notes/deprecation.rst
> @@ -130,3 +130,12 @@ Deprecation Notices
>  * cmdline: ``cmdline`` structure will be made opaque to hide platform-specific
>    content. On Linux and FreeBSD, supported prior to DPDK 20.11,
>    original structure will be kept until DPDK 21.11.
> +
> +* eventdev: The structure ``rte_event_eth_rx_adapter_queue_conf`` will be
> +  extended to include ``rte_event_eth_rx_adapter_event_vector_config`` elements
> +  and the function ``rte_event_eth_rx_adapter_queue_event_vector_config`` will
> +  be removed in DPDK 21.11.
> +
> +  An application can enable event vectorization by passing the desired vector
> +  values to the function ``rte_event_eth_rx_adapter_queue_add`` using
> +  the structure ``rte_event_eth_rx_adapter_queue_add``.

Acked-by: Ray Kinsella <mdr@ashroe.eu>

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v5 1/8] eventdev: introduce event vector capability
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 1/8] eventdev: introduce event vector capability pbhagavatula
  2021-03-24  6:48           ` Jayatheerthan, Jay
@ 2021-03-24  9:16           ` Kinsella, Ray
  1 sibling, 0 replies; 153+ messages in thread
From: Kinsella, Ray @ 2021-03-24  9:16 UTC (permalink / raw)
  To: pbhagavatula, jerinj, jay.jayatheerthan, erik.g.carrillo,
	abhinandan.gujjar, timothy.mcdaniel, hemant.agrawal,
	harry.van.haaren, mattias.ronnblom, liang.j.ma, Neil Horman
  Cc: dev



On 24/03/2021 05:05, pbhagavatula@marvell.com wrote:
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> Introduce rte_event_vector datastructure which is capable of holding
> multiple uintptr_t of the same flow thereby allowing applications
> to vectorize their pipeline and reducing the complexity of pipelining
> the events across multiple stages.
> This approach also reduces the scheduling overhead on a event device.
> 
> Add a event vector mempool create handler to create mempools based on
> the best mempool ops available on a given platform.
> 
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> Acked-by: Jerin Jacob <jerinj@marvell.com>
> ---
>  doc/guides/prog_guide/eventdev.rst     | 36 +++++++++++-
>  doc/guides/rel_notes/release_21_05.rst |  8 +++
>  lib/librte_eventdev/rte_eventdev.c     | 42 +++++++++++++
>  lib/librte_eventdev/rte_eventdev.h     | 81 +++++++++++++++++++++++++-
>  lib/librte_eventdev/version.map        |  3 +
>  5 files changed, 167 insertions(+), 3 deletions(-)
> 
Acked-by: Ray Kinsella <mdr@ashroe.eu>

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v5 1/8] eventdev: introduce event vector capability
  2021-03-24  6:48           ` Jayatheerthan, Jay
@ 2021-03-24 18:20             ` Pavan Nikhilesh Bhagavatula
  0 siblings, 0 replies; 153+ messages in thread
From: Pavan Nikhilesh Bhagavatula @ 2021-03-24 18:20 UTC (permalink / raw)
  To: Jayatheerthan, Jay, Jerin Jacob Kollanukkaran, Carrillo, Erik G,
	Gujjar, Abhinandan S, McDaniel, Timothy, hemant.agrawal,
	Van Haaren, Harry, mattias.ronnblom, Ma, Liang J, Ray Kinsella,
	Neil Horman
  Cc: dev

>> -----Original Message-----
>> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
>> Sent: Wednesday, March 24, 2021 10:35 AM
>> To: jerinj@marvell.com; Jayatheerthan, Jay
><jay.jayatheerthan@intel.com>; Carrillo, Erik G
><erik.g.carrillo@intel.com>; Gujjar,
>> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy
><timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
>> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
><mattias.ronnblom@ericsson.com>; Ma, Liang J
>> <liang.j.ma@intel.com>; Ray Kinsella <mdr@ashroe.eu>; Neil Horman
><nhorman@tuxdriver.com>
>> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
>> Subject: [dpdk-dev] [PATCH v5 1/8] eventdev: introduce event vector
>capability
>>
>> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>>
>> Introduce rte_event_vector datastructure which is capable of holding
>> multiple uintptr_t of the same flow thereby allowing applications
>> to vectorize their pipeline and reducing the complexity of pipelining
>> the events across multiple stages.
>> This approach also reduces the scheduling overhead on a event
>device.
>>
>> Add a event vector mempool create handler to create mempools
>based on
>> the best mempool ops available on a given platform.
>>
>> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
>> Acked-by: Jerin Jacob <jerinj@marvell.com>
>> ---
>>  doc/guides/prog_guide/eventdev.rst     | 36 +++++++++++-
>>  doc/guides/rel_notes/release_21_05.rst |  8 +++
>>  lib/librte_eventdev/rte_eventdev.c     | 42 +++++++++++++
>>  lib/librte_eventdev/rte_eventdev.h     | 81
>+++++++++++++++++++++++++-
>>  lib/librte_eventdev/version.map        |  3 +
>>  5 files changed, 167 insertions(+), 3 deletions(-)
>>
>> diff --git a/doc/guides/prog_guide/eventdev.rst
>b/doc/guides/prog_guide/eventdev.rst
>> index ccde086f6..fda9c3743 100644
>> --- a/doc/guides/prog_guide/eventdev.rst
>> +++ b/doc/guides/prog_guide/eventdev.rst
>> @@ -63,13 +63,45 @@ the actual event being scheduled is. The
>payload is a union of the following:
>>  * ``uint64_t u64``
>>  * ``void *event_ptr``
>>  * ``struct rte_mbuf *mbuf``
>> +* ``struct rte_event_vector *vec``
>>
>> -These three items in a union occupy the same 64 bits at the end of
>the rte_event
>> +These four items in a union occupy the same 64 bits at the end of the
>rte_event
>>  structure. The application can utilize the 64 bits directly by accessing
>the
>> -u64 variable, while the event_ptr and mbuf are provided as
>convenience
>> +u64 variable, while the event_ptr, mbuf, vec are provided as a
>convenience
>>  variables.  For example the mbuf pointer in the union can used to
>schedule a
>>  DPDK packet.
>>
>> +Event Vector
>> +~~~~~~~~~~~~
>> +
>> +The rte_event_vector struct contains a vector of elements defined
>by the event
>> +type specified in the ``rte_event``. The event_vector structure
>contains the
>> +following data:
>> +
>> +* ``nb_elem`` - The number of elements held within the vector.
>> +
>> +Similar to ``rte_event`` the payload of event vector is also a union,
>allowing
>> +flexibility in what the actual vector is.
>> +
>> +* ``struct rte_mbuf *mbufs[0]`` - An array of mbufs.
>> +* ``void *ptrs[0]`` - An array of pointers.
>> +* ``uint64_t *u64s[0]`` - An array of uint64_t elements.
>> +
>> +The size of the event vector is related to the total number of
>elements it is
>> +configured to hold, this is achieved by making `rte_event_vector` a
>variable
>> +length structure.
>> +A helper function is provided to create a mempool that holds event
>vector, which
>> +takes name of the pool, total number of required
>``rte_event_vector``,
>> +cache size, number of elements in each ``rte_event_vector`` and
>socket id.
>> +
>> +.. code-block:: c
>> +
>> +        rte_event_vector_pool_create("vector_pool",
>nb_event_vectors, cache_sz,
>> +                                     nb_elements_per_vector, socket_id);
>> +
>> +The function ``rte_event_vector_pool_create`` creates mempool
>with the best
>> +platform mempool ops.
>> +
>>  Queues
>>  ~~~~~~
>>
>> diff --git a/doc/guides/rel_notes/release_21_05.rst
>b/doc/guides/rel_notes/release_21_05.rst
>> index 8e686cc62..358623f2f 100644
>> --- a/doc/guides/rel_notes/release_21_05.rst
>> +++ b/doc/guides/rel_notes/release_21_05.rst
>> @@ -101,6 +101,14 @@ New Features
>>    * Added command to display Rx queue used descriptor count.
>>      ``show port (port_id) rxq (queue_id) desc used count``
>>
>> +* **Add Event device vector capability.**
>> +
>> +  * Added ``rte_event_vector`` data structure which is capable of
>holding
>> +    multiple ``uintptr_t`` of the same flow thereby allowing
>applications
>> +    to vectorize their pipelines and also reduce the complexity of
>pipelining
>> +    the events across multiple stages.
>> +  * This also reduces the scheduling overhead on a event device.
>> +
>>
>>  Removed Items
>>  -------------
>> diff --git a/lib/librte_eventdev/rte_eventdev.c
>b/lib/librte_eventdev/rte_eventdev.c
>> index b57363f80..f95edc075 100644
>> --- a/lib/librte_eventdev/rte_eventdev.c
>> +++ b/lib/librte_eventdev/rte_eventdev.c
>> @@ -1266,6 +1266,48 @@ int rte_event_dev_selftest(uint8_t dev_id)
>>  	return -ENOTSUP;
>>  }
>>
>> +struct rte_mempool *
>> +rte_event_vector_pool_create(const char *name, unsigned int n,
>> +			     unsigned int cache_size, uint16_t nb_elem,
>> +			     int socket_id)
>> +{
>> +	const char *mp_ops_name;
>> +	struct rte_mempool *mp;
>> +	unsigned int elt_sz;
>> +	int ret;
>> +
>> +	if (!nb_elem) {
>> +		RTE_LOG(ERR, EVENTDEV,
>> +			"Invalid number of elements=%d requested\n",
>nb_elem);
>> +		rte_errno = -EINVAL;
>
>rte_mempool_create_empty() call below returns non-negative EINVAL.
>Should we maintain consistency within same API call?
>
>> +		return NULL;
>> +	}
>> +
>> +	elt_sz =
>> +		sizeof(struct rte_event_vector) + (nb_elem *
>sizeof(uintptr_t));
>> +	mp = rte_mempool_create_empty(name, n, elt_sz, cache_size,
>0, socket_id,
>> +				      0);
>> +	if (mp == NULL)
>> +		return NULL;
>> +
>> +	mp_ops_name = rte_mbuf_best_mempool_ops();
>> +	ret = rte_mempool_set_ops_byname(mp, mp_ops_name,
>NULL);
>> +	if (ret != 0) {
>> +		RTE_LOG(ERR, EVENTDEV, "error setting mempool
>handler\n");
>> +		goto err;
>> +	}
>> +
>> +	ret = rte_mempool_populate_default(mp);
>> +	if (ret < 0)
>> +		goto err;
>> +
>> +	return mp;
>> +err:
>> +	rte_mempool_free(mp);
>> +	rte_errno = -ret;
>
>rte_mempool_set_ops_byname() API already returns negative ret and
>we are making it positive. DPDK has many instances of error/ret being
>negative and positive. Probably a larger effort to make it consistent
>would help in general.
>

Since rte_eventdev uses positive rte_errno, I will use the same here for
consistency.

>> +	return NULL;
>> +}
>> +
>>  int
>>  rte_event_dev_start(uint8_t dev_id)
>>  {
>> diff --git a/lib/librte_eventdev/rte_eventdev.h
>b/lib/librte_eventdev/rte_eventdev.h
>> index ce1fc2ce0..aa4dd3959 100644
>> --- a/lib/librte_eventdev/rte_eventdev.h
>> +++ b/lib/librte_eventdev/rte_eventdev.h
>> @@ -212,8 +212,10 @@ extern "C" {
>>
>>  #include <rte_common.h>
>>  #include <rte_config.h>
>> -#include <rte_memory.h>
>>  #include <rte_errno.h>
>> +#include <rte_mbuf_pool_ops.h>
>> +#include <rte_memory.h>
>> +#include <rte_mempool.h>
>>
>>  #include "rte_eventdev_trace_fp.h"
>>
>> @@ -913,6 +915,31 @@
>rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
>>  int
>>  rte_event_dev_close(uint8_t dev_id);
>>
>> +/**
>> + * Event vector structure.
>> + */
>> +struct rte_event_vector {
>> +	uint64_t nb_elem : 16;
>> +	/**< Number of elements in this event vector. */
>> +	uint64_t rsvd : 48;
>> +	/**< Reserved for future use */
>> +	uint64_t impl_opaque;
>> +	/**< Implementation specific opaque value.
>> +	 * An implementation may use this field to hold implementation
>specific
>> +	 * value to share between dequeue and enqueue operation.
>> +	 * The application should not modify this field.
>> +	 */
>> +	union {
>> +		struct rte_mbuf *mbufs[0];
>> +		void *ptrs[0];
>> +		uint64_t *u64s[0];
>> +	} __rte_aligned(16);
>> +	/**< Start of the vector array union. Depending upon the event
>type the
>> +	 * vector array can be an array of mbufs or pointers or opaque
>u64
>> +	 * values.
>> +	 */
>> +};
>> +
>>  /* Scheduler type definitions */
>>  #define RTE_SCHED_TYPE_ORDERED          0
>>  /**< Ordered scheduling
>> @@ -986,6 +1013,21 @@ rte_event_dev_close(uint8_t dev_id);
>>   */
>>  #define RTE_EVENT_TYPE_ETH_RX_ADAPTER   0x4
>>  /**< The event generated from event eth Rx adapter */
>> +#define RTE_EVENT_TYPE_VECTOR           0x8
>> +/**< Indicates that event is a vector.
>> + * All vector event types should be a logical OR of
>EVENT_TYPE_VECTOR.
>> + * This simplifies the pipeline design as one can split processing the
>events
>> + * between vector events and normal event across event types.
>> + * Example:
>> + *	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
>> + *		// Classify and handle vector event.
>> + *	} else {
>> + *		// Classify and handle event.
>> + *	}
>> + */
>> +#define RTE_EVENT_TYPE_CPU_VECTOR
>(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
>> +/**< The event vector generated from cpu for pipelining. */
>> +
>>  #define RTE_EVENT_TYPE_MAX              0x10
>>  /**< Maximum number of event types */
>>
>> @@ -1108,6 +1150,8 @@ struct rte_event {
>>  		/**< Opaque event pointer */
>>  		struct rte_mbuf *mbuf;
>>  		/**< mbuf pointer if dequeued event is associated with
>mbuf */
>> +		struct rte_event_vector *vec;
>> +		/**< Event vector pointer. */
>>  	};
>>  };
>>
>> @@ -2023,6 +2067,41 @@ rte_event_dev_xstats_reset(uint8_t
>dev_id,
>>   */
>>  int rte_event_dev_selftest(uint8_t dev_id);
>>
>> +/**
>> + * Get the memory required per event vector based on the number of
>elements per
>> + * vector.
>> + * This should be used to create the mempool that holds the event
>vectors.
>> + *
>> + * @param name
>> + *   The name of the vector pool.
>> + * @param n
>> + *   The number of elements in the mbuf pool.
>> + * @param cache_size
>> + *   Size of the per-core object cache. See rte_mempool_create() for
>> + *   details.
>> + * @param nb_elem
>> + *   The number of elements then a single event vector should be
>able to hold.
>
>Typo: that instead of then.
>
>> + * @param socket_id
>> + *   The socket identifier where the memory should be allocated. The
>> + *   value can be *SOCKET_ID_ANY* if there is no NUMA constraint
>for the
>> + *   reserved zone
>> + *
>> + * @return
>> + *   The pointer to the newly allocated mempool, on success. NULL
>on error
>> + *   with rte_errno set appropriately. Possible rte_errno values
>include:
>> + *    - E_RTE_NO_CONFIG - function could not get pointer to
>rte_config structure
>> + *    - E_RTE_SECONDARY - function was called from a secondary
>process instance
>> + *    - EINVAL - cache size provided is too large, or priv_size is not
>aligned.
>> + *    - ENOSPC - the maximum number of memzones has already been
>allocated
>> + *    - EEXIST - a memzone with the same name already exists
>> + *    - ENOMEM - no appropriate memory area found in which to
>create memzone
>
>rte_mempool_create_empty() can return ENAMETOOLONG if name is
>too long.
>
>> + */
>> +__rte_experimental
>> +struct rte_mempool *
>> +rte_event_vector_pool_create(const char *name, unsigned int n,
>> +			     unsigned int cache_size, uint16_t nb_elem,
>> +			     int socket_id);
>> +
>>  #ifdef __cplusplus
>>  }
>>  #endif
>> diff --git a/lib/librte_eventdev/version.map
>b/lib/librte_eventdev/version.map
>> index 3e5c09cfd..a070ef56e 100644
>> --- a/lib/librte_eventdev/version.map
>> +++ b/lib/librte_eventdev/version.map
>> @@ -138,6 +138,9 @@ EXPERIMENTAL {
>>  	__rte_eventdev_trace_port_setup;
>>  	# added in 20.11
>>  	rte_event_pmd_pci_probe_named;
>> +
>> +	#added in 21.05
>> +	rte_event_vector_pool_create;
>>  };
>>
>>  INTERNAL {
>> --
>> 2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v6 0/8] Introduce event vectorization
  2021-03-24  5:05       ` [dpdk-dev] [PATCH v5 " pbhagavatula
                           ` (8 preceding siblings ...)
  2021-03-24  5:39         ` [dpdk-dev] [PATCH v5 0/8] Introduce event vectorization Jayatheerthan, Jay
@ 2021-03-24 19:28         ` pbhagavatula
  2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 1/8] eventdev: introduce event vector capability pbhagavatula
                             ` (8 more replies)
  9 siblings, 9 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-24 19:28 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

In traditional event programming model, events are identified by a
flow-id and a uintptr_t. The flow-id uniquely identifies a given event
and determines the order of scheduling based on schedule type, the
uintptr_t holds a single object.

Event devices also support burst mode with configurable dequeue depth,
i.e. each dequeue call would return multiple events and each event
might be at a different stage of the pipeline.
Having a burst of events belonging to different stages in a dequeue
burst is not only difficult to vectorize but also increases the scheduler
overhead and application overhead of pipelining events further.
Using event vectors we see a performance gain of ~742.3% as shown in [1].

By introducing event vectorization, each event will be capable of holding
multiple uintptr_t of the same flow thereby allowing applications
to vectorize their pipeline and reduce the complexity of pipelining
events across multiple stages. This also reduces the complexity of handling
enqueue and dequeue on an event device.

Since event devices are transparent to the events they are scheduling
so the event producers such as eth_rx_adapter, crypto_adapter , etc..
are responsible for vectorizing the buffers of the same flow into a single
event.

The series also breaks ABI in the patch [8/8] which is targetted to the
v21.11 release.

The dpdk-test-eventdev application has been updated with options to test
multiple vector sizes and timeouts.

[1]
As for performance improvement, with a ARM Cortex-A72 equivalent processer,
software event device (--vdev=event_sw0), single worker core, single stage
and using one service core for Rx adapter, Tx adapter, Scheduling.

Without this patchset applied:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
         --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
         --stlist=a --wlcores=20
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    5.071 mpps

With the patchset applied and Without event vectorization:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
         --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
         --stlist=a --wlcores=20
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    5.123 mpps

With event vectorization:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
        --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
        --stlist=a --wlcores=20 --enable_vector --nb_eth_queues 1
        --vector_size 256
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    42.715 mpps

Having dedicated service cores for each Rx queues and tweaking the vector,
dequeue burst size would further improve performance.

API usage is shown below:

Configuration:

	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

	vector_pool = rte_event_vector_pool_create("vector_pool",
			nb_elem, 0, vector_size, socket_id);

	rte_event_eth_rx_adapter_create(id, event_id, &adptr_conf);
	rte_event_eth_rx_adapter_queue_add(id, eth_id, -1, &queue_conf);
	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
		vec_conf.vector_sz = vector_size;
		vec_conf.vector_timeout_ns = vector_tmo_nsec;
		vec_conf.vector_mp = vector_pool;
		rte_event_eth_rx_adapter_queue_event_vector_config(id,
				eth_id, -1, &vec_conf);
	}

Fastpath:

	num = rte_event_dequeue_burst(event_id, port_id, &ev, 1, 0);
	if (!num)
		continue;

	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
		switch (ev.event_type) {
		case RTE_EVENT_TYPE_ETHDEV_VECTOR:
		case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
			struct rte_mbuf **mbufs;

			mbufs = ev.vector_ev->mbufs;
			for (i = 0; i < ev.vector_ev->nb_elem; i++)
				//Process mbufs.
			break;
		case ...
		}
	}
	...

v6 Changes:
- Make rte_errno sign consistant.(Jay)
- Gramatical and doxygen fixes. (Jay)

v5 Changes:
- Make `rte_event_vector_pool_create non-inline` to ease ABI stability.(Ray)
- Move `rte_event_eth_rx_adapter_queue_event_vector_config` and
  `rte_event_eth_rx_adapter_vector_limits_get` implementation to the patch
  where they are initially defined.(Ray)
- Multiple gramatical and style fixes.(Jerin)
- Add missing release notes.(Jerin)

v4 Changes:
- Fix missing event vector structure in event structure.(Jay)

v3 Changes:
- Fix unintended formatting changes.

v2 Changes:
- Multiple gramatical and style fixes.(Jerin)
- Add parameter to define vector size in power of 2. (Jerin)
- Redo patch series w/o breaking ABI till the last patch.(David)
- Add deprication notice to announce ABI break in 21.11.(David)
- Add vector limits validation to app/test-eventdev.

Pavan Nikhilesh (8):
  eventdev: introduce event vector capability
  eventdev: introduce event vector Rx capability
  eventdev: introduce event vector Tx capability
  eventdev: add Rx adapter event vector support
  eventdev: add Tx adapter event vector support
  app/eventdev: add event vector mode in pipeline test
  doc: announce event Rx adapter config changes
  eventdev: simplify Rx adapter event vector config

 app/test-eventdev/evt_common.h                |   4 +
 app/test-eventdev/evt_options.c               |  52 +++
 app/test-eventdev/evt_options.h               |   4 +
 app/test-eventdev/test_pipeline_atq.c         | 310 +++++++++++++++--
 app/test-eventdev/test_pipeline_common.c      | 105 +++++-
 app/test-eventdev/test_pipeline_common.h      |  18 +
 app/test-eventdev/test_pipeline_queue.c       | 320 ++++++++++++++++--
 .../prog_guide/event_ethernet_rx_adapter.rst  |  38 +++
 .../prog_guide/event_ethernet_tx_adapter.rst  |  12 +
 doc/guides/prog_guide/eventdev.rst            |  36 +-
 doc/guides/rel_notes/deprecation.rst          |   9 +
 doc/guides/rel_notes/release_21_05.rst        |   8 +
 doc/guides/tools/testeventdev.rst             |  45 ++-
 lib/librte_eventdev/eventdev_pmd.h            |  31 +-
 .../rte_event_eth_rx_adapter.c                | 305 ++++++++++++++++-
 .../rte_event_eth_rx_adapter.h                |  78 +++++
 .../rte_event_eth_tx_adapter.c                |  66 +++-
 lib/librte_eventdev/rte_eventdev.c            |  53 ++-
 lib/librte_eventdev/rte_eventdev.h            | 114 ++++++-
 lib/librte_eventdev/version.map               |   4 +
 20 files changed, 1525 insertions(+), 87 deletions(-)

--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v6 1/8] eventdev: introduce event vector capability
  2021-03-24 19:28         ` [dpdk-dev] [PATCH v6 " pbhagavatula
@ 2021-03-24 19:28           ` pbhagavatula
  2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 2/8] eventdev: introduce event vector Rx capability pbhagavatula
                             ` (7 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-24 19:28 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce rte_event_vector datastructure which is capable of holding
multiple uintptr_t of the same flow thereby allowing applications
to vectorize their pipeline and reducing the complexity of pipelining
the events across multiple stages.
This approach also reduces the scheduling overhead on a event device.

Add a event vector mempool create handler to create mempools based on
the best mempool ops available on a given platform.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
Acked-by: Ray Kinsella <mdr@ashroe.eu>
---
 doc/guides/prog_guide/eventdev.rst     | 36 ++++++++++-
 doc/guides/rel_notes/release_21_05.rst |  8 +++
 lib/librte_eventdev/rte_eventdev.c     | 42 +++++++++++++
 lib/librte_eventdev/rte_eventdev.h     | 82 +++++++++++++++++++++++++-
 lib/librte_eventdev/version.map        |  3 +
 5 files changed, 168 insertions(+), 3 deletions(-)

diff --git a/doc/guides/prog_guide/eventdev.rst b/doc/guides/prog_guide/eventdev.rst
index ccde086f6..fda9c3743 100644
--- a/doc/guides/prog_guide/eventdev.rst
+++ b/doc/guides/prog_guide/eventdev.rst
@@ -63,13 +63,45 @@ the actual event being scheduled is. The payload is a union of the following:
 * ``uint64_t u64``
 * ``void *event_ptr``
 * ``struct rte_mbuf *mbuf``
+* ``struct rte_event_vector *vec``
 
-These three items in a union occupy the same 64 bits at the end of the rte_event
+These four items in a union occupy the same 64 bits at the end of the rte_event
 structure. The application can utilize the 64 bits directly by accessing the
-u64 variable, while the event_ptr and mbuf are provided as convenience
+u64 variable, while the event_ptr, mbuf, vec are provided as a convenience
 variables.  For example the mbuf pointer in the union can used to schedule a
 DPDK packet.
 
+Event Vector
+~~~~~~~~~~~~
+
+The rte_event_vector struct contains a vector of elements defined by the event
+type specified in the ``rte_event``. The event_vector structure contains the
+following data:
+
+* ``nb_elem`` - The number of elements held within the vector.
+
+Similar to ``rte_event`` the payload of event vector is also a union, allowing
+flexibility in what the actual vector is.
+
+* ``struct rte_mbuf *mbufs[0]`` - An array of mbufs.
+* ``void *ptrs[0]`` - An array of pointers.
+* ``uint64_t *u64s[0]`` - An array of uint64_t elements.
+
+The size of the event vector is related to the total number of elements it is
+configured to hold, this is achieved by making `rte_event_vector` a variable
+length structure.
+A helper function is provided to create a mempool that holds event vector, which
+takes name of the pool, total number of required ``rte_event_vector``,
+cache size, number of elements in each ``rte_event_vector`` and socket id.
+
+.. code-block:: c
+
+        rte_event_vector_pool_create("vector_pool", nb_event_vectors, cache_sz,
+                                     nb_elements_per_vector, socket_id);
+
+The function ``rte_event_vector_pool_create`` creates mempool with the best
+platform mempool ops.
+
 Queues
 ~~~~~~
 
diff --git a/doc/guides/rel_notes/release_21_05.rst b/doc/guides/rel_notes/release_21_05.rst
index 8a601e0a7..aeb56533a 100644
--- a/doc/guides/rel_notes/release_21_05.rst
+++ b/doc/guides/rel_notes/release_21_05.rst
@@ -106,6 +106,14 @@ New Features
   * Added support for periodic timer mode in eventdev timer adapter.
   * Added support for periodic timer mode in octeontx2 event device driver.
 
+* **Add Event device vector capability.**
+
+  * Added ``rte_event_vector`` data structure which is capable of holding
+    multiple ``uintptr_t`` of the same flow thereby allowing applications
+    to vectorize their pipelines and also reduce the complexity of pipelining
+    the events across multiple stages.
+  * This also reduces the scheduling overhead on a event device.
+
 
 Removed Items
 -------------
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index b57363f80..be0499c52 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -1266,6 +1266,48 @@ int rte_event_dev_selftest(uint8_t dev_id)
 	return -ENOTSUP;
 }
 
+struct rte_mempool *
+rte_event_vector_pool_create(const char *name, unsigned int n,
+			     unsigned int cache_size, uint16_t nb_elem,
+			     int socket_id)
+{
+	const char *mp_ops_name;
+	struct rte_mempool *mp;
+	unsigned int elt_sz;
+	int ret;
+
+	if (!nb_elem) {
+		RTE_LOG(ERR, EVENTDEV,
+			"Invalid number of elements=%d requested\n", nb_elem);
+		rte_errno = EINVAL;
+		return NULL;
+	}
+
+	elt_sz =
+		sizeof(struct rte_event_vector) + (nb_elem * sizeof(uintptr_t));
+	mp = rte_mempool_create_empty(name, n, elt_sz, cache_size, 0, socket_id,
+				      0);
+	if (mp == NULL)
+		return NULL;
+
+	mp_ops_name = rte_mbuf_best_mempool_ops();
+	ret = rte_mempool_set_ops_byname(mp, mp_ops_name, NULL);
+	if (ret != 0) {
+		RTE_LOG(ERR, EVENTDEV, "error setting mempool handler\n");
+		goto err;
+	}
+
+	ret = rte_mempool_populate_default(mp);
+	if (ret < 0)
+		goto err;
+
+	return mp;
+err:
+	rte_mempool_free(mp);
+	rte_errno = -ret;
+	return NULL;
+}
+
 int
 rte_event_dev_start(uint8_t dev_id)
 {
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index 9fc39e9ca..dee468ed0 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -212,8 +212,10 @@ extern "C" {
 
 #include <rte_common.h>
 #include <rte_config.h>
-#include <rte_memory.h>
 #include <rte_errno.h>
+#include <rte_mbuf_pool_ops.h>
+#include <rte_memory.h>
+#include <rte_mempool.h>
 
 #include "rte_eventdev_trace_fp.h"
 
@@ -913,6 +915,31 @@ rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
 int
 rte_event_dev_close(uint8_t dev_id);
 
+/**
+ * Event vector structure.
+ */
+struct rte_event_vector {
+	uint64_t nb_elem : 16;
+	/**< Number of elements in this event vector. */
+	uint64_t rsvd : 48;
+	/**< Reserved for future use */
+	uint64_t impl_opaque;
+	/**< Implementation specific opaque value.
+	 * An implementation may use this field to hold implementation specific
+	 * value to share between dequeue and enqueue operation.
+	 * The application should not modify this field.
+	 */
+	union {
+		struct rte_mbuf *mbufs[0];
+		void *ptrs[0];
+		uint64_t *u64s[0];
+	} __rte_aligned(16);
+	/**< Start of the vector array union. Depending upon the event type the
+	 * vector array can be an array of mbufs or pointers or opaque u64
+	 * values.
+	 */
+};
+
 /* Scheduler type definitions */
 #define RTE_SCHED_TYPE_ORDERED          0
 /**< Ordered scheduling
@@ -986,6 +1013,21 @@ rte_event_dev_close(uint8_t dev_id);
  */
 #define RTE_EVENT_TYPE_ETH_RX_ADAPTER   0x4
 /**< The event generated from event eth Rx adapter */
+#define RTE_EVENT_TYPE_VECTOR           0x8
+/**< Indicates that event is a vector.
+ * All vector event types should be a logical OR of EVENT_TYPE_VECTOR.
+ * This simplifies the pipeline design as one can split processing the events
+ * between vector events and normal event across event types.
+ * Example:
+ *	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
+ *		// Classify and handle vector event.
+ *	} else {
+ *		// Classify and handle event.
+ *	}
+ */
+#define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
+/**< The event vector generated from cpu for pipelining. */
+
 #define RTE_EVENT_TYPE_MAX              0x10
 /**< Maximum number of event types */
 
@@ -1108,6 +1150,8 @@ struct rte_event {
 		/**< Opaque event pointer */
 		struct rte_mbuf *mbuf;
 		/**< mbuf pointer if dequeued event is associated with mbuf */
+		struct rte_event_vector *vec;
+		/**< Event vector pointer. */
 	};
 };
 
@@ -2026,6 +2070,42 @@ rte_event_dev_xstats_reset(uint8_t dev_id,
  */
 int rte_event_dev_selftest(uint8_t dev_id);
 
+/**
+ * Get the memory required per event vector based on the number of elements per
+ * vector.
+ * This should be used to create the mempool that holds the event vectors.
+ *
+ * @param name
+ *   The name of the vector pool.
+ * @param n
+ *   The number of elements in the mbuf pool.
+ * @param cache_size
+ *   Size of the per-core object cache. See rte_mempool_create() for
+ *   details.
+ * @param nb_elem
+ *   The number of elements that a single event vector should be able to hold.
+ * @param socket_id
+ *   The socket identifier where the memory should be allocated. The
+ *   value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
+ *   reserved zone
+ *
+ * @return
+ *   The pointer to the newly allocated mempool, on success. NULL on error
+ *   with rte_errno set appropriately. Possible rte_errno values include:
+ *    - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
+ *    - E_RTE_SECONDARY - function was called from a secondary process instance
+ *    - EINVAL - cache size provided is too large, or priv_size is not aligned.
+ *    - ENOSPC - the maximum number of memzones has already been allocated
+ *    - EEXIST - a memzone with the same name already exists
+ *    - ENOMEM - no appropriate memory area found in which to create memzone
+ *    - ENAMETOOLONG - mempool name requested is too long.
+ */
+__rte_experimental
+struct rte_mempool *
+rte_event_vector_pool_create(const char *name, unsigned int n,
+			     unsigned int cache_size, uint16_t nb_elem,
+			     int socket_id);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index 3e5c09cfd..a070ef56e 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -138,6 +138,9 @@ EXPERIMENTAL {
 	__rte_eventdev_trace_port_setup;
 	# added in 20.11
 	rte_event_pmd_pci_probe_named;
+
+	#added in 21.05
+	rte_event_vector_pool_create;
 };
 
 INTERNAL {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v6 2/8] eventdev: introduce event vector Rx capability
  2021-03-24 19:28         ` [dpdk-dev] [PATCH v6 " pbhagavatula
  2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 1/8] eventdev: introduce event vector capability pbhagavatula
@ 2021-03-24 19:28           ` pbhagavatula
  2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 3/8] eventdev: introduce event vector Tx capability pbhagavatula
                             ` (6 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-24 19:28 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce event ethernet Rx adapter event vector capability.

If an event eth Rx adapter has the capability of
RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR then a given Rx queue
can be configured to enable event vectorization by passing the
flag RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR to
rte_event_eth_rx_adapter_queue_conf::rx_queue_flags while configuring
Rx adapter through rte_event_eth_rx_adapter_queue_add().

The max vector size, vector timeout define the vector size and
mempool used for allocating vector event are configured through
rte_event_eth_rx_adapter_queue_add. The element size of the element
in the vector pool should be equal to
    sizeof(struct rte_event_vector) + (vector_sz * sizeof(uintptr_t))

Application can use `rte_event_vector_pool_create` to create the
vector mempool used for
rte_event_eth_rx_adapter_queue_conf::vector_mp.

The Rx adapter would be responsible for vectorizing the mbufs
based on the flow, the vector limits configured by the application
and add the vector event of mbufs to the event queue set via
rte_event_eth_rx_adapter_queue_conf::ev::queue_id.
It should also mark rte_event_vector::union_valid and fill
rte_event_vector::port, rte_event_vector::queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
Acked-by: Ray Kinsella <mdr@ashroe.eu>
---
 .../prog_guide/event_ethernet_rx_adapter.rst  |  38 ++++++
 lib/librte_eventdev/eventdev_pmd.h            |  53 ++++++++
 .../rte_event_eth_rx_adapter.c                | 114 ++++++++++++++++++
 .../rte_event_eth_rx_adapter.h                | 105 ++++++++++++++++
 lib/librte_eventdev/rte_eventdev.h            |  30 ++++-
 lib/librte_eventdev/version.map               |   2 +
 6 files changed, 340 insertions(+), 2 deletions(-)

diff --git a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
index cb44ce0e4..5eefef355 100644
--- a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
@@ -186,3 +186,41 @@ the event buffer fill level is low. The
 ``rte_event_eth_rx_adapter_cb_register()`` function allow the application
 to register a callback that selects which packets to enqueue to the event
 device.
+
+Rx event vectorization
+~~~~~~~~~~~~~~~~~~~~~~
+
+The event devices, ethernet device pairs which support the capability
+``RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR`` can aggregate packets based on
+flow characteristics and generate a ``rte_event`` containing ``rte_event_vector``
+whose event type is either ``RTE_EVENT_TYPE_ETHDEV_VECTOR`` or
+``RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR``.
+The aggregation size and timeout are configurable at a queue level and the
+maximum, minimum vector sizes and timeouts vary based on the device capability
+and can be queried using ``rte_event_eth_rx_adapter_vector_limits_get``.
+The Rx adapter additionally might include useful data such as ethernet device
+port and queue identifier in the ``rte_event_vector::port`` and
+``rte_event_vector::queue`` and mark ``rte_event_vector::attr_valid`` as true.
+
+A loop processing ``rte_event_vector`` containing mbufs is shown below.
+
+.. code-block:: c
+
+        event = rte_event_dequeue_burst(event_dev, event_port, &event,
+                                        1, 0);
+        if (!event)
+                continue;
+
+        switch (ev.event_type) {
+        case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
+        case RTE_EVENT_TYPE_ETHDEV_VECTOR:
+                struct rte_mbufs **mbufs;
+
+                mbufs = (struct rte_mbufs **)ev[i].vec->mbufs;
+                for (i = 0; i < ev.vec->nb_elem; i++) {
+                        /* Process each mbuf. */
+                }
+        break;
+        case ...
+        ...
+        }
diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 7eb9a7739..9297f1433 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -645,6 +645,53 @@ typedef int (*eventdev_eth_rx_adapter_stats_reset)
  */
 typedef int (*eventdev_selftest)(void);
 
+struct rte_event_eth_rx_adapter_vector_limits;
+/**
+ * Get event vector limits for a given event, ethernet device pair.
+ *
+ * @param dev
+ *   Event device pointer
+ *
+ * @param eth_dev
+ *   Ethernet device pointer
+ *
+ * @param[out] limits
+ *   Pointer to the limits structure to be filled.
+ *
+ * @return
+ *   - 0: Success.
+ *   - <0: Error code returned by the driver function.
+ */
+typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
+	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
+	struct rte_event_eth_rx_adapter_vector_limits *limits);
+
+struct rte_event_eth_rx_adapter_event_vector_config;
+/**
+ * Enable event vector on an given Rx queue of a ethernet devices belonging to
+ * the Rx adapter.
+ *
+ * @param dev
+ *   Event device pointer
+ *
+ * @param eth_dev
+ *   Ethernet device pointer
+ *
+ * @param rx_queue_id
+ *   The Rx queue identifier
+ *
+ * @param config
+ *   Pointer to the event vector configuration structure.
+ *
+ * @return
+ *   - 0: Success.
+ *   - <0: Error code returned by the driver function.
+ */
+typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
+	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
+	int32_t rx_queue_id,
+	const struct rte_event_eth_rx_adapter_event_vector_config *config);
+
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;
 
@@ -1067,6 +1114,12 @@ struct rte_eventdev_ops {
 	/**< Get ethernet Rx stats */
 	eventdev_eth_rx_adapter_stats_reset eth_rx_adapter_stats_reset;
 	/**< Reset ethernet Rx stats */
+	eventdev_eth_rx_adapter_vector_limits_get_t
+		eth_rx_adapter_vector_limits_get;
+	/**< Get event vector limits for the Rx adapter */
+	eventdev_eth_rx_adapter_event_vector_config_t
+		eth_rx_adapter_event_vector_config;
+	/**< Configure Rx adapter with event vector */
 
 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index d8c635e99..ac8ba5bf0 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -2263,6 +2263,120 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }
 
+int
+rte_event_eth_rx_adapter_queue_event_vector_config(
+	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
+	struct rte_event_eth_rx_adapter_event_vector_config *config)
+{
+	struct rte_event_eth_rx_adapter_vector_limits limits;
+	struct rte_event_eth_rx_adapter *rx_adapter;
+	struct rte_eventdev *dev;
+	uint32_t cap;
+	int ret;
+
+	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
+
+	rx_adapter = rxa_id_to_adapter(id);
+	if ((rx_adapter == NULL) || (config == NULL))
+		return -EINVAL;
+
+	dev = &rte_eventdevs[rx_adapter->eventdev_id];
+	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
+						eth_dev_id, &cap);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
+				 "eth port %" PRIu16,
+				 id, eth_dev_id);
+		return ret;
+	}
+
+	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
+		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
+	ret = rte_event_eth_rx_adapter_vector_limits_get(
+		rx_adapter->eventdev_id, eth_dev_id, &limits);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
+				 "eth port %" PRIu16,
+				 rx_adapter->eventdev_id, eth_dev_id);
+		return ret;
+	}
+
+	if (config->vector_sz < limits.min_sz ||
+	    config->vector_sz > limits.max_sz ||
+	    config->vector_timeout_ns < limits.min_timeout_ns ||
+	    config->vector_timeout_ns > limits.max_timeout_ns ||
+	    config->vector_mp == NULL) {
+		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+	if (config->vector_mp->elt_size <
+	    (sizeof(struct rte_event_vector) +
+	     (sizeof(uintptr_t) * config->vector_sz))) {
+		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
+	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
+		RTE_FUNC_PTR_OR_ERR_RET(
+			*dev->dev_ops->eth_rx_adapter_event_vector_config,
+			-ENOTSUP);
+		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
+			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
+	} else {
+		ret = -ENOTSUP;
+	}
+
+	return ret;
+}
+
+int
+rte_event_eth_rx_adapter_vector_limits_get(
+	uint8_t dev_id, uint16_t eth_port_id,
+	struct rte_event_eth_rx_adapter_vector_limits *limits)
+{
+	struct rte_eventdev *dev;
+	uint32_t cap;
+	int ret;
+
+	RTE_EVENTDEV_VALID_DEVID_OR_ERR_RET(dev_id, -EINVAL);
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_port_id, -EINVAL);
+
+	if (limits == NULL)
+		return -EINVAL;
+
+	dev = &rte_eventdevs[dev_id];
+
+	ret = rte_event_eth_rx_adapter_caps_get(dev_id, eth_port_id, &cap);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
+				 "eth port %" PRIu16,
+				 dev_id, eth_port_id);
+		return ret;
+	}
+
+	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
+		RTE_FUNC_PTR_OR_ERR_RET(
+			*dev->dev_ops->eth_rx_adapter_vector_limits_get,
+			-ENOTSUP);
+		ret = dev->dev_ops->eth_rx_adapter_vector_limits_get(
+			dev, &rte_eth_devices[eth_port_id], limits);
+	} else {
+		ret = -ENOTSUP;
+	}
+
+	return ret;
+}
+
 int
 rte_event_eth_rx_adapter_start(uint8_t id)
 {
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
index 21bb1e54c..7407cde00 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
@@ -92,6 +92,10 @@ extern "C" {
 /**< This flag indicates the flow identifier is valid
  * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
  */
+#define RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR	0x2
+/**< This flag indicates that mbufs arriving on the queue need to be vectorized
+ * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+ */
 
 /**
  * Adapter configuration structure that the adapter configuration callback
@@ -169,6 +173,36 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 */
 };
 
+struct rte_event_eth_rx_adapter_event_vector_config {
+	uint16_t vector_sz;
+	/**<
+	 * Indicates the maximum number for mbufs to combine and form a vector.
+	 * Should be within
+	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_sz
+	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_sz
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+	 */
+	uint64_t vector_timeout_ns;
+	/**<
+	 * Indicates the maximum number of nanoseconds to wait for receiving
+	 * mbufs. Should be within vectorization limits of the
+	 * adapter
+	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_ns
+	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_ns
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+	 */
+	struct rte_mempool *vector_mp;
+	/**<
+	 * Indicates the mempool that should be used for allocating
+	 * rte_event_vector container.
+	 * Should be created by using `rte_event_vector_pool_create`.
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags.
+	 */
+};
+
 /**
  * A structure used to retrieve statistics for an eth rx adapter instance.
  */
@@ -199,6 +233,32 @@ struct rte_event_eth_rx_adapter_stats {
 	/**< Received packet count for interrupt mode Rx queues */
 };
 
+/**
+ * A structure used to retrieve eth rx adapter vector limits.
+ */
+struct rte_event_eth_rx_adapter_vector_limits {
+	uint16_t min_sz;
+	/**< Minimum vector limit configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_sz
+	 */
+	uint16_t max_sz;
+	/**< Maximum vector limit configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_sz
+	 */
+	uint8_t log2_sz;
+	/**< True if the size configured should be in log2.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_sz
+	 */
+	uint64_t min_timeout_ns;
+	/**< Minimum vector timeout configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_timeout_ns
+	 */
+	uint64_t max_timeout_ns;
+	/**< Maximum vector timeout configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_timeout_ns
+	 */
+};
+
 /**
  *
  * Callback function invoked by the SW adapter before it continues
@@ -467,6 +527,51 @@ int rte_event_eth_rx_adapter_cb_register(uint8_t id, uint16_t eth_dev_id,
 					 rte_event_eth_rx_adapter_cb_fn cb_fn,
 					 void *cb_arg);
 
+/**
+ * Retrieve vector limits for a given event dev and eth dev pair.
+ * @see rte_event_eth_rx_adapter_vector_limits
+ *
+ * @param dev_id
+ *  Event device identifier.
+ * @param eth_port_id
+ *  Port identifier of the ethernet device.
+ * @param [out] limits
+ *  A pointer to rte_event_eth_rx_adapter_vector_limits structure that has to
+ * be filled.
+ *
+ * @return
+ *  - 0: Success.
+ *  - <0: Error code on failure.
+ */
+__rte_experimental
+int rte_event_eth_rx_adapter_vector_limits_get(
+	uint8_t dev_id, uint16_t eth_port_id,
+	struct rte_event_eth_rx_adapter_vector_limits *limits);
+
+/**
+ * Configure event vectorization for a given ethernet device queue, that has
+ * been added to a event eth Rx adapter.
+ *
+ * @param id
+ *  The identifier of the ethernet Rx event adapter.
+ *
+ * @param eth_dev_id
+ *  The identifier of the ethernet device.
+ *
+ * @param rx_queue_id
+ *  Ethernet device receive queue index.
+ *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
+ *  are configured with event vectorization.
+ *
+ * @return
+ *  - 0: Success, Receive queue configured correctly.
+ *  - <0: Error code on failure.
+ */
+__rte_experimental
+int rte_event_eth_rx_adapter_queue_event_vector_config(
+	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
+	struct rte_event_eth_rx_adapter_event_vector_config *config);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index dee468ed0..936c97cc2 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -919,10 +919,28 @@ rte_event_dev_close(uint8_t dev_id);
  * Event vector structure.
  */
 struct rte_event_vector {
-	uint64_t nb_elem : 16;
+	uint16_t nb_elem;
 	/**< Number of elements in this event vector. */
-	uint64_t rsvd : 48;
+	uint16_t rsvd : 15;
 	/**< Reserved for future use */
+	uint16_t attr_valid : 1;
+	/**< Indicates that the below union attributes have valid information.
+	 */
+	union {
+		/* Used by Rx adapter.
+		 * Indicates that all the elements in this vector belong to the
+		 * same port and queue pair when originating from Rx adapter,
+		 * valid only when event type is ETHDEV_VECTOR or
+		 * ETH_RX_ADAPTER_VECTOR.
+		 */
+		struct {
+			uint16_t port;
+			/* Ethernet device port id. */
+			uint16_t queue;
+			/* Ethernet device queue id. */
+		};
+	};
+	/**< Union to hold common attributes of the vector array. */
 	uint64_t impl_opaque;
 	/**< Implementation specific opaque value.
 	 * An implementation may use this field to hold implementation specific
@@ -1025,8 +1043,14 @@ struct rte_event_vector {
  *		// Classify and handle event.
  *	}
  */
+#define RTE_EVENT_TYPE_ETHDEV_VECTOR                                           \
+	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETHDEV)
+/**< The event vector generated from ethdev subsystem */
 #define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
 /**< The event vector generated from cpu for pipelining. */
+#define RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR                                   \
+	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETH_RX_ADAPTER)
+/**< The event vector generated from eth Rx adapter. */
 
 #define RTE_EVENT_TYPE_MAX              0x10
 /**< Maximum number of event types */
@@ -1171,6 +1195,8 @@ struct rte_event {
  * @see struct rte_event_eth_rx_adapter_queue_conf::ev
  * @see struct rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
  */
+#define RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR	0x8
+/**< Adapter supports event vectorization per ethdev. */
 
 /**
  * Retrieve the event device's ethdev Rx adapter capabilities for the
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index a070ef56e..902df0ae3 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -141,6 +141,8 @@ EXPERIMENTAL {
 
 	#added in 21.05
 	rte_event_vector_pool_create;
+	rte_event_eth_rx_adapter_vector_limits_get;
+	rte_event_eth_rx_adapter_queue_event_vector_config;
 };
 
 INTERNAL {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v6 3/8] eventdev: introduce event vector Tx capability
  2021-03-24 19:28         ` [dpdk-dev] [PATCH v6 " pbhagavatula
  2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 1/8] eventdev: introduce event vector capability pbhagavatula
  2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 2/8] eventdev: introduce event vector Rx capability pbhagavatula
@ 2021-03-24 19:28           ` pbhagavatula
  2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 4/8] eventdev: add Rx adapter event vector support pbhagavatula
                             ` (5 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-24 19:28 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce event vector transmit capability for event eth
tx adapter.

The capability indicates that the Tx adapter is capable of
transmitting event vectors.
When rte_event_vector::union_valid is set, the Tx adapter should
transmit all the packets to the rte_event_vector::port using the
rte_event_vector::queue.
If rte_event_vector::union_valid is not set then the Tx adapter
should peek into each mbuf to get the destination port and queue
pair.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
---
 doc/guides/prog_guide/event_ethernet_tx_adapter.rst | 12 ++++++++++++
 lib/librte_eventdev/rte_eventdev.h                  |  8 +++++++-
 2 files changed, 19 insertions(+), 1 deletion(-)

diff --git a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
index a8c13e136..87277dcaf 100644
--- a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
@@ -164,3 +164,15 @@ The  ``rte_event_eth_tx_adapter_stats_get()`` function reports counters defined
 in struct ``rte_event_eth_tx_adapter_stats``. The counter values are the sum of
 the counts from the eventdev PMD callback if the callback is supported, and
 the counts maintained by the service function, if one exists.
+
+Tx event vectorization
+~~~~~~~~~~~~~~~~~~~~~~
+
+The event device, ethernet device paris which support the capability
+``RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR`` can process event vector of mbufs.
+Additionally, application can provide a hint to that Tx adapter that all the
+mbufs are destined to the same ethernet port and queue by setting the bit
+``rte_event_vector::attr_valid`` and filling `rte_event_vector::port`` and
+``rte_event_vector::queue``.
+If ``rte_event_vector::attr_valid`` is not set then the Tx adapter should peek
+into each mbuf and transmit them to the requested ethernet port and queue pair.
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index 936c97cc2..5f1f544cc 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -927,11 +927,13 @@ struct rte_event_vector {
 	/**< Indicates that the below union attributes have valid information.
 	 */
 	union {
-		/* Used by Rx adapter.
+		/* Used by Rx/Tx adapter.
 		 * Indicates that all the elements in this vector belong to the
 		 * same port and queue pair when originating from Rx adapter,
 		 * valid only when event type is ETHDEV_VECTOR or
 		 * ETH_RX_ADAPTER_VECTOR.
+		 * Can also be used to indicate the Tx adapter the destination
+		 * port and queue of the mbufs in the vector
 		 */
 		struct {
 			uint16_t port;
@@ -1296,6 +1298,10 @@ rte_event_crypto_adapter_caps_get(uint8_t dev_id, uint8_t cdev_id,
 #define RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT	0x1
 /**< This flag is sent when the PMD supports a packet transmit callback
  */
+#define RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR	0x2
+/**< Indicates that the Tx adapter is capable of handling event vector of
+ * mbufs.
+ */
 
 /**
  * Retrieve the event device's eth Tx adapter capabilities
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v6 4/8] eventdev: add Rx adapter event vector support
  2021-03-24 19:28         ` [dpdk-dev] [PATCH v6 " pbhagavatula
                             ` (2 preceding siblings ...)
  2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 3/8] eventdev: introduce event vector Tx capability pbhagavatula
@ 2021-03-24 19:28           ` pbhagavatula
  2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 5/8] eventdev: add Tx " pbhagavatula
                             ` (4 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-24 19:28 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support for event eth Rx adapter, the implementation
creates vector flows based on port and queue identifier of the received
mbufs.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 lib/librte_eventdev/eventdev_pmd.h            |   7 +-
 .../rte_event_eth_rx_adapter.c                | 257 ++++++++++++++++--
 lib/librte_eventdev/rte_eventdev.c            |   6 +-
 3 files changed, 250 insertions(+), 20 deletions(-)

diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 9297f1433..0f724ac85 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -69,9 +69,10 @@ extern "C" {
 	} \
 } while (0)
 
-#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP \
-		((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) | \
-			(RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ))
+#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP                                        \
+	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |                     \
+	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) |                         \
+	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR))
 
 #define RTE_EVENT_CRYPTO_ADAPTER_SW_CAP \
 		RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index ac8ba5bf0..c71990078 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -26,6 +26,10 @@
 #define BATCH_SIZE		32
 #define BLOCK_CNT_THRESHOLD	10
 #define ETH_EVENT_BUFFER_SIZE	(4*BATCH_SIZE)
+#define MAX_VECTOR_SIZE		1024
+#define MIN_VECTOR_SIZE		4
+#define MAX_VECTOR_NS		1E9
+#define MIN_VECTOR_NS		1E5
 
 #define ETH_RX_ADAPTER_SERVICE_NAME_LEN	32
 #define ETH_RX_ADAPTER_MEM_NAME_LEN	32
@@ -59,6 +63,20 @@ struct eth_rx_poll_entry {
 	uint16_t eth_rx_qid;
 };
 
+struct eth_rx_vector_data {
+	TAILQ_ENTRY(eth_rx_vector_data) next;
+	uint16_t port;
+	uint16_t queue;
+	uint16_t max_vector_count;
+	uint64_t event;
+	uint64_t ts;
+	uint64_t vector_timeout_ticks;
+	struct rte_mempool *vector_pool;
+	struct rte_event_vector *vector_ev;
+} __rte_cache_aligned;
+
+TAILQ_HEAD(eth_rx_vector_data_list, eth_rx_vector_data);
+
 /* Instance per adapter */
 struct rte_eth_event_enqueue_buffer {
 	/* Count of events in this buffer */
@@ -92,6 +110,14 @@ struct rte_event_eth_rx_adapter {
 	uint32_t wrr_pos;
 	/* Event burst buffer */
 	struct rte_eth_event_enqueue_buffer event_enqueue_buffer;
+	/* Vector enable flag */
+	uint8_t ena_vector;
+	/* Timestamp of previous vector expiry list traversal */
+	uint64_t prev_expiry_ts;
+	/* Minimum ticks to wait before traversing expiry list */
+	uint64_t vector_tmo_ticks;
+	/* vector list */
+	struct eth_rx_vector_data_list vector_list;
 	/* Per adapter stats */
 	struct rte_event_eth_rx_adapter_stats stats;
 	/* Block count, counts up to BLOCK_CNT_THRESHOLD */
@@ -198,9 +224,11 @@ struct eth_device_info {
 struct eth_rx_queue_info {
 	int queue_enabled;	/* True if added */
 	int intr_enabled;
+	uint8_t ena_vector;
 	uint16_t wt;		/* Polling weight */
 	uint32_t flow_id_mask;	/* Set to ~0 if app provides flow id else 0 */
 	uint64_t event;
+	struct eth_rx_vector_data vector_data;
 };
 
 static struct rte_event_eth_rx_adapter **event_eth_rx_adapter;
@@ -722,6 +750,9 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
 	    &rx_adapter->event_enqueue_buffer;
 	struct rte_event_eth_rx_adapter_stats *stats = &rx_adapter->stats;
 
+	if (!buf->count)
+		return 0;
+
 	uint16_t n = rte_event_enqueue_new_burst(rx_adapter->eventdev_id,
 					rx_adapter->event_port_id,
 					buf->events,
@@ -742,6 +773,72 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
 	return n;
 }
 
+static inline uint16_t
+rxa_create_event_vector(struct rte_event_eth_rx_adapter *rx_adapter,
+			struct eth_rx_queue_info *queue_info,
+			struct rte_eth_event_enqueue_buffer *buf,
+			struct rte_mbuf **mbufs, uint16_t num)
+{
+	struct rte_event *ev = &buf->events[buf->count];
+	struct eth_rx_vector_data *vec;
+	uint16_t filled, space, sz;
+
+	filled = 0;
+	vec = &queue_info->vector_data;
+	while (num) {
+		if (vec->vector_ev == NULL) {
+			if (rte_mempool_get(vec->vector_pool,
+					    (void **)&vec->vector_ev) < 0) {
+				rte_pktmbuf_free_bulk(mbufs, num);
+				return 0;
+			}
+			vec->vector_ev->nb_elem = 0;
+			vec->vector_ev->port = vec->port;
+			vec->vector_ev->queue = vec->queue;
+			vec->vector_ev->attr_valid = true;
+			TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
+		} else if (vec->vector_ev->nb_elem == vec->max_vector_count) {
+			/* Event ready. */
+			ev->event = vec->event;
+			ev->vec = vec->vector_ev;
+			ev++;
+			filled++;
+			vec->vector_ev = NULL;
+			TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+			if (rte_mempool_get(vec->vector_pool,
+					    (void **)&vec->vector_ev) < 0) {
+				rte_pktmbuf_free_bulk(mbufs, num);
+				return 0;
+			}
+			vec->vector_ev->nb_elem = 0;
+			vec->vector_ev->port = vec->port;
+			vec->vector_ev->queue = vec->queue;
+			vec->vector_ev->attr_valid = true;
+			TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
+		}
+
+		space = vec->max_vector_count - vec->vector_ev->nb_elem;
+		sz = num > space ? space : num;
+		memcpy(vec->vector_ev->mbufs + vec->vector_ev->nb_elem, mbufs,
+		       sizeof(void *) * sz);
+		vec->vector_ev->nb_elem += sz;
+		num -= sz;
+		mbufs += sz;
+		vec->ts = rte_rdtsc();
+	}
+
+	if (vec->vector_ev->nb_elem == vec->max_vector_count) {
+		ev->event = vec->event;
+		ev->vec = vec->vector_ev;
+		ev++;
+		filled++;
+		vec->vector_ev = NULL;
+		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+	}
+
+	return filled;
+}
+
 static inline void
 rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
@@ -770,25 +867,30 @@ rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
 	rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
 	do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
 
-	for (i = 0; i < num; i++) {
-		m = mbufs[i];
-
-		rss = do_rss ?
-			rxa_do_softrss(m, rx_adapter->rss_key_be) :
-			m->hash.rss;
-		ev->event = event;
-		ev->flow_id = (rss & ~flow_id_mask) |
-				(ev->flow_id & flow_id_mask);
-		ev->mbuf = m;
-		ev++;
+	if (!eth_rx_queue_info->ena_vector) {
+		for (i = 0; i < num; i++) {
+			m = mbufs[i];
+
+			rss = do_rss ? rxa_do_softrss(m, rx_adapter->rss_key_be)
+				     : m->hash.rss;
+			ev->event = event;
+			ev->flow_id = (rss & ~flow_id_mask) |
+				      (ev->flow_id & flow_id_mask);
+			ev->mbuf = m;
+			ev++;
+		}
+	} else {
+		num = rxa_create_event_vector(rx_adapter, eth_rx_queue_info,
+					      buf, mbufs, num);
 	}
 
-	if (dev_info->cb_fn) {
+	if (num && dev_info->cb_fn) {
 
 		dropped = 0;
 		nb_cb = dev_info->cb_fn(eth_dev_id, rx_queue_id,
-					ETH_EVENT_BUFFER_SIZE, buf->count, ev,
-					num, dev_info->cb_arg, &dropped);
+					ETH_EVENT_BUFFER_SIZE, buf->count,
+					&buf->events[buf->count], num,
+					dev_info->cb_arg, &dropped);
 		if (unlikely(nb_cb > num))
 			RTE_EDEV_LOG_ERR("Rx CB returned %d (> %d) events",
 				nb_cb, num);
@@ -1124,6 +1226,30 @@ rxa_poll(struct rte_event_eth_rx_adapter *rx_adapter)
 	return nb_rx;
 }
 
+static void
+rxa_vector_expire(struct eth_rx_vector_data *vec, void *arg)
+{
+	struct rte_event_eth_rx_adapter *rx_adapter = arg;
+	struct rte_eth_event_enqueue_buffer *buf =
+		&rx_adapter->event_enqueue_buffer;
+	struct rte_event *ev;
+
+	if (buf->count)
+		rxa_flush_event_buffer(rx_adapter);
+
+	if (vec->vector_ev->nb_elem == 0)
+		return;
+	ev = &buf->events[buf->count];
+
+	/* Event ready. */
+	ev->event = vec->event;
+	ev->vec = vec->vector_ev;
+	buf->count++;
+
+	vec->vector_ev = NULL;
+	vec->ts = 0;
+}
+
 static int
 rxa_service_func(void *args)
 {
@@ -1137,6 +1263,24 @@ rxa_service_func(void *args)
 		return 0;
 	}
 
+	if (rx_adapter->ena_vector) {
+		if ((rte_rdtsc() - rx_adapter->prev_expiry_ts) >=
+		    rx_adapter->vector_tmo_ticks) {
+			struct eth_rx_vector_data *vec;
+
+			TAILQ_FOREACH(vec, &rx_adapter->vector_list, next) {
+				uint64_t elapsed_time = rte_rdtsc() - vec->ts;
+
+				if (elapsed_time >= vec->vector_timeout_ticks) {
+					rxa_vector_expire(vec, rx_adapter);
+					TAILQ_REMOVE(&rx_adapter->vector_list,
+						     vec, next);
+				}
+			}
+			rx_adapter->prev_expiry_ts = rte_rdtsc();
+		}
+	}
+
 	stats = &rx_adapter->stats;
 	stats->rx_packets += rxa_intr_ring_dequeue(rx_adapter);
 	stats->rx_packets += rxa_poll(rx_adapter);
@@ -1640,6 +1784,28 @@ rxa_update_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }
 
+static void
+rxa_set_vector_data(struct eth_rx_queue_info *queue_info, uint16_t vector_count,
+		    uint64_t vector_ns, struct rte_mempool *mp, int32_t qid,
+		    uint16_t port_id)
+{
+#define NSEC2TICK(__ns, __freq) (((__ns) * (__freq)) / 1E9)
+	struct eth_rx_vector_data *vector_data;
+	uint32_t flow_id;
+
+	vector_data = &queue_info->vector_data;
+	vector_data->max_vector_count = vector_count;
+	vector_data->port = port_id;
+	vector_data->queue = qid;
+	vector_data->vector_pool = mp;
+	vector_data->vector_timeout_ticks =
+		NSEC2TICK(vector_ns, rte_get_timer_hz());
+	vector_data->ts = 0;
+	flow_id = queue_info->event & 0xFFFFF;
+	flow_id = flow_id == 0 ? (qid & 0xFF) | (port_id & 0xFFFF) : flow_id;
+	vector_data->event = (queue_info->event & ~0xFFFFF) | flow_id;
+}
+
 static void
 rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
 	struct eth_device_info *dev_info,
@@ -1741,6 +1907,44 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }
 
+static void
+rxa_sw_event_vector_configure(
+	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
+	int rx_queue_id,
+	const struct rte_event_eth_rx_adapter_event_vector_config *config)
+{
+	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
+	struct eth_rx_queue_info *queue_info;
+	struct rte_event *qi_ev;
+
+	if (rx_queue_id == -1) {
+		uint16_t nb_rx_queues;
+		uint16_t i;
+
+		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
+		for (i = 0; i < nb_rx_queues; i++)
+			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
+						      config);
+		return;
+	}
+
+	queue_info = &dev_info->rx_queue[rx_queue_id];
+	qi_ev = (struct rte_event *)&queue_info->event;
+	queue_info->ena_vector = 1;
+	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+	rxa_set_vector_data(queue_info, config->vector_sz,
+			    config->vector_timeout_ns, config->vector_mp,
+			    rx_queue_id, dev_info->dev->data->port_id);
+	rx_adapter->ena_vector = 1;
+	rx_adapter->vector_tmo_ticks =
+		rx_adapter->vector_tmo_ticks ?
+			      RTE_MIN(config->vector_timeout_ns << 1,
+				      rx_adapter->vector_tmo_ticks) :
+			      config->vector_timeout_ns << 1;
+	rx_adapter->prev_expiry_ts = 0;
+	TAILQ_INIT(&rx_adapter->vector_list);
+}
+
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -2081,6 +2285,15 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}
 
+	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
+	    (queue_conf->rx_queue_flags &
+	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
+		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
 		(rx_queue_id != -1)) {
 		RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
@@ -2143,6 +2356,17 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	return 0;
 }
 
+static int
+rxa_sw_vector_limits(struct rte_event_eth_rx_adapter_vector_limits *limits)
+{
+	limits->max_sz = MAX_VECTOR_SIZE;
+	limits->min_sz = MIN_VECTOR_SIZE;
+	limits->max_timeout_ns = MAX_VECTOR_NS;
+	limits->min_timeout_ns = MIN_VECTOR_NS;
+
+	return 0;
+}
+
 int
 rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 				int32_t rx_queue_id)
@@ -2333,7 +2557,8 @@ rte_event_eth_rx_adapter_queue_event_vector_config(
 		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
 			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
 	} else {
-		ret = -ENOTSUP;
+		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
+					      rx_queue_id, config);
 	}
 
 	return ret;
@@ -2371,7 +2596,7 @@ rte_event_eth_rx_adapter_vector_limits_get(
 		ret = dev->dev_ops->eth_rx_adapter_vector_limits_get(
 			dev, &rte_eth_devices[eth_port_id], limits);
 	} else {
-		ret = -ENOTSUP;
+		ret = rxa_sw_vector_limits(limits);
 	}
 
 	return ret;
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index be0499c52..62824654b 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -122,7 +122,11 @@ rte_event_eth_rx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
 
 	if (caps == NULL)
 		return -EINVAL;
-	*caps = 0;
+
+	if (dev->dev_ops->eth_rx_adapter_caps_get == NULL)
+		*caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
+	else
+		*caps = 0;
 
 	return dev->dev_ops->eth_rx_adapter_caps_get ?
 				(*dev->dev_ops->eth_rx_adapter_caps_get)(dev,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v6 5/8] eventdev: add Tx adapter event vector support
  2021-03-24 19:28         ` [dpdk-dev] [PATCH v6 " pbhagavatula
                             ` (3 preceding siblings ...)
  2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 4/8] eventdev: add Rx adapter event vector support pbhagavatula
@ 2021-03-24 19:28           ` pbhagavatula
  2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
                             ` (3 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-24 19:28 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support for event eth Tx adapter, the implementation
receives events from the single linked queue and based on
rte_event_vector::union_valid transmits the vector of mbufs to a given
port, queue pair.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 .../rte_event_eth_tx_adapter.c                | 66 ++++++++++++++++---
 lib/librte_eventdev/rte_eventdev.c            |  5 +-
 2 files changed, 60 insertions(+), 11 deletions(-)

diff --git a/lib/librte_eventdev/rte_event_eth_tx_adapter.c b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
index 5b4c42dcf..db260bfb6 100644
--- a/lib/librte_eventdev/rte_event_eth_tx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
@@ -510,6 +510,47 @@ txa_service_buffer_retry(struct rte_mbuf **pkts, uint16_t unsent,
 	stats->tx_dropped += unsent - sent;
 }
 
+static uint16_t
+txa_process_event_vector(struct txa_service_data *txa,
+			 struct rte_event_vector *vec)
+{
+	struct txa_service_queue_info *tqi;
+	uint16_t port, queue, nb_tx = 0;
+	struct rte_mbuf **mbufs;
+	int i;
+
+	mbufs = (struct rte_mbuf **)vec->mbufs;
+	if (vec->attr_valid) {
+		port = vec->port;
+		queue = vec->queue;
+		tqi = txa_service_queue(txa, port, queue);
+		if (unlikely(tqi == NULL || !tqi->added)) {
+			rte_pktmbuf_free_bulk(mbufs, vec->nb_elem);
+			rte_mempool_put(rte_mempool_from_obj(vec), vec);
+			return 0;
+		}
+		for (i = 0; i < vec->nb_elem; i++) {
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
+						   mbufs[i]);
+		}
+	} else {
+		for (i = 0; i < vec->nb_elem; i++) {
+			port = mbufs[i]->port;
+			queue = rte_event_eth_tx_adapter_txq_get(mbufs[i]);
+			tqi = txa_service_queue(txa, port, queue);
+			if (unlikely(tqi == NULL || !tqi->added)) {
+				rte_pktmbuf_free(mbufs[i]);
+				continue;
+			}
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
+						   mbufs[i]);
+		}
+	}
+	rte_mempool_put(rte_mempool_from_obj(vec), vec);
+
+	return nb_tx;
+}
+
 static void
 txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
 	uint32_t n)
@@ -522,22 +563,27 @@ txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
 
 	nb_tx = 0;
 	for (i = 0; i < n; i++) {
-		struct rte_mbuf *m;
 		uint16_t port;
 		uint16_t queue;
 		struct txa_service_queue_info *tqi;
 
-		m = ev[i].mbuf;
-		port = m->port;
-		queue = rte_event_eth_tx_adapter_txq_get(m);
+		if (!(ev[i].event_type & RTE_EVENT_TYPE_VECTOR)) {
+			struct rte_mbuf *m;
 
-		tqi = txa_service_queue(txa, port, queue);
-		if (unlikely(tqi == NULL || !tqi->added)) {
-			rte_pktmbuf_free(m);
-			continue;
-		}
+			m = ev[i].mbuf;
+			port = m->port;
+			queue = rte_event_eth_tx_adapter_txq_get(m);
 
-		nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
+			tqi = txa_service_queue(txa, port, queue);
+			if (unlikely(tqi == NULL || !tqi->added)) {
+				rte_pktmbuf_free(m);
+				continue;
+			}
+
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
+		} else {
+			nb_tx += txa_process_event_vector(txa, ev[i].vec);
+		}
 	}
 
 	stats->tx_packets += nb_tx;
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index 62824654b..c9bb5d227 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -196,7 +196,10 @@ rte_event_eth_tx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
 	if (caps == NULL)
 		return -EINVAL;
 
-	*caps = 0;
+	if (dev->dev_ops->eth_tx_adapter_caps_get == NULL)
+		*caps = RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR;
+	else
+		*caps = 0;
 
 	return dev->dev_ops->eth_tx_adapter_caps_get ?
 			(*dev->dev_ops->eth_tx_adapter_caps_get)(dev,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v6 6/8] app/eventdev: add event vector mode in pipeline test
  2021-03-24 19:28         ` [dpdk-dev] [PATCH v6 " pbhagavatula
                             ` (4 preceding siblings ...)
  2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 5/8] eventdev: add Tx " pbhagavatula
@ 2021-03-24 19:28           ` pbhagavatula
  2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 7/8] doc: announce event Rx adapter config changes pbhagavatula
                             ` (2 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-24 19:28 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support in pipeline tests. By default this mode
is disabled, it can be enabled by using the option --enable_vector.
example:
	dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
	--nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
	--wlcores=20-23  --enable_vector

Additional options to configure vector size and vector timeout are
also implemented and can be used by specifying --vector_size and
--vector_tmo_ns

This patch also adds a new option to set the number of Rx queues
configured per event eth rx adapter.
example:
	dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
	--nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
	--wlcores=20-23  --nb_eth_queues 4

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
---
 app/test-eventdev/evt_common.h           |   4 +
 app/test-eventdev/evt_options.c          |  52 ++++
 app/test-eventdev/evt_options.h          |   4 +
 app/test-eventdev/test_pipeline_atq.c    | 310 ++++++++++++++++++++--
 app/test-eventdev/test_pipeline_common.c | 113 +++++++-
 app/test-eventdev/test_pipeline_common.h |  18 ++
 app/test-eventdev/test_pipeline_queue.c  | 320 +++++++++++++++++++++--
 doc/guides/tools/testeventdev.rst        |  45 +++-
 8 files changed, 811 insertions(+), 55 deletions(-)

diff --git a/app/test-eventdev/evt_common.h b/app/test-eventdev/evt_common.h
index a1da1cf11..0e228258e 100644
--- a/app/test-eventdev/evt_common.h
+++ b/app/test-eventdev/evt_common.h
@@ -58,16 +58,20 @@ struct evt_options {
 	uint8_t sched_type_list[EVT_MAX_STAGES];
 	uint16_t mbuf_sz;
 	uint16_t wkr_deq_dep;
+	uint16_t vector_size;
+	uint16_t eth_queues;
 	uint32_t nb_flows;
 	uint32_t tx_first;
 	uint32_t max_pkt_sz;
 	uint32_t deq_tmo_nsec;
 	uint32_t q_priority:1;
 	uint32_t fwd_latency:1;
+	uint32_t ena_vector : 1;
 	uint64_t nb_pkts;
 	uint64_t nb_timers;
 	uint64_t expiry_nsec;
 	uint64_t max_tmo_nsec;
+	uint64_t vector_tmo_nsec;
 	uint64_t timer_tick_nsec;
 	uint64_t optm_timer_tick_nsec;
 	enum evt_prod_type prod_type;
diff --git a/app/test-eventdev/evt_options.c b/app/test-eventdev/evt_options.c
index 0d04ea9f8..0d5540574 100644
--- a/app/test-eventdev/evt_options.c
+++ b/app/test-eventdev/evt_options.c
@@ -34,6 +34,9 @@ evt_options_default(struct evt_options *opt)
 	opt->max_tmo_nsec = 1E5;  /* 100000ns ~100us */
 	opt->expiry_nsec = 1E4;   /* 10000ns ~10us */
 	opt->prod_type = EVT_PROD_TYPE_SYNT;
+	opt->eth_queues = 1;
+	opt->vector_size = 64;
+	opt->vector_tmo_nsec = 100E3;
 }
 
 typedef int (*option_parser_t)(struct evt_options *opt,
@@ -257,6 +260,43 @@ evt_parse_max_pkt_sz(struct evt_options *opt, const char *arg)
 	return ret;
 }
 
+static int
+evt_parse_ena_vector(struct evt_options *opt, const char *arg __rte_unused)
+{
+	opt->ena_vector = 1;
+	return 0;
+}
+
+static int
+evt_parse_vector_size(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint16(&(opt->vector_size), arg);
+
+	return ret;
+}
+
+static int
+evt_parse_vector_tmo_ns(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint64(&(opt->vector_tmo_nsec), arg);
+
+	return ret;
+}
+
+static int
+evt_parse_eth_queues(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint16(&(opt->eth_queues), arg);
+
+	return ret;
+}
+
 static void
 usage(char *program)
 {
@@ -289,6 +329,10 @@ usage(char *program)
 		"\t--expiry_nsec      : event timer expiry ns.\n"
 		"\t--mbuf_sz          : packet mbuf size.\n"
 		"\t--max_pkt_sz       : max packet size.\n"
+		"\t--nb_eth_queues    : number of ethernet Rx queues.\n"
+		"\t--enable_vector    : enable event vectorization.\n"
+		"\t--vector_size      : Max vector size.\n"
+		"\t--vector_tmo_ns    : Max vector timeout in nanoseconds\n"
 		);
 	printf("available tests:\n");
 	evt_test_dump_names();
@@ -360,6 +404,10 @@ static struct option lgopts[] = {
 	{ EVT_EXPIRY_NSEC,         1, 0, 0 },
 	{ EVT_MBUF_SZ,             1, 0, 0 },
 	{ EVT_MAX_PKT_SZ,          1, 0, 0 },
+	{ EVT_NB_ETH_QUEUES,       1, 0, 0 },
+	{ EVT_ENA_VECTOR,          0, 0, 0 },
+	{ EVT_VECTOR_SZ,           1, 0, 0 },
+	{ EVT_VECTOR_TMO,          1, 0, 0 },
 	{ EVT_HELP,                0, 0, 0 },
 	{ NULL,                    0, 0, 0 }
 };
@@ -394,6 +442,10 @@ evt_opts_parse_long(int opt_idx, struct evt_options *opt)
 		{ EVT_EXPIRY_NSEC, evt_parse_expiry_nsec},
 		{ EVT_MBUF_SZ, evt_parse_mbuf_sz},
 		{ EVT_MAX_PKT_SZ, evt_parse_max_pkt_sz},
+		{ EVT_NB_ETH_QUEUES, evt_parse_eth_queues},
+		{ EVT_ENA_VECTOR, evt_parse_ena_vector},
+		{ EVT_VECTOR_SZ, evt_parse_vector_size},
+		{ EVT_VECTOR_TMO, evt_parse_vector_tmo_ns},
 	};
 
 	for (i = 0; i < RTE_DIM(parsermap); i++) {
diff --git a/app/test-eventdev/evt_options.h b/app/test-eventdev/evt_options.h
index 748e54fae..1cea2a3e1 100644
--- a/app/test-eventdev/evt_options.h
+++ b/app/test-eventdev/evt_options.h
@@ -42,6 +42,10 @@
 #define EVT_EXPIRY_NSEC          ("expiry_nsec")
 #define EVT_MBUF_SZ              ("mbuf_sz")
 #define EVT_MAX_PKT_SZ           ("max_pkt_sz")
+#define EVT_NB_ETH_QUEUES        ("nb_eth_queues")
+#define EVT_ENA_VECTOR           ("enable_vector")
+#define EVT_VECTOR_SZ            ("vector_size")
+#define EVT_VECTOR_TMO           ("vector_tmo_ns")
 #define EVT_HELP                 ("help")
 
 void evt_options_default(struct evt_options *opt);
diff --git a/app/test-eventdev/test_pipeline_atq.c b/app/test-eventdev/test_pipeline_atq.c
index 0872b25b5..84dd4f44e 100644
--- a/app/test-eventdev/test_pipeline_atq.c
+++ b/app/test-eventdev/test_pipeline_atq.c
@@ -15,6 +15,8 @@ pipeline_atq_nb_event_queues(struct evt_options *opt)
 	return rte_eth_dev_count_avail();
 }
 
+typedef int (*pipeline_atq_worker_t)(void *arg);
+
 static __rte_noinline int
 pipeline_atq_worker_single_stage_tx(void *arg)
 {
@@ -113,6 +115,112 @@ pipeline_atq_worker_single_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_atq_worker_single_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+		vector_sz = ev.vec->nb_elem;
+		pipeline_event_tx_vector(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = ev.vec->nb_elem;
+		ev.queue_id = tx_queue[ev.vec->port];
+		ev.vec->queue = 0;
+		pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			vector_sz += ev[i].vec->nb_elem;
+			ev[i].vec->queue = 0;
+		}
+
+		pipeline_event_tx_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			ev[i].queue_id = tx_queue[ev[i].vec->port];
+			ev[i].vec->queue = 0;
+			vector_sz += ev[i].vec->nb_elem;
+			pipeline_fwd_event_vector(&ev[i],
+						  RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
 static __rte_noinline int
 pipeline_atq_worker_multi_stage_tx(void *arg)
 {
@@ -245,6 +353,147 @@ pipeline_atq_worker_multi_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.sub_event_type % nb_stages;
+
+		if (cq_id == last_queue) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+			continue;
+		}
+
+		ev.sub_event_type++;
+		pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.sub_event_type % nb_stages;
+
+		if (cq_id == last_queue) {
+			ev.queue_id = tx_queue[ev.vec->port];
+			ev.vec->queue = 0;
+			vector_sz = ev.vec->nb_elem;
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			pipeline_event_enqueue(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.sub_event_type++;
+			pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+			pipeline_event_enqueue(dev, port, &ev);
+		}
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].sub_event_type % nb_stages;
+
+			if (cq_id == last_queue) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+				continue;
+			}
+
+			ev[i].sub_event_type++;
+			pipeline_fwd_event_vector(&ev[i],
+						  sched_type_list[cq_id]);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].sub_event_type % nb_stages;
+
+			if (cq_id == last_queue) {
+				vector_sz = ev[i].vec->nb_elem;
+				ev[i].queue_id = tx_queue[ev[i].vec->port];
+				ev[i].vec->queue = 0;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].sub_event_type++;
+				pipeline_fwd_event_vector(
+					&ev[i], sched_type_list[cq_id]);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
 static int
 worker_wrapper(void *arg)
 {
@@ -253,27 +502,36 @@ worker_wrapper(void *arg)
 	const bool burst = evt_has_burst_mode(w->dev_id);
 	const bool internal_port = w->t->internal_port;
 	const uint8_t nb_stages = opt->nb_stages;
-	RTE_SET_USED(opt);
+	/*vector/burst/internal_port*/
+	const pipeline_atq_worker_t
+	pipeline_atq_worker_single_stage[2][2][2] = {
+		[0][0][0] = pipeline_atq_worker_single_stage_fwd,
+		[0][0][1] = pipeline_atq_worker_single_stage_tx,
+		[0][1][0] = pipeline_atq_worker_single_stage_burst_fwd,
+		[0][1][1] = pipeline_atq_worker_single_stage_burst_tx,
+		[1][0][0] = pipeline_atq_worker_single_stage_fwd_vector,
+		[1][0][1] = pipeline_atq_worker_single_stage_tx_vector,
+		[1][1][0] = pipeline_atq_worker_single_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_atq_worker_single_stage_burst_tx_vector,
+	};
+	const pipeline_atq_worker_t
+	pipeline_atq_worker_multi_stage[2][2][2] = {
+		[0][0][0] = pipeline_atq_worker_multi_stage_fwd,
+		[0][0][1] = pipeline_atq_worker_multi_stage_tx,
+		[0][1][0] = pipeline_atq_worker_multi_stage_burst_fwd,
+		[0][1][1] = pipeline_atq_worker_multi_stage_burst_tx,
+		[1][0][0] = pipeline_atq_worker_multi_stage_fwd_vector,
+		[1][0][1] = pipeline_atq_worker_multi_stage_tx_vector,
+		[1][1][0] = pipeline_atq_worker_multi_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_atq_worker_multi_stage_burst_tx_vector,
+	};
 
-	if (nb_stages == 1) {
-		if (!burst && internal_port)
-			return pipeline_atq_worker_single_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_atq_worker_single_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_atq_worker_single_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_atq_worker_single_stage_burst_fwd(arg);
-	} else {
-		if (!burst && internal_port)
-			return pipeline_atq_worker_multi_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_atq_worker_multi_stage_fwd(arg);
-		if (burst && internal_port)
-			return pipeline_atq_worker_multi_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_atq_worker_multi_stage_burst_fwd(arg);
-	}
+	if (nb_stages == 1)
+		return (pipeline_atq_worker_single_stage[opt->ena_vector][burst]
+							[internal_port])(arg);
+	else
+		return (pipeline_atq_worker_multi_stage[opt->ena_vector][burst]
+						       [internal_port])(arg);
 
 	rte_panic("invalid worker\n");
 }
@@ -290,7 +548,7 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
 	int ret;
 	int nb_ports;
 	int nb_queues;
-	uint8_t queue;
+	uint8_t queue, is_prod;
 	uint8_t tx_evqueue_id[RTE_MAX_ETHPORTS];
 	uint8_t queue_arr[RTE_EVENT_MAX_QUEUES_PER_DEV];
 	uint8_t nb_worker_queues = 0;
@@ -330,15 +588,19 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
 		q_conf.event_queue_cfg = RTE_EVENT_QUEUE_CFG_ALL_TYPES;
 
 		if (!t->internal_port) {
+			is_prod = false;
 			RTE_ETH_FOREACH_DEV(prod) {
 				if (queue == tx_evqueue_id[prod]) {
 					q_conf.event_queue_cfg =
 						RTE_EVENT_QUEUE_CFG_SINGLE_LINK;
-				} else {
-					queue_arr[nb_worker_queues] = queue;
-					nb_worker_queues++;
+					is_prod = true;
+					break;
 				}
 			}
+			if (!is_prod) {
+				queue_arr[nb_worker_queues] = queue;
+				nb_worker_queues++;
+			}
 		}
 
 		ret = rte_event_queue_setup(opt->dev_id, queue, &q_conf);
diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index b47d76743..d5ef90500 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -36,6 +36,12 @@ pipeline_opt_dump(struct evt_options *opt, uint8_t nb_queues)
 	evt_dump_queue_priority(opt);
 	evt_dump_sched_type_list(opt);
 	evt_dump_producer_type(opt);
+	evt_dump("nb_eth_rx_queues", "%d", opt->eth_queues);
+	evt_dump("event_vector", "%d", opt->ena_vector);
+	if (opt->ena_vector) {
+		evt_dump("vector_size", "%d", opt->vector_size);
+		evt_dump("vector_tmo_ns", "%" PRIu64 "", opt->vector_tmo_nsec);
+	}
 }
 
 static inline uint64_t
@@ -163,7 +169,7 @@ pipeline_opt_check(struct evt_options *opt, uint64_t nb_queues)
 int
 pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 {
-	uint16_t i;
+	uint16_t i, j;
 	int ret;
 	uint8_t nb_queues = 1;
 	struct test_pipeline *t = evt_test_priv(test);
@@ -210,6 +216,16 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 		if (!(caps & RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT))
 			t->internal_port = 0;
 
+		ret = rte_event_eth_rx_adapter_caps_get(opt->dev_id, i, &caps);
+		if (ret != 0) {
+			evt_err("failed to get event tx adapter[%d] caps", i);
+			return ret;
+		}
+
+		if (!(caps & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT))
+			local_port_conf.rxmode.offloads |=
+				DEV_RX_OFFLOAD_RSS_HASH;
+
 		ret = rte_eth_dev_info_get(i, &dev_info);
 		if (ret != 0) {
 			evt_err("Error during getting device (port %u) info: %s\n",
@@ -236,19 +252,22 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 				local_port_conf.rx_adv_conf.rss_conf.rss_hf);
 		}
 
-		if (rte_eth_dev_configure(i, nb_queues, nb_queues,
-					&local_port_conf)
-				< 0) {
+		if (rte_eth_dev_configure(i, opt->eth_queues, nb_queues,
+					  &local_port_conf) < 0) {
 			evt_err("Failed to configure eth port [%d]", i);
 			return -EINVAL;
 		}
 
-		if (rte_eth_rx_queue_setup(i, 0, NB_RX_DESC,
-				rte_socket_id(), &rx_conf, t->pool) < 0) {
-			evt_err("Failed to setup eth port [%d] rx_queue: %d.",
+		for (j = 0; j < opt->eth_queues; j++) {
+			if (rte_eth_rx_queue_setup(i, j, NB_RX_DESC,
+						   rte_socket_id(), &rx_conf,
+						   t->pool) < 0) {
+				evt_err("Failed to setup eth port [%d] rx_queue: %d.",
 					i, 0);
-			return -EINVAL;
+				return -EINVAL;
+			}
 		}
+
 		if (rte_eth_tx_queue_setup(i, 0, NB_TX_DESC,
 					rte_socket_id(), NULL) < 0) {
 			evt_err("Failed to setup eth port [%d] tx_queue: %d.",
@@ -310,12 +329,27 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 {
 	int ret = 0;
 	uint16_t prod;
+	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
+	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;
 
 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
 	queue_conf.ev.sched_type = opt->sched_type_list[0];
+	if (opt->ena_vector) {
+		unsigned int nb_elem = (opt->pool_sz / opt->vector_size) << 1;
+
+		nb_elem = nb_elem ? nb_elem : 1;
+		vector_pool = rte_event_vector_pool_create(
+			"vector_pool", nb_elem, 0, opt->vector_size,
+			opt->socket_id);
+		if (vector_pool == NULL) {
+			evt_err("failed to create event vector pool");
+			return -ENOMEM;
+		}
+	}
 	RTE_ETH_FOREACH_DEV(prod) {
+		struct rte_event_eth_rx_adapter_vector_limits limits;
 		uint32_t cap;
 
 		ret = rte_event_eth_rx_adapter_caps_get(opt->dev_id,
@@ -326,6 +360,50 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 					opt->dev_id);
 			return ret;
 		}
+
+		if (opt->ena_vector) {
+			memset(&limits, 0, sizeof(limits));
+			ret = rte_event_eth_rx_adapter_vector_limits_get(
+				opt->dev_id, prod, &limits);
+			if (ret) {
+				evt_err("failed to get vector limits");
+				return ret;
+			}
+
+			if (opt->vector_size < limits.min_sz ||
+			    opt->vector_size > limits.max_sz) {
+				evt_err("Vector size [%d] not within limits max[%d] min[%d]",
+					opt->vector_size, limits.min_sz,
+					limits.max_sz);
+				return -EINVAL;
+			}
+
+			if (limits.log2_sz &&
+			    !rte_is_power_of_2(opt->vector_size)) {
+				evt_err("Vector size [%d] not power of 2",
+					opt->vector_size);
+				return -EINVAL;
+			}
+
+			if (opt->vector_tmo_nsec > limits.max_timeout_ns ||
+			    opt->vector_tmo_nsec < limits.min_timeout_ns) {
+				evt_err("Vector timeout [%" PRIu64
+					"] not within limits max[%" PRIu64
+					"] min[%" PRIu64 "]",
+					opt->vector_tmo_nsec,
+					limits.max_timeout_ns,
+					limits.min_timeout_ns);
+				return -EINVAL;
+			}
+
+			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.rx_queue_flags |=
+				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+			} else {
+				evt_err("Rx adapter doesn't support event vector");
+				return -EINVAL;
+			}
+		}
 		queue_conf.ev.queue_id = prod * stride;
 		ret = rte_event_eth_rx_adapter_create(prod, opt->dev_id,
 				&prod_conf);
@@ -340,6 +418,17 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}
 
+		if (opt->ena_vector) {
+			vec_conf.vector_sz = opt->vector_size;
+			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
+			vec_conf.vector_mp = vector_pool;
+			if (rte_event_eth_rx_adapter_queue_event_vector_config(
+				    prod, prod, -1, &vec_conf) < 0) {
+				evt_err("Failed to configure event vectorization for Rx adapter");
+				return -EINVAL;
+			}
+		}
+
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;
 
@@ -378,6 +467,14 @@ pipeline_event_tx_adapter_setup(struct evt_options *opt,
 			return ret;
 		}
 
+		if (opt->ena_vector) {
+			if (!(cap &
+			      RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR)) {
+				evt_err("Tx adapter doesn't support event vector");
+				return -EINVAL;
+			}
+		}
+
 		ret = rte_event_eth_tx_adapter_create(consm, opt->dev_id,
 				&port_conf);
 		if (ret) {
diff --git a/app/test-eventdev/test_pipeline_common.h b/app/test-eventdev/test_pipeline_common.h
index 6e73c6ab2..800a90616 100644
--- a/app/test-eventdev/test_pipeline_common.h
+++ b/app/test-eventdev/test_pipeline_common.h
@@ -101,6 +101,14 @@ pipeline_fwd_event(struct rte_event *ev, uint8_t sched)
 	ev->sched_type = sched;
 }
 
+static __rte_always_inline void
+pipeline_fwd_event_vector(struct rte_event *ev, uint8_t sched)
+{
+	ev->event_type = RTE_EVENT_TYPE_CPU_VECTOR;
+	ev->op = RTE_EVENT_OP_FORWARD;
+	ev->sched_type = sched;
+}
+
 static __rte_always_inline void
 pipeline_event_tx(const uint8_t dev, const uint8_t port,
 		struct rte_event * const ev)
@@ -110,6 +118,16 @@ pipeline_event_tx(const uint8_t dev, const uint8_t port,
 		rte_pause();
 }
 
+static __rte_always_inline void
+pipeline_event_tx_vector(const uint8_t dev, const uint8_t port,
+			 struct rte_event *const ev)
+{
+	ev->vec->queue = 0;
+
+	while (!rte_event_eth_tx_adapter_enqueue(dev, port, ev, 1, 0))
+		rte_pause();
+}
+
 static __rte_always_inline void
 pipeline_event_tx_burst(const uint8_t dev, const uint8_t port,
 		struct rte_event *ev, const uint16_t nb_rx)
diff --git a/app/test-eventdev/test_pipeline_queue.c b/app/test-eventdev/test_pipeline_queue.c
index 9a9febb19..f6cc3e358 100644
--- a/app/test-eventdev/test_pipeline_queue.c
+++ b/app/test-eventdev/test_pipeline_queue.c
@@ -15,6 +15,8 @@ pipeline_queue_nb_event_queues(struct evt_options *opt)
 	return (eth_count * opt->nb_stages) + eth_count;
 }
 
+typedef int (*pipeline_queue_worker_t)(void *arg);
+
 static __rte_noinline int
 pipeline_queue_worker_single_stage_tx(void *arg)
 {
@@ -126,6 +128,125 @@ pipeline_queue_worker_single_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_queue_worker_single_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		if (ev.sched_type == RTE_SCHED_TYPE_ATOMIC) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.queue_id++;
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			pipeline_event_enqueue(dev, port, &ev);
+		}
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		ev.queue_id = tx_queue[ev.vec->port];
+		ev.vec->queue = 0;
+		vector_sz = ev.vec->nb_elem;
+		pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			if (ev[i].sched_type == RTE_SCHED_TYPE_ATOMIC) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].queue_id++;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			ev[i].queue_id = tx_queue[ev[i].vec->port];
+			ev[i].vec->queue = 0;
+			vector_sz += ev[i].vec->nb_elem;
+			pipeline_fwd_event_vector(&ev[i],
+						  RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
 
 static __rte_noinline int
 pipeline_queue_worker_multi_stage_tx(void *arg)
@@ -267,6 +388,151 @@ pipeline_queue_worker_multi_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.queue_id % nb_stages;
+
+		if (ev.queue_id == tx_queue[ev.vec->port]) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+			continue;
+		}
+
+		ev.queue_id++;
+		pipeline_fwd_event_vector(&ev, cq_id != last_queue
+						       ? sched_type_list[cq_id]
+						       : RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.queue_id % nb_stages;
+
+		if (cq_id == last_queue) {
+			vector_sz = ev.vec->nb_elem;
+			ev.queue_id = tx_queue[ev.vec->port];
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.queue_id++;
+			pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+		}
+
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].queue_id % nb_stages;
+
+			if (ev[i].queue_id == tx_queue[ev[i].vec->port]) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+				continue;
+			}
+
+			ev[i].queue_id++;
+			pipeline_fwd_event_vector(
+				&ev[i], cq_id != last_queue
+						? sched_type_list[cq_id]
+						: RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].queue_id % nb_stages;
+
+			if (cq_id == last_queue) {
+				ev[i].queue_id = tx_queue[ev[i].vec->port];
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].queue_id++;
+				pipeline_fwd_event_vector(
+					&ev[i], sched_type_list[cq_id]);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
 static int
 worker_wrapper(void *arg)
 {
@@ -275,29 +541,39 @@ worker_wrapper(void *arg)
 	const bool burst = evt_has_burst_mode(w->dev_id);
 	const bool internal_port = w->t->internal_port;
 	const uint8_t nb_stages = opt->nb_stages;
-	RTE_SET_USED(opt);
-
-	if (nb_stages == 1) {
-		if (!burst && internal_port)
-			return pipeline_queue_worker_single_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_queue_worker_single_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_queue_worker_single_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_queue_worker_single_stage_burst_fwd(
-					arg);
-	} else {
-		if (!burst && internal_port)
-			return pipeline_queue_worker_multi_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_queue_worker_multi_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_queue_worker_multi_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_queue_worker_multi_stage_burst_fwd(arg);
+	/*vector/burst/internal_port*/
+	const pipeline_queue_worker_t
+	pipeline_queue_worker_single_stage[2][2][2] = {
+		[0][0][0] = pipeline_queue_worker_single_stage_fwd,
+		[0][0][1] = pipeline_queue_worker_single_stage_tx,
+		[0][1][0] = pipeline_queue_worker_single_stage_burst_fwd,
+		[0][1][1] = pipeline_queue_worker_single_stage_burst_tx,
+		[1][0][0] = pipeline_queue_worker_single_stage_fwd_vector,
+		[1][0][1] = pipeline_queue_worker_single_stage_tx_vector,
+		[1][1][0] = pipeline_queue_worker_single_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_queue_worker_single_stage_burst_tx_vector,
+	};
+	const pipeline_queue_worker_t
+	pipeline_queue_worker_multi_stage[2][2][2] = {
+		[0][0][0] = pipeline_queue_worker_multi_stage_fwd,
+		[0][0][1] = pipeline_queue_worker_multi_stage_tx,
+		[0][1][0] = pipeline_queue_worker_multi_stage_burst_fwd,
+		[0][1][1] = pipeline_queue_worker_multi_stage_burst_tx,
+		[1][0][0] = pipeline_queue_worker_multi_stage_fwd_vector,
+		[1][0][1] = pipeline_queue_worker_multi_stage_tx_vector,
+		[1][1][0] = pipeline_queue_worker_multi_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_queue_worker_multi_stage_burst_tx_vector,
+	};
+
+	if (nb_stages == 1)
+		return (pipeline_queue_worker_single_stage[opt->ena_vector]
+							  [burst]
+							  [internal_port])(arg);
+	else
+		return (pipeline_queue_worker_multi_stage[opt->ena_vector]
+							 [burst]
+							 [internal_port])(arg);
 
-	}
 	rte_panic("invalid worker\n");
 }
 
diff --git a/doc/guides/tools/testeventdev.rst b/doc/guides/tools/testeventdev.rst
index ad1788a3d..f252dc2c1 100644
--- a/doc/guides/tools/testeventdev.rst
+++ b/doc/guides/tools/testeventdev.rst
@@ -158,6 +158,26 @@ The following are the application command-line options:
        Set max packet mbuf size. Can be used configure Rx/Tx scatter gather.
        Only applicable for `pipeline_atq` and `pipeline_queue` tests.
 
+* ``--nb_eth_queues``
+
+       Configure multiple Rx queues per each ethernet port.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--enable_vector``
+
+       Enable event vector for Rx/Tx adapters.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--vector_size``
+
+       Vector size to configure for the Rx adapter.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--vector_tmo_ns``
+
+       Vector timeout nanoseconds to be configured for the Rx adapter.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
 
 Eventdev Tests
 --------------
@@ -607,6 +627,10 @@ Supported application command line options are following::
         --worker_deq_depth
         --prod_type_ethdev
         --deq_tmo_nsec
+        --nb_eth_queues
+        --enable_vector
+        --vector_size
+        --vector_tmo_ns
 
 
 .. Note::
@@ -623,6 +647,13 @@ Example command to run pipeline queue test:
     sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
         --test=pipeline_queue --wlcore=1 --prod_type_ethdev --stlist=a
 
+Example command to run pipeline atq test with vector events:
+
+.. code-block:: console
+
+    sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
+        --test=pipeline_queue --wlcore=1 --prod_type_ethdev --stlist=a \
+        --enable_vector  --vector_size 512
 
 PIPELINE_ATQ Test
 ~~~~~~~~~~~~~~~~~~~
@@ -699,6 +730,10 @@ Supported application command line options are following::
         --worker_deq_depth
         --prod_type_ethdev
         --deq_tmo_nsec
+        --nb_eth_queues
+        --enable_vector
+        --vector_size
+        --vector_tmo_ns
 
 
 .. Note::
@@ -708,9 +743,17 @@ Supported application command line options are following::
 Example
 ^^^^^^^
 
-Example command to run pipeline queue test:
+Example command to run pipeline atq test:
 
 .. code-block:: console
 
     sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
         --test=pipeline_atq --wlcore=1 --prod_type_ethdev --stlist=a
+
+Example command to run pipeline atq test with vector events:
+
+.. code-block:: console
+
+    sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
+        --test=pipeline_atq --wlcore=1 --prod_type_ethdev --stlist=a \
+        --enable_vector  --vector_size 512
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v6 7/8] doc: announce event Rx adapter config changes
  2021-03-24 19:28         ` [dpdk-dev] [PATCH v6 " pbhagavatula
                             ` (5 preceding siblings ...)
  2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
@ 2021-03-24 19:28           ` pbhagavatula
  2021-03-24 19:28           ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v6 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
  2021-03-25 17:10           ` [dpdk-dev] [PATCH v7 0/8] Introduce event vectorization pbhagavatula
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-24 19:28 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

The Rx adapter event vector configuration will be merged into
Rx adapter queue configuration to simplify enabling event
vectorization.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Ray Kinsella <mdr@ashroe.eu>
---
 doc/guides/rel_notes/deprecation.rst | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst
index 64629e064..ef6d1bb18 100644
--- a/doc/guides/rel_notes/deprecation.rst
+++ b/doc/guides/rel_notes/deprecation.rst
@@ -130,3 +130,12 @@ Deprecation Notices
 * cmdline: ``cmdline`` structure will be made opaque to hide platform-specific
   content. On Linux and FreeBSD, supported prior to DPDK 20.11,
   original structure will be kept until DPDK 21.11.
+
+* eventdev: The structure ``rte_event_eth_rx_adapter_queue_conf`` will be
+  extended to include ``rte_event_eth_rx_adapter_event_vector_config`` elements
+  and the function ``rte_event_eth_rx_adapter_queue_event_vector_config`` will
+  be removed in DPDK 21.11.
+
+  An application can enable event vectorization by passing the desired vector
+  values to the function ``rte_event_eth_rx_adapter_queue_add`` using
+  the structure ``rte_event_eth_rx_adapter_queue_add``.
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [dpdk-dev v21.11] [PATCH v6 8/8] eventdev: simplify Rx adapter event vector config
  2021-03-24 19:28         ` [dpdk-dev] [PATCH v6 " pbhagavatula
                             ` (6 preceding siblings ...)
  2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 7/8] doc: announce event Rx adapter config changes pbhagavatula
@ 2021-03-24 19:28           ` pbhagavatula
  2021-03-25 17:10           ` [dpdk-dev] [PATCH v7 0/8] Introduce event vectorization pbhagavatula
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-24 19:28 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Include vector configuration into the structure
``rte_event_eth_rx_adapter_queue_conf`` used when configuring rest
of the Rx adapter ethernet device Rx queue parameters.
This simplifies event vector configuration as it avoids splitting
configuration per Rx queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 app/test-eventdev/test_pipeline_common.c      |  16 +-
 lib/librte_eventdev/eventdev_pmd.h            |  29 ---
 .../rte_event_eth_rx_adapter.c                | 168 ++++++------------
 .../rte_event_eth_rx_adapter.h                |  27 ---
 lib/librte_eventdev/version.map               |   1 -
 5 files changed, 57 insertions(+), 184 deletions(-)

diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index d5ef90500..76aee254b 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -331,7 +331,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 	uint16_t prod;
 	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
-	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
@@ -397,8 +396,12 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			}

 			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.vector_sz = opt->vector_size;
+				queue_conf.vector_timeout_ns =
+					opt->vector_tmo_nsec;
 				queue_conf.rx_queue_flags |=
 				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+				queue_conf.vector_mp = vector_pool;
 			} else {
 				evt_err("Rx adapter doesn't support event vector");
 				return -EINVAL;
@@ -418,17 +421,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}

-		if (opt->ena_vector) {
-			vec_conf.vector_sz = opt->vector_size;
-			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
-			vec_conf.vector_mp = vector_pool;
-			if (rte_event_eth_rx_adapter_queue_event_vector_config(
-				    prod, prod, -1, &vec_conf) < 0) {
-				evt_err("Failed to configure event vectorization for Rx adapter");
-				return -EINVAL;
-			}
-		}
-
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;

diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 0f724ac85..63b3bc4b5 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -667,32 +667,6 @@ typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
 	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-struct rte_event_eth_rx_adapter_event_vector_config;
-/**
- * Enable event vector on an given Rx queue of a ethernet devices belonging to
- * the Rx adapter.
- *
- * @param dev
- *   Event device pointer
- *
- * @param eth_dev
- *   Ethernet device pointer
- *
- * @param rx_queue_id
- *   The Rx queue identifier
- *
- * @param config
- *   Pointer to the event vector configuration structure.
- *
- * @return
- *   - 0: Success.
- *   - <0: Error code returned by the driver function.
- */
-typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
-	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
-	int32_t rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;

@@ -1118,9 +1092,6 @@ struct rte_eventdev_ops {
 	eventdev_eth_rx_adapter_vector_limits_get_t
 		eth_rx_adapter_vector_limits_get;
 	/**< Get event vector limits for the Rx adapter */
-	eventdev_eth_rx_adapter_event_vector_config_t
-		eth_rx_adapter_event_vector_config;
-	/**< Configure Rx adapter with event vector */

 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index c71990078..a1990637f 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -1882,6 +1882,25 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	} else
 		qi_ev->flow_id = 0;

+	if (conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+		queue_info->ena_vector = 1;
+		qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+		rxa_set_vector_data(queue_info, conf->vector_sz,
+				    conf->vector_timeout_ns, conf->vector_mp,
+				    rx_queue_id, dev_info->dev->data->port_id);
+		rx_adapter->ena_vector = 1;
+		rx_adapter->vector_tmo_ticks =
+			rx_adapter->vector_tmo_ticks
+				? RTE_MIN(queue_info->vector_data
+						  .vector_timeout_ticks,
+					  rx_adapter->vector_tmo_ticks)
+				: queue_info->vector_data.vector_timeout_ticks;
+		rx_adapter->vector_tmo_ticks <<= 1;
+		TAILQ_INIT(&rx_adapter->vector_list);
+		rx_adapter->prev_expiry_ts = 0;
+	}
+
 	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
 	if (rxa_polled_queue(dev_info, rx_queue_id)) {
 		rx_adapter->num_rx_polled += !pollq;
@@ -1907,44 +1926,6 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }

-static void
-rxa_sw_event_vector_configure(
-	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
-	int rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
-	struct eth_rx_queue_info *queue_info;
-	struct rte_event *qi_ev;
-
-	if (rx_queue_id == -1) {
-		uint16_t nb_rx_queues;
-		uint16_t i;
-
-		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
-		for (i = 0; i < nb_rx_queues; i++)
-			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
-						      config);
-		return;
-	}
-
-	queue_info = &dev_info->rx_queue[rx_queue_id];
-	qi_ev = (struct rte_event *)&queue_info->event;
-	queue_info->ena_vector = 1;
-	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
-	rxa_set_vector_data(queue_info, config->vector_sz,
-			    config->vector_timeout_ns, config->vector_mp,
-			    rx_queue_id, dev_info->dev->data->port_id);
-	rx_adapter->ena_vector = 1;
-	rx_adapter->vector_tmo_ticks =
-		rx_adapter->vector_tmo_ticks ?
-			      RTE_MIN(config->vector_timeout_ns << 1,
-				      rx_adapter->vector_tmo_ticks) :
-			      config->vector_timeout_ns << 1;
-	rx_adapter->prev_expiry_ts = 0;
-	TAILQ_INIT(&rx_adapter->vector_list);
-}
-
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -2258,6 +2239,7 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	struct rte_event_eth_rx_adapter *rx_adapter;
 	struct rte_eventdev *dev;
 	struct eth_device_info *dev_info;
+	struct rte_event_eth_rx_adapter_vector_limits limits;

 	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
 	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
@@ -2294,6 +2276,39 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}

+	if (queue_conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+		ret = rte_event_eth_rx_adapter_vector_limits_get(
+			rx_adapter->eventdev_id, eth_dev_id, &limits);
+		if (ret < 0) {
+			RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_sz < limits.min_sz ||
+		    queue_conf->vector_sz > limits.max_sz ||
+		    queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
+		    queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
+		    queue_conf->vector_mp == NULL) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_mp->elt_size <
+		    (sizeof(struct rte_event_vector) +
+		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+	}
+
 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
 		(rx_queue_id != -1)) {
 		RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
@@ -2487,83 +2502,6 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }

-int
-rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct rte_event_eth_rx_adapter_vector_limits limits;
-	struct rte_event_eth_rx_adapter *rx_adapter;
-	struct rte_eventdev *dev;
-	uint32_t cap;
-	int ret;
-
-	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
-	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
-
-	rx_adapter = rxa_id_to_adapter(id);
-	if ((rx_adapter == NULL) || (config == NULL))
-		return -EINVAL;
-
-	dev = &rte_eventdevs[rx_adapter->eventdev_id];
-	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
-						eth_dev_id, &cap);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
-				 "eth port %" PRIu16,
-				 id, eth_dev_id);
-		return ret;
-	}
-
-	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
-		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	ret = rte_event_eth_rx_adapter_vector_limits_get(
-		rx_adapter->eventdev_id, eth_dev_id, &limits);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
-				 "eth port %" PRIu16,
-				 rx_adapter->eventdev_id, eth_dev_id);
-		return ret;
-	}
-
-	if (config->vector_sz < limits.min_sz ||
-	    config->vector_sz > limits.max_sz ||
-	    config->vector_timeout_ns < limits.min_timeout_ns ||
-	    config->vector_timeout_ns > limits.max_timeout_ns ||
-	    config->vector_mp == NULL) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-	if (config->vector_mp->elt_size <
-	    (sizeof(struct rte_event_vector) +
-	     (sizeof(uintptr_t) * config->vector_sz))) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
-		RTE_FUNC_PTR_OR_ERR_RET(
-			*dev->dev_ops->eth_rx_adapter_event_vector_config,
-			-ENOTSUP);
-		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
-			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
-	} else {
-		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
-					      rx_queue_id, config);
-	}
-
-	return ret;
-}
-
 int
 rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
index 7407cde00..3f8b36229 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
@@ -171,9 +171,6 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
 	 * enqueued event.
 	 */
-};
-
-struct rte_event_eth_rx_adapter_event_vector_config {
 	uint16_t vector_sz;
 	/**<
 	 * Indicates the maximum number for mbufs to combine and form a vector.
@@ -548,30 +545,6 @@ int rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-/**
- * Configure event vectorization for a given ethernet device queue, that has
- * been added to a event eth Rx adapter.
- *
- * @param id
- *  The identifier of the ethernet Rx event adapter.
- *
- * @param eth_dev_id
- *  The identifier of the ethernet device.
- *
- * @param rx_queue_id
- *  Ethernet device receive queue index.
- *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
- *  are configured with event vectorization.
- *
- * @return
- *  - 0: Success, Receive queue configured correctly.
- *  - <0: Error code on failure.
- */
-__rte_experimental
-int rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index 902df0ae3..34c1c830e 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -142,7 +142,6 @@ EXPERIMENTAL {
 	#added in 21.05
 	rte_event_vector_pool_create;
 	rte_event_eth_rx_adapter_vector_limits_get;
-	rte_event_eth_rx_adapter_queue_event_vector_config;
 };

 INTERNAL {
--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v5 2/8] eventdev: introduce event vector Rx capability
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 2/8] eventdev: introduce event vector Rx capability pbhagavatula
  2021-03-24  9:15           ` Kinsella, Ray
@ 2021-03-25  8:15           ` Jayatheerthan, Jay
  2021-03-25  9:24             ` Pavan Nikhilesh Bhagavatula
  1 sibling, 1 reply; 153+ messages in thread
From: Jayatheerthan, Jay @ 2021-03-25  8:15 UTC (permalink / raw)
  To: pbhagavatula, jerinj, Carrillo, Erik G, Gujjar, Abhinandan S,
	McDaniel, Timothy, hemant.agrawal, Van Haaren, Harry,
	mattias.ronnblom, Ma, Liang J, Ray Kinsella, Neil Horman
  Cc: dev

> -----Original Message-----
> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
> Sent: Wednesday, March 24, 2021 10:35 AM
> To: jerinj@marvell.com; Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Carrillo, Erik G <erik.g.carrillo@intel.com>; Gujjar,
> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy <timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom <mattias.ronnblom@ericsson.com>; Ma, Liang J
> <liang.j.ma@intel.com>; Ray Kinsella <mdr@ashroe.eu>; Neil Horman <nhorman@tuxdriver.com>
> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
> Subject: [dpdk-dev] [PATCH v5 2/8] eventdev: introduce event vector Rx capability
> 
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> Introduce event ethernet Rx adapter event vector capability.
> 
> If an event eth Rx adapter has the capability of
> RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR then a given Rx queue
> can be configured to enable event vectorization by passing the
> flag RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR to
> rte_event_eth_rx_adapter_queue_conf::rx_queue_flags while configuring
> Rx adapter through rte_event_eth_rx_adapter_queue_add().
> 
> The max vector size, vector timeout define the vector size and
> mempool used for allocating vector event are configured through
> rte_event_eth_rx_adapter_queue_add. The element size of the element
> in the vector pool should be equal to
>     sizeof(struct rte_event_vector) + (vector_sz * sizeof(uintptr_t))
> 
> Application can use `rte_event_vector_pool_create` to create the
> vector mempool used for
> rte_event_eth_rx_adapter_queue_conf::vector_mp.
> 
> The Rx adapter would be responsible for vectorizing the mbufs
> based on the flow, the vector limits configured by the application
> and add the vector event of mbufs to the event queue set via
> rte_event_eth_rx_adapter_queue_conf::ev::queue_id.
> It should also mark rte_event_vector::union_valid and fill
> rte_event_vector::port, rte_event_vector::queue.
> 
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> Acked-by: Jerin Jacob <jerinj@marvell.com>
> ---
>  .../prog_guide/event_ethernet_rx_adapter.rst  |  38 ++++++
>  lib/librte_eventdev/eventdev_pmd.h            |  53 ++++++++
>  .../rte_event_eth_rx_adapter.c                | 114 ++++++++++++++++++
>  .../rte_event_eth_rx_adapter.h                | 105 ++++++++++++++++
>  lib/librte_eventdev/rte_eventdev.h            |  30 ++++-
>  lib/librte_eventdev/version.map               |   2 +
>  6 files changed, 340 insertions(+), 2 deletions(-)
> 
> diff --git a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
> index cb44ce0e4..5eefef355 100644
> --- a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
> +++ b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
> @@ -186,3 +186,41 @@ the event buffer fill level is low. The
>  ``rte_event_eth_rx_adapter_cb_register()`` function allow the application
>  to register a callback that selects which packets to enqueue to the event
>  device.
> +
> +Rx event vectorization
> +~~~~~~~~~~~~~~~~~~~~~~
> +
> +The event devices, ethernet device pairs which support the capability
> +``RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR`` can aggregate packets based on
> +flow characteristics and generate a ``rte_event`` containing ``rte_event_vector``
> +whose event type is either ``RTE_EVENT_TYPE_ETHDEV_VECTOR`` or
> +``RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR``.
> +The aggregation size and timeout are configurable at a queue level and the
> +maximum, minimum vector sizes and timeouts vary based on the device capability
> +and can be queried using ``rte_event_eth_rx_adapter_vector_limits_get``.
> +The Rx adapter additionally might include useful data such as ethernet device
> +port and queue identifier in the ``rte_event_vector::port`` and
> +``rte_event_vector::queue`` and mark ``rte_event_vector::attr_valid`` as true.
> +
> +A loop processing ``rte_event_vector`` containing mbufs is shown below.
> +
> +.. code-block:: c
> +
> +        event = rte_event_dequeue_burst(event_dev, event_port, &event,
> +                                        1, 0);
> +        if (!event)
> +                continue;
> +
> +        switch (ev.event_type) {
> +        case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
> +        case RTE_EVENT_TYPE_ETHDEV_VECTOR:
> +                struct rte_mbufs **mbufs;
> +
> +                mbufs = (struct rte_mbufs **)ev[i].vec->mbufs;
> +                for (i = 0; i < ev.vec->nb_elem; i++) {
> +                        /* Process each mbuf. */
> +                }
> +        break;
> +        case ...
> +        ...
> +        }
> diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
> index 7eb9a7739..9297f1433 100644
> --- a/lib/librte_eventdev/eventdev_pmd.h
> +++ b/lib/librte_eventdev/eventdev_pmd.h
> @@ -645,6 +645,53 @@ typedef int (*eventdev_eth_rx_adapter_stats_reset)
>   */
>  typedef int (*eventdev_selftest)(void);
> 
> +struct rte_event_eth_rx_adapter_vector_limits;
> +/**
> + * Get event vector limits for a given event, ethernet device pair.
> + *
> + * @param dev
> + *   Event device pointer
> + *
> + * @param eth_dev
> + *   Ethernet device pointer
> + *
> + * @param[out] limits
> + *   Pointer to the limits structure to be filled.
> + *
> + * @return
> + *   - 0: Success.
> + *   - <0: Error code returned by the driver function.
> + */
> +typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
> +	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
> +	struct rte_event_eth_rx_adapter_vector_limits *limits);
> +
> +struct rte_event_eth_rx_adapter_event_vector_config;
> +/**
> + * Enable event vector on an given Rx queue of a ethernet devices belonging to
> + * the Rx adapter.
> + *
> + * @param dev
> + *   Event device pointer
> + *
> + * @param eth_dev
> + *   Ethernet device pointer
> + *
> + * @param rx_queue_id
> + *   The Rx queue identifier
> + *
> + * @param config
> + *   Pointer to the event vector configuration structure.
> + *
> + * @return
> + *   - 0: Success.
> + *   - <0: Error code returned by the driver function.
> + */
> +typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
> +	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
> +	int32_t rx_queue_id,
> +	const struct rte_event_eth_rx_adapter_event_vector_config *config);
> +
>  typedef uint32_t rte_event_pmd_selftest_seqn_t;
>  extern int rte_event_pmd_selftest_seqn_dynfield_offset;
> 
> @@ -1067,6 +1114,12 @@ struct rte_eventdev_ops {
>  	/**< Get ethernet Rx stats */
>  	eventdev_eth_rx_adapter_stats_reset eth_rx_adapter_stats_reset;
>  	/**< Reset ethernet Rx stats */
> +	eventdev_eth_rx_adapter_vector_limits_get_t
> +		eth_rx_adapter_vector_limits_get;
> +	/**< Get event vector limits for the Rx adapter */
> +	eventdev_eth_rx_adapter_event_vector_config_t
> +		eth_rx_adapter_event_vector_config;
> +	/**< Configure Rx adapter with event vector */
> 
>  	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
>  	/**< Get timer adapter capabilities */
> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> index d8c635e99..ac8ba5bf0 100644
> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> @@ -2263,6 +2263,120 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
>  	return ret;
>  }
> 
> +int
> +rte_event_eth_rx_adapter_queue_event_vector_config(
> +	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
> +	struct rte_event_eth_rx_adapter_event_vector_config *config)
> +{
> +	struct rte_event_eth_rx_adapter_vector_limits limits;
> +	struct rte_event_eth_rx_adapter *rx_adapter;
> +	struct rte_eventdev *dev;
> +	uint32_t cap;
> +	int ret;
> +
> +	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
> +	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
> +
> +	rx_adapter = rxa_id_to_adapter(id);
> +	if ((rx_adapter == NULL) || (config == NULL))
> +		return -EINVAL;
> +
> +	dev = &rte_eventdevs[rx_adapter->eventdev_id];
> +	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
> +						eth_dev_id, &cap);
> +	if (ret) {
> +		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
> +				 "eth port %" PRIu16,
> +				 id, eth_dev_id);
> +		return ret;
> +	}
> +
> +	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
> +		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
> +				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
> +				 eth_dev_id, id);
> +		return -EINVAL;
> +	}
> +
> +	ret = rte_event_eth_rx_adapter_vector_limits_get(
> +		rx_adapter->eventdev_id, eth_dev_id, &limits);
> +	if (ret) {
> +		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
> +				 "eth port %" PRIu16,
> +				 rx_adapter->eventdev_id, eth_dev_id);
> +		return ret;
> +	}
> +
> +	if (config->vector_sz < limits.min_sz ||
> +	    config->vector_sz > limits.max_sz ||
> +	    config->vector_timeout_ns < limits.min_timeout_ns ||
> +	    config->vector_timeout_ns > limits.max_timeout_ns ||
> +	    config->vector_mp == NULL) {
> +		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> +				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
> +				 eth_dev_id, id);
> +		return -EINVAL;
> +	}
> +	if (config->vector_mp->elt_size <
> +	    (sizeof(struct rte_event_vector) +
> +	     (sizeof(uintptr_t) * config->vector_sz))) {
> +		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> +				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
> +				 eth_dev_id, id);
> +		return -EINVAL;
> +	}
> +
> +	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
> +		RTE_FUNC_PTR_OR_ERR_RET(
> +			*dev->dev_ops->eth_rx_adapter_event_vector_config,
> +			-ENOTSUP);
> +		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
> +			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
> +	} else {
> +		ret = -ENOTSUP;
> +	}

Trying to understand why non-INTERNAL_PORT based event device implementation returns ENOTSUP. Do you foresee any issues if such a device implements this function?

> +
> +	return ret;
> +}
> +
> +int
> +rte_event_eth_rx_adapter_vector_limits_get(
> +	uint8_t dev_id, uint16_t eth_port_id,
> +	struct rte_event_eth_rx_adapter_vector_limits *limits)
> +{
> +	struct rte_eventdev *dev;
> +	uint32_t cap;
> +	int ret;
> +
> +	RTE_EVENTDEV_VALID_DEVID_OR_ERR_RET(dev_id, -EINVAL);
> +	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_port_id, -EINVAL);
> +
> +	if (limits == NULL)
> +		return -EINVAL;
> +
> +	dev = &rte_eventdevs[dev_id];
> +
> +	ret = rte_event_eth_rx_adapter_caps_get(dev_id, eth_port_id, &cap);
> +	if (ret) {
> +		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
> +				 "eth port %" PRIu16,
> +				 dev_id, eth_port_id);
> +		return ret;
> +	}
> +
> +	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
> +		RTE_FUNC_PTR_OR_ERR_RET(
> +			*dev->dev_ops->eth_rx_adapter_vector_limits_get,
> +			-ENOTSUP);
> +		ret = dev->dev_ops->eth_rx_adapter_vector_limits_get(
> +			dev, &rte_eth_devices[eth_port_id], limits);
> +	} else {
> +		ret = -ENOTSUP;
> +	}
> +

Same here.

> +	return ret;
> +}
> +
>  int
>  rte_event_eth_rx_adapter_start(uint8_t id)
>  {
> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> index 21bb1e54c..7407cde00 100644
> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> @@ -92,6 +92,10 @@ extern "C" {
>  /**< This flag indicates the flow identifier is valid
>   * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
>   */
> +#define RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR	0x2
> +/**< This flag indicates that mbufs arriving on the queue need to be vectorized
> + * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
> + */
> 
>  /**
>   * Adapter configuration structure that the adapter configuration callback
> @@ -169,6 +173,36 @@ struct rte_event_eth_rx_adapter_queue_conf {
>  	 */
>  };
> 
> +struct rte_event_eth_rx_adapter_event_vector_config {
> +	uint16_t vector_sz;
> +	/**<
> +	 * Indicates the maximum number for mbufs to combine and form a vector.
> +	 * Should be within
> +	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_sz
> +	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_sz
> +	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
> +	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
> +	 */
> +	uint64_t vector_timeout_ns;
> +	/**<
> +	 * Indicates the maximum number of nanoseconds to wait for receiving
> +	 * mbufs. Should be within vectorization limits of the
> +	 * adapter
> +	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_ns
> +	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_ns
> +	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
> +	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
> +	 */
> +	struct rte_mempool *vector_mp;
> +	/**<
> +	 * Indicates the mempool that should be used for allocating
> +	 * rte_event_vector container.
> +	 * Should be created by using `rte_event_vector_pool_create`.
> +	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
> +	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags.
> +	 */
> +};
> +
>  /**
>   * A structure used to retrieve statistics for an eth rx adapter instance.
>   */
> @@ -199,6 +233,32 @@ struct rte_event_eth_rx_adapter_stats {
>  	/**< Received packet count for interrupt mode Rx queues */
>  };
> 
> +/**
> + * A structure used to retrieve eth rx adapter vector limits.
> + */
> +struct rte_event_eth_rx_adapter_vector_limits {
> +	uint16_t min_sz;
> +	/**< Minimum vector limit configurable.
> +	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_sz
> +	 */
> +	uint16_t max_sz;
> +	/**< Maximum vector limit configurable.
> +	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_sz
> +	 */
> +	uint8_t log2_sz;
> +	/**< True if the size configured should be in log2.
> +	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_sz
> +	 */
> +	uint64_t min_timeout_ns;
> +	/**< Minimum vector timeout configurable.
> +	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_timeout_ns
> +	 */
> +	uint64_t max_timeout_ns;
> +	/**< Maximum vector timeout configurable.
> +	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_timeout_ns
> +	 */
> +};
> +
>  /**
>   *
>   * Callback function invoked by the SW adapter before it continues
> @@ -467,6 +527,51 @@ int rte_event_eth_rx_adapter_cb_register(uint8_t id, uint16_t eth_dev_id,
>  					 rte_event_eth_rx_adapter_cb_fn cb_fn,
>  					 void *cb_arg);
> 
> +/**
> + * Retrieve vector limits for a given event dev and eth dev pair.
> + * @see rte_event_eth_rx_adapter_vector_limits
> + *
> + * @param dev_id
> + *  Event device identifier.
> + * @param eth_port_id
> + *  Port identifier of the ethernet device.
> + * @param [out] limits
> + *  A pointer to rte_event_eth_rx_adapter_vector_limits structure that has to
> + * be filled.
> + *
> + * @return
> + *  - 0: Success.
> + *  - <0: Error code on failure.
> + */
> +__rte_experimental
> +int rte_event_eth_rx_adapter_vector_limits_get(
> +	uint8_t dev_id, uint16_t eth_port_id,
> +	struct rte_event_eth_rx_adapter_vector_limits *limits);
> +
> +/**
> + * Configure event vectorization for a given ethernet device queue, that has
> + * been added to a event eth Rx adapter.
> + *
> + * @param id
> + *  The identifier of the ethernet Rx event adapter.
> + *
> + * @param eth_dev_id
> + *  The identifier of the ethernet device.
> + *
> + * @param rx_queue_id
> + *  Ethernet device receive queue index.
> + *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
> + *  are configured with event vectorization.
> + *
> + * @return
> + *  - 0: Success, Receive queue configured correctly.
> + *  - <0: Error code on failure.
> + */
> +__rte_experimental
> +int rte_event_eth_rx_adapter_queue_event_vector_config(
> +	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
> +	struct rte_event_eth_rx_adapter_event_vector_config *config);
> +
>  #ifdef __cplusplus
>  }
>  #endif
> diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
> index aa4dd3959..678338247 100644
> --- a/lib/librte_eventdev/rte_eventdev.h
> +++ b/lib/librte_eventdev/rte_eventdev.h
> @@ -919,10 +919,28 @@ rte_event_dev_close(uint8_t dev_id);
>   * Event vector structure.
>   */
>  struct rte_event_vector {
> -	uint64_t nb_elem : 16;
> +	uint16_t nb_elem;
>  	/**< Number of elements in this event vector. */
> -	uint64_t rsvd : 48;
> +	uint16_t rsvd : 15;
>  	/**< Reserved for future use */
> +	uint16_t attr_valid : 1;
> +	/**< Indicates that the below union attributes have valid information.
> +	 */
> +	union {
> +		/* Used by Rx adapter.
> +		 * Indicates that all the elements in this vector belong to the
> +		 * same port and queue pair when originating from Rx adapter,
> +		 * valid only when event type is ETHDEV_VECTOR or
> +		 * ETH_RX_ADAPTER_VECTOR.
> +		 */
> +		struct {
> +			uint16_t port;
> +			/* Ethernet device port id. */
> +			uint16_t queue;
> +			/* Ethernet device queue id. */
> +		};
> +	};
> +	/**< Union to hold common attributes of the vector array. */
>  	uint64_t impl_opaque;
>  	/**< Implementation specific opaque value.
>  	 * An implementation may use this field to hold implementation specific
> @@ -1025,8 +1043,14 @@ struct rte_event_vector {
>   *		// Classify and handle event.
>   *	}
>   */
> +#define RTE_EVENT_TYPE_ETHDEV_VECTOR                                           \
> +	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETHDEV)
> +/**< The event vector generated from ethdev subsystem */
>  #define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
>  /**< The event vector generated from cpu for pipelining. */
> +#define RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR                                   \
> +	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETH_RX_ADAPTER)
> +/**< The event vector generated from eth Rx adapter. */
> 
>  #define RTE_EVENT_TYPE_MAX              0x10
>  /**< Maximum number of event types */
> @@ -1171,6 +1195,8 @@ struct rte_event {
>   * @see struct rte_event_eth_rx_adapter_queue_conf::ev
>   * @see struct rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
>   */
> +#define RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR	0x8
> +/**< Adapter supports event vectorization per ethdev. */
> 
>  /**
>   * Retrieve the event device's ethdev Rx adapter capabilities for the
> diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
> index a070ef56e..902df0ae3 100644
> --- a/lib/librte_eventdev/version.map
> +++ b/lib/librte_eventdev/version.map
> @@ -141,6 +141,8 @@ EXPERIMENTAL {
> 
>  	#added in 21.05
>  	rte_event_vector_pool_create;
> +	rte_event_eth_rx_adapter_vector_limits_get;
> +	rte_event_eth_rx_adapter_queue_event_vector_config;
>  };
> 
>  INTERNAL {
> --
> 2.17.1

More of informational queries.

Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v5 3/8] eventdev: introduce event vector Tx capability
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 3/8] eventdev: introduce event vector Tx capability pbhagavatula
@ 2021-03-25  8:16           ` Jayatheerthan, Jay
  0 siblings, 0 replies; 153+ messages in thread
From: Jayatheerthan, Jay @ 2021-03-25  8:16 UTC (permalink / raw)
  To: pbhagavatula, jerinj, Carrillo, Erik G, Gujjar, Abhinandan S,
	McDaniel, Timothy, hemant.agrawal, Van Haaren, Harry,
	mattias.ronnblom, Ma, Liang J
  Cc: dev

> -----Original Message-----
> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
> Sent: Wednesday, March 24, 2021 10:35 AM
> To: jerinj@marvell.com; Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Carrillo, Erik G <erik.g.carrillo@intel.com>; Gujjar,
> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy <timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom <mattias.ronnblom@ericsson.com>; Ma, Liang J
> <liang.j.ma@intel.com>
> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
> Subject: [dpdk-dev] [PATCH v5 3/8] eventdev: introduce event vector Tx capability
> 
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> Introduce event vector transmit capability for event eth
> tx adapter.
> 
> The capability indicates that the Tx adapter is capable of
> transmitting event vectors.
> When rte_event_vector::union_valid is set, the Tx adapter should
> transmit all the packets to the rte_event_vector::port using the
> rte_event_vector::queue.
> If rte_event_vector::union_valid is not set then the Tx adapter
> should peek into each mbuf to get the destination port and queue
> pair.
> 
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> Acked-by: Jerin Jacob <jerinj@marvell.com>
> ---
>  doc/guides/prog_guide/event_ethernet_tx_adapter.rst | 12 ++++++++++++
>  lib/librte_eventdev/rte_eventdev.h                  |  8 +++++++-
>  2 files changed, 19 insertions(+), 1 deletion(-)
> 
> diff --git a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
> index a8c13e136..87277dcaf 100644
> --- a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
> +++ b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
> @@ -164,3 +164,15 @@ The  ``rte_event_eth_tx_adapter_stats_get()`` function reports counters defined
>  in struct ``rte_event_eth_tx_adapter_stats``. The counter values are the sum of
>  the counts from the eventdev PMD callback if the callback is supported, and
>  the counts maintained by the service function, if one exists.
> +
> +Tx event vectorization
> +~~~~~~~~~~~~~~~~~~~~~~
> +
> +The event device, ethernet device paris which support the capability


Typo: "pairs" instead of "paris".

> +``RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR`` can process event vector of mbufs.
> +Additionally, application can provide a hint to that Tx adapter that all the


Typo: "hint to the Tx adapter" instead of "hint to that Tx adapter"

> +mbufs are destined to the same ethernet port and queue by setting the bit
> +``rte_event_vector::attr_valid`` and filling `rte_event_vector::port`` and
> +``rte_event_vector::queue``.
> +If ``rte_event_vector::attr_valid`` is not set then the Tx adapter should peek
> +into each mbuf and transmit them to the requested ethernet port and queue pair.
> diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
> index 678338247..67b681ef6 100644
> --- a/lib/librte_eventdev/rte_eventdev.h
> +++ b/lib/librte_eventdev/rte_eventdev.h
> @@ -927,11 +927,13 @@ struct rte_event_vector {
>  	/**< Indicates that the below union attributes have valid information.
>  	 */
>  	union {
> -		/* Used by Rx adapter.
> +		/* Used by Rx/Tx adapter.
>  		 * Indicates that all the elements in this vector belong to the
>  		 * same port and queue pair when originating from Rx adapter,
>  		 * valid only when event type is ETHDEV_VECTOR or
>  		 * ETH_RX_ADAPTER_VECTOR.
> +		 * Can also be used to indicate the Tx adapter the destination
> +		 * port and queue of the mbufs in the vector
>  		 */
>  		struct {
>  			uint16_t port;
> @@ -1293,6 +1295,10 @@ rte_event_crypto_adapter_caps_get(uint8_t dev_id, uint8_t cdev_id,
>  #define RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT	0x1
>  /**< This flag is sent when the PMD supports a packet transmit callback
>   */
> +#define RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR	0x2
> +/**< Indicates that the Tx adapter is capable of handling event vector of
> + * mbufs.
> + */
> 
>  /**
>   * Retrieve the event device's eth Tx adapter capabilities
> --
> 2.17.1

With above changes, you can add Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>.



^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v5 2/8] eventdev: introduce event vector Rx capability
  2021-03-25  8:15           ` Jayatheerthan, Jay
@ 2021-03-25  9:24             ` Pavan Nikhilesh Bhagavatula
  2021-03-25  9:50               ` Jayatheerthan, Jay
  0 siblings, 1 reply; 153+ messages in thread
From: Pavan Nikhilesh Bhagavatula @ 2021-03-25  9:24 UTC (permalink / raw)
  To: Jayatheerthan, Jay, Jerin Jacob Kollanukkaran, Carrillo, Erik G,
	Gujjar, Abhinandan S, McDaniel, Timothy, hemant.agrawal,
	Van Haaren, Harry, mattias.ronnblom, Ma, Liang J, Ray Kinsella,
	Neil Horman
  Cc: dev

>> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
>> Sent: Wednesday, March 24, 2021 10:35 AM
>> To: jerinj@marvell.com; Jayatheerthan, Jay
><jay.jayatheerthan@intel.com>; Carrillo, Erik G
><erik.g.carrillo@intel.com>; Gujjar,
>> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy
><timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
>> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
><mattias.ronnblom@ericsson.com>; Ma, Liang J
>> <liang.j.ma@intel.com>; Ray Kinsella <mdr@ashroe.eu>; Neil Horman
><nhorman@tuxdriver.com>
>> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
>> Subject: [dpdk-dev] [PATCH v5 2/8] eventdev: introduce event vector
>Rx capability
>>
>> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>>
>> Introduce event ethernet Rx adapter event vector capability.
>>
>> If an event eth Rx adapter has the capability of
>> RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR then a given Rx
>queue
>> can be configured to enable event vectorization by passing the
>> flag RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR to
>> rte_event_eth_rx_adapter_queue_conf::rx_queue_flags while
>configuring
>> Rx adapter through rte_event_eth_rx_adapter_queue_add().
>>
>> The max vector size, vector timeout define the vector size and
>> mempool used for allocating vector event are configured through
>> rte_event_eth_rx_adapter_queue_add. The element size of the
>element
>> in the vector pool should be equal to
>>     sizeof(struct rte_event_vector) + (vector_sz * sizeof(uintptr_t))
>>
>> Application can use `rte_event_vector_pool_create` to create the
>> vector mempool used for
>> rte_event_eth_rx_adapter_queue_conf::vector_mp.
>>
>> The Rx adapter would be responsible for vectorizing the mbufs
>> based on the flow, the vector limits configured by the application
>> and add the vector event of mbufs to the event queue set via
>> rte_event_eth_rx_adapter_queue_conf::ev::queue_id.
>> It should also mark rte_event_vector::union_valid and fill
>> rte_event_vector::port, rte_event_vector::queue.
>>
>> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
>> Acked-by: Jerin Jacob <jerinj@marvell.com>
>> ---
>>  .../prog_guide/event_ethernet_rx_adapter.rst  |  38 ++++++
>>  lib/librte_eventdev/eventdev_pmd.h            |  53 ++++++++
>>  .../rte_event_eth_rx_adapter.c                | 114 ++++++++++++++++++
>>  .../rte_event_eth_rx_adapter.h                | 105 ++++++++++++++++
>>  lib/librte_eventdev/rte_eventdev.h            |  30 ++++-
>>  lib/librte_eventdev/version.map               |   2 +
>>  6 files changed, 340 insertions(+), 2 deletions(-)
>>
>> diff --git a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
>b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
>> index cb44ce0e4..5eefef355 100644
>> --- a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
>> +++ b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
>> @@ -186,3 +186,41 @@ the event buffer fill level is low. The
>>  ``rte_event_eth_rx_adapter_cb_register()`` function allow the
>application
>>  to register a callback that selects which packets to enqueue to the
>event
>>  device.
>> +
>> +Rx event vectorization
>> +~~~~~~~~~~~~~~~~~~~~~~
>> +
>> +The event devices, ethernet device pairs which support the capability
>> +``RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR`` can
>aggregate packets based on
>> +flow characteristics and generate a ``rte_event`` containing
>``rte_event_vector``
>> +whose event type is either ``RTE_EVENT_TYPE_ETHDEV_VECTOR`` or
>> +``RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR``.
>> +The aggregation size and timeout are configurable at a queue level
>and the
>> +maximum, minimum vector sizes and timeouts vary based on the
>device capability
>> +and can be queried using
>``rte_event_eth_rx_adapter_vector_limits_get``.
>> +The Rx adapter additionally might include useful data such as
>ethernet device
>> +port and queue identifier in the ``rte_event_vector::port`` and
>> +``rte_event_vector::queue`` and mark
>``rte_event_vector::attr_valid`` as true.
>> +
>> +A loop processing ``rte_event_vector`` containing mbufs is shown
>below.
>> +
>> +.. code-block:: c
>> +
>> +        event = rte_event_dequeue_burst(event_dev, event_port,
>&event,
>> +                                        1, 0);
>> +        if (!event)
>> +                continue;
>> +
>> +        switch (ev.event_type) {
>> +        case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
>> +        case RTE_EVENT_TYPE_ETHDEV_VECTOR:
>> +                struct rte_mbufs **mbufs;
>> +
>> +                mbufs = (struct rte_mbufs **)ev[i].vec->mbufs;
>> +                for (i = 0; i < ev.vec->nb_elem; i++) {
>> +                        /* Process each mbuf. */
>> +                }
>> +        break;
>> +        case ...
>> +        ...
>> +        }
>> diff --git a/lib/librte_eventdev/eventdev_pmd.h
>b/lib/librte_eventdev/eventdev_pmd.h
>> index 7eb9a7739..9297f1433 100644
>> --- a/lib/librte_eventdev/eventdev_pmd.h
>> +++ b/lib/librte_eventdev/eventdev_pmd.h
>> @@ -645,6 +645,53 @@ typedef int
>(*eventdev_eth_rx_adapter_stats_reset)
>>   */
>>  typedef int (*eventdev_selftest)(void);
>>
>> +struct rte_event_eth_rx_adapter_vector_limits;
>> +/**
>> + * Get event vector limits for a given event, ethernet device pair.
>> + *
>> + * @param dev
>> + *   Event device pointer
>> + *
>> + * @param eth_dev
>> + *   Ethernet device pointer
>> + *
>> + * @param[out] limits
>> + *   Pointer to the limits structure to be filled.
>> + *
>> + * @return
>> + *   - 0: Success.
>> + *   - <0: Error code returned by the driver function.
>> + */
>> +typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
>> +	const struct rte_eventdev *dev, const struct rte_eth_dev
>*eth_dev,
>> +	struct rte_event_eth_rx_adapter_vector_limits *limits);
>> +
>> +struct rte_event_eth_rx_adapter_event_vector_config;
>> +/**
>> + * Enable event vector on an given Rx queue of a ethernet devices
>belonging to
>> + * the Rx adapter.
>> + *
>> + * @param dev
>> + *   Event device pointer
>> + *
>> + * @param eth_dev
>> + *   Ethernet device pointer
>> + *
>> + * @param rx_queue_id
>> + *   The Rx queue identifier
>> + *
>> + * @param config
>> + *   Pointer to the event vector configuration structure.
>> + *
>> + * @return
>> + *   - 0: Success.
>> + *   - <0: Error code returned by the driver function.
>> + */
>> +typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
>> +	const struct rte_eventdev *dev, const struct rte_eth_dev
>*eth_dev,
>> +	int32_t rx_queue_id,
>> +	const struct rte_event_eth_rx_adapter_event_vector_config
>*config);
>> +
>>  typedef uint32_t rte_event_pmd_selftest_seqn_t;
>>  extern int rte_event_pmd_selftest_seqn_dynfield_offset;
>>
>> @@ -1067,6 +1114,12 @@ struct rte_eventdev_ops {
>>  	/**< Get ethernet Rx stats */
>>  	eventdev_eth_rx_adapter_stats_reset
>eth_rx_adapter_stats_reset;
>>  	/**< Reset ethernet Rx stats */
>> +	eventdev_eth_rx_adapter_vector_limits_get_t
>> +		eth_rx_adapter_vector_limits_get;
>> +	/**< Get event vector limits for the Rx adapter */
>> +	eventdev_eth_rx_adapter_event_vector_config_t
>> +		eth_rx_adapter_event_vector_config;
>> +	/**< Configure Rx adapter with event vector */
>>
>>  	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
>>  	/**< Get timer adapter capabilities */
>> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> index d8c635e99..ac8ba5bf0 100644
>> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> @@ -2263,6 +2263,120 @@
>rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
>>  	return ret;
>>  }
>>
>> +int
>> +rte_event_eth_rx_adapter_queue_event_vector_config(
>> +	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
>> +	struct rte_event_eth_rx_adapter_event_vector_config *config)
>> +{
>> +	struct rte_event_eth_rx_adapter_vector_limits limits;
>> +	struct rte_event_eth_rx_adapter *rx_adapter;
>> +	struct rte_eventdev *dev;
>> +	uint32_t cap;
>> +	int ret;
>> +
>> +	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -
>EINVAL);
>> +	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
>> +
>> +	rx_adapter = rxa_id_to_adapter(id);
>> +	if ((rx_adapter == NULL) || (config == NULL))
>> +		return -EINVAL;
>> +
>> +	dev = &rte_eventdevs[rx_adapter->eventdev_id];
>> +	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter-
>>eventdev_id,
>> +						eth_dev_id, &cap);
>> +	if (ret) {
>> +		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev
>%" PRIu8
>> +				 "eth port %" PRIu16,
>> +				 id, eth_dev_id);
>> +		return ret;
>> +	}
>> +
>> +	if (!(cap &
>RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
>> +		RTE_EDEV_LOG_ERR("Event vectorization is not
>supported,"
>> +				 " eth port: %" PRIu16 " adapter id: %"
>PRIu8,
>> +				 eth_dev_id, id);
>> +		return -EINVAL;
>> +	}
>> +
>> +	ret = rte_event_eth_rx_adapter_vector_limits_get(
>> +		rx_adapter->eventdev_id, eth_dev_id, &limits);
>> +	if (ret) {
>> +		RTE_EDEV_LOG_ERR("Failed to get vector limits edev
>%" PRIu8
>> +				 "eth port %" PRIu16,
>> +				 rx_adapter->eventdev_id, eth_dev_id);
>> +		return ret;
>> +	}
>> +
>> +	if (config->vector_sz < limits.min_sz ||
>> +	    config->vector_sz > limits.max_sz ||
>> +	    config->vector_timeout_ns < limits.min_timeout_ns ||
>> +	    config->vector_timeout_ns > limits.max_timeout_ns ||
>> +	    config->vector_mp == NULL) {
>> +		RTE_EDEV_LOG_ERR("Invalid event vector
>configuration,"
>> +				 " eth port: %" PRIu16 " adapter id: %"
>PRIu8,
>> +				 eth_dev_id, id);
>> +		return -EINVAL;
>> +	}
>> +	if (config->vector_mp->elt_size <
>> +	    (sizeof(struct rte_event_vector) +
>> +	     (sizeof(uintptr_t) * config->vector_sz))) {
>> +		RTE_EDEV_LOG_ERR("Invalid event vector
>configuration,"
>> +				 " eth port: %" PRIu16 " adapter id: %"
>PRIu8,
>> +				 eth_dev_id, id);
>> +		return -EINVAL;
>> +	}
>> +
>> +	if (cap &
>RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
>> +		RTE_FUNC_PTR_OR_ERR_RET(
>> +			*dev->dev_ops-
>>eth_rx_adapter_event_vector_config,
>> +			-ENOTSUP);
>> +		ret = dev->dev_ops-
>>eth_rx_adapter_event_vector_config(
>> +			dev, &rte_eth_devices[eth_dev_id],
>rx_queue_id, config);
>> +	} else {
>> +		ret = -ENOTSUP;
>> +	}
>
>Trying to understand why non-INTERNAL_PORT based event device
>implementation returns ENOTSUP. Do you foresee any issues if such a
>device implements this function?
>

I just didn't want to move the implementation here as this patch is only 
Intended for spec.
The implementation for non-INTERNAL_PORT is in 4/8

>> +
>> +	return ret;
>> +}
>> +
>> +int
>> +rte_event_eth_rx_adapter_vector_limits_get(
>> +	uint8_t dev_id, uint16_t eth_port_id,
>> +	struct rte_event_eth_rx_adapter_vector_limits *limits)
>> +{
>> +	struct rte_eventdev *dev;
>> +	uint32_t cap;
>> +	int ret;
>> +
>> +	RTE_EVENTDEV_VALID_DEVID_OR_ERR_RET(dev_id, -EINVAL);
>> +	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_port_id, -EINVAL);
>> +
>> +	if (limits == NULL)
>> +		return -EINVAL;
>> +
>> +	dev = &rte_eventdevs[dev_id];
>> +
>> +	ret = rte_event_eth_rx_adapter_caps_get(dev_id, eth_port_id,
>&cap);
>> +	if (ret) {
>> +		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev
>%" PRIu8
>> +				 "eth port %" PRIu16,
>> +				 dev_id, eth_port_id);
>> +		return ret;
>> +	}
>> +
>> +	if (cap &
>RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
>> +		RTE_FUNC_PTR_OR_ERR_RET(
>> +			*dev->dev_ops-
>>eth_rx_adapter_vector_limits_get,
>> +			-ENOTSUP);
>> +		ret = dev->dev_ops-
>>eth_rx_adapter_vector_limits_get(
>> +			dev, &rte_eth_devices[eth_port_id], limits);
>> +	} else {
>> +		ret = -ENOTSUP;
>> +	}
>> +
>
>Same here.
>
>> +	return ret;
>> +}
>> +
>>  int
>>  rte_event_eth_rx_adapter_start(uint8_t id)
>>  {
>> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
>b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
>> index 21bb1e54c..7407cde00 100644
>> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
>> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
>> @@ -92,6 +92,10 @@ extern "C" {
>>  /**< This flag indicates the flow identifier is valid
>>   * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
>>   */
>> +#define RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR
>	0x2
>> +/**< This flag indicates that mbufs arriving on the queue need to be
>vectorized
>> + * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
>> + */
>>
>>  /**
>>   * Adapter configuration structure that the adapter configuration
>callback
>> @@ -169,6 +173,36 @@ struct
>rte_event_eth_rx_adapter_queue_conf {
>>  	 */
>>  };
>>
>> +struct rte_event_eth_rx_adapter_event_vector_config {
>> +	uint16_t vector_sz;
>> +	/**<
>> +	 * Indicates the maximum number for mbufs to combine and
>form a vector.
>> +	 * Should be within
>> +	 * @see
>rte_event_eth_rx_adapter_vector_limits::min_vector_sz
>> +	 * @see
>rte_event_eth_rx_adapter_vector_limits::max_vector_sz
>> +	 * Valid when
>RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
>> +	 * @see
>rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
>> +	 */
>> +	uint64_t vector_timeout_ns;
>> +	/**<
>> +	 * Indicates the maximum number of nanoseconds to wait for
>receiving
>> +	 * mbufs. Should be within vectorization limits of the
>> +	 * adapter
>> +	 * @see
>rte_event_eth_rx_adapter_vector_limits::min_vector_ns
>> +	 * @see
>rte_event_eth_rx_adapter_vector_limits::max_vector_ns
>> +	 * Valid when
>RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
>> +	 * @see
>rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
>> +	 */
>> +	struct rte_mempool *vector_mp;
>> +	/**<
>> +	 * Indicates the mempool that should be used for allocating
>> +	 * rte_event_vector container.
>> +	 * Should be created by using `rte_event_vector_pool_create`.
>> +	 * Valid when
>RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
>> +	 * @see
>rte_event_eth_rx_adapter_queue_conf::rx_queue_flags.
>> +	 */
>> +};
>> +
>>  /**
>>   * A structure used to retrieve statistics for an eth rx adapter
>instance.
>>   */
>> @@ -199,6 +233,32 @@ struct rte_event_eth_rx_adapter_stats {
>>  	/**< Received packet count for interrupt mode Rx queues */
>>  };
>>
>> +/**
>> + * A structure used to retrieve eth rx adapter vector limits.
>> + */
>> +struct rte_event_eth_rx_adapter_vector_limits {
>> +	uint16_t min_sz;
>> +	/**< Minimum vector limit configurable.
>> +	 * @see
>rte_event_eth_rx_adapter_event_vector_config::vector_sz
>> +	 */
>> +	uint16_t max_sz;
>> +	/**< Maximum vector limit configurable.
>> +	 * @see
>rte_event_eth_rx_adapter_event_vector_config::vector_sz
>> +	 */
>> +	uint8_t log2_sz;
>> +	/**< True if the size configured should be in log2.
>> +	 * @see
>rte_event_eth_rx_adapter_event_vector_config::vector_sz
>> +	 */
>> +	uint64_t min_timeout_ns;
>> +	/**< Minimum vector timeout configurable.
>> +	 * @see
>rte_event_eth_rx_adapter_event_vector_config::vector_timeout_ns
>> +	 */
>> +	uint64_t max_timeout_ns;
>> +	/**< Maximum vector timeout configurable.
>> +	 * @see
>rte_event_eth_rx_adapter_event_vector_config::vector_timeout_ns
>> +	 */
>> +};
>> +
>>  /**
>>   *
>>   * Callback function invoked by the SW adapter before it continues
>> @@ -467,6 +527,51 @@ int
>rte_event_eth_rx_adapter_cb_register(uint8_t id, uint16_t eth_dev_id,
>>
>rte_event_eth_rx_adapter_cb_fn cb_fn,
>>  					 void *cb_arg);
>>
>> +/**
>> + * Retrieve vector limits for a given event dev and eth dev pair.
>> + * @see rte_event_eth_rx_adapter_vector_limits
>> + *
>> + * @param dev_id
>> + *  Event device identifier.
>> + * @param eth_port_id
>> + *  Port identifier of the ethernet device.
>> + * @param [out] limits
>> + *  A pointer to rte_event_eth_rx_adapter_vector_limits structure
>that has to
>> + * be filled.
>> + *
>> + * @return
>> + *  - 0: Success.
>> + *  - <0: Error code on failure.
>> + */
>> +__rte_experimental
>> +int rte_event_eth_rx_adapter_vector_limits_get(
>> +	uint8_t dev_id, uint16_t eth_port_id,
>> +	struct rte_event_eth_rx_adapter_vector_limits *limits);
>> +
>> +/**
>> + * Configure event vectorization for a given ethernet device queue,
>that has
>> + * been added to a event eth Rx adapter.
>> + *
>> + * @param id
>> + *  The identifier of the ethernet Rx event adapter.
>> + *
>> + * @param eth_dev_id
>> + *  The identifier of the ethernet device.
>> + *
>> + * @param rx_queue_id
>> + *  Ethernet device receive queue index.
>> + *  If rx_queue_id is -1, then all Rx queues configured for the
>ethernet device
>> + *  are configured with event vectorization.
>> + *
>> + * @return
>> + *  - 0: Success, Receive queue configured correctly.
>> + *  - <0: Error code on failure.
>> + */
>> +__rte_experimental
>> +int rte_event_eth_rx_adapter_queue_event_vector_config(
>> +	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
>> +	struct rte_event_eth_rx_adapter_event_vector_config
>*config);
>> +
>>  #ifdef __cplusplus
>>  }
>>  #endif
>> diff --git a/lib/librte_eventdev/rte_eventdev.h
>b/lib/librte_eventdev/rte_eventdev.h
>> index aa4dd3959..678338247 100644
>> --- a/lib/librte_eventdev/rte_eventdev.h
>> +++ b/lib/librte_eventdev/rte_eventdev.h
>> @@ -919,10 +919,28 @@ rte_event_dev_close(uint8_t dev_id);
>>   * Event vector structure.
>>   */
>>  struct rte_event_vector {
>> -	uint64_t nb_elem : 16;
>> +	uint16_t nb_elem;
>>  	/**< Number of elements in this event vector. */
>> -	uint64_t rsvd : 48;
>> +	uint16_t rsvd : 15;
>>  	/**< Reserved for future use */
>> +	uint16_t attr_valid : 1;
>> +	/**< Indicates that the below union attributes have valid
>information.
>> +	 */
>> +	union {
>> +		/* Used by Rx adapter.
>> +		 * Indicates that all the elements in this vector belong to
>the
>> +		 * same port and queue pair when originating from Rx
>adapter,
>> +		 * valid only when event type is ETHDEV_VECTOR or
>> +		 * ETH_RX_ADAPTER_VECTOR.
>> +		 */
>> +		struct {
>> +			uint16_t port;
>> +			/* Ethernet device port id. */
>> +			uint16_t queue;
>> +			/* Ethernet device queue id. */
>> +		};
>> +	};
>> +	/**< Union to hold common attributes of the vector array. */
>>  	uint64_t impl_opaque;
>>  	/**< Implementation specific opaque value.
>>  	 * An implementation may use this field to hold implementation
>specific
>> @@ -1025,8 +1043,14 @@ struct rte_event_vector {
>>   *		// Classify and handle event.
>>   *	}
>>   */
>> +#define RTE_EVENT_TYPE_ETHDEV_VECTOR
>\
>> +	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETHDEV)
>> +/**< The event vector generated from ethdev subsystem */
>>  #define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR
>| RTE_EVENT_TYPE_CPU)
>>  /**< The event vector generated from cpu for pipelining. */
>> +#define RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR
>\
>> +	(RTE_EVENT_TYPE_VECTOR |
>RTE_EVENT_TYPE_ETH_RX_ADAPTER)
>> +/**< The event vector generated from eth Rx adapter. */
>>
>>  #define RTE_EVENT_TYPE_MAX              0x10
>>  /**< Maximum number of event types */
>> @@ -1171,6 +1195,8 @@ struct rte_event {
>>   * @see struct rte_event_eth_rx_adapter_queue_conf::ev
>>   * @see struct
>rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
>>   */
>> +#define RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR
>	0x8
>> +/**< Adapter supports event vectorization per ethdev. */
>>
>>  /**
>>   * Retrieve the event device's ethdev Rx adapter capabilities for the
>> diff --git a/lib/librte_eventdev/version.map
>b/lib/librte_eventdev/version.map
>> index a070ef56e..902df0ae3 100644
>> --- a/lib/librte_eventdev/version.map
>> +++ b/lib/librte_eventdev/version.map
>> @@ -141,6 +141,8 @@ EXPERIMENTAL {
>>
>>  	#added in 21.05
>>  	rte_event_vector_pool_create;
>> +	rte_event_eth_rx_adapter_vector_limits_get;
>> +	rte_event_eth_rx_adapter_queue_event_vector_config;
>>  };
>>
>>  INTERNAL {
>> --
>> 2.17.1
>
>More of informational queries.
>
>Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v5 2/8] eventdev: introduce event vector Rx capability
  2021-03-25  9:24             ` Pavan Nikhilesh Bhagavatula
@ 2021-03-25  9:50               ` Jayatheerthan, Jay
  0 siblings, 0 replies; 153+ messages in thread
From: Jayatheerthan, Jay @ 2021-03-25  9:50 UTC (permalink / raw)
  To: Pavan Nikhilesh Bhagavatula, Jerin Jacob Kollanukkaran, Carrillo,
	Erik G, Gujjar, Abhinandan S, McDaniel, Timothy, hemant.agrawal,
	Van Haaren, Harry, mattias.ronnblom, Ma, Liang J, Ray Kinsella,
	Neil Horman
  Cc: dev

> -----Original Message-----
> From: Pavan Nikhilesh Bhagavatula <pbhagavatula@marvell.com>
> Sent: Thursday, March 25, 2021 2:54 PM
> To: Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Jerin Jacob Kollanukkaran <jerinj@marvell.com>; Carrillo, Erik G
> <erik.g.carrillo@intel.com>; Gujjar, Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy
> <timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
> <mattias.ronnblom@ericsson.com>; Ma, Liang J <liang.j.ma@intel.com>; Ray Kinsella <mdr@ashroe.eu>; Neil Horman
> <nhorman@tuxdriver.com>
> Cc: dev@dpdk.org
> Subject: RE: [dpdk-dev] [PATCH v5 2/8] eventdev: introduce event vector Rx capability
> 
> >> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
> >> Sent: Wednesday, March 24, 2021 10:35 AM
> >> To: jerinj@marvell.com; Jayatheerthan, Jay
> ><jay.jayatheerthan@intel.com>; Carrillo, Erik G
> ><erik.g.carrillo@intel.com>; Gujjar,
> >> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy
> ><timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
> >> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
> ><mattias.ronnblom@ericsson.com>; Ma, Liang J
> >> <liang.j.ma@intel.com>; Ray Kinsella <mdr@ashroe.eu>; Neil Horman
> ><nhorman@tuxdriver.com>
> >> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
> >> Subject: [dpdk-dev] [PATCH v5 2/8] eventdev: introduce event vector
> >Rx capability
> >>
> >> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> >>
> >> Introduce event ethernet Rx adapter event vector capability.
> >>
> >> If an event eth Rx adapter has the capability of
> >> RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR then a given Rx
> >queue
> >> can be configured to enable event vectorization by passing the
> >> flag RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR to
> >> rte_event_eth_rx_adapter_queue_conf::rx_queue_flags while
> >configuring
> >> Rx adapter through rte_event_eth_rx_adapter_queue_add().
> >>
> >> The max vector size, vector timeout define the vector size and
> >> mempool used for allocating vector event are configured through
> >> rte_event_eth_rx_adapter_queue_add. The element size of the
> >element
> >> in the vector pool should be equal to
> >>     sizeof(struct rte_event_vector) + (vector_sz * sizeof(uintptr_t))
> >>
> >> Application can use `rte_event_vector_pool_create` to create the
> >> vector mempool used for
> >> rte_event_eth_rx_adapter_queue_conf::vector_mp.
> >>
> >> The Rx adapter would be responsible for vectorizing the mbufs
> >> based on the flow, the vector limits configured by the application
> >> and add the vector event of mbufs to the event queue set via
> >> rte_event_eth_rx_adapter_queue_conf::ev::queue_id.
> >> It should also mark rte_event_vector::union_valid and fill
> >> rte_event_vector::port, rte_event_vector::queue.
> >>
> >> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> >> Acked-by: Jerin Jacob <jerinj@marvell.com>
> >> ---
> >>  .../prog_guide/event_ethernet_rx_adapter.rst  |  38 ++++++
> >>  lib/librte_eventdev/eventdev_pmd.h            |  53 ++++++++
> >>  .../rte_event_eth_rx_adapter.c                | 114 ++++++++++++++++++
> >>  .../rte_event_eth_rx_adapter.h                | 105 ++++++++++++++++
> >>  lib/librte_eventdev/rte_eventdev.h            |  30 ++++-
> >>  lib/librte_eventdev/version.map               |   2 +
> >>  6 files changed, 340 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
> >b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
> >> index cb44ce0e4..5eefef355 100644
> >> --- a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
> >> +++ b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
> >> @@ -186,3 +186,41 @@ the event buffer fill level is low. The
> >>  ``rte_event_eth_rx_adapter_cb_register()`` function allow the
> >application
> >>  to register a callback that selects which packets to enqueue to the
> >event
> >>  device.
> >> +
> >> +Rx event vectorization
> >> +~~~~~~~~~~~~~~~~~~~~~~
> >> +
> >> +The event devices, ethernet device pairs which support the capability
> >> +``RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR`` can
> >aggregate packets based on
> >> +flow characteristics and generate a ``rte_event`` containing
> >``rte_event_vector``
> >> +whose event type is either ``RTE_EVENT_TYPE_ETHDEV_VECTOR`` or
> >> +``RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR``.
> >> +The aggregation size and timeout are configurable at a queue level
> >and the
> >> +maximum, minimum vector sizes and timeouts vary based on the
> >device capability
> >> +and can be queried using
> >``rte_event_eth_rx_adapter_vector_limits_get``.
> >> +The Rx adapter additionally might include useful data such as
> >ethernet device
> >> +port and queue identifier in the ``rte_event_vector::port`` and
> >> +``rte_event_vector::queue`` and mark
> >``rte_event_vector::attr_valid`` as true.
> >> +
> >> +A loop processing ``rte_event_vector`` containing mbufs is shown
> >below.
> >> +
> >> +.. code-block:: c
> >> +
> >> +        event = rte_event_dequeue_burst(event_dev, event_port,
> >&event,
> >> +                                        1, 0);
> >> +        if (!event)
> >> +                continue;
> >> +
> >> +        switch (ev.event_type) {
> >> +        case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
> >> +        case RTE_EVENT_TYPE_ETHDEV_VECTOR:
> >> +                struct rte_mbufs **mbufs;
> >> +
> >> +                mbufs = (struct rte_mbufs **)ev[i].vec->mbufs;
> >> +                for (i = 0; i < ev.vec->nb_elem; i++) {
> >> +                        /* Process each mbuf. */
> >> +                }
> >> +        break;
> >> +        case ...
> >> +        ...
> >> +        }
> >> diff --git a/lib/librte_eventdev/eventdev_pmd.h
> >b/lib/librte_eventdev/eventdev_pmd.h
> >> index 7eb9a7739..9297f1433 100644
> >> --- a/lib/librte_eventdev/eventdev_pmd.h
> >> +++ b/lib/librte_eventdev/eventdev_pmd.h
> >> @@ -645,6 +645,53 @@ typedef int
> >(*eventdev_eth_rx_adapter_stats_reset)
> >>   */
> >>  typedef int (*eventdev_selftest)(void);
> >>
> >> +struct rte_event_eth_rx_adapter_vector_limits;
> >> +/**
> >> + * Get event vector limits for a given event, ethernet device pair.
> >> + *
> >> + * @param dev
> >> + *   Event device pointer
> >> + *
> >> + * @param eth_dev
> >> + *   Ethernet device pointer
> >> + *
> >> + * @param[out] limits
> >> + *   Pointer to the limits structure to be filled.
> >> + *
> >> + * @return
> >> + *   - 0: Success.
> >> + *   - <0: Error code returned by the driver function.
> >> + */
> >> +typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
> >> +	const struct rte_eventdev *dev, const struct rte_eth_dev
> >*eth_dev,
> >> +	struct rte_event_eth_rx_adapter_vector_limits *limits);
> >> +
> >> +struct rte_event_eth_rx_adapter_event_vector_config;
> >> +/**
> >> + * Enable event vector on an given Rx queue of a ethernet devices
> >belonging to
> >> + * the Rx adapter.
> >> + *
> >> + * @param dev
> >> + *   Event device pointer
> >> + *
> >> + * @param eth_dev
> >> + *   Ethernet device pointer
> >> + *
> >> + * @param rx_queue_id
> >> + *   The Rx queue identifier
> >> + *
> >> + * @param config
> >> + *   Pointer to the event vector configuration structure.
> >> + *
> >> + * @return
> >> + *   - 0: Success.
> >> + *   - <0: Error code returned by the driver function.
> >> + */
> >> +typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
> >> +	const struct rte_eventdev *dev, const struct rte_eth_dev
> >*eth_dev,
> >> +	int32_t rx_queue_id,
> >> +	const struct rte_event_eth_rx_adapter_event_vector_config
> >*config);
> >> +
> >>  typedef uint32_t rte_event_pmd_selftest_seqn_t;
> >>  extern int rte_event_pmd_selftest_seqn_dynfield_offset;
> >>
> >> @@ -1067,6 +1114,12 @@ struct rte_eventdev_ops {
> >>  	/**< Get ethernet Rx stats */
> >>  	eventdev_eth_rx_adapter_stats_reset
> >eth_rx_adapter_stats_reset;
> >>  	/**< Reset ethernet Rx stats */
> >> +	eventdev_eth_rx_adapter_vector_limits_get_t
> >> +		eth_rx_adapter_vector_limits_get;
> >> +	/**< Get event vector limits for the Rx adapter */
> >> +	eventdev_eth_rx_adapter_event_vector_config_t
> >> +		eth_rx_adapter_event_vector_config;
> >> +	/**< Configure Rx adapter with event vector */
> >>
> >>  	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
> >>  	/**< Get timer adapter capabilities */
> >> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> >b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> >> index d8c635e99..ac8ba5bf0 100644
> >> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> >> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> >> @@ -2263,6 +2263,120 @@
> >rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
> >>  	return ret;
> >>  }
> >>
> >> +int
> >> +rte_event_eth_rx_adapter_queue_event_vector_config(
> >> +	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
> >> +	struct rte_event_eth_rx_adapter_event_vector_config *config)
> >> +{
> >> +	struct rte_event_eth_rx_adapter_vector_limits limits;
> >> +	struct rte_event_eth_rx_adapter *rx_adapter;
> >> +	struct rte_eventdev *dev;
> >> +	uint32_t cap;
> >> +	int ret;
> >> +
> >> +	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -
> >EINVAL);
> >> +	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
> >> +
> >> +	rx_adapter = rxa_id_to_adapter(id);
> >> +	if ((rx_adapter == NULL) || (config == NULL))
> >> +		return -EINVAL;
> >> +
> >> +	dev = &rte_eventdevs[rx_adapter->eventdev_id];
> >> +	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter-
> >>eventdev_id,
> >> +						eth_dev_id, &cap);
> >> +	if (ret) {
> >> +		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev
> >%" PRIu8
> >> +				 "eth port %" PRIu16,
> >> +				 id, eth_dev_id);
> >> +		return ret;
> >> +	}
> >> +
> >> +	if (!(cap &
> >RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
> >> +		RTE_EDEV_LOG_ERR("Event vectorization is not
> >supported,"
> >> +				 " eth port: %" PRIu16 " adapter id: %"
> >PRIu8,
> >> +				 eth_dev_id, id);
> >> +		return -EINVAL;
> >> +	}
> >> +
> >> +	ret = rte_event_eth_rx_adapter_vector_limits_get(
> >> +		rx_adapter->eventdev_id, eth_dev_id, &limits);
> >> +	if (ret) {
> >> +		RTE_EDEV_LOG_ERR("Failed to get vector limits edev
> >%" PRIu8
> >> +				 "eth port %" PRIu16,
> >> +				 rx_adapter->eventdev_id, eth_dev_id);
> >> +		return ret;
> >> +	}
> >> +
> >> +	if (config->vector_sz < limits.min_sz ||
> >> +	    config->vector_sz > limits.max_sz ||
> >> +	    config->vector_timeout_ns < limits.min_timeout_ns ||
> >> +	    config->vector_timeout_ns > limits.max_timeout_ns ||
> >> +	    config->vector_mp == NULL) {
> >> +		RTE_EDEV_LOG_ERR("Invalid event vector
> >configuration,"
> >> +				 " eth port: %" PRIu16 " adapter id: %"
> >PRIu8,
> >> +				 eth_dev_id, id);
> >> +		return -EINVAL;
> >> +	}
> >> +	if (config->vector_mp->elt_size <
> >> +	    (sizeof(struct rte_event_vector) +
> >> +	     (sizeof(uintptr_t) * config->vector_sz))) {
> >> +		RTE_EDEV_LOG_ERR("Invalid event vector
> >configuration,"
> >> +				 " eth port: %" PRIu16 " adapter id: %"
> >PRIu8,
> >> +				 eth_dev_id, id);
> >> +		return -EINVAL;
> >> +	}
> >> +
> >> +	if (cap &
> >RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
> >> +		RTE_FUNC_PTR_OR_ERR_RET(
> >> +			*dev->dev_ops-
> >>eth_rx_adapter_event_vector_config,
> >> +			-ENOTSUP);
> >> +		ret = dev->dev_ops-
> >>eth_rx_adapter_event_vector_config(
> >> +			dev, &rte_eth_devices[eth_dev_id],
> >rx_queue_id, config);
> >> +	} else {
> >> +		ret = -ENOTSUP;
> >> +	}
> >
> >Trying to understand why non-INTERNAL_PORT based event device
> >implementation returns ENOTSUP. Do you foresee any issues if such a
> >device implements this function?
> >
> 
> I just didn't want to move the implementation here as this patch is only
> Intended for spec.
> The implementation for non-INTERNAL_PORT is in 4/8

Makes sense. Thanks!

> 
> >> +
> >> +	return ret;
> >> +}
> >> +
> >> +int
> >> +rte_event_eth_rx_adapter_vector_limits_get(
> >> +	uint8_t dev_id, uint16_t eth_port_id,
> >> +	struct rte_event_eth_rx_adapter_vector_limits *limits)
> >> +{
> >> +	struct rte_eventdev *dev;
> >> +	uint32_t cap;
> >> +	int ret;
> >> +
> >> +	RTE_EVENTDEV_VALID_DEVID_OR_ERR_RET(dev_id, -EINVAL);
> >> +	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_port_id, -EINVAL);
> >> +
> >> +	if (limits == NULL)
> >> +		return -EINVAL;
> >> +
> >> +	dev = &rte_eventdevs[dev_id];
> >> +
> >> +	ret = rte_event_eth_rx_adapter_caps_get(dev_id, eth_port_id,
> >&cap);
> >> +	if (ret) {
> >> +		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev
> >%" PRIu8
> >> +				 "eth port %" PRIu16,
> >> +				 dev_id, eth_port_id);
> >> +		return ret;
> >> +	}
> >> +
> >> +	if (cap &
> >RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
> >> +		RTE_FUNC_PTR_OR_ERR_RET(
> >> +			*dev->dev_ops-
> >>eth_rx_adapter_vector_limits_get,
> >> +			-ENOTSUP);
> >> +		ret = dev->dev_ops-
> >>eth_rx_adapter_vector_limits_get(
> >> +			dev, &rte_eth_devices[eth_port_id], limits);
> >> +	} else {
> >> +		ret = -ENOTSUP;
> >> +	}
> >> +
> >
> >Same here.
> >
> >> +	return ret;
> >> +}
> >> +
> >>  int
> >>  rte_event_eth_rx_adapter_start(uint8_t id)
> >>  {
> >> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> >b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> >> index 21bb1e54c..7407cde00 100644
> >> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> >> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> >> @@ -92,6 +92,10 @@ extern "C" {
> >>  /**< This flag indicates the flow identifier is valid
> >>   * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
> >>   */
> >> +#define RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR
> >	0x2
> >> +/**< This flag indicates that mbufs arriving on the queue need to be
> >vectorized
> >> + * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
> >> + */
> >>
> >>  /**
> >>   * Adapter configuration structure that the adapter configuration
> >callback
> >> @@ -169,6 +173,36 @@ struct
> >rte_event_eth_rx_adapter_queue_conf {
> >>  	 */
> >>  };
> >>
> >> +struct rte_event_eth_rx_adapter_event_vector_config {
> >> +	uint16_t vector_sz;
> >> +	/**<
> >> +	 * Indicates the maximum number for mbufs to combine and
> >form a vector.
> >> +	 * Should be within
> >> +	 * @see
> >rte_event_eth_rx_adapter_vector_limits::min_vector_sz
> >> +	 * @see
> >rte_event_eth_rx_adapter_vector_limits::max_vector_sz
> >> +	 * Valid when
> >RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
> >> +	 * @see
> >rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
> >> +	 */
> >> +	uint64_t vector_timeout_ns;
> >> +	/**<
> >> +	 * Indicates the maximum number of nanoseconds to wait for
> >receiving
> >> +	 * mbufs. Should be within vectorization limits of the
> >> +	 * adapter
> >> +	 * @see
> >rte_event_eth_rx_adapter_vector_limits::min_vector_ns
> >> +	 * @see
> >rte_event_eth_rx_adapter_vector_limits::max_vector_ns
> >> +	 * Valid when
> >RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
> >> +	 * @see
> >rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
> >> +	 */
> >> +	struct rte_mempool *vector_mp;
> >> +	/**<
> >> +	 * Indicates the mempool that should be used for allocating
> >> +	 * rte_event_vector container.
> >> +	 * Should be created by using `rte_event_vector_pool_create`.
> >> +	 * Valid when
> >RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
> >> +	 * @see
> >rte_event_eth_rx_adapter_queue_conf::rx_queue_flags.
> >> +	 */
> >> +};
> >> +
> >>  /**
> >>   * A structure used to retrieve statistics for an eth rx adapter
> >instance.
> >>   */
> >> @@ -199,6 +233,32 @@ struct rte_event_eth_rx_adapter_stats {
> >>  	/**< Received packet count for interrupt mode Rx queues */
> >>  };
> >>
> >> +/**
> >> + * A structure used to retrieve eth rx adapter vector limits.
> >> + */
> >> +struct rte_event_eth_rx_adapter_vector_limits {
> >> +	uint16_t min_sz;
> >> +	/**< Minimum vector limit configurable.
> >> +	 * @see
> >rte_event_eth_rx_adapter_event_vector_config::vector_sz
> >> +	 */
> >> +	uint16_t max_sz;
> >> +	/**< Maximum vector limit configurable.
> >> +	 * @see
> >rte_event_eth_rx_adapter_event_vector_config::vector_sz
> >> +	 */
> >> +	uint8_t log2_sz;
> >> +	/**< True if the size configured should be in log2.
> >> +	 * @see
> >rte_event_eth_rx_adapter_event_vector_config::vector_sz
> >> +	 */
> >> +	uint64_t min_timeout_ns;
> >> +	/**< Minimum vector timeout configurable.
> >> +	 * @see
> >rte_event_eth_rx_adapter_event_vector_config::vector_timeout_ns
> >> +	 */
> >> +	uint64_t max_timeout_ns;
> >> +	/**< Maximum vector timeout configurable.
> >> +	 * @see
> >rte_event_eth_rx_adapter_event_vector_config::vector_timeout_ns
> >> +	 */
> >> +};
> >> +
> >>  /**
> >>   *
> >>   * Callback function invoked by the SW adapter before it continues
> >> @@ -467,6 +527,51 @@ int
> >rte_event_eth_rx_adapter_cb_register(uint8_t id, uint16_t eth_dev_id,
> >>
> >rte_event_eth_rx_adapter_cb_fn cb_fn,
> >>  					 void *cb_arg);
> >>
> >> +/**
> >> + * Retrieve vector limits for a given event dev and eth dev pair.
> >> + * @see rte_event_eth_rx_adapter_vector_limits
> >> + *
> >> + * @param dev_id
> >> + *  Event device identifier.
> >> + * @param eth_port_id
> >> + *  Port identifier of the ethernet device.
> >> + * @param [out] limits
> >> + *  A pointer to rte_event_eth_rx_adapter_vector_limits structure
> >that has to
> >> + * be filled.
> >> + *
> >> + * @return
> >> + *  - 0: Success.
> >> + *  - <0: Error code on failure.
> >> + */
> >> +__rte_experimental
> >> +int rte_event_eth_rx_adapter_vector_limits_get(
> >> +	uint8_t dev_id, uint16_t eth_port_id,
> >> +	struct rte_event_eth_rx_adapter_vector_limits *limits);
> >> +
> >> +/**
> >> + * Configure event vectorization for a given ethernet device queue,
> >that has
> >> + * been added to a event eth Rx adapter.
> >> + *
> >> + * @param id
> >> + *  The identifier of the ethernet Rx event adapter.
> >> + *
> >> + * @param eth_dev_id
> >> + *  The identifier of the ethernet device.
> >> + *
> >> + * @param rx_queue_id
> >> + *  Ethernet device receive queue index.
> >> + *  If rx_queue_id is -1, then all Rx queues configured for the
> >ethernet device
> >> + *  are configured with event vectorization.
> >> + *
> >> + * @return
> >> + *  - 0: Success, Receive queue configured correctly.
> >> + *  - <0: Error code on failure.
> >> + */
> >> +__rte_experimental
> >> +int rte_event_eth_rx_adapter_queue_event_vector_config(
> >> +	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
> >> +	struct rte_event_eth_rx_adapter_event_vector_config
> >*config);
> >> +
> >>  #ifdef __cplusplus
> >>  }
> >>  #endif
> >> diff --git a/lib/librte_eventdev/rte_eventdev.h
> >b/lib/librte_eventdev/rte_eventdev.h
> >> index aa4dd3959..678338247 100644
> >> --- a/lib/librte_eventdev/rte_eventdev.h
> >> +++ b/lib/librte_eventdev/rte_eventdev.h
> >> @@ -919,10 +919,28 @@ rte_event_dev_close(uint8_t dev_id);
> >>   * Event vector structure.
> >>   */
> >>  struct rte_event_vector {
> >> -	uint64_t nb_elem : 16;
> >> +	uint16_t nb_elem;
> >>  	/**< Number of elements in this event vector. */
> >> -	uint64_t rsvd : 48;
> >> +	uint16_t rsvd : 15;
> >>  	/**< Reserved for future use */
> >> +	uint16_t attr_valid : 1;
> >> +	/**< Indicates that the below union attributes have valid
> >information.
> >> +	 */
> >> +	union {
> >> +		/* Used by Rx adapter.
> >> +		 * Indicates that all the elements in this vector belong to
> >the
> >> +		 * same port and queue pair when originating from Rx
> >adapter,
> >> +		 * valid only when event type is ETHDEV_VECTOR or
> >> +		 * ETH_RX_ADAPTER_VECTOR.
> >> +		 */
> >> +		struct {
> >> +			uint16_t port;
> >> +			/* Ethernet device port id. */
> >> +			uint16_t queue;
> >> +			/* Ethernet device queue id. */
> >> +		};
> >> +	};
> >> +	/**< Union to hold common attributes of the vector array. */
> >>  	uint64_t impl_opaque;
> >>  	/**< Implementation specific opaque value.
> >>  	 * An implementation may use this field to hold implementation
> >specific
> >> @@ -1025,8 +1043,14 @@ struct rte_event_vector {
> >>   *		// Classify and handle event.
> >>   *	}
> >>   */
> >> +#define RTE_EVENT_TYPE_ETHDEV_VECTOR
> >\
> >> +	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETHDEV)
> >> +/**< The event vector generated from ethdev subsystem */
> >>  #define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR
> >| RTE_EVENT_TYPE_CPU)
> >>  /**< The event vector generated from cpu for pipelining. */
> >> +#define RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR
> >\
> >> +	(RTE_EVENT_TYPE_VECTOR |
> >RTE_EVENT_TYPE_ETH_RX_ADAPTER)
> >> +/**< The event vector generated from eth Rx adapter. */
> >>
> >>  #define RTE_EVENT_TYPE_MAX              0x10
> >>  /**< Maximum number of event types */
> >> @@ -1171,6 +1195,8 @@ struct rte_event {
> >>   * @see struct rte_event_eth_rx_adapter_queue_conf::ev
> >>   * @see struct
> >rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
> >>   */
> >> +#define RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR
> >	0x8
> >> +/**< Adapter supports event vectorization per ethdev. */
> >>
> >>  /**
> >>   * Retrieve the event device's ethdev Rx adapter capabilities for the
> >> diff --git a/lib/librte_eventdev/version.map
> >b/lib/librte_eventdev/version.map
> >> index a070ef56e..902df0ae3 100644
> >> --- a/lib/librte_eventdev/version.map
> >> +++ b/lib/librte_eventdev/version.map
> >> @@ -141,6 +141,8 @@ EXPERIMENTAL {
> >>
> >>  	#added in 21.05
> >>  	rte_event_vector_pool_create;
> >> +	rte_event_eth_rx_adapter_vector_limits_get;
> >> +	rte_event_eth_rx_adapter_queue_event_vector_config;
> >>  };
> >>
> >>  INTERNAL {
> >> --
> >> 2.17.1
> >
> >More of informational queries.
> >
> >Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v5 4/8] eventdev: add Rx adapter event vector support
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 4/8] eventdev: add Rx adapter event vector support pbhagavatula
@ 2021-03-25 10:37           ` Jayatheerthan, Jay
  2021-03-25 13:14             ` Pavan Nikhilesh Bhagavatula
  0 siblings, 1 reply; 153+ messages in thread
From: Jayatheerthan, Jay @ 2021-03-25 10:37 UTC (permalink / raw)
  To: pbhagavatula, jerinj, Carrillo, Erik G, Gujjar, Abhinandan S,
	McDaniel, Timothy, hemant.agrawal, Van Haaren, Harry,
	mattias.ronnblom, Ma, Liang J
  Cc: dev

> -----Original Message-----
> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
> Sent: Wednesday, March 24, 2021 10:35 AM
> To: jerinj@marvell.com; Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Carrillo, Erik G <erik.g.carrillo@intel.com>; Gujjar,
> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy <timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom <mattias.ronnblom@ericsson.com>; Ma, Liang J
> <liang.j.ma@intel.com>
> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
> Subject: [dpdk-dev] [PATCH v5 4/8] eventdev: add Rx adapter event vector support
> 
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> Add event vector support for event eth Rx adapter, the implementation
> creates vector flows based on port and queue identifier of the received
> mbufs.
> 
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> ---
>  lib/librte_eventdev/eventdev_pmd.h            |   7 +-
>  .../rte_event_eth_rx_adapter.c                | 257 ++++++++++++++++--
>  lib/librte_eventdev/rte_eventdev.c            |   6 +-
>  3 files changed, 250 insertions(+), 20 deletions(-)
> 
> diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
> index 9297f1433..0f724ac85 100644
> --- a/lib/librte_eventdev/eventdev_pmd.h
> +++ b/lib/librte_eventdev/eventdev_pmd.h
> @@ -69,9 +69,10 @@ extern "C" {
>  	} \
>  } while (0)
> 
> -#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP \
> -		((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) | \
> -			(RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ))
> +#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP                                        \
> +	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |                     \
> +	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) |                         \
> +	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR))
> 
>  #define RTE_EVENT_CRYPTO_ADAPTER_SW_CAP \
>  		RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA
> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> index ac8ba5bf0..c71990078 100644
> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> @@ -26,6 +26,10 @@
>  #define BATCH_SIZE		32
>  #define BLOCK_CNT_THRESHOLD	10
>  #define ETH_EVENT_BUFFER_SIZE	(4*BATCH_SIZE)
> +#define MAX_VECTOR_SIZE		1024
> +#define MIN_VECTOR_SIZE		4
> +#define MAX_VECTOR_NS		1E9
> +#define MIN_VECTOR_NS		1E5
> 
>  #define ETH_RX_ADAPTER_SERVICE_NAME_LEN	32
>  #define ETH_RX_ADAPTER_MEM_NAME_LEN	32
> @@ -59,6 +63,20 @@ struct eth_rx_poll_entry {
>  	uint16_t eth_rx_qid;
>  };
> 
> +struct eth_rx_vector_data {
> +	TAILQ_ENTRY(eth_rx_vector_data) next;
> +	uint16_t port;
> +	uint16_t queue;
> +	uint16_t max_vector_count;
> +	uint64_t event;
> +	uint64_t ts;
> +	uint64_t vector_timeout_ticks;
> +	struct rte_mempool *vector_pool;
> +	struct rte_event_vector *vector_ev;
> +} __rte_cache_aligned;
> +
> +TAILQ_HEAD(eth_rx_vector_data_list, eth_rx_vector_data);
> +
>  /* Instance per adapter */
>  struct rte_eth_event_enqueue_buffer {
>  	/* Count of events in this buffer */
> @@ -92,6 +110,14 @@ struct rte_event_eth_rx_adapter {
>  	uint32_t wrr_pos;
>  	/* Event burst buffer */
>  	struct rte_eth_event_enqueue_buffer event_enqueue_buffer;
> +	/* Vector enable flag */
> +	uint8_t ena_vector;
> +	/* Timestamp of previous vector expiry list traversal */
> +	uint64_t prev_expiry_ts;
> +	/* Minimum ticks to wait before traversing expiry list */
> +	uint64_t vector_tmo_ticks;
> +	/* vector list */
> +	struct eth_rx_vector_data_list vector_list;
>  	/* Per adapter stats */
>  	struct rte_event_eth_rx_adapter_stats stats;
>  	/* Block count, counts up to BLOCK_CNT_THRESHOLD */
> @@ -198,9 +224,11 @@ struct eth_device_info {
>  struct eth_rx_queue_info {
>  	int queue_enabled;	/* True if added */
>  	int intr_enabled;
> +	uint8_t ena_vector;
>  	uint16_t wt;		/* Polling weight */
>  	uint32_t flow_id_mask;	/* Set to ~0 if app provides flow id else 0 */
>  	uint64_t event;
> +	struct eth_rx_vector_data vector_data;
>  };
> 
>  static struct rte_event_eth_rx_adapter **event_eth_rx_adapter;
> @@ -722,6 +750,9 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
>  	    &rx_adapter->event_enqueue_buffer;
>  	struct rte_event_eth_rx_adapter_stats *stats = &rx_adapter->stats;
> 
> +	if (!buf->count)
> +		return 0;
> +
>  	uint16_t n = rte_event_enqueue_new_burst(rx_adapter->eventdev_id,
>  					rx_adapter->event_port_id,
>  					buf->events,
> @@ -742,6 +773,72 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
>  	return n;
>  }
> 
> +static inline uint16_t
> +rxa_create_event_vector(struct rte_event_eth_rx_adapter *rx_adapter,
> +			struct eth_rx_queue_info *queue_info,
> +			struct rte_eth_event_enqueue_buffer *buf,
> +			struct rte_mbuf **mbufs, uint16_t num)
> +{
> +	struct rte_event *ev = &buf->events[buf->count];
> +	struct eth_rx_vector_data *vec;
> +	uint16_t filled, space, sz;
> +
> +	filled = 0;
> +	vec = &queue_info->vector_data;
> +	while (num) {
> +		if (vec->vector_ev == NULL) {
> +			if (rte_mempool_get(vec->vector_pool,
> +					    (void **)&vec->vector_ev) < 0) {
> +				rte_pktmbuf_free_bulk(mbufs, num);
> +				return 0;
> +			}
> +			vec->vector_ev->nb_elem = 0;
> +			vec->vector_ev->port = vec->port;
> +			vec->vector_ev->queue = vec->queue;
> +			vec->vector_ev->attr_valid = true;
> +			TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
> +		} else if (vec->vector_ev->nb_elem == vec->max_vector_count) {

Is there a case where nb_elem > max_vector_count as we accumulate sz to it ?

> +			/* Event ready. */
> +			ev->event = vec->event;
> +			ev->vec = vec->vector_ev;
> +			ev++;
> +			filled++;
> +			vec->vector_ev = NULL;
> +			TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
> +			if (rte_mempool_get(vec->vector_pool,
> +					    (void **)&vec->vector_ev) < 0) {
> +				rte_pktmbuf_free_bulk(mbufs, num);
> +				return 0;
> +			}
> +			vec->vector_ev->nb_elem = 0;
> +			vec->vector_ev->port = vec->port;
> +			vec->vector_ev->queue = vec->queue;
> +			vec->vector_ev->attr_valid = true;
> +			TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
> +		}
> +
> +		space = vec->max_vector_count - vec->vector_ev->nb_elem;
> +		sz = num > space ? space : num;
> +		memcpy(vec->vector_ev->mbufs + vec->vector_ev->nb_elem, mbufs,
> +		       sizeof(void *) * sz);
> +		vec->vector_ev->nb_elem += sz;
> +		num -= sz;
> +		mbufs += sz;
> +		vec->ts = rte_rdtsc();
> +	}
> +
> +	if (vec->vector_ev->nb_elem == vec->max_vector_count) {

Same here.

> +		ev->event = vec->event;
> +		ev->vec = vec->vector_ev;
> +		ev++;
> +		filled++;
> +		vec->vector_ev = NULL;
> +		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
> +	}
> +
> +	return filled;
> +}

I am seeing more than one repeating code chunks in this function. Perhaps, you can give it a try to not repeat. We can drop if its performance affecting.

> +
>  static inline void
>  rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
>  		uint16_t eth_dev_id,
> @@ -770,25 +867,30 @@ rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
>  	rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
>  	do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;

The RSS related code is executed for vector case as well. Can this be moved inside ena_vector if condition ?

> 
> -	for (i = 0; i < num; i++) {
> -		m = mbufs[i];
> -
> -		rss = do_rss ?
> -			rxa_do_softrss(m, rx_adapter->rss_key_be) :
> -			m->hash.rss;
> -		ev->event = event;
> -		ev->flow_id = (rss & ~flow_id_mask) |
> -				(ev->flow_id & flow_id_mask);
> -		ev->mbuf = m;
> -		ev++;
> +	if (!eth_rx_queue_info->ena_vector) {
> +		for (i = 0; i < num; i++) {
> +			m = mbufs[i];
> +
> +			rss = do_rss ? rxa_do_softrss(m, rx_adapter->rss_key_be)
> +				     : m->hash.rss;
> +			ev->event = event;
> +			ev->flow_id = (rss & ~flow_id_mask) |
> +				      (ev->flow_id & flow_id_mask);
> +			ev->mbuf = m;
> +			ev++;
> +		}
> +	} else {
> +		num = rxa_create_event_vector(rx_adapter, eth_rx_queue_info,
> +					      buf, mbufs, num);
>  	}
> 
> -	if (dev_info->cb_fn) {
> +	if (num && dev_info->cb_fn) {
> 
>  		dropped = 0;
>  		nb_cb = dev_info->cb_fn(eth_dev_id, rx_queue_id,
> -					ETH_EVENT_BUFFER_SIZE, buf->count, ev,
> -					num, dev_info->cb_arg, &dropped);
> +					ETH_EVENT_BUFFER_SIZE, buf->count,
> +					&buf->events[buf->count], num,
> +					dev_info->cb_arg, &dropped);

Before this patch, we pass ev which is &buf->events[buf->count] + num as fifth param when calling cb_fn. Now, we are passing &buf->events[buf->count] for non-vector case. Do you see this as an issue?

Also, for vector case would it make sense to do pass &buf->events[buf->count] + num ?

>  		if (unlikely(nb_cb > num))
>  			RTE_EDEV_LOG_ERR("Rx CB returned %d (> %d) events",
>  				nb_cb, num);
> @@ -1124,6 +1226,30 @@ rxa_poll(struct rte_event_eth_rx_adapter *rx_adapter)
>  	return nb_rx;
>  }
> 
> +static void
> +rxa_vector_expire(struct eth_rx_vector_data *vec, void *arg)
> +{
> +	struct rte_event_eth_rx_adapter *rx_adapter = arg;
> +	struct rte_eth_event_enqueue_buffer *buf =
> +		&rx_adapter->event_enqueue_buffer;
> +	struct rte_event *ev;
> +
> +	if (buf->count)
> +		rxa_flush_event_buffer(rx_adapter);
> +
> +	if (vec->vector_ev->nb_elem == 0)
> +		return;
> +	ev = &buf->events[buf->count];
> +
> +	/* Event ready. */
> +	ev->event = vec->event;
> +	ev->vec = vec->vector_ev;
> +	buf->count++;
> +
> +	vec->vector_ev = NULL;
> +	vec->ts = 0;
> +}
> +
>  static int
>  rxa_service_func(void *args)
>  {
> @@ -1137,6 +1263,24 @@ rxa_service_func(void *args)
>  		return 0;
>  	}
> 
> +	if (rx_adapter->ena_vector) {
> +		if ((rte_rdtsc() - rx_adapter->prev_expiry_ts) >=
> +		    rx_adapter->vector_tmo_ticks) {
> +			struct eth_rx_vector_data *vec;
> +
> +			TAILQ_FOREACH(vec, &rx_adapter->vector_list, next) {
> +				uint64_t elapsed_time = rte_rdtsc() - vec->ts;
> +
> +				if (elapsed_time >= vec->vector_timeout_ticks) {
> +					rxa_vector_expire(vec, rx_adapter);
> +					TAILQ_REMOVE(&rx_adapter->vector_list,
> +						     vec, next);
> +				}
> +			}
> +			rx_adapter->prev_expiry_ts = rte_rdtsc();
> +		}
> +	}
> +
>  	stats = &rx_adapter->stats;
>  	stats->rx_packets += rxa_intr_ring_dequeue(rx_adapter);
>  	stats->rx_packets += rxa_poll(rx_adapter);
> @@ -1640,6 +1784,28 @@ rxa_update_queue(struct rte_event_eth_rx_adapter *rx_adapter,
>  	}
>  }
> 
> +static void
> +rxa_set_vector_data(struct eth_rx_queue_info *queue_info, uint16_t vector_count,
> +		    uint64_t vector_ns, struct rte_mempool *mp, int32_t qid,
> +		    uint16_t port_id)
> +{
> +#define NSEC2TICK(__ns, __freq) (((__ns) * (__freq)) / 1E9)
> +	struct eth_rx_vector_data *vector_data;
> +	uint32_t flow_id;
> +
> +	vector_data = &queue_info->vector_data;
> +	vector_data->max_vector_count = vector_count;
> +	vector_data->port = port_id;
> +	vector_data->queue = qid;
> +	vector_data->vector_pool = mp;
> +	vector_data->vector_timeout_ticks =
> +		NSEC2TICK(vector_ns, rte_get_timer_hz());
> +	vector_data->ts = 0;
> +	flow_id = queue_info->event & 0xFFFFF;
> +	flow_id = flow_id == 0 ? (qid & 0xFF) | (port_id & 0xFFFF) : flow_id;

Maybe I am missing something here. Looking at the code it looks like qid and port_id may overlap. For e.g., if qid = 0x10 and port_id = 0x11, flow_id would end up being 0x11. Is this the expectation? Also, it may be useful to document flow_id format.
Comparing this format with existing RSS hash based method, are we saying that all mbufs received in a rx burst are part of same flow when vectorization is used?

> +	vector_data->event = (queue_info->event & ~0xFFFFF) | flow_id;
> +}
> +
>  static void
>  rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
>  	struct eth_device_info *dev_info,
> @@ -1741,6 +1907,44 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
>  	}
>  }
> 
> +static void
> +rxa_sw_event_vector_configure(
> +	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
> +	int rx_queue_id,
> +	const struct rte_event_eth_rx_adapter_event_vector_config *config)
> +{
> +	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
> +	struct eth_rx_queue_info *queue_info;
> +	struct rte_event *qi_ev;
> +
> +	if (rx_queue_id == -1) {
> +		uint16_t nb_rx_queues;
> +		uint16_t i;
> +
> +		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
> +		for (i = 0; i < nb_rx_queues; i++)
> +			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
> +						      config);
> +		return;
> +	}
> +
> +	queue_info = &dev_info->rx_queue[rx_queue_id];
> +	qi_ev = (struct rte_event *)&queue_info->event;
> +	queue_info->ena_vector = 1;
> +	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
> +	rxa_set_vector_data(queue_info, config->vector_sz,
> +			    config->vector_timeout_ns, config->vector_mp,
> +			    rx_queue_id, dev_info->dev->data->port_id);
> +	rx_adapter->ena_vector = 1;
> +	rx_adapter->vector_tmo_ticks =
> +		rx_adapter->vector_tmo_ticks ?
> +			      RTE_MIN(config->vector_timeout_ns << 1,
> +				      rx_adapter->vector_tmo_ticks) :
> +			      config->vector_timeout_ns << 1;
> +	rx_adapter->prev_expiry_ts = 0;
> +	TAILQ_INIT(&rx_adapter->vector_list);
> +}
> +
>  static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
>  		uint16_t eth_dev_id,
>  		int rx_queue_id,
> @@ -2081,6 +2285,15 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
>  		return -EINVAL;
>  	}
> 
> +	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
> +	    (queue_conf->rx_queue_flags &
> +	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
> +		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
> +				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
> +				 eth_dev_id, id);
> +		return -EINVAL;
> +	}
> +
>  	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
>  		(rx_queue_id != -1)) {
>  		RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
> @@ -2143,6 +2356,17 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
>  	return 0;
>  }
> 
> +static int
> +rxa_sw_vector_limits(struct rte_event_eth_rx_adapter_vector_limits *limits)
> +{
> +	limits->max_sz = MAX_VECTOR_SIZE;
> +	limits->min_sz = MIN_VECTOR_SIZE;
> +	limits->max_timeout_ns = MAX_VECTOR_NS;
> +	limits->min_timeout_ns = MIN_VECTOR_NS;
> +
> +	return 0;
> +}
> +
>  int
>  rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
>  				int32_t rx_queue_id)
> @@ -2333,7 +2557,8 @@ rte_event_eth_rx_adapter_queue_event_vector_config(
>  		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
>  			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
>  	} else {
> -		ret = -ENOTSUP;
> +		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
> +					      rx_queue_id, config);
>  	}
> 
>  	return ret;
> @@ -2371,7 +2596,7 @@ rte_event_eth_rx_adapter_vector_limits_get(
>  		ret = dev->dev_ops->eth_rx_adapter_vector_limits_get(
>  			dev, &rte_eth_devices[eth_port_id], limits);
>  	} else {
> -		ret = -ENOTSUP;
> +		ret = rxa_sw_vector_limits(limits);
>  	}
> 
>  	return ret;
> diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
> index f95edc075..254a31b1f 100644
> --- a/lib/librte_eventdev/rte_eventdev.c
> +++ b/lib/librte_eventdev/rte_eventdev.c
> @@ -122,7 +122,11 @@ rte_event_eth_rx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
> 
>  	if (caps == NULL)
>  		return -EINVAL;
> -	*caps = 0;
> +
> +	if (dev->dev_ops->eth_rx_adapter_caps_get == NULL)
> +		*caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
> +	else
> +		*caps = 0;

Any reason why we had to set default caps value? I am thinking if sw event device is used, it would set it anyways.

> 
>  	return dev->dev_ops->eth_rx_adapter_caps_get ?
>  				(*dev->dev_ops->eth_rx_adapter_caps_get)(dev,
> --
> 2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v5 5/8] eventdev: add Tx adapter event vector support
  2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 5/8] eventdev: add Tx " pbhagavatula
@ 2021-03-25 11:44           ` Jayatheerthan, Jay
  0 siblings, 0 replies; 153+ messages in thread
From: Jayatheerthan, Jay @ 2021-03-25 11:44 UTC (permalink / raw)
  To: pbhagavatula, jerinj, Carrillo, Erik G, Gujjar, Abhinandan S,
	McDaniel, Timothy, hemant.agrawal, Van Haaren, Harry,
	mattias.ronnblom, Ma, Liang J
  Cc: dev

> -----Original Message-----
> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
> Sent: Wednesday, March 24, 2021 10:35 AM
> To: jerinj@marvell.com; Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Carrillo, Erik G <erik.g.carrillo@intel.com>; Gujjar,
> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy <timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom <mattias.ronnblom@ericsson.com>; Ma, Liang J
> <liang.j.ma@intel.com>
> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
> Subject: [dpdk-dev] [PATCH v5 5/8] eventdev: add Tx adapter event vector support
> 
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> Add event vector support for event eth Tx adapter, the implementation
> receives events from the single linked queue and based on
> rte_event_vector::union_valid transmits the vector of mbufs to a given

Typo: attr_valid instead of union_valid

> port, queue pair.
> 
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> ---
>  .../rte_event_eth_tx_adapter.c                | 66 ++++++++++++++++---
>  lib/librte_eventdev/rte_eventdev.c            |  5 +-
>  2 files changed, 60 insertions(+), 11 deletions(-)
> 
> diff --git a/lib/librte_eventdev/rte_event_eth_tx_adapter.c b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
> index 5b4c42dcf..db260bfb6 100644
> --- a/lib/librte_eventdev/rte_event_eth_tx_adapter.c
> +++ b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
> @@ -510,6 +510,47 @@ txa_service_buffer_retry(struct rte_mbuf **pkts, uint16_t unsent,
>  	stats->tx_dropped += unsent - sent;
>  }
> 
> +static uint16_t
> +txa_process_event_vector(struct txa_service_data *txa,
> +			 struct rte_event_vector *vec)
> +{
> +	struct txa_service_queue_info *tqi;
> +	uint16_t port, queue, nb_tx = 0;
> +	struct rte_mbuf **mbufs;
> +	int i;
> +
> +	mbufs = (struct rte_mbuf **)vec->mbufs;
> +	if (vec->attr_valid) {
> +		port = vec->port;
> +		queue = vec->queue;
> +		tqi = txa_service_queue(txa, port, queue);
> +		if (unlikely(tqi == NULL || !tqi->added)) {
> +			rte_pktmbuf_free_bulk(mbufs, vec->nb_elem);
> +			rte_mempool_put(rte_mempool_from_obj(vec), vec);
> +			return 0;
> +		}
> +		for (i = 0; i < vec->nb_elem; i++) {
> +			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
> +						   mbufs[i]);
> +		}
> +	} else {
> +		for (i = 0; i < vec->nb_elem; i++) {
> +			port = mbufs[i]->port;
> +			queue = rte_event_eth_tx_adapter_txq_get(mbufs[i]);
> +			tqi = txa_service_queue(txa, port, queue);
> +			if (unlikely(tqi == NULL || !tqi->added)) {
> +				rte_pktmbuf_free(mbufs[i]);
> +				continue;
> +			}
> +			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
> +						   mbufs[i]);
> +		}
> +	}
> +	rte_mempool_put(rte_mempool_from_obj(vec), vec);
> +
> +	return nb_tx;
> +}
> +
>  static void
>  txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
>  	uint32_t n)
> @@ -522,22 +563,27 @@ txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
> 
>  	nb_tx = 0;
>  	for (i = 0; i < n; i++) {
> -		struct rte_mbuf *m;
>  		uint16_t port;
>  		uint16_t queue;
>  		struct txa_service_queue_info *tqi;
> 
> -		m = ev[i].mbuf;
> -		port = m->port;
> -		queue = rte_event_eth_tx_adapter_txq_get(m);
> +		if (!(ev[i].event_type & RTE_EVENT_TYPE_VECTOR)) {
> +			struct rte_mbuf *m;
> 
> -		tqi = txa_service_queue(txa, port, queue);
> -		if (unlikely(tqi == NULL || !tqi->added)) {
> -			rte_pktmbuf_free(m);
> -			continue;
> -		}
> +			m = ev[i].mbuf;
> +			port = m->port;
> +			queue = rte_event_eth_tx_adapter_txq_get(m);
> 
> -		nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
> +			tqi = txa_service_queue(txa, port, queue);
> +			if (unlikely(tqi == NULL || !tqi->added)) {
> +				rte_pktmbuf_free(m);
> +				continue;
> +			}
> +
> +			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
> +		} else {
> +			nb_tx += txa_process_event_vector(txa, ev[i].vec);
> +		}
>  	}
> 
>  	stats->tx_packets += nb_tx;
> diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
> index 254a31b1f..ed6b5ba59 100644
> --- a/lib/librte_eventdev/rte_eventdev.c
> +++ b/lib/librte_eventdev/rte_eventdev.c
> @@ -196,7 +196,10 @@ rte_event_eth_tx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
>  	if (caps == NULL)
>  		return -EINVAL;
> 
> -	*caps = 0;
> +	if (dev->dev_ops->eth_tx_adapter_caps_get == NULL)
> +		*caps = RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR;
> +	else
> +		*caps = 0;
> 
>  	return dev->dev_ops->eth_tx_adapter_caps_get ?
>  			(*dev->dev_ops->eth_tx_adapter_caps_get)(dev,
> --
> 2.17.1

With changes above, looks good.

Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [dpdk-dev v21.11] [PATCH v5 8/8] eventdev: simplify Rx adapter event vector config
  2021-03-24  5:05         ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v5 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
@ 2021-03-25 12:27           ` Jayatheerthan, Jay
  2021-03-25 13:55             ` Pavan Nikhilesh Bhagavatula
  0 siblings, 1 reply; 153+ messages in thread
From: Jayatheerthan, Jay @ 2021-03-25 12:27 UTC (permalink / raw)
  To: pbhagavatula, jerinj, Carrillo, Erik G, Gujjar, Abhinandan S,
	McDaniel, Timothy, hemant.agrawal, Van Haaren, Harry,
	mattias.ronnblom, Ma, Liang J, Ray Kinsella, Neil Horman
  Cc: dev

> -----Original Message-----
> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
> Sent: Wednesday, March 24, 2021 10:35 AM
> To: jerinj@marvell.com; Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Carrillo, Erik G <erik.g.carrillo@intel.com>; Gujjar,
> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy <timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom <mattias.ronnblom@ericsson.com>; Ma, Liang J
> <liang.j.ma@intel.com>; Ray Kinsella <mdr@ashroe.eu>; Neil Horman <nhorman@tuxdriver.com>
> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
> Subject: [dpdk-dev v21.11] [PATCH v5 8/8] eventdev: simplify Rx adapter event vector config
> 
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> Include vector configuration into the structure
> ``rte_event_eth_rx_adapter_queue_conf`` used when configuring rest
> of the Rx adapter ethernet device Rx queue parameters.
> This simplifies event vector configuration as it avoids splitting
> configuration per Rx queue.
> 
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> ---
>  app/test-eventdev/test_pipeline_common.c      |  16 +-
>  lib/librte_eventdev/eventdev_pmd.h            |  29 ---
>  .../rte_event_eth_rx_adapter.c                | 168 ++++++------------
>  .../rte_event_eth_rx_adapter.h                |  27 ---
>  lib/librte_eventdev/version.map               |   1 -
>  5 files changed, 57 insertions(+), 184 deletions(-)
> 
> diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
> index d5ef90500..76aee254b 100644
> --- a/app/test-eventdev/test_pipeline_common.c
> +++ b/app/test-eventdev/test_pipeline_common.c
> @@ -331,7 +331,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
>  	uint16_t prod;
>  	struct rte_mempool *vector_pool = NULL;
>  	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
> -	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;
> 
>  	memset(&queue_conf, 0,
>  			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
> @@ -397,8 +396,12 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
>  			}
> 
>  			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
> +				queue_conf.vector_sz = opt->vector_size;
> +				queue_conf.vector_timeout_ns =
> +					opt->vector_tmo_nsec;
>  				queue_conf.rx_queue_flags |=
>  				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
> +				queue_conf.vector_mp = vector_pool;
>  			} else {
>  				evt_err("Rx adapter doesn't support event vector");
>  				return -EINVAL;
> @@ -418,17 +421,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
>  			return ret;
>  		}
> 
> -		if (opt->ena_vector) {
> -			vec_conf.vector_sz = opt->vector_size;
> -			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
> -			vec_conf.vector_mp = vector_pool;
> -			if (rte_event_eth_rx_adapter_queue_event_vector_config(
> -				    prod, prod, -1, &vec_conf) < 0) {
> -				evt_err("Failed to configure event vectorization for Rx adapter");
> -				return -EINVAL;
> -			}
> -		}
> -
>  		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
>  			uint32_t service_id = -1U;
> 
> diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
> index 0f724ac85..63b3bc4b5 100644
> --- a/lib/librte_eventdev/eventdev_pmd.h
> +++ b/lib/librte_eventdev/eventdev_pmd.h
> @@ -667,32 +667,6 @@ typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
>  	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
>  	struct rte_event_eth_rx_adapter_vector_limits *limits);
> 
> -struct rte_event_eth_rx_adapter_event_vector_config;
> -/**
> - * Enable event vector on an given Rx queue of a ethernet devices belonging to
> - * the Rx adapter.
> - *
> - * @param dev
> - *   Event device pointer
> - *
> - * @param eth_dev
> - *   Ethernet device pointer
> - *
> - * @param rx_queue_id
> - *   The Rx queue identifier
> - *
> - * @param config
> - *   Pointer to the event vector configuration structure.
> - *
> - * @return
> - *   - 0: Success.
> - *   - <0: Error code returned by the driver function.
> - */
> -typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
> -	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
> -	int32_t rx_queue_id,
> -	const struct rte_event_eth_rx_adapter_event_vector_config *config);
> -
>  typedef uint32_t rte_event_pmd_selftest_seqn_t;
>  extern int rte_event_pmd_selftest_seqn_dynfield_offset;
> 
> @@ -1118,9 +1092,6 @@ struct rte_eventdev_ops {
>  	eventdev_eth_rx_adapter_vector_limits_get_t
>  		eth_rx_adapter_vector_limits_get;
>  	/**< Get event vector limits for the Rx adapter */
> -	eventdev_eth_rx_adapter_event_vector_config_t
> -		eth_rx_adapter_event_vector_config;
> -	/**< Configure Rx adapter with event vector */
> 
>  	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
>  	/**< Get timer adapter capabilities */
> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> index c71990078..a1990637f 100644
> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> @@ -1882,6 +1882,25 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
>  	} else
>  		qi_ev->flow_id = 0;
> 
> +	if (conf->rx_queue_flags &
> +	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
> +		queue_info->ena_vector = 1;
> +		qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
> +		rxa_set_vector_data(queue_info, conf->vector_sz,
> +				    conf->vector_timeout_ns, conf->vector_mp,
> +				    rx_queue_id, dev_info->dev->data->port_id);
> +		rx_adapter->ena_vector = 1;
> +		rx_adapter->vector_tmo_ticks =
> +			rx_adapter->vector_tmo_ticks
> +				? RTE_MIN(queue_info->vector_data
> +						  .vector_timeout_ticks,
> +					  rx_adapter->vector_tmo_ticks)
> +				: queue_info->vector_data.vector_timeout_ticks;
> +		rx_adapter->vector_tmo_ticks <<= 1;

Any reason why we left shift here ? Applicable in patch 4/8 as well.

> +		TAILQ_INIT(&rx_adapter->vector_list);

Can doing TAILQ_INIT every time a queue is added cause existing elements to be wiped out ? Applicable in patch 4/8 as well.

> +		rx_adapter->prev_expiry_ts = 0;

Can setting this every time a queue is added affect existing queues created with vector support and passing traffic ? Applicable in patch 4/8 as well.

> +	}
> +
>  	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
>  	if (rxa_polled_queue(dev_info, rx_queue_id)) {
>  		rx_adapter->num_rx_polled += !pollq;
> @@ -1907,44 +1926,6 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
>  	}
>  }
> 
> -static void
> -rxa_sw_event_vector_configure(
> -	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
> -	int rx_queue_id,
> -	const struct rte_event_eth_rx_adapter_event_vector_config *config)
> -{
> -	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
> -	struct eth_rx_queue_info *queue_info;
> -	struct rte_event *qi_ev;
> -
> -	if (rx_queue_id == -1) {
> -		uint16_t nb_rx_queues;
> -		uint16_t i;
> -
> -		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
> -		for (i = 0; i < nb_rx_queues; i++)
> -			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
> -						      config);
> -		return;
> -	}
> -
> -	queue_info = &dev_info->rx_queue[rx_queue_id];
> -	qi_ev = (struct rte_event *)&queue_info->event;
> -	queue_info->ena_vector = 1;
> -	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
> -	rxa_set_vector_data(queue_info, config->vector_sz,
> -			    config->vector_timeout_ns, config->vector_mp,
> -			    rx_queue_id, dev_info->dev->data->port_id);
> -	rx_adapter->ena_vector = 1;
> -	rx_adapter->vector_tmo_ticks =
> -		rx_adapter->vector_tmo_ticks ?
> -			      RTE_MIN(config->vector_timeout_ns << 1,
> -				      rx_adapter->vector_tmo_ticks) :
> -			      config->vector_timeout_ns << 1;
> -	rx_adapter->prev_expiry_ts = 0;
> -	TAILQ_INIT(&rx_adapter->vector_list);
> -}
> -
>  static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
>  		uint16_t eth_dev_id,
>  		int rx_queue_id,
> @@ -2258,6 +2239,7 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
>  	struct rte_event_eth_rx_adapter *rx_adapter;
>  	struct rte_eventdev *dev;
>  	struct eth_device_info *dev_info;
> +	struct rte_event_eth_rx_adapter_vector_limits limits;
> 
>  	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
>  	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
> @@ -2294,6 +2276,39 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
>  		return -EINVAL;
>  	}
> 
> +	if (queue_conf->rx_queue_flags &
> +	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {

Perhaps, you could move the previous if condition here and just check for (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)

> +		ret = rte_event_eth_rx_adapter_vector_limits_get(
> +			rx_adapter->eventdev_id, eth_dev_id, &limits);
> +		if (ret < 0) {
> +			RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
> +					 " eth port: %" PRIu16
> +					 " adapter id: %" PRIu8,
> +					 eth_dev_id, id);
> +			return -EINVAL;
> +		}
> +		if (queue_conf->vector_sz < limits.min_sz ||
> +		    queue_conf->vector_sz > limits.max_sz ||
> +		    queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
> +		    queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
> +		    queue_conf->vector_mp == NULL) {
> +			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> +					 " eth port: %" PRIu16
> +					 " adapter id: %" PRIu8,
> +					 eth_dev_id, id);
> +			return -EINVAL;
> +		}
> +		if (queue_conf->vector_mp->elt_size <
> +		    (sizeof(struct rte_event_vector) +
> +		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
> +			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> +					 " eth port: %" PRIu16
> +					 " adapter id: %" PRIu8,
> +					 eth_dev_id, id);
> +			return -EINVAL;
> +		}
> +	}
> +
>  	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
>  		(rx_queue_id != -1)) {
>  		RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
> @@ -2487,83 +2502,6 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
>  	return ret;
>  }
> 
> -int
> -rte_event_eth_rx_adapter_queue_event_vector_config(
> -	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
> -	struct rte_event_eth_rx_adapter_event_vector_config *config)
> -{
> -	struct rte_event_eth_rx_adapter_vector_limits limits;
> -	struct rte_event_eth_rx_adapter *rx_adapter;
> -	struct rte_eventdev *dev;
> -	uint32_t cap;
> -	int ret;
> -
> -	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
> -	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
> -
> -	rx_adapter = rxa_id_to_adapter(id);
> -	if ((rx_adapter == NULL) || (config == NULL))
> -		return -EINVAL;
> -
> -	dev = &rte_eventdevs[rx_adapter->eventdev_id];
> -	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
> -						eth_dev_id, &cap);
> -	if (ret) {
> -		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
> -				 "eth port %" PRIu16,
> -				 id, eth_dev_id);
> -		return ret;
> -	}
> -
> -	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
> -		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
> -				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
> -				 eth_dev_id, id);
> -		return -EINVAL;
> -	}
> -
> -	ret = rte_event_eth_rx_adapter_vector_limits_get(
> -		rx_adapter->eventdev_id, eth_dev_id, &limits);
> -	if (ret) {
> -		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
> -				 "eth port %" PRIu16,
> -				 rx_adapter->eventdev_id, eth_dev_id);
> -		return ret;
> -	}
> -
> -	if (config->vector_sz < limits.min_sz ||
> -	    config->vector_sz > limits.max_sz ||
> -	    config->vector_timeout_ns < limits.min_timeout_ns ||
> -	    config->vector_timeout_ns > limits.max_timeout_ns ||
> -	    config->vector_mp == NULL) {
> -		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> -				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
> -				 eth_dev_id, id);
> -		return -EINVAL;
> -	}
> -	if (config->vector_mp->elt_size <
> -	    (sizeof(struct rte_event_vector) +
> -	     (sizeof(uintptr_t) * config->vector_sz))) {
> -		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> -				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
> -				 eth_dev_id, id);
> -		return -EINVAL;
> -	}
> -
> -	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
> -		RTE_FUNC_PTR_OR_ERR_RET(
> -			*dev->dev_ops->eth_rx_adapter_event_vector_config,
> -			-ENOTSUP);
> -		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
> -			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
> -	} else {
> -		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
> -					      rx_queue_id, config);
> -	}
> -
> -	return ret;
> -}
> -
>  int
>  rte_event_eth_rx_adapter_vector_limits_get(
>  	uint8_t dev_id, uint16_t eth_port_id,
> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> index 7407cde00..3f8b36229 100644
> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> @@ -171,9 +171,6 @@ struct rte_event_eth_rx_adapter_queue_conf {
>  	 * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
>  	 * enqueued event.
>  	 */
> -};
> -
> -struct rte_event_eth_rx_adapter_event_vector_config {
>  	uint16_t vector_sz;
>  	/**<
>  	 * Indicates the maximum number for mbufs to combine and form a vector.
> @@ -548,30 +545,6 @@ int rte_event_eth_rx_adapter_vector_limits_get(
>  	uint8_t dev_id, uint16_t eth_port_id,
>  	struct rte_event_eth_rx_adapter_vector_limits *limits);
> 
> -/**
> - * Configure event vectorization for a given ethernet device queue, that has
> - * been added to a event eth Rx adapter.
> - *
> - * @param id
> - *  The identifier of the ethernet Rx event adapter.
> - *
> - * @param eth_dev_id
> - *  The identifier of the ethernet device.
> - *
> - * @param rx_queue_id
> - *  Ethernet device receive queue index.
> - *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
> - *  are configured with event vectorization.
> - *
> - * @return
> - *  - 0: Success, Receive queue configured correctly.
> - *  - <0: Error code on failure.
> - */
> -__rte_experimental
> -int rte_event_eth_rx_adapter_queue_event_vector_config(
> -	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
> -	struct rte_event_eth_rx_adapter_event_vector_config *config);
> -
>  #ifdef __cplusplus
>  }
>  #endif
> diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
> index 902df0ae3..34c1c830e 100644
> --- a/lib/librte_eventdev/version.map
> +++ b/lib/librte_eventdev/version.map
> @@ -142,7 +142,6 @@ EXPERIMENTAL {
>  	#added in 21.05
>  	rte_event_vector_pool_create;
>  	rte_event_eth_rx_adapter_vector_limits_get;
> -	rte_event_eth_rx_adapter_queue_event_vector_config;
>  };
> 
>  INTERNAL {
> --
> 2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v5 4/8] eventdev: add Rx adapter event vector support
  2021-03-25 10:37           ` Jayatheerthan, Jay
@ 2021-03-25 13:14             ` Pavan Nikhilesh Bhagavatula
  2021-03-26  6:26               ` Jayatheerthan, Jay
  0 siblings, 1 reply; 153+ messages in thread
From: Pavan Nikhilesh Bhagavatula @ 2021-03-25 13:14 UTC (permalink / raw)
  To: Jayatheerthan, Jay, Jerin Jacob Kollanukkaran, Carrillo, Erik G,
	Gujjar, Abhinandan S, McDaniel, Timothy, hemant.agrawal,
	Van Haaren, Harry, mattias.ronnblom, Ma, Liang J
  Cc: dev



>-----Original Message-----
>From: Jayatheerthan, Jay <jay.jayatheerthan@intel.com>
>Sent: Thursday, March 25, 2021 4:07 PM
>To: Pavan Nikhilesh Bhagavatula <pbhagavatula@marvell.com>; Jerin
>Jacob Kollanukkaran <jerinj@marvell.com>; Carrillo, Erik G
><erik.g.carrillo@intel.com>; Gujjar, Abhinandan S
><abhinandan.gujjar@intel.com>; McDaniel, Timothy
><timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
>Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
><mattias.ronnblom@ericsson.com>; Ma, Liang J
><liang.j.ma@intel.com>
>Cc: dev@dpdk.org
>Subject: [EXT] RE: [dpdk-dev] [PATCH v5 4/8] eventdev: add Rx adapter
>event vector support
>
>External Email
>
>----------------------------------------------------------------------
>> -----Original Message-----
>> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
>> Sent: Wednesday, March 24, 2021 10:35 AM
>> To: jerinj@marvell.com; Jayatheerthan, Jay
><jay.jayatheerthan@intel.com>; Carrillo, Erik G
><erik.g.carrillo@intel.com>; Gujjar,
>> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy
><timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
>> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
><mattias.ronnblom@ericsson.com>; Ma, Liang J
>> <liang.j.ma@intel.com>
>> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
>> Subject: [dpdk-dev] [PATCH v5 4/8] eventdev: add Rx adapter event
>vector support
>>
>> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>>
>> Add event vector support for event eth Rx adapter, the
>implementation
>> creates vector flows based on port and queue identifier of the
>received
>> mbufs.
>>
>> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
>> ---
>>  lib/librte_eventdev/eventdev_pmd.h            |   7 +-
>>  .../rte_event_eth_rx_adapter.c                | 257 ++++++++++++++++--
>>  lib/librte_eventdev/rte_eventdev.c            |   6 +-
>>  3 files changed, 250 insertions(+), 20 deletions(-)
>>
>> diff --git a/lib/librte_eventdev/eventdev_pmd.h
>b/lib/librte_eventdev/eventdev_pmd.h
>> index 9297f1433..0f724ac85 100644
>> --- a/lib/librte_eventdev/eventdev_pmd.h
>> +++ b/lib/librte_eventdev/eventdev_pmd.h
>> @@ -69,9 +69,10 @@ extern "C" {
>>  	} \
>>  } while (0)
>>
>> -#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP \
>> -
>	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |
>\
>> -
>	(RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ))
>> +#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP
>\
>> +	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |
>\
>> +	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) |
>\
>> +	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR))
>>
>>  #define RTE_EVENT_CRYPTO_ADAPTER_SW_CAP \
>>
>	RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA
>> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> index ac8ba5bf0..c71990078 100644
>> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> @@ -26,6 +26,10 @@
>>  #define BATCH_SIZE		32
>>  #define BLOCK_CNT_THRESHOLD	10
>>  #define ETH_EVENT_BUFFER_SIZE	(4*BATCH_SIZE)
>> +#define MAX_VECTOR_SIZE		1024
>> +#define MIN_VECTOR_SIZE		4
>> +#define MAX_VECTOR_NS		1E9
>> +#define MIN_VECTOR_NS		1E5
>>
>>  #define ETH_RX_ADAPTER_SERVICE_NAME_LEN	32
>>  #define ETH_RX_ADAPTER_MEM_NAME_LEN	32
>> @@ -59,6 +63,20 @@ struct eth_rx_poll_entry {
>>  	uint16_t eth_rx_qid;
>>  };
>>
>> +struct eth_rx_vector_data {
>> +	TAILQ_ENTRY(eth_rx_vector_data) next;
>> +	uint16_t port;
>> +	uint16_t queue;
>> +	uint16_t max_vector_count;
>> +	uint64_t event;
>> +	uint64_t ts;
>> +	uint64_t vector_timeout_ticks;
>> +	struct rte_mempool *vector_pool;
>> +	struct rte_event_vector *vector_ev;
>> +} __rte_cache_aligned;
>> +
>> +TAILQ_HEAD(eth_rx_vector_data_list, eth_rx_vector_data);
>> +
>>  /* Instance per adapter */
>>  struct rte_eth_event_enqueue_buffer {
>>  	/* Count of events in this buffer */
>> @@ -92,6 +110,14 @@ struct rte_event_eth_rx_adapter {
>>  	uint32_t wrr_pos;
>>  	/* Event burst buffer */
>>  	struct rte_eth_event_enqueue_buffer event_enqueue_buffer;
>> +	/* Vector enable flag */
>> +	uint8_t ena_vector;
>> +	/* Timestamp of previous vector expiry list traversal */
>> +	uint64_t prev_expiry_ts;
>> +	/* Minimum ticks to wait before traversing expiry list */
>> +	uint64_t vector_tmo_ticks;
>> +	/* vector list */
>> +	struct eth_rx_vector_data_list vector_list;
>>  	/* Per adapter stats */
>>  	struct rte_event_eth_rx_adapter_stats stats;
>>  	/* Block count, counts up to BLOCK_CNT_THRESHOLD */
>> @@ -198,9 +224,11 @@ struct eth_device_info {
>>  struct eth_rx_queue_info {
>>  	int queue_enabled;	/* True if added */
>>  	int intr_enabled;
>> +	uint8_t ena_vector;
>>  	uint16_t wt;		/* Polling weight */
>>  	uint32_t flow_id_mask;	/* Set to ~0 if app provides flow id else
>0 */
>>  	uint64_t event;
>> +	struct eth_rx_vector_data vector_data;
>>  };
>>
>>  static struct rte_event_eth_rx_adapter **event_eth_rx_adapter;
>> @@ -722,6 +750,9 @@ rxa_flush_event_buffer(struct
>rte_event_eth_rx_adapter *rx_adapter)
>>  	    &rx_adapter->event_enqueue_buffer;
>>  	struct rte_event_eth_rx_adapter_stats *stats = &rx_adapter-
>>stats;
>>
>> +	if (!buf->count)
>> +		return 0;
>> +
>>  	uint16_t n = rte_event_enqueue_new_burst(rx_adapter-
>>eventdev_id,
>>  					rx_adapter->event_port_id,
>>  					buf->events,
>> @@ -742,6 +773,72 @@ rxa_flush_event_buffer(struct
>rte_event_eth_rx_adapter *rx_adapter)
>>  	return n;
>>  }
>>
>> +static inline uint16_t
>> +rxa_create_event_vector(struct rte_event_eth_rx_adapter
>*rx_adapter,
>> +			struct eth_rx_queue_info *queue_info,
>> +			struct rte_eth_event_enqueue_buffer *buf,
>> +			struct rte_mbuf **mbufs, uint16_t num)
>> +{
>> +	struct rte_event *ev = &buf->events[buf->count];
>> +	struct eth_rx_vector_data *vec;
>> +	uint16_t filled, space, sz;
>> +
>> +	filled = 0;
>> +	vec = &queue_info->vector_data;
>> +	while (num) {
>> +		if (vec->vector_ev == NULL) {
>> +			if (rte_mempool_get(vec->vector_pool,
>> +					    (void **)&vec->vector_ev) <
>0) {
>> +				rte_pktmbuf_free_bulk(mbufs, num);
>> +				return 0;
>> +			}
>> +			vec->vector_ev->nb_elem = 0;
>> +			vec->vector_ev->port = vec->port;
>> +			vec->vector_ev->queue = vec->queue;
>> +			vec->vector_ev->attr_valid = true;
>> +			TAILQ_INSERT_TAIL(&rx_adapter->vector_list,
>vec, next);
>> +		} else if (vec->vector_ev->nb_elem == vec-
>>max_vector_count) {
>
>Is there a case where nb_elem > max_vector_count as we accumulate
>sz to it ?

I don't think so, that would overflow the vector event.

>
>> +			/* Event ready. */
>> +			ev->event = vec->event;
>> +			ev->vec = vec->vector_ev;
>> +			ev++;
>> +			filled++;
>> +			vec->vector_ev = NULL;
>> +			TAILQ_REMOVE(&rx_adapter->vector_list, vec,
>next);
>> +			if (rte_mempool_get(vec->vector_pool,
>> +					    (void **)&vec->vector_ev) <
>0) {
>> +				rte_pktmbuf_free_bulk(mbufs, num);
>> +				return 0;
>> +			}
>> +			vec->vector_ev->nb_elem = 0;
>> +			vec->vector_ev->port = vec->port;
>> +			vec->vector_ev->queue = vec->queue;
>> +			vec->vector_ev->attr_valid = true;
>> +			TAILQ_INSERT_TAIL(&rx_adapter->vector_list,
>vec, next);
>> +		}
>> +
>> +		space = vec->max_vector_count - vec->vector_ev-
>>nb_elem;
>> +		sz = num > space ? space : num;
>> +		memcpy(vec->vector_ev->mbufs + vec->vector_ev-
>>nb_elem, mbufs,
>> +		       sizeof(void *) * sz);
>> +		vec->vector_ev->nb_elem += sz;
>> +		num -= sz;
>> +		mbufs += sz;
>> +		vec->ts = rte_rdtsc();
>> +	}
>> +
>> +	if (vec->vector_ev->nb_elem == vec->max_vector_count) {
>
>Same here.
>
>> +		ev->event = vec->event;
>> +		ev->vec = vec->vector_ev;
>> +		ev++;
>> +		filled++;
>> +		vec->vector_ev = NULL;
>> +		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
>> +	}
>> +
>> +	return filled;
>> +}
>
>I am seeing more than one repeating code chunks in this function.
>Perhaps, you can give it a try to not repeat. We can drop if its
>performance affecting.

I will try to move them to inline functions and test.

>
>> +
>>  static inline void
>>  rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
>>  		uint16_t eth_dev_id,
>> @@ -770,25 +867,30 @@ rxa_buffer_mbufs(struct
>rte_event_eth_rx_adapter *rx_adapter,
>>  	rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
>>  	do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
>
>The RSS related code is executed for vector case as well. Can this be
>moved inside ena_vector if condition ?

RSS is used to generate the event flowid, in vector case the flow id 
Will be a combination of port and queue id. 
The idea is that flows having the same RSS LSB will end up in the same 
queue.

>
>>
>> -	for (i = 0; i < num; i++) {
>> -		m = mbufs[i];
>> -
>> -		rss = do_rss ?
>> -			rxa_do_softrss(m, rx_adapter->rss_key_be) :
>> -			m->hash.rss;
>> -		ev->event = event;
>> -		ev->flow_id = (rss & ~flow_id_mask) |
>> -				(ev->flow_id & flow_id_mask);
>> -		ev->mbuf = m;
>> -		ev++;
>> +	if (!eth_rx_queue_info->ena_vector) {
>> +		for (i = 0; i < num; i++) {
>> +			m = mbufs[i];
>> +
>> +			rss = do_rss ? rxa_do_softrss(m, rx_adapter-
>>rss_key_be)
>> +				     : m->hash.rss;
>> +			ev->event = event;
>> +			ev->flow_id = (rss & ~flow_id_mask) |
>> +				      (ev->flow_id & flow_id_mask);
>> +			ev->mbuf = m;
>> +			ev++;
>> +		}
>> +	} else {
>> +		num = rxa_create_event_vector(rx_adapter,
>eth_rx_queue_info,
>> +					      buf, mbufs, num);
>>  	}
>>
>> -	if (dev_info->cb_fn) {
>> +	if (num && dev_info->cb_fn) {
>>
>>  		dropped = 0;
>>  		nb_cb = dev_info->cb_fn(eth_dev_id, rx_queue_id,
>> -					ETH_EVENT_BUFFER_SIZE, buf-
>>count, ev,
>> -					num, dev_info->cb_arg,
>&dropped);
>> +					ETH_EVENT_BUFFER_SIZE, buf-
>>count,
>> +					&buf->events[buf->count],
>num,
>> +					dev_info->cb_arg, &dropped);
>
>Before this patch, we pass ev which is &buf->events[buf->count] + num
>as fifth param when calling cb_fn. Now, we are passing &buf-
>>events[buf->count] for non-vector case. Do you see this as an issue?
>

The callback function takes in the array newly formed events i.e. we need
to pass the start of array and the count.

the previous code had a bug where it passes the end of the event list.

>Also, for vector case would it make sense to do pass &buf->events[buf-
>>count] + num ?
>
>>  		if (unlikely(nb_cb > num))
>>  			RTE_EDEV_LOG_ERR("Rx CB returned %d (> %d)
>events",
>>  				nb_cb, num);
>> @@ -1124,6 +1226,30 @@ rxa_poll(struct rte_event_eth_rx_adapter
>*rx_adapter)
>>  	return nb_rx;
>>  }
>>
>> +static void
>> +rxa_vector_expire(struct eth_rx_vector_data *vec, void *arg)
>> +{
>> +	struct rte_event_eth_rx_adapter *rx_adapter = arg;
>> +	struct rte_eth_event_enqueue_buffer *buf =
>> +		&rx_adapter->event_enqueue_buffer;
>> +	struct rte_event *ev;
>> +
>> +	if (buf->count)
>> +		rxa_flush_event_buffer(rx_adapter);
>> +
>> +	if (vec->vector_ev->nb_elem == 0)
>> +		return;
>> +	ev = &buf->events[buf->count];
>> +
>> +	/* Event ready. */
>> +	ev->event = vec->event;
>> +	ev->vec = vec->vector_ev;
>> +	buf->count++;
>> +
>> +	vec->vector_ev = NULL;
>> +	vec->ts = 0;
>> +}
>> +
>>  static int
>>  rxa_service_func(void *args)
>>  {
>> @@ -1137,6 +1263,24 @@ rxa_service_func(void *args)
>>  		return 0;
>>  	}
>>
>> +	if (rx_adapter->ena_vector) {
>> +		if ((rte_rdtsc() - rx_adapter->prev_expiry_ts) >=
>> +		    rx_adapter->vector_tmo_ticks) {
>> +			struct eth_rx_vector_data *vec;
>> +
>> +			TAILQ_FOREACH(vec, &rx_adapter->vector_list,
>next) {
>> +				uint64_t elapsed_time = rte_rdtsc() -
>vec->ts;
>> +
>> +				if (elapsed_time >= vec-
>>vector_timeout_ticks) {
>> +					rxa_vector_expire(vec,
>rx_adapter);
>> +					TAILQ_REMOVE(&rx_adapter-
>>vector_list,
>> +						     vec, next);
>> +				}
>> +			}
>> +			rx_adapter->prev_expiry_ts = rte_rdtsc();
>> +		}
>> +	}
>> +
>>  	stats = &rx_adapter->stats;
>>  	stats->rx_packets += rxa_intr_ring_dequeue(rx_adapter);
>>  	stats->rx_packets += rxa_poll(rx_adapter);
>> @@ -1640,6 +1784,28 @@ rxa_update_queue(struct
>rte_event_eth_rx_adapter *rx_adapter,
>>  	}
>>  }
>>
>> +static void
>> +rxa_set_vector_data(struct eth_rx_queue_info *queue_info,
>uint16_t vector_count,
>> +		    uint64_t vector_ns, struct rte_mempool *mp, int32_t
>qid,
>> +		    uint16_t port_id)
>> +{
>> +#define NSEC2TICK(__ns, __freq) (((__ns) * (__freq)) / 1E9)
>> +	struct eth_rx_vector_data *vector_data;
>> +	uint32_t flow_id;
>> +
>> +	vector_data = &queue_info->vector_data;
>> +	vector_data->max_vector_count = vector_count;
>> +	vector_data->port = port_id;
>> +	vector_data->queue = qid;
>> +	vector_data->vector_pool = mp;
>> +	vector_data->vector_timeout_ticks =
>> +		NSEC2TICK(vector_ns, rte_get_timer_hz());
>> +	vector_data->ts = 0;
>> +	flow_id = queue_info->event & 0xFFFFF;
>> +	flow_id = flow_id == 0 ? (qid & 0xFF) | (port_id & 0xFFFF) :
>flow_id;
>
>Maybe I am missing something here. Looking at the code it looks like
>qid and port_id may overlap. For e.g., if qid = 0x10 and port_id = 0x11,
>flow_id would end up being 0x11. Is this the expectation? Also, it may
>be useful to document flow_id format.

The flow_id is 20 bit, I guess we could do 12bit queue_id and 8bit port
as a flow.

>Comparing this format with existing RSS hash based method, are we
>saying that all mbufs received in a rx burst are part of same flow when
>vectorization is used?

Yes, the hard way to do this is to use a hash table and treating each 
mbuf having an unique flow.

>
>> +	vector_data->event = (queue_info->event & ~0xFFFFF) |
>flow_id;
>> +}
>> +
>>  static void
>>  rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
>>  	struct eth_device_info *dev_info,
>> @@ -1741,6 +1907,44 @@ rxa_add_queue(struct
>rte_event_eth_rx_adapter *rx_adapter,
>>  	}
>>  }
>>
>> +static void
>> +rxa_sw_event_vector_configure(
>> +	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t
>eth_dev_id,
>> +	int rx_queue_id,
>> +	const struct rte_event_eth_rx_adapter_event_vector_config
>*config)
>> +{
>> +	struct eth_device_info *dev_info = &rx_adapter-
>>eth_devices[eth_dev_id];
>> +	struct eth_rx_queue_info *queue_info;
>> +	struct rte_event *qi_ev;
>> +
>> +	if (rx_queue_id == -1) {
>> +		uint16_t nb_rx_queues;
>> +		uint16_t i;
>> +
>> +		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
>> +		for (i = 0; i < nb_rx_queues; i++)
>> +			rxa_sw_event_vector_configure(rx_adapter,
>eth_dev_id, i,
>> +						      config);
>> +		return;
>> +	}
>> +
>> +	queue_info = &dev_info->rx_queue[rx_queue_id];
>> +	qi_ev = (struct rte_event *)&queue_info->event;
>> +	queue_info->ena_vector = 1;
>> +	qi_ev->event_type =
>RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
>> +	rxa_set_vector_data(queue_info, config->vector_sz,
>> +			    config->vector_timeout_ns, config-
>>vector_mp,
>> +			    rx_queue_id, dev_info->dev->data->port_id);
>> +	rx_adapter->ena_vector = 1;
>> +	rx_adapter->vector_tmo_ticks =
>> +		rx_adapter->vector_tmo_ticks ?
>> +			      RTE_MIN(config->vector_timeout_ns << 1,
>> +				      rx_adapter->vector_tmo_ticks) :
>> +			      config->vector_timeout_ns << 1;
>> +	rx_adapter->prev_expiry_ts = 0;
>> +	TAILQ_INIT(&rx_adapter->vector_list);
>> +}
>> +
>>  static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
>>  		uint16_t eth_dev_id,
>>  		int rx_queue_id,
>> @@ -2081,6 +2285,15 @@
>rte_event_eth_rx_adapter_queue_add(uint8_t id,
>>  		return -EINVAL;
>>  	}
>>
>> +	if ((cap &
>RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
>> +	    (queue_conf->rx_queue_flags &
>> +	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
>> +		RTE_EDEV_LOG_ERR("Event vectorization is not
>supported,"
>> +				 " eth port: %" PRIu16 " adapter id: %"
>PRIu8,
>> +				 eth_dev_id, id);
>> +		return -EINVAL;
>> +	}
>> +
>>  	if ((cap &
>RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
>>  		(rx_queue_id != -1)) {
>>  		RTE_EDEV_LOG_ERR("Rx queues can only be connected
>to single "
>> @@ -2143,6 +2356,17 @@
>rte_event_eth_rx_adapter_queue_add(uint8_t id,
>>  	return 0;
>>  }
>>
>> +static int
>> +rxa_sw_vector_limits(struct
>rte_event_eth_rx_adapter_vector_limits *limits)
>> +{
>> +	limits->max_sz = MAX_VECTOR_SIZE;
>> +	limits->min_sz = MIN_VECTOR_SIZE;
>> +	limits->max_timeout_ns = MAX_VECTOR_NS;
>> +	limits->min_timeout_ns = MIN_VECTOR_NS;
>> +
>> +	return 0;
>> +}
>> +
>>  int
>>  rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t
>eth_dev_id,
>>  				int32_t rx_queue_id)
>> @@ -2333,7 +2557,8 @@
>rte_event_eth_rx_adapter_queue_event_vector_config(
>>  		ret = dev->dev_ops-
>>eth_rx_adapter_event_vector_config(
>>  			dev, &rte_eth_devices[eth_dev_id],
>rx_queue_id, config);
>>  	} else {
>> -		ret = -ENOTSUP;
>> +		rxa_sw_event_vector_configure(rx_adapter,
>eth_dev_id,
>> +					      rx_queue_id, config);
>>  	}
>>
>>  	return ret;
>> @@ -2371,7 +2596,7 @@
>rte_event_eth_rx_adapter_vector_limits_get(
>>  		ret = dev->dev_ops-
>>eth_rx_adapter_vector_limits_get(
>>  			dev, &rte_eth_devices[eth_port_id], limits);
>>  	} else {
>> -		ret = -ENOTSUP;
>> +		ret = rxa_sw_vector_limits(limits);
>>  	}
>>
>>  	return ret;
>> diff --git a/lib/librte_eventdev/rte_eventdev.c
>b/lib/librte_eventdev/rte_eventdev.c
>> index f95edc075..254a31b1f 100644
>> --- a/lib/librte_eventdev/rte_eventdev.c
>> +++ b/lib/librte_eventdev/rte_eventdev.c
>> @@ -122,7 +122,11 @@ rte_event_eth_rx_adapter_caps_get(uint8_t
>dev_id, uint16_t eth_port_id,
>>
>>  	if (caps == NULL)
>>  		return -EINVAL;
>> -	*caps = 0;
>> +
>> +	if (dev->dev_ops->eth_rx_adapter_caps_get == NULL)
>> +		*caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
>> +	else
>> +		*caps = 0;
>
>Any reason why we had to set default caps value? I am thinking if sw
>event device is used, it would set it anyways.
>

There are multiple sw event devices which don't implement caps_get 
function, this changes solves that.

>>
>>  	return dev->dev_ops->eth_rx_adapter_caps_get ?
>>  				(*dev->dev_ops-
>>eth_rx_adapter_caps_get)(dev,
>> --
>> 2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [dpdk-dev v21.11] [PATCH v5 8/8] eventdev: simplify Rx adapter event vector config
  2021-03-25 12:27           ` Jayatheerthan, Jay
@ 2021-03-25 13:55             ` Pavan Nikhilesh Bhagavatula
  2021-03-26  7:09               ` Jayatheerthan, Jay
  0 siblings, 1 reply; 153+ messages in thread
From: Pavan Nikhilesh Bhagavatula @ 2021-03-25 13:55 UTC (permalink / raw)
  To: Jayatheerthan, Jay, Jerin Jacob Kollanukkaran, Carrillo, Erik G,
	Gujjar, Abhinandan S, McDaniel, Timothy, hemant.agrawal,
	Van Haaren, Harry, mattias.ronnblom, Ma, Liang J, Ray Kinsella,
	Neil Horman
  Cc: dev

>> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
>> Sent: Wednesday, March 24, 2021 10:35 AM
>> To: jerinj@marvell.com; Jayatheerthan, Jay
><jay.jayatheerthan@intel.com>; Carrillo, Erik G
><erik.g.carrillo@intel.com>; Gujjar,
>> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy
><timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
>> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
><mattias.ronnblom@ericsson.com>; Ma, Liang J
>> <liang.j.ma@intel.com>; Ray Kinsella <mdr@ashroe.eu>; Neil Horman
><nhorman@tuxdriver.com>
>> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
>> Subject: [dpdk-dev v21.11] [PATCH v5 8/8] eventdev: simplify Rx
>adapter event vector config
>>
>> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>>
>> Include vector configuration into the structure
>> ``rte_event_eth_rx_adapter_queue_conf`` used when configuring rest
>> of the Rx adapter ethernet device Rx queue parameters.
>> This simplifies event vector configuration as it avoids splitting
>> configuration per Rx queue.
>>
>> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
>> ---
>>  app/test-eventdev/test_pipeline_common.c      |  16 +-
>>  lib/librte_eventdev/eventdev_pmd.h            |  29 ---
>>  .../rte_event_eth_rx_adapter.c                | 168 ++++++------------
>>  .../rte_event_eth_rx_adapter.h                |  27 ---
>>  lib/librte_eventdev/version.map               |   1 -
>>  5 files changed, 57 insertions(+), 184 deletions(-)
>>
>> diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-
>eventdev/test_pipeline_common.c
>> index d5ef90500..76aee254b 100644
>> --- a/app/test-eventdev/test_pipeline_common.c
>> +++ b/app/test-eventdev/test_pipeline_common.c
>> @@ -331,7 +331,6 @@ pipeline_event_rx_adapter_setup(struct
>evt_options *opt, uint8_t stride,
>>  	uint16_t prod;
>>  	struct rte_mempool *vector_pool = NULL;
>>  	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
>> -	struct rte_event_eth_rx_adapter_event_vector_config
>vec_conf;
>>
>>  	memset(&queue_conf, 0,
>>  			sizeof(struct
>rte_event_eth_rx_adapter_queue_conf));
>> @@ -397,8 +396,12 @@ pipeline_event_rx_adapter_setup(struct
>evt_options *opt, uint8_t stride,
>>  			}
>>
>>  			if (cap &
>RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
>> +				queue_conf.vector_sz = opt-
>>vector_size;
>> +				queue_conf.vector_timeout_ns =
>> +					opt->vector_tmo_nsec;
>>  				queue_conf.rx_queue_flags |=
>>
>	RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
>> +				queue_conf.vector_mp = vector_pool;
>>  			} else {
>>  				evt_err("Rx adapter doesn't support
>event vector");
>>  				return -EINVAL;
>> @@ -418,17 +421,6 @@ pipeline_event_rx_adapter_setup(struct
>evt_options *opt, uint8_t stride,
>>  			return ret;
>>  		}
>>
>> -		if (opt->ena_vector) {
>> -			vec_conf.vector_sz = opt->vector_size;
>> -			vec_conf.vector_timeout_ns = opt-
>>vector_tmo_nsec;
>> -			vec_conf.vector_mp = vector_pool;
>> -			if
>(rte_event_eth_rx_adapter_queue_event_vector_config(
>> -				    prod, prod, -1, &vec_conf) < 0) {
>> -				evt_err("Failed to configure event
>vectorization for Rx adapter");
>> -				return -EINVAL;
>> -			}
>> -		}
>> -
>>  		if (!(cap &
>RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
>>  			uint32_t service_id = -1U;
>>
>> diff --git a/lib/librte_eventdev/eventdev_pmd.h
>b/lib/librte_eventdev/eventdev_pmd.h
>> index 0f724ac85..63b3bc4b5 100644
>> --- a/lib/librte_eventdev/eventdev_pmd.h
>> +++ b/lib/librte_eventdev/eventdev_pmd.h
>> @@ -667,32 +667,6 @@ typedef int
>(*eventdev_eth_rx_adapter_vector_limits_get_t)(
>>  	const struct rte_eventdev *dev, const struct rte_eth_dev
>*eth_dev,
>>  	struct rte_event_eth_rx_adapter_vector_limits *limits);
>>
>> -struct rte_event_eth_rx_adapter_event_vector_config;
>> -/**
>> - * Enable event vector on an given Rx queue of a ethernet devices
>belonging to
>> - * the Rx adapter.
>> - *
>> - * @param dev
>> - *   Event device pointer
>> - *
>> - * @param eth_dev
>> - *   Ethernet device pointer
>> - *
>> - * @param rx_queue_id
>> - *   The Rx queue identifier
>> - *
>> - * @param config
>> - *   Pointer to the event vector configuration structure.
>> - *
>> - * @return
>> - *   - 0: Success.
>> - *   - <0: Error code returned by the driver function.
>> - */
>> -typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
>> -	const struct rte_eventdev *dev, const struct rte_eth_dev
>*eth_dev,
>> -	int32_t rx_queue_id,
>> -	const struct rte_event_eth_rx_adapter_event_vector_config
>*config);
>> -
>>  typedef uint32_t rte_event_pmd_selftest_seqn_t;
>>  extern int rte_event_pmd_selftest_seqn_dynfield_offset;
>>
>> @@ -1118,9 +1092,6 @@ struct rte_eventdev_ops {
>>  	eventdev_eth_rx_adapter_vector_limits_get_t
>>  		eth_rx_adapter_vector_limits_get;
>>  	/**< Get event vector limits for the Rx adapter */
>> -	eventdev_eth_rx_adapter_event_vector_config_t
>> -		eth_rx_adapter_event_vector_config;
>> -	/**< Configure Rx adapter with event vector */
>>
>>  	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
>>  	/**< Get timer adapter capabilities */
>> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> index c71990078..a1990637f 100644
>> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> @@ -1882,6 +1882,25 @@ rxa_add_queue(struct
>rte_event_eth_rx_adapter *rx_adapter,
>>  	} else
>>  		qi_ev->flow_id = 0;
>>
>> +	if (conf->rx_queue_flags &
>> +	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
>> +		queue_info->ena_vector = 1;
>> +		qi_ev->event_type =
>RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
>> +		rxa_set_vector_data(queue_info, conf->vector_sz,
>> +				    conf->vector_timeout_ns, conf-
>>vector_mp,
>> +				    rx_queue_id, dev_info->dev->data-
>>port_id);
>> +		rx_adapter->ena_vector = 1;
>> +		rx_adapter->vector_tmo_ticks =
>> +			rx_adapter->vector_tmo_ticks
>> +				? RTE_MIN(queue_info->vector_data
>> +						  .vector_timeout_ticks,
>> +					  rx_adapter-
>>vector_tmo_ticks)
>> +				: queue_info-
>>vector_data.vector_timeout_ticks;
>> +		rx_adapter->vector_tmo_ticks <<= 1;
>
>Any reason why we left shift here ? Applicable in patch 4/8 as well.

Just so that we have half the precision of the lowest timeout, helps to 
Maintain accuracy.

>
>> +		TAILQ_INIT(&rx_adapter->vector_list);
>
>Can doing TAILQ_INIT every time a queue is added cause existing
>elements to be wiped out ? Applicable in patch 4/8 as well.

I don't think queues can be added when adapter is already started
rxa_sw_add Isn't thready safe.

>
>> +		rx_adapter->prev_expiry_ts = 0;
>
>Can setting this every time a queue is added affect existing queues
>created with vector support and passing traffic ? Applicable in patch 4/8
>as well.

Same as above.

>
>> +	}
>> +
>>  	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
>>  	if (rxa_polled_queue(dev_info, rx_queue_id)) {
>>  		rx_adapter->num_rx_polled += !pollq;
>> @@ -1907,44 +1926,6 @@ rxa_add_queue(struct
>rte_event_eth_rx_adapter *rx_adapter,
>>  	}
>>  }
>>
>> -static void
>> -rxa_sw_event_vector_configure(
>> -	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t
>eth_dev_id,
>> -	int rx_queue_id,
>> -	const struct rte_event_eth_rx_adapter_event_vector_config
>*config)
>> -{
>> -	struct eth_device_info *dev_info = &rx_adapter-
>>eth_devices[eth_dev_id];
>> -	struct eth_rx_queue_info *queue_info;
>> -	struct rte_event *qi_ev;
>> -
>> -	if (rx_queue_id == -1) {
>> -		uint16_t nb_rx_queues;
>> -		uint16_t i;
>> -
>> -		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
>> -		for (i = 0; i < nb_rx_queues; i++)
>> -			rxa_sw_event_vector_configure(rx_adapter,
>eth_dev_id, i,
>> -						      config);
>> -		return;
>> -	}
>> -
>> -	queue_info = &dev_info->rx_queue[rx_queue_id];
>> -	qi_ev = (struct rte_event *)&queue_info->event;
>> -	queue_info->ena_vector = 1;
>> -	qi_ev->event_type =
>RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
>> -	rxa_set_vector_data(queue_info, config->vector_sz,
>> -			    config->vector_timeout_ns, config-
>>vector_mp,
>> -			    rx_queue_id, dev_info->dev->data->port_id);
>> -	rx_adapter->ena_vector = 1;
>> -	rx_adapter->vector_tmo_ticks =
>> -		rx_adapter->vector_tmo_ticks ?
>> -			      RTE_MIN(config->vector_timeout_ns << 1,
>> -				      rx_adapter->vector_tmo_ticks) :
>> -			      config->vector_timeout_ns << 1;
>> -	rx_adapter->prev_expiry_ts = 0;
>> -	TAILQ_INIT(&rx_adapter->vector_list);
>> -}
>> -
>>  static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
>>  		uint16_t eth_dev_id,
>>  		int rx_queue_id,
>> @@ -2258,6 +2239,7 @@
>rte_event_eth_rx_adapter_queue_add(uint8_t id,
>>  	struct rte_event_eth_rx_adapter *rx_adapter;
>>  	struct rte_eventdev *dev;
>>  	struct eth_device_info *dev_info;
>> +	struct rte_event_eth_rx_adapter_vector_limits limits;
>>
>>  	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -
>EINVAL);
>>  	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
>> @@ -2294,6 +2276,39 @@
>rte_event_eth_rx_adapter_queue_add(uint8_t id,
>>  		return -EINVAL;
>>  	}
>>
>> +	if (queue_conf->rx_queue_flags &
>> +	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
>
>Perhaps, you could move the previous if condition here and just check
>for (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)

No, the above check is not a capability check, it is to check if application
requested to enable vectorization for this queue or not.

>
>> +		ret = rte_event_eth_rx_adapter_vector_limits_get(
>> +			rx_adapter->eventdev_id, eth_dev_id, &limits);
>> +		if (ret < 0) {
>> +			RTE_EDEV_LOG_ERR("Failed to get event
>device vector limits,"
>> +					 " eth port: %" PRIu16
>> +					 " adapter id: %" PRIu8,
>> +					 eth_dev_id, id);
>> +			return -EINVAL;
>> +		}
>> +		if (queue_conf->vector_sz < limits.min_sz ||
>> +		    queue_conf->vector_sz > limits.max_sz ||
>> +		    queue_conf->vector_timeout_ns <
>limits.min_timeout_ns ||
>> +		    queue_conf->vector_timeout_ns >
>limits.max_timeout_ns ||
>> +		    queue_conf->vector_mp == NULL) {
>> +			RTE_EDEV_LOG_ERR("Invalid event vector
>configuration,"
>> +					 " eth port: %" PRIu16
>> +					 " adapter id: %" PRIu8,
>> +					 eth_dev_id, id);
>> +			return -EINVAL;
>> +		}
>> +		if (queue_conf->vector_mp->elt_size <
>> +		    (sizeof(struct rte_event_vector) +
>> +		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
>> +			RTE_EDEV_LOG_ERR("Invalid event vector
>configuration,"
>> +					 " eth port: %" PRIu16
>> +					 " adapter id: %" PRIu8,
>> +					 eth_dev_id, id);
>> +			return -EINVAL;
>> +		}
>> +	}
>> +
>>  	if ((cap &
>RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
>>  		(rx_queue_id != -1)) {
>>  		RTE_EDEV_LOG_ERR("Rx queues can only be connected
>to single "
>> @@ -2487,83 +2502,6 @@
>rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
>>  	return ret;
>>  }
>>
>> -int
>> -rte_event_eth_rx_adapter_queue_event_vector_config(
>> -	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
>> -	struct rte_event_eth_rx_adapter_event_vector_config *config)
>> -{
>> -	struct rte_event_eth_rx_adapter_vector_limits limits;
>> -	struct rte_event_eth_rx_adapter *rx_adapter;
>> -	struct rte_eventdev *dev;
>> -	uint32_t cap;
>> -	int ret;
>> -
>> -	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -
>EINVAL);
>> -	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
>> -
>> -	rx_adapter = rxa_id_to_adapter(id);
>> -	if ((rx_adapter == NULL) || (config == NULL))
>> -		return -EINVAL;
>> -
>> -	dev = &rte_eventdevs[rx_adapter->eventdev_id];
>> -	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter-
>>eventdev_id,
>> -						eth_dev_id, &cap);
>> -	if (ret) {
>> -		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev
>%" PRIu8
>> -				 "eth port %" PRIu16,
>> -				 id, eth_dev_id);
>> -		return ret;
>> -	}
>> -
>> -	if (!(cap &
>RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
>> -		RTE_EDEV_LOG_ERR("Event vectorization is not
>supported,"
>> -				 " eth port: %" PRIu16 " adapter id: %"
>PRIu8,
>> -				 eth_dev_id, id);
>> -		return -EINVAL;
>> -	}
>> -
>> -	ret = rte_event_eth_rx_adapter_vector_limits_get(
>> -		rx_adapter->eventdev_id, eth_dev_id, &limits);
>> -	if (ret) {
>> -		RTE_EDEV_LOG_ERR("Failed to get vector limits edev
>%" PRIu8
>> -				 "eth port %" PRIu16,
>> -				 rx_adapter->eventdev_id, eth_dev_id);
>> -		return ret;
>> -	}
>> -
>> -	if (config->vector_sz < limits.min_sz ||
>> -	    config->vector_sz > limits.max_sz ||
>> -	    config->vector_timeout_ns < limits.min_timeout_ns ||
>> -	    config->vector_timeout_ns > limits.max_timeout_ns ||
>> -	    config->vector_mp == NULL) {
>> -		RTE_EDEV_LOG_ERR("Invalid event vector
>configuration,"
>> -				 " eth port: %" PRIu16 " adapter id: %"
>PRIu8,
>> -				 eth_dev_id, id);
>> -		return -EINVAL;
>> -	}
>> -	if (config->vector_mp->elt_size <
>> -	    (sizeof(struct rte_event_vector) +
>> -	     (sizeof(uintptr_t) * config->vector_sz))) {
>> -		RTE_EDEV_LOG_ERR("Invalid event vector
>configuration,"
>> -				 " eth port: %" PRIu16 " adapter id: %"
>PRIu8,
>> -				 eth_dev_id, id);
>> -		return -EINVAL;
>> -	}
>> -
>> -	if (cap &
>RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
>> -		RTE_FUNC_PTR_OR_ERR_RET(
>> -			*dev->dev_ops-
>>eth_rx_adapter_event_vector_config,
>> -			-ENOTSUP);
>> -		ret = dev->dev_ops-
>>eth_rx_adapter_event_vector_config(
>> -			dev, &rte_eth_devices[eth_dev_id],
>rx_queue_id, config);
>> -	} else {
>> -		rxa_sw_event_vector_configure(rx_adapter,
>eth_dev_id,
>> -					      rx_queue_id, config);
>> -	}
>> -
>> -	return ret;
>> -}
>> -
>>  int
>>  rte_event_eth_rx_adapter_vector_limits_get(
>>  	uint8_t dev_id, uint16_t eth_port_id,
>> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
>b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
>> index 7407cde00..3f8b36229 100644
>> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
>> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
>> @@ -171,9 +171,6 @@ struct rte_event_eth_rx_adapter_queue_conf
>{
>>  	 * The event adapter sets ev.event_type to
>RTE_EVENT_TYPE_ETHDEV in the
>>  	 * enqueued event.
>>  	 */
>> -};
>> -
>> -struct rte_event_eth_rx_adapter_event_vector_config {
>>  	uint16_t vector_sz;
>>  	/**<
>>  	 * Indicates the maximum number for mbufs to combine and
>form a vector.
>> @@ -548,30 +545,6 @@ int
>rte_event_eth_rx_adapter_vector_limits_get(
>>  	uint8_t dev_id, uint16_t eth_port_id,
>>  	struct rte_event_eth_rx_adapter_vector_limits *limits);
>>
>> -/**
>> - * Configure event vectorization for a given ethernet device queue,
>that has
>> - * been added to a event eth Rx adapter.
>> - *
>> - * @param id
>> - *  The identifier of the ethernet Rx event adapter.
>> - *
>> - * @param eth_dev_id
>> - *  The identifier of the ethernet device.
>> - *
>> - * @param rx_queue_id
>> - *  Ethernet device receive queue index.
>> - *  If rx_queue_id is -1, then all Rx queues configured for the ethernet
>device
>> - *  are configured with event vectorization.
>> - *
>> - * @return
>> - *  - 0: Success, Receive queue configured correctly.
>> - *  - <0: Error code on failure.
>> - */
>> -__rte_experimental
>> -int rte_event_eth_rx_adapter_queue_event_vector_config(
>> -	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
>> -	struct rte_event_eth_rx_adapter_event_vector_config
>*config);
>> -
>>  #ifdef __cplusplus
>>  }
>>  #endif
>> diff --git a/lib/librte_eventdev/version.map
>b/lib/librte_eventdev/version.map
>> index 902df0ae3..34c1c830e 100644
>> --- a/lib/librte_eventdev/version.map
>> +++ b/lib/librte_eventdev/version.map
>> @@ -142,7 +142,6 @@ EXPERIMENTAL {
>>  	#added in 21.05
>>  	rte_event_vector_pool_create;
>>  	rte_event_eth_rx_adapter_vector_limits_get;
>> -	rte_event_eth_rx_adapter_queue_event_vector_config;
>>  };
>>
>>  INTERNAL {
>> --
>> 2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v7 0/8] Introduce event vectorization
  2021-03-24 19:28         ` [dpdk-dev] [PATCH v6 " pbhagavatula
                             ` (7 preceding siblings ...)
  2021-03-24 19:28           ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v6 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
@ 2021-03-25 17:10           ` pbhagavatula
  2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 1/8] eventdev: introduce event vector capability pbhagavatula
                               ` (8 more replies)
  8 siblings, 9 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-25 17:10 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

In traditional event programming model, events are identified by a
flow-id and a uintptr_t. The flow-id uniquely identifies a given event
and determines the order of scheduling based on schedule type, the
uintptr_t holds a single object.

Event devices also support burst mode with configurable dequeue depth,
i.e. each dequeue call would return multiple events and each event
might be at a different stage of the pipeline.
Having a burst of events belonging to different stages in a dequeue
burst is not only difficult to vectorize but also increases the scheduler
overhead and application overhead of pipelining events further.
Using event vectors we see a performance gain of ~742.3% as shown in [1].

By introducing event vectorization, each event will be capable of holding
multiple uintptr_t of the same flow thereby allowing applications
to vectorize their pipeline and reduce the complexity of pipelining
events across multiple stages. This also reduces the complexity of handling
enqueue and dequeue on an event device.

Since event devices are transparent to the events they are scheduling
so the event producers such as eth_rx_adapter, crypto_adapter , etc..
are responsible for vectorizing the buffers of the same flow into a single
event.

The series also breaks ABI in the patch [8/8] which is targetted to the
v21.11 release.

The dpdk-test-eventdev application has been updated with options to test
multiple vector sizes and timeouts.

[1]
As for performance improvement, with a ARM Cortex-A72 equivalent processer,
software event device (--vdev=event_sw0), single worker core, single stage
and using one service core for Rx adapter, Tx adapter, Scheduling.

Without this patchset applied:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
         --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
         --stlist=a --wlcores=20
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    5.071 mpps

With the patchset applied and Without event vectorization:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
         --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
         --stlist=a --wlcores=20
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    5.123 mpps

With event vectorization:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
        --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
        --stlist=a --wlcores=20 --enable_vector --nb_eth_queues 1
        --vector_size 256
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    42.715 mpps

Having dedicated service cores for each Rx queues and tweaking the vector,
dequeue burst size would further improve performance.

API usage is shown below:

Configuration:

	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

	vector_pool = rte_event_vector_pool_create("vector_pool",
			nb_elem, 0, vector_size, socket_id);

	rte_event_eth_rx_adapter_create(id, event_id, &adptr_conf);
	rte_event_eth_rx_adapter_queue_add(id, eth_id, -1, &queue_conf);
	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
		vec_conf.vector_sz = vector_size;
		vec_conf.vector_timeout_ns = vector_tmo_nsec;
		vec_conf.vector_mp = vector_pool;
		rte_event_eth_rx_adapter_queue_event_vector_config(id,
				eth_id, -1, &vec_conf);
	}

Fastpath:

	num = rte_event_dequeue_burst(event_id, port_id, &ev, 1, 0);
	if (!num)
		continue;

	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
		switch (ev.event_type) {
		case RTE_EVENT_TYPE_ETHDEV_VECTOR:
		case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
			struct rte_mbuf **mbufs;

			mbufs = ev.vector_ev->mbufs;
			for (i = 0; i < ev.vector_ev->nb_elem; i++)
				//Process mbufs.
			break;
		case ...
		}
	}
	...

v7 Changes:
- More doxygen fixes.(Jay)
- Reduce code duplication in 4/8.(Jay)

v6 Changes:
- Make rte_errno sign consistant.(Jay)
- Gramatical and doxygen fixes. (Jay)

v5 Changes:
- Make `rte_event_vector_pool_create non-inline` to ease ABI stability.(Ray)
- Move `rte_event_eth_rx_adapter_queue_event_vector_config` and
  `rte_event_eth_rx_adapter_vector_limits_get` implementation to the patch
  where they are initially defined.(Ray)
- Multiple gramatical and style fixes.(Jerin)
- Add missing release notes.(Jerin)

v4 Changes:
- Fix missing event vector structure in event structure.(Jay)

v3 Changes:
- Fix unintended formatting changes.

v2 Changes:
- Multiple gramatical and style fixes.(Jerin)
- Add parameter to define vector size in power of 2. (Jerin)
- Redo patch series w/o breaking ABI till the last patch.(David)
- Add deprication notice to announce ABI break in 21.11.(David)
- Add vector limits validation to app/test-eventdev.

Pavan Nikhilesh (8):
  eventdev: introduce event vector capability
  eventdev: introduce event vector Rx capability
  eventdev: introduce event vector Tx capability
  eventdev: add Rx adapter event vector support
  eventdev: add Tx adapter event vector support
  app/eventdev: add event vector mode in pipeline test
  doc: announce event Rx adapter config changes
  eventdev: simplify Rx adapter event vector config

 app/test-eventdev/evt_common.h                |   4 +
 app/test-eventdev/evt_options.c               |  52 +++
 app/test-eventdev/evt_options.h               |   4 +
 app/test-eventdev/test_pipeline_atq.c         | 310 +++++++++++++++--
 app/test-eventdev/test_pipeline_common.c      | 105 +++++-
 app/test-eventdev/test_pipeline_common.h      |  18 +
 app/test-eventdev/test_pipeline_queue.c       | 320 ++++++++++++++++--
 .../prog_guide/event_ethernet_rx_adapter.rst  |  38 +++
 .../prog_guide/event_ethernet_tx_adapter.rst  |  12 +
 doc/guides/prog_guide/eventdev.rst            |  36 +-
 doc/guides/rel_notes/deprecation.rst          |   9 +
 doc/guides/rel_notes/release_21_05.rst        |   8 +
 doc/guides/tools/testeventdev.rst             |  45 ++-
 lib/librte_eventdev/eventdev_pmd.h            |  31 +-
 .../rte_event_eth_rx_adapter.c                | 311 ++++++++++++++++-
 .../rte_event_eth_rx_adapter.h                |  78 +++++
 .../rte_event_eth_tx_adapter.c                |  66 +++-
 lib/librte_eventdev/rte_eventdev.c            |  53 ++-
 lib/librte_eventdev/rte_eventdev.h            | 114 ++++++-
 lib/librte_eventdev/version.map               |   4 +
 20 files changed, 1531 insertions(+), 87 deletions(-)

--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v7 1/8] eventdev: introduce event vector capability
  2021-03-25 17:10           ` [dpdk-dev] [PATCH v7 0/8] Introduce event vectorization pbhagavatula
@ 2021-03-25 17:10             ` pbhagavatula
  2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 2/8] eventdev: introduce event vector Rx capability pbhagavatula
                               ` (7 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-25 17:10 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce rte_event_vector datastructure which is capable of holding
multiple uintptr_t of the same flow thereby allowing applications
to vectorize their pipeline and reducing the complexity of pipelining
the events across multiple stages.
This approach also reduces the scheduling overhead on a event device.

Add a event vector mempool create handler to create mempools based on
the best mempool ops available on a given platform.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
Acked-by: Ray Kinsella <mdr@ashroe.eu>
---
 doc/guides/prog_guide/eventdev.rst     | 36 ++++++++++-
 doc/guides/rel_notes/release_21_05.rst |  8 +++
 lib/librte_eventdev/rte_eventdev.c     | 42 +++++++++++++
 lib/librte_eventdev/rte_eventdev.h     | 82 +++++++++++++++++++++++++-
 lib/librte_eventdev/version.map        |  3 +
 5 files changed, 168 insertions(+), 3 deletions(-)

diff --git a/doc/guides/prog_guide/eventdev.rst b/doc/guides/prog_guide/eventdev.rst
index ccde086f6..fda9c3743 100644
--- a/doc/guides/prog_guide/eventdev.rst
+++ b/doc/guides/prog_guide/eventdev.rst
@@ -63,13 +63,45 @@ the actual event being scheduled is. The payload is a union of the following:
 * ``uint64_t u64``
 * ``void *event_ptr``
 * ``struct rte_mbuf *mbuf``
+* ``struct rte_event_vector *vec``
 
-These three items in a union occupy the same 64 bits at the end of the rte_event
+These four items in a union occupy the same 64 bits at the end of the rte_event
 structure. The application can utilize the 64 bits directly by accessing the
-u64 variable, while the event_ptr and mbuf are provided as convenience
+u64 variable, while the event_ptr, mbuf, vec are provided as a convenience
 variables.  For example the mbuf pointer in the union can used to schedule a
 DPDK packet.
 
+Event Vector
+~~~~~~~~~~~~
+
+The rte_event_vector struct contains a vector of elements defined by the event
+type specified in the ``rte_event``. The event_vector structure contains the
+following data:
+
+* ``nb_elem`` - The number of elements held within the vector.
+
+Similar to ``rte_event`` the payload of event vector is also a union, allowing
+flexibility in what the actual vector is.
+
+* ``struct rte_mbuf *mbufs[0]`` - An array of mbufs.
+* ``void *ptrs[0]`` - An array of pointers.
+* ``uint64_t *u64s[0]`` - An array of uint64_t elements.
+
+The size of the event vector is related to the total number of elements it is
+configured to hold, this is achieved by making `rte_event_vector` a variable
+length structure.
+A helper function is provided to create a mempool that holds event vector, which
+takes name of the pool, total number of required ``rte_event_vector``,
+cache size, number of elements in each ``rte_event_vector`` and socket id.
+
+.. code-block:: c
+
+        rte_event_vector_pool_create("vector_pool", nb_event_vectors, cache_sz,
+                                     nb_elements_per_vector, socket_id);
+
+The function ``rte_event_vector_pool_create`` creates mempool with the best
+platform mempool ops.
+
 Queues
 ~~~~~~
 
diff --git a/doc/guides/rel_notes/release_21_05.rst b/doc/guides/rel_notes/release_21_05.rst
index 8a601e0a7..aeb56533a 100644
--- a/doc/guides/rel_notes/release_21_05.rst
+++ b/doc/guides/rel_notes/release_21_05.rst
@@ -106,6 +106,14 @@ New Features
   * Added support for periodic timer mode in eventdev timer adapter.
   * Added support for periodic timer mode in octeontx2 event device driver.
 
+* **Add Event device vector capability.**
+
+  * Added ``rte_event_vector`` data structure which is capable of holding
+    multiple ``uintptr_t`` of the same flow thereby allowing applications
+    to vectorize their pipelines and also reduce the complexity of pipelining
+    the events across multiple stages.
+  * This also reduces the scheduling overhead on a event device.
+
 
 Removed Items
 -------------
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index b57363f80..be0499c52 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -1266,6 +1266,48 @@ int rte_event_dev_selftest(uint8_t dev_id)
 	return -ENOTSUP;
 }
 
+struct rte_mempool *
+rte_event_vector_pool_create(const char *name, unsigned int n,
+			     unsigned int cache_size, uint16_t nb_elem,
+			     int socket_id)
+{
+	const char *mp_ops_name;
+	struct rte_mempool *mp;
+	unsigned int elt_sz;
+	int ret;
+
+	if (!nb_elem) {
+		RTE_LOG(ERR, EVENTDEV,
+			"Invalid number of elements=%d requested\n", nb_elem);
+		rte_errno = EINVAL;
+		return NULL;
+	}
+
+	elt_sz =
+		sizeof(struct rte_event_vector) + (nb_elem * sizeof(uintptr_t));
+	mp = rte_mempool_create_empty(name, n, elt_sz, cache_size, 0, socket_id,
+				      0);
+	if (mp == NULL)
+		return NULL;
+
+	mp_ops_name = rte_mbuf_best_mempool_ops();
+	ret = rte_mempool_set_ops_byname(mp, mp_ops_name, NULL);
+	if (ret != 0) {
+		RTE_LOG(ERR, EVENTDEV, "error setting mempool handler\n");
+		goto err;
+	}
+
+	ret = rte_mempool_populate_default(mp);
+	if (ret < 0)
+		goto err;
+
+	return mp;
+err:
+	rte_mempool_free(mp);
+	rte_errno = -ret;
+	return NULL;
+}
+
 int
 rte_event_dev_start(uint8_t dev_id)
 {
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index 9fc39e9ca..dee468ed0 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -212,8 +212,10 @@ extern "C" {
 
 #include <rte_common.h>
 #include <rte_config.h>
-#include <rte_memory.h>
 #include <rte_errno.h>
+#include <rte_mbuf_pool_ops.h>
+#include <rte_memory.h>
+#include <rte_mempool.h>
 
 #include "rte_eventdev_trace_fp.h"
 
@@ -913,6 +915,31 @@ rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
 int
 rte_event_dev_close(uint8_t dev_id);
 
+/**
+ * Event vector structure.
+ */
+struct rte_event_vector {
+	uint64_t nb_elem : 16;
+	/**< Number of elements in this event vector. */
+	uint64_t rsvd : 48;
+	/**< Reserved for future use */
+	uint64_t impl_opaque;
+	/**< Implementation specific opaque value.
+	 * An implementation may use this field to hold implementation specific
+	 * value to share between dequeue and enqueue operation.
+	 * The application should not modify this field.
+	 */
+	union {
+		struct rte_mbuf *mbufs[0];
+		void *ptrs[0];
+		uint64_t *u64s[0];
+	} __rte_aligned(16);
+	/**< Start of the vector array union. Depending upon the event type the
+	 * vector array can be an array of mbufs or pointers or opaque u64
+	 * values.
+	 */
+};
+
 /* Scheduler type definitions */
 #define RTE_SCHED_TYPE_ORDERED          0
 /**< Ordered scheduling
@@ -986,6 +1013,21 @@ rte_event_dev_close(uint8_t dev_id);
  */
 #define RTE_EVENT_TYPE_ETH_RX_ADAPTER   0x4
 /**< The event generated from event eth Rx adapter */
+#define RTE_EVENT_TYPE_VECTOR           0x8
+/**< Indicates that event is a vector.
+ * All vector event types should be a logical OR of EVENT_TYPE_VECTOR.
+ * This simplifies the pipeline design as one can split processing the events
+ * between vector events and normal event across event types.
+ * Example:
+ *	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
+ *		// Classify and handle vector event.
+ *	} else {
+ *		// Classify and handle event.
+ *	}
+ */
+#define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
+/**< The event vector generated from cpu for pipelining. */
+
 #define RTE_EVENT_TYPE_MAX              0x10
 /**< Maximum number of event types */
 
@@ -1108,6 +1150,8 @@ struct rte_event {
 		/**< Opaque event pointer */
 		struct rte_mbuf *mbuf;
 		/**< mbuf pointer if dequeued event is associated with mbuf */
+		struct rte_event_vector *vec;
+		/**< Event vector pointer. */
 	};
 };
 
@@ -2026,6 +2070,42 @@ rte_event_dev_xstats_reset(uint8_t dev_id,
  */
 int rte_event_dev_selftest(uint8_t dev_id);
 
+/**
+ * Get the memory required per event vector based on the number of elements per
+ * vector.
+ * This should be used to create the mempool that holds the event vectors.
+ *
+ * @param name
+ *   The name of the vector pool.
+ * @param n
+ *   The number of elements in the mbuf pool.
+ * @param cache_size
+ *   Size of the per-core object cache. See rte_mempool_create() for
+ *   details.
+ * @param nb_elem
+ *   The number of elements that a single event vector should be able to hold.
+ * @param socket_id
+ *   The socket identifier where the memory should be allocated. The
+ *   value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
+ *   reserved zone
+ *
+ * @return
+ *   The pointer to the newly allocated mempool, on success. NULL on error
+ *   with rte_errno set appropriately. Possible rte_errno values include:
+ *    - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
+ *    - E_RTE_SECONDARY - function was called from a secondary process instance
+ *    - EINVAL - cache size provided is too large, or priv_size is not aligned.
+ *    - ENOSPC - the maximum number of memzones has already been allocated
+ *    - EEXIST - a memzone with the same name already exists
+ *    - ENOMEM - no appropriate memory area found in which to create memzone
+ *    - ENAMETOOLONG - mempool name requested is too long.
+ */
+__rte_experimental
+struct rte_mempool *
+rte_event_vector_pool_create(const char *name, unsigned int n,
+			     unsigned int cache_size, uint16_t nb_elem,
+			     int socket_id);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index 3e5c09cfd..a070ef56e 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -138,6 +138,9 @@ EXPERIMENTAL {
 	__rte_eventdev_trace_port_setup;
 	# added in 20.11
 	rte_event_pmd_pci_probe_named;
+
+	#added in 21.05
+	rte_event_vector_pool_create;
 };
 
 INTERNAL {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v7 2/8] eventdev: introduce event vector Rx capability
  2021-03-25 17:10           ` [dpdk-dev] [PATCH v7 0/8] Introduce event vectorization pbhagavatula
  2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 1/8] eventdev: introduce event vector capability pbhagavatula
@ 2021-03-25 17:10             ` pbhagavatula
  2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 3/8] eventdev: introduce event vector Tx capability pbhagavatula
                               ` (6 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-25 17:10 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce event ethernet Rx adapter event vector capability.

If an event eth Rx adapter has the capability of
RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR then a given Rx queue
can be configured to enable event vectorization by passing the
flag RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR to
rte_event_eth_rx_adapter_queue_conf::rx_queue_flags while configuring
Rx adapter through rte_event_eth_rx_adapter_queue_add().

The max vector size, vector timeout define the vector size and
mempool used for allocating vector event are configured through
rte_event_eth_rx_adapter_queue_add. The element size of the element
in the vector pool should be equal to
    sizeof(struct rte_event_vector) + (vector_sz * sizeof(uintptr_t))

Application can use `rte_event_vector_pool_create` to create the
vector mempool used for
rte_event_eth_rx_adapter_queue_conf::vector_mp.

The Rx adapter would be responsible for vectorizing the mbufs
based on the flow, the vector limits configured by the application
and add the vector event of mbufs to the event queue set via
rte_event_eth_rx_adapter_queue_conf::ev::queue_id.
It should also mark rte_event_vector::union_valid and fill
rte_event_vector::port, rte_event_vector::queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
Acked-by: Ray Kinsella <mdr@ashroe.eu>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
---
 .../prog_guide/event_ethernet_rx_adapter.rst  |  38 ++++++
 lib/librte_eventdev/eventdev_pmd.h            |  53 ++++++++
 .../rte_event_eth_rx_adapter.c                | 114 ++++++++++++++++++
 .../rte_event_eth_rx_adapter.h                | 105 ++++++++++++++++
 lib/librte_eventdev/rte_eventdev.h            |  30 ++++-
 lib/librte_eventdev/version.map               |   2 +
 6 files changed, 340 insertions(+), 2 deletions(-)

diff --git a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
index cb44ce0e4..5eefef355 100644
--- a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
@@ -186,3 +186,41 @@ the event buffer fill level is low. The
 ``rte_event_eth_rx_adapter_cb_register()`` function allow the application
 to register a callback that selects which packets to enqueue to the event
 device.
+
+Rx event vectorization
+~~~~~~~~~~~~~~~~~~~~~~
+
+The event devices, ethernet device pairs which support the capability
+``RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR`` can aggregate packets based on
+flow characteristics and generate a ``rte_event`` containing ``rte_event_vector``
+whose event type is either ``RTE_EVENT_TYPE_ETHDEV_VECTOR`` or
+``RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR``.
+The aggregation size and timeout are configurable at a queue level and the
+maximum, minimum vector sizes and timeouts vary based on the device capability
+and can be queried using ``rte_event_eth_rx_adapter_vector_limits_get``.
+The Rx adapter additionally might include useful data such as ethernet device
+port and queue identifier in the ``rte_event_vector::port`` and
+``rte_event_vector::queue`` and mark ``rte_event_vector::attr_valid`` as true.
+
+A loop processing ``rte_event_vector`` containing mbufs is shown below.
+
+.. code-block:: c
+
+        event = rte_event_dequeue_burst(event_dev, event_port, &event,
+                                        1, 0);
+        if (!event)
+                continue;
+
+        switch (ev.event_type) {
+        case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
+        case RTE_EVENT_TYPE_ETHDEV_VECTOR:
+                struct rte_mbufs **mbufs;
+
+                mbufs = (struct rte_mbufs **)ev[i].vec->mbufs;
+                for (i = 0; i < ev.vec->nb_elem; i++) {
+                        /* Process each mbuf. */
+                }
+        break;
+        case ...
+        ...
+        }
diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 7eb9a7739..9297f1433 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -645,6 +645,53 @@ typedef int (*eventdev_eth_rx_adapter_stats_reset)
  */
 typedef int (*eventdev_selftest)(void);
 
+struct rte_event_eth_rx_adapter_vector_limits;
+/**
+ * Get event vector limits for a given event, ethernet device pair.
+ *
+ * @param dev
+ *   Event device pointer
+ *
+ * @param eth_dev
+ *   Ethernet device pointer
+ *
+ * @param[out] limits
+ *   Pointer to the limits structure to be filled.
+ *
+ * @return
+ *   - 0: Success.
+ *   - <0: Error code returned by the driver function.
+ */
+typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
+	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
+	struct rte_event_eth_rx_adapter_vector_limits *limits);
+
+struct rte_event_eth_rx_adapter_event_vector_config;
+/**
+ * Enable event vector on an given Rx queue of a ethernet devices belonging to
+ * the Rx adapter.
+ *
+ * @param dev
+ *   Event device pointer
+ *
+ * @param eth_dev
+ *   Ethernet device pointer
+ *
+ * @param rx_queue_id
+ *   The Rx queue identifier
+ *
+ * @param config
+ *   Pointer to the event vector configuration structure.
+ *
+ * @return
+ *   - 0: Success.
+ *   - <0: Error code returned by the driver function.
+ */
+typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
+	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
+	int32_t rx_queue_id,
+	const struct rte_event_eth_rx_adapter_event_vector_config *config);
+
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;
 
@@ -1067,6 +1114,12 @@ struct rte_eventdev_ops {
 	/**< Get ethernet Rx stats */
 	eventdev_eth_rx_adapter_stats_reset eth_rx_adapter_stats_reset;
 	/**< Reset ethernet Rx stats */
+	eventdev_eth_rx_adapter_vector_limits_get_t
+		eth_rx_adapter_vector_limits_get;
+	/**< Get event vector limits for the Rx adapter */
+	eventdev_eth_rx_adapter_event_vector_config_t
+		eth_rx_adapter_event_vector_config;
+	/**< Configure Rx adapter with event vector */
 
 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index d8c635e99..ac8ba5bf0 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -2263,6 +2263,120 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }
 
+int
+rte_event_eth_rx_adapter_queue_event_vector_config(
+	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
+	struct rte_event_eth_rx_adapter_event_vector_config *config)
+{
+	struct rte_event_eth_rx_adapter_vector_limits limits;
+	struct rte_event_eth_rx_adapter *rx_adapter;
+	struct rte_eventdev *dev;
+	uint32_t cap;
+	int ret;
+
+	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
+
+	rx_adapter = rxa_id_to_adapter(id);
+	if ((rx_adapter == NULL) || (config == NULL))
+		return -EINVAL;
+
+	dev = &rte_eventdevs[rx_adapter->eventdev_id];
+	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
+						eth_dev_id, &cap);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
+				 "eth port %" PRIu16,
+				 id, eth_dev_id);
+		return ret;
+	}
+
+	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
+		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
+	ret = rte_event_eth_rx_adapter_vector_limits_get(
+		rx_adapter->eventdev_id, eth_dev_id, &limits);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
+				 "eth port %" PRIu16,
+				 rx_adapter->eventdev_id, eth_dev_id);
+		return ret;
+	}
+
+	if (config->vector_sz < limits.min_sz ||
+	    config->vector_sz > limits.max_sz ||
+	    config->vector_timeout_ns < limits.min_timeout_ns ||
+	    config->vector_timeout_ns > limits.max_timeout_ns ||
+	    config->vector_mp == NULL) {
+		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+	if (config->vector_mp->elt_size <
+	    (sizeof(struct rte_event_vector) +
+	     (sizeof(uintptr_t) * config->vector_sz))) {
+		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
+	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
+		RTE_FUNC_PTR_OR_ERR_RET(
+			*dev->dev_ops->eth_rx_adapter_event_vector_config,
+			-ENOTSUP);
+		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
+			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
+	} else {
+		ret = -ENOTSUP;
+	}
+
+	return ret;
+}
+
+int
+rte_event_eth_rx_adapter_vector_limits_get(
+	uint8_t dev_id, uint16_t eth_port_id,
+	struct rte_event_eth_rx_adapter_vector_limits *limits)
+{
+	struct rte_eventdev *dev;
+	uint32_t cap;
+	int ret;
+
+	RTE_EVENTDEV_VALID_DEVID_OR_ERR_RET(dev_id, -EINVAL);
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_port_id, -EINVAL);
+
+	if (limits == NULL)
+		return -EINVAL;
+
+	dev = &rte_eventdevs[dev_id];
+
+	ret = rte_event_eth_rx_adapter_caps_get(dev_id, eth_port_id, &cap);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
+				 "eth port %" PRIu16,
+				 dev_id, eth_port_id);
+		return ret;
+	}
+
+	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
+		RTE_FUNC_PTR_OR_ERR_RET(
+			*dev->dev_ops->eth_rx_adapter_vector_limits_get,
+			-ENOTSUP);
+		ret = dev->dev_ops->eth_rx_adapter_vector_limits_get(
+			dev, &rte_eth_devices[eth_port_id], limits);
+	} else {
+		ret = -ENOTSUP;
+	}
+
+	return ret;
+}
+
 int
 rte_event_eth_rx_adapter_start(uint8_t id)
 {
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
index 21bb1e54c..7407cde00 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
@@ -92,6 +92,10 @@ extern "C" {
 /**< This flag indicates the flow identifier is valid
  * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
  */
+#define RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR	0x2
+/**< This flag indicates that mbufs arriving on the queue need to be vectorized
+ * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+ */
 
 /**
  * Adapter configuration structure that the adapter configuration callback
@@ -169,6 +173,36 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 */
 };
 
+struct rte_event_eth_rx_adapter_event_vector_config {
+	uint16_t vector_sz;
+	/**<
+	 * Indicates the maximum number for mbufs to combine and form a vector.
+	 * Should be within
+	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_sz
+	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_sz
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+	 */
+	uint64_t vector_timeout_ns;
+	/**<
+	 * Indicates the maximum number of nanoseconds to wait for receiving
+	 * mbufs. Should be within vectorization limits of the
+	 * adapter
+	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_ns
+	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_ns
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+	 */
+	struct rte_mempool *vector_mp;
+	/**<
+	 * Indicates the mempool that should be used for allocating
+	 * rte_event_vector container.
+	 * Should be created by using `rte_event_vector_pool_create`.
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags.
+	 */
+};
+
 /**
  * A structure used to retrieve statistics for an eth rx adapter instance.
  */
@@ -199,6 +233,32 @@ struct rte_event_eth_rx_adapter_stats {
 	/**< Received packet count for interrupt mode Rx queues */
 };
 
+/**
+ * A structure used to retrieve eth rx adapter vector limits.
+ */
+struct rte_event_eth_rx_adapter_vector_limits {
+	uint16_t min_sz;
+	/**< Minimum vector limit configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_sz
+	 */
+	uint16_t max_sz;
+	/**< Maximum vector limit configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_sz
+	 */
+	uint8_t log2_sz;
+	/**< True if the size configured should be in log2.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_sz
+	 */
+	uint64_t min_timeout_ns;
+	/**< Minimum vector timeout configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_timeout_ns
+	 */
+	uint64_t max_timeout_ns;
+	/**< Maximum vector timeout configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_timeout_ns
+	 */
+};
+
 /**
  *
  * Callback function invoked by the SW adapter before it continues
@@ -467,6 +527,51 @@ int rte_event_eth_rx_adapter_cb_register(uint8_t id, uint16_t eth_dev_id,
 					 rte_event_eth_rx_adapter_cb_fn cb_fn,
 					 void *cb_arg);
 
+/**
+ * Retrieve vector limits for a given event dev and eth dev pair.
+ * @see rte_event_eth_rx_adapter_vector_limits
+ *
+ * @param dev_id
+ *  Event device identifier.
+ * @param eth_port_id
+ *  Port identifier of the ethernet device.
+ * @param [out] limits
+ *  A pointer to rte_event_eth_rx_adapter_vector_limits structure that has to
+ * be filled.
+ *
+ * @return
+ *  - 0: Success.
+ *  - <0: Error code on failure.
+ */
+__rte_experimental
+int rte_event_eth_rx_adapter_vector_limits_get(
+	uint8_t dev_id, uint16_t eth_port_id,
+	struct rte_event_eth_rx_adapter_vector_limits *limits);
+
+/**
+ * Configure event vectorization for a given ethernet device queue, that has
+ * been added to a event eth Rx adapter.
+ *
+ * @param id
+ *  The identifier of the ethernet Rx event adapter.
+ *
+ * @param eth_dev_id
+ *  The identifier of the ethernet device.
+ *
+ * @param rx_queue_id
+ *  Ethernet device receive queue index.
+ *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
+ *  are configured with event vectorization.
+ *
+ * @return
+ *  - 0: Success, Receive queue configured correctly.
+ *  - <0: Error code on failure.
+ */
+__rte_experimental
+int rte_event_eth_rx_adapter_queue_event_vector_config(
+	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
+	struct rte_event_eth_rx_adapter_event_vector_config *config);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index dee468ed0..936c97cc2 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -919,10 +919,28 @@ rte_event_dev_close(uint8_t dev_id);
  * Event vector structure.
  */
 struct rte_event_vector {
-	uint64_t nb_elem : 16;
+	uint16_t nb_elem;
 	/**< Number of elements in this event vector. */
-	uint64_t rsvd : 48;
+	uint16_t rsvd : 15;
 	/**< Reserved for future use */
+	uint16_t attr_valid : 1;
+	/**< Indicates that the below union attributes have valid information.
+	 */
+	union {
+		/* Used by Rx adapter.
+		 * Indicates that all the elements in this vector belong to the
+		 * same port and queue pair when originating from Rx adapter,
+		 * valid only when event type is ETHDEV_VECTOR or
+		 * ETH_RX_ADAPTER_VECTOR.
+		 */
+		struct {
+			uint16_t port;
+			/* Ethernet device port id. */
+			uint16_t queue;
+			/* Ethernet device queue id. */
+		};
+	};
+	/**< Union to hold common attributes of the vector array. */
 	uint64_t impl_opaque;
 	/**< Implementation specific opaque value.
 	 * An implementation may use this field to hold implementation specific
@@ -1025,8 +1043,14 @@ struct rte_event_vector {
  *		// Classify and handle event.
  *	}
  */
+#define RTE_EVENT_TYPE_ETHDEV_VECTOR                                           \
+	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETHDEV)
+/**< The event vector generated from ethdev subsystem */
 #define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
 /**< The event vector generated from cpu for pipelining. */
+#define RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR                                   \
+	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETH_RX_ADAPTER)
+/**< The event vector generated from eth Rx adapter. */
 
 #define RTE_EVENT_TYPE_MAX              0x10
 /**< Maximum number of event types */
@@ -1171,6 +1195,8 @@ struct rte_event {
  * @see struct rte_event_eth_rx_adapter_queue_conf::ev
  * @see struct rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
  */
+#define RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR	0x8
+/**< Adapter supports event vectorization per ethdev. */
 
 /**
  * Retrieve the event device's ethdev Rx adapter capabilities for the
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index a070ef56e..902df0ae3 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -141,6 +141,8 @@ EXPERIMENTAL {
 
 	#added in 21.05
 	rte_event_vector_pool_create;
+	rte_event_eth_rx_adapter_vector_limits_get;
+	rte_event_eth_rx_adapter_queue_event_vector_config;
 };
 
 INTERNAL {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v7 3/8] eventdev: introduce event vector Tx capability
  2021-03-25 17:10           ` [dpdk-dev] [PATCH v7 0/8] Introduce event vectorization pbhagavatula
  2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 1/8] eventdev: introduce event vector capability pbhagavatula
  2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 2/8] eventdev: introduce event vector Rx capability pbhagavatula
@ 2021-03-25 17:10             ` pbhagavatula
  2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 4/8] eventdev: add Rx adapter event vector support pbhagavatula
                               ` (5 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-25 17:10 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce event vector transmit capability for event eth
tx adapter.

The capability indicates that the Tx adapter is capable of
transmitting event vectors.
When rte_event_vector::union_valid is set, the Tx adapter should
transmit all the packets to the rte_event_vector::port using the
rte_event_vector::queue.
If rte_event_vector::union_valid is not set then the Tx adapter
should peek into each mbuf to get the destination port and queue
pair.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
---
 doc/guides/prog_guide/event_ethernet_tx_adapter.rst | 12 ++++++++++++
 lib/librte_eventdev/rte_eventdev.h                  |  8 +++++++-
 2 files changed, 19 insertions(+), 1 deletion(-)

diff --git a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
index a8c13e136..f80d22621 100644
--- a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
@@ -164,3 +164,15 @@ The  ``rte_event_eth_tx_adapter_stats_get()`` function reports counters defined
 in struct ``rte_event_eth_tx_adapter_stats``. The counter values are the sum of
 the counts from the eventdev PMD callback if the callback is supported, and
 the counts maintained by the service function, if one exists.
+
+Tx event vectorization
+~~~~~~~~~~~~~~~~~~~~~~
+
+The event device, ethernet device pairs which support the capability
+``RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR`` can process event vector of mbufs.
+Additionally, application can provide a hint to the Tx adapter that all the
+mbufs are destined to the same ethernet port and queue by setting the bit
+``rte_event_vector::attr_valid`` and filling `rte_event_vector::port`` and
+``rte_event_vector::queue``.
+If ``rte_event_vector::attr_valid`` is not set then the Tx adapter should peek
+into each mbuf and transmit them to the requested ethernet port and queue pair.
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index 936c97cc2..5f1f544cc 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -927,11 +927,13 @@ struct rte_event_vector {
 	/**< Indicates that the below union attributes have valid information.
 	 */
 	union {
-		/* Used by Rx adapter.
+		/* Used by Rx/Tx adapter.
 		 * Indicates that all the elements in this vector belong to the
 		 * same port and queue pair when originating from Rx adapter,
 		 * valid only when event type is ETHDEV_VECTOR or
 		 * ETH_RX_ADAPTER_VECTOR.
+		 * Can also be used to indicate the Tx adapter the destination
+		 * port and queue of the mbufs in the vector
 		 */
 		struct {
 			uint16_t port;
@@ -1296,6 +1298,10 @@ rte_event_crypto_adapter_caps_get(uint8_t dev_id, uint8_t cdev_id,
 #define RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT	0x1
 /**< This flag is sent when the PMD supports a packet transmit callback
  */
+#define RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR	0x2
+/**< Indicates that the Tx adapter is capable of handling event vector of
+ * mbufs.
+ */
 
 /**
  * Retrieve the event device's eth Tx adapter capabilities
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v7 4/8] eventdev: add Rx adapter event vector support
  2021-03-25 17:10           ` [dpdk-dev] [PATCH v7 0/8] Introduce event vectorization pbhagavatula
                               ` (2 preceding siblings ...)
  2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 3/8] eventdev: introduce event vector Tx capability pbhagavatula
@ 2021-03-25 17:10             ` pbhagavatula
  2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 5/8] eventdev: add Tx " pbhagavatula
                               ` (4 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-25 17:10 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support for event eth Rx adapter, the implementation
creates vector flows based on port and queue identifier of the received
mbufs.
The flow id for SW Rx event vectorization will use the queue and port
identifier for simplicity.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 lib/librte_eventdev/eventdev_pmd.h            |   7 +-
 .../rte_event_eth_rx_adapter.c                | 263 ++++++++++++++++--
 lib/librte_eventdev/rte_eventdev.c            |   6 +-
 3 files changed, 256 insertions(+), 20 deletions(-)

diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 9297f1433..0f724ac85 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -69,9 +69,10 @@ extern "C" {
 	} \
 } while (0)

-#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP \
-		((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) | \
-			(RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ))
+#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP                                        \
+	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |                     \
+	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) |                         \
+	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR))

 #define RTE_EVENT_CRYPTO_ADAPTER_SW_CAP \
 		RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index ac8ba5bf0..a3227df92 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -26,6 +26,10 @@
 #define BATCH_SIZE		32
 #define BLOCK_CNT_THRESHOLD	10
 #define ETH_EVENT_BUFFER_SIZE	(4*BATCH_SIZE)
+#define MAX_VECTOR_SIZE		1024
+#define MIN_VECTOR_SIZE		4
+#define MAX_VECTOR_NS		1E9
+#define MIN_VECTOR_NS		1E5

 #define ETH_RX_ADAPTER_SERVICE_NAME_LEN	32
 #define ETH_RX_ADAPTER_MEM_NAME_LEN	32
@@ -59,6 +63,20 @@ struct eth_rx_poll_entry {
 	uint16_t eth_rx_qid;
 };

+struct eth_rx_vector_data {
+	TAILQ_ENTRY(eth_rx_vector_data) next;
+	uint16_t port;
+	uint16_t queue;
+	uint16_t max_vector_count;
+	uint64_t event;
+	uint64_t ts;
+	uint64_t vector_timeout_ticks;
+	struct rte_mempool *vector_pool;
+	struct rte_event_vector *vector_ev;
+} __rte_cache_aligned;
+
+TAILQ_HEAD(eth_rx_vector_data_list, eth_rx_vector_data);
+
 /* Instance per adapter */
 struct rte_eth_event_enqueue_buffer {
 	/* Count of events in this buffer */
@@ -92,6 +110,14 @@ struct rte_event_eth_rx_adapter {
 	uint32_t wrr_pos;
 	/* Event burst buffer */
 	struct rte_eth_event_enqueue_buffer event_enqueue_buffer;
+	/* Vector enable flag */
+	uint8_t ena_vector;
+	/* Timestamp of previous vector expiry list traversal */
+	uint64_t prev_expiry_ts;
+	/* Minimum ticks to wait before traversing expiry list */
+	uint64_t vector_tmo_ticks;
+	/* vector list */
+	struct eth_rx_vector_data_list vector_list;
 	/* Per adapter stats */
 	struct rte_event_eth_rx_adapter_stats stats;
 	/* Block count, counts up to BLOCK_CNT_THRESHOLD */
@@ -198,9 +224,11 @@ struct eth_device_info {
 struct eth_rx_queue_info {
 	int queue_enabled;	/* True if added */
 	int intr_enabled;
+	uint8_t ena_vector;
 	uint16_t wt;		/* Polling weight */
 	uint32_t flow_id_mask;	/* Set to ~0 if app provides flow id else 0 */
 	uint64_t event;
+	struct eth_rx_vector_data vector_data;
 };

 static struct rte_event_eth_rx_adapter **event_eth_rx_adapter;
@@ -722,6 +750,9 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
 	    &rx_adapter->event_enqueue_buffer;
 	struct rte_event_eth_rx_adapter_stats *stats = &rx_adapter->stats;

+	if (!buf->count)
+		return 0;
+
 	uint16_t n = rte_event_enqueue_new_burst(rx_adapter->eventdev_id,
 					rx_adapter->event_port_id,
 					buf->events,
@@ -742,6 +773,77 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
 	return n;
 }

+static inline void
+rxa_init_vector(struct rte_event_eth_rx_adapter *rx_adapter,
+		struct eth_rx_vector_data *vec)
+{
+	vec->vector_ev->nb_elem = 0;
+	vec->vector_ev->port = vec->port;
+	vec->vector_ev->queue = vec->queue;
+	vec->vector_ev->attr_valid = true;
+	TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
+}
+
+static inline uint16_t
+rxa_create_event_vector(struct rte_event_eth_rx_adapter *rx_adapter,
+			struct eth_rx_queue_info *queue_info,
+			struct rte_eth_event_enqueue_buffer *buf,
+			struct rte_mbuf **mbufs, uint16_t num)
+{
+	struct rte_event *ev = &buf->events[buf->count];
+	struct eth_rx_vector_data *vec;
+	uint16_t filled, space, sz;
+
+	filled = 0;
+	vec = &queue_info->vector_data;
+
+	if (vec->vector_ev == NULL) {
+		if (rte_mempool_get(vec->vector_pool,
+				    (void **)&vec->vector_ev) < 0) {
+			rte_pktmbuf_free_bulk(mbufs, num);
+			return 0;
+		}
+		rxa_init_vector(rx_adapter, vec);
+	}
+	while (num) {
+		if (vec->vector_ev->nb_elem == vec->max_vector_count) {
+			/* Event ready. */
+			ev->event = vec->event;
+			ev->vec = vec->vector_ev;
+			ev++;
+			filled++;
+			vec->vector_ev = NULL;
+			TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+			if (rte_mempool_get(vec->vector_pool,
+					    (void **)&vec->vector_ev) < 0) {
+				rte_pktmbuf_free_bulk(mbufs, num);
+				return 0;
+			}
+			rxa_init_vector(rx_adapter, vec);
+		}
+
+		space = vec->max_vector_count - vec->vector_ev->nb_elem;
+		sz = num > space ? space : num;
+		memcpy(vec->vector_ev->mbufs + vec->vector_ev->nb_elem, mbufs,
+		       sizeof(void *) * sz);
+		vec->vector_ev->nb_elem += sz;
+		num -= sz;
+		mbufs += sz;
+		vec->ts = rte_rdtsc();
+	}
+
+	if (vec->vector_ev->nb_elem == vec->max_vector_count) {
+		ev->event = vec->event;
+		ev->vec = vec->vector_ev;
+		ev++;
+		filled++;
+		vec->vector_ev = NULL;
+		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+	}
+
+	return filled;
+}
+
 static inline void
 rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
@@ -770,25 +872,30 @@ rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
 	rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
 	do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;

-	for (i = 0; i < num; i++) {
-		m = mbufs[i];
-
-		rss = do_rss ?
-			rxa_do_softrss(m, rx_adapter->rss_key_be) :
-			m->hash.rss;
-		ev->event = event;
-		ev->flow_id = (rss & ~flow_id_mask) |
-				(ev->flow_id & flow_id_mask);
-		ev->mbuf = m;
-		ev++;
+	if (!eth_rx_queue_info->ena_vector) {
+		for (i = 0; i < num; i++) {
+			m = mbufs[i];
+
+			rss = do_rss ? rxa_do_softrss(m, rx_adapter->rss_key_be)
+				     : m->hash.rss;
+			ev->event = event;
+			ev->flow_id = (rss & ~flow_id_mask) |
+				      (ev->flow_id & flow_id_mask);
+			ev->mbuf = m;
+			ev++;
+		}
+	} else {
+		num = rxa_create_event_vector(rx_adapter, eth_rx_queue_info,
+					      buf, mbufs, num);
 	}

-	if (dev_info->cb_fn) {
+	if (num && dev_info->cb_fn) {

 		dropped = 0;
 		nb_cb = dev_info->cb_fn(eth_dev_id, rx_queue_id,
-					ETH_EVENT_BUFFER_SIZE, buf->count, ev,
-					num, dev_info->cb_arg, &dropped);
+					ETH_EVENT_BUFFER_SIZE, buf->count,
+					&buf->events[buf->count], num,
+					dev_info->cb_arg, &dropped);
 		if (unlikely(nb_cb > num))
 			RTE_EDEV_LOG_ERR("Rx CB returned %d (> %d) events",
 				nb_cb, num);
@@ -1124,6 +1231,30 @@ rxa_poll(struct rte_event_eth_rx_adapter *rx_adapter)
 	return nb_rx;
 }

+static void
+rxa_vector_expire(struct eth_rx_vector_data *vec, void *arg)
+{
+	struct rte_event_eth_rx_adapter *rx_adapter = arg;
+	struct rte_eth_event_enqueue_buffer *buf =
+		&rx_adapter->event_enqueue_buffer;
+	struct rte_event *ev;
+
+	if (buf->count)
+		rxa_flush_event_buffer(rx_adapter);
+
+	if (vec->vector_ev->nb_elem == 0)
+		return;
+	ev = &buf->events[buf->count];
+
+	/* Event ready. */
+	ev->event = vec->event;
+	ev->vec = vec->vector_ev;
+	buf->count++;
+
+	vec->vector_ev = NULL;
+	vec->ts = 0;
+}
+
 static int
 rxa_service_func(void *args)
 {
@@ -1137,6 +1268,24 @@ rxa_service_func(void *args)
 		return 0;
 	}

+	if (rx_adapter->ena_vector) {
+		if ((rte_rdtsc() - rx_adapter->prev_expiry_ts) >=
+		    rx_adapter->vector_tmo_ticks) {
+			struct eth_rx_vector_data *vec;
+
+			TAILQ_FOREACH(vec, &rx_adapter->vector_list, next) {
+				uint64_t elapsed_time = rte_rdtsc() - vec->ts;
+
+				if (elapsed_time >= vec->vector_timeout_ticks) {
+					rxa_vector_expire(vec, rx_adapter);
+					TAILQ_REMOVE(&rx_adapter->vector_list,
+						     vec, next);
+				}
+			}
+			rx_adapter->prev_expiry_ts = rte_rdtsc();
+		}
+	}
+
 	stats = &rx_adapter->stats;
 	stats->rx_packets += rxa_intr_ring_dequeue(rx_adapter);
 	stats->rx_packets += rxa_poll(rx_adapter);
@@ -1640,6 +1789,29 @@ rxa_update_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }

+static void
+rxa_set_vector_data(struct eth_rx_queue_info *queue_info, uint16_t vector_count,
+		    uint64_t vector_ns, struct rte_mempool *mp, int32_t qid,
+		    uint16_t port_id)
+{
+#define NSEC2TICK(__ns, __freq) (((__ns) * (__freq)) / 1E9)
+	struct eth_rx_vector_data *vector_data;
+	uint32_t flow_id;
+
+	vector_data = &queue_info->vector_data;
+	vector_data->max_vector_count = vector_count;
+	vector_data->port = port_id;
+	vector_data->queue = qid;
+	vector_data->vector_pool = mp;
+	vector_data->vector_timeout_ticks =
+		NSEC2TICK(vector_ns, rte_get_timer_hz());
+	vector_data->ts = 0;
+	flow_id = queue_info->event & 0xFFFFF;
+	flow_id =
+		flow_id == 0 ? (qid & 0xFFF) | (port_id & 0xFF) << 12 : flow_id;
+	vector_data->event = (queue_info->event & ~0xFFFFF) | flow_id;
+}
+
 static void
 rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
 	struct eth_device_info *dev_info,
@@ -1741,6 +1913,44 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }

+static void
+rxa_sw_event_vector_configure(
+	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
+	int rx_queue_id,
+	const struct rte_event_eth_rx_adapter_event_vector_config *config)
+{
+	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
+	struct eth_rx_queue_info *queue_info;
+	struct rte_event *qi_ev;
+
+	if (rx_queue_id == -1) {
+		uint16_t nb_rx_queues;
+		uint16_t i;
+
+		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
+		for (i = 0; i < nb_rx_queues; i++)
+			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
+						      config);
+		return;
+	}
+
+	queue_info = &dev_info->rx_queue[rx_queue_id];
+	qi_ev = (struct rte_event *)&queue_info->event;
+	queue_info->ena_vector = 1;
+	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+	rxa_set_vector_data(queue_info, config->vector_sz,
+			    config->vector_timeout_ns, config->vector_mp,
+			    rx_queue_id, dev_info->dev->data->port_id);
+	rx_adapter->ena_vector = 1;
+	rx_adapter->vector_tmo_ticks =
+		rx_adapter->vector_tmo_ticks ?
+			      RTE_MIN(config->vector_timeout_ns << 1,
+				      rx_adapter->vector_tmo_ticks) :
+			      config->vector_timeout_ns << 1;
+	rx_adapter->prev_expiry_ts = 0;
+	TAILQ_INIT(&rx_adapter->vector_list);
+}
+
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -2081,6 +2291,15 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}

+	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
+	    (queue_conf->rx_queue_flags &
+	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
+		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
 		(rx_queue_id != -1)) {
 		RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
@@ -2143,6 +2362,17 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	return 0;
 }

+static int
+rxa_sw_vector_limits(struct rte_event_eth_rx_adapter_vector_limits *limits)
+{
+	limits->max_sz = MAX_VECTOR_SIZE;
+	limits->min_sz = MIN_VECTOR_SIZE;
+	limits->max_timeout_ns = MAX_VECTOR_NS;
+	limits->min_timeout_ns = MIN_VECTOR_NS;
+
+	return 0;
+}
+
 int
 rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 				int32_t rx_queue_id)
@@ -2333,7 +2563,8 @@ rte_event_eth_rx_adapter_queue_event_vector_config(
 		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
 			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
 	} else {
-		ret = -ENOTSUP;
+		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
+					      rx_queue_id, config);
 	}

 	return ret;
@@ -2371,7 +2602,7 @@ rte_event_eth_rx_adapter_vector_limits_get(
 		ret = dev->dev_ops->eth_rx_adapter_vector_limits_get(
 			dev, &rte_eth_devices[eth_port_id], limits);
 	} else {
-		ret = -ENOTSUP;
+		ret = rxa_sw_vector_limits(limits);
 	}

 	return ret;
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index be0499c52..62824654b 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -122,7 +122,11 @@ rte_event_eth_rx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,

 	if (caps == NULL)
 		return -EINVAL;
-	*caps = 0;
+
+	if (dev->dev_ops->eth_rx_adapter_caps_get == NULL)
+		*caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
+	else
+		*caps = 0;

 	return dev->dev_ops->eth_rx_adapter_caps_get ?
 				(*dev->dev_ops->eth_rx_adapter_caps_get)(dev,
--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v7 5/8] eventdev: add Tx adapter event vector support
  2021-03-25 17:10           ` [dpdk-dev] [PATCH v7 0/8] Introduce event vectorization pbhagavatula
                               ` (3 preceding siblings ...)
  2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 4/8] eventdev: add Rx adapter event vector support pbhagavatula
@ 2021-03-25 17:10             ` pbhagavatula
  2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
                               ` (3 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-25 17:10 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support for event eth Tx adapter, the implementation
receives events from the single linked queue and based on
rte_event_vector::attr_valid transmits the vector of mbufs to a given
port, queue pair.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
---
 .../rte_event_eth_tx_adapter.c                | 66 ++++++++++++++++---
 lib/librte_eventdev/rte_eventdev.c            |  5 +-
 2 files changed, 60 insertions(+), 11 deletions(-)

diff --git a/lib/librte_eventdev/rte_event_eth_tx_adapter.c b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
index 5b4c42dcf..db260bfb6 100644
--- a/lib/librte_eventdev/rte_event_eth_tx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
@@ -510,6 +510,47 @@ txa_service_buffer_retry(struct rte_mbuf **pkts, uint16_t unsent,
 	stats->tx_dropped += unsent - sent;
 }
 
+static uint16_t
+txa_process_event_vector(struct txa_service_data *txa,
+			 struct rte_event_vector *vec)
+{
+	struct txa_service_queue_info *tqi;
+	uint16_t port, queue, nb_tx = 0;
+	struct rte_mbuf **mbufs;
+	int i;
+
+	mbufs = (struct rte_mbuf **)vec->mbufs;
+	if (vec->attr_valid) {
+		port = vec->port;
+		queue = vec->queue;
+		tqi = txa_service_queue(txa, port, queue);
+		if (unlikely(tqi == NULL || !tqi->added)) {
+			rte_pktmbuf_free_bulk(mbufs, vec->nb_elem);
+			rte_mempool_put(rte_mempool_from_obj(vec), vec);
+			return 0;
+		}
+		for (i = 0; i < vec->nb_elem; i++) {
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
+						   mbufs[i]);
+		}
+	} else {
+		for (i = 0; i < vec->nb_elem; i++) {
+			port = mbufs[i]->port;
+			queue = rte_event_eth_tx_adapter_txq_get(mbufs[i]);
+			tqi = txa_service_queue(txa, port, queue);
+			if (unlikely(tqi == NULL || !tqi->added)) {
+				rte_pktmbuf_free(mbufs[i]);
+				continue;
+			}
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
+						   mbufs[i]);
+		}
+	}
+	rte_mempool_put(rte_mempool_from_obj(vec), vec);
+
+	return nb_tx;
+}
+
 static void
 txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
 	uint32_t n)
@@ -522,22 +563,27 @@ txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
 
 	nb_tx = 0;
 	for (i = 0; i < n; i++) {
-		struct rte_mbuf *m;
 		uint16_t port;
 		uint16_t queue;
 		struct txa_service_queue_info *tqi;
 
-		m = ev[i].mbuf;
-		port = m->port;
-		queue = rte_event_eth_tx_adapter_txq_get(m);
+		if (!(ev[i].event_type & RTE_EVENT_TYPE_VECTOR)) {
+			struct rte_mbuf *m;
 
-		tqi = txa_service_queue(txa, port, queue);
-		if (unlikely(tqi == NULL || !tqi->added)) {
-			rte_pktmbuf_free(m);
-			continue;
-		}
+			m = ev[i].mbuf;
+			port = m->port;
+			queue = rte_event_eth_tx_adapter_txq_get(m);
 
-		nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
+			tqi = txa_service_queue(txa, port, queue);
+			if (unlikely(tqi == NULL || !tqi->added)) {
+				rte_pktmbuf_free(m);
+				continue;
+			}
+
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
+		} else {
+			nb_tx += txa_process_event_vector(txa, ev[i].vec);
+		}
 	}
 
 	stats->tx_packets += nb_tx;
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index 62824654b..c9bb5d227 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -196,7 +196,10 @@ rte_event_eth_tx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
 	if (caps == NULL)
 		return -EINVAL;
 
-	*caps = 0;
+	if (dev->dev_ops->eth_tx_adapter_caps_get == NULL)
+		*caps = RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR;
+	else
+		*caps = 0;
 
 	return dev->dev_ops->eth_tx_adapter_caps_get ?
 			(*dev->dev_ops->eth_tx_adapter_caps_get)(dev,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v7 6/8] app/eventdev: add event vector mode in pipeline test
  2021-03-25 17:10           ` [dpdk-dev] [PATCH v7 0/8] Introduce event vectorization pbhagavatula
                               ` (4 preceding siblings ...)
  2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 5/8] eventdev: add Tx " pbhagavatula
@ 2021-03-25 17:10             ` pbhagavatula
  2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 7/8] doc: announce event Rx adapter config changes pbhagavatula
                               ` (2 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-25 17:10 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support in pipeline tests. By default this mode
is disabled, it can be enabled by using the option --enable_vector.
example:
	dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
	--nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
	--wlcores=20-23  --enable_vector

Additional options to configure vector size and vector timeout are
also implemented and can be used by specifying --vector_size and
--vector_tmo_ns

This patch also adds a new option to set the number of Rx queues
configured per event eth rx adapter.
example:
	dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
	--nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
	--wlcores=20-23  --nb_eth_queues 4

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
---
 app/test-eventdev/evt_common.h           |   4 +
 app/test-eventdev/evt_options.c          |  52 ++++
 app/test-eventdev/evt_options.h          |   4 +
 app/test-eventdev/test_pipeline_atq.c    | 310 ++++++++++++++++++++--
 app/test-eventdev/test_pipeline_common.c | 113 +++++++-
 app/test-eventdev/test_pipeline_common.h |  18 ++
 app/test-eventdev/test_pipeline_queue.c  | 320 +++++++++++++++++++++--
 doc/guides/tools/testeventdev.rst        |  45 +++-
 8 files changed, 811 insertions(+), 55 deletions(-)

diff --git a/app/test-eventdev/evt_common.h b/app/test-eventdev/evt_common.h
index a1da1cf11..0e228258e 100644
--- a/app/test-eventdev/evt_common.h
+++ b/app/test-eventdev/evt_common.h
@@ -58,16 +58,20 @@ struct evt_options {
 	uint8_t sched_type_list[EVT_MAX_STAGES];
 	uint16_t mbuf_sz;
 	uint16_t wkr_deq_dep;
+	uint16_t vector_size;
+	uint16_t eth_queues;
 	uint32_t nb_flows;
 	uint32_t tx_first;
 	uint32_t max_pkt_sz;
 	uint32_t deq_tmo_nsec;
 	uint32_t q_priority:1;
 	uint32_t fwd_latency:1;
+	uint32_t ena_vector : 1;
 	uint64_t nb_pkts;
 	uint64_t nb_timers;
 	uint64_t expiry_nsec;
 	uint64_t max_tmo_nsec;
+	uint64_t vector_tmo_nsec;
 	uint64_t timer_tick_nsec;
 	uint64_t optm_timer_tick_nsec;
 	enum evt_prod_type prod_type;
diff --git a/app/test-eventdev/evt_options.c b/app/test-eventdev/evt_options.c
index 0d04ea9f8..0d5540574 100644
--- a/app/test-eventdev/evt_options.c
+++ b/app/test-eventdev/evt_options.c
@@ -34,6 +34,9 @@ evt_options_default(struct evt_options *opt)
 	opt->max_tmo_nsec = 1E5;  /* 100000ns ~100us */
 	opt->expiry_nsec = 1E4;   /* 10000ns ~10us */
 	opt->prod_type = EVT_PROD_TYPE_SYNT;
+	opt->eth_queues = 1;
+	opt->vector_size = 64;
+	opt->vector_tmo_nsec = 100E3;
 }
 
 typedef int (*option_parser_t)(struct evt_options *opt,
@@ -257,6 +260,43 @@ evt_parse_max_pkt_sz(struct evt_options *opt, const char *arg)
 	return ret;
 }
 
+static int
+evt_parse_ena_vector(struct evt_options *opt, const char *arg __rte_unused)
+{
+	opt->ena_vector = 1;
+	return 0;
+}
+
+static int
+evt_parse_vector_size(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint16(&(opt->vector_size), arg);
+
+	return ret;
+}
+
+static int
+evt_parse_vector_tmo_ns(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint64(&(opt->vector_tmo_nsec), arg);
+
+	return ret;
+}
+
+static int
+evt_parse_eth_queues(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint16(&(opt->eth_queues), arg);
+
+	return ret;
+}
+
 static void
 usage(char *program)
 {
@@ -289,6 +329,10 @@ usage(char *program)
 		"\t--expiry_nsec      : event timer expiry ns.\n"
 		"\t--mbuf_sz          : packet mbuf size.\n"
 		"\t--max_pkt_sz       : max packet size.\n"
+		"\t--nb_eth_queues    : number of ethernet Rx queues.\n"
+		"\t--enable_vector    : enable event vectorization.\n"
+		"\t--vector_size      : Max vector size.\n"
+		"\t--vector_tmo_ns    : Max vector timeout in nanoseconds\n"
 		);
 	printf("available tests:\n");
 	evt_test_dump_names();
@@ -360,6 +404,10 @@ static struct option lgopts[] = {
 	{ EVT_EXPIRY_NSEC,         1, 0, 0 },
 	{ EVT_MBUF_SZ,             1, 0, 0 },
 	{ EVT_MAX_PKT_SZ,          1, 0, 0 },
+	{ EVT_NB_ETH_QUEUES,       1, 0, 0 },
+	{ EVT_ENA_VECTOR,          0, 0, 0 },
+	{ EVT_VECTOR_SZ,           1, 0, 0 },
+	{ EVT_VECTOR_TMO,          1, 0, 0 },
 	{ EVT_HELP,                0, 0, 0 },
 	{ NULL,                    0, 0, 0 }
 };
@@ -394,6 +442,10 @@ evt_opts_parse_long(int opt_idx, struct evt_options *opt)
 		{ EVT_EXPIRY_NSEC, evt_parse_expiry_nsec},
 		{ EVT_MBUF_SZ, evt_parse_mbuf_sz},
 		{ EVT_MAX_PKT_SZ, evt_parse_max_pkt_sz},
+		{ EVT_NB_ETH_QUEUES, evt_parse_eth_queues},
+		{ EVT_ENA_VECTOR, evt_parse_ena_vector},
+		{ EVT_VECTOR_SZ, evt_parse_vector_size},
+		{ EVT_VECTOR_TMO, evt_parse_vector_tmo_ns},
 	};
 
 	for (i = 0; i < RTE_DIM(parsermap); i++) {
diff --git a/app/test-eventdev/evt_options.h b/app/test-eventdev/evt_options.h
index 748e54fae..1cea2a3e1 100644
--- a/app/test-eventdev/evt_options.h
+++ b/app/test-eventdev/evt_options.h
@@ -42,6 +42,10 @@
 #define EVT_EXPIRY_NSEC          ("expiry_nsec")
 #define EVT_MBUF_SZ              ("mbuf_sz")
 #define EVT_MAX_PKT_SZ           ("max_pkt_sz")
+#define EVT_NB_ETH_QUEUES        ("nb_eth_queues")
+#define EVT_ENA_VECTOR           ("enable_vector")
+#define EVT_VECTOR_SZ            ("vector_size")
+#define EVT_VECTOR_TMO           ("vector_tmo_ns")
 #define EVT_HELP                 ("help")
 
 void evt_options_default(struct evt_options *opt);
diff --git a/app/test-eventdev/test_pipeline_atq.c b/app/test-eventdev/test_pipeline_atq.c
index 0872b25b5..84dd4f44e 100644
--- a/app/test-eventdev/test_pipeline_atq.c
+++ b/app/test-eventdev/test_pipeline_atq.c
@@ -15,6 +15,8 @@ pipeline_atq_nb_event_queues(struct evt_options *opt)
 	return rte_eth_dev_count_avail();
 }
 
+typedef int (*pipeline_atq_worker_t)(void *arg);
+
 static __rte_noinline int
 pipeline_atq_worker_single_stage_tx(void *arg)
 {
@@ -113,6 +115,112 @@ pipeline_atq_worker_single_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_atq_worker_single_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+		vector_sz = ev.vec->nb_elem;
+		pipeline_event_tx_vector(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = ev.vec->nb_elem;
+		ev.queue_id = tx_queue[ev.vec->port];
+		ev.vec->queue = 0;
+		pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			vector_sz += ev[i].vec->nb_elem;
+			ev[i].vec->queue = 0;
+		}
+
+		pipeline_event_tx_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			ev[i].queue_id = tx_queue[ev[i].vec->port];
+			ev[i].vec->queue = 0;
+			vector_sz += ev[i].vec->nb_elem;
+			pipeline_fwd_event_vector(&ev[i],
+						  RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
 static __rte_noinline int
 pipeline_atq_worker_multi_stage_tx(void *arg)
 {
@@ -245,6 +353,147 @@ pipeline_atq_worker_multi_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.sub_event_type % nb_stages;
+
+		if (cq_id == last_queue) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+			continue;
+		}
+
+		ev.sub_event_type++;
+		pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.sub_event_type % nb_stages;
+
+		if (cq_id == last_queue) {
+			ev.queue_id = tx_queue[ev.vec->port];
+			ev.vec->queue = 0;
+			vector_sz = ev.vec->nb_elem;
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			pipeline_event_enqueue(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.sub_event_type++;
+			pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+			pipeline_event_enqueue(dev, port, &ev);
+		}
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].sub_event_type % nb_stages;
+
+			if (cq_id == last_queue) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+				continue;
+			}
+
+			ev[i].sub_event_type++;
+			pipeline_fwd_event_vector(&ev[i],
+						  sched_type_list[cq_id]);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].sub_event_type % nb_stages;
+
+			if (cq_id == last_queue) {
+				vector_sz = ev[i].vec->nb_elem;
+				ev[i].queue_id = tx_queue[ev[i].vec->port];
+				ev[i].vec->queue = 0;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].sub_event_type++;
+				pipeline_fwd_event_vector(
+					&ev[i], sched_type_list[cq_id]);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
 static int
 worker_wrapper(void *arg)
 {
@@ -253,27 +502,36 @@ worker_wrapper(void *arg)
 	const bool burst = evt_has_burst_mode(w->dev_id);
 	const bool internal_port = w->t->internal_port;
 	const uint8_t nb_stages = opt->nb_stages;
-	RTE_SET_USED(opt);
+	/*vector/burst/internal_port*/
+	const pipeline_atq_worker_t
+	pipeline_atq_worker_single_stage[2][2][2] = {
+		[0][0][0] = pipeline_atq_worker_single_stage_fwd,
+		[0][0][1] = pipeline_atq_worker_single_stage_tx,
+		[0][1][0] = pipeline_atq_worker_single_stage_burst_fwd,
+		[0][1][1] = pipeline_atq_worker_single_stage_burst_tx,
+		[1][0][0] = pipeline_atq_worker_single_stage_fwd_vector,
+		[1][0][1] = pipeline_atq_worker_single_stage_tx_vector,
+		[1][1][0] = pipeline_atq_worker_single_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_atq_worker_single_stage_burst_tx_vector,
+	};
+	const pipeline_atq_worker_t
+	pipeline_atq_worker_multi_stage[2][2][2] = {
+		[0][0][0] = pipeline_atq_worker_multi_stage_fwd,
+		[0][0][1] = pipeline_atq_worker_multi_stage_tx,
+		[0][1][0] = pipeline_atq_worker_multi_stage_burst_fwd,
+		[0][1][1] = pipeline_atq_worker_multi_stage_burst_tx,
+		[1][0][0] = pipeline_atq_worker_multi_stage_fwd_vector,
+		[1][0][1] = pipeline_atq_worker_multi_stage_tx_vector,
+		[1][1][0] = pipeline_atq_worker_multi_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_atq_worker_multi_stage_burst_tx_vector,
+	};
 
-	if (nb_stages == 1) {
-		if (!burst && internal_port)
-			return pipeline_atq_worker_single_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_atq_worker_single_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_atq_worker_single_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_atq_worker_single_stage_burst_fwd(arg);
-	} else {
-		if (!burst && internal_port)
-			return pipeline_atq_worker_multi_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_atq_worker_multi_stage_fwd(arg);
-		if (burst && internal_port)
-			return pipeline_atq_worker_multi_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_atq_worker_multi_stage_burst_fwd(arg);
-	}
+	if (nb_stages == 1)
+		return (pipeline_atq_worker_single_stage[opt->ena_vector][burst]
+							[internal_port])(arg);
+	else
+		return (pipeline_atq_worker_multi_stage[opt->ena_vector][burst]
+						       [internal_port])(arg);
 
 	rte_panic("invalid worker\n");
 }
@@ -290,7 +548,7 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
 	int ret;
 	int nb_ports;
 	int nb_queues;
-	uint8_t queue;
+	uint8_t queue, is_prod;
 	uint8_t tx_evqueue_id[RTE_MAX_ETHPORTS];
 	uint8_t queue_arr[RTE_EVENT_MAX_QUEUES_PER_DEV];
 	uint8_t nb_worker_queues = 0;
@@ -330,15 +588,19 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
 		q_conf.event_queue_cfg = RTE_EVENT_QUEUE_CFG_ALL_TYPES;
 
 		if (!t->internal_port) {
+			is_prod = false;
 			RTE_ETH_FOREACH_DEV(prod) {
 				if (queue == tx_evqueue_id[prod]) {
 					q_conf.event_queue_cfg =
 						RTE_EVENT_QUEUE_CFG_SINGLE_LINK;
-				} else {
-					queue_arr[nb_worker_queues] = queue;
-					nb_worker_queues++;
+					is_prod = true;
+					break;
 				}
 			}
+			if (!is_prod) {
+				queue_arr[nb_worker_queues] = queue;
+				nb_worker_queues++;
+			}
 		}
 
 		ret = rte_event_queue_setup(opt->dev_id, queue, &q_conf);
diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index b47d76743..d5ef90500 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -36,6 +36,12 @@ pipeline_opt_dump(struct evt_options *opt, uint8_t nb_queues)
 	evt_dump_queue_priority(opt);
 	evt_dump_sched_type_list(opt);
 	evt_dump_producer_type(opt);
+	evt_dump("nb_eth_rx_queues", "%d", opt->eth_queues);
+	evt_dump("event_vector", "%d", opt->ena_vector);
+	if (opt->ena_vector) {
+		evt_dump("vector_size", "%d", opt->vector_size);
+		evt_dump("vector_tmo_ns", "%" PRIu64 "", opt->vector_tmo_nsec);
+	}
 }
 
 static inline uint64_t
@@ -163,7 +169,7 @@ pipeline_opt_check(struct evt_options *opt, uint64_t nb_queues)
 int
 pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 {
-	uint16_t i;
+	uint16_t i, j;
 	int ret;
 	uint8_t nb_queues = 1;
 	struct test_pipeline *t = evt_test_priv(test);
@@ -210,6 +216,16 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 		if (!(caps & RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT))
 			t->internal_port = 0;
 
+		ret = rte_event_eth_rx_adapter_caps_get(opt->dev_id, i, &caps);
+		if (ret != 0) {
+			evt_err("failed to get event tx adapter[%d] caps", i);
+			return ret;
+		}
+
+		if (!(caps & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT))
+			local_port_conf.rxmode.offloads |=
+				DEV_RX_OFFLOAD_RSS_HASH;
+
 		ret = rte_eth_dev_info_get(i, &dev_info);
 		if (ret != 0) {
 			evt_err("Error during getting device (port %u) info: %s\n",
@@ -236,19 +252,22 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 				local_port_conf.rx_adv_conf.rss_conf.rss_hf);
 		}
 
-		if (rte_eth_dev_configure(i, nb_queues, nb_queues,
-					&local_port_conf)
-				< 0) {
+		if (rte_eth_dev_configure(i, opt->eth_queues, nb_queues,
+					  &local_port_conf) < 0) {
 			evt_err("Failed to configure eth port [%d]", i);
 			return -EINVAL;
 		}
 
-		if (rte_eth_rx_queue_setup(i, 0, NB_RX_DESC,
-				rte_socket_id(), &rx_conf, t->pool) < 0) {
-			evt_err("Failed to setup eth port [%d] rx_queue: %d.",
+		for (j = 0; j < opt->eth_queues; j++) {
+			if (rte_eth_rx_queue_setup(i, j, NB_RX_DESC,
+						   rte_socket_id(), &rx_conf,
+						   t->pool) < 0) {
+				evt_err("Failed to setup eth port [%d] rx_queue: %d.",
 					i, 0);
-			return -EINVAL;
+				return -EINVAL;
+			}
 		}
+
 		if (rte_eth_tx_queue_setup(i, 0, NB_TX_DESC,
 					rte_socket_id(), NULL) < 0) {
 			evt_err("Failed to setup eth port [%d] tx_queue: %d.",
@@ -310,12 +329,27 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 {
 	int ret = 0;
 	uint16_t prod;
+	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
+	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;
 
 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
 	queue_conf.ev.sched_type = opt->sched_type_list[0];
+	if (opt->ena_vector) {
+		unsigned int nb_elem = (opt->pool_sz / opt->vector_size) << 1;
+
+		nb_elem = nb_elem ? nb_elem : 1;
+		vector_pool = rte_event_vector_pool_create(
+			"vector_pool", nb_elem, 0, opt->vector_size,
+			opt->socket_id);
+		if (vector_pool == NULL) {
+			evt_err("failed to create event vector pool");
+			return -ENOMEM;
+		}
+	}
 	RTE_ETH_FOREACH_DEV(prod) {
+		struct rte_event_eth_rx_adapter_vector_limits limits;
 		uint32_t cap;
 
 		ret = rte_event_eth_rx_adapter_caps_get(opt->dev_id,
@@ -326,6 +360,50 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 					opt->dev_id);
 			return ret;
 		}
+
+		if (opt->ena_vector) {
+			memset(&limits, 0, sizeof(limits));
+			ret = rte_event_eth_rx_adapter_vector_limits_get(
+				opt->dev_id, prod, &limits);
+			if (ret) {
+				evt_err("failed to get vector limits");
+				return ret;
+			}
+
+			if (opt->vector_size < limits.min_sz ||
+			    opt->vector_size > limits.max_sz) {
+				evt_err("Vector size [%d] not within limits max[%d] min[%d]",
+					opt->vector_size, limits.min_sz,
+					limits.max_sz);
+				return -EINVAL;
+			}
+
+			if (limits.log2_sz &&
+			    !rte_is_power_of_2(opt->vector_size)) {
+				evt_err("Vector size [%d] not power of 2",
+					opt->vector_size);
+				return -EINVAL;
+			}
+
+			if (opt->vector_tmo_nsec > limits.max_timeout_ns ||
+			    opt->vector_tmo_nsec < limits.min_timeout_ns) {
+				evt_err("Vector timeout [%" PRIu64
+					"] not within limits max[%" PRIu64
+					"] min[%" PRIu64 "]",
+					opt->vector_tmo_nsec,
+					limits.max_timeout_ns,
+					limits.min_timeout_ns);
+				return -EINVAL;
+			}
+
+			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.rx_queue_flags |=
+				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+			} else {
+				evt_err("Rx adapter doesn't support event vector");
+				return -EINVAL;
+			}
+		}
 		queue_conf.ev.queue_id = prod * stride;
 		ret = rte_event_eth_rx_adapter_create(prod, opt->dev_id,
 				&prod_conf);
@@ -340,6 +418,17 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}
 
+		if (opt->ena_vector) {
+			vec_conf.vector_sz = opt->vector_size;
+			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
+			vec_conf.vector_mp = vector_pool;
+			if (rte_event_eth_rx_adapter_queue_event_vector_config(
+				    prod, prod, -1, &vec_conf) < 0) {
+				evt_err("Failed to configure event vectorization for Rx adapter");
+				return -EINVAL;
+			}
+		}
+
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;
 
@@ -378,6 +467,14 @@ pipeline_event_tx_adapter_setup(struct evt_options *opt,
 			return ret;
 		}
 
+		if (opt->ena_vector) {
+			if (!(cap &
+			      RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR)) {
+				evt_err("Tx adapter doesn't support event vector");
+				return -EINVAL;
+			}
+		}
+
 		ret = rte_event_eth_tx_adapter_create(consm, opt->dev_id,
 				&port_conf);
 		if (ret) {
diff --git a/app/test-eventdev/test_pipeline_common.h b/app/test-eventdev/test_pipeline_common.h
index 6e73c6ab2..800a90616 100644
--- a/app/test-eventdev/test_pipeline_common.h
+++ b/app/test-eventdev/test_pipeline_common.h
@@ -101,6 +101,14 @@ pipeline_fwd_event(struct rte_event *ev, uint8_t sched)
 	ev->sched_type = sched;
 }
 
+static __rte_always_inline void
+pipeline_fwd_event_vector(struct rte_event *ev, uint8_t sched)
+{
+	ev->event_type = RTE_EVENT_TYPE_CPU_VECTOR;
+	ev->op = RTE_EVENT_OP_FORWARD;
+	ev->sched_type = sched;
+}
+
 static __rte_always_inline void
 pipeline_event_tx(const uint8_t dev, const uint8_t port,
 		struct rte_event * const ev)
@@ -110,6 +118,16 @@ pipeline_event_tx(const uint8_t dev, const uint8_t port,
 		rte_pause();
 }
 
+static __rte_always_inline void
+pipeline_event_tx_vector(const uint8_t dev, const uint8_t port,
+			 struct rte_event *const ev)
+{
+	ev->vec->queue = 0;
+
+	while (!rte_event_eth_tx_adapter_enqueue(dev, port, ev, 1, 0))
+		rte_pause();
+}
+
 static __rte_always_inline void
 pipeline_event_tx_burst(const uint8_t dev, const uint8_t port,
 		struct rte_event *ev, const uint16_t nb_rx)
diff --git a/app/test-eventdev/test_pipeline_queue.c b/app/test-eventdev/test_pipeline_queue.c
index 9a9febb19..f6cc3e358 100644
--- a/app/test-eventdev/test_pipeline_queue.c
+++ b/app/test-eventdev/test_pipeline_queue.c
@@ -15,6 +15,8 @@ pipeline_queue_nb_event_queues(struct evt_options *opt)
 	return (eth_count * opt->nb_stages) + eth_count;
 }
 
+typedef int (*pipeline_queue_worker_t)(void *arg);
+
 static __rte_noinline int
 pipeline_queue_worker_single_stage_tx(void *arg)
 {
@@ -126,6 +128,125 @@ pipeline_queue_worker_single_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_queue_worker_single_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		if (ev.sched_type == RTE_SCHED_TYPE_ATOMIC) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.queue_id++;
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			pipeline_event_enqueue(dev, port, &ev);
+		}
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		ev.queue_id = tx_queue[ev.vec->port];
+		ev.vec->queue = 0;
+		vector_sz = ev.vec->nb_elem;
+		pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			if (ev[i].sched_type == RTE_SCHED_TYPE_ATOMIC) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].queue_id++;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			ev[i].queue_id = tx_queue[ev[i].vec->port];
+			ev[i].vec->queue = 0;
+			vector_sz += ev[i].vec->nb_elem;
+			pipeline_fwd_event_vector(&ev[i],
+						  RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
 
 static __rte_noinline int
 pipeline_queue_worker_multi_stage_tx(void *arg)
@@ -267,6 +388,151 @@ pipeline_queue_worker_multi_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.queue_id % nb_stages;
+
+		if (ev.queue_id == tx_queue[ev.vec->port]) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+			continue;
+		}
+
+		ev.queue_id++;
+		pipeline_fwd_event_vector(&ev, cq_id != last_queue
+						       ? sched_type_list[cq_id]
+						       : RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.queue_id % nb_stages;
+
+		if (cq_id == last_queue) {
+			vector_sz = ev.vec->nb_elem;
+			ev.queue_id = tx_queue[ev.vec->port];
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.queue_id++;
+			pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+		}
+
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].queue_id % nb_stages;
+
+			if (ev[i].queue_id == tx_queue[ev[i].vec->port]) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+				continue;
+			}
+
+			ev[i].queue_id++;
+			pipeline_fwd_event_vector(
+				&ev[i], cq_id != last_queue
+						? sched_type_list[cq_id]
+						: RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].queue_id % nb_stages;
+
+			if (cq_id == last_queue) {
+				ev[i].queue_id = tx_queue[ev[i].vec->port];
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].queue_id++;
+				pipeline_fwd_event_vector(
+					&ev[i], sched_type_list[cq_id]);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
 static int
 worker_wrapper(void *arg)
 {
@@ -275,29 +541,39 @@ worker_wrapper(void *arg)
 	const bool burst = evt_has_burst_mode(w->dev_id);
 	const bool internal_port = w->t->internal_port;
 	const uint8_t nb_stages = opt->nb_stages;
-	RTE_SET_USED(opt);
-
-	if (nb_stages == 1) {
-		if (!burst && internal_port)
-			return pipeline_queue_worker_single_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_queue_worker_single_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_queue_worker_single_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_queue_worker_single_stage_burst_fwd(
-					arg);
-	} else {
-		if (!burst && internal_port)
-			return pipeline_queue_worker_multi_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_queue_worker_multi_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_queue_worker_multi_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_queue_worker_multi_stage_burst_fwd(arg);
+	/*vector/burst/internal_port*/
+	const pipeline_queue_worker_t
+	pipeline_queue_worker_single_stage[2][2][2] = {
+		[0][0][0] = pipeline_queue_worker_single_stage_fwd,
+		[0][0][1] = pipeline_queue_worker_single_stage_tx,
+		[0][1][0] = pipeline_queue_worker_single_stage_burst_fwd,
+		[0][1][1] = pipeline_queue_worker_single_stage_burst_tx,
+		[1][0][0] = pipeline_queue_worker_single_stage_fwd_vector,
+		[1][0][1] = pipeline_queue_worker_single_stage_tx_vector,
+		[1][1][0] = pipeline_queue_worker_single_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_queue_worker_single_stage_burst_tx_vector,
+	};
+	const pipeline_queue_worker_t
+	pipeline_queue_worker_multi_stage[2][2][2] = {
+		[0][0][0] = pipeline_queue_worker_multi_stage_fwd,
+		[0][0][1] = pipeline_queue_worker_multi_stage_tx,
+		[0][1][0] = pipeline_queue_worker_multi_stage_burst_fwd,
+		[0][1][1] = pipeline_queue_worker_multi_stage_burst_tx,
+		[1][0][0] = pipeline_queue_worker_multi_stage_fwd_vector,
+		[1][0][1] = pipeline_queue_worker_multi_stage_tx_vector,
+		[1][1][0] = pipeline_queue_worker_multi_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_queue_worker_multi_stage_burst_tx_vector,
+	};
+
+	if (nb_stages == 1)
+		return (pipeline_queue_worker_single_stage[opt->ena_vector]
+							  [burst]
+							  [internal_port])(arg);
+	else
+		return (pipeline_queue_worker_multi_stage[opt->ena_vector]
+							 [burst]
+							 [internal_port])(arg);
 
-	}
 	rte_panic("invalid worker\n");
 }
 
diff --git a/doc/guides/tools/testeventdev.rst b/doc/guides/tools/testeventdev.rst
index ad1788a3d..f252dc2c1 100644
--- a/doc/guides/tools/testeventdev.rst
+++ b/doc/guides/tools/testeventdev.rst
@@ -158,6 +158,26 @@ The following are the application command-line options:
        Set max packet mbuf size. Can be used configure Rx/Tx scatter gather.
        Only applicable for `pipeline_atq` and `pipeline_queue` tests.
 
+* ``--nb_eth_queues``
+
+       Configure multiple Rx queues per each ethernet port.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--enable_vector``
+
+       Enable event vector for Rx/Tx adapters.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--vector_size``
+
+       Vector size to configure for the Rx adapter.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--vector_tmo_ns``
+
+       Vector timeout nanoseconds to be configured for the Rx adapter.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
 
 Eventdev Tests
 --------------
@@ -607,6 +627,10 @@ Supported application command line options are following::
         --worker_deq_depth
         --prod_type_ethdev
         --deq_tmo_nsec
+        --nb_eth_queues
+        --enable_vector
+        --vector_size
+        --vector_tmo_ns
 
 
 .. Note::
@@ -623,6 +647,13 @@ Example command to run pipeline queue test:
     sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
         --test=pipeline_queue --wlcore=1 --prod_type_ethdev --stlist=a
 
+Example command to run pipeline atq test with vector events:
+
+.. code-block:: console
+
+    sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
+        --test=pipeline_queue --wlcore=1 --prod_type_ethdev --stlist=a \
+        --enable_vector  --vector_size 512
 
 PIPELINE_ATQ Test
 ~~~~~~~~~~~~~~~~~~~
@@ -699,6 +730,10 @@ Supported application command line options are following::
         --worker_deq_depth
         --prod_type_ethdev
         --deq_tmo_nsec
+        --nb_eth_queues
+        --enable_vector
+        --vector_size
+        --vector_tmo_ns
 
 
 .. Note::
@@ -708,9 +743,17 @@ Supported application command line options are following::
 Example
 ^^^^^^^
 
-Example command to run pipeline queue test:
+Example command to run pipeline atq test:
 
 .. code-block:: console
 
     sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
         --test=pipeline_atq --wlcore=1 --prod_type_ethdev --stlist=a
+
+Example command to run pipeline atq test with vector events:
+
+.. code-block:: console
+
+    sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
+        --test=pipeline_atq --wlcore=1 --prod_type_ethdev --stlist=a \
+        --enable_vector  --vector_size 512
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v7 7/8] doc: announce event Rx adapter config changes
  2021-03-25 17:10           ` [dpdk-dev] [PATCH v7 0/8] Introduce event vectorization pbhagavatula
                               ` (5 preceding siblings ...)
  2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
@ 2021-03-25 17:10             ` pbhagavatula
  2021-03-25 17:10             ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v7 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
  2021-03-26 14:08             ` [dpdk-dev] [PATCH v8 0/8] Introduce event vectorization pbhagavatula
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-25 17:10 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

The Rx adapter event vector configuration will be merged into
Rx adapter queue configuration to simplify enabling event
vectorization.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Ray Kinsella <mdr@ashroe.eu>
---
 doc/guides/rel_notes/deprecation.rst | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst
index 64629e064..ef6d1bb18 100644
--- a/doc/guides/rel_notes/deprecation.rst
+++ b/doc/guides/rel_notes/deprecation.rst
@@ -130,3 +130,12 @@ Deprecation Notices
 * cmdline: ``cmdline`` structure will be made opaque to hide platform-specific
   content. On Linux and FreeBSD, supported prior to DPDK 20.11,
   original structure will be kept until DPDK 21.11.
+
+* eventdev: The structure ``rte_event_eth_rx_adapter_queue_conf`` will be
+  extended to include ``rte_event_eth_rx_adapter_event_vector_config`` elements
+  and the function ``rte_event_eth_rx_adapter_queue_event_vector_config`` will
+  be removed in DPDK 21.11.
+
+  An application can enable event vectorization by passing the desired vector
+  values to the function ``rte_event_eth_rx_adapter_queue_add`` using
+  the structure ``rte_event_eth_rx_adapter_queue_add``.
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [dpdk-dev v21.11] [PATCH v7 8/8] eventdev: simplify Rx adapter event vector config
  2021-03-25 17:10           ` [dpdk-dev] [PATCH v7 0/8] Introduce event vectorization pbhagavatula
                               ` (6 preceding siblings ...)
  2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 7/8] doc: announce event Rx adapter config changes pbhagavatula
@ 2021-03-25 17:10             ` pbhagavatula
  2021-03-26 14:08             ` [dpdk-dev] [PATCH v8 0/8] Introduce event vectorization pbhagavatula
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-25 17:10 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Include vector configuration into the structure
``rte_event_eth_rx_adapter_queue_conf`` used when configuring rest
of the Rx adapter ethernet device Rx queue parameters.
This simplifies event vector configuration as it avoids splitting
configuration per Rx queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 app/test-eventdev/test_pipeline_common.c      |  16 +-
 lib/librte_eventdev/eventdev_pmd.h            |  29 ---
 .../rte_event_eth_rx_adapter.c                | 168 ++++++------------
 .../rte_event_eth_rx_adapter.h                |  27 ---
 lib/librte_eventdev/version.map               |   1 -
 5 files changed, 57 insertions(+), 184 deletions(-)

diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index d5ef90500..76aee254b 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -331,7 +331,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 	uint16_t prod;
 	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
-	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
@@ -397,8 +396,12 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			}

 			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.vector_sz = opt->vector_size;
+				queue_conf.vector_timeout_ns =
+					opt->vector_tmo_nsec;
 				queue_conf.rx_queue_flags |=
 				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+				queue_conf.vector_mp = vector_pool;
 			} else {
 				evt_err("Rx adapter doesn't support event vector");
 				return -EINVAL;
@@ -418,17 +421,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}

-		if (opt->ena_vector) {
-			vec_conf.vector_sz = opt->vector_size;
-			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
-			vec_conf.vector_mp = vector_pool;
-			if (rte_event_eth_rx_adapter_queue_event_vector_config(
-				    prod, prod, -1, &vec_conf) < 0) {
-				evt_err("Failed to configure event vectorization for Rx adapter");
-				return -EINVAL;
-			}
-		}
-
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;

diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 0f724ac85..63b3bc4b5 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -667,32 +667,6 @@ typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
 	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-struct rte_event_eth_rx_adapter_event_vector_config;
-/**
- * Enable event vector on an given Rx queue of a ethernet devices belonging to
- * the Rx adapter.
- *
- * @param dev
- *   Event device pointer
- *
- * @param eth_dev
- *   Ethernet device pointer
- *
- * @param rx_queue_id
- *   The Rx queue identifier
- *
- * @param config
- *   Pointer to the event vector configuration structure.
- *
- * @return
- *   - 0: Success.
- *   - <0: Error code returned by the driver function.
- */
-typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
-	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
-	int32_t rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;

@@ -1118,9 +1092,6 @@ struct rte_eventdev_ops {
 	eventdev_eth_rx_adapter_vector_limits_get_t
 		eth_rx_adapter_vector_limits_get;
 	/**< Get event vector limits for the Rx adapter */
-	eventdev_eth_rx_adapter_event_vector_config_t
-		eth_rx_adapter_event_vector_config;
-	/**< Configure Rx adapter with event vector */

 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index a3227df92..e18e59c74 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -1888,6 +1888,25 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	} else
 		qi_ev->flow_id = 0;

+	if (conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+		queue_info->ena_vector = 1;
+		qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+		rxa_set_vector_data(queue_info, conf->vector_sz,
+				    conf->vector_timeout_ns, conf->vector_mp,
+				    rx_queue_id, dev_info->dev->data->port_id);
+		rx_adapter->ena_vector = 1;
+		rx_adapter->vector_tmo_ticks =
+			rx_adapter->vector_tmo_ticks
+				? RTE_MIN(queue_info->vector_data
+						  .vector_timeout_ticks,
+					  rx_adapter->vector_tmo_ticks)
+				: queue_info->vector_data.vector_timeout_ticks;
+		rx_adapter->vector_tmo_ticks <<= 1;
+		TAILQ_INIT(&rx_adapter->vector_list);
+		rx_adapter->prev_expiry_ts = 0;
+	}
+
 	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
 	if (rxa_polled_queue(dev_info, rx_queue_id)) {
 		rx_adapter->num_rx_polled += !pollq;
@@ -1913,44 +1932,6 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }

-static void
-rxa_sw_event_vector_configure(
-	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
-	int rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
-	struct eth_rx_queue_info *queue_info;
-	struct rte_event *qi_ev;
-
-	if (rx_queue_id == -1) {
-		uint16_t nb_rx_queues;
-		uint16_t i;
-
-		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
-		for (i = 0; i < nb_rx_queues; i++)
-			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
-						      config);
-		return;
-	}
-
-	queue_info = &dev_info->rx_queue[rx_queue_id];
-	qi_ev = (struct rte_event *)&queue_info->event;
-	queue_info->ena_vector = 1;
-	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
-	rxa_set_vector_data(queue_info, config->vector_sz,
-			    config->vector_timeout_ns, config->vector_mp,
-			    rx_queue_id, dev_info->dev->data->port_id);
-	rx_adapter->ena_vector = 1;
-	rx_adapter->vector_tmo_ticks =
-		rx_adapter->vector_tmo_ticks ?
-			      RTE_MIN(config->vector_timeout_ns << 1,
-				      rx_adapter->vector_tmo_ticks) :
-			      config->vector_timeout_ns << 1;
-	rx_adapter->prev_expiry_ts = 0;
-	TAILQ_INIT(&rx_adapter->vector_list);
-}
-
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -2264,6 +2245,7 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	struct rte_event_eth_rx_adapter *rx_adapter;
 	struct rte_eventdev *dev;
 	struct eth_device_info *dev_info;
+	struct rte_event_eth_rx_adapter_vector_limits limits;

 	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
 	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
@@ -2300,6 +2282,39 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}

+	if (queue_conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+		ret = rte_event_eth_rx_adapter_vector_limits_get(
+			rx_adapter->eventdev_id, eth_dev_id, &limits);
+		if (ret < 0) {
+			RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_sz < limits.min_sz ||
+		    queue_conf->vector_sz > limits.max_sz ||
+		    queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
+		    queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
+		    queue_conf->vector_mp == NULL) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_mp->elt_size <
+		    (sizeof(struct rte_event_vector) +
+		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+	}
+
 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
 		(rx_queue_id != -1)) {
 		RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
@@ -2493,83 +2508,6 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }

-int
-rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct rte_event_eth_rx_adapter_vector_limits limits;
-	struct rte_event_eth_rx_adapter *rx_adapter;
-	struct rte_eventdev *dev;
-	uint32_t cap;
-	int ret;
-
-	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
-	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
-
-	rx_adapter = rxa_id_to_adapter(id);
-	if ((rx_adapter == NULL) || (config == NULL))
-		return -EINVAL;
-
-	dev = &rte_eventdevs[rx_adapter->eventdev_id];
-	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
-						eth_dev_id, &cap);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
-				 "eth port %" PRIu16,
-				 id, eth_dev_id);
-		return ret;
-	}
-
-	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
-		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	ret = rte_event_eth_rx_adapter_vector_limits_get(
-		rx_adapter->eventdev_id, eth_dev_id, &limits);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
-				 "eth port %" PRIu16,
-				 rx_adapter->eventdev_id, eth_dev_id);
-		return ret;
-	}
-
-	if (config->vector_sz < limits.min_sz ||
-	    config->vector_sz > limits.max_sz ||
-	    config->vector_timeout_ns < limits.min_timeout_ns ||
-	    config->vector_timeout_ns > limits.max_timeout_ns ||
-	    config->vector_mp == NULL) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-	if (config->vector_mp->elt_size <
-	    (sizeof(struct rte_event_vector) +
-	     (sizeof(uintptr_t) * config->vector_sz))) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
-		RTE_FUNC_PTR_OR_ERR_RET(
-			*dev->dev_ops->eth_rx_adapter_event_vector_config,
-			-ENOTSUP);
-		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
-			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
-	} else {
-		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
-					      rx_queue_id, config);
-	}
-
-	return ret;
-}
-
 int
 rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
index 7407cde00..3f8b36229 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
@@ -171,9 +171,6 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
 	 * enqueued event.
 	 */
-};
-
-struct rte_event_eth_rx_adapter_event_vector_config {
 	uint16_t vector_sz;
 	/**<
 	 * Indicates the maximum number for mbufs to combine and form a vector.
@@ -548,30 +545,6 @@ int rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-/**
- * Configure event vectorization for a given ethernet device queue, that has
- * been added to a event eth Rx adapter.
- *
- * @param id
- *  The identifier of the ethernet Rx event adapter.
- *
- * @param eth_dev_id
- *  The identifier of the ethernet device.
- *
- * @param rx_queue_id
- *  Ethernet device receive queue index.
- *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
- *  are configured with event vectorization.
- *
- * @return
- *  - 0: Success, Receive queue configured correctly.
- *  - <0: Error code on failure.
- */
-__rte_experimental
-int rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index 902df0ae3..34c1c830e 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -142,7 +142,6 @@ EXPERIMENTAL {
 	#added in 21.05
 	rte_event_vector_pool_create;
 	rte_event_eth_rx_adapter_vector_limits_get;
-	rte_event_eth_rx_adapter_queue_event_vector_config;
 };

 INTERNAL {
--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v5 4/8] eventdev: add Rx adapter event vector support
  2021-03-25 13:14             ` Pavan Nikhilesh Bhagavatula
@ 2021-03-26  6:26               ` Jayatheerthan, Jay
  2021-03-26  9:00                 ` Pavan Nikhilesh Bhagavatula
  0 siblings, 1 reply; 153+ messages in thread
From: Jayatheerthan, Jay @ 2021-03-26  6:26 UTC (permalink / raw)
  To: Pavan Nikhilesh Bhagavatula, Jerin Jacob Kollanukkaran, Carrillo,
	Erik G, Gujjar, Abhinandan S, McDaniel, Timothy, hemant.agrawal,
	Van Haaren, Harry, mattias.ronnblom, Ma, Liang J
  Cc: dev

> -----Original Message-----
> From: Pavan Nikhilesh Bhagavatula <pbhagavatula@marvell.com>
> Sent: Thursday, March 25, 2021 6:44 PM
> To: Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Jerin Jacob Kollanukkaran <jerinj@marvell.com>; Carrillo, Erik G
> <erik.g.carrillo@intel.com>; Gujjar, Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy
> <timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
> <mattias.ronnblom@ericsson.com>; Ma, Liang J <liang.j.ma@intel.com>
> Cc: dev@dpdk.org
> Subject: RE: [dpdk-dev] [PATCH v5 4/8] eventdev: add Rx adapter event vector support
> 
> 
> 
> >-----Original Message-----
> >From: Jayatheerthan, Jay <jay.jayatheerthan@intel.com>
> >Sent: Thursday, March 25, 2021 4:07 PM
> >To: Pavan Nikhilesh Bhagavatula <pbhagavatula@marvell.com>; Jerin
> >Jacob Kollanukkaran <jerinj@marvell.com>; Carrillo, Erik G
> ><erik.g.carrillo@intel.com>; Gujjar, Abhinandan S
> ><abhinandan.gujjar@intel.com>; McDaniel, Timothy
> ><timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
> >Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
> ><mattias.ronnblom@ericsson.com>; Ma, Liang J
> ><liang.j.ma@intel.com>
> >Cc: dev@dpdk.org
> >Subject: [EXT] RE: [dpdk-dev] [PATCH v5 4/8] eventdev: add Rx adapter
> >event vector support
> >
> >External Email
> >
> >----------------------------------------------------------------------
> >> -----Original Message-----
> >> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
> >> Sent: Wednesday, March 24, 2021 10:35 AM
> >> To: jerinj@marvell.com; Jayatheerthan, Jay
> ><jay.jayatheerthan@intel.com>; Carrillo, Erik G
> ><erik.g.carrillo@intel.com>; Gujjar,
> >> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy
> ><timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
> >> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
> ><mattias.ronnblom@ericsson.com>; Ma, Liang J
> >> <liang.j.ma@intel.com>
> >> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
> >> Subject: [dpdk-dev] [PATCH v5 4/8] eventdev: add Rx adapter event
> >vector support
> >>
> >> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> >>
> >> Add event vector support for event eth Rx adapter, the
> >implementation
> >> creates vector flows based on port and queue identifier of the
> >received
> >> mbufs.
> >>
> >> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> >> ---
> >>  lib/librte_eventdev/eventdev_pmd.h            |   7 +-
> >>  .../rte_event_eth_rx_adapter.c                | 257 ++++++++++++++++--
> >>  lib/librte_eventdev/rte_eventdev.c            |   6 +-
> >>  3 files changed, 250 insertions(+), 20 deletions(-)
> >>
> >> diff --git a/lib/librte_eventdev/eventdev_pmd.h
> >b/lib/librte_eventdev/eventdev_pmd.h
> >> index 9297f1433..0f724ac85 100644
> >> --- a/lib/librte_eventdev/eventdev_pmd.h
> >> +++ b/lib/librte_eventdev/eventdev_pmd.h
> >> @@ -69,9 +69,10 @@ extern "C" {
> >>  	} \
> >>  } while (0)
> >>
> >> -#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP \
> >> -
> >	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |
> >\
> >> -
> >	(RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ))
> >> +#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP
> >\
> >> +	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |
> >\
> >> +	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) |
> >\
> >> +	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR))
> >>
> >>  #define RTE_EVENT_CRYPTO_ADAPTER_SW_CAP \
> >>
> >	RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA
> >> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> >b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> >> index ac8ba5bf0..c71990078 100644
> >> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> >> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> >> @@ -26,6 +26,10 @@
> >>  #define BATCH_SIZE		32
> >>  #define BLOCK_CNT_THRESHOLD	10
> >>  #define ETH_EVENT_BUFFER_SIZE	(4*BATCH_SIZE)
> >> +#define MAX_VECTOR_SIZE		1024
> >> +#define MIN_VECTOR_SIZE		4
> >> +#define MAX_VECTOR_NS		1E9
> >> +#define MIN_VECTOR_NS		1E5
> >>
> >>  #define ETH_RX_ADAPTER_SERVICE_NAME_LEN	32
> >>  #define ETH_RX_ADAPTER_MEM_NAME_LEN	32
> >> @@ -59,6 +63,20 @@ struct eth_rx_poll_entry {
> >>  	uint16_t eth_rx_qid;
> >>  };
> >>
> >> +struct eth_rx_vector_data {
> >> +	TAILQ_ENTRY(eth_rx_vector_data) next;
> >> +	uint16_t port;
> >> +	uint16_t queue;
> >> +	uint16_t max_vector_count;
> >> +	uint64_t event;
> >> +	uint64_t ts;
> >> +	uint64_t vector_timeout_ticks;
> >> +	struct rte_mempool *vector_pool;
> >> +	struct rte_event_vector *vector_ev;
> >> +} __rte_cache_aligned;
> >> +
> >> +TAILQ_HEAD(eth_rx_vector_data_list, eth_rx_vector_data);
> >> +
> >>  /* Instance per adapter */
> >>  struct rte_eth_event_enqueue_buffer {
> >>  	/* Count of events in this buffer */
> >> @@ -92,6 +110,14 @@ struct rte_event_eth_rx_adapter {
> >>  	uint32_t wrr_pos;
> >>  	/* Event burst buffer */
> >>  	struct rte_eth_event_enqueue_buffer event_enqueue_buffer;
> >> +	/* Vector enable flag */
> >> +	uint8_t ena_vector;
> >> +	/* Timestamp of previous vector expiry list traversal */
> >> +	uint64_t prev_expiry_ts;
> >> +	/* Minimum ticks to wait before traversing expiry list */
> >> +	uint64_t vector_tmo_ticks;
> >> +	/* vector list */
> >> +	struct eth_rx_vector_data_list vector_list;
> >>  	/* Per adapter stats */
> >>  	struct rte_event_eth_rx_adapter_stats stats;
> >>  	/* Block count, counts up to BLOCK_CNT_THRESHOLD */
> >> @@ -198,9 +224,11 @@ struct eth_device_info {
> >>  struct eth_rx_queue_info {
> >>  	int queue_enabled;	/* True if added */
> >>  	int intr_enabled;
> >> +	uint8_t ena_vector;
> >>  	uint16_t wt;		/* Polling weight */
> >>  	uint32_t flow_id_mask;	/* Set to ~0 if app provides flow id else
> >0 */
> >>  	uint64_t event;
> >> +	struct eth_rx_vector_data vector_data;
> >>  };
> >>
> >>  static struct rte_event_eth_rx_adapter **event_eth_rx_adapter;
> >> @@ -722,6 +750,9 @@ rxa_flush_event_buffer(struct
> >rte_event_eth_rx_adapter *rx_adapter)
> >>  	    &rx_adapter->event_enqueue_buffer;
> >>  	struct rte_event_eth_rx_adapter_stats *stats = &rx_adapter-
> >>stats;
> >>
> >> +	if (!buf->count)
> >> +		return 0;
> >> +
> >>  	uint16_t n = rte_event_enqueue_new_burst(rx_adapter-
> >>eventdev_id,
> >>  					rx_adapter->event_port_id,
> >>  					buf->events,
> >> @@ -742,6 +773,72 @@ rxa_flush_event_buffer(struct
> >rte_event_eth_rx_adapter *rx_adapter)
> >>  	return n;
> >>  }
> >>
> >> +static inline uint16_t
> >> +rxa_create_event_vector(struct rte_event_eth_rx_adapter
> >*rx_adapter,
> >> +			struct eth_rx_queue_info *queue_info,
> >> +			struct rte_eth_event_enqueue_buffer *buf,
> >> +			struct rte_mbuf **mbufs, uint16_t num)
> >> +{
> >> +	struct rte_event *ev = &buf->events[buf->count];
> >> +	struct eth_rx_vector_data *vec;
> >> +	uint16_t filled, space, sz;
> >> +
> >> +	filled = 0;
> >> +	vec = &queue_info->vector_data;
> >> +	while (num) {
> >> +		if (vec->vector_ev == NULL) {
> >> +			if (rte_mempool_get(vec->vector_pool,
> >> +					    (void **)&vec->vector_ev) <
> >0) {
> >> +				rte_pktmbuf_free_bulk(mbufs, num);
> >> +				return 0;
> >> +			}
> >> +			vec->vector_ev->nb_elem = 0;
> >> +			vec->vector_ev->port = vec->port;
> >> +			vec->vector_ev->queue = vec->queue;
> >> +			vec->vector_ev->attr_valid = true;
> >> +			TAILQ_INSERT_TAIL(&rx_adapter->vector_list,
> >vec, next);
> >> +		} else if (vec->vector_ev->nb_elem == vec-
> >>max_vector_count) {
> >
> >Is there a case where nb_elem > max_vector_count as we accumulate
> >sz to it ?
> 
> I don't think so, that would overflow the vector event.
> 
> >
> >> +			/* Event ready. */
> >> +			ev->event = vec->event;
> >> +			ev->vec = vec->vector_ev;
> >> +			ev++;
> >> +			filled++;
> >> +			vec->vector_ev = NULL;
> >> +			TAILQ_REMOVE(&rx_adapter->vector_list, vec,
> >next);
> >> +			if (rte_mempool_get(vec->vector_pool,
> >> +					    (void **)&vec->vector_ev) <
> >0) {
> >> +				rte_pktmbuf_free_bulk(mbufs, num);
> >> +				return 0;
> >> +			}
> >> +			vec->vector_ev->nb_elem = 0;
> >> +			vec->vector_ev->port = vec->port;
> >> +			vec->vector_ev->queue = vec->queue;
> >> +			vec->vector_ev->attr_valid = true;
> >> +			TAILQ_INSERT_TAIL(&rx_adapter->vector_list,
> >vec, next);
> >> +		}
> >> +
> >> +		space = vec->max_vector_count - vec->vector_ev-
> >>nb_elem;
> >> +		sz = num > space ? space : num;
> >> +		memcpy(vec->vector_ev->mbufs + vec->vector_ev-
> >>nb_elem, mbufs,
> >> +		       sizeof(void *) * sz);
> >> +		vec->vector_ev->nb_elem += sz;
> >> +		num -= sz;
> >> +		mbufs += sz;
> >> +		vec->ts = rte_rdtsc();
> >> +	}
> >> +
> >> +	if (vec->vector_ev->nb_elem == vec->max_vector_count) {
> >
> >Same here.
> >
> >> +		ev->event = vec->event;
> >> +		ev->vec = vec->vector_ev;
> >> +		ev++;
> >> +		filled++;
> >> +		vec->vector_ev = NULL;
> >> +		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
> >> +	}
> >> +
> >> +	return filled;
> >> +}
> >
> >I am seeing more than one repeating code chunks in this function.
> >Perhaps, you can give it a try to not repeat. We can drop if its
> >performance affecting.
> 
> I will try to move them to inline functions and test.
> 
> >
> >> +
> >>  static inline void
> >>  rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
> >>  		uint16_t eth_dev_id,
> >> @@ -770,25 +867,30 @@ rxa_buffer_mbufs(struct
> >rte_event_eth_rx_adapter *rx_adapter,
> >>  	rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
> >>  	do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
> >
> >The RSS related code is executed for vector case as well. Can this be
> >moved inside ena_vector if condition ?
> 
> RSS is used to generate the event flowid, in vector case the flow id
> Will be a combination of port and queue id.
> The idea is that flows having the same RSS LSB will end up in the same
> queue.
> 

I meant to say, rss_mask and do_rss are used only when ena_vector is false. Could be moved inside the appropriate condition ?

> >
> >>
> >> -	for (i = 0; i < num; i++) {
> >> -		m = mbufs[i];
> >> -
> >> -		rss = do_rss ?
> >> -			rxa_do_softrss(m, rx_adapter->rss_key_be) :
> >> -			m->hash.rss;
> >> -		ev->event = event;
> >> -		ev->flow_id = (rss & ~flow_id_mask) |
> >> -				(ev->flow_id & flow_id_mask);
> >> -		ev->mbuf = m;
> >> -		ev++;
> >> +	if (!eth_rx_queue_info->ena_vector) {
> >> +		for (i = 0; i < num; i++) {
> >> +			m = mbufs[i];
> >> +
> >> +			rss = do_rss ? rxa_do_softrss(m, rx_adapter-
> >>rss_key_be)
> >> +				     : m->hash.rss;
> >> +			ev->event = event;
> >> +			ev->flow_id = (rss & ~flow_id_mask) |
> >> +				      (ev->flow_id & flow_id_mask);
> >> +			ev->mbuf = m;
> >> +			ev++;
> >> +		}
> >> +	} else {
> >> +		num = rxa_create_event_vector(rx_adapter,
> >eth_rx_queue_info,
> >> +					      buf, mbufs, num);
> >>  	}
> >>
> >> -	if (dev_info->cb_fn) {
> >> +	if (num && dev_info->cb_fn) {
> >>
> >>  		dropped = 0;
> >>  		nb_cb = dev_info->cb_fn(eth_dev_id, rx_queue_id,
> >> -					ETH_EVENT_BUFFER_SIZE, buf-
> >>count, ev,
> >> -					num, dev_info->cb_arg,
> >&dropped);
> >> +					ETH_EVENT_BUFFER_SIZE, buf-
> >>count,
> >> +					&buf->events[buf->count],
> >num,
> >> +					dev_info->cb_arg, &dropped);
> >
> >Before this patch, we pass ev which is &buf->events[buf->count] + num
> >as fifth param when calling cb_fn. Now, we are passing &buf-
> >>events[buf->count] for non-vector case. Do you see this as an issue?
> >
> 
> The callback function takes in the array newly formed events i.e. we need
> to pass the start of array and the count.
> 
> the previous code had a bug where it passes the end of the event list.

ok, that makes sense.

> 
> >Also, for vector case would it make sense to do pass &buf->events[buf-
> >>count] + num ?
> >
> >>  		if (unlikely(nb_cb > num))
> >>  			RTE_EDEV_LOG_ERR("Rx CB returned %d (> %d)
> >events",
> >>  				nb_cb, num);
> >> @@ -1124,6 +1226,30 @@ rxa_poll(struct rte_event_eth_rx_adapter
> >*rx_adapter)
> >>  	return nb_rx;
> >>  }
> >>
> >> +static void
> >> +rxa_vector_expire(struct eth_rx_vector_data *vec, void *arg)
> >> +{
> >> +	struct rte_event_eth_rx_adapter *rx_adapter = arg;
> >> +	struct rte_eth_event_enqueue_buffer *buf =
> >> +		&rx_adapter->event_enqueue_buffer;
> >> +	struct rte_event *ev;
> >> +
> >> +	if (buf->count)
> >> +		rxa_flush_event_buffer(rx_adapter);
> >> +
> >> +	if (vec->vector_ev->nb_elem == 0)
> >> +		return;
> >> +	ev = &buf->events[buf->count];
> >> +
> >> +	/* Event ready. */
> >> +	ev->event = vec->event;
> >> +	ev->vec = vec->vector_ev;
> >> +	buf->count++;
> >> +
> >> +	vec->vector_ev = NULL;
> >> +	vec->ts = 0;
> >> +}
> >> +
> >>  static int
> >>  rxa_service_func(void *args)
> >>  {
> >> @@ -1137,6 +1263,24 @@ rxa_service_func(void *args)
> >>  		return 0;
> >>  	}
> >>
> >> +	if (rx_adapter->ena_vector) {
> >> +		if ((rte_rdtsc() - rx_adapter->prev_expiry_ts) >=
> >> +		    rx_adapter->vector_tmo_ticks) {
> >> +			struct eth_rx_vector_data *vec;
> >> +
> >> +			TAILQ_FOREACH(vec, &rx_adapter->vector_list,
> >next) {
> >> +				uint64_t elapsed_time = rte_rdtsc() -
> >vec->ts;
> >> +
> >> +				if (elapsed_time >= vec-
> >>vector_timeout_ticks) {
> >> +					rxa_vector_expire(vec,
> >rx_adapter);
> >> +					TAILQ_REMOVE(&rx_adapter-
> >>vector_list,
> >> +						     vec, next);
> >> +				}
> >> +			}
> >> +			rx_adapter->prev_expiry_ts = rte_rdtsc();
> >> +		}
> >> +	}
> >> +
> >>  	stats = &rx_adapter->stats;
> >>  	stats->rx_packets += rxa_intr_ring_dequeue(rx_adapter);
> >>  	stats->rx_packets += rxa_poll(rx_adapter);
> >> @@ -1640,6 +1784,28 @@ rxa_update_queue(struct
> >rte_event_eth_rx_adapter *rx_adapter,
> >>  	}
> >>  }
> >>
> >> +static void
> >> +rxa_set_vector_data(struct eth_rx_queue_info *queue_info,
> >uint16_t vector_count,
> >> +		    uint64_t vector_ns, struct rte_mempool *mp, int32_t
> >qid,
> >> +		    uint16_t port_id)
> >> +{
> >> +#define NSEC2TICK(__ns, __freq) (((__ns) * (__freq)) / 1E9)
> >> +	struct eth_rx_vector_data *vector_data;
> >> +	uint32_t flow_id;
> >> +
> >> +	vector_data = &queue_info->vector_data;
> >> +	vector_data->max_vector_count = vector_count;
> >> +	vector_data->port = port_id;
> >> +	vector_data->queue = qid;
> >> +	vector_data->vector_pool = mp;
> >> +	vector_data->vector_timeout_ticks =
> >> +		NSEC2TICK(vector_ns, rte_get_timer_hz());
> >> +	vector_data->ts = 0;
> >> +	flow_id = queue_info->event & 0xFFFFF;
> >> +	flow_id = flow_id == 0 ? (qid & 0xFF) | (port_id & 0xFFFF) :
> >flow_id;
> >
> >Maybe I am missing something here. Looking at the code it looks like
> >qid and port_id may overlap. For e.g., if qid = 0x10 and port_id = 0x11,
> >flow_id would end up being 0x11. Is this the expectation? Also, it may
> >be useful to document flow_id format.
> 
> The flow_id is 20 bit, I guess we could do 12bit queue_id and 8bit port
> as a flow.

This sounds reasonable to me. It would be useful to have the flow_id format and how it is used for vectorization in Rx/Tx adapter documentation.

> 
> >Comparing this format with existing RSS hash based method, are we
> >saying that all mbufs received in a rx burst are part of same flow when
> >vectorization is used?
> 
> Yes, the hard way to do this is to use a hash table and treating each
> mbuf having an unique flow.
> 
> >
> >> +	vector_data->event = (queue_info->event & ~0xFFFFF) |
> >flow_id;
> >> +}
> >> +
> >>  static void
> >>  rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
> >>  	struct eth_device_info *dev_info,
> >> @@ -1741,6 +1907,44 @@ rxa_add_queue(struct
> >rte_event_eth_rx_adapter *rx_adapter,
> >>  	}
> >>  }
> >>
> >> +static void
> >> +rxa_sw_event_vector_configure(
> >> +	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t
> >eth_dev_id,
> >> +	int rx_queue_id,
> >> +	const struct rte_event_eth_rx_adapter_event_vector_config
> >*config)
> >> +{
> >> +	struct eth_device_info *dev_info = &rx_adapter-
> >>eth_devices[eth_dev_id];
> >> +	struct eth_rx_queue_info *queue_info;
> >> +	struct rte_event *qi_ev;
> >> +
> >> +	if (rx_queue_id == -1) {
> >> +		uint16_t nb_rx_queues;
> >> +		uint16_t i;
> >> +
> >> +		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
> >> +		for (i = 0; i < nb_rx_queues; i++)
> >> +			rxa_sw_event_vector_configure(rx_adapter,
> >eth_dev_id, i,
> >> +						      config);
> >> +		return;
> >> +	}
> >> +
> >> +	queue_info = &dev_info->rx_queue[rx_queue_id];
> >> +	qi_ev = (struct rte_event *)&queue_info->event;
> >> +	queue_info->ena_vector = 1;
> >> +	qi_ev->event_type =
> >RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
> >> +	rxa_set_vector_data(queue_info, config->vector_sz,
> >> +			    config->vector_timeout_ns, config-
> >>vector_mp,
> >> +			    rx_queue_id, dev_info->dev->data->port_id);
> >> +	rx_adapter->ena_vector = 1;
> >> +	rx_adapter->vector_tmo_ticks =
> >> +		rx_adapter->vector_tmo_ticks ?
> >> +			      RTE_MIN(config->vector_timeout_ns << 1,
> >> +				      rx_adapter->vector_tmo_ticks) :
> >> +			      config->vector_timeout_ns << 1;
> >> +	rx_adapter->prev_expiry_ts = 0;
> >> +	TAILQ_INIT(&rx_adapter->vector_list);
> >> +}
> >> +
> >>  static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
> >>  		uint16_t eth_dev_id,
> >>  		int rx_queue_id,
> >> @@ -2081,6 +2285,15 @@
> >rte_event_eth_rx_adapter_queue_add(uint8_t id,
> >>  		return -EINVAL;
> >>  	}
> >>
> >> +	if ((cap &
> >RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
> >> +	    (queue_conf->rx_queue_flags &
> >> +	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
> >> +		RTE_EDEV_LOG_ERR("Event vectorization is not
> >supported,"
> >> +				 " eth port: %" PRIu16 " adapter id: %"
> >PRIu8,
> >> +				 eth_dev_id, id);
> >> +		return -EINVAL;
> >> +	}
> >> +
> >>  	if ((cap &
> >RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
> >>  		(rx_queue_id != -1)) {
> >>  		RTE_EDEV_LOG_ERR("Rx queues can only be connected
> >to single "
> >> @@ -2143,6 +2356,17 @@
> >rte_event_eth_rx_adapter_queue_add(uint8_t id,
> >>  	return 0;
> >>  }
> >>
> >> +static int
> >> +rxa_sw_vector_limits(struct
> >rte_event_eth_rx_adapter_vector_limits *limits)
> >> +{
> >> +	limits->max_sz = MAX_VECTOR_SIZE;
> >> +	limits->min_sz = MIN_VECTOR_SIZE;
> >> +	limits->max_timeout_ns = MAX_VECTOR_NS;
> >> +	limits->min_timeout_ns = MIN_VECTOR_NS;
> >> +
> >> +	return 0;
> >> +}
> >> +
> >>  int
> >>  rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t
> >eth_dev_id,
> >>  				int32_t rx_queue_id)
> >> @@ -2333,7 +2557,8 @@
> >rte_event_eth_rx_adapter_queue_event_vector_config(
> >>  		ret = dev->dev_ops-
> >>eth_rx_adapter_event_vector_config(
> >>  			dev, &rte_eth_devices[eth_dev_id],
> >rx_queue_id, config);
> >>  	} else {
> >> -		ret = -ENOTSUP;
> >> +		rxa_sw_event_vector_configure(rx_adapter,
> >eth_dev_id,
> >> +					      rx_queue_id, config);
> >>  	}
> >>
> >>  	return ret;
> >> @@ -2371,7 +2596,7 @@
> >rte_event_eth_rx_adapter_vector_limits_get(
> >>  		ret = dev->dev_ops-
> >>eth_rx_adapter_vector_limits_get(
> >>  			dev, &rte_eth_devices[eth_port_id], limits);
> >>  	} else {
> >> -		ret = -ENOTSUP;
> >> +		ret = rxa_sw_vector_limits(limits);
> >>  	}
> >>
> >>  	return ret;
> >> diff --git a/lib/librte_eventdev/rte_eventdev.c
> >b/lib/librte_eventdev/rte_eventdev.c
> >> index f95edc075..254a31b1f 100644
> >> --- a/lib/librte_eventdev/rte_eventdev.c
> >> +++ b/lib/librte_eventdev/rte_eventdev.c
> >> @@ -122,7 +122,11 @@ rte_event_eth_rx_adapter_caps_get(uint8_t
> >dev_id, uint16_t eth_port_id,
> >>
> >>  	if (caps == NULL)
> >>  		return -EINVAL;
> >> -	*caps = 0;
> >> +
> >> +	if (dev->dev_ops->eth_rx_adapter_caps_get == NULL)
> >> +		*caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
> >> +	else
> >> +		*caps = 0;
> >
> >Any reason why we had to set default caps value? I am thinking if sw
> >event device is used, it would set it anyways.
> >
> 
> There are multiple sw event devices which don't implement caps_get
> function, this changes solves that.
> 
> >>
> >>  	return dev->dev_ops->eth_rx_adapter_caps_get ?
> >>  				(*dev->dev_ops-
> >>eth_rx_adapter_caps_get)(dev,
> >> --
> >> 2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [dpdk-dev v21.11] [PATCH v5 8/8] eventdev: simplify Rx adapter event vector config
  2021-03-25 13:55             ` Pavan Nikhilesh Bhagavatula
@ 2021-03-26  7:09               ` Jayatheerthan, Jay
  2021-03-26  9:44                 ` Pavan Nikhilesh Bhagavatula
  0 siblings, 1 reply; 153+ messages in thread
From: Jayatheerthan, Jay @ 2021-03-26  7:09 UTC (permalink / raw)
  To: Pavan Nikhilesh Bhagavatula, Jerin Jacob Kollanukkaran, Carrillo,
	Erik G, Gujjar, Abhinandan S, McDaniel, Timothy, hemant.agrawal,
	Van Haaren, Harry, mattias.ronnblom, Ma, Liang J, Ray Kinsella,
	Neil Horman
  Cc: dev

> -----Original Message-----
> From: Pavan Nikhilesh Bhagavatula <pbhagavatula@marvell.com>
> Sent: Thursday, March 25, 2021 7:25 PM
> To: Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Jerin Jacob Kollanukkaran <jerinj@marvell.com>; Carrillo, Erik G
> <erik.g.carrillo@intel.com>; Gujjar, Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy
> <timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
> <mattias.ronnblom@ericsson.com>; Ma, Liang J <liang.j.ma@intel.com>; Ray Kinsella <mdr@ashroe.eu>; Neil Horman
> <nhorman@tuxdriver.com>
> Cc: dev@dpdk.org
> Subject: RE: [dpdk-dev v21.11] [PATCH v5 8/8] eventdev: simplify Rx adapter event vector config
> 
> >> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
> >> Sent: Wednesday, March 24, 2021 10:35 AM
> >> To: jerinj@marvell.com; Jayatheerthan, Jay
> ><jay.jayatheerthan@intel.com>; Carrillo, Erik G
> ><erik.g.carrillo@intel.com>; Gujjar,
> >> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy
> ><timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
> >> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
> ><mattias.ronnblom@ericsson.com>; Ma, Liang J
> >> <liang.j.ma@intel.com>; Ray Kinsella <mdr@ashroe.eu>; Neil Horman
> ><nhorman@tuxdriver.com>
> >> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
> >> Subject: [dpdk-dev v21.11] [PATCH v5 8/8] eventdev: simplify Rx
> >adapter event vector config
> >>
> >> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> >>
> >> Include vector configuration into the structure
> >> ``rte_event_eth_rx_adapter_queue_conf`` used when configuring rest
> >> of the Rx adapter ethernet device Rx queue parameters.
> >> This simplifies event vector configuration as it avoids splitting
> >> configuration per Rx queue.
> >>
> >> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> >> ---
> >>  app/test-eventdev/test_pipeline_common.c      |  16 +-
> >>  lib/librte_eventdev/eventdev_pmd.h            |  29 ---
> >>  .../rte_event_eth_rx_adapter.c                | 168 ++++++------------
> >>  .../rte_event_eth_rx_adapter.h                |  27 ---
> >>  lib/librte_eventdev/version.map               |   1 -
> >>  5 files changed, 57 insertions(+), 184 deletions(-)
> >>
> >> diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-
> >eventdev/test_pipeline_common.c
> >> index d5ef90500..76aee254b 100644
> >> --- a/app/test-eventdev/test_pipeline_common.c
> >> +++ b/app/test-eventdev/test_pipeline_common.c
> >> @@ -331,7 +331,6 @@ pipeline_event_rx_adapter_setup(struct
> >evt_options *opt, uint8_t stride,
> >>  	uint16_t prod;
> >>  	struct rte_mempool *vector_pool = NULL;
> >>  	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
> >> -	struct rte_event_eth_rx_adapter_event_vector_config
> >vec_conf;
> >>
> >>  	memset(&queue_conf, 0,
> >>  			sizeof(struct
> >rte_event_eth_rx_adapter_queue_conf));
> >> @@ -397,8 +396,12 @@ pipeline_event_rx_adapter_setup(struct
> >evt_options *opt, uint8_t stride,
> >>  			}
> >>
> >>  			if (cap &
> >RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
> >> +				queue_conf.vector_sz = opt-
> >>vector_size;
> >> +				queue_conf.vector_timeout_ns =
> >> +					opt->vector_tmo_nsec;
> >>  				queue_conf.rx_queue_flags |=
> >>
> >	RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
> >> +				queue_conf.vector_mp = vector_pool;
> >>  			} else {
> >>  				evt_err("Rx adapter doesn't support
> >event vector");
> >>  				return -EINVAL;
> >> @@ -418,17 +421,6 @@ pipeline_event_rx_adapter_setup(struct
> >evt_options *opt, uint8_t stride,
> >>  			return ret;
> >>  		}
> >>
> >> -		if (opt->ena_vector) {
> >> -			vec_conf.vector_sz = opt->vector_size;
> >> -			vec_conf.vector_timeout_ns = opt-
> >>vector_tmo_nsec;
> >> -			vec_conf.vector_mp = vector_pool;
> >> -			if
> >(rte_event_eth_rx_adapter_queue_event_vector_config(
> >> -				    prod, prod, -1, &vec_conf) < 0) {
> >> -				evt_err("Failed to configure event
> >vectorization for Rx adapter");
> >> -				return -EINVAL;
> >> -			}
> >> -		}
> >> -
> >>  		if (!(cap &
> >RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
> >>  			uint32_t service_id = -1U;
> >>
> >> diff --git a/lib/librte_eventdev/eventdev_pmd.h
> >b/lib/librte_eventdev/eventdev_pmd.h
> >> index 0f724ac85..63b3bc4b5 100644
> >> --- a/lib/librte_eventdev/eventdev_pmd.h
> >> +++ b/lib/librte_eventdev/eventdev_pmd.h
> >> @@ -667,32 +667,6 @@ typedef int
> >(*eventdev_eth_rx_adapter_vector_limits_get_t)(
> >>  	const struct rte_eventdev *dev, const struct rte_eth_dev
> >*eth_dev,
> >>  	struct rte_event_eth_rx_adapter_vector_limits *limits);
> >>
> >> -struct rte_event_eth_rx_adapter_event_vector_config;
> >> -/**
> >> - * Enable event vector on an given Rx queue of a ethernet devices
> >belonging to
> >> - * the Rx adapter.
> >> - *
> >> - * @param dev
> >> - *   Event device pointer
> >> - *
> >> - * @param eth_dev
> >> - *   Ethernet device pointer
> >> - *
> >> - * @param rx_queue_id
> >> - *   The Rx queue identifier
> >> - *
> >> - * @param config
> >> - *   Pointer to the event vector configuration structure.
> >> - *
> >> - * @return
> >> - *   - 0: Success.
> >> - *   - <0: Error code returned by the driver function.
> >> - */
> >> -typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
> >> -	const struct rte_eventdev *dev, const struct rte_eth_dev
> >*eth_dev,
> >> -	int32_t rx_queue_id,
> >> -	const struct rte_event_eth_rx_adapter_event_vector_config
> >*config);
> >> -
> >>  typedef uint32_t rte_event_pmd_selftest_seqn_t;
> >>  extern int rte_event_pmd_selftest_seqn_dynfield_offset;
> >>
> >> @@ -1118,9 +1092,6 @@ struct rte_eventdev_ops {
> >>  	eventdev_eth_rx_adapter_vector_limits_get_t
> >>  		eth_rx_adapter_vector_limits_get;
> >>  	/**< Get event vector limits for the Rx adapter */
> >> -	eventdev_eth_rx_adapter_event_vector_config_t
> >> -		eth_rx_adapter_event_vector_config;
> >> -	/**< Configure Rx adapter with event vector */
> >>
> >>  	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
> >>  	/**< Get timer adapter capabilities */
> >> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> >b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> >> index c71990078..a1990637f 100644
> >> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> >> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> >> @@ -1882,6 +1882,25 @@ rxa_add_queue(struct
> >rte_event_eth_rx_adapter *rx_adapter,
> >>  	} else
> >>  		qi_ev->flow_id = 0;
> >>
> >> +	if (conf->rx_queue_flags &
> >> +	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
> >> +		queue_info->ena_vector = 1;
> >> +		qi_ev->event_type =
> >RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
> >> +		rxa_set_vector_data(queue_info, conf->vector_sz,
> >> +				    conf->vector_timeout_ns, conf-
> >>vector_mp,
> >> +				    rx_queue_id, dev_info->dev->data-
> >>port_id);
> >> +		rx_adapter->ena_vector = 1;
> >> +		rx_adapter->vector_tmo_ticks =
> >> +			rx_adapter->vector_tmo_ticks
> >> +				? RTE_MIN(queue_info->vector_data
> >> +						  .vector_timeout_ticks,
> >> +					  rx_adapter-
> >>vector_tmo_ticks)
> >> +				: queue_info-
> >>vector_data.vector_timeout_ticks;
> >> +		rx_adapter->vector_tmo_ticks <<= 1;
> >
> >Any reason why we left shift here ? Applicable in patch 4/8 as well.
> 
> Just so that we have half the precision of the lowest timeout, helps to
> Maintain accuracy.

Maybe I am missing something here. For e.g. if the lowest timeout is 20 ticks, would you want 10 or 40 ? Currently, its set to 40.

> 
> >
> >> +		TAILQ_INIT(&rx_adapter->vector_list);
> >
> >Can doing TAILQ_INIT every time a queue is added cause existing
> >elements to be wiped out ? Applicable in patch 4/8 as well.
> 
> I don't think queues can be added when adapter is already started
> rxa_sw_add Isn't thready safe.

Yes, it takes lock before calling rxa_sa_add. For internal_port implementation, I don't see any lock being taken. Besides, since its adapter related, it would be more relevant in adapter create than in queue add.

> 
> >
> >> +		rx_adapter->prev_expiry_ts = 0;
> >
> >Can setting this every time a queue is added affect existing queues
> >created with vector support and passing traffic ? Applicable in patch 4/8
> >as well.
> 
> Same as above.
> 

Same reasoning as above.

> >
> >> +	}
> >> +
> >>  	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
> >>  	if (rxa_polled_queue(dev_info, rx_queue_id)) {
> >>  		rx_adapter->num_rx_polled += !pollq;
> >> @@ -1907,44 +1926,6 @@ rxa_add_queue(struct
> >rte_event_eth_rx_adapter *rx_adapter,
> >>  	}
> >>  }
> >>
> >> -static void
> >> -rxa_sw_event_vector_configure(
> >> -	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t
> >eth_dev_id,
> >> -	int rx_queue_id,
> >> -	const struct rte_event_eth_rx_adapter_event_vector_config
> >*config)
> >> -{
> >> -	struct eth_device_info *dev_info = &rx_adapter-
> >>eth_devices[eth_dev_id];
> >> -	struct eth_rx_queue_info *queue_info;
> >> -	struct rte_event *qi_ev;
> >> -
> >> -	if (rx_queue_id == -1) {
> >> -		uint16_t nb_rx_queues;
> >> -		uint16_t i;
> >> -
> >> -		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
> >> -		for (i = 0; i < nb_rx_queues; i++)
> >> -			rxa_sw_event_vector_configure(rx_adapter,
> >eth_dev_id, i,
> >> -						      config);
> >> -		return;
> >> -	}
> >> -
> >> -	queue_info = &dev_info->rx_queue[rx_queue_id];
> >> -	qi_ev = (struct rte_event *)&queue_info->event;
> >> -	queue_info->ena_vector = 1;
> >> -	qi_ev->event_type =
> >RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
> >> -	rxa_set_vector_data(queue_info, config->vector_sz,
> >> -			    config->vector_timeout_ns, config-
> >>vector_mp,
> >> -			    rx_queue_id, dev_info->dev->data->port_id);
> >> -	rx_adapter->ena_vector = 1;
> >> -	rx_adapter->vector_tmo_ticks =
> >> -		rx_adapter->vector_tmo_ticks ?
> >> -			      RTE_MIN(config->vector_timeout_ns << 1,
> >> -				      rx_adapter->vector_tmo_ticks) :
> >> -			      config->vector_timeout_ns << 1;
> >> -	rx_adapter->prev_expiry_ts = 0;
> >> -	TAILQ_INIT(&rx_adapter->vector_list);
> >> -}
> >> -
> >>  static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
> >>  		uint16_t eth_dev_id,
> >>  		int rx_queue_id,
> >> @@ -2258,6 +2239,7 @@
> >rte_event_eth_rx_adapter_queue_add(uint8_t id,
> >>  	struct rte_event_eth_rx_adapter *rx_adapter;
> >>  	struct rte_eventdev *dev;
> >>  	struct eth_device_info *dev_info;
> >> +	struct rte_event_eth_rx_adapter_vector_limits limits;
> >>
> >>  	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -
> >EINVAL);
> >>  	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
> >> @@ -2294,6 +2276,39 @@
> >rte_event_eth_rx_adapter_queue_add(uint8_t id,
> >>  		return -EINVAL;
> >>  	}
> >>
> >> +	if (queue_conf->rx_queue_flags &
> >> +	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
> >
> >Perhaps, you could move the previous if condition here and just check
> >for (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)
> 
> No, the above check is not a capability check, it is to check if application
> requested to enable vectorization for this queue or not.

In current code, same condition check (queue_conf->rx_queue_flags & RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR ) is done twice next to each other, which could be avoided.

	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
---->	    (queue_conf->rx_queue_flags &
	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
				 eth_dev_id, id);
		return -EINVAL;
	}

---->	if (queue_conf->rx_queue_flags &
	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
	...
	}

> 
> >
> >> +		ret = rte_event_eth_rx_adapter_vector_limits_get(
> >> +			rx_adapter->eventdev_id, eth_dev_id, &limits);
> >> +		if (ret < 0) {
> >> +			RTE_EDEV_LOG_ERR("Failed to get event
> >device vector limits,"
> >> +					 " eth port: %" PRIu16
> >> +					 " adapter id: %" PRIu8,
> >> +					 eth_dev_id, id);
> >> +			return -EINVAL;
> >> +		}
> >> +		if (queue_conf->vector_sz < limits.min_sz ||
> >> +		    queue_conf->vector_sz > limits.max_sz ||
> >> +		    queue_conf->vector_timeout_ns <
> >limits.min_timeout_ns ||
> >> +		    queue_conf->vector_timeout_ns >
> >limits.max_timeout_ns ||
> >> +		    queue_conf->vector_mp == NULL) {
> >> +			RTE_EDEV_LOG_ERR("Invalid event vector
> >configuration,"
> >> +					 " eth port: %" PRIu16
> >> +					 " adapter id: %" PRIu8,
> >> +					 eth_dev_id, id);
> >> +			return -EINVAL;
> >> +		}
> >> +		if (queue_conf->vector_mp->elt_size <
> >> +		    (sizeof(struct rte_event_vector) +
> >> +		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
> >> +			RTE_EDEV_LOG_ERR("Invalid event vector
> >configuration,"
> >> +					 " eth port: %" PRIu16
> >> +					 " adapter id: %" PRIu8,
> >> +					 eth_dev_id, id);
> >> +			return -EINVAL;
> >> +		}
> >> +	}
> >> +
> >>  	if ((cap &
> >RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
> >>  		(rx_queue_id != -1)) {
> >>  		RTE_EDEV_LOG_ERR("Rx queues can only be connected
> >to single "
> >> @@ -2487,83 +2502,6 @@
> >rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
> >>  	return ret;
> >>  }
> >>
> >> -int
> >> -rte_event_eth_rx_adapter_queue_event_vector_config(
> >> -	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
> >> -	struct rte_event_eth_rx_adapter_event_vector_config *config)
> >> -{
> >> -	struct rte_event_eth_rx_adapter_vector_limits limits;
> >> -	struct rte_event_eth_rx_adapter *rx_adapter;
> >> -	struct rte_eventdev *dev;
> >> -	uint32_t cap;
> >> -	int ret;
> >> -
> >> -	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -
> >EINVAL);
> >> -	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
> >> -
> >> -	rx_adapter = rxa_id_to_adapter(id);
> >> -	if ((rx_adapter == NULL) || (config == NULL))
> >> -		return -EINVAL;
> >> -
> >> -	dev = &rte_eventdevs[rx_adapter->eventdev_id];
> >> -	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter-
> >>eventdev_id,
> >> -						eth_dev_id, &cap);
> >> -	if (ret) {
> >> -		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev
> >%" PRIu8
> >> -				 "eth port %" PRIu16,
> >> -				 id, eth_dev_id);
> >> -		return ret;
> >> -	}
> >> -
> >> -	if (!(cap &
> >RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
> >> -		RTE_EDEV_LOG_ERR("Event vectorization is not
> >supported,"
> >> -				 " eth port: %" PRIu16 " adapter id: %"
> >PRIu8,
> >> -				 eth_dev_id, id);
> >> -		return -EINVAL;
> >> -	}
> >> -
> >> -	ret = rte_event_eth_rx_adapter_vector_limits_get(
> >> -		rx_adapter->eventdev_id, eth_dev_id, &limits);
> >> -	if (ret) {
> >> -		RTE_EDEV_LOG_ERR("Failed to get vector limits edev
> >%" PRIu8
> >> -				 "eth port %" PRIu16,
> >> -				 rx_adapter->eventdev_id, eth_dev_id);
> >> -		return ret;
> >> -	}
> >> -
> >> -	if (config->vector_sz < limits.min_sz ||
> >> -	    config->vector_sz > limits.max_sz ||
> >> -	    config->vector_timeout_ns < limits.min_timeout_ns ||
> >> -	    config->vector_timeout_ns > limits.max_timeout_ns ||
> >> -	    config->vector_mp == NULL) {
> >> -		RTE_EDEV_LOG_ERR("Invalid event vector
> >configuration,"
> >> -				 " eth port: %" PRIu16 " adapter id: %"
> >PRIu8,
> >> -				 eth_dev_id, id);
> >> -		return -EINVAL;
> >> -	}
> >> -	if (config->vector_mp->elt_size <
> >> -	    (sizeof(struct rte_event_vector) +
> >> -	     (sizeof(uintptr_t) * config->vector_sz))) {
> >> -		RTE_EDEV_LOG_ERR("Invalid event vector
> >configuration,"
> >> -				 " eth port: %" PRIu16 " adapter id: %"
> >PRIu8,
> >> -				 eth_dev_id, id);
> >> -		return -EINVAL;
> >> -	}
> >> -
> >> -	if (cap &
> >RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
> >> -		RTE_FUNC_PTR_OR_ERR_RET(
> >> -			*dev->dev_ops-
> >>eth_rx_adapter_event_vector_config,
> >> -			-ENOTSUP);
> >> -		ret = dev->dev_ops-
> >>eth_rx_adapter_event_vector_config(
> >> -			dev, &rte_eth_devices[eth_dev_id],
> >rx_queue_id, config);
> >> -	} else {
> >> -		rxa_sw_event_vector_configure(rx_adapter,
> >eth_dev_id,
> >> -					      rx_queue_id, config);
> >> -	}
> >> -
> >> -	return ret;
> >> -}
> >> -
> >>  int
> >>  rte_event_eth_rx_adapter_vector_limits_get(
> >>  	uint8_t dev_id, uint16_t eth_port_id,
> >> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> >b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> >> index 7407cde00..3f8b36229 100644
> >> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> >> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> >> @@ -171,9 +171,6 @@ struct rte_event_eth_rx_adapter_queue_conf
> >{
> >>  	 * The event adapter sets ev.event_type to
> >RTE_EVENT_TYPE_ETHDEV in the
> >>  	 * enqueued event.
> >>  	 */
> >> -};
> >> -
> >> -struct rte_event_eth_rx_adapter_event_vector_config {
> >>  	uint16_t vector_sz;
> >>  	/**<
> >>  	 * Indicates the maximum number for mbufs to combine and
> >form a vector.
> >> @@ -548,30 +545,6 @@ int
> >rte_event_eth_rx_adapter_vector_limits_get(
> >>  	uint8_t dev_id, uint16_t eth_port_id,
> >>  	struct rte_event_eth_rx_adapter_vector_limits *limits);
> >>
> >> -/**
> >> - * Configure event vectorization for a given ethernet device queue,
> >that has
> >> - * been added to a event eth Rx adapter.
> >> - *
> >> - * @param id
> >> - *  The identifier of the ethernet Rx event adapter.
> >> - *
> >> - * @param eth_dev_id
> >> - *  The identifier of the ethernet device.
> >> - *
> >> - * @param rx_queue_id
> >> - *  Ethernet device receive queue index.
> >> - *  If rx_queue_id is -1, then all Rx queues configured for the ethernet
> >device
> >> - *  are configured with event vectorization.
> >> - *
> >> - * @return
> >> - *  - 0: Success, Receive queue configured correctly.
> >> - *  - <0: Error code on failure.
> >> - */
> >> -__rte_experimental
> >> -int rte_event_eth_rx_adapter_queue_event_vector_config(
> >> -	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
> >> -	struct rte_event_eth_rx_adapter_event_vector_config
> >*config);
> >> -
> >>  #ifdef __cplusplus
> >>  }
> >>  #endif
> >> diff --git a/lib/librte_eventdev/version.map
> >b/lib/librte_eventdev/version.map
> >> index 902df0ae3..34c1c830e 100644
> >> --- a/lib/librte_eventdev/version.map
> >> +++ b/lib/librte_eventdev/version.map
> >> @@ -142,7 +142,6 @@ EXPERIMENTAL {
> >>  	#added in 21.05
> >>  	rte_event_vector_pool_create;
> >>  	rte_event_eth_rx_adapter_vector_limits_get;
> >> -	rte_event_eth_rx_adapter_queue_event_vector_config;
> >>  };
> >>
> >>  INTERNAL {
> >> --
> >> 2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v5 4/8] eventdev: add Rx adapter event vector support
  2021-03-26  6:26               ` Jayatheerthan, Jay
@ 2021-03-26  9:00                 ` Pavan Nikhilesh Bhagavatula
  0 siblings, 0 replies; 153+ messages in thread
From: Pavan Nikhilesh Bhagavatula @ 2021-03-26  9:00 UTC (permalink / raw)
  To: Jayatheerthan, Jay, Jerin Jacob Kollanukkaran, Carrillo, Erik G,
	Gujjar, Abhinandan S, McDaniel, Timothy, hemant.agrawal,
	Van Haaren, Harry, mattias.ronnblom, Ma, Liang J
  Cc: dev

>> From: Pavan Nikhilesh Bhagavatula <pbhagavatula@marvell.com>
>> Sent: Thursday, March 25, 2021 6:44 PM
>> To: Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Jerin Jacob
>Kollanukkaran <jerinj@marvell.com>; Carrillo, Erik G
>> <erik.g.carrillo@intel.com>; Gujjar, Abhinandan S
><abhinandan.gujjar@intel.com>; McDaniel, Timothy
>> <timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
>Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
>> <mattias.ronnblom@ericsson.com>; Ma, Liang J
><liang.j.ma@intel.com>
>> Cc: dev@dpdk.org
>> Subject: RE: [dpdk-dev] [PATCH v5 4/8] eventdev: add Rx adapter
>event vector support
>>
>>
>>
>> >-----Original Message-----
>> >From: Jayatheerthan, Jay <jay.jayatheerthan@intel.com>
>> >Sent: Thursday, March 25, 2021 4:07 PM
>> >To: Pavan Nikhilesh Bhagavatula <pbhagavatula@marvell.com>;
>Jerin
>> >Jacob Kollanukkaran <jerinj@marvell.com>; Carrillo, Erik G
>> ><erik.g.carrillo@intel.com>; Gujjar, Abhinandan S
>> ><abhinandan.gujjar@intel.com>; McDaniel, Timothy
>> ><timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
>> >Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
>> ><mattias.ronnblom@ericsson.com>; Ma, Liang J
>> ><liang.j.ma@intel.com>
>> >Cc: dev@dpdk.org
>> >Subject: [EXT] RE: [dpdk-dev] [PATCH v5 4/8] eventdev: add Rx
>adapter
>> >event vector support
>> >
>> >External Email
>> >
>> >----------------------------------------------------------------------
>> >> -----Original Message-----
>> >> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
>> >> Sent: Wednesday, March 24, 2021 10:35 AM
>> >> To: jerinj@marvell.com; Jayatheerthan, Jay
>> ><jay.jayatheerthan@intel.com>; Carrillo, Erik G
>> ><erik.g.carrillo@intel.com>; Gujjar,
>> >> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy
>> ><timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
>> >> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
>> ><mattias.ronnblom@ericsson.com>; Ma, Liang J
>> >> <liang.j.ma@intel.com>
>> >> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
>> >> Subject: [dpdk-dev] [PATCH v5 4/8] eventdev: add Rx adapter
>event
>> >vector support
>> >>
>> >> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>> >>
>> >> Add event vector support for event eth Rx adapter, the
>> >implementation
>> >> creates vector flows based on port and queue identifier of the
>> >received
>> >> mbufs.
>> >>
>> >> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
>> >> ---
>> >>  lib/librte_eventdev/eventdev_pmd.h            |   7 +-
>> >>  .../rte_event_eth_rx_adapter.c                | 257 ++++++++++++++++-
>-
>> >>  lib/librte_eventdev/rte_eventdev.c            |   6 +-
>> >>  3 files changed, 250 insertions(+), 20 deletions(-)
>> >>
>> >> diff --git a/lib/librte_eventdev/eventdev_pmd.h
>> >b/lib/librte_eventdev/eventdev_pmd.h
>> >> index 9297f1433..0f724ac85 100644
>> >> --- a/lib/librte_eventdev/eventdev_pmd.h
>> >> +++ b/lib/librte_eventdev/eventdev_pmd.h
>> >> @@ -69,9 +69,10 @@ extern "C" {
>> >>  	} \
>> >>  } while (0)
>> >>
>> >> -#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP \
>> >> -
>> >	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |
>> >\
>> >> -
>> >	(RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ))
>> >> +#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP
>> >\
>> >> +	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |
>> >\
>> >> +	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) |
>> >\
>> >> +	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR))
>> >>
>> >>  #define RTE_EVENT_CRYPTO_ADAPTER_SW_CAP \
>> >>
>> >	RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA
>> >> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> >b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> >> index ac8ba5bf0..c71990078 100644
>> >> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> >> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> >> @@ -26,6 +26,10 @@
>> >>  #define BATCH_SIZE		32
>> >>  #define BLOCK_CNT_THRESHOLD	10
>> >>  #define ETH_EVENT_BUFFER_SIZE	(4*BATCH_SIZE)
>> >> +#define MAX_VECTOR_SIZE		1024
>> >> +#define MIN_VECTOR_SIZE		4
>> >> +#define MAX_VECTOR_NS		1E9
>> >> +#define MIN_VECTOR_NS		1E5
>> >>
>> >>  #define ETH_RX_ADAPTER_SERVICE_NAME_LEN	32
>> >>  #define ETH_RX_ADAPTER_MEM_NAME_LEN	32
>> >> @@ -59,6 +63,20 @@ struct eth_rx_poll_entry {
>> >>  	uint16_t eth_rx_qid;
>> >>  };
>> >>
>> >> +struct eth_rx_vector_data {
>> >> +	TAILQ_ENTRY(eth_rx_vector_data) next;
>> >> +	uint16_t port;
>> >> +	uint16_t queue;
>> >> +	uint16_t max_vector_count;
>> >> +	uint64_t event;
>> >> +	uint64_t ts;
>> >> +	uint64_t vector_timeout_ticks;
>> >> +	struct rte_mempool *vector_pool;
>> >> +	struct rte_event_vector *vector_ev;
>> >> +} __rte_cache_aligned;
>> >> +
>> >> +TAILQ_HEAD(eth_rx_vector_data_list, eth_rx_vector_data);
>> >> +
>> >>  /* Instance per adapter */
>> >>  struct rte_eth_event_enqueue_buffer {
>> >>  	/* Count of events in this buffer */
>> >> @@ -92,6 +110,14 @@ struct rte_event_eth_rx_adapter {
>> >>  	uint32_t wrr_pos;
>> >>  	/* Event burst buffer */
>> >>  	struct rte_eth_event_enqueue_buffer event_enqueue_buffer;
>> >> +	/* Vector enable flag */
>> >> +	uint8_t ena_vector;
>> >> +	/* Timestamp of previous vector expiry list traversal */
>> >> +	uint64_t prev_expiry_ts;
>> >> +	/* Minimum ticks to wait before traversing expiry list */
>> >> +	uint64_t vector_tmo_ticks;
>> >> +	/* vector list */
>> >> +	struct eth_rx_vector_data_list vector_list;
>> >>  	/* Per adapter stats */
>> >>  	struct rte_event_eth_rx_adapter_stats stats;
>> >>  	/* Block count, counts up to BLOCK_CNT_THRESHOLD */
>> >> @@ -198,9 +224,11 @@ struct eth_device_info {
>> >>  struct eth_rx_queue_info {
>> >>  	int queue_enabled;	/* True if added */
>> >>  	int intr_enabled;
>> >> +	uint8_t ena_vector;
>> >>  	uint16_t wt;		/* Polling weight */
>> >>  	uint32_t flow_id_mask;	/* Set to ~0 if app provides flow id else
>> >0 */
>> >>  	uint64_t event;
>> >> +	struct eth_rx_vector_data vector_data;
>> >>  };
>> >>
>> >>  static struct rte_event_eth_rx_adapter **event_eth_rx_adapter;
>> >> @@ -722,6 +750,9 @@ rxa_flush_event_buffer(struct
>> >rte_event_eth_rx_adapter *rx_adapter)
>> >>  	    &rx_adapter->event_enqueue_buffer;
>> >>  	struct rte_event_eth_rx_adapter_stats *stats = &rx_adapter-
>> >>stats;
>> >>
>> >> +	if (!buf->count)
>> >> +		return 0;
>> >> +
>> >>  	uint16_t n = rte_event_enqueue_new_burst(rx_adapter-
>> >>eventdev_id,
>> >>  					rx_adapter->event_port_id,
>> >>  					buf->events,
>> >> @@ -742,6 +773,72 @@ rxa_flush_event_buffer(struct
>> >rte_event_eth_rx_adapter *rx_adapter)
>> >>  	return n;
>> >>  }
>> >>
>> >> +static inline uint16_t
>> >> +rxa_create_event_vector(struct rte_event_eth_rx_adapter
>> >*rx_adapter,
>> >> +			struct eth_rx_queue_info *queue_info,
>> >> +			struct rte_eth_event_enqueue_buffer *buf,
>> >> +			struct rte_mbuf **mbufs, uint16_t num)
>> >> +{
>> >> +	struct rte_event *ev = &buf->events[buf->count];
>> >> +	struct eth_rx_vector_data *vec;
>> >> +	uint16_t filled, space, sz;
>> >> +
>> >> +	filled = 0;
>> >> +	vec = &queue_info->vector_data;
>> >> +	while (num) {
>> >> +		if (vec->vector_ev == NULL) {
>> >> +			if (rte_mempool_get(vec->vector_pool,
>> >> +					    (void **)&vec->vector_ev) <
>> >0) {
>> >> +				rte_pktmbuf_free_bulk(mbufs, num);
>> >> +				return 0;
>> >> +			}
>> >> +			vec->vector_ev->nb_elem = 0;
>> >> +			vec->vector_ev->port = vec->port;
>> >> +			vec->vector_ev->queue = vec->queue;
>> >> +			vec->vector_ev->attr_valid = true;
>> >> +			TAILQ_INSERT_TAIL(&rx_adapter->vector_list,
>> >vec, next);
>> >> +		} else if (vec->vector_ev->nb_elem == vec-
>> >>max_vector_count) {
>> >
>> >Is there a case where nb_elem > max_vector_count as we
>accumulate
>> >sz to it ?
>>
>> I don't think so, that would overflow the vector event.
>>
>> >
>> >> +			/* Event ready. */
>> >> +			ev->event = vec->event;
>> >> +			ev->vec = vec->vector_ev;
>> >> +			ev++;
>> >> +			filled++;
>> >> +			vec->vector_ev = NULL;
>> >> +			TAILQ_REMOVE(&rx_adapter->vector_list, vec,
>> >next);
>> >> +			if (rte_mempool_get(vec->vector_pool,
>> >> +					    (void **)&vec->vector_ev) <
>> >0) {
>> >> +				rte_pktmbuf_free_bulk(mbufs, num);
>> >> +				return 0;
>> >> +			}
>> >> +			vec->vector_ev->nb_elem = 0;
>> >> +			vec->vector_ev->port = vec->port;
>> >> +			vec->vector_ev->queue = vec->queue;
>> >> +			vec->vector_ev->attr_valid = true;
>> >> +			TAILQ_INSERT_TAIL(&rx_adapter->vector_list,
>> >vec, next);
>> >> +		}
>> >> +
>> >> +		space = vec->max_vector_count - vec->vector_ev-
>> >>nb_elem;
>> >> +		sz = num > space ? space : num;
>> >> +		memcpy(vec->vector_ev->mbufs + vec->vector_ev-
>> >>nb_elem, mbufs,
>> >> +		       sizeof(void *) * sz);
>> >> +		vec->vector_ev->nb_elem += sz;
>> >> +		num -= sz;
>> >> +		mbufs += sz;
>> >> +		vec->ts = rte_rdtsc();
>> >> +	}
>> >> +
>> >> +	if (vec->vector_ev->nb_elem == vec->max_vector_count) {
>> >
>> >Same here.
>> >
>> >> +		ev->event = vec->event;
>> >> +		ev->vec = vec->vector_ev;
>> >> +		ev++;
>> >> +		filled++;
>> >> +		vec->vector_ev = NULL;
>> >> +		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
>> >> +	}
>> >> +
>> >> +	return filled;
>> >> +}
>> >
>> >I am seeing more than one repeating code chunks in this function.
>> >Perhaps, you can give it a try to not repeat. We can drop if its
>> >performance affecting.
>>
>> I will try to move them to inline functions and test.
>>
>> >
>> >> +
>> >>  static inline void
>> >>  rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
>> >>  		uint16_t eth_dev_id,
>> >> @@ -770,25 +867,30 @@ rxa_buffer_mbufs(struct
>> >rte_event_eth_rx_adapter *rx_adapter,
>> >>  	rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
>> >>  	do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
>> >
>> >The RSS related code is executed for vector case as well. Can this be
>> >moved inside ena_vector if condition ?
>>
>> RSS is used to generate the event flowid, in vector case the flow id
>> Will be a combination of port and queue id.
>> The idea is that flows having the same RSS LSB will end up in the same
>> queue.
>>
>
>I meant to say, rss_mask and do_rss are used only when ena_vector is
>false. Could be moved inside the appropriate condition ?
>

Ah! I see I will move them inside the conditional.

>> >
>> >>
>> >> -	for (i = 0; i < num; i++) {
>> >> -		m = mbufs[i];
>> >> -
>> >> -		rss = do_rss ?
>> >> -			rxa_do_softrss(m, rx_adapter->rss_key_be) :
>> >> -			m->hash.rss;
>> >> -		ev->event = event;
>> >> -		ev->flow_id = (rss & ~flow_id_mask) |
>> >> -				(ev->flow_id & flow_id_mask);
>> >> -		ev->mbuf = m;
>> >> -		ev++;
>> >> +	if (!eth_rx_queue_info->ena_vector) {
>> >> +		for (i = 0; i < num; i++) {
>> >> +			m = mbufs[i];
>> >> +
>> >> +			rss = do_rss ? rxa_do_softrss(m, rx_adapter-
>> >>rss_key_be)
>> >> +				     : m->hash.rss;
>> >> +			ev->event = event;
>> >> +			ev->flow_id = (rss & ~flow_id_mask) |
>> >> +				      (ev->flow_id & flow_id_mask);
>> >> +			ev->mbuf = m;
>> >> +			ev++;
>> >> +		}
>> >> +	} else {
>> >> +		num = rxa_create_event_vector(rx_adapter,
>> >eth_rx_queue_info,
>> >> +					      buf, mbufs, num);
>> >>  	}
>> >>
>> >> -	if (dev_info->cb_fn) {
>> >> +	if (num && dev_info->cb_fn) {
>> >>
>> >>  		dropped = 0;
>> >>  		nb_cb = dev_info->cb_fn(eth_dev_id, rx_queue_id,
>> >> -					ETH_EVENT_BUFFER_SIZE, buf-
>> >>count, ev,
>> >> -					num, dev_info->cb_arg,
>> >&dropped);
>> >> +					ETH_EVENT_BUFFER_SIZE, buf-
>> >>count,
>> >> +					&buf->events[buf->count],
>> >num,
>> >> +					dev_info->cb_arg, &dropped);
>> >
>> >Before this patch, we pass ev which is &buf->events[buf->count] +
>num
>> >as fifth param when calling cb_fn. Now, we are passing &buf-
>> >>events[buf->count] for non-vector case. Do you see this as an
>issue?
>> >
>>
>> The callback function takes in the array newly formed events i.e. we
>need
>> to pass the start of array and the count.
>>
>> the previous code had a bug where it passes the end of the event list.
>
>ok, that makes sense.
>
>>
>> >Also, for vector case would it make sense to do pass &buf-
>>events[buf-
>> >>count] + num ?
>> >
>> >>  		if (unlikely(nb_cb > num))
>> >>  			RTE_EDEV_LOG_ERR("Rx CB returned %d (> %d)
>> >events",
>> >>  				nb_cb, num);
>> >> @@ -1124,6 +1226,30 @@ rxa_poll(struct
>rte_event_eth_rx_adapter
>> >*rx_adapter)
>> >>  	return nb_rx;
>> >>  }
>> >>
>> >> +static void
>> >> +rxa_vector_expire(struct eth_rx_vector_data *vec, void *arg)
>> >> +{
>> >> +	struct rte_event_eth_rx_adapter *rx_adapter = arg;
>> >> +	struct rte_eth_event_enqueue_buffer *buf =
>> >> +		&rx_adapter->event_enqueue_buffer;
>> >> +	struct rte_event *ev;
>> >> +
>> >> +	if (buf->count)
>> >> +		rxa_flush_event_buffer(rx_adapter);
>> >> +
>> >> +	if (vec->vector_ev->nb_elem == 0)
>> >> +		return;
>> >> +	ev = &buf->events[buf->count];
>> >> +
>> >> +	/* Event ready. */
>> >> +	ev->event = vec->event;
>> >> +	ev->vec = vec->vector_ev;
>> >> +	buf->count++;
>> >> +
>> >> +	vec->vector_ev = NULL;
>> >> +	vec->ts = 0;
>> >> +}
>> >> +
>> >>  static int
>> >>  rxa_service_func(void *args)
>> >>  {
>> >> @@ -1137,6 +1263,24 @@ rxa_service_func(void *args)
>> >>  		return 0;
>> >>  	}
>> >>
>> >> +	if (rx_adapter->ena_vector) {
>> >> +		if ((rte_rdtsc() - rx_adapter->prev_expiry_ts) >=
>> >> +		    rx_adapter->vector_tmo_ticks) {
>> >> +			struct eth_rx_vector_data *vec;
>> >> +
>> >> +			TAILQ_FOREACH(vec, &rx_adapter->vector_list,
>> >next) {
>> >> +				uint64_t elapsed_time = rte_rdtsc() -
>> >vec->ts;
>> >> +
>> >> +				if (elapsed_time >= vec-
>> >>vector_timeout_ticks) {
>> >> +					rxa_vector_expire(vec,
>> >rx_adapter);
>> >> +					TAILQ_REMOVE(&rx_adapter-
>> >>vector_list,
>> >> +						     vec, next);
>> >> +				}
>> >> +			}
>> >> +			rx_adapter->prev_expiry_ts = rte_rdtsc();
>> >> +		}
>> >> +	}
>> >> +
>> >>  	stats = &rx_adapter->stats;
>> >>  	stats->rx_packets += rxa_intr_ring_dequeue(rx_adapter);
>> >>  	stats->rx_packets += rxa_poll(rx_adapter);
>> >> @@ -1640,6 +1784,28 @@ rxa_update_queue(struct
>> >rte_event_eth_rx_adapter *rx_adapter,
>> >>  	}
>> >>  }
>> >>
>> >> +static void
>> >> +rxa_set_vector_data(struct eth_rx_queue_info *queue_info,
>> >uint16_t vector_count,
>> >> +		    uint64_t vector_ns, struct rte_mempool *mp, int32_t
>> >qid,
>> >> +		    uint16_t port_id)
>> >> +{
>> >> +#define NSEC2TICK(__ns, __freq) (((__ns) * (__freq)) / 1E9)
>> >> +	struct eth_rx_vector_data *vector_data;
>> >> +	uint32_t flow_id;
>> >> +
>> >> +	vector_data = &queue_info->vector_data;
>> >> +	vector_data->max_vector_count = vector_count;
>> >> +	vector_data->port = port_id;
>> >> +	vector_data->queue = qid;
>> >> +	vector_data->vector_pool = mp;
>> >> +	vector_data->vector_timeout_ticks =
>> >> +		NSEC2TICK(vector_ns, rte_get_timer_hz());
>> >> +	vector_data->ts = 0;
>> >> +	flow_id = queue_info->event & 0xFFFFF;
>> >> +	flow_id = flow_id == 0 ? (qid & 0xFF) | (port_id & 0xFFFF) :
>> >flow_id;
>> >
>> >Maybe I am missing something here. Looking at the code it looks like
>> >qid and port_id may overlap. For e.g., if qid = 0x10 and port_id =
>0x11,
>> >flow_id would end up being 0x11. Is this the expectation? Also, it
>may
>> >be useful to document flow_id format.
>>
>> The flow_id is 20 bit, I guess we could do 12bit queue_id and 8bit port
>> as a flow.
>
>This sounds reasonable to me. It would be useful to have the flow_id
>format and how it is used for vectorization in Rx/Tx adapter
>documentation.

This is only applicable to the SW Rx adapter implementation a HW implementation might have
its own way of implementing the flow aggregation.
There is no documentation specific to SW Rx adapter, I will add it to the commit log.

>
>>
>> >Comparing this format with existing RSS hash based method, are we
>> >saying that all mbufs received in a rx burst are part of same flow
>when
>> >vectorization is used?
>>
>> Yes, the hard way to do this is to use a hash table and treating each
>> mbuf having an unique flow.
>>
>> >
>> >> +	vector_data->event = (queue_info->event & ~0xFFFFF) |
>> >flow_id;
>> >> +}
>> >> +
>> >>  static void
>> >>  rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
>> >>  	struct eth_device_info *dev_info,
>> >> @@ -1741,6 +1907,44 @@ rxa_add_queue(struct
>> >rte_event_eth_rx_adapter *rx_adapter,
>> >>  	}
>> >>  }
>> >>
>> >> +static void
>> >> +rxa_sw_event_vector_configure(
>> >> +	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t
>> >eth_dev_id,
>> >> +	int rx_queue_id,
>> >> +	const struct rte_event_eth_rx_adapter_event_vector_config
>> >*config)
>> >> +{
>> >> +	struct eth_device_info *dev_info = &rx_adapter-
>> >>eth_devices[eth_dev_id];
>> >> +	struct eth_rx_queue_info *queue_info;
>> >> +	struct rte_event *qi_ev;
>> >> +
>> >> +	if (rx_queue_id == -1) {
>> >> +		uint16_t nb_rx_queues;
>> >> +		uint16_t i;
>> >> +
>> >> +		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
>> >> +		for (i = 0; i < nb_rx_queues; i++)
>> >> +			rxa_sw_event_vector_configure(rx_adapter,
>> >eth_dev_id, i,
>> >> +						      config);
>> >> +		return;
>> >> +	}
>> >> +
>> >> +	queue_info = &dev_info->rx_queue[rx_queue_id];
>> >> +	qi_ev = (struct rte_event *)&queue_info->event;
>> >> +	queue_info->ena_vector = 1;
>> >> +	qi_ev->event_type =
>> >RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
>> >> +	rxa_set_vector_data(queue_info, config->vector_sz,
>> >> +			    config->vector_timeout_ns, config-
>> >>vector_mp,
>> >> +			    rx_queue_id, dev_info->dev->data->port_id);
>> >> +	rx_adapter->ena_vector = 1;
>> >> +	rx_adapter->vector_tmo_ticks =
>> >> +		rx_adapter->vector_tmo_ticks ?
>> >> +			      RTE_MIN(config->vector_timeout_ns << 1,
>> >> +				      rx_adapter->vector_tmo_ticks) :
>> >> +			      config->vector_timeout_ns << 1;
>> >> +	rx_adapter->prev_expiry_ts = 0;
>> >> +	TAILQ_INIT(&rx_adapter->vector_list);
>> >> +}
>> >> +
>> >>  static int rxa_sw_add(struct rte_event_eth_rx_adapter
>*rx_adapter,
>> >>  		uint16_t eth_dev_id,
>> >>  		int rx_queue_id,
>> >> @@ -2081,6 +2285,15 @@
>> >rte_event_eth_rx_adapter_queue_add(uint8_t id,
>> >>  		return -EINVAL;
>> >>  	}
>> >>
>> >> +	if ((cap &
>> >RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
>> >> +	    (queue_conf->rx_queue_flags &
>> >> +	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
>> >> +		RTE_EDEV_LOG_ERR("Event vectorization is not
>> >supported,"
>> >> +				 " eth port: %" PRIu16 " adapter id: %"
>> >PRIu8,
>> >> +				 eth_dev_id, id);
>> >> +		return -EINVAL;
>> >> +	}
>> >> +
>> >>  	if ((cap &
>> >RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
>> >>  		(rx_queue_id != -1)) {
>> >>  		RTE_EDEV_LOG_ERR("Rx queues can only be connected
>> >to single "
>> >> @@ -2143,6 +2356,17 @@
>> >rte_event_eth_rx_adapter_queue_add(uint8_t id,
>> >>  	return 0;
>> >>  }
>> >>
>> >> +static int
>> >> +rxa_sw_vector_limits(struct
>> >rte_event_eth_rx_adapter_vector_limits *limits)
>> >> +{
>> >> +	limits->max_sz = MAX_VECTOR_SIZE;
>> >> +	limits->min_sz = MIN_VECTOR_SIZE;
>> >> +	limits->max_timeout_ns = MAX_VECTOR_NS;
>> >> +	limits->min_timeout_ns = MIN_VECTOR_NS;
>> >> +
>> >> +	return 0;
>> >> +}
>> >> +
>> >>  int
>> >>  rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t
>> >eth_dev_id,
>> >>  				int32_t rx_queue_id)
>> >> @@ -2333,7 +2557,8 @@
>> >rte_event_eth_rx_adapter_queue_event_vector_config(
>> >>  		ret = dev->dev_ops-
>> >>eth_rx_adapter_event_vector_config(
>> >>  			dev, &rte_eth_devices[eth_dev_id],
>> >rx_queue_id, config);
>> >>  	} else {
>> >> -		ret = -ENOTSUP;
>> >> +		rxa_sw_event_vector_configure(rx_adapter,
>> >eth_dev_id,
>> >> +					      rx_queue_id, config);
>> >>  	}
>> >>
>> >>  	return ret;
>> >> @@ -2371,7 +2596,7 @@
>> >rte_event_eth_rx_adapter_vector_limits_get(
>> >>  		ret = dev->dev_ops-
>> >>eth_rx_adapter_vector_limits_get(
>> >>  			dev, &rte_eth_devices[eth_port_id], limits);
>> >>  	} else {
>> >> -		ret = -ENOTSUP;
>> >> +		ret = rxa_sw_vector_limits(limits);
>> >>  	}
>> >>
>> >>  	return ret;
>> >> diff --git a/lib/librte_eventdev/rte_eventdev.c
>> >b/lib/librte_eventdev/rte_eventdev.c
>> >> index f95edc075..254a31b1f 100644
>> >> --- a/lib/librte_eventdev/rte_eventdev.c
>> >> +++ b/lib/librte_eventdev/rte_eventdev.c
>> >> @@ -122,7 +122,11 @@
>rte_event_eth_rx_adapter_caps_get(uint8_t
>> >dev_id, uint16_t eth_port_id,
>> >>
>> >>  	if (caps == NULL)
>> >>  		return -EINVAL;
>> >> -	*caps = 0;
>> >> +
>> >> +	if (dev->dev_ops->eth_rx_adapter_caps_get == NULL)
>> >> +		*caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
>> >> +	else
>> >> +		*caps = 0;
>> >
>> >Any reason why we had to set default caps value? I am thinking if sw
>> >event device is used, it would set it anyways.
>> >
>>
>> There are multiple sw event devices which don't implement caps_get
>> function, this changes solves that.
>>
>> >>
>> >>  	return dev->dev_ops->eth_rx_adapter_caps_get ?
>> >>  				(*dev->dev_ops-
>> >>eth_rx_adapter_caps_get)(dev,
>> >> --
>> >> 2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [dpdk-dev v21.11] [PATCH v5 8/8] eventdev: simplify Rx adapter event vector config
  2021-03-26  7:09               ` Jayatheerthan, Jay
@ 2021-03-26  9:44                 ` Pavan Nikhilesh Bhagavatula
  0 siblings, 0 replies; 153+ messages in thread
From: Pavan Nikhilesh Bhagavatula @ 2021-03-26  9:44 UTC (permalink / raw)
  To: Jayatheerthan, Jay, Jerin Jacob Kollanukkaran, Carrillo, Erik G,
	Gujjar, Abhinandan S, McDaniel, Timothy, hemant.agrawal,
	Van Haaren, Harry, mattias.ronnblom, Ma, Liang J, Ray Kinsella,
	Neil Horman
  Cc: dev

>> -----Original Message-----
>> From: Pavan Nikhilesh Bhagavatula <pbhagavatula@marvell.com>
>> Sent: Thursday, March 25, 2021 7:25 PM
>> To: Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Jerin Jacob
>Kollanukkaran <jerinj@marvell.com>; Carrillo, Erik G
>> <erik.g.carrillo@intel.com>; Gujjar, Abhinandan S
><abhinandan.gujjar@intel.com>; McDaniel, Timothy
>> <timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
>Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
>> <mattias.ronnblom@ericsson.com>; Ma, Liang J
><liang.j.ma@intel.com>; Ray Kinsella <mdr@ashroe.eu>; Neil Horman
>> <nhorman@tuxdriver.com>
>> Cc: dev@dpdk.org
>> Subject: RE: [dpdk-dev v21.11] [PATCH v5 8/8] eventdev: simplify Rx
>adapter event vector config
>>
>> >> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
>> >> Sent: Wednesday, March 24, 2021 10:35 AM
>> >> To: jerinj@marvell.com; Jayatheerthan, Jay
>> ><jay.jayatheerthan@intel.com>; Carrillo, Erik G
>> ><erik.g.carrillo@intel.com>; Gujjar,
>> >> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy
>> ><timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
>> >> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
>> ><mattias.ronnblom@ericsson.com>; Ma, Liang J
>> >> <liang.j.ma@intel.com>; Ray Kinsella <mdr@ashroe.eu>; Neil
>Horman
>> ><nhorman@tuxdriver.com>
>> >> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
>> >> Subject: [dpdk-dev v21.11] [PATCH v5 8/8] eventdev: simplify Rx
>> >adapter event vector config
>> >>
>> >> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>> >>
>> >> Include vector configuration into the structure
>> >> ``rte_event_eth_rx_adapter_queue_conf`` used when configuring
>rest
>> >> of the Rx adapter ethernet device Rx queue parameters.
>> >> This simplifies event vector configuration as it avoids splitting
>> >> configuration per Rx queue.
>> >>
>> >> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
>> >> ---
>> >>  app/test-eventdev/test_pipeline_common.c      |  16 +-
>> >>  lib/librte_eventdev/eventdev_pmd.h            |  29 ---
>> >>  .../rte_event_eth_rx_adapter.c                | 168 ++++++------------
>> >>  .../rte_event_eth_rx_adapter.h                |  27 ---
>> >>  lib/librte_eventdev/version.map               |   1 -
>> >>  5 files changed, 57 insertions(+), 184 deletions(-)
>> >>
>> >> diff --git a/app/test-eventdev/test_pipeline_common.c
>b/app/test-
>> >eventdev/test_pipeline_common.c
>> >> index d5ef90500..76aee254b 100644
>> >> --- a/app/test-eventdev/test_pipeline_common.c
>> >> +++ b/app/test-eventdev/test_pipeline_common.c
>> >> @@ -331,7 +331,6 @@ pipeline_event_rx_adapter_setup(struct
>> >evt_options *opt, uint8_t stride,
>> >>  	uint16_t prod;
>> >>  	struct rte_mempool *vector_pool = NULL;
>> >>  	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
>> >> -	struct rte_event_eth_rx_adapter_event_vector_config
>> >vec_conf;
>> >>
>> >>  	memset(&queue_conf, 0,
>> >>  			sizeof(struct
>> >rte_event_eth_rx_adapter_queue_conf));
>> >> @@ -397,8 +396,12 @@ pipeline_event_rx_adapter_setup(struct
>> >evt_options *opt, uint8_t stride,
>> >>  			}
>> >>
>> >>  			if (cap &
>> >RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
>> >> +				queue_conf.vector_sz = opt-
>> >>vector_size;
>> >> +				queue_conf.vector_timeout_ns =
>> >> +					opt->vector_tmo_nsec;
>> >>  				queue_conf.rx_queue_flags |=
>> >>
>> >	RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
>> >> +				queue_conf.vector_mp = vector_pool;
>> >>  			} else {
>> >>  				evt_err("Rx adapter doesn't support
>> >event vector");
>> >>  				return -EINVAL;
>> >> @@ -418,17 +421,6 @@ pipeline_event_rx_adapter_setup(struct
>> >evt_options *opt, uint8_t stride,
>> >>  			return ret;
>> >>  		}
>> >>
>> >> -		if (opt->ena_vector) {
>> >> -			vec_conf.vector_sz = opt->vector_size;
>> >> -			vec_conf.vector_timeout_ns = opt-
>> >>vector_tmo_nsec;
>> >> -			vec_conf.vector_mp = vector_pool;
>> >> -			if
>> >(rte_event_eth_rx_adapter_queue_event_vector_config(
>> >> -				    prod, prod, -1, &vec_conf) < 0) {
>> >> -				evt_err("Failed to configure event
>> >vectorization for Rx adapter");
>> >> -				return -EINVAL;
>> >> -			}
>> >> -		}
>> >> -
>> >>  		if (!(cap &
>> >RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
>> >>  			uint32_t service_id = -1U;
>> >>
>> >> diff --git a/lib/librte_eventdev/eventdev_pmd.h
>> >b/lib/librte_eventdev/eventdev_pmd.h
>> >> index 0f724ac85..63b3bc4b5 100644
>> >> --- a/lib/librte_eventdev/eventdev_pmd.h
>> >> +++ b/lib/librte_eventdev/eventdev_pmd.h
>> >> @@ -667,32 +667,6 @@ typedef int
>> >(*eventdev_eth_rx_adapter_vector_limits_get_t)(
>> >>  	const struct rte_eventdev *dev, const struct rte_eth_dev
>> >*eth_dev,
>> >>  	struct rte_event_eth_rx_adapter_vector_limits *limits);
>> >>
>> >> -struct rte_event_eth_rx_adapter_event_vector_config;
>> >> -/**
>> >> - * Enable event vector on an given Rx queue of a ethernet devices
>> >belonging to
>> >> - * the Rx adapter.
>> >> - *
>> >> - * @param dev
>> >> - *   Event device pointer
>> >> - *
>> >> - * @param eth_dev
>> >> - *   Ethernet device pointer
>> >> - *
>> >> - * @param rx_queue_id
>> >> - *   The Rx queue identifier
>> >> - *
>> >> - * @param config
>> >> - *   Pointer to the event vector configuration structure.
>> >> - *
>> >> - * @return
>> >> - *   - 0: Success.
>> >> - *   - <0: Error code returned by the driver function.
>> >> - */
>> >> -typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
>> >> -	const struct rte_eventdev *dev, const struct rte_eth_dev
>> >*eth_dev,
>> >> -	int32_t rx_queue_id,
>> >> -	const struct rte_event_eth_rx_adapter_event_vector_config
>> >*config);
>> >> -
>> >>  typedef uint32_t rte_event_pmd_selftest_seqn_t;
>> >>  extern int rte_event_pmd_selftest_seqn_dynfield_offset;
>> >>
>> >> @@ -1118,9 +1092,6 @@ struct rte_eventdev_ops {
>> >>  	eventdev_eth_rx_adapter_vector_limits_get_t
>> >>  		eth_rx_adapter_vector_limits_get;
>> >>  	/**< Get event vector limits for the Rx adapter */
>> >> -	eventdev_eth_rx_adapter_event_vector_config_t
>> >> -		eth_rx_adapter_event_vector_config;
>> >> -	/**< Configure Rx adapter with event vector */
>> >>
>> >>  	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
>> >>  	/**< Get timer adapter capabilities */
>> >> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> >b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> >> index c71990078..a1990637f 100644
>> >> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> >> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> >> @@ -1882,6 +1882,25 @@ rxa_add_queue(struct
>> >rte_event_eth_rx_adapter *rx_adapter,
>> >>  	} else
>> >>  		qi_ev->flow_id = 0;
>> >>
>> >> +	if (conf->rx_queue_flags &
>> >> +	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
>> >> +		queue_info->ena_vector = 1;
>> >> +		qi_ev->event_type =
>> >RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
>> >> +		rxa_set_vector_data(queue_info, conf->vector_sz,
>> >> +				    conf->vector_timeout_ns, conf-
>> >>vector_mp,
>> >> +				    rx_queue_id, dev_info->dev->data-
>> >>port_id);
>> >> +		rx_adapter->ena_vector = 1;
>> >> +		rx_adapter->vector_tmo_ticks =
>> >> +			rx_adapter->vector_tmo_ticks
>> >> +				? RTE_MIN(queue_info->vector_data
>> >> +						  .vector_timeout_ticks,
>> >> +					  rx_adapter-
>> >>vector_tmo_ticks)
>> >> +				: queue_info-
>> >>vector_data.vector_timeout_ticks;
>> >> +		rx_adapter->vector_tmo_ticks <<= 1;
>> >
>> >Any reason why we left shift here ? Applicable in patch 4/8 as well.
>>
>> Just so that we have half the precision of the lowest timeout, helps to
>> Maintain accuracy.
>
>Maybe I am missing something here. For e.g. if the lowest timeout is 20
>ticks, would you want 10 or 40 ? Currently, its set to 40.

My bad this should be Rshift, I will fix it in 4/8 too.

>
>>
>> >
>> >> +		TAILQ_INIT(&rx_adapter->vector_list);
>> >
>> >Can doing TAILQ_INIT every time a queue is added cause existing
>> >elements to be wiped out ? Applicable in patch 4/8 as well.
>>
>> I don't think queues can be added when adapter is already started
>> rxa_sw_add Isn't thready safe.
>
>Yes, it takes lock before calling rxa_sa_add. For internal_port
>implementation, I don't see any lock being taken. Besides, since its
>adapter related, it would be more relevant in adapter create than in
>queue add.

Sure, I will move the init code to adapter create.

>
>>
>> >
>> >> +		rx_adapter->prev_expiry_ts = 0;
>> >
>> >Can setting this every time a queue is added affect existing queues
>> >created with vector support and passing traffic ? Applicable in patch
>4/8
>> >as well.
>>
>> Same as above.
>>
>
>Same reasoning as above.
>
>> >
>> >> +	}
>> >> +
>> >>  	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
>> >>  	if (rxa_polled_queue(dev_info, rx_queue_id)) {
>> >>  		rx_adapter->num_rx_polled += !pollq;
>> >> @@ -1907,44 +1926,6 @@ rxa_add_queue(struct
>> >rte_event_eth_rx_adapter *rx_adapter,
>> >>  	}
>> >>  }
>> >>
>> >> -static void
>> >> -rxa_sw_event_vector_configure(
>> >> -	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t
>> >eth_dev_id,
>> >> -	int rx_queue_id,
>> >> -	const struct rte_event_eth_rx_adapter_event_vector_config
>> >*config)
>> >> -{
>> >> -	struct eth_device_info *dev_info = &rx_adapter-
>> >>eth_devices[eth_dev_id];
>> >> -	struct eth_rx_queue_info *queue_info;
>> >> -	struct rte_event *qi_ev;
>> >> -
>> >> -	if (rx_queue_id == -1) {
>> >> -		uint16_t nb_rx_queues;
>> >> -		uint16_t i;
>> >> -
>> >> -		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
>> >> -		for (i = 0; i < nb_rx_queues; i++)
>> >> -			rxa_sw_event_vector_configure(rx_adapter,
>> >eth_dev_id, i,
>> >> -						      config);
>> >> -		return;
>> >> -	}
>> >> -
>> >> -	queue_info = &dev_info->rx_queue[rx_queue_id];
>> >> -	qi_ev = (struct rte_event *)&queue_info->event;
>> >> -	queue_info->ena_vector = 1;
>> >> -	qi_ev->event_type =
>> >RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
>> >> -	rxa_set_vector_data(queue_info, config->vector_sz,
>> >> -			    config->vector_timeout_ns, config-
>> >>vector_mp,
>> >> -			    rx_queue_id, dev_info->dev->data->port_id);
>> >> -	rx_adapter->ena_vector = 1;
>> >> -	rx_adapter->vector_tmo_ticks =
>> >> -		rx_adapter->vector_tmo_ticks ?
>> >> -			      RTE_MIN(config->vector_timeout_ns << 1,
>> >> -				      rx_adapter->vector_tmo_ticks) :
>> >> -			      config->vector_timeout_ns << 1;
>> >> -	rx_adapter->prev_expiry_ts = 0;
>> >> -	TAILQ_INIT(&rx_adapter->vector_list);
>> >> -}
>> >> -
>> >>  static int rxa_sw_add(struct rte_event_eth_rx_adapter
>*rx_adapter,
>> >>  		uint16_t eth_dev_id,
>> >>  		int rx_queue_id,
>> >> @@ -2258,6 +2239,7 @@
>> >rte_event_eth_rx_adapter_queue_add(uint8_t id,
>> >>  	struct rte_event_eth_rx_adapter *rx_adapter;
>> >>  	struct rte_eventdev *dev;
>> >>  	struct eth_device_info *dev_info;
>> >> +	struct rte_event_eth_rx_adapter_vector_limits limits;
>> >>
>> >>  	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -
>> >EINVAL);
>> >>  	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
>> >> @@ -2294,6 +2276,39 @@
>> >rte_event_eth_rx_adapter_queue_add(uint8_t id,
>> >>  		return -EINVAL;
>> >>  	}
>> >>
>> >> +	if (queue_conf->rx_queue_flags &
>> >> +	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
>> >
>> >Perhaps, you could move the previous if condition here and just
>check
>> >for (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)
>>
>> No, the above check is not a capability check, it is to check if
>application
>> requested to enable vectorization for this queue or not.
>
>In current code, same condition check (queue_conf->rx_queue_flags &
>RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR ) is done
>twice next to each other, which could be avoided.
>
>	if ((cap &
>RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
>---->	    (queue_conf->rx_queue_flags &
>	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
>		RTE_EDEV_LOG_ERR("Event vectorization is not
>supported,"
>				 " eth port: %" PRIu16 " adapter id: %"
>PRIu8,
>				 eth_dev_id, id);
>		return -EINVAL;
>	}
>
>---->	if (queue_conf->rx_queue_flags &
>	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
>	...
>	}
>

Ah, I see it now, will fix it in next version.

>>
>> >
>> >> +		ret = rte_event_eth_rx_adapter_vector_limits_get(
>> >> +			rx_adapter->eventdev_id, eth_dev_id, &limits);
>> >> +		if (ret < 0) {
>> >> +			RTE_EDEV_LOG_ERR("Failed to get event
>> >device vector limits,"
>> >> +					 " eth port: %" PRIu16
>> >> +					 " adapter id: %" PRIu8,
>> >> +					 eth_dev_id, id);
>> >> +			return -EINVAL;
>> >> +		}
>> >> +		if (queue_conf->vector_sz < limits.min_sz ||
>> >> +		    queue_conf->vector_sz > limits.max_sz ||
>> >> +		    queue_conf->vector_timeout_ns <
>> >limits.min_timeout_ns ||
>> >> +		    queue_conf->vector_timeout_ns >
>> >limits.max_timeout_ns ||
>> >> +		    queue_conf->vector_mp == NULL) {
>> >> +			RTE_EDEV_LOG_ERR("Invalid event vector
>> >configuration,"
>> >> +					 " eth port: %" PRIu16
>> >> +					 " adapter id: %" PRIu8,
>> >> +					 eth_dev_id, id);
>> >> +			return -EINVAL;
>> >> +		}
>> >> +		if (queue_conf->vector_mp->elt_size <
>> >> +		    (sizeof(struct rte_event_vector) +
>> >> +		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
>> >> +			RTE_EDEV_LOG_ERR("Invalid event vector
>> >configuration,"
>> >> +					 " eth port: %" PRIu16
>> >> +					 " adapter id: %" PRIu8,
>> >> +					 eth_dev_id, id);
>> >> +			return -EINVAL;
>> >> +		}
>> >> +	}
>> >> +
>> >>  	if ((cap &
>> >RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
>> >>  		(rx_queue_id != -1)) {
>> >>  		RTE_EDEV_LOG_ERR("Rx queues can only be connected
>> >to single "
>> >> @@ -2487,83 +2502,6 @@
>> >rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t
>eth_dev_id,
>> >>  	return ret;
>> >>  }
>> >>
>> >> -int
>> >> -rte_event_eth_rx_adapter_queue_event_vector_config(
>> >> -	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
>> >> -	struct rte_event_eth_rx_adapter_event_vector_config *config)
>> >> -{
>> >> -	struct rte_event_eth_rx_adapter_vector_limits limits;
>> >> -	struct rte_event_eth_rx_adapter *rx_adapter;
>> >> -	struct rte_eventdev *dev;
>> >> -	uint32_t cap;
>> >> -	int ret;
>> >> -
>> >> -	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -
>> >EINVAL);
>> >> -	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
>> >> -
>> >> -	rx_adapter = rxa_id_to_adapter(id);
>> >> -	if ((rx_adapter == NULL) || (config == NULL))
>> >> -		return -EINVAL;
>> >> -
>> >> -	dev = &rte_eventdevs[rx_adapter->eventdev_id];
>> >> -	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter-
>> >>eventdev_id,
>> >> -						eth_dev_id, &cap);
>> >> -	if (ret) {
>> >> -		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev
>> >%" PRIu8
>> >> -				 "eth port %" PRIu16,
>> >> -				 id, eth_dev_id);
>> >> -		return ret;
>> >> -	}
>> >> -
>> >> -	if (!(cap &
>> >RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
>> >> -		RTE_EDEV_LOG_ERR("Event vectorization is not
>> >supported,"
>> >> -				 " eth port: %" PRIu16 " adapter id: %"
>> >PRIu8,
>> >> -				 eth_dev_id, id);
>> >> -		return -EINVAL;
>> >> -	}
>> >> -
>> >> -	ret = rte_event_eth_rx_adapter_vector_limits_get(
>> >> -		rx_adapter->eventdev_id, eth_dev_id, &limits);
>> >> -	if (ret) {
>> >> -		RTE_EDEV_LOG_ERR("Failed to get vector limits edev
>> >%" PRIu8
>> >> -				 "eth port %" PRIu16,
>> >> -				 rx_adapter->eventdev_id, eth_dev_id);
>> >> -		return ret;
>> >> -	}
>> >> -
>> >> -	if (config->vector_sz < limits.min_sz ||
>> >> -	    config->vector_sz > limits.max_sz ||
>> >> -	    config->vector_timeout_ns < limits.min_timeout_ns ||
>> >> -	    config->vector_timeout_ns > limits.max_timeout_ns ||
>> >> -	    config->vector_mp == NULL) {
>> >> -		RTE_EDEV_LOG_ERR("Invalid event vector
>> >configuration,"
>> >> -				 " eth port: %" PRIu16 " adapter id: %"
>> >PRIu8,
>> >> -				 eth_dev_id, id);
>> >> -		return -EINVAL;
>> >> -	}
>> >> -	if (config->vector_mp->elt_size <
>> >> -	    (sizeof(struct rte_event_vector) +
>> >> -	     (sizeof(uintptr_t) * config->vector_sz))) {
>> >> -		RTE_EDEV_LOG_ERR("Invalid event vector
>> >configuration,"
>> >> -				 " eth port: %" PRIu16 " adapter id: %"
>> >PRIu8,
>> >> -				 eth_dev_id, id);
>> >> -		return -EINVAL;
>> >> -	}
>> >> -
>> >> -	if (cap &
>> >RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
>> >> -		RTE_FUNC_PTR_OR_ERR_RET(
>> >> -			*dev->dev_ops-
>> >>eth_rx_adapter_event_vector_config,
>> >> -			-ENOTSUP);
>> >> -		ret = dev->dev_ops-
>> >>eth_rx_adapter_event_vector_config(
>> >> -			dev, &rte_eth_devices[eth_dev_id],
>> >rx_queue_id, config);
>> >> -	} else {
>> >> -		rxa_sw_event_vector_configure(rx_adapter,
>> >eth_dev_id,
>> >> -					      rx_queue_id, config);
>> >> -	}
>> >> -
>> >> -	return ret;
>> >> -}
>> >> -
>> >>  int
>> >>  rte_event_eth_rx_adapter_vector_limits_get(
>> >>  	uint8_t dev_id, uint16_t eth_port_id,
>> >> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
>> >b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
>> >> index 7407cde00..3f8b36229 100644
>> >> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
>> >> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
>> >> @@ -171,9 +171,6 @@ struct
>rte_event_eth_rx_adapter_queue_conf
>> >{
>> >>  	 * The event adapter sets ev.event_type to
>> >RTE_EVENT_TYPE_ETHDEV in the
>> >>  	 * enqueued event.
>> >>  	 */
>> >> -};
>> >> -
>> >> -struct rte_event_eth_rx_adapter_event_vector_config {
>> >>  	uint16_t vector_sz;
>> >>  	/**<
>> >>  	 * Indicates the maximum number for mbufs to combine and
>> >form a vector.
>> >> @@ -548,30 +545,6 @@ int
>> >rte_event_eth_rx_adapter_vector_limits_get(
>> >>  	uint8_t dev_id, uint16_t eth_port_id,
>> >>  	struct rte_event_eth_rx_adapter_vector_limits *limits);
>> >>
>> >> -/**
>> >> - * Configure event vectorization for a given ethernet device
>queue,
>> >that has
>> >> - * been added to a event eth Rx adapter.
>> >> - *
>> >> - * @param id
>> >> - *  The identifier of the ethernet Rx event adapter.
>> >> - *
>> >> - * @param eth_dev_id
>> >> - *  The identifier of the ethernet device.
>> >> - *
>> >> - * @param rx_queue_id
>> >> - *  Ethernet device receive queue index.
>> >> - *  If rx_queue_id is -1, then all Rx queues configured for the
>ethernet
>> >device
>> >> - *  are configured with event vectorization.
>> >> - *
>> >> - * @return
>> >> - *  - 0: Success, Receive queue configured correctly.
>> >> - *  - <0: Error code on failure.
>> >> - */
>> >> -__rte_experimental
>> >> -int rte_event_eth_rx_adapter_queue_event_vector_config(
>> >> -	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
>> >> -	struct rte_event_eth_rx_adapter_event_vector_config
>> >*config);
>> >> -
>> >>  #ifdef __cplusplus
>> >>  }
>> >>  #endif
>> >> diff --git a/lib/librte_eventdev/version.map
>> >b/lib/librte_eventdev/version.map
>> >> index 902df0ae3..34c1c830e 100644
>> >> --- a/lib/librte_eventdev/version.map
>> >> +++ b/lib/librte_eventdev/version.map
>> >> @@ -142,7 +142,6 @@ EXPERIMENTAL {
>> >>  	#added in 21.05
>> >>  	rte_event_vector_pool_create;
>> >>  	rte_event_eth_rx_adapter_vector_limits_get;
>> >> -	rte_event_eth_rx_adapter_queue_event_vector_config;
>> >>  };
>> >>
>> >>  INTERNAL {
>> >> --
>> >> 2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v8 0/8] Introduce event vectorization
  2021-03-25 17:10           ` [dpdk-dev] [PATCH v7 0/8] Introduce event vectorization pbhagavatula
                               ` (7 preceding siblings ...)
  2021-03-25 17:10             ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v7 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
@ 2021-03-26 14:08             ` pbhagavatula
  2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 1/8] eventdev: introduce event vector capability pbhagavatula
                                 ` (8 more replies)
  8 siblings, 9 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-26 14:08 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

In traditional event programming model, events are identified by a
flow-id and a uintptr_t. The flow-id uniquely identifies a given event
and determines the order of scheduling based on schedule type, the
uintptr_t holds a single object.

Event devices also support burst mode with configurable dequeue depth,
i.e. each dequeue call would return multiple events and each event
might be at a different stage of the pipeline.
Having a burst of events belonging to different stages in a dequeue
burst is not only difficult to vectorize but also increases the scheduler
overhead and application overhead of pipelining events further.
Using event vectors we see a performance gain of ~742.3% as shown in [1].

By introducing event vectorization, each event will be capable of holding
multiple uintptr_t of the same flow thereby allowing applications
to vectorize their pipeline and reduce the complexity of pipelining
events across multiple stages. This also reduces the complexity of handling
enqueue and dequeue on an event device.

Since event devices are transparent to the events they are scheduling
so the event producers such as eth_rx_adapter, crypto_adapter , etc..
are responsible for vectorizing the buffers of the same flow into a single
event.

The series also breaks ABI in the patch [8/8] which is targetted to the
v21.11 release.

The dpdk-test-eventdev application has been updated with options to test
multiple vector sizes and timeouts.

[1]
As for performance improvement, with a ARM Cortex-A72 equivalent processer,
software event device (--vdev=event_sw0), single worker core, single stage
and using one service core for Rx adapter, Tx adapter, Scheduling.

Without this patchset applied:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
         --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
         --stlist=a --wlcores=20
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    5.071 mpps

With the patchset applied and Without event vectorization:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
         --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
         --stlist=a --wlcores=20
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    5.123 mpps

With event vectorization:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
        --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
        --stlist=a --wlcores=20 --enable_vector --nb_eth_queues 1
        --vector_size 256
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    42.715 mpps

Having dedicated service cores for each Rx queues and tweaking the vector,
dequeue burst size would further improve performance.

API usage is shown below:

Configuration:

	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

	vector_pool = rte_event_vector_pool_create("vector_pool",
			nb_elem, 0, vector_size, socket_id);

	rte_event_eth_rx_adapter_create(id, event_id, &adptr_conf);
	rte_event_eth_rx_adapter_queue_add(id, eth_id, -1, &queue_conf);
	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
		vec_conf.vector_sz = vector_size;
		vec_conf.vector_timeout_ns = vector_tmo_nsec;
		vec_conf.vector_mp = vector_pool;
		rte_event_eth_rx_adapter_queue_event_vector_config(id,
				eth_id, -1, &vec_conf);
	}

Fastpath:

	num = rte_event_dequeue_burst(event_id, port_id, &ev, 1, 0);
	if (!num)
		continue;

	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
		switch (ev.event_type) {
		case RTE_EVENT_TYPE_ETHDEV_VECTOR:
		case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
			struct rte_mbuf **mbufs;

			mbufs = ev.vector_ev->mbufs;
			for (i = 0; i < ev.vector_ev->nb_elem; i++)
				//Process mbufs.
			break;
		case ...
		}
	}
	...

v8 Changes:
- Fix incorrect shift for vector timeout interval.(Jay)
- Code reallocation.(Jay)

v7 Changes:
- More doxygen fixes.(Jay)
- Reduce code duplication in 4/8.(Jay)

v6 Changes:
- Make rte_errno sign consistant.(Jay)
- Gramatical and doxygen fixes. (Jay)

v5 Changes:
- Make `rte_event_vector_pool_create non-inline` to ease ABI stability.(Ray)
- Move `rte_event_eth_rx_adapter_queue_event_vector_config` and
  `rte_event_eth_rx_adapter_vector_limits_get` implementation to the patch
  where they are initially defined.(Ray)
- Multiple gramatical and style fixes.(Jerin)
- Add missing release notes.(Jerin)

v4 Changes:
- Fix missing event vector structure in event structure.(Jay)

v3 Changes:
- Fix unintended formatting changes.

v2 Changes:
- Multiple gramatical and style fixes.(Jerin)
- Add parameter to define vector size in power of 2. (Jerin)
- Redo patch series w/o breaking ABI till the last patch.(David)
- Add deprication notice to announce ABI break in 21.11.(David)
- Add vector limits validation to app/test-eventdev.

Pavan Nikhilesh (8):
  eventdev: introduce event vector capability
  eventdev: introduce event vector Rx capability
  eventdev: introduce event vector Tx capability
  eventdev: add Rx adapter event vector support
  eventdev: add Tx adapter event vector support
  app/eventdev: add event vector mode in pipeline test
  doc: announce event Rx adapter config changes
  eventdev: simplify Rx adapter event vector config

 app/test-eventdev/evt_common.h                |   4 +
 app/test-eventdev/evt_options.c               |  52 +++
 app/test-eventdev/evt_options.h               |   4 +
 app/test-eventdev/test_pipeline_atq.c         | 310 +++++++++++++++--
 app/test-eventdev/test_pipeline_common.c      | 105 +++++-
 app/test-eventdev/test_pipeline_common.h      |  18 +
 app/test-eventdev/test_pipeline_queue.c       | 320 ++++++++++++++++--
 .../prog_guide/event_ethernet_rx_adapter.rst  |  38 +++
 .../prog_guide/event_ethernet_tx_adapter.rst  |  12 +
 doc/guides/prog_guide/eventdev.rst            |  36 +-
 doc/guides/rel_notes/deprecation.rst          |   9 +
 doc/guides/rel_notes/release_21_05.rst        |   8 +
 doc/guides/tools/testeventdev.rst             |  45 ++-
 lib/librte_eventdev/eventdev_pmd.h            |  31 +-
 .../rte_event_eth_rx_adapter.c                | 318 ++++++++++++++++-
 .../rte_event_eth_rx_adapter.h                |  78 +++++
 .../rte_event_eth_tx_adapter.c                |  66 +++-
 lib/librte_eventdev/rte_eventdev.c            |  53 ++-
 lib/librte_eventdev/rte_eventdev.h            | 114 ++++++-
 lib/librte_eventdev/version.map               |   4 +
 20 files changed, 1534 insertions(+), 91 deletions(-)

--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v8 1/8] eventdev: introduce event vector capability
  2021-03-26 14:08             ` [dpdk-dev] [PATCH v8 0/8] Introduce event vectorization pbhagavatula
@ 2021-03-26 14:08               ` pbhagavatula
  2021-03-27 12:07                 ` Jayatheerthan, Jay
  2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 2/8] eventdev: introduce event vector Rx capability pbhagavatula
                                 ` (7 subsequent siblings)
  8 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-03-26 14:08 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce rte_event_vector datastructure which is capable of holding
multiple uintptr_t of the same flow thereby allowing applications
to vectorize their pipeline and reducing the complexity of pipelining
the events across multiple stages.
This approach also reduces the scheduling overhead on a event device.

Add a event vector mempool create handler to create mempools based on
the best mempool ops available on a given platform.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
Acked-by: Ray Kinsella <mdr@ashroe.eu>
---
 doc/guides/prog_guide/eventdev.rst     | 36 ++++++++++-
 doc/guides/rel_notes/release_21_05.rst |  8 +++
 lib/librte_eventdev/rte_eventdev.c     | 42 +++++++++++++
 lib/librte_eventdev/rte_eventdev.h     | 82 +++++++++++++++++++++++++-
 lib/librte_eventdev/version.map        |  3 +
 5 files changed, 168 insertions(+), 3 deletions(-)

diff --git a/doc/guides/prog_guide/eventdev.rst b/doc/guides/prog_guide/eventdev.rst
index ccde086f6..fda9c3743 100644
--- a/doc/guides/prog_guide/eventdev.rst
+++ b/doc/guides/prog_guide/eventdev.rst
@@ -63,13 +63,45 @@ the actual event being scheduled is. The payload is a union of the following:
 * ``uint64_t u64``
 * ``void *event_ptr``
 * ``struct rte_mbuf *mbuf``
+* ``struct rte_event_vector *vec``
 
-These three items in a union occupy the same 64 bits at the end of the rte_event
+These four items in a union occupy the same 64 bits at the end of the rte_event
 structure. The application can utilize the 64 bits directly by accessing the
-u64 variable, while the event_ptr and mbuf are provided as convenience
+u64 variable, while the event_ptr, mbuf, vec are provided as a convenience
 variables.  For example the mbuf pointer in the union can used to schedule a
 DPDK packet.
 
+Event Vector
+~~~~~~~~~~~~
+
+The rte_event_vector struct contains a vector of elements defined by the event
+type specified in the ``rte_event``. The event_vector structure contains the
+following data:
+
+* ``nb_elem`` - The number of elements held within the vector.
+
+Similar to ``rte_event`` the payload of event vector is also a union, allowing
+flexibility in what the actual vector is.
+
+* ``struct rte_mbuf *mbufs[0]`` - An array of mbufs.
+* ``void *ptrs[0]`` - An array of pointers.
+* ``uint64_t *u64s[0]`` - An array of uint64_t elements.
+
+The size of the event vector is related to the total number of elements it is
+configured to hold, this is achieved by making `rte_event_vector` a variable
+length structure.
+A helper function is provided to create a mempool that holds event vector, which
+takes name of the pool, total number of required ``rte_event_vector``,
+cache size, number of elements in each ``rte_event_vector`` and socket id.
+
+.. code-block:: c
+
+        rte_event_vector_pool_create("vector_pool", nb_event_vectors, cache_sz,
+                                     nb_elements_per_vector, socket_id);
+
+The function ``rte_event_vector_pool_create`` creates mempool with the best
+platform mempool ops.
+
 Queues
 ~~~~~~
 
diff --git a/doc/guides/rel_notes/release_21_05.rst b/doc/guides/rel_notes/release_21_05.rst
index 8a601e0a7..aeb56533a 100644
--- a/doc/guides/rel_notes/release_21_05.rst
+++ b/doc/guides/rel_notes/release_21_05.rst
@@ -106,6 +106,14 @@ New Features
   * Added support for periodic timer mode in eventdev timer adapter.
   * Added support for periodic timer mode in octeontx2 event device driver.
 
+* **Add Event device vector capability.**
+
+  * Added ``rte_event_vector`` data structure which is capable of holding
+    multiple ``uintptr_t`` of the same flow thereby allowing applications
+    to vectorize their pipelines and also reduce the complexity of pipelining
+    the events across multiple stages.
+  * This also reduces the scheduling overhead on a event device.
+
 
 Removed Items
 -------------
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index b57363f80..be0499c52 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -1266,6 +1266,48 @@ int rte_event_dev_selftest(uint8_t dev_id)
 	return -ENOTSUP;
 }
 
+struct rte_mempool *
+rte_event_vector_pool_create(const char *name, unsigned int n,
+			     unsigned int cache_size, uint16_t nb_elem,
+			     int socket_id)
+{
+	const char *mp_ops_name;
+	struct rte_mempool *mp;
+	unsigned int elt_sz;
+	int ret;
+
+	if (!nb_elem) {
+		RTE_LOG(ERR, EVENTDEV,
+			"Invalid number of elements=%d requested\n", nb_elem);
+		rte_errno = EINVAL;
+		return NULL;
+	}
+
+	elt_sz =
+		sizeof(struct rte_event_vector) + (nb_elem * sizeof(uintptr_t));
+	mp = rte_mempool_create_empty(name, n, elt_sz, cache_size, 0, socket_id,
+				      0);
+	if (mp == NULL)
+		return NULL;
+
+	mp_ops_name = rte_mbuf_best_mempool_ops();
+	ret = rte_mempool_set_ops_byname(mp, mp_ops_name, NULL);
+	if (ret != 0) {
+		RTE_LOG(ERR, EVENTDEV, "error setting mempool handler\n");
+		goto err;
+	}
+
+	ret = rte_mempool_populate_default(mp);
+	if (ret < 0)
+		goto err;
+
+	return mp;
+err:
+	rte_mempool_free(mp);
+	rte_errno = -ret;
+	return NULL;
+}
+
 int
 rte_event_dev_start(uint8_t dev_id)
 {
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index 9fc39e9ca..dee468ed0 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -212,8 +212,10 @@ extern "C" {
 
 #include <rte_common.h>
 #include <rte_config.h>
-#include <rte_memory.h>
 #include <rte_errno.h>
+#include <rte_mbuf_pool_ops.h>
+#include <rte_memory.h>
+#include <rte_mempool.h>
 
 #include "rte_eventdev_trace_fp.h"
 
@@ -913,6 +915,31 @@ rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
 int
 rte_event_dev_close(uint8_t dev_id);
 
+/**
+ * Event vector structure.
+ */
+struct rte_event_vector {
+	uint64_t nb_elem : 16;
+	/**< Number of elements in this event vector. */
+	uint64_t rsvd : 48;
+	/**< Reserved for future use */
+	uint64_t impl_opaque;
+	/**< Implementation specific opaque value.
+	 * An implementation may use this field to hold implementation specific
+	 * value to share between dequeue and enqueue operation.
+	 * The application should not modify this field.
+	 */
+	union {
+		struct rte_mbuf *mbufs[0];
+		void *ptrs[0];
+		uint64_t *u64s[0];
+	} __rte_aligned(16);
+	/**< Start of the vector array union. Depending upon the event type the
+	 * vector array can be an array of mbufs or pointers or opaque u64
+	 * values.
+	 */
+};
+
 /* Scheduler type definitions */
 #define RTE_SCHED_TYPE_ORDERED          0
 /**< Ordered scheduling
@@ -986,6 +1013,21 @@ rte_event_dev_close(uint8_t dev_id);
  */
 #define RTE_EVENT_TYPE_ETH_RX_ADAPTER   0x4
 /**< The event generated from event eth Rx adapter */
+#define RTE_EVENT_TYPE_VECTOR           0x8
+/**< Indicates that event is a vector.
+ * All vector event types should be a logical OR of EVENT_TYPE_VECTOR.
+ * This simplifies the pipeline design as one can split processing the events
+ * between vector events and normal event across event types.
+ * Example:
+ *	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
+ *		// Classify and handle vector event.
+ *	} else {
+ *		// Classify and handle event.
+ *	}
+ */
+#define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
+/**< The event vector generated from cpu for pipelining. */
+
 #define RTE_EVENT_TYPE_MAX              0x10
 /**< Maximum number of event types */
 
@@ -1108,6 +1150,8 @@ struct rte_event {
 		/**< Opaque event pointer */
 		struct rte_mbuf *mbuf;
 		/**< mbuf pointer if dequeued event is associated with mbuf */
+		struct rte_event_vector *vec;
+		/**< Event vector pointer. */
 	};
 };
 
@@ -2026,6 +2070,42 @@ rte_event_dev_xstats_reset(uint8_t dev_id,
  */
 int rte_event_dev_selftest(uint8_t dev_id);
 
+/**
+ * Get the memory required per event vector based on the number of elements per
+ * vector.
+ * This should be used to create the mempool that holds the event vectors.
+ *
+ * @param name
+ *   The name of the vector pool.
+ * @param n
+ *   The number of elements in the mbuf pool.
+ * @param cache_size
+ *   Size of the per-core object cache. See rte_mempool_create() for
+ *   details.
+ * @param nb_elem
+ *   The number of elements that a single event vector should be able to hold.
+ * @param socket_id
+ *   The socket identifier where the memory should be allocated. The
+ *   value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
+ *   reserved zone
+ *
+ * @return
+ *   The pointer to the newly allocated mempool, on success. NULL on error
+ *   with rte_errno set appropriately. Possible rte_errno values include:
+ *    - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
+ *    - E_RTE_SECONDARY - function was called from a secondary process instance
+ *    - EINVAL - cache size provided is too large, or priv_size is not aligned.
+ *    - ENOSPC - the maximum number of memzones has already been allocated
+ *    - EEXIST - a memzone with the same name already exists
+ *    - ENOMEM - no appropriate memory area found in which to create memzone
+ *    - ENAMETOOLONG - mempool name requested is too long.
+ */
+__rte_experimental
+struct rte_mempool *
+rte_event_vector_pool_create(const char *name, unsigned int n,
+			     unsigned int cache_size, uint16_t nb_elem,
+			     int socket_id);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index 3e5c09cfd..a070ef56e 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -138,6 +138,9 @@ EXPERIMENTAL {
 	__rte_eventdev_trace_port_setup;
 	# added in 20.11
 	rte_event_pmd_pci_probe_named;
+
+	#added in 21.05
+	rte_event_vector_pool_create;
 };
 
 INTERNAL {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v8 2/8] eventdev: introduce event vector Rx capability
  2021-03-26 14:08             ` [dpdk-dev] [PATCH v8 0/8] Introduce event vectorization pbhagavatula
  2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 1/8] eventdev: introduce event vector capability pbhagavatula
@ 2021-03-26 14:08               ` pbhagavatula
  2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 3/8] eventdev: introduce event vector Tx capability pbhagavatula
                                 ` (6 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-26 14:08 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce event ethernet Rx adapter event vector capability.

If an event eth Rx adapter has the capability of
RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR then a given Rx queue
can be configured to enable event vectorization by passing the
flag RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR to
rte_event_eth_rx_adapter_queue_conf::rx_queue_flags while configuring
Rx adapter through rte_event_eth_rx_adapter_queue_add().

The max vector size, vector timeout define the vector size and
mempool used for allocating vector event are configured through
rte_event_eth_rx_adapter_queue_add. The element size of the element
in the vector pool should be equal to
    sizeof(struct rte_event_vector) + (vector_sz * sizeof(uintptr_t))

Application can use `rte_event_vector_pool_create` to create the
vector mempool used for
rte_event_eth_rx_adapter_queue_conf::vector_mp.

The Rx adapter would be responsible for vectorizing the mbufs
based on the flow, the vector limits configured by the application
and add the vector event of mbufs to the event queue set via
rte_event_eth_rx_adapter_queue_conf::ev::queue_id.
It should also mark rte_event_vector::union_valid and fill
rte_event_vector::port, rte_event_vector::queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
Acked-by: Ray Kinsella <mdr@ashroe.eu>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
---
 .../prog_guide/event_ethernet_rx_adapter.rst  |  38 ++++++
 lib/librte_eventdev/eventdev_pmd.h            |  53 ++++++++
 .../rte_event_eth_rx_adapter.c                | 114 ++++++++++++++++++
 .../rte_event_eth_rx_adapter.h                | 105 ++++++++++++++++
 lib/librte_eventdev/rte_eventdev.h            |  30 ++++-
 lib/librte_eventdev/version.map               |   2 +
 6 files changed, 340 insertions(+), 2 deletions(-)

diff --git a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
index cb44ce0e4..5eefef355 100644
--- a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
@@ -186,3 +186,41 @@ the event buffer fill level is low. The
 ``rte_event_eth_rx_adapter_cb_register()`` function allow the application
 to register a callback that selects which packets to enqueue to the event
 device.
+
+Rx event vectorization
+~~~~~~~~~~~~~~~~~~~~~~
+
+The event devices, ethernet device pairs which support the capability
+``RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR`` can aggregate packets based on
+flow characteristics and generate a ``rte_event`` containing ``rte_event_vector``
+whose event type is either ``RTE_EVENT_TYPE_ETHDEV_VECTOR`` or
+``RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR``.
+The aggregation size and timeout are configurable at a queue level and the
+maximum, minimum vector sizes and timeouts vary based on the device capability
+and can be queried using ``rte_event_eth_rx_adapter_vector_limits_get``.
+The Rx adapter additionally might include useful data such as ethernet device
+port and queue identifier in the ``rte_event_vector::port`` and
+``rte_event_vector::queue`` and mark ``rte_event_vector::attr_valid`` as true.
+
+A loop processing ``rte_event_vector`` containing mbufs is shown below.
+
+.. code-block:: c
+
+        event = rte_event_dequeue_burst(event_dev, event_port, &event,
+                                        1, 0);
+        if (!event)
+                continue;
+
+        switch (ev.event_type) {
+        case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
+        case RTE_EVENT_TYPE_ETHDEV_VECTOR:
+                struct rte_mbufs **mbufs;
+
+                mbufs = (struct rte_mbufs **)ev[i].vec->mbufs;
+                for (i = 0; i < ev.vec->nb_elem; i++) {
+                        /* Process each mbuf. */
+                }
+        break;
+        case ...
+        ...
+        }
diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 7eb9a7739..9297f1433 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -645,6 +645,53 @@ typedef int (*eventdev_eth_rx_adapter_stats_reset)
  */
 typedef int (*eventdev_selftest)(void);
 
+struct rte_event_eth_rx_adapter_vector_limits;
+/**
+ * Get event vector limits for a given event, ethernet device pair.
+ *
+ * @param dev
+ *   Event device pointer
+ *
+ * @param eth_dev
+ *   Ethernet device pointer
+ *
+ * @param[out] limits
+ *   Pointer to the limits structure to be filled.
+ *
+ * @return
+ *   - 0: Success.
+ *   - <0: Error code returned by the driver function.
+ */
+typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
+	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
+	struct rte_event_eth_rx_adapter_vector_limits *limits);
+
+struct rte_event_eth_rx_adapter_event_vector_config;
+/**
+ * Enable event vector on an given Rx queue of a ethernet devices belonging to
+ * the Rx adapter.
+ *
+ * @param dev
+ *   Event device pointer
+ *
+ * @param eth_dev
+ *   Ethernet device pointer
+ *
+ * @param rx_queue_id
+ *   The Rx queue identifier
+ *
+ * @param config
+ *   Pointer to the event vector configuration structure.
+ *
+ * @return
+ *   - 0: Success.
+ *   - <0: Error code returned by the driver function.
+ */
+typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
+	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
+	int32_t rx_queue_id,
+	const struct rte_event_eth_rx_adapter_event_vector_config *config);
+
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;
 
@@ -1067,6 +1114,12 @@ struct rte_eventdev_ops {
 	/**< Get ethernet Rx stats */
 	eventdev_eth_rx_adapter_stats_reset eth_rx_adapter_stats_reset;
 	/**< Reset ethernet Rx stats */
+	eventdev_eth_rx_adapter_vector_limits_get_t
+		eth_rx_adapter_vector_limits_get;
+	/**< Get event vector limits for the Rx adapter */
+	eventdev_eth_rx_adapter_event_vector_config_t
+		eth_rx_adapter_event_vector_config;
+	/**< Configure Rx adapter with event vector */
 
 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index d8c635e99..ac8ba5bf0 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -2263,6 +2263,120 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }
 
+int
+rte_event_eth_rx_adapter_queue_event_vector_config(
+	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
+	struct rte_event_eth_rx_adapter_event_vector_config *config)
+{
+	struct rte_event_eth_rx_adapter_vector_limits limits;
+	struct rte_event_eth_rx_adapter *rx_adapter;
+	struct rte_eventdev *dev;
+	uint32_t cap;
+	int ret;
+
+	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
+
+	rx_adapter = rxa_id_to_adapter(id);
+	if ((rx_adapter == NULL) || (config == NULL))
+		return -EINVAL;
+
+	dev = &rte_eventdevs[rx_adapter->eventdev_id];
+	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
+						eth_dev_id, &cap);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
+				 "eth port %" PRIu16,
+				 id, eth_dev_id);
+		return ret;
+	}
+
+	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
+		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
+	ret = rte_event_eth_rx_adapter_vector_limits_get(
+		rx_adapter->eventdev_id, eth_dev_id, &limits);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
+				 "eth port %" PRIu16,
+				 rx_adapter->eventdev_id, eth_dev_id);
+		return ret;
+	}
+
+	if (config->vector_sz < limits.min_sz ||
+	    config->vector_sz > limits.max_sz ||
+	    config->vector_timeout_ns < limits.min_timeout_ns ||
+	    config->vector_timeout_ns > limits.max_timeout_ns ||
+	    config->vector_mp == NULL) {
+		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+	if (config->vector_mp->elt_size <
+	    (sizeof(struct rte_event_vector) +
+	     (sizeof(uintptr_t) * config->vector_sz))) {
+		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
+	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
+		RTE_FUNC_PTR_OR_ERR_RET(
+			*dev->dev_ops->eth_rx_adapter_event_vector_config,
+			-ENOTSUP);
+		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
+			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
+	} else {
+		ret = -ENOTSUP;
+	}
+
+	return ret;
+}
+
+int
+rte_event_eth_rx_adapter_vector_limits_get(
+	uint8_t dev_id, uint16_t eth_port_id,
+	struct rte_event_eth_rx_adapter_vector_limits *limits)
+{
+	struct rte_eventdev *dev;
+	uint32_t cap;
+	int ret;
+
+	RTE_EVENTDEV_VALID_DEVID_OR_ERR_RET(dev_id, -EINVAL);
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_port_id, -EINVAL);
+
+	if (limits == NULL)
+		return -EINVAL;
+
+	dev = &rte_eventdevs[dev_id];
+
+	ret = rte_event_eth_rx_adapter_caps_get(dev_id, eth_port_id, &cap);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
+				 "eth port %" PRIu16,
+				 dev_id, eth_port_id);
+		return ret;
+	}
+
+	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
+		RTE_FUNC_PTR_OR_ERR_RET(
+			*dev->dev_ops->eth_rx_adapter_vector_limits_get,
+			-ENOTSUP);
+		ret = dev->dev_ops->eth_rx_adapter_vector_limits_get(
+			dev, &rte_eth_devices[eth_port_id], limits);
+	} else {
+		ret = -ENOTSUP;
+	}
+
+	return ret;
+}
+
 int
 rte_event_eth_rx_adapter_start(uint8_t id)
 {
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
index 21bb1e54c..7407cde00 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
@@ -92,6 +92,10 @@ extern "C" {
 /**< This flag indicates the flow identifier is valid
  * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
  */
+#define RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR	0x2
+/**< This flag indicates that mbufs arriving on the queue need to be vectorized
+ * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+ */
 
 /**
  * Adapter configuration structure that the adapter configuration callback
@@ -169,6 +173,36 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 */
 };
 
+struct rte_event_eth_rx_adapter_event_vector_config {
+	uint16_t vector_sz;
+	/**<
+	 * Indicates the maximum number for mbufs to combine and form a vector.
+	 * Should be within
+	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_sz
+	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_sz
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+	 */
+	uint64_t vector_timeout_ns;
+	/**<
+	 * Indicates the maximum number of nanoseconds to wait for receiving
+	 * mbufs. Should be within vectorization limits of the
+	 * adapter
+	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_ns
+	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_ns
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+	 */
+	struct rte_mempool *vector_mp;
+	/**<
+	 * Indicates the mempool that should be used for allocating
+	 * rte_event_vector container.
+	 * Should be created by using `rte_event_vector_pool_create`.
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags.
+	 */
+};
+
 /**
  * A structure used to retrieve statistics for an eth rx adapter instance.
  */
@@ -199,6 +233,32 @@ struct rte_event_eth_rx_adapter_stats {
 	/**< Received packet count for interrupt mode Rx queues */
 };
 
+/**
+ * A structure used to retrieve eth rx adapter vector limits.
+ */
+struct rte_event_eth_rx_adapter_vector_limits {
+	uint16_t min_sz;
+	/**< Minimum vector limit configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_sz
+	 */
+	uint16_t max_sz;
+	/**< Maximum vector limit configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_sz
+	 */
+	uint8_t log2_sz;
+	/**< True if the size configured should be in log2.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_sz
+	 */
+	uint64_t min_timeout_ns;
+	/**< Minimum vector timeout configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_timeout_ns
+	 */
+	uint64_t max_timeout_ns;
+	/**< Maximum vector timeout configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_timeout_ns
+	 */
+};
+
 /**
  *
  * Callback function invoked by the SW adapter before it continues
@@ -467,6 +527,51 @@ int rte_event_eth_rx_adapter_cb_register(uint8_t id, uint16_t eth_dev_id,
 					 rte_event_eth_rx_adapter_cb_fn cb_fn,
 					 void *cb_arg);
 
+/**
+ * Retrieve vector limits for a given event dev and eth dev pair.
+ * @see rte_event_eth_rx_adapter_vector_limits
+ *
+ * @param dev_id
+ *  Event device identifier.
+ * @param eth_port_id
+ *  Port identifier of the ethernet device.
+ * @param [out] limits
+ *  A pointer to rte_event_eth_rx_adapter_vector_limits structure that has to
+ * be filled.
+ *
+ * @return
+ *  - 0: Success.
+ *  - <0: Error code on failure.
+ */
+__rte_experimental
+int rte_event_eth_rx_adapter_vector_limits_get(
+	uint8_t dev_id, uint16_t eth_port_id,
+	struct rte_event_eth_rx_adapter_vector_limits *limits);
+
+/**
+ * Configure event vectorization for a given ethernet device queue, that has
+ * been added to a event eth Rx adapter.
+ *
+ * @param id
+ *  The identifier of the ethernet Rx event adapter.
+ *
+ * @param eth_dev_id
+ *  The identifier of the ethernet device.
+ *
+ * @param rx_queue_id
+ *  Ethernet device receive queue index.
+ *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
+ *  are configured with event vectorization.
+ *
+ * @return
+ *  - 0: Success, Receive queue configured correctly.
+ *  - <0: Error code on failure.
+ */
+__rte_experimental
+int rte_event_eth_rx_adapter_queue_event_vector_config(
+	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
+	struct rte_event_eth_rx_adapter_event_vector_config *config);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index dee468ed0..936c97cc2 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -919,10 +919,28 @@ rte_event_dev_close(uint8_t dev_id);
  * Event vector structure.
  */
 struct rte_event_vector {
-	uint64_t nb_elem : 16;
+	uint16_t nb_elem;
 	/**< Number of elements in this event vector. */
-	uint64_t rsvd : 48;
+	uint16_t rsvd : 15;
 	/**< Reserved for future use */
+	uint16_t attr_valid : 1;
+	/**< Indicates that the below union attributes have valid information.
+	 */
+	union {
+		/* Used by Rx adapter.
+		 * Indicates that all the elements in this vector belong to the
+		 * same port and queue pair when originating from Rx adapter,
+		 * valid only when event type is ETHDEV_VECTOR or
+		 * ETH_RX_ADAPTER_VECTOR.
+		 */
+		struct {
+			uint16_t port;
+			/* Ethernet device port id. */
+			uint16_t queue;
+			/* Ethernet device queue id. */
+		};
+	};
+	/**< Union to hold common attributes of the vector array. */
 	uint64_t impl_opaque;
 	/**< Implementation specific opaque value.
 	 * An implementation may use this field to hold implementation specific
@@ -1025,8 +1043,14 @@ struct rte_event_vector {
  *		// Classify and handle event.
  *	}
  */
+#define RTE_EVENT_TYPE_ETHDEV_VECTOR                                           \
+	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETHDEV)
+/**< The event vector generated from ethdev subsystem */
 #define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
 /**< The event vector generated from cpu for pipelining. */
+#define RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR                                   \
+	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETH_RX_ADAPTER)
+/**< The event vector generated from eth Rx adapter. */
 
 #define RTE_EVENT_TYPE_MAX              0x10
 /**< Maximum number of event types */
@@ -1171,6 +1195,8 @@ struct rte_event {
  * @see struct rte_event_eth_rx_adapter_queue_conf::ev
  * @see struct rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
  */
+#define RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR	0x8
+/**< Adapter supports event vectorization per ethdev. */
 
 /**
  * Retrieve the event device's ethdev Rx adapter capabilities for the
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index a070ef56e..902df0ae3 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -141,6 +141,8 @@ EXPERIMENTAL {
 
 	#added in 21.05
 	rte_event_vector_pool_create;
+	rte_event_eth_rx_adapter_vector_limits_get;
+	rte_event_eth_rx_adapter_queue_event_vector_config;
 };
 
 INTERNAL {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v8 3/8] eventdev: introduce event vector Tx capability
  2021-03-26 14:08             ` [dpdk-dev] [PATCH v8 0/8] Introduce event vectorization pbhagavatula
  2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 1/8] eventdev: introduce event vector capability pbhagavatula
  2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 2/8] eventdev: introduce event vector Rx capability pbhagavatula
@ 2021-03-26 14:08               ` pbhagavatula
  2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 4/8] eventdev: add Rx adapter event vector support pbhagavatula
                                 ` (5 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-26 14:08 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce event vector transmit capability for event eth
tx adapter.

The capability indicates that the Tx adapter is capable of
transmitting event vectors.
When rte_event_vector::union_valid is set, the Tx adapter should
transmit all the packets to the rte_event_vector::port using the
rte_event_vector::queue.
If rte_event_vector::union_valid is not set then the Tx adapter
should peek into each mbuf to get the destination port and queue
pair.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
---
 doc/guides/prog_guide/event_ethernet_tx_adapter.rst | 12 ++++++++++++
 lib/librte_eventdev/rte_eventdev.h                  |  8 +++++++-
 2 files changed, 19 insertions(+), 1 deletion(-)

diff --git a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
index a8c13e136..f80d22621 100644
--- a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
@@ -164,3 +164,15 @@ The  ``rte_event_eth_tx_adapter_stats_get()`` function reports counters defined
 in struct ``rte_event_eth_tx_adapter_stats``. The counter values are the sum of
 the counts from the eventdev PMD callback if the callback is supported, and
 the counts maintained by the service function, if one exists.
+
+Tx event vectorization
+~~~~~~~~~~~~~~~~~~~~~~
+
+The event device, ethernet device pairs which support the capability
+``RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR`` can process event vector of mbufs.
+Additionally, application can provide a hint to the Tx adapter that all the
+mbufs are destined to the same ethernet port and queue by setting the bit
+``rte_event_vector::attr_valid`` and filling `rte_event_vector::port`` and
+``rte_event_vector::queue``.
+If ``rte_event_vector::attr_valid`` is not set then the Tx adapter should peek
+into each mbuf and transmit them to the requested ethernet port and queue pair.
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index 936c97cc2..5f1f544cc 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -927,11 +927,13 @@ struct rte_event_vector {
 	/**< Indicates that the below union attributes have valid information.
 	 */
 	union {
-		/* Used by Rx adapter.
+		/* Used by Rx/Tx adapter.
 		 * Indicates that all the elements in this vector belong to the
 		 * same port and queue pair when originating from Rx adapter,
 		 * valid only when event type is ETHDEV_VECTOR or
 		 * ETH_RX_ADAPTER_VECTOR.
+		 * Can also be used to indicate the Tx adapter the destination
+		 * port and queue of the mbufs in the vector
 		 */
 		struct {
 			uint16_t port;
@@ -1296,6 +1298,10 @@ rte_event_crypto_adapter_caps_get(uint8_t dev_id, uint8_t cdev_id,
 #define RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT	0x1
 /**< This flag is sent when the PMD supports a packet transmit callback
  */
+#define RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR	0x2
+/**< Indicates that the Tx adapter is capable of handling event vector of
+ * mbufs.
+ */
 
 /**
  * Retrieve the event device's eth Tx adapter capabilities
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v8 4/8] eventdev: add Rx adapter event vector support
  2021-03-26 14:08             ` [dpdk-dev] [PATCH v8 0/8] Introduce event vectorization pbhagavatula
                                 ` (2 preceding siblings ...)
  2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 3/8] eventdev: introduce event vector Tx capability pbhagavatula
@ 2021-03-26 14:08               ` pbhagavatula
  2021-03-28  8:18                 ` Jerin Jacob
  2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 5/8] eventdev: add Tx " pbhagavatula
                                 ` (4 subsequent siblings)
  8 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-03-26 14:08 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support for event eth Rx adapter, the implementation
creates vector flows based on port and queue identifier of the received
mbufs.
The flow id for SW Rx event vectorization will use 12-bits of queue
identifier and 8-bits port identifier when custom flow id is not set
for simplicity.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 lib/librte_eventdev/eventdev_pmd.h            |   7 +-
 .../rte_event_eth_rx_adapter.c                | 269 ++++++++++++++++--
 lib/librte_eventdev/rte_eventdev.c            |   6 +-
 3 files changed, 258 insertions(+), 24 deletions(-)

diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 9297f1433..0f724ac85 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -69,9 +69,10 @@ extern "C" {
 	} \
 } while (0)
 
-#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP \
-		((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) | \
-			(RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ))
+#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP                                        \
+	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |                     \
+	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) |                         \
+	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR))
 
 #define RTE_EVENT_CRYPTO_ADAPTER_SW_CAP \
 		RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index ac8ba5bf0..fba3b5ec3 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -26,6 +26,10 @@
 #define BATCH_SIZE		32
 #define BLOCK_CNT_THRESHOLD	10
 #define ETH_EVENT_BUFFER_SIZE	(4*BATCH_SIZE)
+#define MAX_VECTOR_SIZE		1024
+#define MIN_VECTOR_SIZE		4
+#define MAX_VECTOR_NS		1E9
+#define MIN_VECTOR_NS		1E5
 
 #define ETH_RX_ADAPTER_SERVICE_NAME_LEN	32
 #define ETH_RX_ADAPTER_MEM_NAME_LEN	32
@@ -59,6 +63,20 @@ struct eth_rx_poll_entry {
 	uint16_t eth_rx_qid;
 };
 
+struct eth_rx_vector_data {
+	TAILQ_ENTRY(eth_rx_vector_data) next;
+	uint16_t port;
+	uint16_t queue;
+	uint16_t max_vector_count;
+	uint64_t event;
+	uint64_t ts;
+	uint64_t vector_timeout_ticks;
+	struct rte_mempool *vector_pool;
+	struct rte_event_vector *vector_ev;
+} __rte_cache_aligned;
+
+TAILQ_HEAD(eth_rx_vector_data_list, eth_rx_vector_data);
+
 /* Instance per adapter */
 struct rte_eth_event_enqueue_buffer {
 	/* Count of events in this buffer */
@@ -92,6 +110,14 @@ struct rte_event_eth_rx_adapter {
 	uint32_t wrr_pos;
 	/* Event burst buffer */
 	struct rte_eth_event_enqueue_buffer event_enqueue_buffer;
+	/* Vector enable flag */
+	uint8_t ena_vector;
+	/* Timestamp of previous vector expiry list traversal */
+	uint64_t prev_expiry_ts;
+	/* Minimum ticks to wait before traversing expiry list */
+	uint64_t vector_tmo_ticks;
+	/* vector list */
+	struct eth_rx_vector_data_list vector_list;
 	/* Per adapter stats */
 	struct rte_event_eth_rx_adapter_stats stats;
 	/* Block count, counts up to BLOCK_CNT_THRESHOLD */
@@ -198,9 +224,11 @@ struct eth_device_info {
 struct eth_rx_queue_info {
 	int queue_enabled;	/* True if added */
 	int intr_enabled;
+	uint8_t ena_vector;
 	uint16_t wt;		/* Polling weight */
 	uint32_t flow_id_mask;	/* Set to ~0 if app provides flow id else 0 */
 	uint64_t event;
+	struct eth_rx_vector_data vector_data;
 };
 
 static struct rte_event_eth_rx_adapter **event_eth_rx_adapter;
@@ -722,6 +750,9 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
 	    &rx_adapter->event_enqueue_buffer;
 	struct rte_event_eth_rx_adapter_stats *stats = &rx_adapter->stats;
 
+	if (!buf->count)
+		return 0;
+
 	uint16_t n = rte_event_enqueue_new_burst(rx_adapter->eventdev_id,
 					rx_adapter->event_port_id,
 					buf->events,
@@ -742,6 +773,77 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
 	return n;
 }
 
+static inline void
+rxa_init_vector(struct rte_event_eth_rx_adapter *rx_adapter,
+		struct eth_rx_vector_data *vec)
+{
+	vec->vector_ev->nb_elem = 0;
+	vec->vector_ev->port = vec->port;
+	vec->vector_ev->queue = vec->queue;
+	vec->vector_ev->attr_valid = true;
+	TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
+}
+
+static inline uint16_t
+rxa_create_event_vector(struct rte_event_eth_rx_adapter *rx_adapter,
+			struct eth_rx_queue_info *queue_info,
+			struct rte_eth_event_enqueue_buffer *buf,
+			struct rte_mbuf **mbufs, uint16_t num)
+{
+	struct rte_event *ev = &buf->events[buf->count];
+	struct eth_rx_vector_data *vec;
+	uint16_t filled, space, sz;
+
+	filled = 0;
+	vec = &queue_info->vector_data;
+
+	if (vec->vector_ev == NULL) {
+		if (rte_mempool_get(vec->vector_pool,
+				    (void **)&vec->vector_ev) < 0) {
+			rte_pktmbuf_free_bulk(mbufs, num);
+			return 0;
+		}
+		rxa_init_vector(rx_adapter, vec);
+	}
+	while (num) {
+		if (vec->vector_ev->nb_elem == vec->max_vector_count) {
+			/* Event ready. */
+			ev->event = vec->event;
+			ev->vec = vec->vector_ev;
+			ev++;
+			filled++;
+			vec->vector_ev = NULL;
+			TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+			if (rte_mempool_get(vec->vector_pool,
+					    (void **)&vec->vector_ev) < 0) {
+				rte_pktmbuf_free_bulk(mbufs, num);
+				return 0;
+			}
+			rxa_init_vector(rx_adapter, vec);
+		}
+
+		space = vec->max_vector_count - vec->vector_ev->nb_elem;
+		sz = num > space ? space : num;
+		memcpy(vec->vector_ev->mbufs + vec->vector_ev->nb_elem, mbufs,
+		       sizeof(void *) * sz);
+		vec->vector_ev->nb_elem += sz;
+		num -= sz;
+		mbufs += sz;
+		vec->ts = rte_rdtsc();
+	}
+
+	if (vec->vector_ev->nb_elem == vec->max_vector_count) {
+		ev->event = vec->event;
+		ev->vec = vec->vector_ev;
+		ev++;
+		filled++;
+		vec->vector_ev = NULL;
+		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+	}
+
+	return filled;
+}
+
 static inline void
 rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
@@ -766,29 +868,33 @@ rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
 	uint16_t nb_cb;
 	uint16_t dropped;
 
-	/* 0xffff ffff if PKT_RX_RSS_HASH is set, otherwise 0 */
-	rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
-	do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
-
-	for (i = 0; i < num; i++) {
-		m = mbufs[i];
-
-		rss = do_rss ?
-			rxa_do_softrss(m, rx_adapter->rss_key_be) :
-			m->hash.rss;
-		ev->event = event;
-		ev->flow_id = (rss & ~flow_id_mask) |
-				(ev->flow_id & flow_id_mask);
-		ev->mbuf = m;
-		ev++;
+	if (!eth_rx_queue_info->ena_vector) {
+		/* 0xffff ffff if PKT_RX_RSS_HASH is set, otherwise 0 */
+		rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
+		do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
+		for (i = 0; i < num; i++) {
+			m = mbufs[i];
+
+			rss = do_rss ? rxa_do_softrss(m, rx_adapter->rss_key_be)
+				     : m->hash.rss;
+			ev->event = event;
+			ev->flow_id = (rss & ~flow_id_mask) |
+				      (ev->flow_id & flow_id_mask);
+			ev->mbuf = m;
+			ev++;
+		}
+	} else {
+		num = rxa_create_event_vector(rx_adapter, eth_rx_queue_info,
+					      buf, mbufs, num);
 	}
 
-	if (dev_info->cb_fn) {
+	if (num && dev_info->cb_fn) {
 
 		dropped = 0;
 		nb_cb = dev_info->cb_fn(eth_dev_id, rx_queue_id,
-					ETH_EVENT_BUFFER_SIZE, buf->count, ev,
-					num, dev_info->cb_arg, &dropped);
+					ETH_EVENT_BUFFER_SIZE, buf->count,
+					&buf->events[buf->count], num,
+					dev_info->cb_arg, &dropped);
 		if (unlikely(nb_cb > num))
 			RTE_EDEV_LOG_ERR("Rx CB returned %d (> %d) events",
 				nb_cb, num);
@@ -1124,6 +1230,30 @@ rxa_poll(struct rte_event_eth_rx_adapter *rx_adapter)
 	return nb_rx;
 }
 
+static void
+rxa_vector_expire(struct eth_rx_vector_data *vec, void *arg)
+{
+	struct rte_event_eth_rx_adapter *rx_adapter = arg;
+	struct rte_eth_event_enqueue_buffer *buf =
+		&rx_adapter->event_enqueue_buffer;
+	struct rte_event *ev;
+
+	if (buf->count)
+		rxa_flush_event_buffer(rx_adapter);
+
+	if (vec->vector_ev->nb_elem == 0)
+		return;
+	ev = &buf->events[buf->count];
+
+	/* Event ready. */
+	ev->event = vec->event;
+	ev->vec = vec->vector_ev;
+	buf->count++;
+
+	vec->vector_ev = NULL;
+	vec->ts = 0;
+}
+
 static int
 rxa_service_func(void *args)
 {
@@ -1137,6 +1267,24 @@ rxa_service_func(void *args)
 		return 0;
 	}
 
+	if (rx_adapter->ena_vector) {
+		if ((rte_rdtsc() - rx_adapter->prev_expiry_ts) >=
+		    rx_adapter->vector_tmo_ticks) {
+			struct eth_rx_vector_data *vec;
+
+			TAILQ_FOREACH(vec, &rx_adapter->vector_list, next) {
+				uint64_t elapsed_time = rte_rdtsc() - vec->ts;
+
+				if (elapsed_time >= vec->vector_timeout_ticks) {
+					rxa_vector_expire(vec, rx_adapter);
+					TAILQ_REMOVE(&rx_adapter->vector_list,
+						     vec, next);
+				}
+			}
+			rx_adapter->prev_expiry_ts = rte_rdtsc();
+		}
+	}
+
 	stats = &rx_adapter->stats;
 	stats->rx_packets += rxa_intr_ring_dequeue(rx_adapter);
 	stats->rx_packets += rxa_poll(rx_adapter);
@@ -1640,6 +1788,29 @@ rxa_update_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }
 
+static void
+rxa_set_vector_data(struct eth_rx_queue_info *queue_info, uint16_t vector_count,
+		    uint64_t vector_ns, struct rte_mempool *mp, int32_t qid,
+		    uint16_t port_id)
+{
+#define NSEC2TICK(__ns, __freq) (((__ns) * (__freq)) / 1E9)
+	struct eth_rx_vector_data *vector_data;
+	uint32_t flow_id;
+
+	vector_data = &queue_info->vector_data;
+	vector_data->max_vector_count = vector_count;
+	vector_data->port = port_id;
+	vector_data->queue = qid;
+	vector_data->vector_pool = mp;
+	vector_data->vector_timeout_ticks =
+		NSEC2TICK(vector_ns, rte_get_timer_hz());
+	vector_data->ts = 0;
+	flow_id = queue_info->event & 0xFFFFF;
+	flow_id =
+		flow_id == 0 ? (qid & 0xFFF) | (port_id & 0xFF) << 12 : flow_id;
+	vector_data->event = (queue_info->event & ~0xFFFFF) | flow_id;
+}
+
 static void
 rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
 	struct eth_device_info *dev_info,
@@ -1741,6 +1912,42 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }
 
+static void
+rxa_sw_event_vector_configure(
+	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
+	int rx_queue_id,
+	const struct rte_event_eth_rx_adapter_event_vector_config *config)
+{
+	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
+	struct eth_rx_queue_info *queue_info;
+	struct rte_event *qi_ev;
+
+	if (rx_queue_id == -1) {
+		uint16_t nb_rx_queues;
+		uint16_t i;
+
+		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
+		for (i = 0; i < nb_rx_queues; i++)
+			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
+						      config);
+		return;
+	}
+
+	queue_info = &dev_info->rx_queue[rx_queue_id];
+	qi_ev = (struct rte_event *)&queue_info->event;
+	queue_info->ena_vector = 1;
+	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+	rxa_set_vector_data(queue_info, config->vector_sz,
+			    config->vector_timeout_ns, config->vector_mp,
+			    rx_queue_id, dev_info->dev->data->port_id);
+	rx_adapter->ena_vector = 1;
+	rx_adapter->vector_tmo_ticks =
+		rx_adapter->vector_tmo_ticks ?
+			      RTE_MIN(config->vector_timeout_ns >> 1,
+				rx_adapter->vector_tmo_ticks) :
+			      config->vector_timeout_ns >> 1;
+}
+
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -1967,6 +2174,7 @@ rte_event_eth_rx_adapter_create_ext(uint8_t id, uint8_t dev_id,
 	rx_adapter->conf_cb = conf_cb;
 	rx_adapter->conf_arg = conf_arg;
 	rx_adapter->id = id;
+	TAILQ_INIT(&rx_adapter->vector_list);
 	strcpy(rx_adapter->mem_name, mem_name);
 	rx_adapter->eth_devices = rte_zmalloc_socket(rx_adapter->mem_name,
 					RTE_MAX_ETHPORTS *
@@ -2081,6 +2289,15 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}
 
+	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
+	    (queue_conf->rx_queue_flags &
+	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
+		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
 		(rx_queue_id != -1)) {
 		RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
@@ -2143,6 +2360,17 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	return 0;
 }
 
+static int
+rxa_sw_vector_limits(struct rte_event_eth_rx_adapter_vector_limits *limits)
+{
+	limits->max_sz = MAX_VECTOR_SIZE;
+	limits->min_sz = MIN_VECTOR_SIZE;
+	limits->max_timeout_ns = MAX_VECTOR_NS;
+	limits->min_timeout_ns = MIN_VECTOR_NS;
+
+	return 0;
+}
+
 int
 rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 				int32_t rx_queue_id)
@@ -2333,7 +2561,8 @@ rte_event_eth_rx_adapter_queue_event_vector_config(
 		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
 			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
 	} else {
-		ret = -ENOTSUP;
+		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
+					      rx_queue_id, config);
 	}
 
 	return ret;
@@ -2371,7 +2600,7 @@ rte_event_eth_rx_adapter_vector_limits_get(
 		ret = dev->dev_ops->eth_rx_adapter_vector_limits_get(
 			dev, &rte_eth_devices[eth_port_id], limits);
 	} else {
-		ret = -ENOTSUP;
+		ret = rxa_sw_vector_limits(limits);
 	}
 
 	return ret;
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index be0499c52..62824654b 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -122,7 +122,11 @@ rte_event_eth_rx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
 
 	if (caps == NULL)
 		return -EINVAL;
-	*caps = 0;
+
+	if (dev->dev_ops->eth_rx_adapter_caps_get == NULL)
+		*caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
+	else
+		*caps = 0;
 
 	return dev->dev_ops->eth_rx_adapter_caps_get ?
 				(*dev->dev_ops->eth_rx_adapter_caps_get)(dev,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v8 5/8] eventdev: add Tx adapter event vector support
  2021-03-26 14:08             ` [dpdk-dev] [PATCH v8 0/8] Introduce event vectorization pbhagavatula
                                 ` (3 preceding siblings ...)
  2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 4/8] eventdev: add Rx adapter event vector support pbhagavatula
@ 2021-03-26 14:08               ` pbhagavatula
  2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
                                 ` (3 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-26 14:08 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support for event eth Tx adapter, the implementation
receives events from the single linked queue and based on
rte_event_vector::attr_valid transmits the vector of mbufs to a given
port, queue pair.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
---
 .../rte_event_eth_tx_adapter.c                | 66 ++++++++++++++++---
 lib/librte_eventdev/rte_eventdev.c            |  5 +-
 2 files changed, 60 insertions(+), 11 deletions(-)

diff --git a/lib/librte_eventdev/rte_event_eth_tx_adapter.c b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
index 5b4c42dcf..db260bfb6 100644
--- a/lib/librte_eventdev/rte_event_eth_tx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
@@ -510,6 +510,47 @@ txa_service_buffer_retry(struct rte_mbuf **pkts, uint16_t unsent,
 	stats->tx_dropped += unsent - sent;
 }
 
+static uint16_t
+txa_process_event_vector(struct txa_service_data *txa,
+			 struct rte_event_vector *vec)
+{
+	struct txa_service_queue_info *tqi;
+	uint16_t port, queue, nb_tx = 0;
+	struct rte_mbuf **mbufs;
+	int i;
+
+	mbufs = (struct rte_mbuf **)vec->mbufs;
+	if (vec->attr_valid) {
+		port = vec->port;
+		queue = vec->queue;
+		tqi = txa_service_queue(txa, port, queue);
+		if (unlikely(tqi == NULL || !tqi->added)) {
+			rte_pktmbuf_free_bulk(mbufs, vec->nb_elem);
+			rte_mempool_put(rte_mempool_from_obj(vec), vec);
+			return 0;
+		}
+		for (i = 0; i < vec->nb_elem; i++) {
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
+						   mbufs[i]);
+		}
+	} else {
+		for (i = 0; i < vec->nb_elem; i++) {
+			port = mbufs[i]->port;
+			queue = rte_event_eth_tx_adapter_txq_get(mbufs[i]);
+			tqi = txa_service_queue(txa, port, queue);
+			if (unlikely(tqi == NULL || !tqi->added)) {
+				rte_pktmbuf_free(mbufs[i]);
+				continue;
+			}
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
+						   mbufs[i]);
+		}
+	}
+	rte_mempool_put(rte_mempool_from_obj(vec), vec);
+
+	return nb_tx;
+}
+
 static void
 txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
 	uint32_t n)
@@ -522,22 +563,27 @@ txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
 
 	nb_tx = 0;
 	for (i = 0; i < n; i++) {
-		struct rte_mbuf *m;
 		uint16_t port;
 		uint16_t queue;
 		struct txa_service_queue_info *tqi;
 
-		m = ev[i].mbuf;
-		port = m->port;
-		queue = rte_event_eth_tx_adapter_txq_get(m);
+		if (!(ev[i].event_type & RTE_EVENT_TYPE_VECTOR)) {
+			struct rte_mbuf *m;
 
-		tqi = txa_service_queue(txa, port, queue);
-		if (unlikely(tqi == NULL || !tqi->added)) {
-			rte_pktmbuf_free(m);
-			continue;
-		}
+			m = ev[i].mbuf;
+			port = m->port;
+			queue = rte_event_eth_tx_adapter_txq_get(m);
 
-		nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
+			tqi = txa_service_queue(txa, port, queue);
+			if (unlikely(tqi == NULL || !tqi->added)) {
+				rte_pktmbuf_free(m);
+				continue;
+			}
+
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
+		} else {
+			nb_tx += txa_process_event_vector(txa, ev[i].vec);
+		}
 	}
 
 	stats->tx_packets += nb_tx;
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index 62824654b..c9bb5d227 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -196,7 +196,10 @@ rte_event_eth_tx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
 	if (caps == NULL)
 		return -EINVAL;
 
-	*caps = 0;
+	if (dev->dev_ops->eth_tx_adapter_caps_get == NULL)
+		*caps = RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR;
+	else
+		*caps = 0;
 
 	return dev->dev_ops->eth_tx_adapter_caps_get ?
 			(*dev->dev_ops->eth_tx_adapter_caps_get)(dev,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v8 6/8] app/eventdev: add event vector mode in pipeline test
  2021-03-26 14:08             ` [dpdk-dev] [PATCH v8 0/8] Introduce event vectorization pbhagavatula
                                 ` (4 preceding siblings ...)
  2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 5/8] eventdev: add Tx " pbhagavatula
@ 2021-03-26 14:08               ` pbhagavatula
  2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 7/8] doc: announce event Rx adapter config changes pbhagavatula
                                 ` (2 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-26 14:08 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support in pipeline tests. By default this mode
is disabled, it can be enabled by using the option --enable_vector.
example:
	dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
	--nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
	--wlcores=20-23  --enable_vector

Additional options to configure vector size and vector timeout are
also implemented and can be used by specifying --vector_size and
--vector_tmo_ns

This patch also adds a new option to set the number of Rx queues
configured per event eth rx adapter.
example:
	dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
	--nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
	--wlcores=20-23  --nb_eth_queues 4

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
---
 app/test-eventdev/evt_common.h           |   4 +
 app/test-eventdev/evt_options.c          |  52 ++++
 app/test-eventdev/evt_options.h          |   4 +
 app/test-eventdev/test_pipeline_atq.c    | 310 ++++++++++++++++++++--
 app/test-eventdev/test_pipeline_common.c | 113 +++++++-
 app/test-eventdev/test_pipeline_common.h |  18 ++
 app/test-eventdev/test_pipeline_queue.c  | 320 +++++++++++++++++++++--
 doc/guides/tools/testeventdev.rst        |  45 +++-
 8 files changed, 811 insertions(+), 55 deletions(-)

diff --git a/app/test-eventdev/evt_common.h b/app/test-eventdev/evt_common.h
index a1da1cf11..0e228258e 100644
--- a/app/test-eventdev/evt_common.h
+++ b/app/test-eventdev/evt_common.h
@@ -58,16 +58,20 @@ struct evt_options {
 	uint8_t sched_type_list[EVT_MAX_STAGES];
 	uint16_t mbuf_sz;
 	uint16_t wkr_deq_dep;
+	uint16_t vector_size;
+	uint16_t eth_queues;
 	uint32_t nb_flows;
 	uint32_t tx_first;
 	uint32_t max_pkt_sz;
 	uint32_t deq_tmo_nsec;
 	uint32_t q_priority:1;
 	uint32_t fwd_latency:1;
+	uint32_t ena_vector : 1;
 	uint64_t nb_pkts;
 	uint64_t nb_timers;
 	uint64_t expiry_nsec;
 	uint64_t max_tmo_nsec;
+	uint64_t vector_tmo_nsec;
 	uint64_t timer_tick_nsec;
 	uint64_t optm_timer_tick_nsec;
 	enum evt_prod_type prod_type;
diff --git a/app/test-eventdev/evt_options.c b/app/test-eventdev/evt_options.c
index 0d04ea9f8..0d5540574 100644
--- a/app/test-eventdev/evt_options.c
+++ b/app/test-eventdev/evt_options.c
@@ -34,6 +34,9 @@ evt_options_default(struct evt_options *opt)
 	opt->max_tmo_nsec = 1E5;  /* 100000ns ~100us */
 	opt->expiry_nsec = 1E4;   /* 10000ns ~10us */
 	opt->prod_type = EVT_PROD_TYPE_SYNT;
+	opt->eth_queues = 1;
+	opt->vector_size = 64;
+	opt->vector_tmo_nsec = 100E3;
 }
 
 typedef int (*option_parser_t)(struct evt_options *opt,
@@ -257,6 +260,43 @@ evt_parse_max_pkt_sz(struct evt_options *opt, const char *arg)
 	return ret;
 }
 
+static int
+evt_parse_ena_vector(struct evt_options *opt, const char *arg __rte_unused)
+{
+	opt->ena_vector = 1;
+	return 0;
+}
+
+static int
+evt_parse_vector_size(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint16(&(opt->vector_size), arg);
+
+	return ret;
+}
+
+static int
+evt_parse_vector_tmo_ns(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint64(&(opt->vector_tmo_nsec), arg);
+
+	return ret;
+}
+
+static int
+evt_parse_eth_queues(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint16(&(opt->eth_queues), arg);
+
+	return ret;
+}
+
 static void
 usage(char *program)
 {
@@ -289,6 +329,10 @@ usage(char *program)
 		"\t--expiry_nsec      : event timer expiry ns.\n"
 		"\t--mbuf_sz          : packet mbuf size.\n"
 		"\t--max_pkt_sz       : max packet size.\n"
+		"\t--nb_eth_queues    : number of ethernet Rx queues.\n"
+		"\t--enable_vector    : enable event vectorization.\n"
+		"\t--vector_size      : Max vector size.\n"
+		"\t--vector_tmo_ns    : Max vector timeout in nanoseconds\n"
 		);
 	printf("available tests:\n");
 	evt_test_dump_names();
@@ -360,6 +404,10 @@ static struct option lgopts[] = {
 	{ EVT_EXPIRY_NSEC,         1, 0, 0 },
 	{ EVT_MBUF_SZ,             1, 0, 0 },
 	{ EVT_MAX_PKT_SZ,          1, 0, 0 },
+	{ EVT_NB_ETH_QUEUES,       1, 0, 0 },
+	{ EVT_ENA_VECTOR,          0, 0, 0 },
+	{ EVT_VECTOR_SZ,           1, 0, 0 },
+	{ EVT_VECTOR_TMO,          1, 0, 0 },
 	{ EVT_HELP,                0, 0, 0 },
 	{ NULL,                    0, 0, 0 }
 };
@@ -394,6 +442,10 @@ evt_opts_parse_long(int opt_idx, struct evt_options *opt)
 		{ EVT_EXPIRY_NSEC, evt_parse_expiry_nsec},
 		{ EVT_MBUF_SZ, evt_parse_mbuf_sz},
 		{ EVT_MAX_PKT_SZ, evt_parse_max_pkt_sz},
+		{ EVT_NB_ETH_QUEUES, evt_parse_eth_queues},
+		{ EVT_ENA_VECTOR, evt_parse_ena_vector},
+		{ EVT_VECTOR_SZ, evt_parse_vector_size},
+		{ EVT_VECTOR_TMO, evt_parse_vector_tmo_ns},
 	};
 
 	for (i = 0; i < RTE_DIM(parsermap); i++) {
diff --git a/app/test-eventdev/evt_options.h b/app/test-eventdev/evt_options.h
index 748e54fae..1cea2a3e1 100644
--- a/app/test-eventdev/evt_options.h
+++ b/app/test-eventdev/evt_options.h
@@ -42,6 +42,10 @@
 #define EVT_EXPIRY_NSEC          ("expiry_nsec")
 #define EVT_MBUF_SZ              ("mbuf_sz")
 #define EVT_MAX_PKT_SZ           ("max_pkt_sz")
+#define EVT_NB_ETH_QUEUES        ("nb_eth_queues")
+#define EVT_ENA_VECTOR           ("enable_vector")
+#define EVT_VECTOR_SZ            ("vector_size")
+#define EVT_VECTOR_TMO           ("vector_tmo_ns")
 #define EVT_HELP                 ("help")
 
 void evt_options_default(struct evt_options *opt);
diff --git a/app/test-eventdev/test_pipeline_atq.c b/app/test-eventdev/test_pipeline_atq.c
index 0872b25b5..84dd4f44e 100644
--- a/app/test-eventdev/test_pipeline_atq.c
+++ b/app/test-eventdev/test_pipeline_atq.c
@@ -15,6 +15,8 @@ pipeline_atq_nb_event_queues(struct evt_options *opt)
 	return rte_eth_dev_count_avail();
 }
 
+typedef int (*pipeline_atq_worker_t)(void *arg);
+
 static __rte_noinline int
 pipeline_atq_worker_single_stage_tx(void *arg)
 {
@@ -113,6 +115,112 @@ pipeline_atq_worker_single_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_atq_worker_single_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+		vector_sz = ev.vec->nb_elem;
+		pipeline_event_tx_vector(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = ev.vec->nb_elem;
+		ev.queue_id = tx_queue[ev.vec->port];
+		ev.vec->queue = 0;
+		pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			vector_sz += ev[i].vec->nb_elem;
+			ev[i].vec->queue = 0;
+		}
+
+		pipeline_event_tx_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			ev[i].queue_id = tx_queue[ev[i].vec->port];
+			ev[i].vec->queue = 0;
+			vector_sz += ev[i].vec->nb_elem;
+			pipeline_fwd_event_vector(&ev[i],
+						  RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
 static __rte_noinline int
 pipeline_atq_worker_multi_stage_tx(void *arg)
 {
@@ -245,6 +353,147 @@ pipeline_atq_worker_multi_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.sub_event_type % nb_stages;
+
+		if (cq_id == last_queue) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+			continue;
+		}
+
+		ev.sub_event_type++;
+		pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.sub_event_type % nb_stages;
+
+		if (cq_id == last_queue) {
+			ev.queue_id = tx_queue[ev.vec->port];
+			ev.vec->queue = 0;
+			vector_sz = ev.vec->nb_elem;
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			pipeline_event_enqueue(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.sub_event_type++;
+			pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+			pipeline_event_enqueue(dev, port, &ev);
+		}
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].sub_event_type % nb_stages;
+
+			if (cq_id == last_queue) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+				continue;
+			}
+
+			ev[i].sub_event_type++;
+			pipeline_fwd_event_vector(&ev[i],
+						  sched_type_list[cq_id]);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].sub_event_type % nb_stages;
+
+			if (cq_id == last_queue) {
+				vector_sz = ev[i].vec->nb_elem;
+				ev[i].queue_id = tx_queue[ev[i].vec->port];
+				ev[i].vec->queue = 0;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].sub_event_type++;
+				pipeline_fwd_event_vector(
+					&ev[i], sched_type_list[cq_id]);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
 static int
 worker_wrapper(void *arg)
 {
@@ -253,27 +502,36 @@ worker_wrapper(void *arg)
 	const bool burst = evt_has_burst_mode(w->dev_id);
 	const bool internal_port = w->t->internal_port;
 	const uint8_t nb_stages = opt->nb_stages;
-	RTE_SET_USED(opt);
+	/*vector/burst/internal_port*/
+	const pipeline_atq_worker_t
+	pipeline_atq_worker_single_stage[2][2][2] = {
+		[0][0][0] = pipeline_atq_worker_single_stage_fwd,
+		[0][0][1] = pipeline_atq_worker_single_stage_tx,
+		[0][1][0] = pipeline_atq_worker_single_stage_burst_fwd,
+		[0][1][1] = pipeline_atq_worker_single_stage_burst_tx,
+		[1][0][0] = pipeline_atq_worker_single_stage_fwd_vector,
+		[1][0][1] = pipeline_atq_worker_single_stage_tx_vector,
+		[1][1][0] = pipeline_atq_worker_single_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_atq_worker_single_stage_burst_tx_vector,
+	};
+	const pipeline_atq_worker_t
+	pipeline_atq_worker_multi_stage[2][2][2] = {
+		[0][0][0] = pipeline_atq_worker_multi_stage_fwd,
+		[0][0][1] = pipeline_atq_worker_multi_stage_tx,
+		[0][1][0] = pipeline_atq_worker_multi_stage_burst_fwd,
+		[0][1][1] = pipeline_atq_worker_multi_stage_burst_tx,
+		[1][0][0] = pipeline_atq_worker_multi_stage_fwd_vector,
+		[1][0][1] = pipeline_atq_worker_multi_stage_tx_vector,
+		[1][1][0] = pipeline_atq_worker_multi_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_atq_worker_multi_stage_burst_tx_vector,
+	};
 
-	if (nb_stages == 1) {
-		if (!burst && internal_port)
-			return pipeline_atq_worker_single_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_atq_worker_single_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_atq_worker_single_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_atq_worker_single_stage_burst_fwd(arg);
-	} else {
-		if (!burst && internal_port)
-			return pipeline_atq_worker_multi_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_atq_worker_multi_stage_fwd(arg);
-		if (burst && internal_port)
-			return pipeline_atq_worker_multi_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_atq_worker_multi_stage_burst_fwd(arg);
-	}
+	if (nb_stages == 1)
+		return (pipeline_atq_worker_single_stage[opt->ena_vector][burst]
+							[internal_port])(arg);
+	else
+		return (pipeline_atq_worker_multi_stage[opt->ena_vector][burst]
+						       [internal_port])(arg);
 
 	rte_panic("invalid worker\n");
 }
@@ -290,7 +548,7 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
 	int ret;
 	int nb_ports;
 	int nb_queues;
-	uint8_t queue;
+	uint8_t queue, is_prod;
 	uint8_t tx_evqueue_id[RTE_MAX_ETHPORTS];
 	uint8_t queue_arr[RTE_EVENT_MAX_QUEUES_PER_DEV];
 	uint8_t nb_worker_queues = 0;
@@ -330,15 +588,19 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
 		q_conf.event_queue_cfg = RTE_EVENT_QUEUE_CFG_ALL_TYPES;
 
 		if (!t->internal_port) {
+			is_prod = false;
 			RTE_ETH_FOREACH_DEV(prod) {
 				if (queue == tx_evqueue_id[prod]) {
 					q_conf.event_queue_cfg =
 						RTE_EVENT_QUEUE_CFG_SINGLE_LINK;
-				} else {
-					queue_arr[nb_worker_queues] = queue;
-					nb_worker_queues++;
+					is_prod = true;
+					break;
 				}
 			}
+			if (!is_prod) {
+				queue_arr[nb_worker_queues] = queue;
+				nb_worker_queues++;
+			}
 		}
 
 		ret = rte_event_queue_setup(opt->dev_id, queue, &q_conf);
diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index b47d76743..d5ef90500 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -36,6 +36,12 @@ pipeline_opt_dump(struct evt_options *opt, uint8_t nb_queues)
 	evt_dump_queue_priority(opt);
 	evt_dump_sched_type_list(opt);
 	evt_dump_producer_type(opt);
+	evt_dump("nb_eth_rx_queues", "%d", opt->eth_queues);
+	evt_dump("event_vector", "%d", opt->ena_vector);
+	if (opt->ena_vector) {
+		evt_dump("vector_size", "%d", opt->vector_size);
+		evt_dump("vector_tmo_ns", "%" PRIu64 "", opt->vector_tmo_nsec);
+	}
 }
 
 static inline uint64_t
@@ -163,7 +169,7 @@ pipeline_opt_check(struct evt_options *opt, uint64_t nb_queues)
 int
 pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 {
-	uint16_t i;
+	uint16_t i, j;
 	int ret;
 	uint8_t nb_queues = 1;
 	struct test_pipeline *t = evt_test_priv(test);
@@ -210,6 +216,16 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 		if (!(caps & RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT))
 			t->internal_port = 0;
 
+		ret = rte_event_eth_rx_adapter_caps_get(opt->dev_id, i, &caps);
+		if (ret != 0) {
+			evt_err("failed to get event tx adapter[%d] caps", i);
+			return ret;
+		}
+
+		if (!(caps & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT))
+			local_port_conf.rxmode.offloads |=
+				DEV_RX_OFFLOAD_RSS_HASH;
+
 		ret = rte_eth_dev_info_get(i, &dev_info);
 		if (ret != 0) {
 			evt_err("Error during getting device (port %u) info: %s\n",
@@ -236,19 +252,22 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 				local_port_conf.rx_adv_conf.rss_conf.rss_hf);
 		}
 
-		if (rte_eth_dev_configure(i, nb_queues, nb_queues,
-					&local_port_conf)
-				< 0) {
+		if (rte_eth_dev_configure(i, opt->eth_queues, nb_queues,
+					  &local_port_conf) < 0) {
 			evt_err("Failed to configure eth port [%d]", i);
 			return -EINVAL;
 		}
 
-		if (rte_eth_rx_queue_setup(i, 0, NB_RX_DESC,
-				rte_socket_id(), &rx_conf, t->pool) < 0) {
-			evt_err("Failed to setup eth port [%d] rx_queue: %d.",
+		for (j = 0; j < opt->eth_queues; j++) {
+			if (rte_eth_rx_queue_setup(i, j, NB_RX_DESC,
+						   rte_socket_id(), &rx_conf,
+						   t->pool) < 0) {
+				evt_err("Failed to setup eth port [%d] rx_queue: %d.",
 					i, 0);
-			return -EINVAL;
+				return -EINVAL;
+			}
 		}
+
 		if (rte_eth_tx_queue_setup(i, 0, NB_TX_DESC,
 					rte_socket_id(), NULL) < 0) {
 			evt_err("Failed to setup eth port [%d] tx_queue: %d.",
@@ -310,12 +329,27 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 {
 	int ret = 0;
 	uint16_t prod;
+	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
+	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;
 
 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
 	queue_conf.ev.sched_type = opt->sched_type_list[0];
+	if (opt->ena_vector) {
+		unsigned int nb_elem = (opt->pool_sz / opt->vector_size) << 1;
+
+		nb_elem = nb_elem ? nb_elem : 1;
+		vector_pool = rte_event_vector_pool_create(
+			"vector_pool", nb_elem, 0, opt->vector_size,
+			opt->socket_id);
+		if (vector_pool == NULL) {
+			evt_err("failed to create event vector pool");
+			return -ENOMEM;
+		}
+	}
 	RTE_ETH_FOREACH_DEV(prod) {
+		struct rte_event_eth_rx_adapter_vector_limits limits;
 		uint32_t cap;
 
 		ret = rte_event_eth_rx_adapter_caps_get(opt->dev_id,
@@ -326,6 +360,50 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 					opt->dev_id);
 			return ret;
 		}
+
+		if (opt->ena_vector) {
+			memset(&limits, 0, sizeof(limits));
+			ret = rte_event_eth_rx_adapter_vector_limits_get(
+				opt->dev_id, prod, &limits);
+			if (ret) {
+				evt_err("failed to get vector limits");
+				return ret;
+			}
+
+			if (opt->vector_size < limits.min_sz ||
+			    opt->vector_size > limits.max_sz) {
+				evt_err("Vector size [%d] not within limits max[%d] min[%d]",
+					opt->vector_size, limits.min_sz,
+					limits.max_sz);
+				return -EINVAL;
+			}
+
+			if (limits.log2_sz &&
+			    !rte_is_power_of_2(opt->vector_size)) {
+				evt_err("Vector size [%d] not power of 2",
+					opt->vector_size);
+				return -EINVAL;
+			}
+
+			if (opt->vector_tmo_nsec > limits.max_timeout_ns ||
+			    opt->vector_tmo_nsec < limits.min_timeout_ns) {
+				evt_err("Vector timeout [%" PRIu64
+					"] not within limits max[%" PRIu64
+					"] min[%" PRIu64 "]",
+					opt->vector_tmo_nsec,
+					limits.max_timeout_ns,
+					limits.min_timeout_ns);
+				return -EINVAL;
+			}
+
+			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.rx_queue_flags |=
+				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+			} else {
+				evt_err("Rx adapter doesn't support event vector");
+				return -EINVAL;
+			}
+		}
 		queue_conf.ev.queue_id = prod * stride;
 		ret = rte_event_eth_rx_adapter_create(prod, opt->dev_id,
 				&prod_conf);
@@ -340,6 +418,17 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}
 
+		if (opt->ena_vector) {
+			vec_conf.vector_sz = opt->vector_size;
+			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
+			vec_conf.vector_mp = vector_pool;
+			if (rte_event_eth_rx_adapter_queue_event_vector_config(
+				    prod, prod, -1, &vec_conf) < 0) {
+				evt_err("Failed to configure event vectorization for Rx adapter");
+				return -EINVAL;
+			}
+		}
+
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;
 
@@ -378,6 +467,14 @@ pipeline_event_tx_adapter_setup(struct evt_options *opt,
 			return ret;
 		}
 
+		if (opt->ena_vector) {
+			if (!(cap &
+			      RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR)) {
+				evt_err("Tx adapter doesn't support event vector");
+				return -EINVAL;
+			}
+		}
+
 		ret = rte_event_eth_tx_adapter_create(consm, opt->dev_id,
 				&port_conf);
 		if (ret) {
diff --git a/app/test-eventdev/test_pipeline_common.h b/app/test-eventdev/test_pipeline_common.h
index 6e73c6ab2..800a90616 100644
--- a/app/test-eventdev/test_pipeline_common.h
+++ b/app/test-eventdev/test_pipeline_common.h
@@ -101,6 +101,14 @@ pipeline_fwd_event(struct rte_event *ev, uint8_t sched)
 	ev->sched_type = sched;
 }
 
+static __rte_always_inline void
+pipeline_fwd_event_vector(struct rte_event *ev, uint8_t sched)
+{
+	ev->event_type = RTE_EVENT_TYPE_CPU_VECTOR;
+	ev->op = RTE_EVENT_OP_FORWARD;
+	ev->sched_type = sched;
+}
+
 static __rte_always_inline void
 pipeline_event_tx(const uint8_t dev, const uint8_t port,
 		struct rte_event * const ev)
@@ -110,6 +118,16 @@ pipeline_event_tx(const uint8_t dev, const uint8_t port,
 		rte_pause();
 }
 
+static __rte_always_inline void
+pipeline_event_tx_vector(const uint8_t dev, const uint8_t port,
+			 struct rte_event *const ev)
+{
+	ev->vec->queue = 0;
+
+	while (!rte_event_eth_tx_adapter_enqueue(dev, port, ev, 1, 0))
+		rte_pause();
+}
+
 static __rte_always_inline void
 pipeline_event_tx_burst(const uint8_t dev, const uint8_t port,
 		struct rte_event *ev, const uint16_t nb_rx)
diff --git a/app/test-eventdev/test_pipeline_queue.c b/app/test-eventdev/test_pipeline_queue.c
index 9a9febb19..f6cc3e358 100644
--- a/app/test-eventdev/test_pipeline_queue.c
+++ b/app/test-eventdev/test_pipeline_queue.c
@@ -15,6 +15,8 @@ pipeline_queue_nb_event_queues(struct evt_options *opt)
 	return (eth_count * opt->nb_stages) + eth_count;
 }
 
+typedef int (*pipeline_queue_worker_t)(void *arg);
+
 static __rte_noinline int
 pipeline_queue_worker_single_stage_tx(void *arg)
 {
@@ -126,6 +128,125 @@ pipeline_queue_worker_single_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_queue_worker_single_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		if (ev.sched_type == RTE_SCHED_TYPE_ATOMIC) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.queue_id++;
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			pipeline_event_enqueue(dev, port, &ev);
+		}
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		ev.queue_id = tx_queue[ev.vec->port];
+		ev.vec->queue = 0;
+		vector_sz = ev.vec->nb_elem;
+		pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			if (ev[i].sched_type == RTE_SCHED_TYPE_ATOMIC) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].queue_id++;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			ev[i].queue_id = tx_queue[ev[i].vec->port];
+			ev[i].vec->queue = 0;
+			vector_sz += ev[i].vec->nb_elem;
+			pipeline_fwd_event_vector(&ev[i],
+						  RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
 
 static __rte_noinline int
 pipeline_queue_worker_multi_stage_tx(void *arg)
@@ -267,6 +388,151 @@ pipeline_queue_worker_multi_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.queue_id % nb_stages;
+
+		if (ev.queue_id == tx_queue[ev.vec->port]) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+			continue;
+		}
+
+		ev.queue_id++;
+		pipeline_fwd_event_vector(&ev, cq_id != last_queue
+						       ? sched_type_list[cq_id]
+						       : RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.queue_id % nb_stages;
+
+		if (cq_id == last_queue) {
+			vector_sz = ev.vec->nb_elem;
+			ev.queue_id = tx_queue[ev.vec->port];
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.queue_id++;
+			pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+		}
+
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].queue_id % nb_stages;
+
+			if (ev[i].queue_id == tx_queue[ev[i].vec->port]) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+				continue;
+			}
+
+			ev[i].queue_id++;
+			pipeline_fwd_event_vector(
+				&ev[i], cq_id != last_queue
+						? sched_type_list[cq_id]
+						: RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].queue_id % nb_stages;
+
+			if (cq_id == last_queue) {
+				ev[i].queue_id = tx_queue[ev[i].vec->port];
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].queue_id++;
+				pipeline_fwd_event_vector(
+					&ev[i], sched_type_list[cq_id]);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
 static int
 worker_wrapper(void *arg)
 {
@@ -275,29 +541,39 @@ worker_wrapper(void *arg)
 	const bool burst = evt_has_burst_mode(w->dev_id);
 	const bool internal_port = w->t->internal_port;
 	const uint8_t nb_stages = opt->nb_stages;
-	RTE_SET_USED(opt);
-
-	if (nb_stages == 1) {
-		if (!burst && internal_port)
-			return pipeline_queue_worker_single_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_queue_worker_single_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_queue_worker_single_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_queue_worker_single_stage_burst_fwd(
-					arg);
-	} else {
-		if (!burst && internal_port)
-			return pipeline_queue_worker_multi_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_queue_worker_multi_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_queue_worker_multi_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_queue_worker_multi_stage_burst_fwd(arg);
+	/*vector/burst/internal_port*/
+	const pipeline_queue_worker_t
+	pipeline_queue_worker_single_stage[2][2][2] = {
+		[0][0][0] = pipeline_queue_worker_single_stage_fwd,
+		[0][0][1] = pipeline_queue_worker_single_stage_tx,
+		[0][1][0] = pipeline_queue_worker_single_stage_burst_fwd,
+		[0][1][1] = pipeline_queue_worker_single_stage_burst_tx,
+		[1][0][0] = pipeline_queue_worker_single_stage_fwd_vector,
+		[1][0][1] = pipeline_queue_worker_single_stage_tx_vector,
+		[1][1][0] = pipeline_queue_worker_single_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_queue_worker_single_stage_burst_tx_vector,
+	};
+	const pipeline_queue_worker_t
+	pipeline_queue_worker_multi_stage[2][2][2] = {
+		[0][0][0] = pipeline_queue_worker_multi_stage_fwd,
+		[0][0][1] = pipeline_queue_worker_multi_stage_tx,
+		[0][1][0] = pipeline_queue_worker_multi_stage_burst_fwd,
+		[0][1][1] = pipeline_queue_worker_multi_stage_burst_tx,
+		[1][0][0] = pipeline_queue_worker_multi_stage_fwd_vector,
+		[1][0][1] = pipeline_queue_worker_multi_stage_tx_vector,
+		[1][1][0] = pipeline_queue_worker_multi_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_queue_worker_multi_stage_burst_tx_vector,
+	};
+
+	if (nb_stages == 1)
+		return (pipeline_queue_worker_single_stage[opt->ena_vector]
+							  [burst]
+							  [internal_port])(arg);
+	else
+		return (pipeline_queue_worker_multi_stage[opt->ena_vector]
+							 [burst]
+							 [internal_port])(arg);
 
-	}
 	rte_panic("invalid worker\n");
 }
 
diff --git a/doc/guides/tools/testeventdev.rst b/doc/guides/tools/testeventdev.rst
index ad1788a3d..f252dc2c1 100644
--- a/doc/guides/tools/testeventdev.rst
+++ b/doc/guides/tools/testeventdev.rst
@@ -158,6 +158,26 @@ The following are the application command-line options:
        Set max packet mbuf size. Can be used configure Rx/Tx scatter gather.
        Only applicable for `pipeline_atq` and `pipeline_queue` tests.
 
+* ``--nb_eth_queues``
+
+       Configure multiple Rx queues per each ethernet port.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--enable_vector``
+
+       Enable event vector for Rx/Tx adapters.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--vector_size``
+
+       Vector size to configure for the Rx adapter.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--vector_tmo_ns``
+
+       Vector timeout nanoseconds to be configured for the Rx adapter.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
 
 Eventdev Tests
 --------------
@@ -607,6 +627,10 @@ Supported application command line options are following::
         --worker_deq_depth
         --prod_type_ethdev
         --deq_tmo_nsec
+        --nb_eth_queues
+        --enable_vector
+        --vector_size
+        --vector_tmo_ns
 
 
 .. Note::
@@ -623,6 +647,13 @@ Example command to run pipeline queue test:
     sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
         --test=pipeline_queue --wlcore=1 --prod_type_ethdev --stlist=a
 
+Example command to run pipeline atq test with vector events:
+
+.. code-block:: console
+
+    sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
+        --test=pipeline_queue --wlcore=1 --prod_type_ethdev --stlist=a \
+        --enable_vector  --vector_size 512
 
 PIPELINE_ATQ Test
 ~~~~~~~~~~~~~~~~~~~
@@ -699,6 +730,10 @@ Supported application command line options are following::
         --worker_deq_depth
         --prod_type_ethdev
         --deq_tmo_nsec
+        --nb_eth_queues
+        --enable_vector
+        --vector_size
+        --vector_tmo_ns
 
 
 .. Note::
@@ -708,9 +743,17 @@ Supported application command line options are following::
 Example
 ^^^^^^^
 
-Example command to run pipeline queue test:
+Example command to run pipeline atq test:
 
 .. code-block:: console
 
     sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
         --test=pipeline_atq --wlcore=1 --prod_type_ethdev --stlist=a
+
+Example command to run pipeline atq test with vector events:
+
+.. code-block:: console
+
+    sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
+        --test=pipeline_atq --wlcore=1 --prod_type_ethdev --stlist=a \
+        --enable_vector  --vector_size 512
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v8 7/8] doc: announce event Rx adapter config changes
  2021-03-26 14:08             ` [dpdk-dev] [PATCH v8 0/8] Introduce event vectorization pbhagavatula
                                 ` (5 preceding siblings ...)
  2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
@ 2021-03-26 14:08               ` pbhagavatula
  2021-03-26 14:43                 ` Jerin Jacob
  2021-03-26 14:08               ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v8 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
  2021-03-30  8:22               ` [dpdk-dev] [PATCH v9 0/8] Introduce event vectorization pbhagavatula
  8 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-03-26 14:08 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

The Rx adapter event vector configuration will be merged into
Rx adapter queue configuration to simplify enabling event
vectorization.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Ray Kinsella <mdr@ashroe.eu>
---
 doc/guides/rel_notes/deprecation.rst | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst
index 64629e064..ef6d1bb18 100644
--- a/doc/guides/rel_notes/deprecation.rst
+++ b/doc/guides/rel_notes/deprecation.rst
@@ -130,3 +130,12 @@ Deprecation Notices
 * cmdline: ``cmdline`` structure will be made opaque to hide platform-specific
   content. On Linux and FreeBSD, supported prior to DPDK 20.11,
   original structure will be kept until DPDK 21.11.
+
+* eventdev: The structure ``rte_event_eth_rx_adapter_queue_conf`` will be
+  extended to include ``rte_event_eth_rx_adapter_event_vector_config`` elements
+  and the function ``rte_event_eth_rx_adapter_queue_event_vector_config`` will
+  be removed in DPDK 21.11.
+
+  An application can enable event vectorization by passing the desired vector
+  values to the function ``rte_event_eth_rx_adapter_queue_add`` using
+  the structure ``rte_event_eth_rx_adapter_queue_add``.
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [dpdk-dev v21.11] [PATCH v8 8/8] eventdev: simplify Rx adapter event vector config
  2021-03-26 14:08             ` [dpdk-dev] [PATCH v8 0/8] Introduce event vectorization pbhagavatula
                                 ` (6 preceding siblings ...)
  2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 7/8] doc: announce event Rx adapter config changes pbhagavatula
@ 2021-03-26 14:08               ` pbhagavatula
  2021-03-30  8:22               ` [dpdk-dev] [PATCH v9 0/8] Introduce event vectorization pbhagavatula
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-26 14:08 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Include vector configuration into the structure
``rte_event_eth_rx_adapter_queue_conf`` used when configuring rest
of the Rx adapter ethernet device Rx queue parameters.
This simplifies event vector configuration as it avoids splitting
configuration per Rx queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 app/test-eventdev/test_pipeline_common.c      |  16 +-
 lib/librte_eventdev/eventdev_pmd.h            |  29 ---
 .../rte_event_eth_rx_adapter.c                | 179 ++++++------------
 .../rte_event_eth_rx_adapter.h                |  27 ---
 lib/librte_eventdev/version.map               |   1 -
 5 files changed, 63 insertions(+), 189 deletions(-)

diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index d5ef90500..76aee254b 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -331,7 +331,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 	uint16_t prod;
 	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
-	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
@@ -397,8 +396,12 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			}

 			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.vector_sz = opt->vector_size;
+				queue_conf.vector_timeout_ns =
+					opt->vector_tmo_nsec;
 				queue_conf.rx_queue_flags |=
 				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+				queue_conf.vector_mp = vector_pool;
 			} else {
 				evt_err("Rx adapter doesn't support event vector");
 				return -EINVAL;
@@ -418,17 +421,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}

-		if (opt->ena_vector) {
-			vec_conf.vector_sz = opt->vector_size;
-			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
-			vec_conf.vector_mp = vector_pool;
-			if (rte_event_eth_rx_adapter_queue_event_vector_config(
-				    prod, prod, -1, &vec_conf) < 0) {
-				evt_err("Failed to configure event vectorization for Rx adapter");
-				return -EINVAL;
-			}
-		}
-
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;

diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 0f724ac85..63b3bc4b5 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -667,32 +667,6 @@ typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
 	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-struct rte_event_eth_rx_adapter_event_vector_config;
-/**
- * Enable event vector on an given Rx queue of a ethernet devices belonging to
- * the Rx adapter.
- *
- * @param dev
- *   Event device pointer
- *
- * @param eth_dev
- *   Ethernet device pointer
- *
- * @param rx_queue_id
- *   The Rx queue identifier
- *
- * @param config
- *   Pointer to the event vector configuration structure.
- *
- * @return
- *   - 0: Success.
- *   - <0: Error code returned by the driver function.
- */
-typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
-	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
-	int32_t rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;

@@ -1118,9 +1092,6 @@ struct rte_eventdev_ops {
 	eventdev_eth_rx_adapter_vector_limits_get_t
 		eth_rx_adapter_vector_limits_get;
 	/**< Get event vector limits for the Rx adapter */
-	eventdev_eth_rx_adapter_event_vector_config_t
-		eth_rx_adapter_event_vector_config;
-	/**< Configure Rx adapter with event vector */

 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index fba3b5ec3..e4e1f3c7d 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -1887,6 +1887,24 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	} else
 		qi_ev->flow_id = 0;

+	if (conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+		queue_info->ena_vector = 1;
+		qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+		rxa_set_vector_data(queue_info, conf->vector_sz,
+				    conf->vector_timeout_ns, conf->vector_mp,
+				    rx_queue_id, dev_info->dev->data->port_id);
+		rx_adapter->ena_vector = 1;
+		rx_adapter->vector_tmo_ticks =
+			rx_adapter->vector_tmo_ticks ?
+				      RTE_MIN(queue_info->vector_data
+							.vector_timeout_ticks >>
+						1,
+					rx_adapter->vector_tmo_ticks) :
+				      queue_info->vector_data.vector_timeout_ticks >>
+					1;
+	}
+
 	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
 	if (rxa_polled_queue(dev_info, rx_queue_id)) {
 		rx_adapter->num_rx_polled += !pollq;
@@ -1912,42 +1930,6 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }

-static void
-rxa_sw_event_vector_configure(
-	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
-	int rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
-	struct eth_rx_queue_info *queue_info;
-	struct rte_event *qi_ev;
-
-	if (rx_queue_id == -1) {
-		uint16_t nb_rx_queues;
-		uint16_t i;
-
-		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
-		for (i = 0; i < nb_rx_queues; i++)
-			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
-						      config);
-		return;
-	}
-
-	queue_info = &dev_info->rx_queue[rx_queue_id];
-	qi_ev = (struct rte_event *)&queue_info->event;
-	queue_info->ena_vector = 1;
-	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
-	rxa_set_vector_data(queue_info, config->vector_sz,
-			    config->vector_timeout_ns, config->vector_mp,
-			    rx_queue_id, dev_info->dev->data->port_id);
-	rx_adapter->ena_vector = 1;
-	rx_adapter->vector_tmo_ticks =
-		rx_adapter->vector_tmo_ticks ?
-			      RTE_MIN(config->vector_timeout_ns >> 1,
-				rx_adapter->vector_tmo_ticks) :
-			      config->vector_timeout_ns >> 1;
-}
-
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -2262,6 +2244,7 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	struct rte_event_eth_rx_adapter *rx_adapter;
 	struct rte_eventdev *dev;
 	struct eth_device_info *dev_info;
+	struct rte_event_eth_rx_adapter_vector_limits limits;

 	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
 	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
@@ -2289,13 +2272,46 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}

-	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
-	    (queue_conf->rx_queue_flags &
-	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
-		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
+	if (queue_conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+
+		if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0) {
+			RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+
+		ret = rte_event_eth_rx_adapter_vector_limits_get(
+			rx_adapter->eventdev_id, eth_dev_id, &limits);
+		if (ret < 0) {
+			RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_sz < limits.min_sz ||
+		    queue_conf->vector_sz > limits.max_sz ||
+		    queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
+		    queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
+		    queue_conf->vector_mp == NULL) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_mp->elt_size <
+		    (sizeof(struct rte_event_vector) +
+		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
 	}

 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
@@ -2491,83 +2507,6 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }

-int
-rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct rte_event_eth_rx_adapter_vector_limits limits;
-	struct rte_event_eth_rx_adapter *rx_adapter;
-	struct rte_eventdev *dev;
-	uint32_t cap;
-	int ret;
-
-	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
-	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
-
-	rx_adapter = rxa_id_to_adapter(id);
-	if ((rx_adapter == NULL) || (config == NULL))
-		return -EINVAL;
-
-	dev = &rte_eventdevs[rx_adapter->eventdev_id];
-	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
-						eth_dev_id, &cap);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
-				 "eth port %" PRIu16,
-				 id, eth_dev_id);
-		return ret;
-	}
-
-	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
-		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	ret = rte_event_eth_rx_adapter_vector_limits_get(
-		rx_adapter->eventdev_id, eth_dev_id, &limits);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
-				 "eth port %" PRIu16,
-				 rx_adapter->eventdev_id, eth_dev_id);
-		return ret;
-	}
-
-	if (config->vector_sz < limits.min_sz ||
-	    config->vector_sz > limits.max_sz ||
-	    config->vector_timeout_ns < limits.min_timeout_ns ||
-	    config->vector_timeout_ns > limits.max_timeout_ns ||
-	    config->vector_mp == NULL) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-	if (config->vector_mp->elt_size <
-	    (sizeof(struct rte_event_vector) +
-	     (sizeof(uintptr_t) * config->vector_sz))) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
-		RTE_FUNC_PTR_OR_ERR_RET(
-			*dev->dev_ops->eth_rx_adapter_event_vector_config,
-			-ENOTSUP);
-		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
-			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
-	} else {
-		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
-					      rx_queue_id, config);
-	}
-
-	return ret;
-}
-
 int
 rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
index 7407cde00..3f8b36229 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
@@ -171,9 +171,6 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
 	 * enqueued event.
 	 */
-};
-
-struct rte_event_eth_rx_adapter_event_vector_config {
 	uint16_t vector_sz;
 	/**<
 	 * Indicates the maximum number for mbufs to combine and form a vector.
@@ -548,30 +545,6 @@ int rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-/**
- * Configure event vectorization for a given ethernet device queue, that has
- * been added to a event eth Rx adapter.
- *
- * @param id
- *  The identifier of the ethernet Rx event adapter.
- *
- * @param eth_dev_id
- *  The identifier of the ethernet device.
- *
- * @param rx_queue_id
- *  Ethernet device receive queue index.
- *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
- *  are configured with event vectorization.
- *
- * @return
- *  - 0: Success, Receive queue configured correctly.
- *  - <0: Error code on failure.
- */
-__rte_experimental
-int rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index 902df0ae3..34c1c830e 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -142,7 +142,6 @@ EXPERIMENTAL {
 	#added in 21.05
 	rte_event_vector_pool_create;
 	rte_event_eth_rx_adapter_vector_limits_get;
-	rte_event_eth_rx_adapter_queue_event_vector_config;
 };

 INTERNAL {
--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v8 7/8] doc: announce event Rx adapter config changes
  2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 7/8] doc: announce event Rx adapter config changes pbhagavatula
@ 2021-03-26 14:43                 ` Jerin Jacob
  2021-03-27 12:07                   ` Jayatheerthan, Jay
  0 siblings, 1 reply; 153+ messages in thread
From: Jerin Jacob @ 2021-03-26 14:43 UTC (permalink / raw)
  To: Pavan Nikhilesh
  Cc: Jerin Jacob, Jayatheerthan, Jay, Erik Gabriel Carrillo, Gujjar,
	Abhinandan S, McDaniel, Timothy, Hemant Agrawal, Van Haaren,
	Harry, Mattias Rönnblom, Liang Ma, Ray Kinsella,
	Neil Horman, dpdk-dev

On Fri, Mar 26, 2021 at 7:41 PM <pbhagavatula@marvell.com> wrote:
>
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>
> The Rx adapter event vector configuration will be merged into
> Rx adapter queue configuration to simplify enabling event
> vectorization.
>
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> Acked-by: Ray Kinsella <mdr@ashroe.eu>

Acked-by: Jerin Jacob <jerinj@marvell.com>

@Jayatheerthan, Jay

Any comment on this?


> ---
>  doc/guides/rel_notes/deprecation.rst | 9 +++++++++
>  1 file changed, 9 insertions(+)
>
> diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst
> index 64629e064..ef6d1bb18 100644
> --- a/doc/guides/rel_notes/deprecation.rst
> +++ b/doc/guides/rel_notes/deprecation.rst
> @@ -130,3 +130,12 @@ Deprecation Notices
>  * cmdline: ``cmdline`` structure will be made opaque to hide platform-specific
>    content. On Linux and FreeBSD, supported prior to DPDK 20.11,
>    original structure will be kept until DPDK 21.11.
> +
> +* eventdev: The structure ``rte_event_eth_rx_adapter_queue_conf`` will be
> +  extended to include ``rte_event_eth_rx_adapter_event_vector_config`` elements
> +  and the function ``rte_event_eth_rx_adapter_queue_event_vector_config`` will
> +  be removed in DPDK 21.11.
> +
> +  An application can enable event vectorization by passing the desired vector
> +  values to the function ``rte_event_eth_rx_adapter_queue_add`` using
> +  the structure ``rte_event_eth_rx_adapter_queue_add``.
> --
> 2.17.1
>

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v8 1/8] eventdev: introduce event vector capability
  2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 1/8] eventdev: introduce event vector capability pbhagavatula
@ 2021-03-27 12:07                 ` Jayatheerthan, Jay
  0 siblings, 0 replies; 153+ messages in thread
From: Jayatheerthan, Jay @ 2021-03-27 12:07 UTC (permalink / raw)
  To: pbhagavatula, jerinj, Carrillo, Erik G, Gujjar, Abhinandan S,
	McDaniel, Timothy, hemant.agrawal, Van Haaren, Harry,
	mattias.ronnblom, Ma, Liang J, Ray Kinsella, Neil Horman
  Cc: dev

> -----Original Message-----
> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
> Sent: Friday, March 26, 2021 7:39 PM
> To: jerinj@marvell.com; Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Carrillo, Erik G <erik.g.carrillo@intel.com>; Gujjar,
> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy <timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom <mattias.ronnblom@ericsson.com>; Ma, Liang J
> <liang.j.ma@intel.com>; Ray Kinsella <mdr@ashroe.eu>; Neil Horman <nhorman@tuxdriver.com>
> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
> Subject: [dpdk-dev] [PATCH v8 1/8] eventdev: introduce event vector capability
> 
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> Introduce rte_event_vector datastructure which is capable of holding
> multiple uintptr_t of the same flow thereby allowing applications
> to vectorize their pipeline and reducing the complexity of pipelining
> the events across multiple stages.
> This approach also reduces the scheduling overhead on a event device.
> 
> Add a event vector mempool create handler to create mempools based on
> the best mempool ops available on a given platform.
> 
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> Acked-by: Jerin Jacob <jerinj@marvell.com>
> Acked-by: Ray Kinsella <mdr@ashroe.eu>
> ---
>  doc/guides/prog_guide/eventdev.rst     | 36 ++++++++++-
>  doc/guides/rel_notes/release_21_05.rst |  8 +++
>  lib/librte_eventdev/rte_eventdev.c     | 42 +++++++++++++
>  lib/librte_eventdev/rte_eventdev.h     | 82 +++++++++++++++++++++++++-
>  lib/librte_eventdev/version.map        |  3 +
>  5 files changed, 168 insertions(+), 3 deletions(-)
> 
> diff --git a/doc/guides/prog_guide/eventdev.rst b/doc/guides/prog_guide/eventdev.rst
> index ccde086f6..fda9c3743 100644
> --- a/doc/guides/prog_guide/eventdev.rst
> +++ b/doc/guides/prog_guide/eventdev.rst
> @@ -63,13 +63,45 @@ the actual event being scheduled is. The payload is a union of the following:
>  * ``uint64_t u64``
>  * ``void *event_ptr``
>  * ``struct rte_mbuf *mbuf``
> +* ``struct rte_event_vector *vec``
> 
> -These three items in a union occupy the same 64 bits at the end of the rte_event
> +These four items in a union occupy the same 64 bits at the end of the rte_event
>  structure. The application can utilize the 64 bits directly by accessing the
> -u64 variable, while the event_ptr and mbuf are provided as convenience
> +u64 variable, while the event_ptr, mbuf, vec are provided as a convenience
>  variables.  For example the mbuf pointer in the union can used to schedule a
>  DPDK packet.
> 
> +Event Vector
> +~~~~~~~~~~~~
> +
> +The rte_event_vector struct contains a vector of elements defined by the event
> +type specified in the ``rte_event``. The event_vector structure contains the
> +following data:
> +
> +* ``nb_elem`` - The number of elements held within the vector.
> +
> +Similar to ``rte_event`` the payload of event vector is also a union, allowing
> +flexibility in what the actual vector is.
> +
> +* ``struct rte_mbuf *mbufs[0]`` - An array of mbufs.
> +* ``void *ptrs[0]`` - An array of pointers.
> +* ``uint64_t *u64s[0]`` - An array of uint64_t elements.
> +
> +The size of the event vector is related to the total number of elements it is
> +configured to hold, this is achieved by making `rte_event_vector` a variable
> +length structure.
> +A helper function is provided to create a mempool that holds event vector, which
> +takes name of the pool, total number of required ``rte_event_vector``,
> +cache size, number of elements in each ``rte_event_vector`` and socket id.
> +
> +.. code-block:: c
> +
> +        rte_event_vector_pool_create("vector_pool", nb_event_vectors, cache_sz,
> +                                     nb_elements_per_vector, socket_id);
> +
> +The function ``rte_event_vector_pool_create`` creates mempool with the best
> +platform mempool ops.
> +
>  Queues
>  ~~~~~~
> 
> diff --git a/doc/guides/rel_notes/release_21_05.rst b/doc/guides/rel_notes/release_21_05.rst
> index 8a601e0a7..aeb56533a 100644
> --- a/doc/guides/rel_notes/release_21_05.rst
> +++ b/doc/guides/rel_notes/release_21_05.rst
> @@ -106,6 +106,14 @@ New Features
>    * Added support for periodic timer mode in eventdev timer adapter.
>    * Added support for periodic timer mode in octeontx2 event device driver.
> 
> +* **Add Event device vector capability.**
> +
> +  * Added ``rte_event_vector`` data structure which is capable of holding
> +    multiple ``uintptr_t`` of the same flow thereby allowing applications
> +    to vectorize their pipelines and also reduce the complexity of pipelining
> +    the events across multiple stages.
> +  * This also reduces the scheduling overhead on a event device.
> +
> 
>  Removed Items
>  -------------
> diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
> index b57363f80..be0499c52 100644
> --- a/lib/librte_eventdev/rte_eventdev.c
> +++ b/lib/librte_eventdev/rte_eventdev.c
> @@ -1266,6 +1266,48 @@ int rte_event_dev_selftest(uint8_t dev_id)
>  	return -ENOTSUP;
>  }
> 
> +struct rte_mempool *
> +rte_event_vector_pool_create(const char *name, unsigned int n,
> +			     unsigned int cache_size, uint16_t nb_elem,
> +			     int socket_id)
> +{
> +	const char *mp_ops_name;
> +	struct rte_mempool *mp;
> +	unsigned int elt_sz;
> +	int ret;
> +
> +	if (!nb_elem) {
> +		RTE_LOG(ERR, EVENTDEV,
> +			"Invalid number of elements=%d requested\n", nb_elem);
> +		rte_errno = EINVAL;
> +		return NULL;
> +	}
> +
> +	elt_sz =
> +		sizeof(struct rte_event_vector) + (nb_elem * sizeof(uintptr_t));
> +	mp = rte_mempool_create_empty(name, n, elt_sz, cache_size, 0, socket_id,
> +				      0);
> +	if (mp == NULL)
> +		return NULL;
> +
> +	mp_ops_name = rte_mbuf_best_mempool_ops();
> +	ret = rte_mempool_set_ops_byname(mp, mp_ops_name, NULL);
> +	if (ret != 0) {
> +		RTE_LOG(ERR, EVENTDEV, "error setting mempool handler\n");
> +		goto err;
> +	}
> +
> +	ret = rte_mempool_populate_default(mp);
> +	if (ret < 0)
> +		goto err;
> +
> +	return mp;
> +err:
> +	rte_mempool_free(mp);
> +	rte_errno = -ret;
> +	return NULL;
> +}
> +
>  int
>  rte_event_dev_start(uint8_t dev_id)
>  {
> diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
> index 9fc39e9ca..dee468ed0 100644
> --- a/lib/librte_eventdev/rte_eventdev.h
> +++ b/lib/librte_eventdev/rte_eventdev.h
> @@ -212,8 +212,10 @@ extern "C" {
> 
>  #include <rte_common.h>
>  #include <rte_config.h>
> -#include <rte_memory.h>
>  #include <rte_errno.h>
> +#include <rte_mbuf_pool_ops.h>
> +#include <rte_memory.h>
> +#include <rte_mempool.h>
> 
>  #include "rte_eventdev_trace_fp.h"
> 
> @@ -913,6 +915,31 @@ rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
>  int
>  rte_event_dev_close(uint8_t dev_id);
> 
> +/**
> + * Event vector structure.
> + */
> +struct rte_event_vector {
> +	uint64_t nb_elem : 16;
> +	/**< Number of elements in this event vector. */
> +	uint64_t rsvd : 48;
> +	/**< Reserved for future use */
> +	uint64_t impl_opaque;
> +	/**< Implementation specific opaque value.
> +	 * An implementation may use this field to hold implementation specific
> +	 * value to share between dequeue and enqueue operation.
> +	 * The application should not modify this field.
> +	 */
> +	union {
> +		struct rte_mbuf *mbufs[0];
> +		void *ptrs[0];
> +		uint64_t *u64s[0];
> +	} __rte_aligned(16);
> +	/**< Start of the vector array union. Depending upon the event type the
> +	 * vector array can be an array of mbufs or pointers or opaque u64
> +	 * values.
> +	 */
> +};
> +
>  /* Scheduler type definitions */
>  #define RTE_SCHED_TYPE_ORDERED          0
>  /**< Ordered scheduling
> @@ -986,6 +1013,21 @@ rte_event_dev_close(uint8_t dev_id);
>   */
>  #define RTE_EVENT_TYPE_ETH_RX_ADAPTER   0x4
>  /**< The event generated from event eth Rx adapter */
> +#define RTE_EVENT_TYPE_VECTOR           0x8
> +/**< Indicates that event is a vector.
> + * All vector event types should be a logical OR of EVENT_TYPE_VECTOR.
> + * This simplifies the pipeline design as one can split processing the events
> + * between vector events and normal event across event types.
> + * Example:
> + *	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
> + *		// Classify and handle vector event.
> + *	} else {
> + *		// Classify and handle event.
> + *	}
> + */
> +#define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
> +/**< The event vector generated from cpu for pipelining. */
> +
>  #define RTE_EVENT_TYPE_MAX              0x10
>  /**< Maximum number of event types */
> 
> @@ -1108,6 +1150,8 @@ struct rte_event {
>  		/**< Opaque event pointer */
>  		struct rte_mbuf *mbuf;
>  		/**< mbuf pointer if dequeued event is associated with mbuf */
> +		struct rte_event_vector *vec;
> +		/**< Event vector pointer. */
>  	};
>  };
> 
> @@ -2026,6 +2070,42 @@ rte_event_dev_xstats_reset(uint8_t dev_id,
>   */
>  int rte_event_dev_selftest(uint8_t dev_id);
> 
> +/**
> + * Get the memory required per event vector based on the number of elements per
> + * vector.
> + * This should be used to create the mempool that holds the event vectors.
> + *
> + * @param name
> + *   The name of the vector pool.
> + * @param n
> + *   The number of elements in the mbuf pool.
> + * @param cache_size
> + *   Size of the per-core object cache. See rte_mempool_create() for
> + *   details.
> + * @param nb_elem
> + *   The number of elements that a single event vector should be able to hold.
> + * @param socket_id
> + *   The socket identifier where the memory should be allocated. The
> + *   value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
> + *   reserved zone
> + *
> + * @return
> + *   The pointer to the newly allocated mempool, on success. NULL on error
> + *   with rte_errno set appropriately. Possible rte_errno values include:
> + *    - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
> + *    - E_RTE_SECONDARY - function was called from a secondary process instance
> + *    - EINVAL - cache size provided is too large, or priv_size is not aligned.
> + *    - ENOSPC - the maximum number of memzones has already been allocated
> + *    - EEXIST - a memzone with the same name already exists
> + *    - ENOMEM - no appropriate memory area found in which to create memzone
> + *    - ENAMETOOLONG - mempool name requested is too long.
> + */
> +__rte_experimental
> +struct rte_mempool *
> +rte_event_vector_pool_create(const char *name, unsigned int n,
> +			     unsigned int cache_size, uint16_t nb_elem,
> +			     int socket_id);
> +
>  #ifdef __cplusplus
>  }
>  #endif
> diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
> index 3e5c09cfd..a070ef56e 100644
> --- a/lib/librte_eventdev/version.map
> +++ b/lib/librte_eventdev/version.map
> @@ -138,6 +138,9 @@ EXPERIMENTAL {
>  	__rte_eventdev_trace_port_setup;
>  	# added in 20.11
>  	rte_event_pmd_pci_probe_named;
> +
> +	#added in 21.05
> +	rte_event_vector_pool_create;
>  };
> 
>  INTERNAL {
> --
> 2.17.1

Looks good.

Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v8 7/8] doc: announce event Rx adapter config changes
  2021-03-26 14:43                 ` Jerin Jacob
@ 2021-03-27 12:07                   ` Jayatheerthan, Jay
  0 siblings, 0 replies; 153+ messages in thread
From: Jayatheerthan, Jay @ 2021-03-27 12:07 UTC (permalink / raw)
  To: Jerin Jacob, Pavan Nikhilesh
  Cc: Jerin Jacob, Carrillo, Erik G, Gujjar, Abhinandan S, McDaniel,
	Timothy, Hemant Agrawal, Van Haaren, Harry, mattias.ronnblom, Ma,
	Liang J, Ray Kinsella, Neil Horman, dpdk-dev

> -----Original Message-----
> From: Jerin Jacob <jerinjacobk@gmail.com>
> Sent: Friday, March 26, 2021 8:13 PM
> To: Pavan Nikhilesh <pbhagavatula@marvell.com>
> Cc: Jerin Jacob <jerinj@marvell.com>; Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Carrillo, Erik G <erik.g.carrillo@intel.com>;
> Gujjar, Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy <timothy.mcdaniel@intel.com>; Hemant Agrawal
> <hemant.agrawal@nxp.com>; Van Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
> <mattias.ronnblom@ericsson.com>; Ma, Liang J <liang.j.ma@intel.com>; Ray Kinsella <mdr@ashroe.eu>; Neil Horman
> <nhorman@tuxdriver.com>; dpdk-dev <dev@dpdk.org>
> Subject: Re: [dpdk-dev] [PATCH v8 7/8] doc: announce event Rx adapter config changes
> 
> On Fri, Mar 26, 2021 at 7:41 PM <pbhagavatula@marvell.com> wrote:
> >
> > From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> >
> > The Rx adapter event vector configuration will be merged into
> > Rx adapter queue configuration to simplify enabling event
> > vectorization.
> >
> > Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> > Acked-by: Ray Kinsella <mdr@ashroe.eu>
> 
> Acked-by: Jerin Jacob <jerinj@marvell.com>
> 
> @Jayatheerthan, Jay
> 
> Any comment on this?

Yes, looks good.

Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>

> 
> 
> > ---
> >  doc/guides/rel_notes/deprecation.rst | 9 +++++++++
> >  1 file changed, 9 insertions(+)
> >
> > diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst
> > index 64629e064..ef6d1bb18 100644
> > --- a/doc/guides/rel_notes/deprecation.rst
> > +++ b/doc/guides/rel_notes/deprecation.rst
> > @@ -130,3 +130,12 @@ Deprecation Notices
> >  * cmdline: ``cmdline`` structure will be made opaque to hide platform-specific
> >    content. On Linux and FreeBSD, supported prior to DPDK 20.11,
> >    original structure will be kept until DPDK 21.11.
> > +
> > +* eventdev: The structure ``rte_event_eth_rx_adapter_queue_conf`` will be
> > +  extended to include ``rte_event_eth_rx_adapter_event_vector_config`` elements
> > +  and the function ``rte_event_eth_rx_adapter_queue_event_vector_config`` will
> > +  be removed in DPDK 21.11.
> > +
> > +  An application can enable event vectorization by passing the desired vector
> > +  values to the function ``rte_event_eth_rx_adapter_queue_add`` using
> > +  the structure ``rte_event_eth_rx_adapter_queue_add``.
> > --
> > 2.17.1
> >

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v8 4/8] eventdev: add Rx adapter event vector support
  2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 4/8] eventdev: add Rx adapter event vector support pbhagavatula
@ 2021-03-28  8:18                 ` Jerin Jacob
  2021-03-29  6:09                   ` Jayatheerthan, Jay
  0 siblings, 1 reply; 153+ messages in thread
From: Jerin Jacob @ 2021-03-28  8:18 UTC (permalink / raw)
  To: Pavan Nikhilesh
  Cc: Jerin Jacob, Jayatheerthan, Jay, Erik Gabriel Carrillo, Gujjar,
	Abhinandan S, McDaniel, Timothy, Hemant Agrawal, Van Haaren,
	Harry, Mattias Rönnblom, Liang Ma, dpdk-dev

On Fri, Mar 26, 2021 at 7:40 PM <pbhagavatula@marvell.com> wrote:
>
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>
> Add event vector support for event eth Rx adapter, the implementation
> creates vector flows based on port and queue identifier of the received
> mbufs.
> The flow id for SW Rx event vectorization will use 12-bits of queue
> identifier and 8-bits port identifier when custom flow id is not set
> for simplicity.
>
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>


@Jayatheerthan, Jay
Let me know if you have any comments on this patch.


> ---
>  lib/librte_eventdev/eventdev_pmd.h            |   7 +-
>  .../rte_event_eth_rx_adapter.c                | 269 ++++++++++++++++--
>  lib/librte_eventdev/rte_eventdev.c            |   6 +-
>  3 files changed, 258 insertions(+), 24 deletions(-)
>
> diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
> index 9297f1433..0f724ac85 100644
> --- a/lib/librte_eventdev/eventdev_pmd.h
> +++ b/lib/librte_eventdev/eventdev_pmd.h
> @@ -69,9 +69,10 @@ extern "C" {
>         } \
>  } while (0)
>
> -#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP \
> -               ((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) | \
> -                       (RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ))
> +#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP                                        \
> +       ((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |                     \
> +        (RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) |                         \
> +        (RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR))
>
>  #define RTE_EVENT_CRYPTO_ADAPTER_SW_CAP \
>                 RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA
> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> index ac8ba5bf0..fba3b5ec3 100644
> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> @@ -26,6 +26,10 @@
>  #define BATCH_SIZE             32
>  #define BLOCK_CNT_THRESHOLD    10
>  #define ETH_EVENT_BUFFER_SIZE  (4*BATCH_SIZE)
> +#define MAX_VECTOR_SIZE                1024
> +#define MIN_VECTOR_SIZE                4
> +#define MAX_VECTOR_NS          1E9
> +#define MIN_VECTOR_NS          1E5
>
>  #define ETH_RX_ADAPTER_SERVICE_NAME_LEN        32
>  #define ETH_RX_ADAPTER_MEM_NAME_LEN    32
> @@ -59,6 +63,20 @@ struct eth_rx_poll_entry {
>         uint16_t eth_rx_qid;
>  };
>
> +struct eth_rx_vector_data {
> +       TAILQ_ENTRY(eth_rx_vector_data) next;
> +       uint16_t port;
> +       uint16_t queue;
> +       uint16_t max_vector_count;
> +       uint64_t event;
> +       uint64_t ts;
> +       uint64_t vector_timeout_ticks;
> +       struct rte_mempool *vector_pool;
> +       struct rte_event_vector *vector_ev;
> +} __rte_cache_aligned;
> +
> +TAILQ_HEAD(eth_rx_vector_data_list, eth_rx_vector_data);
> +
>  /* Instance per adapter */
>  struct rte_eth_event_enqueue_buffer {
>         /* Count of events in this buffer */
> @@ -92,6 +110,14 @@ struct rte_event_eth_rx_adapter {
>         uint32_t wrr_pos;
>         /* Event burst buffer */
>         struct rte_eth_event_enqueue_buffer event_enqueue_buffer;
> +       /* Vector enable flag */
> +       uint8_t ena_vector;
> +       /* Timestamp of previous vector expiry list traversal */
> +       uint64_t prev_expiry_ts;
> +       /* Minimum ticks to wait before traversing expiry list */
> +       uint64_t vector_tmo_ticks;
> +       /* vector list */
> +       struct eth_rx_vector_data_list vector_list;
>         /* Per adapter stats */
>         struct rte_event_eth_rx_adapter_stats stats;
>         /* Block count, counts up to BLOCK_CNT_THRESHOLD */
> @@ -198,9 +224,11 @@ struct eth_device_info {
>  struct eth_rx_queue_info {
>         int queue_enabled;      /* True if added */
>         int intr_enabled;
> +       uint8_t ena_vector;
>         uint16_t wt;            /* Polling weight */
>         uint32_t flow_id_mask;  /* Set to ~0 if app provides flow id else 0 */
>         uint64_t event;
> +       struct eth_rx_vector_data vector_data;
>  };
>
>  static struct rte_event_eth_rx_adapter **event_eth_rx_adapter;
> @@ -722,6 +750,9 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
>             &rx_adapter->event_enqueue_buffer;
>         struct rte_event_eth_rx_adapter_stats *stats = &rx_adapter->stats;
>
> +       if (!buf->count)
> +               return 0;
> +
>         uint16_t n = rte_event_enqueue_new_burst(rx_adapter->eventdev_id,
>                                         rx_adapter->event_port_id,
>                                         buf->events,
> @@ -742,6 +773,77 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
>         return n;
>  }
>
> +static inline void
> +rxa_init_vector(struct rte_event_eth_rx_adapter *rx_adapter,
> +               struct eth_rx_vector_data *vec)
> +{
> +       vec->vector_ev->nb_elem = 0;
> +       vec->vector_ev->port = vec->port;
> +       vec->vector_ev->queue = vec->queue;
> +       vec->vector_ev->attr_valid = true;
> +       TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
> +}
> +
> +static inline uint16_t
> +rxa_create_event_vector(struct rte_event_eth_rx_adapter *rx_adapter,
> +                       struct eth_rx_queue_info *queue_info,
> +                       struct rte_eth_event_enqueue_buffer *buf,
> +                       struct rte_mbuf **mbufs, uint16_t num)
> +{
> +       struct rte_event *ev = &buf->events[buf->count];
> +       struct eth_rx_vector_data *vec;
> +       uint16_t filled, space, sz;
> +
> +       filled = 0;
> +       vec = &queue_info->vector_data;
> +
> +       if (vec->vector_ev == NULL) {
> +               if (rte_mempool_get(vec->vector_pool,
> +                                   (void **)&vec->vector_ev) < 0) {
> +                       rte_pktmbuf_free_bulk(mbufs, num);
> +                       return 0;
> +               }
> +               rxa_init_vector(rx_adapter, vec);
> +       }
> +       while (num) {
> +               if (vec->vector_ev->nb_elem == vec->max_vector_count) {
> +                       /* Event ready. */
> +                       ev->event = vec->event;
> +                       ev->vec = vec->vector_ev;
> +                       ev++;
> +                       filled++;
> +                       vec->vector_ev = NULL;
> +                       TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
> +                       if (rte_mempool_get(vec->vector_pool,
> +                                           (void **)&vec->vector_ev) < 0) {
> +                               rte_pktmbuf_free_bulk(mbufs, num);
> +                               return 0;
> +                       }
> +                       rxa_init_vector(rx_adapter, vec);
> +               }
> +
> +               space = vec->max_vector_count - vec->vector_ev->nb_elem;
> +               sz = num > space ? space : num;
> +               memcpy(vec->vector_ev->mbufs + vec->vector_ev->nb_elem, mbufs,
> +                      sizeof(void *) * sz);
> +               vec->vector_ev->nb_elem += sz;
> +               num -= sz;
> +               mbufs += sz;
> +               vec->ts = rte_rdtsc();
> +       }
> +
> +       if (vec->vector_ev->nb_elem == vec->max_vector_count) {
> +               ev->event = vec->event;
> +               ev->vec = vec->vector_ev;
> +               ev++;
> +               filled++;
> +               vec->vector_ev = NULL;
> +               TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
> +       }
> +
> +       return filled;
> +}
> +
>  static inline void
>  rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
>                 uint16_t eth_dev_id,
> @@ -766,29 +868,33 @@ rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
>         uint16_t nb_cb;
>         uint16_t dropped;
>
> -       /* 0xffff ffff if PKT_RX_RSS_HASH is set, otherwise 0 */
> -       rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
> -       do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
> -
> -       for (i = 0; i < num; i++) {
> -               m = mbufs[i];
> -
> -               rss = do_rss ?
> -                       rxa_do_softrss(m, rx_adapter->rss_key_be) :
> -                       m->hash.rss;
> -               ev->event = event;
> -               ev->flow_id = (rss & ~flow_id_mask) |
> -                               (ev->flow_id & flow_id_mask);
> -               ev->mbuf = m;
> -               ev++;
> +       if (!eth_rx_queue_info->ena_vector) {
> +               /* 0xffff ffff if PKT_RX_RSS_HASH is set, otherwise 0 */
> +               rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
> +               do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
> +               for (i = 0; i < num; i++) {
> +                       m = mbufs[i];
> +
> +                       rss = do_rss ? rxa_do_softrss(m, rx_adapter->rss_key_be)
> +                                    : m->hash.rss;
> +                       ev->event = event;
> +                       ev->flow_id = (rss & ~flow_id_mask) |
> +                                     (ev->flow_id & flow_id_mask);
> +                       ev->mbuf = m;
> +                       ev++;
> +               }
> +       } else {
> +               num = rxa_create_event_vector(rx_adapter, eth_rx_queue_info,
> +                                             buf, mbufs, num);
>         }
>
> -       if (dev_info->cb_fn) {
> +       if (num && dev_info->cb_fn) {
>
>                 dropped = 0;
>                 nb_cb = dev_info->cb_fn(eth_dev_id, rx_queue_id,
> -                                       ETH_EVENT_BUFFER_SIZE, buf->count, ev,
> -                                       num, dev_info->cb_arg, &dropped);
> +                                       ETH_EVENT_BUFFER_SIZE, buf->count,
> +                                       &buf->events[buf->count], num,
> +                                       dev_info->cb_arg, &dropped);
>                 if (unlikely(nb_cb > num))
>                         RTE_EDEV_LOG_ERR("Rx CB returned %d (> %d) events",
>                                 nb_cb, num);
> @@ -1124,6 +1230,30 @@ rxa_poll(struct rte_event_eth_rx_adapter *rx_adapter)
>         return nb_rx;
>  }
>
> +static void
> +rxa_vector_expire(struct eth_rx_vector_data *vec, void *arg)
> +{
> +       struct rte_event_eth_rx_adapter *rx_adapter = arg;
> +       struct rte_eth_event_enqueue_buffer *buf =
> +               &rx_adapter->event_enqueue_buffer;
> +       struct rte_event *ev;
> +
> +       if (buf->count)
> +               rxa_flush_event_buffer(rx_adapter);
> +
> +       if (vec->vector_ev->nb_elem == 0)
> +               return;
> +       ev = &buf->events[buf->count];
> +
> +       /* Event ready. */
> +       ev->event = vec->event;
> +       ev->vec = vec->vector_ev;
> +       buf->count++;
> +
> +       vec->vector_ev = NULL;
> +       vec->ts = 0;
> +}
> +
>  static int
>  rxa_service_func(void *args)
>  {
> @@ -1137,6 +1267,24 @@ rxa_service_func(void *args)
>                 return 0;
>         }
>
> +       if (rx_adapter->ena_vector) {
> +               if ((rte_rdtsc() - rx_adapter->prev_expiry_ts) >=
> +                   rx_adapter->vector_tmo_ticks) {
> +                       struct eth_rx_vector_data *vec;
> +
> +                       TAILQ_FOREACH(vec, &rx_adapter->vector_list, next) {
> +                               uint64_t elapsed_time = rte_rdtsc() - vec->ts;
> +
> +                               if (elapsed_time >= vec->vector_timeout_ticks) {
> +                                       rxa_vector_expire(vec, rx_adapter);
> +                                       TAILQ_REMOVE(&rx_adapter->vector_list,
> +                                                    vec, next);
> +                               }
> +                       }
> +                       rx_adapter->prev_expiry_ts = rte_rdtsc();
> +               }
> +       }
> +
>         stats = &rx_adapter->stats;
>         stats->rx_packets += rxa_intr_ring_dequeue(rx_adapter);
>         stats->rx_packets += rxa_poll(rx_adapter);
> @@ -1640,6 +1788,29 @@ rxa_update_queue(struct rte_event_eth_rx_adapter *rx_adapter,
>         }
>  }
>
> +static void
> +rxa_set_vector_data(struct eth_rx_queue_info *queue_info, uint16_t vector_count,
> +                   uint64_t vector_ns, struct rte_mempool *mp, int32_t qid,
> +                   uint16_t port_id)
> +{
> +#define NSEC2TICK(__ns, __freq) (((__ns) * (__freq)) / 1E9)
> +       struct eth_rx_vector_data *vector_data;
> +       uint32_t flow_id;
> +
> +       vector_data = &queue_info->vector_data;
> +       vector_data->max_vector_count = vector_count;
> +       vector_data->port = port_id;
> +       vector_data->queue = qid;
> +       vector_data->vector_pool = mp;
> +       vector_data->vector_timeout_ticks =
> +               NSEC2TICK(vector_ns, rte_get_timer_hz());
> +       vector_data->ts = 0;
> +       flow_id = queue_info->event & 0xFFFFF;
> +       flow_id =
> +               flow_id == 0 ? (qid & 0xFFF) | (port_id & 0xFF) << 12 : flow_id;
> +       vector_data->event = (queue_info->event & ~0xFFFFF) | flow_id;
> +}
> +
>  static void
>  rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
>         struct eth_device_info *dev_info,
> @@ -1741,6 +1912,42 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
>         }
>  }
>
> +static void
> +rxa_sw_event_vector_configure(
> +       struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
> +       int rx_queue_id,
> +       const struct rte_event_eth_rx_adapter_event_vector_config *config)
> +{
> +       struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
> +       struct eth_rx_queue_info *queue_info;
> +       struct rte_event *qi_ev;
> +
> +       if (rx_queue_id == -1) {
> +               uint16_t nb_rx_queues;
> +               uint16_t i;
> +
> +               nb_rx_queues = dev_info->dev->data->nb_rx_queues;
> +               for (i = 0; i < nb_rx_queues; i++)
> +                       rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
> +                                                     config);
> +               return;
> +       }
> +
> +       queue_info = &dev_info->rx_queue[rx_queue_id];
> +       qi_ev = (struct rte_event *)&queue_info->event;
> +       queue_info->ena_vector = 1;
> +       qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
> +       rxa_set_vector_data(queue_info, config->vector_sz,
> +                           config->vector_timeout_ns, config->vector_mp,
> +                           rx_queue_id, dev_info->dev->data->port_id);
> +       rx_adapter->ena_vector = 1;
> +       rx_adapter->vector_tmo_ticks =
> +               rx_adapter->vector_tmo_ticks ?
> +                             RTE_MIN(config->vector_timeout_ns >> 1,
> +                               rx_adapter->vector_tmo_ticks) :
> +                             config->vector_timeout_ns >> 1;
> +}
> +
>  static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
>                 uint16_t eth_dev_id,
>                 int rx_queue_id,
> @@ -1967,6 +2174,7 @@ rte_event_eth_rx_adapter_create_ext(uint8_t id, uint8_t dev_id,
>         rx_adapter->conf_cb = conf_cb;
>         rx_adapter->conf_arg = conf_arg;
>         rx_adapter->id = id;
> +       TAILQ_INIT(&rx_adapter->vector_list);
>         strcpy(rx_adapter->mem_name, mem_name);
>         rx_adapter->eth_devices = rte_zmalloc_socket(rx_adapter->mem_name,
>                                         RTE_MAX_ETHPORTS *
> @@ -2081,6 +2289,15 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
>                 return -EINVAL;
>         }
>
> +       if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
> +           (queue_conf->rx_queue_flags &
> +            RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
> +               RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
> +                                " eth port: %" PRIu16 " adapter id: %" PRIu8,
> +                                eth_dev_id, id);
> +               return -EINVAL;
> +       }
> +
>         if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
>                 (rx_queue_id != -1)) {
>                 RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
> @@ -2143,6 +2360,17 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
>         return 0;
>  }
>
> +static int
> +rxa_sw_vector_limits(struct rte_event_eth_rx_adapter_vector_limits *limits)
> +{
> +       limits->max_sz = MAX_VECTOR_SIZE;
> +       limits->min_sz = MIN_VECTOR_SIZE;
> +       limits->max_timeout_ns = MAX_VECTOR_NS;
> +       limits->min_timeout_ns = MIN_VECTOR_NS;
> +
> +       return 0;
> +}
> +
>  int
>  rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
>                                 int32_t rx_queue_id)
> @@ -2333,7 +2561,8 @@ rte_event_eth_rx_adapter_queue_event_vector_config(
>                 ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
>                         dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
>         } else {
> -               ret = -ENOTSUP;
> +               rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
> +                                             rx_queue_id, config);
>         }
>
>         return ret;
> @@ -2371,7 +2600,7 @@ rte_event_eth_rx_adapter_vector_limits_get(
>                 ret = dev->dev_ops->eth_rx_adapter_vector_limits_get(
>                         dev, &rte_eth_devices[eth_port_id], limits);
>         } else {
> -               ret = -ENOTSUP;
> +               ret = rxa_sw_vector_limits(limits);
>         }
>
>         return ret;
> diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
> index be0499c52..62824654b 100644
> --- a/lib/librte_eventdev/rte_eventdev.c
> +++ b/lib/librte_eventdev/rte_eventdev.c
> @@ -122,7 +122,11 @@ rte_event_eth_rx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
>
>         if (caps == NULL)
>                 return -EINVAL;
> -       *caps = 0;
> +
> +       if (dev->dev_ops->eth_rx_adapter_caps_get == NULL)
> +               *caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
> +       else
> +               *caps = 0;
>
>         return dev->dev_ops->eth_rx_adapter_caps_get ?
>                                 (*dev->dev_ops->eth_rx_adapter_caps_get)(dev,
> --
> 2.17.1
>

^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v8 4/8] eventdev: add Rx adapter event vector support
  2021-03-28  8:18                 ` Jerin Jacob
@ 2021-03-29  6:09                   ` Jayatheerthan, Jay
  0 siblings, 0 replies; 153+ messages in thread
From: Jayatheerthan, Jay @ 2021-03-29  6:09 UTC (permalink / raw)
  To: Jerin Jacob, Pavan Nikhilesh
  Cc: Jerin Jacob, Carrillo, Erik G, Gujjar, Abhinandan S, McDaniel,
	Timothy, Hemant Agrawal, Van Haaren, Harry, mattias.ronnblom, Ma,
	Liang J, dpdk-dev

> -----Original Message-----
> From: Jerin Jacob <jerinjacobk@gmail.com>
> Sent: Sunday, March 28, 2021 1:49 PM
> To: Pavan Nikhilesh <pbhagavatula@marvell.com>
> Cc: Jerin Jacob <jerinj@marvell.com>; Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Carrillo, Erik G <erik.g.carrillo@intel.com>;
> Gujjar, Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy <timothy.mcdaniel@intel.com>; Hemant Agrawal
> <hemant.agrawal@nxp.com>; Van Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
> <mattias.ronnblom@ericsson.com>; Ma, Liang J <liang.j.ma@intel.com>; dpdk-dev <dev@dpdk.org>
> Subject: Re: [dpdk-dev] [PATCH v8 4/8] eventdev: add Rx adapter event vector support
> 
> On Fri, Mar 26, 2021 at 7:40 PM <pbhagavatula@marvell.com> wrote:
> >
> > From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> >
> > Add event vector support for event eth Rx adapter, the implementation
> > creates vector flows based on port and queue identifier of the received
> > mbufs.
> > The flow id for SW Rx event vectorization will use 12-bits of queue
> > identifier and 8-bits port identifier when custom flow id is not set
> > for simplicity.

Since Rx adapter code implements SW case, it may be relevant to put the flow_id details in Rx adapter documentation. Please also include the details of how we are forming the flow_id with queue id and port. Having it in commit log, may not be readily available when needed.

> >
> > Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> 
> @Jayatheerthan, Jay
> Let me know if you have any comments on this patch.
> 
> 
> > ---
> >  lib/librte_eventdev/eventdev_pmd.h            |   7 +-
> >  .../rte_event_eth_rx_adapter.c                | 269 ++++++++++++++++--
> >  lib/librte_eventdev/rte_eventdev.c            |   6 +-
> >  3 files changed, 258 insertions(+), 24 deletions(-)
> >
> > diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
> > index 9297f1433..0f724ac85 100644
> > --- a/lib/librte_eventdev/eventdev_pmd.h
> > +++ b/lib/librte_eventdev/eventdev_pmd.h
> > @@ -69,9 +69,10 @@ extern "C" {
> >         } \
> >  } while (0)
> >
> > -#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP \
> > -               ((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) | \
> > -                       (RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ))
> > +#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP                                        \
> > +       ((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |                     \
> > +        (RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) |                         \
> > +        (RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR))
> >
> >  #define RTE_EVENT_CRYPTO_ADAPTER_SW_CAP \
> >                 RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA
> > diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> > index ac8ba5bf0..fba3b5ec3 100644
> > --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> > +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> > @@ -26,6 +26,10 @@
> >  #define BATCH_SIZE             32
> >  #define BLOCK_CNT_THRESHOLD    10
> >  #define ETH_EVENT_BUFFER_SIZE  (4*BATCH_SIZE)
> > +#define MAX_VECTOR_SIZE                1024
> > +#define MIN_VECTOR_SIZE                4
> > +#define MAX_VECTOR_NS          1E9
> > +#define MIN_VECTOR_NS          1E5
> >
> >  #define ETH_RX_ADAPTER_SERVICE_NAME_LEN        32
> >  #define ETH_RX_ADAPTER_MEM_NAME_LEN    32
> > @@ -59,6 +63,20 @@ struct eth_rx_poll_entry {
> >         uint16_t eth_rx_qid;
> >  };
> >
> > +struct eth_rx_vector_data {
> > +       TAILQ_ENTRY(eth_rx_vector_data) next;
> > +       uint16_t port;
> > +       uint16_t queue;
> > +       uint16_t max_vector_count;
> > +       uint64_t event;
> > +       uint64_t ts;
> > +       uint64_t vector_timeout_ticks;
> > +       struct rte_mempool *vector_pool;
> > +       struct rte_event_vector *vector_ev;
> > +} __rte_cache_aligned;
> > +
> > +TAILQ_HEAD(eth_rx_vector_data_list, eth_rx_vector_data);
> > +
> >  /* Instance per adapter */
> >  struct rte_eth_event_enqueue_buffer {
> >         /* Count of events in this buffer */
> > @@ -92,6 +110,14 @@ struct rte_event_eth_rx_adapter {
> >         uint32_t wrr_pos;
> >         /* Event burst buffer */
> >         struct rte_eth_event_enqueue_buffer event_enqueue_buffer;
> > +       /* Vector enable flag */
> > +       uint8_t ena_vector;
> > +       /* Timestamp of previous vector expiry list traversal */
> > +       uint64_t prev_expiry_ts;
> > +       /* Minimum ticks to wait before traversing expiry list */
> > +       uint64_t vector_tmo_ticks;
> > +       /* vector list */
> > +       struct eth_rx_vector_data_list vector_list;

We are maintaining a list of vector data that is allocated as we are performing vectorization. Any need for freeing these in adapter stop/free or queue del? For e.g. vector_ev that is allocated from vector_pool.

> >         /* Per adapter stats */
> >         struct rte_event_eth_rx_adapter_stats stats;
> >         /* Block count, counts up to BLOCK_CNT_THRESHOLD */
> > @@ -198,9 +224,11 @@ struct eth_device_info {
> >  struct eth_rx_queue_info {
> >         int queue_enabled;      /* True if added */
> >         int intr_enabled;
> > +       uint8_t ena_vector;
> >         uint16_t wt;            /* Polling weight */
> >         uint32_t flow_id_mask;  /* Set to ~0 if app provides flow id else 0 */
> >         uint64_t event;
> > +       struct eth_rx_vector_data vector_data;
> >  };
> >
> >  static struct rte_event_eth_rx_adapter **event_eth_rx_adapter;
> > @@ -722,6 +750,9 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
> >             &rx_adapter->event_enqueue_buffer;
> >         struct rte_event_eth_rx_adapter_stats *stats = &rx_adapter->stats;
> >
> > +       if (!buf->count)
> > +               return 0;
> > +
> >         uint16_t n = rte_event_enqueue_new_burst(rx_adapter->eventdev_id,
> >                                         rx_adapter->event_port_id,
> >                                         buf->events,
> > @@ -742,6 +773,77 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
> >         return n;
> >  }
> >
> > +static inline void
> > +rxa_init_vector(struct rte_event_eth_rx_adapter *rx_adapter,
> > +               struct eth_rx_vector_data *vec)
> > +{
> > +       vec->vector_ev->nb_elem = 0;
> > +       vec->vector_ev->port = vec->port;
> > +       vec->vector_ev->queue = vec->queue;
> > +       vec->vector_ev->attr_valid = true;
> > +       TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
> > +}
> > +
> > +static inline uint16_t
> > +rxa_create_event_vector(struct rte_event_eth_rx_adapter *rx_adapter,
> > +                       struct eth_rx_queue_info *queue_info,
> > +                       struct rte_eth_event_enqueue_buffer *buf,
> > +                       struct rte_mbuf **mbufs, uint16_t num)
> > +{
> > +       struct rte_event *ev = &buf->events[buf->count];
> > +       struct eth_rx_vector_data *vec;
> > +       uint16_t filled, space, sz;
> > +
> > +       filled = 0;
> > +       vec = &queue_info->vector_data;
> > +
> > +       if (vec->vector_ev == NULL) {
> > +               if (rte_mempool_get(vec->vector_pool,
> > +                                   (void **)&vec->vector_ev) < 0) {
> > +                       rte_pktmbuf_free_bulk(mbufs, num);
> > +                       return 0;
> > +               }
> > +               rxa_init_vector(rx_adapter, vec);
> > +       }
> > +       while (num) {
> > +               if (vec->vector_ev->nb_elem == vec->max_vector_count) {
> > +                       /* Event ready. */
> > +                       ev->event = vec->event;
> > +                       ev->vec = vec->vector_ev;
> > +                       ev++;
> > +                       filled++;
> > +                       vec->vector_ev = NULL;
> > +                       TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
> > +                       if (rte_mempool_get(vec->vector_pool,
> > +                                           (void **)&vec->vector_ev) < 0) {
> > +                               rte_pktmbuf_free_bulk(mbufs, num);
> > +                               return 0;
> > +                       }
> > +                       rxa_init_vector(rx_adapter, vec);
> > +               }
> > +
> > +               space = vec->max_vector_count - vec->vector_ev->nb_elem;
> > +               sz = num > space ? space : num;
> > +               memcpy(vec->vector_ev->mbufs + vec->vector_ev->nb_elem, mbufs,
> > +                      sizeof(void *) * sz);
> > +               vec->vector_ev->nb_elem += sz;
> > +               num -= sz;
> > +               mbufs += sz;
> > +               vec->ts = rte_rdtsc();
> > +       }
> > +
> > +       if (vec->vector_ev->nb_elem == vec->max_vector_count) {
> > +               ev->event = vec->event;
> > +               ev->vec = vec->vector_ev;
> > +               ev++;
> > +               filled++;
> > +               vec->vector_ev = NULL;
> > +               TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
> > +       }
> > +
> > +       return filled;
> > +}
> > +
> >  static inline void
> >  rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
> >                 uint16_t eth_dev_id,
> > @@ -766,29 +868,33 @@ rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
> >         uint16_t nb_cb;
> >         uint16_t dropped;
> >
> > -       /* 0xffff ffff if PKT_RX_RSS_HASH is set, otherwise 0 */
> > -       rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
> > -       do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
> > -
> > -       for (i = 0; i < num; i++) {
> > -               m = mbufs[i];
> > -
> > -               rss = do_rss ?
> > -                       rxa_do_softrss(m, rx_adapter->rss_key_be) :
> > -                       m->hash.rss;
> > -               ev->event = event;
> > -               ev->flow_id = (rss & ~flow_id_mask) |
> > -                               (ev->flow_id & flow_id_mask);
> > -               ev->mbuf = m;
> > -               ev++;
> > +       if (!eth_rx_queue_info->ena_vector) {
> > +               /* 0xffff ffff if PKT_RX_RSS_HASH is set, otherwise 0 */
> > +               rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
> > +               do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
> > +               for (i = 0; i < num; i++) {
> > +                       m = mbufs[i];
> > +
> > +                       rss = do_rss ? rxa_do_softrss(m, rx_adapter->rss_key_be)
> > +                                    : m->hash.rss;
> > +                       ev->event = event;
> > +                       ev->flow_id = (rss & ~flow_id_mask) |
> > +                                     (ev->flow_id & flow_id_mask);
> > +                       ev->mbuf = m;
> > +                       ev++;
> > +               }
> > +       } else {
> > +               num = rxa_create_event_vector(rx_adapter, eth_rx_queue_info,
> > +                                             buf, mbufs, num);
> >         }
> >
> > -       if (dev_info->cb_fn) {
> > +       if (num && dev_info->cb_fn) {
> >
> >                 dropped = 0;
> >                 nb_cb = dev_info->cb_fn(eth_dev_id, rx_queue_id,
> > -                                       ETH_EVENT_BUFFER_SIZE, buf->count, ev,
> > -                                       num, dev_info->cb_arg, &dropped);
> > +                                       ETH_EVENT_BUFFER_SIZE, buf->count,
> > +                                       &buf->events[buf->count], num,
> > +                                       dev_info->cb_arg, &dropped);
> >                 if (unlikely(nb_cb > num))
> >                         RTE_EDEV_LOG_ERR("Rx CB returned %d (> %d) events",
> >                                 nb_cb, num);
> > @@ -1124,6 +1230,30 @@ rxa_poll(struct rte_event_eth_rx_adapter *rx_adapter)
> >         return nb_rx;
> >  }
> >
> > +static void
> > +rxa_vector_expire(struct eth_rx_vector_data *vec, void *arg)
> > +{
> > +       struct rte_event_eth_rx_adapter *rx_adapter = arg;
> > +       struct rte_eth_event_enqueue_buffer *buf =
> > +               &rx_adapter->event_enqueue_buffer;
> > +       struct rte_event *ev;
> > +
> > +       if (buf->count)
> > +               rxa_flush_event_buffer(rx_adapter);
> > +
> > +       if (vec->vector_ev->nb_elem == 0)
> > +               return;
> > +       ev = &buf->events[buf->count];
> > +
> > +       /* Event ready. */
> > +       ev->event = vec->event;
> > +       ev->vec = vec->vector_ev;
> > +       buf->count++;
> > +
> > +       vec->vector_ev = NULL;
> > +       vec->ts = 0;
> > +}
> > +
> >  static int
> >  rxa_service_func(void *args)
> >  {
> > @@ -1137,6 +1267,24 @@ rxa_service_func(void *args)
> >                 return 0;
> >         }
> >
> > +       if (rx_adapter->ena_vector) {
> > +               if ((rte_rdtsc() - rx_adapter->prev_expiry_ts) >=
> > +                   rx_adapter->vector_tmo_ticks) {
> > +                       struct eth_rx_vector_data *vec;
> > +
> > +                       TAILQ_FOREACH(vec, &rx_adapter->vector_list, next) {
> > +                               uint64_t elapsed_time = rte_rdtsc() - vec->ts;
> > +
> > +                               if (elapsed_time >= vec->vector_timeout_ticks) {
> > +                                       rxa_vector_expire(vec, rx_adapter);
> > +                                       TAILQ_REMOVE(&rx_adapter->vector_list,
> > +                                                    vec, next);
> > +                               }
> > +                       }
> > +                       rx_adapter->prev_expiry_ts = rte_rdtsc();
> > +               }
> > +       }
> > +
> >         stats = &rx_adapter->stats;
> >         stats->rx_packets += rxa_intr_ring_dequeue(rx_adapter);
> >         stats->rx_packets += rxa_poll(rx_adapter);
> > @@ -1640,6 +1788,29 @@ rxa_update_queue(struct rte_event_eth_rx_adapter *rx_adapter,
> >         }
> >  }
> >
> > +static void
> > +rxa_set_vector_data(struct eth_rx_queue_info *queue_info, uint16_t vector_count,
> > +                   uint64_t vector_ns, struct rte_mempool *mp, int32_t qid,
> > +                   uint16_t port_id)
> > +{
> > +#define NSEC2TICK(__ns, __freq) (((__ns) * (__freq)) / 1E9)
> > +       struct eth_rx_vector_data *vector_data;
> > +       uint32_t flow_id;
> > +
> > +       vector_data = &queue_info->vector_data;
> > +       vector_data->max_vector_count = vector_count;
> > +       vector_data->port = port_id;
> > +       vector_data->queue = qid;
> > +       vector_data->vector_pool = mp;
> > +       vector_data->vector_timeout_ticks =
> > +               NSEC2TICK(vector_ns, rte_get_timer_hz());
> > +       vector_data->ts = 0;
> > +       flow_id = queue_info->event & 0xFFFFF;
> > +       flow_id =
> > +               flow_id == 0 ? (qid & 0xFFF) | (port_id & 0xFF) << 12 : flow_id;
> > +       vector_data->event = (queue_info->event & ~0xFFFFF) | flow_id;
> > +}
> > +
> >  static void
> >  rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
> >         struct eth_device_info *dev_info,
> > @@ -1741,6 +1912,42 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
> >         }
> >  }
> >
> > +static void
> > +rxa_sw_event_vector_configure(
> > +       struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
> > +       int rx_queue_id,
> > +       const struct rte_event_eth_rx_adapter_event_vector_config *config)
> > +{
> > +       struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
> > +       struct eth_rx_queue_info *queue_info;
> > +       struct rte_event *qi_ev;
> > +
> > +       if (rx_queue_id == -1) {
> > +               uint16_t nb_rx_queues;
> > +               uint16_t i;
> > +
> > +               nb_rx_queues = dev_info->dev->data->nb_rx_queues;
> > +               for (i = 0; i < nb_rx_queues; i++)
> > +                       rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
> > +                                                     config);
> > +               return;
> > +       }
> > +
> > +       queue_info = &dev_info->rx_queue[rx_queue_id];
> > +       qi_ev = (struct rte_event *)&queue_info->event;
> > +       queue_info->ena_vector = 1;
> > +       qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
> > +       rxa_set_vector_data(queue_info, config->vector_sz,
> > +                           config->vector_timeout_ns, config->vector_mp,
> > +                           rx_queue_id, dev_info->dev->data->port_id);
> > +       rx_adapter->ena_vector = 1;
> > +       rx_adapter->vector_tmo_ticks =
> > +               rx_adapter->vector_tmo_ticks ?
> > +                             RTE_MIN(config->vector_timeout_ns >> 1,
> > +                               rx_adapter->vector_tmo_ticks) :
> > +                             config->vector_timeout_ns >> 1;
> > +}
> > +
> >  static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
> >                 uint16_t eth_dev_id,
> >                 int rx_queue_id,
> > @@ -1967,6 +2174,7 @@ rte_event_eth_rx_adapter_create_ext(uint8_t id, uint8_t dev_id,
> >         rx_adapter->conf_cb = conf_cb;
> >         rx_adapter->conf_arg = conf_arg;
> >         rx_adapter->id = id;
> > +       TAILQ_INIT(&rx_adapter->vector_list);
> >         strcpy(rx_adapter->mem_name, mem_name);
> >         rx_adapter->eth_devices = rte_zmalloc_socket(rx_adapter->mem_name,
> >                                         RTE_MAX_ETHPORTS *
> > @@ -2081,6 +2289,15 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
> >                 return -EINVAL;
> >         }
> >
> > +       if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
> > +           (queue_conf->rx_queue_flags &
> > +            RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
> > +               RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
> > +                                " eth port: %" PRIu16 " adapter id: %" PRIu8,
> > +                                eth_dev_id, id);
> > +               return -EINVAL;
> > +       }
> > +
> >         if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
> >                 (rx_queue_id != -1)) {
> >                 RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
> > @@ -2143,6 +2360,17 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
> >         return 0;
> >  }
> >
> > +static int
> > +rxa_sw_vector_limits(struct rte_event_eth_rx_adapter_vector_limits *limits)
> > +{
> > +       limits->max_sz = MAX_VECTOR_SIZE;
> > +       limits->min_sz = MIN_VECTOR_SIZE;
> > +       limits->max_timeout_ns = MAX_VECTOR_NS;
> > +       limits->min_timeout_ns = MIN_VECTOR_NS;
> > +
> > +       return 0;
> > +}
> > +
> >  int
> >  rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
> >                                 int32_t rx_queue_id)
> > @@ -2333,7 +2561,8 @@ rte_event_eth_rx_adapter_queue_event_vector_config(
> >                 ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
> >                         dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
> >         } else {
> > -               ret = -ENOTSUP;
> > +               rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
> > +                                             rx_queue_id, config);
> >         }
> >
> >         return ret;
> > @@ -2371,7 +2600,7 @@ rte_event_eth_rx_adapter_vector_limits_get(
> >                 ret = dev->dev_ops->eth_rx_adapter_vector_limits_get(
> >                         dev, &rte_eth_devices[eth_port_id], limits);
> >         } else {
> > -               ret = -ENOTSUP;
> > +               ret = rxa_sw_vector_limits(limits);
> >         }
> >
> >         return ret;
> > diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
> > index be0499c52..62824654b 100644
> > --- a/lib/librte_eventdev/rte_eventdev.c
> > +++ b/lib/librte_eventdev/rte_eventdev.c
> > @@ -122,7 +122,11 @@ rte_event_eth_rx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
> >
> >         if (caps == NULL)
> >                 return -EINVAL;
> > -       *caps = 0;
> > +
> > +       if (dev->dev_ops->eth_rx_adapter_caps_get == NULL)
> > +               *caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
> > +       else
> > +               *caps = 0;
> >
> >         return dev->dev_ops->eth_rx_adapter_caps_get ?
> >                                 (*dev->dev_ops->eth_rx_adapter_caps_get)(dev,
> > --
> > 2.17.1
> >



^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v9 0/8] Introduce event vectorization
  2021-03-26 14:08             ` [dpdk-dev] [PATCH v8 0/8] Introduce event vectorization pbhagavatula
                                 ` (7 preceding siblings ...)
  2021-03-26 14:08               ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v8 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
@ 2021-03-30  8:22               ` pbhagavatula
  2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 1/8] eventdev: introduce event vector capability pbhagavatula
                                   ` (8 more replies)
  8 siblings, 9 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-30  8:22 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

In traditional event programming model, events are identified by a
flow-id and a uintptr_t. The flow-id uniquely identifies a given event
and determines the order of scheduling based on schedule type, the
uintptr_t holds a single object.

Event devices also support burst mode with configurable dequeue depth,
i.e. each dequeue call would return multiple events and each event
might be at a different stage of the pipeline.
Having a burst of events belonging to different stages in a dequeue
burst is not only difficult to vectorize but also increases the scheduler
overhead and application overhead of pipelining events further.
Using event vectors we see a performance gain of ~742.3% as shown in [1].

By introducing event vectorization, each event will be capable of holding
multiple uintptr_t of the same flow thereby allowing applications
to vectorize their pipeline and reduce the complexity of pipelining
events across multiple stages. This also reduces the complexity of handling
enqueue and dequeue on an event device.

Since event devices are transparent to the events they are scheduling
so the event producers such as eth_rx_adapter, crypto_adapter , etc..
are responsible for vectorizing the buffers of the same flow into a single
event.

The series also breaks ABI in the patch [8/8] which is targetted to the
v21.11 release.

The dpdk-test-eventdev application has been updated with options to test
multiple vector sizes and timeouts.

[1]
As for performance improvement, with a ARM Cortex-A72 equivalent processer,
software event device (--vdev=event_sw0), single worker core, single stage
and using one service core for Rx adapter, Tx adapter, Scheduling.

Without this patchset applied:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
         --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
         --stlist=a --wlcores=20
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    5.071 mpps

With the patchset applied and Without event vectorization:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
         --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
         --stlist=a --wlcores=20
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    5.123 mpps

With event vectorization:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
        --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
        --stlist=a --wlcores=20 --enable_vector --nb_eth_queues 1
        --vector_size 256
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    42.715 mpps

Having dedicated service cores for each Rx queues and tweaking the vector,
dequeue burst size would further improve performance.

API usage is shown below:

Configuration:

	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

	vector_pool = rte_event_vector_pool_create("vector_pool",
			nb_elem, 0, vector_size, socket_id);

	rte_event_eth_rx_adapter_create(id, event_id, &adptr_conf);
	rte_event_eth_rx_adapter_queue_add(id, eth_id, -1, &queue_conf);
	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
		vec_conf.vector_sz = vector_size;
		vec_conf.vector_timeout_ns = vector_tmo_nsec;
		vec_conf.vector_mp = vector_pool;
		rte_event_eth_rx_adapter_queue_event_vector_config(id,
				eth_id, -1, &vec_conf);
	}

Fastpath:

	num = rte_event_dequeue_burst(event_id, port_id, &ev, 1, 0);
	if (!num)
		continue;

	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
		switch (ev.event_type) {
		case RTE_EVENT_TYPE_ETHDEV_VECTOR:
		case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
			struct rte_mbuf **mbufs;

			mbufs = ev.vector_ev->mbufs;
			for (i = 0; i < ev.vector_ev->nb_elem; i++)
				//Process mbufs.
			break;
		case ...
		}
	}
	...

v9 Changes:
- Update Rx adapter documentation w.r.t SW event vectorizations. (Jay)
- Push partial vectors to event device on queue delete. (Jay)

v8 Changes:
- Fix incorrect shift for vector timeout interval.(Jay)
- Code reallocation.(Jay)

v7 Changes:
- More doxygen fixes.(Jay)
- Reduce code duplication in 4/8.(Jay)

v6 Changes:
- Make rte_errno sign consistant.(Jay)
- Gramatical and doxygen fixes. (Jay)

v5 Changes:
- Make `rte_event_vector_pool_create non-inline` to ease ABI stability.(Ray)
- Move `rte_event_eth_rx_adapter_queue_event_vector_config` and
  `rte_event_eth_rx_adapter_vector_limits_get` implementation to the patch
  where they are initially defined.(Ray)
- Multiple gramatical and style fixes.(Jerin)
- Add missing release notes.(Jerin)

v4 Changes:
- Fix missing event vector structure in event structure.(Jay)

v3 Changes:
- Fix unintended formatting changes.

v2 Changes:
- Multiple gramatical and style fixes.(Jerin)
- Add parameter to define vector size in power of 2. (Jerin)
- Redo patch series w/o breaking ABI till the last patch.(David)
- Add deprication notice to announce ABI break in 21.11.(David)
- Add vector limits validation to app/test-eventdev.

Pavan Nikhilesh (8):
  eventdev: introduce event vector capability
  eventdev: introduce event vector Rx capability
  eventdev: introduce event vector Tx capability
  eventdev: add Rx adapter event vector support
  eventdev: add Tx adapter event vector support
  app/eventdev: add event vector mode in pipeline test
  doc: announce event Rx adapter config changes
  eventdev: simplify Rx adapter event vector config

 app/test-eventdev/evt_common.h                |   4 +
 app/test-eventdev/evt_options.c               |  52 +++
 app/test-eventdev/evt_options.h               |   4 +
 app/test-eventdev/test_pipeline_atq.c         | 310 +++++++++++++++--
 app/test-eventdev/test_pipeline_common.c      | 105 +++++-
 app/test-eventdev/test_pipeline_common.h      |  18 +
 app/test-eventdev/test_pipeline_queue.c       | 320 +++++++++++++++--
 .../prog_guide/event_ethernet_rx_adapter.rst  |  49 +++
 .../prog_guide/event_ethernet_tx_adapter.rst  |  12 +
 doc/guides/prog_guide/eventdev.rst            |  36 +-
 doc/guides/rel_notes/deprecation.rst          |   9 +
 doc/guides/rel_notes/release_21_05.rst        |   8 +
 doc/guides/tools/testeventdev.rst             |  45 ++-
 lib/librte_eventdev/eventdev_pmd.h            |  31 +-
 .../rte_event_eth_rx_adapter.c                | 327 +++++++++++++++++-
 .../rte_event_eth_rx_adapter.h                |  78 +++++
 .../rte_event_eth_tx_adapter.c                |  66 +++-
 lib/librte_eventdev/rte_eventdev.c            |  53 ++-
 lib/librte_eventdev/rte_eventdev.h            | 114 +++++-
 lib/librte_eventdev/version.map               |   4 +
 20 files changed, 1554 insertions(+), 91 deletions(-)

--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v9 1/8] eventdev: introduce event vector capability
  2021-03-30  8:22               ` [dpdk-dev] [PATCH v9 0/8] Introduce event vectorization pbhagavatula
@ 2021-03-30  8:22                 ` pbhagavatula
  2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 2/8] eventdev: introduce event vector Rx capability pbhagavatula
                                   ` (7 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-30  8:22 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce rte_event_vector datastructure which is capable of holding
multiple uintptr_t of the same flow thereby allowing applications
to vectorize their pipeline and reducing the complexity of pipelining
the events across multiple stages.
This approach also reduces the scheduling overhead on a event device.

Add a event vector mempool create handler to create mempools based on
the best mempool ops available on a given platform.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
Acked-by: Ray Kinsella <mdr@ashroe.eu>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
---
 doc/guides/prog_guide/eventdev.rst     | 36 ++++++++++-
 doc/guides/rel_notes/release_21_05.rst |  8 +++
 lib/librte_eventdev/rte_eventdev.c     | 42 +++++++++++++
 lib/librte_eventdev/rte_eventdev.h     | 82 +++++++++++++++++++++++++-
 lib/librte_eventdev/version.map        |  3 +
 5 files changed, 168 insertions(+), 3 deletions(-)

diff --git a/doc/guides/prog_guide/eventdev.rst b/doc/guides/prog_guide/eventdev.rst
index ccde086f6..fda9c3743 100644
--- a/doc/guides/prog_guide/eventdev.rst
+++ b/doc/guides/prog_guide/eventdev.rst
@@ -63,13 +63,45 @@ the actual event being scheduled is. The payload is a union of the following:
 * ``uint64_t u64``
 * ``void *event_ptr``
 * ``struct rte_mbuf *mbuf``
+* ``struct rte_event_vector *vec``
 
-These three items in a union occupy the same 64 bits at the end of the rte_event
+These four items in a union occupy the same 64 bits at the end of the rte_event
 structure. The application can utilize the 64 bits directly by accessing the
-u64 variable, while the event_ptr and mbuf are provided as convenience
+u64 variable, while the event_ptr, mbuf, vec are provided as a convenience
 variables.  For example the mbuf pointer in the union can used to schedule a
 DPDK packet.
 
+Event Vector
+~~~~~~~~~~~~
+
+The rte_event_vector struct contains a vector of elements defined by the event
+type specified in the ``rte_event``. The event_vector structure contains the
+following data:
+
+* ``nb_elem`` - The number of elements held within the vector.
+
+Similar to ``rte_event`` the payload of event vector is also a union, allowing
+flexibility in what the actual vector is.
+
+* ``struct rte_mbuf *mbufs[0]`` - An array of mbufs.
+* ``void *ptrs[0]`` - An array of pointers.
+* ``uint64_t *u64s[0]`` - An array of uint64_t elements.
+
+The size of the event vector is related to the total number of elements it is
+configured to hold, this is achieved by making `rte_event_vector` a variable
+length structure.
+A helper function is provided to create a mempool that holds event vector, which
+takes name of the pool, total number of required ``rte_event_vector``,
+cache size, number of elements in each ``rte_event_vector`` and socket id.
+
+.. code-block:: c
+
+        rte_event_vector_pool_create("vector_pool", nb_event_vectors, cache_sz,
+                                     nb_elements_per_vector, socket_id);
+
+The function ``rte_event_vector_pool_create`` creates mempool with the best
+platform mempool ops.
+
 Queues
 ~~~~~~
 
diff --git a/doc/guides/rel_notes/release_21_05.rst b/doc/guides/rel_notes/release_21_05.rst
index e2b0886a9..4cbf2769f 100644
--- a/doc/guides/rel_notes/release_21_05.rst
+++ b/doc/guides/rel_notes/release_21_05.rst
@@ -106,6 +106,14 @@ New Features
   * Added support for periodic timer mode in eventdev timer adapter.
   * Added support for periodic timer mode in octeontx2 event device driver.
 
+* **Add Event device vector capability.**
+
+  * Added ``rte_event_vector`` data structure which is capable of holding
+    multiple ``uintptr_t`` of the same flow thereby allowing applications
+    to vectorize their pipelines and also reduce the complexity of pipelining
+    the events across multiple stages.
+  * This also reduces the scheduling overhead on a event device.
+
 
 Removed Items
 -------------
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index b57363f80..be0499c52 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -1266,6 +1266,48 @@ int rte_event_dev_selftest(uint8_t dev_id)
 	return -ENOTSUP;
 }
 
+struct rte_mempool *
+rte_event_vector_pool_create(const char *name, unsigned int n,
+			     unsigned int cache_size, uint16_t nb_elem,
+			     int socket_id)
+{
+	const char *mp_ops_name;
+	struct rte_mempool *mp;
+	unsigned int elt_sz;
+	int ret;
+
+	if (!nb_elem) {
+		RTE_LOG(ERR, EVENTDEV,
+			"Invalid number of elements=%d requested\n", nb_elem);
+		rte_errno = EINVAL;
+		return NULL;
+	}
+
+	elt_sz =
+		sizeof(struct rte_event_vector) + (nb_elem * sizeof(uintptr_t));
+	mp = rte_mempool_create_empty(name, n, elt_sz, cache_size, 0, socket_id,
+				      0);
+	if (mp == NULL)
+		return NULL;
+
+	mp_ops_name = rte_mbuf_best_mempool_ops();
+	ret = rte_mempool_set_ops_byname(mp, mp_ops_name, NULL);
+	if (ret != 0) {
+		RTE_LOG(ERR, EVENTDEV, "error setting mempool handler\n");
+		goto err;
+	}
+
+	ret = rte_mempool_populate_default(mp);
+	if (ret < 0)
+		goto err;
+
+	return mp;
+err:
+	rte_mempool_free(mp);
+	rte_errno = -ret;
+	return NULL;
+}
+
 int
 rte_event_dev_start(uint8_t dev_id)
 {
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index 9fc39e9ca..dee468ed0 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -212,8 +212,10 @@ extern "C" {
 
 #include <rte_common.h>
 #include <rte_config.h>
-#include <rte_memory.h>
 #include <rte_errno.h>
+#include <rte_mbuf_pool_ops.h>
+#include <rte_memory.h>
+#include <rte_mempool.h>
 
 #include "rte_eventdev_trace_fp.h"
 
@@ -913,6 +915,31 @@ rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
 int
 rte_event_dev_close(uint8_t dev_id);
 
+/**
+ * Event vector structure.
+ */
+struct rte_event_vector {
+	uint64_t nb_elem : 16;
+	/**< Number of elements in this event vector. */
+	uint64_t rsvd : 48;
+	/**< Reserved for future use */
+	uint64_t impl_opaque;
+	/**< Implementation specific opaque value.
+	 * An implementation may use this field to hold implementation specific
+	 * value to share between dequeue and enqueue operation.
+	 * The application should not modify this field.
+	 */
+	union {
+		struct rte_mbuf *mbufs[0];
+		void *ptrs[0];
+		uint64_t *u64s[0];
+	} __rte_aligned(16);
+	/**< Start of the vector array union. Depending upon the event type the
+	 * vector array can be an array of mbufs or pointers or opaque u64
+	 * values.
+	 */
+};
+
 /* Scheduler type definitions */
 #define RTE_SCHED_TYPE_ORDERED          0
 /**< Ordered scheduling
@@ -986,6 +1013,21 @@ rte_event_dev_close(uint8_t dev_id);
  */
 #define RTE_EVENT_TYPE_ETH_RX_ADAPTER   0x4
 /**< The event generated from event eth Rx adapter */
+#define RTE_EVENT_TYPE_VECTOR           0x8
+/**< Indicates that event is a vector.
+ * All vector event types should be a logical OR of EVENT_TYPE_VECTOR.
+ * This simplifies the pipeline design as one can split processing the events
+ * between vector events and normal event across event types.
+ * Example:
+ *	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
+ *		// Classify and handle vector event.
+ *	} else {
+ *		// Classify and handle event.
+ *	}
+ */
+#define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
+/**< The event vector generated from cpu for pipelining. */
+
 #define RTE_EVENT_TYPE_MAX              0x10
 /**< Maximum number of event types */
 
@@ -1108,6 +1150,8 @@ struct rte_event {
 		/**< Opaque event pointer */
 		struct rte_mbuf *mbuf;
 		/**< mbuf pointer if dequeued event is associated with mbuf */
+		struct rte_event_vector *vec;
+		/**< Event vector pointer. */
 	};
 };
 
@@ -2026,6 +2070,42 @@ rte_event_dev_xstats_reset(uint8_t dev_id,
  */
 int rte_event_dev_selftest(uint8_t dev_id);
 
+/**
+ * Get the memory required per event vector based on the number of elements per
+ * vector.
+ * This should be used to create the mempool that holds the event vectors.
+ *
+ * @param name
+ *   The name of the vector pool.
+ * @param n
+ *   The number of elements in the mbuf pool.
+ * @param cache_size
+ *   Size of the per-core object cache. See rte_mempool_create() for
+ *   details.
+ * @param nb_elem
+ *   The number of elements that a single event vector should be able to hold.
+ * @param socket_id
+ *   The socket identifier where the memory should be allocated. The
+ *   value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
+ *   reserved zone
+ *
+ * @return
+ *   The pointer to the newly allocated mempool, on success. NULL on error
+ *   with rte_errno set appropriately. Possible rte_errno values include:
+ *    - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
+ *    - E_RTE_SECONDARY - function was called from a secondary process instance
+ *    - EINVAL - cache size provided is too large, or priv_size is not aligned.
+ *    - ENOSPC - the maximum number of memzones has already been allocated
+ *    - EEXIST - a memzone with the same name already exists
+ *    - ENOMEM - no appropriate memory area found in which to create memzone
+ *    - ENAMETOOLONG - mempool name requested is too long.
+ */
+__rte_experimental
+struct rte_mempool *
+rte_event_vector_pool_create(const char *name, unsigned int n,
+			     unsigned int cache_size, uint16_t nb_elem,
+			     int socket_id);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index 3e5c09cfd..a070ef56e 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -138,6 +138,9 @@ EXPERIMENTAL {
 	__rte_eventdev_trace_port_setup;
 	# added in 20.11
 	rte_event_pmd_pci_probe_named;
+
+	#added in 21.05
+	rte_event_vector_pool_create;
 };
 
 INTERNAL {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v9 2/8] eventdev: introduce event vector Rx capability
  2021-03-30  8:22               ` [dpdk-dev] [PATCH v9 0/8] Introduce event vectorization pbhagavatula
  2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 1/8] eventdev: introduce event vector capability pbhagavatula
@ 2021-03-30  8:22                 ` pbhagavatula
  2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 3/8] eventdev: introduce event vector Tx capability pbhagavatula
                                   ` (6 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-30  8:22 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce event ethernet Rx adapter event vector capability.

If an event eth Rx adapter has the capability of
RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR then a given Rx queue
can be configured to enable event vectorization by passing the
flag RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR to
rte_event_eth_rx_adapter_queue_conf::rx_queue_flags while configuring
Rx adapter through rte_event_eth_rx_adapter_queue_add().

The max vector size, vector timeout define the vector size and
mempool used for allocating vector event are configured through
rte_event_eth_rx_adapter_queue_add. The element size of the element
in the vector pool should be equal to
    sizeof(struct rte_event_vector) + (vector_sz * sizeof(uintptr_t))

Application can use `rte_event_vector_pool_create` to create the
vector mempool used for
rte_event_eth_rx_adapter_queue_conf::vector_mp.

The Rx adapter would be responsible for vectorizing the mbufs
based on the flow, the vector limits configured by the application
and add the vector event of mbufs to the event queue set via
rte_event_eth_rx_adapter_queue_conf::ev::queue_id.
It should also mark rte_event_vector::union_valid and fill
rte_event_vector::port, rte_event_vector::queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
Acked-by: Ray Kinsella <mdr@ashroe.eu>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
---
 .../prog_guide/event_ethernet_rx_adapter.rst  |  38 ++++++
 lib/librte_eventdev/eventdev_pmd.h            |  53 ++++++++
 .../rte_event_eth_rx_adapter.c                | 114 ++++++++++++++++++
 .../rte_event_eth_rx_adapter.h                | 105 ++++++++++++++++
 lib/librte_eventdev/rte_eventdev.h            |  30 ++++-
 lib/librte_eventdev/version.map               |   2 +
 6 files changed, 340 insertions(+), 2 deletions(-)

diff --git a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
index cb44ce0e4..5eefef355 100644
--- a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
@@ -186,3 +186,41 @@ the event buffer fill level is low. The
 ``rte_event_eth_rx_adapter_cb_register()`` function allow the application
 to register a callback that selects which packets to enqueue to the event
 device.
+
+Rx event vectorization
+~~~~~~~~~~~~~~~~~~~~~~
+
+The event devices, ethernet device pairs which support the capability
+``RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR`` can aggregate packets based on
+flow characteristics and generate a ``rte_event`` containing ``rte_event_vector``
+whose event type is either ``RTE_EVENT_TYPE_ETHDEV_VECTOR`` or
+``RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR``.
+The aggregation size and timeout are configurable at a queue level and the
+maximum, minimum vector sizes and timeouts vary based on the device capability
+and can be queried using ``rte_event_eth_rx_adapter_vector_limits_get``.
+The Rx adapter additionally might include useful data such as ethernet device
+port and queue identifier in the ``rte_event_vector::port`` and
+``rte_event_vector::queue`` and mark ``rte_event_vector::attr_valid`` as true.
+
+A loop processing ``rte_event_vector`` containing mbufs is shown below.
+
+.. code-block:: c
+
+        event = rte_event_dequeue_burst(event_dev, event_port, &event,
+                                        1, 0);
+        if (!event)
+                continue;
+
+        switch (ev.event_type) {
+        case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
+        case RTE_EVENT_TYPE_ETHDEV_VECTOR:
+                struct rte_mbufs **mbufs;
+
+                mbufs = (struct rte_mbufs **)ev[i].vec->mbufs;
+                for (i = 0; i < ev.vec->nb_elem; i++) {
+                        /* Process each mbuf. */
+                }
+        break;
+        case ...
+        ...
+        }
diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 7eb9a7739..9297f1433 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -645,6 +645,53 @@ typedef int (*eventdev_eth_rx_adapter_stats_reset)
  */
 typedef int (*eventdev_selftest)(void);
 
+struct rte_event_eth_rx_adapter_vector_limits;
+/**
+ * Get event vector limits for a given event, ethernet device pair.
+ *
+ * @param dev
+ *   Event device pointer
+ *
+ * @param eth_dev
+ *   Ethernet device pointer
+ *
+ * @param[out] limits
+ *   Pointer to the limits structure to be filled.
+ *
+ * @return
+ *   - 0: Success.
+ *   - <0: Error code returned by the driver function.
+ */
+typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
+	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
+	struct rte_event_eth_rx_adapter_vector_limits *limits);
+
+struct rte_event_eth_rx_adapter_event_vector_config;
+/**
+ * Enable event vector on an given Rx queue of a ethernet devices belonging to
+ * the Rx adapter.
+ *
+ * @param dev
+ *   Event device pointer
+ *
+ * @param eth_dev
+ *   Ethernet device pointer
+ *
+ * @param rx_queue_id
+ *   The Rx queue identifier
+ *
+ * @param config
+ *   Pointer to the event vector configuration structure.
+ *
+ * @return
+ *   - 0: Success.
+ *   - <0: Error code returned by the driver function.
+ */
+typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
+	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
+	int32_t rx_queue_id,
+	const struct rte_event_eth_rx_adapter_event_vector_config *config);
+
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;
 
@@ -1067,6 +1114,12 @@ struct rte_eventdev_ops {
 	/**< Get ethernet Rx stats */
 	eventdev_eth_rx_adapter_stats_reset eth_rx_adapter_stats_reset;
 	/**< Reset ethernet Rx stats */
+	eventdev_eth_rx_adapter_vector_limits_get_t
+		eth_rx_adapter_vector_limits_get;
+	/**< Get event vector limits for the Rx adapter */
+	eventdev_eth_rx_adapter_event_vector_config_t
+		eth_rx_adapter_event_vector_config;
+	/**< Configure Rx adapter with event vector */
 
 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index d8c635e99..ac8ba5bf0 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -2263,6 +2263,120 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }
 
+int
+rte_event_eth_rx_adapter_queue_event_vector_config(
+	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
+	struct rte_event_eth_rx_adapter_event_vector_config *config)
+{
+	struct rte_event_eth_rx_adapter_vector_limits limits;
+	struct rte_event_eth_rx_adapter *rx_adapter;
+	struct rte_eventdev *dev;
+	uint32_t cap;
+	int ret;
+
+	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
+
+	rx_adapter = rxa_id_to_adapter(id);
+	if ((rx_adapter == NULL) || (config == NULL))
+		return -EINVAL;
+
+	dev = &rte_eventdevs[rx_adapter->eventdev_id];
+	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
+						eth_dev_id, &cap);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
+				 "eth port %" PRIu16,
+				 id, eth_dev_id);
+		return ret;
+	}
+
+	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
+		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
+	ret = rte_event_eth_rx_adapter_vector_limits_get(
+		rx_adapter->eventdev_id, eth_dev_id, &limits);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
+				 "eth port %" PRIu16,
+				 rx_adapter->eventdev_id, eth_dev_id);
+		return ret;
+	}
+
+	if (config->vector_sz < limits.min_sz ||
+	    config->vector_sz > limits.max_sz ||
+	    config->vector_timeout_ns < limits.min_timeout_ns ||
+	    config->vector_timeout_ns > limits.max_timeout_ns ||
+	    config->vector_mp == NULL) {
+		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+	if (config->vector_mp->elt_size <
+	    (sizeof(struct rte_event_vector) +
+	     (sizeof(uintptr_t) * config->vector_sz))) {
+		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
+	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
+		RTE_FUNC_PTR_OR_ERR_RET(
+			*dev->dev_ops->eth_rx_adapter_event_vector_config,
+			-ENOTSUP);
+		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
+			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
+	} else {
+		ret = -ENOTSUP;
+	}
+
+	return ret;
+}
+
+int
+rte_event_eth_rx_adapter_vector_limits_get(
+	uint8_t dev_id, uint16_t eth_port_id,
+	struct rte_event_eth_rx_adapter_vector_limits *limits)
+{
+	struct rte_eventdev *dev;
+	uint32_t cap;
+	int ret;
+
+	RTE_EVENTDEV_VALID_DEVID_OR_ERR_RET(dev_id, -EINVAL);
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_port_id, -EINVAL);
+
+	if (limits == NULL)
+		return -EINVAL;
+
+	dev = &rte_eventdevs[dev_id];
+
+	ret = rte_event_eth_rx_adapter_caps_get(dev_id, eth_port_id, &cap);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
+				 "eth port %" PRIu16,
+				 dev_id, eth_port_id);
+		return ret;
+	}
+
+	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
+		RTE_FUNC_PTR_OR_ERR_RET(
+			*dev->dev_ops->eth_rx_adapter_vector_limits_get,
+			-ENOTSUP);
+		ret = dev->dev_ops->eth_rx_adapter_vector_limits_get(
+			dev, &rte_eth_devices[eth_port_id], limits);
+	} else {
+		ret = -ENOTSUP;
+	}
+
+	return ret;
+}
+
 int
 rte_event_eth_rx_adapter_start(uint8_t id)
 {
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
index 21bb1e54c..7407cde00 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
@@ -92,6 +92,10 @@ extern "C" {
 /**< This flag indicates the flow identifier is valid
  * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
  */
+#define RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR	0x2
+/**< This flag indicates that mbufs arriving on the queue need to be vectorized
+ * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+ */
 
 /**
  * Adapter configuration structure that the adapter configuration callback
@@ -169,6 +173,36 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 */
 };
 
+struct rte_event_eth_rx_adapter_event_vector_config {
+	uint16_t vector_sz;
+	/**<
+	 * Indicates the maximum number for mbufs to combine and form a vector.
+	 * Should be within
+	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_sz
+	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_sz
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+	 */
+	uint64_t vector_timeout_ns;
+	/**<
+	 * Indicates the maximum number of nanoseconds to wait for receiving
+	 * mbufs. Should be within vectorization limits of the
+	 * adapter
+	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_ns
+	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_ns
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+	 */
+	struct rte_mempool *vector_mp;
+	/**<
+	 * Indicates the mempool that should be used for allocating
+	 * rte_event_vector container.
+	 * Should be created by using `rte_event_vector_pool_create`.
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags.
+	 */
+};
+
 /**
  * A structure used to retrieve statistics for an eth rx adapter instance.
  */
@@ -199,6 +233,32 @@ struct rte_event_eth_rx_adapter_stats {
 	/**< Received packet count for interrupt mode Rx queues */
 };
 
+/**
+ * A structure used to retrieve eth rx adapter vector limits.
+ */
+struct rte_event_eth_rx_adapter_vector_limits {
+	uint16_t min_sz;
+	/**< Minimum vector limit configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_sz
+	 */
+	uint16_t max_sz;
+	/**< Maximum vector limit configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_sz
+	 */
+	uint8_t log2_sz;
+	/**< True if the size configured should be in log2.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_sz
+	 */
+	uint64_t min_timeout_ns;
+	/**< Minimum vector timeout configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_timeout_ns
+	 */
+	uint64_t max_timeout_ns;
+	/**< Maximum vector timeout configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_timeout_ns
+	 */
+};
+
 /**
  *
  * Callback function invoked by the SW adapter before it continues
@@ -467,6 +527,51 @@ int rte_event_eth_rx_adapter_cb_register(uint8_t id, uint16_t eth_dev_id,
 					 rte_event_eth_rx_adapter_cb_fn cb_fn,
 					 void *cb_arg);
 
+/**
+ * Retrieve vector limits for a given event dev and eth dev pair.
+ * @see rte_event_eth_rx_adapter_vector_limits
+ *
+ * @param dev_id
+ *  Event device identifier.
+ * @param eth_port_id
+ *  Port identifier of the ethernet device.
+ * @param [out] limits
+ *  A pointer to rte_event_eth_rx_adapter_vector_limits structure that has to
+ * be filled.
+ *
+ * @return
+ *  - 0: Success.
+ *  - <0: Error code on failure.
+ */
+__rte_experimental
+int rte_event_eth_rx_adapter_vector_limits_get(
+	uint8_t dev_id, uint16_t eth_port_id,
+	struct rte_event_eth_rx_adapter_vector_limits *limits);
+
+/**
+ * Configure event vectorization for a given ethernet device queue, that has
+ * been added to a event eth Rx adapter.
+ *
+ * @param id
+ *  The identifier of the ethernet Rx event adapter.
+ *
+ * @param eth_dev_id
+ *  The identifier of the ethernet device.
+ *
+ * @param rx_queue_id
+ *  Ethernet device receive queue index.
+ *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
+ *  are configured with event vectorization.
+ *
+ * @return
+ *  - 0: Success, Receive queue configured correctly.
+ *  - <0: Error code on failure.
+ */
+__rte_experimental
+int rte_event_eth_rx_adapter_queue_event_vector_config(
+	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
+	struct rte_event_eth_rx_adapter_event_vector_config *config);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index dee468ed0..936c97cc2 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -919,10 +919,28 @@ rte_event_dev_close(uint8_t dev_id);
  * Event vector structure.
  */
 struct rte_event_vector {
-	uint64_t nb_elem : 16;
+	uint16_t nb_elem;
 	/**< Number of elements in this event vector. */
-	uint64_t rsvd : 48;
+	uint16_t rsvd : 15;
 	/**< Reserved for future use */
+	uint16_t attr_valid : 1;
+	/**< Indicates that the below union attributes have valid information.
+	 */
+	union {
+		/* Used by Rx adapter.
+		 * Indicates that all the elements in this vector belong to the
+		 * same port and queue pair when originating from Rx adapter,
+		 * valid only when event type is ETHDEV_VECTOR or
+		 * ETH_RX_ADAPTER_VECTOR.
+		 */
+		struct {
+			uint16_t port;
+			/* Ethernet device port id. */
+			uint16_t queue;
+			/* Ethernet device queue id. */
+		};
+	};
+	/**< Union to hold common attributes of the vector array. */
 	uint64_t impl_opaque;
 	/**< Implementation specific opaque value.
 	 * An implementation may use this field to hold implementation specific
@@ -1025,8 +1043,14 @@ struct rte_event_vector {
  *		// Classify and handle event.
  *	}
  */
+#define RTE_EVENT_TYPE_ETHDEV_VECTOR                                           \
+	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETHDEV)
+/**< The event vector generated from ethdev subsystem */
 #define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
 /**< The event vector generated from cpu for pipelining. */
+#define RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR                                   \
+	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETH_RX_ADAPTER)
+/**< The event vector generated from eth Rx adapter. */
 
 #define RTE_EVENT_TYPE_MAX              0x10
 /**< Maximum number of event types */
@@ -1171,6 +1195,8 @@ struct rte_event {
  * @see struct rte_event_eth_rx_adapter_queue_conf::ev
  * @see struct rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
  */
+#define RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR	0x8
+/**< Adapter supports event vectorization per ethdev. */
 
 /**
  * Retrieve the event device's ethdev Rx adapter capabilities for the
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index a070ef56e..902df0ae3 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -141,6 +141,8 @@ EXPERIMENTAL {
 
 	#added in 21.05
 	rte_event_vector_pool_create;
+	rte_event_eth_rx_adapter_vector_limits_get;
+	rte_event_eth_rx_adapter_queue_event_vector_config;
 };
 
 INTERNAL {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v9 3/8] eventdev: introduce event vector Tx capability
  2021-03-30  8:22               ` [dpdk-dev] [PATCH v9 0/8] Introduce event vectorization pbhagavatula
  2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 1/8] eventdev: introduce event vector capability pbhagavatula
  2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 2/8] eventdev: introduce event vector Rx capability pbhagavatula
@ 2021-03-30  8:22                 ` pbhagavatula
  2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 4/8] eventdev: add Rx adapter event vector support pbhagavatula
                                   ` (5 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-30  8:22 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce event vector transmit capability for event eth
tx adapter.

The capability indicates that the Tx adapter is capable of
transmitting event vectors.
When rte_event_vector::union_valid is set, the Tx adapter should
transmit all the packets to the rte_event_vector::port using the
rte_event_vector::queue.
If rte_event_vector::union_valid is not set then the Tx adapter
should peek into each mbuf to get the destination port and queue
pair.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
---
 doc/guides/prog_guide/event_ethernet_tx_adapter.rst | 12 ++++++++++++
 lib/librte_eventdev/rte_eventdev.h                  |  8 +++++++-
 2 files changed, 19 insertions(+), 1 deletion(-)

diff --git a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
index a8c13e136..f80d22621 100644
--- a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
@@ -164,3 +164,15 @@ The  ``rte_event_eth_tx_adapter_stats_get()`` function reports counters defined
 in struct ``rte_event_eth_tx_adapter_stats``. The counter values are the sum of
 the counts from the eventdev PMD callback if the callback is supported, and
 the counts maintained by the service function, if one exists.
+
+Tx event vectorization
+~~~~~~~~~~~~~~~~~~~~~~
+
+The event device, ethernet device pairs which support the capability
+``RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR`` can process event vector of mbufs.
+Additionally, application can provide a hint to the Tx adapter that all the
+mbufs are destined to the same ethernet port and queue by setting the bit
+``rte_event_vector::attr_valid`` and filling `rte_event_vector::port`` and
+``rte_event_vector::queue``.
+If ``rte_event_vector::attr_valid`` is not set then the Tx adapter should peek
+into each mbuf and transmit them to the requested ethernet port and queue pair.
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index 936c97cc2..5f1f544cc 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -927,11 +927,13 @@ struct rte_event_vector {
 	/**< Indicates that the below union attributes have valid information.
 	 */
 	union {
-		/* Used by Rx adapter.
+		/* Used by Rx/Tx adapter.
 		 * Indicates that all the elements in this vector belong to the
 		 * same port and queue pair when originating from Rx adapter,
 		 * valid only when event type is ETHDEV_VECTOR or
 		 * ETH_RX_ADAPTER_VECTOR.
+		 * Can also be used to indicate the Tx adapter the destination
+		 * port and queue of the mbufs in the vector
 		 */
 		struct {
 			uint16_t port;
@@ -1296,6 +1298,10 @@ rte_event_crypto_adapter_caps_get(uint8_t dev_id, uint8_t cdev_id,
 #define RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT	0x1
 /**< This flag is sent when the PMD supports a packet transmit callback
  */
+#define RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR	0x2
+/**< Indicates that the Tx adapter is capable of handling event vector of
+ * mbufs.
+ */
 
 /**
  * Retrieve the event device's eth Tx adapter capabilities
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v9 4/8] eventdev: add Rx adapter event vector support
  2021-03-30  8:22               ` [dpdk-dev] [PATCH v9 0/8] Introduce event vectorization pbhagavatula
                                   ` (2 preceding siblings ...)
  2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 3/8] eventdev: introduce event vector Tx capability pbhagavatula
@ 2021-03-30  8:22                 ` pbhagavatula
  2021-03-31  6:35                   ` Jayatheerthan, Jay
  2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 5/8] eventdev: add Tx " pbhagavatula
                                   ` (4 subsequent siblings)
  8 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-03-30  8:22 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support for event eth Rx adapter, the implementation
creates vector flows based on port and queue identifier of the received
mbufs.
The flow id for SW Rx event vectorization will use 12-bits of queue
identifier and 8-bits port identifier when custom flow id is not set
for simplicity.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 .../prog_guide/event_ethernet_rx_adapter.rst  |  11 +
 lib/librte_eventdev/eventdev_pmd.h            |   7 +-
 .../rte_event_eth_rx_adapter.c                | 278 ++++++++++++++++--
 lib/librte_eventdev/rte_eventdev.c            |   6 +-
 4 files changed, 278 insertions(+), 24 deletions(-)

diff --git a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
index 5eefef355..06fa864fa 100644
--- a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
@@ -224,3 +224,14 @@ A loop processing ``rte_event_vector`` containing mbufs is shown below.
         case ...
         ...
         }
+
+Rx event vectorization for SW Rx adapter
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+For SW based event vectorization, i.e., when the
+``RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT`` is not set in the adapter's
+capabilities flags for a particular ethernet device, the service function
+creates a single event vector flow for all the mbufs arriving on the given
+Rx queue.
+The 20-bit event flow identifier is set to 12-bits of Rx queue identifier
+and 8-bits of ethernet device identifier.
diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 9297f1433..0f724ac85 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -69,9 +69,10 @@ extern "C" {
 	} \
 } while (0)
 
-#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP \
-		((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) | \
-			(RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ))
+#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP                                        \
+	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |                     \
+	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) |                         \
+	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR))
 
 #define RTE_EVENT_CRYPTO_ADAPTER_SW_CAP \
 		RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index ac8ba5bf0..e273b3acf 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -26,6 +26,10 @@
 #define BATCH_SIZE		32
 #define BLOCK_CNT_THRESHOLD	10
 #define ETH_EVENT_BUFFER_SIZE	(4*BATCH_SIZE)
+#define MAX_VECTOR_SIZE		1024
+#define MIN_VECTOR_SIZE		4
+#define MAX_VECTOR_NS		1E9
+#define MIN_VECTOR_NS		1E5
 
 #define ETH_RX_ADAPTER_SERVICE_NAME_LEN	32
 #define ETH_RX_ADAPTER_MEM_NAME_LEN	32
@@ -59,6 +63,20 @@ struct eth_rx_poll_entry {
 	uint16_t eth_rx_qid;
 };
 
+struct eth_rx_vector_data {
+	TAILQ_ENTRY(eth_rx_vector_data) next;
+	uint16_t port;
+	uint16_t queue;
+	uint16_t max_vector_count;
+	uint64_t event;
+	uint64_t ts;
+	uint64_t vector_timeout_ticks;
+	struct rte_mempool *vector_pool;
+	struct rte_event_vector *vector_ev;
+} __rte_cache_aligned;
+
+TAILQ_HEAD(eth_rx_vector_data_list, eth_rx_vector_data);
+
 /* Instance per adapter */
 struct rte_eth_event_enqueue_buffer {
 	/* Count of events in this buffer */
@@ -92,6 +110,14 @@ struct rte_event_eth_rx_adapter {
 	uint32_t wrr_pos;
 	/* Event burst buffer */
 	struct rte_eth_event_enqueue_buffer event_enqueue_buffer;
+	/* Vector enable flag */
+	uint8_t ena_vector;
+	/* Timestamp of previous vector expiry list traversal */
+	uint64_t prev_expiry_ts;
+	/* Minimum ticks to wait before traversing expiry list */
+	uint64_t vector_tmo_ticks;
+	/* vector list */
+	struct eth_rx_vector_data_list vector_list;
 	/* Per adapter stats */
 	struct rte_event_eth_rx_adapter_stats stats;
 	/* Block count, counts up to BLOCK_CNT_THRESHOLD */
@@ -198,9 +224,11 @@ struct eth_device_info {
 struct eth_rx_queue_info {
 	int queue_enabled;	/* True if added */
 	int intr_enabled;
+	uint8_t ena_vector;
 	uint16_t wt;		/* Polling weight */
 	uint32_t flow_id_mask;	/* Set to ~0 if app provides flow id else 0 */
 	uint64_t event;
+	struct eth_rx_vector_data vector_data;
 };
 
 static struct rte_event_eth_rx_adapter **event_eth_rx_adapter;
@@ -722,6 +750,9 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
 	    &rx_adapter->event_enqueue_buffer;
 	struct rte_event_eth_rx_adapter_stats *stats = &rx_adapter->stats;
 
+	if (!buf->count)
+		return 0;
+
 	uint16_t n = rte_event_enqueue_new_burst(rx_adapter->eventdev_id,
 					rx_adapter->event_port_id,
 					buf->events,
@@ -742,6 +773,77 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
 	return n;
 }
 
+static inline void
+rxa_init_vector(struct rte_event_eth_rx_adapter *rx_adapter,
+		struct eth_rx_vector_data *vec)
+{
+	vec->vector_ev->nb_elem = 0;
+	vec->vector_ev->port = vec->port;
+	vec->vector_ev->queue = vec->queue;
+	vec->vector_ev->attr_valid = true;
+	TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
+}
+
+static inline uint16_t
+rxa_create_event_vector(struct rte_event_eth_rx_adapter *rx_adapter,
+			struct eth_rx_queue_info *queue_info,
+			struct rte_eth_event_enqueue_buffer *buf,
+			struct rte_mbuf **mbufs, uint16_t num)
+{
+	struct rte_event *ev = &buf->events[buf->count];
+	struct eth_rx_vector_data *vec;
+	uint16_t filled, space, sz;
+
+	filled = 0;
+	vec = &queue_info->vector_data;
+
+	if (vec->vector_ev == NULL) {
+		if (rte_mempool_get(vec->vector_pool,
+				    (void **)&vec->vector_ev) < 0) {
+			rte_pktmbuf_free_bulk(mbufs, num);
+			return 0;
+		}
+		rxa_init_vector(rx_adapter, vec);
+	}
+	while (num) {
+		if (vec->vector_ev->nb_elem == vec->max_vector_count) {
+			/* Event ready. */
+			ev->event = vec->event;
+			ev->vec = vec->vector_ev;
+			ev++;
+			filled++;
+			vec->vector_ev = NULL;
+			TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+			if (rte_mempool_get(vec->vector_pool,
+					    (void **)&vec->vector_ev) < 0) {
+				rte_pktmbuf_free_bulk(mbufs, num);
+				return 0;
+			}
+			rxa_init_vector(rx_adapter, vec);
+		}
+
+		space = vec->max_vector_count - vec->vector_ev->nb_elem;
+		sz = num > space ? space : num;
+		memcpy(vec->vector_ev->mbufs + vec->vector_ev->nb_elem, mbufs,
+		       sizeof(void *) * sz);
+		vec->vector_ev->nb_elem += sz;
+		num -= sz;
+		mbufs += sz;
+		vec->ts = rte_rdtsc();
+	}
+
+	if (vec->vector_ev->nb_elem == vec->max_vector_count) {
+		ev->event = vec->event;
+		ev->vec = vec->vector_ev;
+		ev++;
+		filled++;
+		vec->vector_ev = NULL;
+		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+	}
+
+	return filled;
+}
+
 static inline void
 rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
@@ -766,29 +868,33 @@ rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
 	uint16_t nb_cb;
 	uint16_t dropped;
 
-	/* 0xffff ffff if PKT_RX_RSS_HASH is set, otherwise 0 */
-	rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
-	do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
-
-	for (i = 0; i < num; i++) {
-		m = mbufs[i];
-
-		rss = do_rss ?
-			rxa_do_softrss(m, rx_adapter->rss_key_be) :
-			m->hash.rss;
-		ev->event = event;
-		ev->flow_id = (rss & ~flow_id_mask) |
-				(ev->flow_id & flow_id_mask);
-		ev->mbuf = m;
-		ev++;
+	if (!eth_rx_queue_info->ena_vector) {
+		/* 0xffff ffff if PKT_RX_RSS_HASH is set, otherwise 0 */
+		rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
+		do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
+		for (i = 0; i < num; i++) {
+			m = mbufs[i];
+
+			rss = do_rss ? rxa_do_softrss(m, rx_adapter->rss_key_be)
+				     : m->hash.rss;
+			ev->event = event;
+			ev->flow_id = (rss & ~flow_id_mask) |
+				      (ev->flow_id & flow_id_mask);
+			ev->mbuf = m;
+			ev++;
+		}
+	} else {
+		num = rxa_create_event_vector(rx_adapter, eth_rx_queue_info,
+					      buf, mbufs, num);
 	}
 
-	if (dev_info->cb_fn) {
+	if (num && dev_info->cb_fn) {
 
 		dropped = 0;
 		nb_cb = dev_info->cb_fn(eth_dev_id, rx_queue_id,
-					ETH_EVENT_BUFFER_SIZE, buf->count, ev,
-					num, dev_info->cb_arg, &dropped);
+					ETH_EVENT_BUFFER_SIZE, buf->count,
+					&buf->events[buf->count], num,
+					dev_info->cb_arg, &dropped);
 		if (unlikely(nb_cb > num))
 			RTE_EDEV_LOG_ERR("Rx CB returned %d (> %d) events",
 				nb_cb, num);
@@ -1124,6 +1230,30 @@ rxa_poll(struct rte_event_eth_rx_adapter *rx_adapter)
 	return nb_rx;
 }
 
+static void
+rxa_vector_expire(struct eth_rx_vector_data *vec, void *arg)
+{
+	struct rte_event_eth_rx_adapter *rx_adapter = arg;
+	struct rte_eth_event_enqueue_buffer *buf =
+		&rx_adapter->event_enqueue_buffer;
+	struct rte_event *ev;
+
+	if (buf->count)
+		rxa_flush_event_buffer(rx_adapter);
+
+	if (vec->vector_ev->nb_elem == 0)
+		return;
+	ev = &buf->events[buf->count];
+
+	/* Event ready. */
+	ev->event = vec->event;
+	ev->vec = vec->vector_ev;
+	buf->count++;
+
+	vec->vector_ev = NULL;
+	vec->ts = 0;
+}
+
 static int
 rxa_service_func(void *args)
 {
@@ -1137,6 +1267,24 @@ rxa_service_func(void *args)
 		return 0;
 	}
 
+	if (rx_adapter->ena_vector) {
+		if ((rte_rdtsc() - rx_adapter->prev_expiry_ts) >=
+		    rx_adapter->vector_tmo_ticks) {
+			struct eth_rx_vector_data *vec;
+
+			TAILQ_FOREACH(vec, &rx_adapter->vector_list, next) {
+				uint64_t elapsed_time = rte_rdtsc() - vec->ts;
+
+				if (elapsed_time >= vec->vector_timeout_ticks) {
+					rxa_vector_expire(vec, rx_adapter);
+					TAILQ_REMOVE(&rx_adapter->vector_list,
+						     vec, next);
+				}
+			}
+			rx_adapter->prev_expiry_ts = rte_rdtsc();
+		}
+	}
+
 	stats = &rx_adapter->stats;
 	stats->rx_packets += rxa_intr_ring_dequeue(rx_adapter);
 	stats->rx_packets += rxa_poll(rx_adapter);
@@ -1640,11 +1788,35 @@ rxa_update_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }
 
+static void
+rxa_set_vector_data(struct eth_rx_queue_info *queue_info, uint16_t vector_count,
+		    uint64_t vector_ns, struct rte_mempool *mp, int32_t qid,
+		    uint16_t port_id)
+{
+#define NSEC2TICK(__ns, __freq) (((__ns) * (__freq)) / 1E9)
+	struct eth_rx_vector_data *vector_data;
+	uint32_t flow_id;
+
+	vector_data = &queue_info->vector_data;
+	vector_data->max_vector_count = vector_count;
+	vector_data->port = port_id;
+	vector_data->queue = qid;
+	vector_data->vector_pool = mp;
+	vector_data->vector_timeout_ticks =
+		NSEC2TICK(vector_ns, rte_get_timer_hz());
+	vector_data->ts = 0;
+	flow_id = queue_info->event & 0xFFFFF;
+	flow_id =
+		flow_id == 0 ? (qid & 0xFFF) | (port_id & 0xFF) << 12 : flow_id;
+	vector_data->event = (queue_info->event & ~0xFFFFF) | flow_id;
+}
+
 static void
 rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
 	struct eth_device_info *dev_info,
 	int32_t rx_queue_id)
 {
+	struct eth_rx_vector_data *vec;
 	int pollq;
 	int intrq;
 	int sintrq;
@@ -1663,6 +1835,14 @@ rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
 		return;
 	}
 
+	/* Push all the partial event vectors to event device. */
+	TAILQ_FOREACH(vec, &rx_adapter->vector_list, next) {
+		if (vec->queue != rx_queue_id)
+			continue;
+		rxa_vector_expire(vec, rx_adapter);
+		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+	}
+
 	pollq = rxa_polled_queue(dev_info, rx_queue_id);
 	intrq = rxa_intr_queue(dev_info, rx_queue_id);
 	sintrq = rxa_shared_intr(dev_info, rx_queue_id);
@@ -1741,6 +1921,42 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }
 
+static void
+rxa_sw_event_vector_configure(
+	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
+	int rx_queue_id,
+	const struct rte_event_eth_rx_adapter_event_vector_config *config)
+{
+	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
+	struct eth_rx_queue_info *queue_info;
+	struct rte_event *qi_ev;
+
+	if (rx_queue_id == -1) {
+		uint16_t nb_rx_queues;
+		uint16_t i;
+
+		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
+		for (i = 0; i < nb_rx_queues; i++)
+			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
+						      config);
+		return;
+	}
+
+	queue_info = &dev_info->rx_queue[rx_queue_id];
+	qi_ev = (struct rte_event *)&queue_info->event;
+	queue_info->ena_vector = 1;
+	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+	rxa_set_vector_data(queue_info, config->vector_sz,
+			    config->vector_timeout_ns, config->vector_mp,
+			    rx_queue_id, dev_info->dev->data->port_id);
+	rx_adapter->ena_vector = 1;
+	rx_adapter->vector_tmo_ticks =
+		rx_adapter->vector_tmo_ticks ?
+			      RTE_MIN(config->vector_timeout_ns >> 1,
+				rx_adapter->vector_tmo_ticks) :
+			      config->vector_timeout_ns >> 1;
+}
+
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -1967,6 +2183,7 @@ rte_event_eth_rx_adapter_create_ext(uint8_t id, uint8_t dev_id,
 	rx_adapter->conf_cb = conf_cb;
 	rx_adapter->conf_arg = conf_arg;
 	rx_adapter->id = id;
+	TAILQ_INIT(&rx_adapter->vector_list);
 	strcpy(rx_adapter->mem_name, mem_name);
 	rx_adapter->eth_devices = rte_zmalloc_socket(rx_adapter->mem_name,
 					RTE_MAX_ETHPORTS *
@@ -2081,6 +2298,15 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}
 
+	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
+	    (queue_conf->rx_queue_flags &
+	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
+		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
 		(rx_queue_id != -1)) {
 		RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
@@ -2143,6 +2369,17 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	return 0;
 }
 
+static int
+rxa_sw_vector_limits(struct rte_event_eth_rx_adapter_vector_limits *limits)
+{
+	limits->max_sz = MAX_VECTOR_SIZE;
+	limits->min_sz = MIN_VECTOR_SIZE;
+	limits->max_timeout_ns = MAX_VECTOR_NS;
+	limits->min_timeout_ns = MIN_VECTOR_NS;
+
+	return 0;
+}
+
 int
 rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 				int32_t rx_queue_id)
@@ -2333,7 +2570,8 @@ rte_event_eth_rx_adapter_queue_event_vector_config(
 		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
 			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
 	} else {
-		ret = -ENOTSUP;
+		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
+					      rx_queue_id, config);
 	}
 
 	return ret;
@@ -2371,7 +2609,7 @@ rte_event_eth_rx_adapter_vector_limits_get(
 		ret = dev->dev_ops->eth_rx_adapter_vector_limits_get(
 			dev, &rte_eth_devices[eth_port_id], limits);
 	} else {
-		ret = -ENOTSUP;
+		ret = rxa_sw_vector_limits(limits);
 	}
 
 	return ret;
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index be0499c52..62824654b 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -122,7 +122,11 @@ rte_event_eth_rx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
 
 	if (caps == NULL)
 		return -EINVAL;
-	*caps = 0;
+
+	if (dev->dev_ops->eth_rx_adapter_caps_get == NULL)
+		*caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
+	else
+		*caps = 0;
 
 	return dev->dev_ops->eth_rx_adapter_caps_get ?
 				(*dev->dev_ops->eth_rx_adapter_caps_get)(dev,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v9 5/8] eventdev: add Tx adapter event vector support
  2021-03-30  8:22               ` [dpdk-dev] [PATCH v9 0/8] Introduce event vectorization pbhagavatula
                                   ` (3 preceding siblings ...)
  2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 4/8] eventdev: add Rx adapter event vector support pbhagavatula
@ 2021-03-30  8:22                 ` pbhagavatula
  2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
                                   ` (3 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-30  8:22 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support for event eth Tx adapter, the implementation
receives events from the single linked queue and based on
rte_event_vector::attr_valid transmits the vector of mbufs to a given
port, queue pair.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
---
 .../rte_event_eth_tx_adapter.c                | 66 ++++++++++++++++---
 lib/librte_eventdev/rte_eventdev.c            |  5 +-
 2 files changed, 60 insertions(+), 11 deletions(-)

diff --git a/lib/librte_eventdev/rte_event_eth_tx_adapter.c b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
index 5b4c42dcf..db260bfb6 100644
--- a/lib/librte_eventdev/rte_event_eth_tx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
@@ -510,6 +510,47 @@ txa_service_buffer_retry(struct rte_mbuf **pkts, uint16_t unsent,
 	stats->tx_dropped += unsent - sent;
 }
 
+static uint16_t
+txa_process_event_vector(struct txa_service_data *txa,
+			 struct rte_event_vector *vec)
+{
+	struct txa_service_queue_info *tqi;
+	uint16_t port, queue, nb_tx = 0;
+	struct rte_mbuf **mbufs;
+	int i;
+
+	mbufs = (struct rte_mbuf **)vec->mbufs;
+	if (vec->attr_valid) {
+		port = vec->port;
+		queue = vec->queue;
+		tqi = txa_service_queue(txa, port, queue);
+		if (unlikely(tqi == NULL || !tqi->added)) {
+			rte_pktmbuf_free_bulk(mbufs, vec->nb_elem);
+			rte_mempool_put(rte_mempool_from_obj(vec), vec);
+			return 0;
+		}
+		for (i = 0; i < vec->nb_elem; i++) {
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
+						   mbufs[i]);
+		}
+	} else {
+		for (i = 0; i < vec->nb_elem; i++) {
+			port = mbufs[i]->port;
+			queue = rte_event_eth_tx_adapter_txq_get(mbufs[i]);
+			tqi = txa_service_queue(txa, port, queue);
+			if (unlikely(tqi == NULL || !tqi->added)) {
+				rte_pktmbuf_free(mbufs[i]);
+				continue;
+			}
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
+						   mbufs[i]);
+		}
+	}
+	rte_mempool_put(rte_mempool_from_obj(vec), vec);
+
+	return nb_tx;
+}
+
 static void
 txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
 	uint32_t n)
@@ -522,22 +563,27 @@ txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
 
 	nb_tx = 0;
 	for (i = 0; i < n; i++) {
-		struct rte_mbuf *m;
 		uint16_t port;
 		uint16_t queue;
 		struct txa_service_queue_info *tqi;
 
-		m = ev[i].mbuf;
-		port = m->port;
-		queue = rte_event_eth_tx_adapter_txq_get(m);
+		if (!(ev[i].event_type & RTE_EVENT_TYPE_VECTOR)) {
+			struct rte_mbuf *m;
 
-		tqi = txa_service_queue(txa, port, queue);
-		if (unlikely(tqi == NULL || !tqi->added)) {
-			rte_pktmbuf_free(m);
-			continue;
-		}
+			m = ev[i].mbuf;
+			port = m->port;
+			queue = rte_event_eth_tx_adapter_txq_get(m);
 
-		nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
+			tqi = txa_service_queue(txa, port, queue);
+			if (unlikely(tqi == NULL || !tqi->added)) {
+				rte_pktmbuf_free(m);
+				continue;
+			}
+
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
+		} else {
+			nb_tx += txa_process_event_vector(txa, ev[i].vec);
+		}
 	}
 
 	stats->tx_packets += nb_tx;
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index 62824654b..c9bb5d227 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -196,7 +196,10 @@ rte_event_eth_tx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
 	if (caps == NULL)
 		return -EINVAL;
 
-	*caps = 0;
+	if (dev->dev_ops->eth_tx_adapter_caps_get == NULL)
+		*caps = RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR;
+	else
+		*caps = 0;
 
 	return dev->dev_ops->eth_tx_adapter_caps_get ?
 			(*dev->dev_ops->eth_tx_adapter_caps_get)(dev,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v9 6/8] app/eventdev: add event vector mode in pipeline test
  2021-03-30  8:22               ` [dpdk-dev] [PATCH v9 0/8] Introduce event vectorization pbhagavatula
                                   ` (4 preceding siblings ...)
  2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 5/8] eventdev: add Tx " pbhagavatula
@ 2021-03-30  8:22                 ` pbhagavatula
  2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 7/8] doc: announce event Rx adapter config changes pbhagavatula
                                   ` (2 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-30  8:22 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support in pipeline tests. By default this mode
is disabled, it can be enabled by using the option --enable_vector.
example:
	dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
	--nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
	--wlcores=20-23  --enable_vector

Additional options to configure vector size and vector timeout are
also implemented and can be used by specifying --vector_size and
--vector_tmo_ns

This patch also adds a new option to set the number of Rx queues
configured per event eth rx adapter.
example:
	dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
	--nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
	--wlcores=20-23  --nb_eth_queues 4

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
---
 app/test-eventdev/evt_common.h           |   4 +
 app/test-eventdev/evt_options.c          |  52 ++++
 app/test-eventdev/evt_options.h          |   4 +
 app/test-eventdev/test_pipeline_atq.c    | 310 ++++++++++++++++++++--
 app/test-eventdev/test_pipeline_common.c | 113 +++++++-
 app/test-eventdev/test_pipeline_common.h |  18 ++
 app/test-eventdev/test_pipeline_queue.c  | 320 +++++++++++++++++++++--
 doc/guides/tools/testeventdev.rst        |  45 +++-
 8 files changed, 811 insertions(+), 55 deletions(-)

diff --git a/app/test-eventdev/evt_common.h b/app/test-eventdev/evt_common.h
index a1da1cf11..0e228258e 100644
--- a/app/test-eventdev/evt_common.h
+++ b/app/test-eventdev/evt_common.h
@@ -58,16 +58,20 @@ struct evt_options {
 	uint8_t sched_type_list[EVT_MAX_STAGES];
 	uint16_t mbuf_sz;
 	uint16_t wkr_deq_dep;
+	uint16_t vector_size;
+	uint16_t eth_queues;
 	uint32_t nb_flows;
 	uint32_t tx_first;
 	uint32_t max_pkt_sz;
 	uint32_t deq_tmo_nsec;
 	uint32_t q_priority:1;
 	uint32_t fwd_latency:1;
+	uint32_t ena_vector : 1;
 	uint64_t nb_pkts;
 	uint64_t nb_timers;
 	uint64_t expiry_nsec;
 	uint64_t max_tmo_nsec;
+	uint64_t vector_tmo_nsec;
 	uint64_t timer_tick_nsec;
 	uint64_t optm_timer_tick_nsec;
 	enum evt_prod_type prod_type;
diff --git a/app/test-eventdev/evt_options.c b/app/test-eventdev/evt_options.c
index 0d04ea9f8..0d5540574 100644
--- a/app/test-eventdev/evt_options.c
+++ b/app/test-eventdev/evt_options.c
@@ -34,6 +34,9 @@ evt_options_default(struct evt_options *opt)
 	opt->max_tmo_nsec = 1E5;  /* 100000ns ~100us */
 	opt->expiry_nsec = 1E4;   /* 10000ns ~10us */
 	opt->prod_type = EVT_PROD_TYPE_SYNT;
+	opt->eth_queues = 1;
+	opt->vector_size = 64;
+	opt->vector_tmo_nsec = 100E3;
 }
 
 typedef int (*option_parser_t)(struct evt_options *opt,
@@ -257,6 +260,43 @@ evt_parse_max_pkt_sz(struct evt_options *opt, const char *arg)
 	return ret;
 }
 
+static int
+evt_parse_ena_vector(struct evt_options *opt, const char *arg __rte_unused)
+{
+	opt->ena_vector = 1;
+	return 0;
+}
+
+static int
+evt_parse_vector_size(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint16(&(opt->vector_size), arg);
+
+	return ret;
+}
+
+static int
+evt_parse_vector_tmo_ns(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint64(&(opt->vector_tmo_nsec), arg);
+
+	return ret;
+}
+
+static int
+evt_parse_eth_queues(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint16(&(opt->eth_queues), arg);
+
+	return ret;
+}
+
 static void
 usage(char *program)
 {
@@ -289,6 +329,10 @@ usage(char *program)
 		"\t--expiry_nsec      : event timer expiry ns.\n"
 		"\t--mbuf_sz          : packet mbuf size.\n"
 		"\t--max_pkt_sz       : max packet size.\n"
+		"\t--nb_eth_queues    : number of ethernet Rx queues.\n"
+		"\t--enable_vector    : enable event vectorization.\n"
+		"\t--vector_size      : Max vector size.\n"
+		"\t--vector_tmo_ns    : Max vector timeout in nanoseconds\n"
 		);
 	printf("available tests:\n");
 	evt_test_dump_names();
@@ -360,6 +404,10 @@ static struct option lgopts[] = {
 	{ EVT_EXPIRY_NSEC,         1, 0, 0 },
 	{ EVT_MBUF_SZ,             1, 0, 0 },
 	{ EVT_MAX_PKT_SZ,          1, 0, 0 },
+	{ EVT_NB_ETH_QUEUES,       1, 0, 0 },
+	{ EVT_ENA_VECTOR,          0, 0, 0 },
+	{ EVT_VECTOR_SZ,           1, 0, 0 },
+	{ EVT_VECTOR_TMO,          1, 0, 0 },
 	{ EVT_HELP,                0, 0, 0 },
 	{ NULL,                    0, 0, 0 }
 };
@@ -394,6 +442,10 @@ evt_opts_parse_long(int opt_idx, struct evt_options *opt)
 		{ EVT_EXPIRY_NSEC, evt_parse_expiry_nsec},
 		{ EVT_MBUF_SZ, evt_parse_mbuf_sz},
 		{ EVT_MAX_PKT_SZ, evt_parse_max_pkt_sz},
+		{ EVT_NB_ETH_QUEUES, evt_parse_eth_queues},
+		{ EVT_ENA_VECTOR, evt_parse_ena_vector},
+		{ EVT_VECTOR_SZ, evt_parse_vector_size},
+		{ EVT_VECTOR_TMO, evt_parse_vector_tmo_ns},
 	};
 
 	for (i = 0; i < RTE_DIM(parsermap); i++) {
diff --git a/app/test-eventdev/evt_options.h b/app/test-eventdev/evt_options.h
index 748e54fae..1cea2a3e1 100644
--- a/app/test-eventdev/evt_options.h
+++ b/app/test-eventdev/evt_options.h
@@ -42,6 +42,10 @@
 #define EVT_EXPIRY_NSEC          ("expiry_nsec")
 #define EVT_MBUF_SZ              ("mbuf_sz")
 #define EVT_MAX_PKT_SZ           ("max_pkt_sz")
+#define EVT_NB_ETH_QUEUES        ("nb_eth_queues")
+#define EVT_ENA_VECTOR           ("enable_vector")
+#define EVT_VECTOR_SZ            ("vector_size")
+#define EVT_VECTOR_TMO           ("vector_tmo_ns")
 #define EVT_HELP                 ("help")
 
 void evt_options_default(struct evt_options *opt);
diff --git a/app/test-eventdev/test_pipeline_atq.c b/app/test-eventdev/test_pipeline_atq.c
index 0872b25b5..84dd4f44e 100644
--- a/app/test-eventdev/test_pipeline_atq.c
+++ b/app/test-eventdev/test_pipeline_atq.c
@@ -15,6 +15,8 @@ pipeline_atq_nb_event_queues(struct evt_options *opt)
 	return rte_eth_dev_count_avail();
 }
 
+typedef int (*pipeline_atq_worker_t)(void *arg);
+
 static __rte_noinline int
 pipeline_atq_worker_single_stage_tx(void *arg)
 {
@@ -113,6 +115,112 @@ pipeline_atq_worker_single_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_atq_worker_single_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+		vector_sz = ev.vec->nb_elem;
+		pipeline_event_tx_vector(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = ev.vec->nb_elem;
+		ev.queue_id = tx_queue[ev.vec->port];
+		ev.vec->queue = 0;
+		pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			vector_sz += ev[i].vec->nb_elem;
+			ev[i].vec->queue = 0;
+		}
+
+		pipeline_event_tx_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			ev[i].queue_id = tx_queue[ev[i].vec->port];
+			ev[i].vec->queue = 0;
+			vector_sz += ev[i].vec->nb_elem;
+			pipeline_fwd_event_vector(&ev[i],
+						  RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
 static __rte_noinline int
 pipeline_atq_worker_multi_stage_tx(void *arg)
 {
@@ -245,6 +353,147 @@ pipeline_atq_worker_multi_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.sub_event_type % nb_stages;
+
+		if (cq_id == last_queue) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+			continue;
+		}
+
+		ev.sub_event_type++;
+		pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.sub_event_type % nb_stages;
+
+		if (cq_id == last_queue) {
+			ev.queue_id = tx_queue[ev.vec->port];
+			ev.vec->queue = 0;
+			vector_sz = ev.vec->nb_elem;
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			pipeline_event_enqueue(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.sub_event_type++;
+			pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+			pipeline_event_enqueue(dev, port, &ev);
+		}
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].sub_event_type % nb_stages;
+
+			if (cq_id == last_queue) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+				continue;
+			}
+
+			ev[i].sub_event_type++;
+			pipeline_fwd_event_vector(&ev[i],
+						  sched_type_list[cq_id]);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].sub_event_type % nb_stages;
+
+			if (cq_id == last_queue) {
+				vector_sz = ev[i].vec->nb_elem;
+				ev[i].queue_id = tx_queue[ev[i].vec->port];
+				ev[i].vec->queue = 0;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].sub_event_type++;
+				pipeline_fwd_event_vector(
+					&ev[i], sched_type_list[cq_id]);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
 static int
 worker_wrapper(void *arg)
 {
@@ -253,27 +502,36 @@ worker_wrapper(void *arg)
 	const bool burst = evt_has_burst_mode(w->dev_id);
 	const bool internal_port = w->t->internal_port;
 	const uint8_t nb_stages = opt->nb_stages;
-	RTE_SET_USED(opt);
+	/*vector/burst/internal_port*/
+	const pipeline_atq_worker_t
+	pipeline_atq_worker_single_stage[2][2][2] = {
+		[0][0][0] = pipeline_atq_worker_single_stage_fwd,
+		[0][0][1] = pipeline_atq_worker_single_stage_tx,
+		[0][1][0] = pipeline_atq_worker_single_stage_burst_fwd,
+		[0][1][1] = pipeline_atq_worker_single_stage_burst_tx,
+		[1][0][0] = pipeline_atq_worker_single_stage_fwd_vector,
+		[1][0][1] = pipeline_atq_worker_single_stage_tx_vector,
+		[1][1][0] = pipeline_atq_worker_single_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_atq_worker_single_stage_burst_tx_vector,
+	};
+	const pipeline_atq_worker_t
+	pipeline_atq_worker_multi_stage[2][2][2] = {
+		[0][0][0] = pipeline_atq_worker_multi_stage_fwd,
+		[0][0][1] = pipeline_atq_worker_multi_stage_tx,
+		[0][1][0] = pipeline_atq_worker_multi_stage_burst_fwd,
+		[0][1][1] = pipeline_atq_worker_multi_stage_burst_tx,
+		[1][0][0] = pipeline_atq_worker_multi_stage_fwd_vector,
+		[1][0][1] = pipeline_atq_worker_multi_stage_tx_vector,
+		[1][1][0] = pipeline_atq_worker_multi_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_atq_worker_multi_stage_burst_tx_vector,
+	};
 
-	if (nb_stages == 1) {
-		if (!burst && internal_port)
-			return pipeline_atq_worker_single_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_atq_worker_single_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_atq_worker_single_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_atq_worker_single_stage_burst_fwd(arg);
-	} else {
-		if (!burst && internal_port)
-			return pipeline_atq_worker_multi_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_atq_worker_multi_stage_fwd(arg);
-		if (burst && internal_port)
-			return pipeline_atq_worker_multi_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_atq_worker_multi_stage_burst_fwd(arg);
-	}
+	if (nb_stages == 1)
+		return (pipeline_atq_worker_single_stage[opt->ena_vector][burst]
+							[internal_port])(arg);
+	else
+		return (pipeline_atq_worker_multi_stage[opt->ena_vector][burst]
+						       [internal_port])(arg);
 
 	rte_panic("invalid worker\n");
 }
@@ -290,7 +548,7 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
 	int ret;
 	int nb_ports;
 	int nb_queues;
-	uint8_t queue;
+	uint8_t queue, is_prod;
 	uint8_t tx_evqueue_id[RTE_MAX_ETHPORTS];
 	uint8_t queue_arr[RTE_EVENT_MAX_QUEUES_PER_DEV];
 	uint8_t nb_worker_queues = 0;
@@ -330,15 +588,19 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
 		q_conf.event_queue_cfg = RTE_EVENT_QUEUE_CFG_ALL_TYPES;
 
 		if (!t->internal_port) {
+			is_prod = false;
 			RTE_ETH_FOREACH_DEV(prod) {
 				if (queue == tx_evqueue_id[prod]) {
 					q_conf.event_queue_cfg =
 						RTE_EVENT_QUEUE_CFG_SINGLE_LINK;
-				} else {
-					queue_arr[nb_worker_queues] = queue;
-					nb_worker_queues++;
+					is_prod = true;
+					break;
 				}
 			}
+			if (!is_prod) {
+				queue_arr[nb_worker_queues] = queue;
+				nb_worker_queues++;
+			}
 		}
 
 		ret = rte_event_queue_setup(opt->dev_id, queue, &q_conf);
diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index b47d76743..d5ef90500 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -36,6 +36,12 @@ pipeline_opt_dump(struct evt_options *opt, uint8_t nb_queues)
 	evt_dump_queue_priority(opt);
 	evt_dump_sched_type_list(opt);
 	evt_dump_producer_type(opt);
+	evt_dump("nb_eth_rx_queues", "%d", opt->eth_queues);
+	evt_dump("event_vector", "%d", opt->ena_vector);
+	if (opt->ena_vector) {
+		evt_dump("vector_size", "%d", opt->vector_size);
+		evt_dump("vector_tmo_ns", "%" PRIu64 "", opt->vector_tmo_nsec);
+	}
 }
 
 static inline uint64_t
@@ -163,7 +169,7 @@ pipeline_opt_check(struct evt_options *opt, uint64_t nb_queues)
 int
 pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 {
-	uint16_t i;
+	uint16_t i, j;
 	int ret;
 	uint8_t nb_queues = 1;
 	struct test_pipeline *t = evt_test_priv(test);
@@ -210,6 +216,16 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 		if (!(caps & RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT))
 			t->internal_port = 0;
 
+		ret = rte_event_eth_rx_adapter_caps_get(opt->dev_id, i, &caps);
+		if (ret != 0) {
+			evt_err("failed to get event tx adapter[%d] caps", i);
+			return ret;
+		}
+
+		if (!(caps & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT))
+			local_port_conf.rxmode.offloads |=
+				DEV_RX_OFFLOAD_RSS_HASH;
+
 		ret = rte_eth_dev_info_get(i, &dev_info);
 		if (ret != 0) {
 			evt_err("Error during getting device (port %u) info: %s\n",
@@ -236,19 +252,22 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 				local_port_conf.rx_adv_conf.rss_conf.rss_hf);
 		}
 
-		if (rte_eth_dev_configure(i, nb_queues, nb_queues,
-					&local_port_conf)
-				< 0) {
+		if (rte_eth_dev_configure(i, opt->eth_queues, nb_queues,
+					  &local_port_conf) < 0) {
 			evt_err("Failed to configure eth port [%d]", i);
 			return -EINVAL;
 		}
 
-		if (rte_eth_rx_queue_setup(i, 0, NB_RX_DESC,
-				rte_socket_id(), &rx_conf, t->pool) < 0) {
-			evt_err("Failed to setup eth port [%d] rx_queue: %d.",
+		for (j = 0; j < opt->eth_queues; j++) {
+			if (rte_eth_rx_queue_setup(i, j, NB_RX_DESC,
+						   rte_socket_id(), &rx_conf,
+						   t->pool) < 0) {
+				evt_err("Failed to setup eth port [%d] rx_queue: %d.",
 					i, 0);
-			return -EINVAL;
+				return -EINVAL;
+			}
 		}
+
 		if (rte_eth_tx_queue_setup(i, 0, NB_TX_DESC,
 					rte_socket_id(), NULL) < 0) {
 			evt_err("Failed to setup eth port [%d] tx_queue: %d.",
@@ -310,12 +329,27 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 {
 	int ret = 0;
 	uint16_t prod;
+	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
+	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;
 
 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
 	queue_conf.ev.sched_type = opt->sched_type_list[0];
+	if (opt->ena_vector) {
+		unsigned int nb_elem = (opt->pool_sz / opt->vector_size) << 1;
+
+		nb_elem = nb_elem ? nb_elem : 1;
+		vector_pool = rte_event_vector_pool_create(
+			"vector_pool", nb_elem, 0, opt->vector_size,
+			opt->socket_id);
+		if (vector_pool == NULL) {
+			evt_err("failed to create event vector pool");
+			return -ENOMEM;
+		}
+	}
 	RTE_ETH_FOREACH_DEV(prod) {
+		struct rte_event_eth_rx_adapter_vector_limits limits;
 		uint32_t cap;
 
 		ret = rte_event_eth_rx_adapter_caps_get(opt->dev_id,
@@ -326,6 +360,50 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 					opt->dev_id);
 			return ret;
 		}
+
+		if (opt->ena_vector) {
+			memset(&limits, 0, sizeof(limits));
+			ret = rte_event_eth_rx_adapter_vector_limits_get(
+				opt->dev_id, prod, &limits);
+			if (ret) {
+				evt_err("failed to get vector limits");
+				return ret;
+			}
+
+			if (opt->vector_size < limits.min_sz ||
+			    opt->vector_size > limits.max_sz) {
+				evt_err("Vector size [%d] not within limits max[%d] min[%d]",
+					opt->vector_size, limits.min_sz,
+					limits.max_sz);
+				return -EINVAL;
+			}
+
+			if (limits.log2_sz &&
+			    !rte_is_power_of_2(opt->vector_size)) {
+				evt_err("Vector size [%d] not power of 2",
+					opt->vector_size);
+				return -EINVAL;
+			}
+
+			if (opt->vector_tmo_nsec > limits.max_timeout_ns ||
+			    opt->vector_tmo_nsec < limits.min_timeout_ns) {
+				evt_err("Vector timeout [%" PRIu64
+					"] not within limits max[%" PRIu64
+					"] min[%" PRIu64 "]",
+					opt->vector_tmo_nsec,
+					limits.max_timeout_ns,
+					limits.min_timeout_ns);
+				return -EINVAL;
+			}
+
+			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.rx_queue_flags |=
+				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+			} else {
+				evt_err("Rx adapter doesn't support event vector");
+				return -EINVAL;
+			}
+		}
 		queue_conf.ev.queue_id = prod * stride;
 		ret = rte_event_eth_rx_adapter_create(prod, opt->dev_id,
 				&prod_conf);
@@ -340,6 +418,17 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}
 
+		if (opt->ena_vector) {
+			vec_conf.vector_sz = opt->vector_size;
+			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
+			vec_conf.vector_mp = vector_pool;
+			if (rte_event_eth_rx_adapter_queue_event_vector_config(
+				    prod, prod, -1, &vec_conf) < 0) {
+				evt_err("Failed to configure event vectorization for Rx adapter");
+				return -EINVAL;
+			}
+		}
+
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;
 
@@ -378,6 +467,14 @@ pipeline_event_tx_adapter_setup(struct evt_options *opt,
 			return ret;
 		}
 
+		if (opt->ena_vector) {
+			if (!(cap &
+			      RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR)) {
+				evt_err("Tx adapter doesn't support event vector");
+				return -EINVAL;
+			}
+		}
+
 		ret = rte_event_eth_tx_adapter_create(consm, opt->dev_id,
 				&port_conf);
 		if (ret) {
diff --git a/app/test-eventdev/test_pipeline_common.h b/app/test-eventdev/test_pipeline_common.h
index 6e73c6ab2..800a90616 100644
--- a/app/test-eventdev/test_pipeline_common.h
+++ b/app/test-eventdev/test_pipeline_common.h
@@ -101,6 +101,14 @@ pipeline_fwd_event(struct rte_event *ev, uint8_t sched)
 	ev->sched_type = sched;
 }
 
+static __rte_always_inline void
+pipeline_fwd_event_vector(struct rte_event *ev, uint8_t sched)
+{
+	ev->event_type = RTE_EVENT_TYPE_CPU_VECTOR;
+	ev->op = RTE_EVENT_OP_FORWARD;
+	ev->sched_type = sched;
+}
+
 static __rte_always_inline void
 pipeline_event_tx(const uint8_t dev, const uint8_t port,
 		struct rte_event * const ev)
@@ -110,6 +118,16 @@ pipeline_event_tx(const uint8_t dev, const uint8_t port,
 		rte_pause();
 }
 
+static __rte_always_inline void
+pipeline_event_tx_vector(const uint8_t dev, const uint8_t port,
+			 struct rte_event *const ev)
+{
+	ev->vec->queue = 0;
+
+	while (!rte_event_eth_tx_adapter_enqueue(dev, port, ev, 1, 0))
+		rte_pause();
+}
+
 static __rte_always_inline void
 pipeline_event_tx_burst(const uint8_t dev, const uint8_t port,
 		struct rte_event *ev, const uint16_t nb_rx)
diff --git a/app/test-eventdev/test_pipeline_queue.c b/app/test-eventdev/test_pipeline_queue.c
index 9a9febb19..f6cc3e358 100644
--- a/app/test-eventdev/test_pipeline_queue.c
+++ b/app/test-eventdev/test_pipeline_queue.c
@@ -15,6 +15,8 @@ pipeline_queue_nb_event_queues(struct evt_options *opt)
 	return (eth_count * opt->nb_stages) + eth_count;
 }
 
+typedef int (*pipeline_queue_worker_t)(void *arg);
+
 static __rte_noinline int
 pipeline_queue_worker_single_stage_tx(void *arg)
 {
@@ -126,6 +128,125 @@ pipeline_queue_worker_single_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_queue_worker_single_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		if (ev.sched_type == RTE_SCHED_TYPE_ATOMIC) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.queue_id++;
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			pipeline_event_enqueue(dev, port, &ev);
+		}
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		ev.queue_id = tx_queue[ev.vec->port];
+		ev.vec->queue = 0;
+		vector_sz = ev.vec->nb_elem;
+		pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			if (ev[i].sched_type == RTE_SCHED_TYPE_ATOMIC) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].queue_id++;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			ev[i].queue_id = tx_queue[ev[i].vec->port];
+			ev[i].vec->queue = 0;
+			vector_sz += ev[i].vec->nb_elem;
+			pipeline_fwd_event_vector(&ev[i],
+						  RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
 
 static __rte_noinline int
 pipeline_queue_worker_multi_stage_tx(void *arg)
@@ -267,6 +388,151 @@ pipeline_queue_worker_multi_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.queue_id % nb_stages;
+
+		if (ev.queue_id == tx_queue[ev.vec->port]) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+			continue;
+		}
+
+		ev.queue_id++;
+		pipeline_fwd_event_vector(&ev, cq_id != last_queue
+						       ? sched_type_list[cq_id]
+						       : RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.queue_id % nb_stages;
+
+		if (cq_id == last_queue) {
+			vector_sz = ev.vec->nb_elem;
+			ev.queue_id = tx_queue[ev.vec->port];
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.queue_id++;
+			pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+		}
+
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].queue_id % nb_stages;
+
+			if (ev[i].queue_id == tx_queue[ev[i].vec->port]) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+				continue;
+			}
+
+			ev[i].queue_id++;
+			pipeline_fwd_event_vector(
+				&ev[i], cq_id != last_queue
+						? sched_type_list[cq_id]
+						: RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].queue_id % nb_stages;
+
+			if (cq_id == last_queue) {
+				ev[i].queue_id = tx_queue[ev[i].vec->port];
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].queue_id++;
+				pipeline_fwd_event_vector(
+					&ev[i], sched_type_list[cq_id]);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
 static int
 worker_wrapper(void *arg)
 {
@@ -275,29 +541,39 @@ worker_wrapper(void *arg)
 	const bool burst = evt_has_burst_mode(w->dev_id);
 	const bool internal_port = w->t->internal_port;
 	const uint8_t nb_stages = opt->nb_stages;
-	RTE_SET_USED(opt);
-
-	if (nb_stages == 1) {
-		if (!burst && internal_port)
-			return pipeline_queue_worker_single_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_queue_worker_single_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_queue_worker_single_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_queue_worker_single_stage_burst_fwd(
-					arg);
-	} else {
-		if (!burst && internal_port)
-			return pipeline_queue_worker_multi_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_queue_worker_multi_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_queue_worker_multi_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_queue_worker_multi_stage_burst_fwd(arg);
+	/*vector/burst/internal_port*/
+	const pipeline_queue_worker_t
+	pipeline_queue_worker_single_stage[2][2][2] = {
+		[0][0][0] = pipeline_queue_worker_single_stage_fwd,
+		[0][0][1] = pipeline_queue_worker_single_stage_tx,
+		[0][1][0] = pipeline_queue_worker_single_stage_burst_fwd,
+		[0][1][1] = pipeline_queue_worker_single_stage_burst_tx,
+		[1][0][0] = pipeline_queue_worker_single_stage_fwd_vector,
+		[1][0][1] = pipeline_queue_worker_single_stage_tx_vector,
+		[1][1][0] = pipeline_queue_worker_single_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_queue_worker_single_stage_burst_tx_vector,
+	};
+	const pipeline_queue_worker_t
+	pipeline_queue_worker_multi_stage[2][2][2] = {
+		[0][0][0] = pipeline_queue_worker_multi_stage_fwd,
+		[0][0][1] = pipeline_queue_worker_multi_stage_tx,
+		[0][1][0] = pipeline_queue_worker_multi_stage_burst_fwd,
+		[0][1][1] = pipeline_queue_worker_multi_stage_burst_tx,
+		[1][0][0] = pipeline_queue_worker_multi_stage_fwd_vector,
+		[1][0][1] = pipeline_queue_worker_multi_stage_tx_vector,
+		[1][1][0] = pipeline_queue_worker_multi_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_queue_worker_multi_stage_burst_tx_vector,
+	};
+
+	if (nb_stages == 1)
+		return (pipeline_queue_worker_single_stage[opt->ena_vector]
+							  [burst]
+							  [internal_port])(arg);
+	else
+		return (pipeline_queue_worker_multi_stage[opt->ena_vector]
+							 [burst]
+							 [internal_port])(arg);
 
-	}
 	rte_panic("invalid worker\n");
 }
 
diff --git a/doc/guides/tools/testeventdev.rst b/doc/guides/tools/testeventdev.rst
index ad1788a3d..f252dc2c1 100644
--- a/doc/guides/tools/testeventdev.rst
+++ b/doc/guides/tools/testeventdev.rst
@@ -158,6 +158,26 @@ The following are the application command-line options:
        Set max packet mbuf size. Can be used configure Rx/Tx scatter gather.
        Only applicable for `pipeline_atq` and `pipeline_queue` tests.
 
+* ``--nb_eth_queues``
+
+       Configure multiple Rx queues per each ethernet port.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--enable_vector``
+
+       Enable event vector for Rx/Tx adapters.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--vector_size``
+
+       Vector size to configure for the Rx adapter.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--vector_tmo_ns``
+
+       Vector timeout nanoseconds to be configured for the Rx adapter.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
 
 Eventdev Tests
 --------------
@@ -607,6 +627,10 @@ Supported application command line options are following::
         --worker_deq_depth
         --prod_type_ethdev
         --deq_tmo_nsec
+        --nb_eth_queues
+        --enable_vector
+        --vector_size
+        --vector_tmo_ns
 
 
 .. Note::
@@ -623,6 +647,13 @@ Example command to run pipeline queue test:
     sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
         --test=pipeline_queue --wlcore=1 --prod_type_ethdev --stlist=a
 
+Example command to run pipeline atq test with vector events:
+
+.. code-block:: console
+
+    sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
+        --test=pipeline_queue --wlcore=1 --prod_type_ethdev --stlist=a \
+        --enable_vector  --vector_size 512
 
 PIPELINE_ATQ Test
 ~~~~~~~~~~~~~~~~~~~
@@ -699,6 +730,10 @@ Supported application command line options are following::
         --worker_deq_depth
         --prod_type_ethdev
         --deq_tmo_nsec
+        --nb_eth_queues
+        --enable_vector
+        --vector_size
+        --vector_tmo_ns
 
 
 .. Note::
@@ -708,9 +743,17 @@ Supported application command line options are following::
 Example
 ^^^^^^^
 
-Example command to run pipeline queue test:
+Example command to run pipeline atq test:
 
 .. code-block:: console
 
     sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
         --test=pipeline_atq --wlcore=1 --prod_type_ethdev --stlist=a
+
+Example command to run pipeline atq test with vector events:
+
+.. code-block:: console
+
+    sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
+        --test=pipeline_atq --wlcore=1 --prod_type_ethdev --stlist=a \
+        --enable_vector  --vector_size 512
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v9 7/8] doc: announce event Rx adapter config changes
  2021-03-30  8:22               ` [dpdk-dev] [PATCH v9 0/8] Introduce event vectorization pbhagavatula
                                   ` (5 preceding siblings ...)
  2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
@ 2021-03-30  8:22                 ` pbhagavatula
  2021-03-30  8:22                 ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v9 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
  2021-03-31  9:29                 ` [dpdk-dev] [PATCH v10 0/8] Introduce event vectorization pbhagavatula
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-30  8:22 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

The Rx adapter event vector configuration will be merged into
Rx adapter queue configuration to simplify enabling event
vectorization.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Ray Kinsella <mdr@ashroe.eu>
Acked-by: Jerin Jacob <jerinj@marvell.com>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
---
 doc/guides/rel_notes/deprecation.rst | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst
index 64629e064..ef6d1bb18 100644
--- a/doc/guides/rel_notes/deprecation.rst
+++ b/doc/guides/rel_notes/deprecation.rst
@@ -130,3 +130,12 @@ Deprecation Notices
 * cmdline: ``cmdline`` structure will be made opaque to hide platform-specific
   content. On Linux and FreeBSD, supported prior to DPDK 20.11,
   original structure will be kept until DPDK 21.11.
+
+* eventdev: The structure ``rte_event_eth_rx_adapter_queue_conf`` will be
+  extended to include ``rte_event_eth_rx_adapter_event_vector_config`` elements
+  and the function ``rte_event_eth_rx_adapter_queue_event_vector_config`` will
+  be removed in DPDK 21.11.
+
+  An application can enable event vectorization by passing the desired vector
+  values to the function ``rte_event_eth_rx_adapter_queue_add`` using
+  the structure ``rte_event_eth_rx_adapter_queue_add``.
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [dpdk-dev v21.11] [PATCH v9 8/8] eventdev: simplify Rx adapter event vector config
  2021-03-30  8:22               ` [dpdk-dev] [PATCH v9 0/8] Introduce event vectorization pbhagavatula
                                   ` (6 preceding siblings ...)
  2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 7/8] doc: announce event Rx adapter config changes pbhagavatula
@ 2021-03-30  8:22                 ` pbhagavatula
  2021-03-31  6:55                   ` Jayatheerthan, Jay
  2021-03-31  9:29                 ` [dpdk-dev] [PATCH v10 0/8] Introduce event vectorization pbhagavatula
  8 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-03-30  8:22 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Include vector configuration into the structure
``rte_event_eth_rx_adapter_queue_conf`` used when configuring rest
of the Rx adapter ethernet device Rx queue parameters.
This simplifies event vector configuration as it avoids splitting
configuration per Rx queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 app/test-eventdev/test_pipeline_common.c      |  16 +-
 lib/librte_eventdev/eventdev_pmd.h            |  29 ---
 .../rte_event_eth_rx_adapter.c                | 179 ++++++------------
 .../rte_event_eth_rx_adapter.h                |  27 ---
 lib/librte_eventdev/version.map               |   1 -
 5 files changed, 63 insertions(+), 189 deletions(-)

diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index d5ef90500..76aee254b 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -331,7 +331,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 	uint16_t prod;
 	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
-	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
@@ -397,8 +396,12 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			}

 			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.vector_sz = opt->vector_size;
+				queue_conf.vector_timeout_ns =
+					opt->vector_tmo_nsec;
 				queue_conf.rx_queue_flags |=
 				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+				queue_conf.vector_mp = vector_pool;
 			} else {
 				evt_err("Rx adapter doesn't support event vector");
 				return -EINVAL;
@@ -418,17 +421,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}

-		if (opt->ena_vector) {
-			vec_conf.vector_sz = opt->vector_size;
-			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
-			vec_conf.vector_mp = vector_pool;
-			if (rte_event_eth_rx_adapter_queue_event_vector_config(
-				    prod, prod, -1, &vec_conf) < 0) {
-				evt_err("Failed to configure event vectorization for Rx adapter");
-				return -EINVAL;
-			}
-		}
-
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;

diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 0f724ac85..63b3bc4b5 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -667,32 +667,6 @@ typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
 	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-struct rte_event_eth_rx_adapter_event_vector_config;
-/**
- * Enable event vector on an given Rx queue of a ethernet devices belonging to
- * the Rx adapter.
- *
- * @param dev
- *   Event device pointer
- *
- * @param eth_dev
- *   Ethernet device pointer
- *
- * @param rx_queue_id
- *   The Rx queue identifier
- *
- * @param config
- *   Pointer to the event vector configuration structure.
- *
- * @return
- *   - 0: Success.
- *   - <0: Error code returned by the driver function.
- */
-typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
-	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
-	int32_t rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;

@@ -1118,9 +1092,6 @@ struct rte_eventdev_ops {
 	eventdev_eth_rx_adapter_vector_limits_get_t
 		eth_rx_adapter_vector_limits_get;
 	/**< Get event vector limits for the Rx adapter */
-	eventdev_eth_rx_adapter_event_vector_config_t
-		eth_rx_adapter_event_vector_config;
-	/**< Configure Rx adapter with event vector */

 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index e273b3acf..5cdb3c3cb 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -1896,6 +1896,24 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	} else
 		qi_ev->flow_id = 0;

+	if (conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+		queue_info->ena_vector = 1;
+		qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+		rxa_set_vector_data(queue_info, conf->vector_sz,
+				    conf->vector_timeout_ns, conf->vector_mp,
+				    rx_queue_id, dev_info->dev->data->port_id);
+		rx_adapter->ena_vector = 1;
+		rx_adapter->vector_tmo_ticks =
+			rx_adapter->vector_tmo_ticks ?
+				      RTE_MIN(queue_info->vector_data
+							.vector_timeout_ticks >>
+						1,
+					rx_adapter->vector_tmo_ticks) :
+				queue_info->vector_data.vector_timeout_ticks >>
+					1;
+	}
+
 	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
 	if (rxa_polled_queue(dev_info, rx_queue_id)) {
 		rx_adapter->num_rx_polled += !pollq;
@@ -1921,42 +1939,6 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }

-static void
-rxa_sw_event_vector_configure(
-	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
-	int rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
-	struct eth_rx_queue_info *queue_info;
-	struct rte_event *qi_ev;
-
-	if (rx_queue_id == -1) {
-		uint16_t nb_rx_queues;
-		uint16_t i;
-
-		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
-		for (i = 0; i < nb_rx_queues; i++)
-			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
-						      config);
-		return;
-	}
-
-	queue_info = &dev_info->rx_queue[rx_queue_id];
-	qi_ev = (struct rte_event *)&queue_info->event;
-	queue_info->ena_vector = 1;
-	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
-	rxa_set_vector_data(queue_info, config->vector_sz,
-			    config->vector_timeout_ns, config->vector_mp,
-			    rx_queue_id, dev_info->dev->data->port_id);
-	rx_adapter->ena_vector = 1;
-	rx_adapter->vector_tmo_ticks =
-		rx_adapter->vector_tmo_ticks ?
-			      RTE_MIN(config->vector_timeout_ns >> 1,
-				rx_adapter->vector_tmo_ticks) :
-			      config->vector_timeout_ns >> 1;
-}
-
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -2271,6 +2253,7 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	struct rte_event_eth_rx_adapter *rx_adapter;
 	struct rte_eventdev *dev;
 	struct eth_device_info *dev_info;
+	struct rte_event_eth_rx_adapter_vector_limits limits;

 	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
 	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
@@ -2298,13 +2281,46 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}

-	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
-	    (queue_conf->rx_queue_flags &
-	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
-		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
+	if (queue_conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+
+		if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0) {
+			RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+
+		ret = rte_event_eth_rx_adapter_vector_limits_get(
+			rx_adapter->eventdev_id, eth_dev_id, &limits);
+		if (ret < 0) {
+			RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_sz < limits.min_sz ||
+		    queue_conf->vector_sz > limits.max_sz ||
+		    queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
+		    queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
+		    queue_conf->vector_mp == NULL) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_mp->elt_size <
+		    (sizeof(struct rte_event_vector) +
+		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
 	}

 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
@@ -2500,83 +2516,6 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }

-int
-rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct rte_event_eth_rx_adapter_vector_limits limits;
-	struct rte_event_eth_rx_adapter *rx_adapter;
-	struct rte_eventdev *dev;
-	uint32_t cap;
-	int ret;
-
-	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
-	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
-
-	rx_adapter = rxa_id_to_adapter(id);
-	if ((rx_adapter == NULL) || (config == NULL))
-		return -EINVAL;
-
-	dev = &rte_eventdevs[rx_adapter->eventdev_id];
-	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
-						eth_dev_id, &cap);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
-				 "eth port %" PRIu16,
-				 id, eth_dev_id);
-		return ret;
-	}
-
-	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
-		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	ret = rte_event_eth_rx_adapter_vector_limits_get(
-		rx_adapter->eventdev_id, eth_dev_id, &limits);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
-				 "eth port %" PRIu16,
-				 rx_adapter->eventdev_id, eth_dev_id);
-		return ret;
-	}
-
-	if (config->vector_sz < limits.min_sz ||
-	    config->vector_sz > limits.max_sz ||
-	    config->vector_timeout_ns < limits.min_timeout_ns ||
-	    config->vector_timeout_ns > limits.max_timeout_ns ||
-	    config->vector_mp == NULL) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-	if (config->vector_mp->elt_size <
-	    (sizeof(struct rte_event_vector) +
-	     (sizeof(uintptr_t) * config->vector_sz))) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
-		RTE_FUNC_PTR_OR_ERR_RET(
-			*dev->dev_ops->eth_rx_adapter_event_vector_config,
-			-ENOTSUP);
-		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
-			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
-	} else {
-		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
-					      rx_queue_id, config);
-	}
-
-	return ret;
-}
-
 int
 rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
index 7407cde00..3f8b36229 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
@@ -171,9 +171,6 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
 	 * enqueued event.
 	 */
-};
-
-struct rte_event_eth_rx_adapter_event_vector_config {
 	uint16_t vector_sz;
 	/**<
 	 * Indicates the maximum number for mbufs to combine and form a vector.
@@ -548,30 +545,6 @@ int rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-/**
- * Configure event vectorization for a given ethernet device queue, that has
- * been added to a event eth Rx adapter.
- *
- * @param id
- *  The identifier of the ethernet Rx event adapter.
- *
- * @param eth_dev_id
- *  The identifier of the ethernet device.
- *
- * @param rx_queue_id
- *  Ethernet device receive queue index.
- *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
- *  are configured with event vectorization.
- *
- * @return
- *  - 0: Success, Receive queue configured correctly.
- *  - <0: Error code on failure.
- */
-__rte_experimental
-int rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index 902df0ae3..34c1c830e 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -142,7 +142,6 @@ EXPERIMENTAL {
 	#added in 21.05
 	rte_event_vector_pool_create;
 	rte_event_eth_rx_adapter_vector_limits_get;
-	rte_event_eth_rx_adapter_queue_event_vector_config;
 };

 INTERNAL {
--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v9 4/8] eventdev: add Rx adapter event vector support
  2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 4/8] eventdev: add Rx adapter event vector support pbhagavatula
@ 2021-03-31  6:35                   ` Jayatheerthan, Jay
  2021-03-31  6:40                     ` Pavan Nikhilesh Bhagavatula
  0 siblings, 1 reply; 153+ messages in thread
From: Jayatheerthan, Jay @ 2021-03-31  6:35 UTC (permalink / raw)
  To: pbhagavatula, jerinj, Carrillo, Erik G, Gujjar, Abhinandan S,
	McDaniel, Timothy, hemant.agrawal, Van Haaren, Harry,
	mattias.ronnblom, Ma, Liang J
  Cc: dev

> -----Original Message-----
> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
> Sent: Tuesday, March 30, 2021 1:52 PM
> To: jerinj@marvell.com; Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Carrillo, Erik G <erik.g.carrillo@intel.com>; Gujjar,
> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy <timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom <mattias.ronnblom@ericsson.com>; Ma, Liang J
> <liang.j.ma@intel.com>
> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
> Subject: [dpdk-dev] [PATCH v9 4/8] eventdev: add Rx adapter event vector support
> 
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> Add event vector support for event eth Rx adapter, the implementation
> creates vector flows based on port and queue identifier of the received
> mbufs.
> The flow id for SW Rx event vectorization will use 12-bits of queue
> identifier and 8-bits port identifier when custom flow id is not set
> for simplicity.
> 
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> ---
>  .../prog_guide/event_ethernet_rx_adapter.rst  |  11 +
>  lib/librte_eventdev/eventdev_pmd.h            |   7 +-
>  .../rte_event_eth_rx_adapter.c                | 278 ++++++++++++++++--
>  lib/librte_eventdev/rte_eventdev.c            |   6 +-
>  4 files changed, 278 insertions(+), 24 deletions(-)
> 
> diff --git a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
> index 5eefef355..06fa864fa 100644
> --- a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
> +++ b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
> @@ -224,3 +224,14 @@ A loop processing ``rte_event_vector`` containing mbufs is shown below.
>          case ...
>          ...
>          }
> +
> +Rx event vectorization for SW Rx adapter
> +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> +
> +For SW based event vectorization, i.e., when the
> +``RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT`` is not set in the adapter's
> +capabilities flags for a particular ethernet device, the service function
> +creates a single event vector flow for all the mbufs arriving on the given
> +Rx queue.
> +The 20-bit event flow identifier is set to 12-bits of Rx queue identifier
> +and 8-bits of ethernet device identifier.
> diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
> index 9297f1433..0f724ac85 100644
> --- a/lib/librte_eventdev/eventdev_pmd.h
> +++ b/lib/librte_eventdev/eventdev_pmd.h
> @@ -69,9 +69,10 @@ extern "C" {
>  	} \
>  } while (0)
> 
> -#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP \
> -		((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) | \
> -			(RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ))
> +#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP                                        \
> +	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |                     \
> +	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) |                         \
> +	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR))
> 
>  #define RTE_EVENT_CRYPTO_ADAPTER_SW_CAP \
>  		RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA
> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> index ac8ba5bf0..e273b3acf 100644
> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> @@ -26,6 +26,10 @@
>  #define BATCH_SIZE		32
>  #define BLOCK_CNT_THRESHOLD	10
>  #define ETH_EVENT_BUFFER_SIZE	(4*BATCH_SIZE)
> +#define MAX_VECTOR_SIZE		1024
> +#define MIN_VECTOR_SIZE		4
> +#define MAX_VECTOR_NS		1E9
> +#define MIN_VECTOR_NS		1E5
> 
>  #define ETH_RX_ADAPTER_SERVICE_NAME_LEN	32
>  #define ETH_RX_ADAPTER_MEM_NAME_LEN	32
> @@ -59,6 +63,20 @@ struct eth_rx_poll_entry {
>  	uint16_t eth_rx_qid;
>  };
> 
> +struct eth_rx_vector_data {
> +	TAILQ_ENTRY(eth_rx_vector_data) next;
> +	uint16_t port;
> +	uint16_t queue;
> +	uint16_t max_vector_count;
> +	uint64_t event;
> +	uint64_t ts;
> +	uint64_t vector_timeout_ticks;
> +	struct rte_mempool *vector_pool;
> +	struct rte_event_vector *vector_ev;
> +} __rte_cache_aligned;
> +
> +TAILQ_HEAD(eth_rx_vector_data_list, eth_rx_vector_data);
> +
>  /* Instance per adapter */
>  struct rte_eth_event_enqueue_buffer {
>  	/* Count of events in this buffer */
> @@ -92,6 +110,14 @@ struct rte_event_eth_rx_adapter {
>  	uint32_t wrr_pos;
>  	/* Event burst buffer */
>  	struct rte_eth_event_enqueue_buffer event_enqueue_buffer;
> +	/* Vector enable flag */
> +	uint8_t ena_vector;
> +	/* Timestamp of previous vector expiry list traversal */
> +	uint64_t prev_expiry_ts;
> +	/* Minimum ticks to wait before traversing expiry list */
> +	uint64_t vector_tmo_ticks;
> +	/* vector list */
> +	struct eth_rx_vector_data_list vector_list;
>  	/* Per adapter stats */
>  	struct rte_event_eth_rx_adapter_stats stats;
>  	/* Block count, counts up to BLOCK_CNT_THRESHOLD */
> @@ -198,9 +224,11 @@ struct eth_device_info {
>  struct eth_rx_queue_info {
>  	int queue_enabled;	/* True if added */
>  	int intr_enabled;
> +	uint8_t ena_vector;
>  	uint16_t wt;		/* Polling weight */
>  	uint32_t flow_id_mask;	/* Set to ~0 if app provides flow id else 0 */
>  	uint64_t event;
> +	struct eth_rx_vector_data vector_data;
>  };
> 
>  static struct rte_event_eth_rx_adapter **event_eth_rx_adapter;
> @@ -722,6 +750,9 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
>  	    &rx_adapter->event_enqueue_buffer;
>  	struct rte_event_eth_rx_adapter_stats *stats = &rx_adapter->stats;
> 
> +	if (!buf->count)
> +		return 0;
> +
>  	uint16_t n = rte_event_enqueue_new_burst(rx_adapter->eventdev_id,
>  					rx_adapter->event_port_id,
>  					buf->events,
> @@ -742,6 +773,77 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
>  	return n;
>  }
> 
> +static inline void
> +rxa_init_vector(struct rte_event_eth_rx_adapter *rx_adapter,
> +		struct eth_rx_vector_data *vec)
> +{
> +	vec->vector_ev->nb_elem = 0;
> +	vec->vector_ev->port = vec->port;
> +	vec->vector_ev->queue = vec->queue;
> +	vec->vector_ev->attr_valid = true;
> +	TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
> +}
> +
> +static inline uint16_t
> +rxa_create_event_vector(struct rte_event_eth_rx_adapter *rx_adapter,
> +			struct eth_rx_queue_info *queue_info,
> +			struct rte_eth_event_enqueue_buffer *buf,
> +			struct rte_mbuf **mbufs, uint16_t num)
> +{
> +	struct rte_event *ev = &buf->events[buf->count];
> +	struct eth_rx_vector_data *vec;
> +	uint16_t filled, space, sz;
> +
> +	filled = 0;
> +	vec = &queue_info->vector_data;
> +
> +	if (vec->vector_ev == NULL) {
> +		if (rte_mempool_get(vec->vector_pool,
> +				    (void **)&vec->vector_ev) < 0) {
> +			rte_pktmbuf_free_bulk(mbufs, num);
> +			return 0;
> +		}
> +		rxa_init_vector(rx_adapter, vec);
> +	}
> +	while (num) {
> +		if (vec->vector_ev->nb_elem == vec->max_vector_count) {
> +			/* Event ready. */
> +			ev->event = vec->event;
> +			ev->vec = vec->vector_ev;
> +			ev++;
> +			filled++;
> +			vec->vector_ev = NULL;
> +			TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
> +			if (rte_mempool_get(vec->vector_pool,
> +					    (void **)&vec->vector_ev) < 0) {
> +				rte_pktmbuf_free_bulk(mbufs, num);
> +				return 0;
> +			}
> +			rxa_init_vector(rx_adapter, vec);
> +		}
> +
> +		space = vec->max_vector_count - vec->vector_ev->nb_elem;
> +		sz = num > space ? space : num;
> +		memcpy(vec->vector_ev->mbufs + vec->vector_ev->nb_elem, mbufs,
> +		       sizeof(void *) * sz);
> +		vec->vector_ev->nb_elem += sz;
> +		num -= sz;
> +		mbufs += sz;
> +		vec->ts = rte_rdtsc();
> +	}
> +
> +	if (vec->vector_ev->nb_elem == vec->max_vector_count) {
> +		ev->event = vec->event;
> +		ev->vec = vec->vector_ev;
> +		ev++;
> +		filled++;
> +		vec->vector_ev = NULL;
> +		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
> +	}
> +
> +	return filled;
> +}
> +
>  static inline void
>  rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
>  		uint16_t eth_dev_id,
> @@ -766,29 +868,33 @@ rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
>  	uint16_t nb_cb;
>  	uint16_t dropped;
> 
> -	/* 0xffff ffff if PKT_RX_RSS_HASH is set, otherwise 0 */
> -	rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
> -	do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
> -
> -	for (i = 0; i < num; i++) {
> -		m = mbufs[i];
> -
> -		rss = do_rss ?
> -			rxa_do_softrss(m, rx_adapter->rss_key_be) :
> -			m->hash.rss;
> -		ev->event = event;
> -		ev->flow_id = (rss & ~flow_id_mask) |
> -				(ev->flow_id & flow_id_mask);
> -		ev->mbuf = m;
> -		ev++;
> +	if (!eth_rx_queue_info->ena_vector) {
> +		/* 0xffff ffff if PKT_RX_RSS_HASH is set, otherwise 0 */
> +		rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
> +		do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
> +		for (i = 0; i < num; i++) {
> +			m = mbufs[i];
> +
> +			rss = do_rss ? rxa_do_softrss(m, rx_adapter->rss_key_be)
> +				     : m->hash.rss;
> +			ev->event = event;
> +			ev->flow_id = (rss & ~flow_id_mask) |
> +				      (ev->flow_id & flow_id_mask);
> +			ev->mbuf = m;
> +			ev++;
> +		}
> +	} else {
> +		num = rxa_create_event_vector(rx_adapter, eth_rx_queue_info,
> +					      buf, mbufs, num);
>  	}
> 
> -	if (dev_info->cb_fn) {
> +	if (num && dev_info->cb_fn) {
> 
>  		dropped = 0;
>  		nb_cb = dev_info->cb_fn(eth_dev_id, rx_queue_id,
> -					ETH_EVENT_BUFFER_SIZE, buf->count, ev,
> -					num, dev_info->cb_arg, &dropped);
> +					ETH_EVENT_BUFFER_SIZE, buf->count,
> +					&buf->events[buf->count], num,
> +					dev_info->cb_arg, &dropped);
>  		if (unlikely(nb_cb > num))
>  			RTE_EDEV_LOG_ERR("Rx CB returned %d (> %d) events",
>  				nb_cb, num);
> @@ -1124,6 +1230,30 @@ rxa_poll(struct rte_event_eth_rx_adapter *rx_adapter)
>  	return nb_rx;
>  }
> 
> +static void
> +rxa_vector_expire(struct eth_rx_vector_data *vec, void *arg)
> +{
> +	struct rte_event_eth_rx_adapter *rx_adapter = arg;
> +	struct rte_eth_event_enqueue_buffer *buf =
> +		&rx_adapter->event_enqueue_buffer;
> +	struct rte_event *ev;
> +
> +	if (buf->count)
> +		rxa_flush_event_buffer(rx_adapter);
> +
> +	if (vec->vector_ev->nb_elem == 0)
> +		return;
> +	ev = &buf->events[buf->count];
> +
> +	/* Event ready. */
> +	ev->event = vec->event;
> +	ev->vec = vec->vector_ev;
> +	buf->count++;
> +
> +	vec->vector_ev = NULL;
> +	vec->ts = 0;
> +}
> +
>  static int
>  rxa_service_func(void *args)
>  {
> @@ -1137,6 +1267,24 @@ rxa_service_func(void *args)
>  		return 0;
>  	}
> 
> +	if (rx_adapter->ena_vector) {
> +		if ((rte_rdtsc() - rx_adapter->prev_expiry_ts) >=
> +		    rx_adapter->vector_tmo_ticks) {
> +			struct eth_rx_vector_data *vec;
> +
> +			TAILQ_FOREACH(vec, &rx_adapter->vector_list, next) {
> +				uint64_t elapsed_time = rte_rdtsc() - vec->ts;
> +
> +				if (elapsed_time >= vec->vector_timeout_ticks) {
> +					rxa_vector_expire(vec, rx_adapter);
> +					TAILQ_REMOVE(&rx_adapter->vector_list,
> +						     vec, next);
> +				}
> +			}
> +			rx_adapter->prev_expiry_ts = rte_rdtsc();
> +		}
> +	}
> +
>  	stats = &rx_adapter->stats;
>  	stats->rx_packets += rxa_intr_ring_dequeue(rx_adapter);
>  	stats->rx_packets += rxa_poll(rx_adapter);
> @@ -1640,11 +1788,35 @@ rxa_update_queue(struct rte_event_eth_rx_adapter *rx_adapter,
>  	}
>  }
> 
> +static void
> +rxa_set_vector_data(struct eth_rx_queue_info *queue_info, uint16_t vector_count,
> +		    uint64_t vector_ns, struct rte_mempool *mp, int32_t qid,
> +		    uint16_t port_id)
> +{
> +#define NSEC2TICK(__ns, __freq) (((__ns) * (__freq)) / 1E9)
> +	struct eth_rx_vector_data *vector_data;
> +	uint32_t flow_id;
> +
> +	vector_data = &queue_info->vector_data;
> +	vector_data->max_vector_count = vector_count;
> +	vector_data->port = port_id;
> +	vector_data->queue = qid;
> +	vector_data->vector_pool = mp;
> +	vector_data->vector_timeout_ticks =
> +		NSEC2TICK(vector_ns, rte_get_timer_hz());
> +	vector_data->ts = 0;
> +	flow_id = queue_info->event & 0xFFFFF;
> +	flow_id =
> +		flow_id == 0 ? (qid & 0xFFF) | (port_id & 0xFF) << 12 : flow_id;
> +	vector_data->event = (queue_info->event & ~0xFFFFF) | flow_id;
> +}
> +
>  static void
>  rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
>  	struct eth_device_info *dev_info,
>  	int32_t rx_queue_id)
>  {
> +	struct eth_rx_vector_data *vec;
>  	int pollq;
>  	int intrq;
>  	int sintrq;
> @@ -1663,6 +1835,14 @@ rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
>  		return;
>  	}
> 
> +	/* Push all the partial event vectors to event device. */
> +	TAILQ_FOREACH(vec, &rx_adapter->vector_list, next) {
> +		if (vec->queue != rx_queue_id)
> +			continue;
> +		rxa_vector_expire(vec, rx_adapter);
> +		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
> +	}
> +

We are doing packet related activity (rxa_flush_event_buffer()) outside of rxa_service_func() although it wouldn't be running since queue del code has the lock. It would also be done in the context of a control thread. I don't know if there is a precedence for this. What do you think of just freeing the vector data and mbufs ?

>  	pollq = rxa_polled_queue(dev_info, rx_queue_id);
>  	intrq = rxa_intr_queue(dev_info, rx_queue_id);
>  	sintrq = rxa_shared_intr(dev_info, rx_queue_id);
> @@ -1741,6 +1921,42 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
>  	}
>  }
> 
> +static void
> +rxa_sw_event_vector_configure(
> +	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
> +	int rx_queue_id,
> +	const struct rte_event_eth_rx_adapter_event_vector_config *config)
> +{
> +	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
> +	struct eth_rx_queue_info *queue_info;
> +	struct rte_event *qi_ev;
> +
> +	if (rx_queue_id == -1) {
> +		uint16_t nb_rx_queues;
> +		uint16_t i;
> +
> +		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
> +		for (i = 0; i < nb_rx_queues; i++)
> +			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
> +						      config);
> +		return;
> +	}
> +
> +	queue_info = &dev_info->rx_queue[rx_queue_id];
> +	qi_ev = (struct rte_event *)&queue_info->event;
> +	queue_info->ena_vector = 1;
> +	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
> +	rxa_set_vector_data(queue_info, config->vector_sz,
> +			    config->vector_timeout_ns, config->vector_mp,
> +			    rx_queue_id, dev_info->dev->data->port_id);
> +	rx_adapter->ena_vector = 1;
> +	rx_adapter->vector_tmo_ticks =
> +		rx_adapter->vector_tmo_ticks ?
> +			      RTE_MIN(config->vector_timeout_ns >> 1,
> +				rx_adapter->vector_tmo_ticks) :
> +			      config->vector_timeout_ns >> 1;
> +}
> +
>  static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
>  		uint16_t eth_dev_id,
>  		int rx_queue_id,
> @@ -1967,6 +2183,7 @@ rte_event_eth_rx_adapter_create_ext(uint8_t id, uint8_t dev_id,
>  	rx_adapter->conf_cb = conf_cb;
>  	rx_adapter->conf_arg = conf_arg;
>  	rx_adapter->id = id;
> +	TAILQ_INIT(&rx_adapter->vector_list);
>  	strcpy(rx_adapter->mem_name, mem_name);
>  	rx_adapter->eth_devices = rte_zmalloc_socket(rx_adapter->mem_name,
>  					RTE_MAX_ETHPORTS *
> @@ -2081,6 +2298,15 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
>  		return -EINVAL;
>  	}
> 
> +	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
> +	    (queue_conf->rx_queue_flags &
> +	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
> +		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
> +				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
> +				 eth_dev_id, id);
> +		return -EINVAL;
> +	}
> +
>  	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
>  		(rx_queue_id != -1)) {
>  		RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
> @@ -2143,6 +2369,17 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
>  	return 0;
>  }
> 
> +static int
> +rxa_sw_vector_limits(struct rte_event_eth_rx_adapter_vector_limits *limits)
> +{
> +	limits->max_sz = MAX_VECTOR_SIZE;
> +	limits->min_sz = MIN_VECTOR_SIZE;
> +	limits->max_timeout_ns = MAX_VECTOR_NS;
> +	limits->min_timeout_ns = MIN_VECTOR_NS;
> +
> +	return 0;
> +}
> +
>  int
>  rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
>  				int32_t rx_queue_id)
> @@ -2333,7 +2570,8 @@ rte_event_eth_rx_adapter_queue_event_vector_config(
>  		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
>  			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
>  	} else {
> -		ret = -ENOTSUP;
> +		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
> +					      rx_queue_id, config);
>  	}
> 
>  	return ret;
> @@ -2371,7 +2609,7 @@ rte_event_eth_rx_adapter_vector_limits_get(
>  		ret = dev->dev_ops->eth_rx_adapter_vector_limits_get(
>  			dev, &rte_eth_devices[eth_port_id], limits);
>  	} else {
> -		ret = -ENOTSUP;
> +		ret = rxa_sw_vector_limits(limits);
>  	}
> 
>  	return ret;
> diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
> index be0499c52..62824654b 100644
> --- a/lib/librte_eventdev/rte_eventdev.c
> +++ b/lib/librte_eventdev/rte_eventdev.c
> @@ -122,7 +122,11 @@ rte_event_eth_rx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
> 
>  	if (caps == NULL)
>  		return -EINVAL;
> -	*caps = 0;
> +
> +	if (dev->dev_ops->eth_rx_adapter_caps_get == NULL)
> +		*caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
> +	else
> +		*caps = 0;
> 
>  	return dev->dev_ops->eth_rx_adapter_caps_get ?
>  				(*dev->dev_ops->eth_rx_adapter_caps_get)(dev,
> --
> 2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v9 4/8] eventdev: add Rx adapter event vector support
  2021-03-31  6:35                   ` Jayatheerthan, Jay
@ 2021-03-31  6:40                     ` Pavan Nikhilesh Bhagavatula
  2021-03-31  6:55                       ` Jayatheerthan, Jay
  0 siblings, 1 reply; 153+ messages in thread
From: Pavan Nikhilesh Bhagavatula @ 2021-03-31  6:40 UTC (permalink / raw)
  To: Jayatheerthan, Jay, Jerin Jacob Kollanukkaran, Carrillo, Erik G,
	Gujjar, Abhinandan S, McDaniel, Timothy, hemant.agrawal,
	Van Haaren, Harry, mattias.ronnblom, Ma, Liang J
  Cc: dev

>> -----Original Message-----
>> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
>> Sent: Tuesday, March 30, 2021 1:52 PM
>> To: jerinj@marvell.com; Jayatheerthan, Jay
><jay.jayatheerthan@intel.com>; Carrillo, Erik G
><erik.g.carrillo@intel.com>; Gujjar,
>> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy
><timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
>> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
><mattias.ronnblom@ericsson.com>; Ma, Liang J
>> <liang.j.ma@intel.com>
>> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
>> Subject: [dpdk-dev] [PATCH v9 4/8] eventdev: add Rx adapter event
>vector support
>>
>> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>>
>> Add event vector support for event eth Rx adapter, the
>implementation
>> creates vector flows based on port and queue identifier of the
>received
>> mbufs.
>> The flow id for SW Rx event vectorization will use 12-bits of queue
>> identifier and 8-bits port identifier when custom flow id is not set
>> for simplicity.
>>
>> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
>> ---
>>  .../prog_guide/event_ethernet_rx_adapter.rst  |  11 +
>>  lib/librte_eventdev/eventdev_pmd.h            |   7 +-
>>  .../rte_event_eth_rx_adapter.c                | 278 ++++++++++++++++--
>>  lib/librte_eventdev/rte_eventdev.c            |   6 +-
>>  4 files changed, 278 insertions(+), 24 deletions(-)
>>
>> diff --git a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
>b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
>> index 5eefef355..06fa864fa 100644
>> --- a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
>> +++ b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
>> @@ -224,3 +224,14 @@ A loop processing ``rte_event_vector``
>containing mbufs is shown below.
>>          case ...
>>          ...
>>          }
>> +
>> +Rx event vectorization for SW Rx adapter
>> +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>> +
>> +For SW based event vectorization, i.e., when the
>> +``RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT`` is not set
>in the adapter's
>> +capabilities flags for a particular ethernet device, the service
>function
>> +creates a single event vector flow for all the mbufs arriving on the
>given
>> +Rx queue.
>> +The 20-bit event flow identifier is set to 12-bits of Rx queue identifier
>> +and 8-bits of ethernet device identifier.
>> diff --git a/lib/librte_eventdev/eventdev_pmd.h
>b/lib/librte_eventdev/eventdev_pmd.h
>> index 9297f1433..0f724ac85 100644
>> --- a/lib/librte_eventdev/eventdev_pmd.h
>> +++ b/lib/librte_eventdev/eventdev_pmd.h
>> @@ -69,9 +69,10 @@ extern "C" {
>>  	} \
>>  } while (0)
>>
>> -#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP \
>> -
>	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |
>\
>> -
>	(RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ))
>> +#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP
>\
>> +	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |
>\
>> +	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) |
>\
>> +	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR))
>>
>>  #define RTE_EVENT_CRYPTO_ADAPTER_SW_CAP \
>>
>	RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA
>> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> index ac8ba5bf0..e273b3acf 100644
>> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
>> @@ -26,6 +26,10 @@
>>  #define BATCH_SIZE		32
>>  #define BLOCK_CNT_THRESHOLD	10
>>  #define ETH_EVENT_BUFFER_SIZE	(4*BATCH_SIZE)
>> +#define MAX_VECTOR_SIZE		1024
>> +#define MIN_VECTOR_SIZE		4
>> +#define MAX_VECTOR_NS		1E9
>> +#define MIN_VECTOR_NS		1E5
>>
>>  #define ETH_RX_ADAPTER_SERVICE_NAME_LEN	32
>>  #define ETH_RX_ADAPTER_MEM_NAME_LEN	32
>> @@ -59,6 +63,20 @@ struct eth_rx_poll_entry {
>>  	uint16_t eth_rx_qid;
>>  };
>>
>> +struct eth_rx_vector_data {
>> +	TAILQ_ENTRY(eth_rx_vector_data) next;
>> +	uint16_t port;
>> +	uint16_t queue;
>> +	uint16_t max_vector_count;
>> +	uint64_t event;
>> +	uint64_t ts;
>> +	uint64_t vector_timeout_ticks;
>> +	struct rte_mempool *vector_pool;
>> +	struct rte_event_vector *vector_ev;
>> +} __rte_cache_aligned;
>> +
>> +TAILQ_HEAD(eth_rx_vector_data_list, eth_rx_vector_data);
>> +
>>  /* Instance per adapter */
>>  struct rte_eth_event_enqueue_buffer {
>>  	/* Count of events in this buffer */
>> @@ -92,6 +110,14 @@ struct rte_event_eth_rx_adapter {
>>  	uint32_t wrr_pos;
>>  	/* Event burst buffer */
>>  	struct rte_eth_event_enqueue_buffer event_enqueue_buffer;
>> +	/* Vector enable flag */
>> +	uint8_t ena_vector;
>> +	/* Timestamp of previous vector expiry list traversal */
>> +	uint64_t prev_expiry_ts;
>> +	/* Minimum ticks to wait before traversing expiry list */
>> +	uint64_t vector_tmo_ticks;
>> +	/* vector list */
>> +	struct eth_rx_vector_data_list vector_list;
>>  	/* Per adapter stats */
>>  	struct rte_event_eth_rx_adapter_stats stats;
>>  	/* Block count, counts up to BLOCK_CNT_THRESHOLD */
>> @@ -198,9 +224,11 @@ struct eth_device_info {
>>  struct eth_rx_queue_info {
>>  	int queue_enabled;	/* True if added */
>>  	int intr_enabled;
>> +	uint8_t ena_vector;
>>  	uint16_t wt;		/* Polling weight */
>>  	uint32_t flow_id_mask;	/* Set to ~0 if app provides flow id else
>0 */
>>  	uint64_t event;
>> +	struct eth_rx_vector_data vector_data;
>>  };
>>
>>  static struct rte_event_eth_rx_adapter **event_eth_rx_adapter;
>> @@ -722,6 +750,9 @@ rxa_flush_event_buffer(struct
>rte_event_eth_rx_adapter *rx_adapter)
>>  	    &rx_adapter->event_enqueue_buffer;
>>  	struct rte_event_eth_rx_adapter_stats *stats = &rx_adapter-
>>stats;
>>
>> +	if (!buf->count)
>> +		return 0;
>> +
>>  	uint16_t n = rte_event_enqueue_new_burst(rx_adapter-
>>eventdev_id,
>>  					rx_adapter->event_port_id,
>>  					buf->events,
>> @@ -742,6 +773,77 @@ rxa_flush_event_buffer(struct
>rte_event_eth_rx_adapter *rx_adapter)
>>  	return n;
>>  }
>>
>> +static inline void
>> +rxa_init_vector(struct rte_event_eth_rx_adapter *rx_adapter,
>> +		struct eth_rx_vector_data *vec)
>> +{
>> +	vec->vector_ev->nb_elem = 0;
>> +	vec->vector_ev->port = vec->port;
>> +	vec->vector_ev->queue = vec->queue;
>> +	vec->vector_ev->attr_valid = true;
>> +	TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
>> +}
>> +
>> +static inline uint16_t
>> +rxa_create_event_vector(struct rte_event_eth_rx_adapter
>*rx_adapter,
>> +			struct eth_rx_queue_info *queue_info,
>> +			struct rte_eth_event_enqueue_buffer *buf,
>> +			struct rte_mbuf **mbufs, uint16_t num)
>> +{
>> +	struct rte_event *ev = &buf->events[buf->count];
>> +	struct eth_rx_vector_data *vec;
>> +	uint16_t filled, space, sz;
>> +
>> +	filled = 0;
>> +	vec = &queue_info->vector_data;
>> +
>> +	if (vec->vector_ev == NULL) {
>> +		if (rte_mempool_get(vec->vector_pool,
>> +				    (void **)&vec->vector_ev) < 0) {
>> +			rte_pktmbuf_free_bulk(mbufs, num);
>> +			return 0;
>> +		}
>> +		rxa_init_vector(rx_adapter, vec);
>> +	}
>> +	while (num) {
>> +		if (vec->vector_ev->nb_elem == vec-
>>max_vector_count) {
>> +			/* Event ready. */
>> +			ev->event = vec->event;
>> +			ev->vec = vec->vector_ev;
>> +			ev++;
>> +			filled++;
>> +			vec->vector_ev = NULL;
>> +			TAILQ_REMOVE(&rx_adapter->vector_list, vec,
>next);
>> +			if (rte_mempool_get(vec->vector_pool,
>> +					    (void **)&vec->vector_ev) <
>0) {
>> +				rte_pktmbuf_free_bulk(mbufs, num);
>> +				return 0;
>> +			}
>> +			rxa_init_vector(rx_adapter, vec);
>> +		}
>> +
>> +		space = vec->max_vector_count - vec->vector_ev-
>>nb_elem;
>> +		sz = num > space ? space : num;
>> +		memcpy(vec->vector_ev->mbufs + vec->vector_ev-
>>nb_elem, mbufs,
>> +		       sizeof(void *) * sz);
>> +		vec->vector_ev->nb_elem += sz;
>> +		num -= sz;
>> +		mbufs += sz;
>> +		vec->ts = rte_rdtsc();
>> +	}
>> +
>> +	if (vec->vector_ev->nb_elem == vec->max_vector_count) {
>> +		ev->event = vec->event;
>> +		ev->vec = vec->vector_ev;
>> +		ev++;
>> +		filled++;
>> +		vec->vector_ev = NULL;
>> +		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
>> +	}
>> +
>> +	return filled;
>> +}
>> +
>>  static inline void
>>  rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
>>  		uint16_t eth_dev_id,
>> @@ -766,29 +868,33 @@ rxa_buffer_mbufs(struct
>rte_event_eth_rx_adapter *rx_adapter,
>>  	uint16_t nb_cb;
>>  	uint16_t dropped;
>>
>> -	/* 0xffff ffff if PKT_RX_RSS_HASH is set, otherwise 0 */
>> -	rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
>> -	do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
>> -
>> -	for (i = 0; i < num; i++) {
>> -		m = mbufs[i];
>> -
>> -		rss = do_rss ?
>> -			rxa_do_softrss(m, rx_adapter->rss_key_be) :
>> -			m->hash.rss;
>> -		ev->event = event;
>> -		ev->flow_id = (rss & ~flow_id_mask) |
>> -				(ev->flow_id & flow_id_mask);
>> -		ev->mbuf = m;
>> -		ev++;
>> +	if (!eth_rx_queue_info->ena_vector) {
>> +		/* 0xffff ffff if PKT_RX_RSS_HASH is set, otherwise 0 */
>> +		rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) -
>1);
>> +		do_rss = !rss_mask && !eth_rx_queue_info-
>>flow_id_mask;
>> +		for (i = 0; i < num; i++) {
>> +			m = mbufs[i];
>> +
>> +			rss = do_rss ? rxa_do_softrss(m, rx_adapter-
>>rss_key_be)
>> +				     : m->hash.rss;
>> +			ev->event = event;
>> +			ev->flow_id = (rss & ~flow_id_mask) |
>> +				      (ev->flow_id & flow_id_mask);
>> +			ev->mbuf = m;
>> +			ev++;
>> +		}
>> +	} else {
>> +		num = rxa_create_event_vector(rx_adapter,
>eth_rx_queue_info,
>> +					      buf, mbufs, num);
>>  	}
>>
>> -	if (dev_info->cb_fn) {
>> +	if (num && dev_info->cb_fn) {
>>
>>  		dropped = 0;
>>  		nb_cb = dev_info->cb_fn(eth_dev_id, rx_queue_id,
>> -					ETH_EVENT_BUFFER_SIZE, buf-
>>count, ev,
>> -					num, dev_info->cb_arg,
>&dropped);
>> +					ETH_EVENT_BUFFER_SIZE, buf-
>>count,
>> +					&buf->events[buf->count],
>num,
>> +					dev_info->cb_arg, &dropped);
>>  		if (unlikely(nb_cb > num))
>>  			RTE_EDEV_LOG_ERR("Rx CB returned %d (> %d)
>events",
>>  				nb_cb, num);
>> @@ -1124,6 +1230,30 @@ rxa_poll(struct rte_event_eth_rx_adapter
>*rx_adapter)
>>  	return nb_rx;
>>  }
>>
>> +static void
>> +rxa_vector_expire(struct eth_rx_vector_data *vec, void *arg)
>> +{
>> +	struct rte_event_eth_rx_adapter *rx_adapter = arg;
>> +	struct rte_eth_event_enqueue_buffer *buf =
>> +		&rx_adapter->event_enqueue_buffer;
>> +	struct rte_event *ev;
>> +
>> +	if (buf->count)
>> +		rxa_flush_event_buffer(rx_adapter);
>> +
>> +	if (vec->vector_ev->nb_elem == 0)
>> +		return;
>> +	ev = &buf->events[buf->count];
>> +
>> +	/* Event ready. */
>> +	ev->event = vec->event;
>> +	ev->vec = vec->vector_ev;
>> +	buf->count++;
>> +
>> +	vec->vector_ev = NULL;
>> +	vec->ts = 0;
>> +}
>> +
>>  static int
>>  rxa_service_func(void *args)
>>  {
>> @@ -1137,6 +1267,24 @@ rxa_service_func(void *args)
>>  		return 0;
>>  	}
>>
>> +	if (rx_adapter->ena_vector) {
>> +		if ((rte_rdtsc() - rx_adapter->prev_expiry_ts) >=
>> +		    rx_adapter->vector_tmo_ticks) {
>> +			struct eth_rx_vector_data *vec;
>> +
>> +			TAILQ_FOREACH(vec, &rx_adapter->vector_list,
>next) {
>> +				uint64_t elapsed_time = rte_rdtsc() -
>vec->ts;
>> +
>> +				if (elapsed_time >= vec-
>>vector_timeout_ticks) {
>> +					rxa_vector_expire(vec,
>rx_adapter);
>> +					TAILQ_REMOVE(&rx_adapter-
>>vector_list,
>> +						     vec, next);
>> +				}
>> +			}
>> +			rx_adapter->prev_expiry_ts = rte_rdtsc();
>> +		}
>> +	}
>> +
>>  	stats = &rx_adapter->stats;
>>  	stats->rx_packets += rxa_intr_ring_dequeue(rx_adapter);
>>  	stats->rx_packets += rxa_poll(rx_adapter);
>> @@ -1640,11 +1788,35 @@ rxa_update_queue(struct
>rte_event_eth_rx_adapter *rx_adapter,
>>  	}
>>  }
>>
>> +static void
>> +rxa_set_vector_data(struct eth_rx_queue_info *queue_info,
>uint16_t vector_count,
>> +		    uint64_t vector_ns, struct rte_mempool *mp, int32_t
>qid,
>> +		    uint16_t port_id)
>> +{
>> +#define NSEC2TICK(__ns, __freq) (((__ns) * (__freq)) / 1E9)
>> +	struct eth_rx_vector_data *vector_data;
>> +	uint32_t flow_id;
>> +
>> +	vector_data = &queue_info->vector_data;
>> +	vector_data->max_vector_count = vector_count;
>> +	vector_data->port = port_id;
>> +	vector_data->queue = qid;
>> +	vector_data->vector_pool = mp;
>> +	vector_data->vector_timeout_ticks =
>> +		NSEC2TICK(vector_ns, rte_get_timer_hz());
>> +	vector_data->ts = 0;
>> +	flow_id = queue_info->event & 0xFFFFF;
>> +	flow_id =
>> +		flow_id == 0 ? (qid & 0xFFF) | (port_id & 0xFF) << 12 :
>flow_id;
>> +	vector_data->event = (queue_info->event & ~0xFFFFF) |
>flow_id;
>> +}
>> +
>>  static void
>>  rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
>>  	struct eth_device_info *dev_info,
>>  	int32_t rx_queue_id)
>>  {
>> +	struct eth_rx_vector_data *vec;
>>  	int pollq;
>>  	int intrq;
>>  	int sintrq;
>> @@ -1663,6 +1835,14 @@ rxa_sw_del(struct
>rte_event_eth_rx_adapter *rx_adapter,
>>  		return;
>>  	}
>>
>> +	/* Push all the partial event vectors to event device. */
>> +	TAILQ_FOREACH(vec, &rx_adapter->vector_list, next) {
>> +		if (vec->queue != rx_queue_id)
>> +			continue;
>> +		rxa_vector_expire(vec, rx_adapter);
>> +		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
>> +	}
>> +
>
>We are doing packet related activity (rxa_flush_event_buffer()) outside
>of rxa_service_func() although it wouldn't be running since queue del
>code has the lock. It would also be done in the context of a control
>thread. I don't know if there is a precedence for this. What do you think
>of just freeing the vector data and mbufs ?

Since we are just enqueueing to event device it should work fine.
The teardown sequence for event devices is to quiesce all producers first
and finally the event device.
This will make sure that the packets get handled properly with the stop 
callback registered through `rte_event_dev_stop_flush_callback_register`. 

>
>>  	pollq = rxa_polled_queue(dev_info, rx_queue_id);
>>  	intrq = rxa_intr_queue(dev_info, rx_queue_id);
>>  	sintrq = rxa_shared_intr(dev_info, rx_queue_id);
>> @@ -1741,6 +1921,42 @@ rxa_add_queue(struct
>rte_event_eth_rx_adapter *rx_adapter,
>>  	}
>>  }
>>
>> +static void
>> +rxa_sw_event_vector_configure(
>> +	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t
>eth_dev_id,
>> +	int rx_queue_id,
>> +	const struct rte_event_eth_rx_adapter_event_vector_config
>*config)
>> +{
>> +	struct eth_device_info *dev_info = &rx_adapter-
>>eth_devices[eth_dev_id];
>> +	struct eth_rx_queue_info *queue_info;
>> +	struct rte_event *qi_ev;
>> +
>> +	if (rx_queue_id == -1) {
>> +		uint16_t nb_rx_queues;
>> +		uint16_t i;
>> +
>> +		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
>> +		for (i = 0; i < nb_rx_queues; i++)
>> +			rxa_sw_event_vector_configure(rx_adapter,
>eth_dev_id, i,
>> +						      config);
>> +		return;
>> +	}
>> +
>> +	queue_info = &dev_info->rx_queue[rx_queue_id];
>> +	qi_ev = (struct rte_event *)&queue_info->event;
>> +	queue_info->ena_vector = 1;
>> +	qi_ev->event_type =
>RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
>> +	rxa_set_vector_data(queue_info, config->vector_sz,
>> +			    config->vector_timeout_ns, config-
>>vector_mp,
>> +			    rx_queue_id, dev_info->dev->data->port_id);
>> +	rx_adapter->ena_vector = 1;
>> +	rx_adapter->vector_tmo_ticks =
>> +		rx_adapter->vector_tmo_ticks ?
>> +			      RTE_MIN(config->vector_timeout_ns >> 1,
>> +				rx_adapter->vector_tmo_ticks) :
>> +			      config->vector_timeout_ns >> 1;
>> +}
>> +
>>  static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
>>  		uint16_t eth_dev_id,
>>  		int rx_queue_id,
>> @@ -1967,6 +2183,7 @@
>rte_event_eth_rx_adapter_create_ext(uint8_t id, uint8_t dev_id,
>>  	rx_adapter->conf_cb = conf_cb;
>>  	rx_adapter->conf_arg = conf_arg;
>>  	rx_adapter->id = id;
>> +	TAILQ_INIT(&rx_adapter->vector_list);
>>  	strcpy(rx_adapter->mem_name, mem_name);
>>  	rx_adapter->eth_devices = rte_zmalloc_socket(rx_adapter-
>>mem_name,
>>  					RTE_MAX_ETHPORTS *
>> @@ -2081,6 +2298,15 @@
>rte_event_eth_rx_adapter_queue_add(uint8_t id,
>>  		return -EINVAL;
>>  	}
>>
>> +	if ((cap &
>RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
>> +	    (queue_conf->rx_queue_flags &
>> +	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
>> +		RTE_EDEV_LOG_ERR("Event vectorization is not
>supported,"
>> +				 " eth port: %" PRIu16 " adapter id: %"
>PRIu8,
>> +				 eth_dev_id, id);
>> +		return -EINVAL;
>> +	}
>> +
>>  	if ((cap &
>RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
>>  		(rx_queue_id != -1)) {
>>  		RTE_EDEV_LOG_ERR("Rx queues can only be connected
>to single "
>> @@ -2143,6 +2369,17 @@
>rte_event_eth_rx_adapter_queue_add(uint8_t id,
>>  	return 0;
>>  }
>>
>> +static int
>> +rxa_sw_vector_limits(struct
>rte_event_eth_rx_adapter_vector_limits *limits)
>> +{
>> +	limits->max_sz = MAX_VECTOR_SIZE;
>> +	limits->min_sz = MIN_VECTOR_SIZE;
>> +	limits->max_timeout_ns = MAX_VECTOR_NS;
>> +	limits->min_timeout_ns = MIN_VECTOR_NS;
>> +
>> +	return 0;
>> +}
>> +
>>  int
>>  rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t
>eth_dev_id,
>>  				int32_t rx_queue_id)
>> @@ -2333,7 +2570,8 @@
>rte_event_eth_rx_adapter_queue_event_vector_config(
>>  		ret = dev->dev_ops-
>>eth_rx_adapter_event_vector_config(
>>  			dev, &rte_eth_devices[eth_dev_id],
>rx_queue_id, config);
>>  	} else {
>> -		ret = -ENOTSUP;
>> +		rxa_sw_event_vector_configure(rx_adapter,
>eth_dev_id,
>> +					      rx_queue_id, config);
>>  	}
>>
>>  	return ret;
>> @@ -2371,7 +2609,7 @@
>rte_event_eth_rx_adapter_vector_limits_get(
>>  		ret = dev->dev_ops-
>>eth_rx_adapter_vector_limits_get(
>>  			dev, &rte_eth_devices[eth_port_id], limits);
>>  	} else {
>> -		ret = -ENOTSUP;
>> +		ret = rxa_sw_vector_limits(limits);
>>  	}
>>
>>  	return ret;
>> diff --git a/lib/librte_eventdev/rte_eventdev.c
>b/lib/librte_eventdev/rte_eventdev.c
>> index be0499c52..62824654b 100644
>> --- a/lib/librte_eventdev/rte_eventdev.c
>> +++ b/lib/librte_eventdev/rte_eventdev.c
>> @@ -122,7 +122,11 @@ rte_event_eth_rx_adapter_caps_get(uint8_t
>dev_id, uint16_t eth_port_id,
>>
>>  	if (caps == NULL)
>>  		return -EINVAL;
>> -	*caps = 0;
>> +
>> +	if (dev->dev_ops->eth_rx_adapter_caps_get == NULL)
>> +		*caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
>> +	else
>> +		*caps = 0;
>>
>>  	return dev->dev_ops->eth_rx_adapter_caps_get ?
>>  				(*dev->dev_ops-
>>eth_rx_adapter_caps_get)(dev,
>> --
>> 2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v9 4/8] eventdev: add Rx adapter event vector support
  2021-03-31  6:40                     ` Pavan Nikhilesh Bhagavatula
@ 2021-03-31  6:55                       ` Jayatheerthan, Jay
  0 siblings, 0 replies; 153+ messages in thread
From: Jayatheerthan, Jay @ 2021-03-31  6:55 UTC (permalink / raw)
  To: Pavan Nikhilesh Bhagavatula, Jerin Jacob Kollanukkaran, Carrillo,
	Erik G, Gujjar, Abhinandan S, McDaniel, Timothy, hemant.agrawal,
	Van Haaren, Harry, mattias.ronnblom, Ma, Liang J
  Cc: dev

> -----Original Message-----
> From: Pavan Nikhilesh Bhagavatula <pbhagavatula@marvell.com>
> Sent: Wednesday, March 31, 2021 12:10 PM
> To: Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Jerin Jacob Kollanukkaran <jerinj@marvell.com>; Carrillo, Erik G
> <erik.g.carrillo@intel.com>; Gujjar, Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy
> <timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
> <mattias.ronnblom@ericsson.com>; Ma, Liang J <liang.j.ma@intel.com>
> Cc: dev@dpdk.org
> Subject: RE: [dpdk-dev] [PATCH v9 4/8] eventdev: add Rx adapter event vector support
> 
> >> -----Original Message-----
> >> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
> >> Sent: Tuesday, March 30, 2021 1:52 PM
> >> To: jerinj@marvell.com; Jayatheerthan, Jay
> ><jay.jayatheerthan@intel.com>; Carrillo, Erik G
> ><erik.g.carrillo@intel.com>; Gujjar,
> >> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy
> ><timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
> >> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom
> ><mattias.ronnblom@ericsson.com>; Ma, Liang J
> >> <liang.j.ma@intel.com>
> >> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
> >> Subject: [dpdk-dev] [PATCH v9 4/8] eventdev: add Rx adapter event
> >vector support
> >>
> >> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> >>
> >> Add event vector support for event eth Rx adapter, the
> >implementation
> >> creates vector flows based on port and queue identifier of the
> >received
> >> mbufs.
> >> The flow id for SW Rx event vectorization will use 12-bits of queue
> >> identifier and 8-bits port identifier when custom flow id is not set
> >> for simplicity.
> >>
> >> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> >> ---
> >>  .../prog_guide/event_ethernet_rx_adapter.rst  |  11 +
> >>  lib/librte_eventdev/eventdev_pmd.h            |   7 +-
> >>  .../rte_event_eth_rx_adapter.c                | 278 ++++++++++++++++--
> >>  lib/librte_eventdev/rte_eventdev.c            |   6 +-
> >>  4 files changed, 278 insertions(+), 24 deletions(-)
> >>
> >> diff --git a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
> >b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
> >> index 5eefef355..06fa864fa 100644
> >> --- a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
> >> +++ b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
> >> @@ -224,3 +224,14 @@ A loop processing ``rte_event_vector``
> >containing mbufs is shown below.
> >>          case ...
> >>          ...
> >>          }
> >> +
> >> +Rx event vectorization for SW Rx adapter
> >> +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> >> +
> >> +For SW based event vectorization, i.e., when the
> >> +``RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT`` is not set
> >in the adapter's
> >> +capabilities flags for a particular ethernet device, the service
> >function
> >> +creates a single event vector flow for all the mbufs arriving on the
> >given
> >> +Rx queue.
> >> +The 20-bit event flow identifier is set to 12-bits of Rx queue identifier
> >> +and 8-bits of ethernet device identifier.

Could you add a format like below ?

19        12,11                 0
+-----------+--------------+
| port_id | queue_id |
+-----------+--------------+

> >> diff --git a/lib/librte_eventdev/eventdev_pmd.h
> >b/lib/librte_eventdev/eventdev_pmd.h
> >> index 9297f1433..0f724ac85 100644
> >> --- a/lib/librte_eventdev/eventdev_pmd.h
> >> +++ b/lib/librte_eventdev/eventdev_pmd.h
> >> @@ -69,9 +69,10 @@ extern "C" {
> >>  	} \
> >>  } while (0)
> >>
> >> -#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP \
> >> -
> >	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |
> >\
> >> -
> >	(RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ))
> >> +#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP
> >\
> >> +	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |
> >\
> >> +	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) |
> >\
> >> +	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR))
> >>
> >>  #define RTE_EVENT_CRYPTO_ADAPTER_SW_CAP \
> >>
> >	RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA
> >> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> >b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> >> index ac8ba5bf0..e273b3acf 100644
> >> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> >> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> >> @@ -26,6 +26,10 @@
> >>  #define BATCH_SIZE		32
> >>  #define BLOCK_CNT_THRESHOLD	10
> >>  #define ETH_EVENT_BUFFER_SIZE	(4*BATCH_SIZE)
> >> +#define MAX_VECTOR_SIZE		1024
> >> +#define MIN_VECTOR_SIZE		4
> >> +#define MAX_VECTOR_NS		1E9
> >> +#define MIN_VECTOR_NS		1E5
> >>
> >>  #define ETH_RX_ADAPTER_SERVICE_NAME_LEN	32
> >>  #define ETH_RX_ADAPTER_MEM_NAME_LEN	32
> >> @@ -59,6 +63,20 @@ struct eth_rx_poll_entry {
> >>  	uint16_t eth_rx_qid;
> >>  };
> >>
> >> +struct eth_rx_vector_data {
> >> +	TAILQ_ENTRY(eth_rx_vector_data) next;
> >> +	uint16_t port;
> >> +	uint16_t queue;
> >> +	uint16_t max_vector_count;
> >> +	uint64_t event;
> >> +	uint64_t ts;
> >> +	uint64_t vector_timeout_ticks;
> >> +	struct rte_mempool *vector_pool;
> >> +	struct rte_event_vector *vector_ev;
> >> +} __rte_cache_aligned;
> >> +
> >> +TAILQ_HEAD(eth_rx_vector_data_list, eth_rx_vector_data);
> >> +
> >>  /* Instance per adapter */
> >>  struct rte_eth_event_enqueue_buffer {
> >>  	/* Count of events in this buffer */
> >> @@ -92,6 +110,14 @@ struct rte_event_eth_rx_adapter {
> >>  	uint32_t wrr_pos;
> >>  	/* Event burst buffer */
> >>  	struct rte_eth_event_enqueue_buffer event_enqueue_buffer;
> >> +	/* Vector enable flag */
> >> +	uint8_t ena_vector;
> >> +	/* Timestamp of previous vector expiry list traversal */
> >> +	uint64_t prev_expiry_ts;
> >> +	/* Minimum ticks to wait before traversing expiry list */
> >> +	uint64_t vector_tmo_ticks;
> >> +	/* vector list */
> >> +	struct eth_rx_vector_data_list vector_list;
> >>  	/* Per adapter stats */
> >>  	struct rte_event_eth_rx_adapter_stats stats;
> >>  	/* Block count, counts up to BLOCK_CNT_THRESHOLD */
> >> @@ -198,9 +224,11 @@ struct eth_device_info {
> >>  struct eth_rx_queue_info {
> >>  	int queue_enabled;	/* True if added */
> >>  	int intr_enabled;
> >> +	uint8_t ena_vector;
> >>  	uint16_t wt;		/* Polling weight */
> >>  	uint32_t flow_id_mask;	/* Set to ~0 if app provides flow id else
> >0 */
> >>  	uint64_t event;
> >> +	struct eth_rx_vector_data vector_data;
> >>  };
> >>
> >>  static struct rte_event_eth_rx_adapter **event_eth_rx_adapter;
> >> @@ -722,6 +750,9 @@ rxa_flush_event_buffer(struct
> >rte_event_eth_rx_adapter *rx_adapter)
> >>  	    &rx_adapter->event_enqueue_buffer;
> >>  	struct rte_event_eth_rx_adapter_stats *stats = &rx_adapter-
> >>stats;
> >>
> >> +	if (!buf->count)
> >> +		return 0;
> >> +
> >>  	uint16_t n = rte_event_enqueue_new_burst(rx_adapter-
> >>eventdev_id,
> >>  					rx_adapter->event_port_id,
> >>  					buf->events,
> >> @@ -742,6 +773,77 @@ rxa_flush_event_buffer(struct
> >rte_event_eth_rx_adapter *rx_adapter)
> >>  	return n;
> >>  }
> >>
> >> +static inline void
> >> +rxa_init_vector(struct rte_event_eth_rx_adapter *rx_adapter,
> >> +		struct eth_rx_vector_data *vec)
> >> +{
> >> +	vec->vector_ev->nb_elem = 0;
> >> +	vec->vector_ev->port = vec->port;
> >> +	vec->vector_ev->queue = vec->queue;
> >> +	vec->vector_ev->attr_valid = true;
> >> +	TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
> >> +}
> >> +
> >> +static inline uint16_t
> >> +rxa_create_event_vector(struct rte_event_eth_rx_adapter
> >*rx_adapter,
> >> +			struct eth_rx_queue_info *queue_info,
> >> +			struct rte_eth_event_enqueue_buffer *buf,
> >> +			struct rte_mbuf **mbufs, uint16_t num)
> >> +{
> >> +	struct rte_event *ev = &buf->events[buf->count];
> >> +	struct eth_rx_vector_data *vec;
> >> +	uint16_t filled, space, sz;
> >> +
> >> +	filled = 0;
> >> +	vec = &queue_info->vector_data;
> >> +
> >> +	if (vec->vector_ev == NULL) {
> >> +		if (rte_mempool_get(vec->vector_pool,
> >> +				    (void **)&vec->vector_ev) < 0) {
> >> +			rte_pktmbuf_free_bulk(mbufs, num);
> >> +			return 0;
> >> +		}
> >> +		rxa_init_vector(rx_adapter, vec);
> >> +	}
> >> +	while (num) {
> >> +		if (vec->vector_ev->nb_elem == vec-
> >>max_vector_count) {
> >> +			/* Event ready. */
> >> +			ev->event = vec->event;
> >> +			ev->vec = vec->vector_ev;
> >> +			ev++;
> >> +			filled++;
> >> +			vec->vector_ev = NULL;
> >> +			TAILQ_REMOVE(&rx_adapter->vector_list, vec,
> >next);
> >> +			if (rte_mempool_get(vec->vector_pool,
> >> +					    (void **)&vec->vector_ev) <
> >0) {
> >> +				rte_pktmbuf_free_bulk(mbufs, num);
> >> +				return 0;
> >> +			}
> >> +			rxa_init_vector(rx_adapter, vec);
> >> +		}
> >> +
> >> +		space = vec->max_vector_count - vec->vector_ev-
> >>nb_elem;
> >> +		sz = num > space ? space : num;
> >> +		memcpy(vec->vector_ev->mbufs + vec->vector_ev-
> >>nb_elem, mbufs,
> >> +		       sizeof(void *) * sz);
> >> +		vec->vector_ev->nb_elem += sz;
> >> +		num -= sz;
> >> +		mbufs += sz;
> >> +		vec->ts = rte_rdtsc();
> >> +	}
> >> +
> >> +	if (vec->vector_ev->nb_elem == vec->max_vector_count) {
> >> +		ev->event = vec->event;
> >> +		ev->vec = vec->vector_ev;
> >> +		ev++;
> >> +		filled++;
> >> +		vec->vector_ev = NULL;
> >> +		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
> >> +	}
> >> +
> >> +	return filled;
> >> +}
> >> +
> >>  static inline void
> >>  rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
> >>  		uint16_t eth_dev_id,
> >> @@ -766,29 +868,33 @@ rxa_buffer_mbufs(struct
> >rte_event_eth_rx_adapter *rx_adapter,
> >>  	uint16_t nb_cb;
> >>  	uint16_t dropped;
> >>
> >> -	/* 0xffff ffff if PKT_RX_RSS_HASH is set, otherwise 0 */
> >> -	rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
> >> -	do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
> >> -
> >> -	for (i = 0; i < num; i++) {
> >> -		m = mbufs[i];
> >> -
> >> -		rss = do_rss ?
> >> -			rxa_do_softrss(m, rx_adapter->rss_key_be) :
> >> -			m->hash.rss;
> >> -		ev->event = event;
> >> -		ev->flow_id = (rss & ~flow_id_mask) |
> >> -				(ev->flow_id & flow_id_mask);
> >> -		ev->mbuf = m;
> >> -		ev++;
> >> +	if (!eth_rx_queue_info->ena_vector) {
> >> +		/* 0xffff ffff if PKT_RX_RSS_HASH is set, otherwise 0 */
> >> +		rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) -
> >1);
> >> +		do_rss = !rss_mask && !eth_rx_queue_info-
> >>flow_id_mask;
> >> +		for (i = 0; i < num; i++) {
> >> +			m = mbufs[i];
> >> +
> >> +			rss = do_rss ? rxa_do_softrss(m, rx_adapter-
> >>rss_key_be)
> >> +				     : m->hash.rss;
> >> +			ev->event = event;
> >> +			ev->flow_id = (rss & ~flow_id_mask) |
> >> +				      (ev->flow_id & flow_id_mask);
> >> +			ev->mbuf = m;
> >> +			ev++;
> >> +		}
> >> +	} else {
> >> +		num = rxa_create_event_vector(rx_adapter,
> >eth_rx_queue_info,
> >> +					      buf, mbufs, num);
> >>  	}
> >>
> >> -	if (dev_info->cb_fn) {
> >> +	if (num && dev_info->cb_fn) {
> >>
> >>  		dropped = 0;
> >>  		nb_cb = dev_info->cb_fn(eth_dev_id, rx_queue_id,
> >> -					ETH_EVENT_BUFFER_SIZE, buf-
> >>count, ev,
> >> -					num, dev_info->cb_arg,
> >&dropped);
> >> +					ETH_EVENT_BUFFER_SIZE, buf-
> >>count,
> >> +					&buf->events[buf->count],
> >num,
> >> +					dev_info->cb_arg, &dropped);
> >>  		if (unlikely(nb_cb > num))
> >>  			RTE_EDEV_LOG_ERR("Rx CB returned %d (> %d)
> >events",
> >>  				nb_cb, num);
> >> @@ -1124,6 +1230,30 @@ rxa_poll(struct rte_event_eth_rx_adapter
> >*rx_adapter)
> >>  	return nb_rx;
> >>  }
> >>
> >> +static void
> >> +rxa_vector_expire(struct eth_rx_vector_data *vec, void *arg)
> >> +{
> >> +	struct rte_event_eth_rx_adapter *rx_adapter = arg;
> >> +	struct rte_eth_event_enqueue_buffer *buf =
> >> +		&rx_adapter->event_enqueue_buffer;
> >> +	struct rte_event *ev;
> >> +
> >> +	if (buf->count)
> >> +		rxa_flush_event_buffer(rx_adapter);
> >> +
> >> +	if (vec->vector_ev->nb_elem == 0)
> >> +		return;
> >> +	ev = &buf->events[buf->count];
> >> +
> >> +	/* Event ready. */
> >> +	ev->event = vec->event;
> >> +	ev->vec = vec->vector_ev;
> >> +	buf->count++;
> >> +
> >> +	vec->vector_ev = NULL;
> >> +	vec->ts = 0;
> >> +}
> >> +
> >>  static int
> >>  rxa_service_func(void *args)
> >>  {
> >> @@ -1137,6 +1267,24 @@ rxa_service_func(void *args)
> >>  		return 0;
> >>  	}
> >>
> >> +	if (rx_adapter->ena_vector) {
> >> +		if ((rte_rdtsc() - rx_adapter->prev_expiry_ts) >=
> >> +		    rx_adapter->vector_tmo_ticks) {
> >> +			struct eth_rx_vector_data *vec;
> >> +
> >> +			TAILQ_FOREACH(vec, &rx_adapter->vector_list,
> >next) {
> >> +				uint64_t elapsed_time = rte_rdtsc() -
> >vec->ts;
> >> +
> >> +				if (elapsed_time >= vec-
> >>vector_timeout_ticks) {
> >> +					rxa_vector_expire(vec,
> >rx_adapter);
> >> +					TAILQ_REMOVE(&rx_adapter-
> >>vector_list,
> >> +						     vec, next);
> >> +				}
> >> +			}
> >> +			rx_adapter->prev_expiry_ts = rte_rdtsc();
> >> +		}
> >> +	}
> >> +
> >>  	stats = &rx_adapter->stats;
> >>  	stats->rx_packets += rxa_intr_ring_dequeue(rx_adapter);
> >>  	stats->rx_packets += rxa_poll(rx_adapter);
> >> @@ -1640,11 +1788,35 @@ rxa_update_queue(struct
> >rte_event_eth_rx_adapter *rx_adapter,
> >>  	}
> >>  }
> >>
> >> +static void
> >> +rxa_set_vector_data(struct eth_rx_queue_info *queue_info,
> >uint16_t vector_count,
> >> +		    uint64_t vector_ns, struct rte_mempool *mp, int32_t
> >qid,
> >> +		    uint16_t port_id)
> >> +{
> >> +#define NSEC2TICK(__ns, __freq) (((__ns) * (__freq)) / 1E9)
> >> +	struct eth_rx_vector_data *vector_data;
> >> +	uint32_t flow_id;
> >> +
> >> +	vector_data = &queue_info->vector_data;
> >> +	vector_data->max_vector_count = vector_count;
> >> +	vector_data->port = port_id;
> >> +	vector_data->queue = qid;
> >> +	vector_data->vector_pool = mp;
> >> +	vector_data->vector_timeout_ticks =
> >> +		NSEC2TICK(vector_ns, rte_get_timer_hz());
> >> +	vector_data->ts = 0;
> >> +	flow_id = queue_info->event & 0xFFFFF;
> >> +	flow_id =
> >> +		flow_id == 0 ? (qid & 0xFFF) | (port_id & 0xFF) << 12 :
> >flow_id;
> >> +	vector_data->event = (queue_info->event & ~0xFFFFF) |
> >flow_id;
> >> +}
> >> +
> >>  static void
> >>  rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
> >>  	struct eth_device_info *dev_info,
> >>  	int32_t rx_queue_id)
> >>  {
> >> +	struct eth_rx_vector_data *vec;
> >>  	int pollq;
> >>  	int intrq;
> >>  	int sintrq;
> >> @@ -1663,6 +1835,14 @@ rxa_sw_del(struct
> >rte_event_eth_rx_adapter *rx_adapter,
> >>  		return;
> >>  	}
> >>
> >> +	/* Push all the partial event vectors to event device. */
> >> +	TAILQ_FOREACH(vec, &rx_adapter->vector_list, next) {
> >> +		if (vec->queue != rx_queue_id)
> >> +			continue;
> >> +		rxa_vector_expire(vec, rx_adapter);
> >> +		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
> >> +	}
> >> +
> >
> >We are doing packet related activity (rxa_flush_event_buffer()) outside
> >of rxa_service_func() although it wouldn't be running since queue del
> >code has the lock. It would also be done in the context of a control
> >thread. I don't know if there is a precedence for this. What do you think
> >of just freeing the vector data and mbufs ?
> 
> Since we are just enqueueing to event device it should work fine.
> The teardown sequence for event devices is to quiesce all producers first
> and finally the event device.
> This will make sure that the packets get handled properly with the stop
> callback registered through `rte_event_dev_stop_flush_callback_register`.
> 
> >
> >>  	pollq = rxa_polled_queue(dev_info, rx_queue_id);
> >>  	intrq = rxa_intr_queue(dev_info, rx_queue_id);
> >>  	sintrq = rxa_shared_intr(dev_info, rx_queue_id);
> >> @@ -1741,6 +1921,42 @@ rxa_add_queue(struct
> >rte_event_eth_rx_adapter *rx_adapter,
> >>  	}
> >>  }
> >>
> >> +static void
> >> +rxa_sw_event_vector_configure(
> >> +	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t
> >eth_dev_id,
> >> +	int rx_queue_id,
> >> +	const struct rte_event_eth_rx_adapter_event_vector_config
> >*config)
> >> +{
> >> +	struct eth_device_info *dev_info = &rx_adapter-
> >>eth_devices[eth_dev_id];
> >> +	struct eth_rx_queue_info *queue_info;
> >> +	struct rte_event *qi_ev;
> >> +
> >> +	if (rx_queue_id == -1) {
> >> +		uint16_t nb_rx_queues;
> >> +		uint16_t i;
> >> +
> >> +		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
> >> +		for (i = 0; i < nb_rx_queues; i++)
> >> +			rxa_sw_event_vector_configure(rx_adapter,
> >eth_dev_id, i,
> >> +						      config);
> >> +		return;
> >> +	}
> >> +
> >> +	queue_info = &dev_info->rx_queue[rx_queue_id];
> >> +	qi_ev = (struct rte_event *)&queue_info->event;
> >> +	queue_info->ena_vector = 1;
> >> +	qi_ev->event_type =
> >RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
> >> +	rxa_set_vector_data(queue_info, config->vector_sz,
> >> +			    config->vector_timeout_ns, config-
> >>vector_mp,
> >> +			    rx_queue_id, dev_info->dev->data->port_id);
> >> +	rx_adapter->ena_vector = 1;
> >> +	rx_adapter->vector_tmo_ticks =
> >> +		rx_adapter->vector_tmo_ticks ?
> >> +			      RTE_MIN(config->vector_timeout_ns >> 1,
> >> +				rx_adapter->vector_tmo_ticks) :
> >> +			      config->vector_timeout_ns >> 1;
> >> +}
> >> +
> >>  static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
> >>  		uint16_t eth_dev_id,
> >>  		int rx_queue_id,
> >> @@ -1967,6 +2183,7 @@
> >rte_event_eth_rx_adapter_create_ext(uint8_t id, uint8_t dev_id,
> >>  	rx_adapter->conf_cb = conf_cb;
> >>  	rx_adapter->conf_arg = conf_arg;
> >>  	rx_adapter->id = id;
> >> +	TAILQ_INIT(&rx_adapter->vector_list);
> >>  	strcpy(rx_adapter->mem_name, mem_name);
> >>  	rx_adapter->eth_devices = rte_zmalloc_socket(rx_adapter-
> >>mem_name,
> >>  					RTE_MAX_ETHPORTS *
> >> @@ -2081,6 +2298,15 @@
> >rte_event_eth_rx_adapter_queue_add(uint8_t id,
> >>  		return -EINVAL;
> >>  	}
> >>
> >> +	if ((cap &
> >RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
> >> +	    (queue_conf->rx_queue_flags &
> >> +	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
> >> +		RTE_EDEV_LOG_ERR("Event vectorization is not
> >supported,"
> >> +				 " eth port: %" PRIu16 " adapter id: %"
> >PRIu8,
> >> +				 eth_dev_id, id);
> >> +		return -EINVAL;
> >> +	}
> >> +
> >>  	if ((cap &
> >RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
> >>  		(rx_queue_id != -1)) {
> >>  		RTE_EDEV_LOG_ERR("Rx queues can only be connected
> >to single "
> >> @@ -2143,6 +2369,17 @@
> >rte_event_eth_rx_adapter_queue_add(uint8_t id,
> >>  	return 0;
> >>  }
> >>
> >> +static int
> >> +rxa_sw_vector_limits(struct
> >rte_event_eth_rx_adapter_vector_limits *limits)
> >> +{
> >> +	limits->max_sz = MAX_VECTOR_SIZE;
> >> +	limits->min_sz = MIN_VECTOR_SIZE;
> >> +	limits->max_timeout_ns = MAX_VECTOR_NS;
> >> +	limits->min_timeout_ns = MIN_VECTOR_NS;
> >> +
> >> +	return 0;
> >> +}
> >> +
> >>  int
> >>  rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t
> >eth_dev_id,
> >>  				int32_t rx_queue_id)
> >> @@ -2333,7 +2570,8 @@
> >rte_event_eth_rx_adapter_queue_event_vector_config(
> >>  		ret = dev->dev_ops-
> >>eth_rx_adapter_event_vector_config(
> >>  			dev, &rte_eth_devices[eth_dev_id],
> >rx_queue_id, config);
> >>  	} else {
> >> -		ret = -ENOTSUP;
> >> +		rxa_sw_event_vector_configure(rx_adapter,
> >eth_dev_id,
> >> +					      rx_queue_id, config);
> >>  	}
> >>
> >>  	return ret;
> >> @@ -2371,7 +2609,7 @@
> >rte_event_eth_rx_adapter_vector_limits_get(
> >>  		ret = dev->dev_ops-
> >>eth_rx_adapter_vector_limits_get(
> >>  			dev, &rte_eth_devices[eth_port_id], limits);
> >>  	} else {
> >> -		ret = -ENOTSUP;
> >> +		ret = rxa_sw_vector_limits(limits);
> >>  	}
> >>
> >>  	return ret;
> >> diff --git a/lib/librte_eventdev/rte_eventdev.c
> >b/lib/librte_eventdev/rte_eventdev.c
> >> index be0499c52..62824654b 100644
> >> --- a/lib/librte_eventdev/rte_eventdev.c
> >> +++ b/lib/librte_eventdev/rte_eventdev.c
> >> @@ -122,7 +122,11 @@ rte_event_eth_rx_adapter_caps_get(uint8_t
> >dev_id, uint16_t eth_port_id,
> >>
> >>  	if (caps == NULL)
> >>  		return -EINVAL;
> >> -	*caps = 0;
> >> +
> >> +	if (dev->dev_ops->eth_rx_adapter_caps_get == NULL)
> >> +		*caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
> >> +	else
> >> +		*caps = 0;
> >>
> >>  	return dev->dev_ops->eth_rx_adapter_caps_get ?
> >>  				(*dev->dev_ops-
> >>eth_rx_adapter_caps_get)(dev,
> >> --
> >> 2.17.1

With the above change, you can add Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>.


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [dpdk-dev v21.11] [PATCH v9 8/8] eventdev: simplify Rx adapter event vector config
  2021-03-30  8:22                 ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v9 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
@ 2021-03-31  6:55                   ` Jayatheerthan, Jay
  0 siblings, 0 replies; 153+ messages in thread
From: Jayatheerthan, Jay @ 2021-03-31  6:55 UTC (permalink / raw)
  To: pbhagavatula, jerinj, Carrillo, Erik G, Gujjar, Abhinandan S,
	McDaniel, Timothy, hemant.agrawal, Van Haaren, Harry,
	mattias.ronnblom, Ma, Liang J, Ray Kinsella, Neil Horman
  Cc: dev

> -----Original Message-----
> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
> Sent: Tuesday, March 30, 2021 1:52 PM
> To: jerinj@marvell.com; Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Carrillo, Erik G <erik.g.carrillo@intel.com>; Gujjar,
> Abhinandan S <abhinandan.gujjar@intel.com>; McDaniel, Timothy <timothy.mcdaniel@intel.com>; hemant.agrawal@nxp.com; Van
> Haaren, Harry <harry.van.haaren@intel.com>; mattias.ronnblom <mattias.ronnblom@ericsson.com>; Ma, Liang J
> <liang.j.ma@intel.com>; Ray Kinsella <mdr@ashroe.eu>; Neil Horman <nhorman@tuxdriver.com>
> Cc: dev@dpdk.org; Pavan Nikhilesh <pbhagavatula@marvell.com>
> Subject: [dpdk-dev v21.11] [PATCH v9 8/8] eventdev: simplify Rx adapter event vector config
> 
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> Include vector configuration into the structure
> ``rte_event_eth_rx_adapter_queue_conf`` used when configuring rest
> of the Rx adapter ethernet device Rx queue parameters.
> This simplifies event vector configuration as it avoids splitting
> configuration per Rx queue.
> 
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> ---
>  app/test-eventdev/test_pipeline_common.c      |  16 +-
>  lib/librte_eventdev/eventdev_pmd.h            |  29 ---
>  .../rte_event_eth_rx_adapter.c                | 179 ++++++------------
>  .../rte_event_eth_rx_adapter.h                |  27 ---
>  lib/librte_eventdev/version.map               |   1 -
>  5 files changed, 63 insertions(+), 189 deletions(-)
> 
> diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
> index d5ef90500..76aee254b 100644
> --- a/app/test-eventdev/test_pipeline_common.c
> +++ b/app/test-eventdev/test_pipeline_common.c
> @@ -331,7 +331,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
>  	uint16_t prod;
>  	struct rte_mempool *vector_pool = NULL;
>  	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
> -	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;
> 
>  	memset(&queue_conf, 0,
>  			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
> @@ -397,8 +396,12 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
>  			}
> 
>  			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
> +				queue_conf.vector_sz = opt->vector_size;
> +				queue_conf.vector_timeout_ns =
> +					opt->vector_tmo_nsec;
>  				queue_conf.rx_queue_flags |=
>  				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
> +				queue_conf.vector_mp = vector_pool;
>  			} else {
>  				evt_err("Rx adapter doesn't support event vector");
>  				return -EINVAL;
> @@ -418,17 +421,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
>  			return ret;
>  		}
> 
> -		if (opt->ena_vector) {
> -			vec_conf.vector_sz = opt->vector_size;
> -			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
> -			vec_conf.vector_mp = vector_pool;
> -			if (rte_event_eth_rx_adapter_queue_event_vector_config(
> -				    prod, prod, -1, &vec_conf) < 0) {
> -				evt_err("Failed to configure event vectorization for Rx adapter");
> -				return -EINVAL;
> -			}
> -		}
> -
>  		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
>  			uint32_t service_id = -1U;
> 
> diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
> index 0f724ac85..63b3bc4b5 100644
> --- a/lib/librte_eventdev/eventdev_pmd.h
> +++ b/lib/librte_eventdev/eventdev_pmd.h
> @@ -667,32 +667,6 @@ typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
>  	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
>  	struct rte_event_eth_rx_adapter_vector_limits *limits);
> 
> -struct rte_event_eth_rx_adapter_event_vector_config;
> -/**
> - * Enable event vector on an given Rx queue of a ethernet devices belonging to
> - * the Rx adapter.
> - *
> - * @param dev
> - *   Event device pointer
> - *
> - * @param eth_dev
> - *   Ethernet device pointer
> - *
> - * @param rx_queue_id
> - *   The Rx queue identifier
> - *
> - * @param config
> - *   Pointer to the event vector configuration structure.
> - *
> - * @return
> - *   - 0: Success.
> - *   - <0: Error code returned by the driver function.
> - */
> -typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
> -	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
> -	int32_t rx_queue_id,
> -	const struct rte_event_eth_rx_adapter_event_vector_config *config);
> -
>  typedef uint32_t rte_event_pmd_selftest_seqn_t;
>  extern int rte_event_pmd_selftest_seqn_dynfield_offset;
> 
> @@ -1118,9 +1092,6 @@ struct rte_eventdev_ops {
>  	eventdev_eth_rx_adapter_vector_limits_get_t
>  		eth_rx_adapter_vector_limits_get;
>  	/**< Get event vector limits for the Rx adapter */
> -	eventdev_eth_rx_adapter_event_vector_config_t
> -		eth_rx_adapter_event_vector_config;
> -	/**< Configure Rx adapter with event vector */
> 
>  	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
>  	/**< Get timer adapter capabilities */
> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> index e273b3acf..5cdb3c3cb 100644
> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> @@ -1896,6 +1896,24 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
>  	} else
>  		qi_ev->flow_id = 0;
> 
> +	if (conf->rx_queue_flags &
> +	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
> +		queue_info->ena_vector = 1;
> +		qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
> +		rxa_set_vector_data(queue_info, conf->vector_sz,
> +				    conf->vector_timeout_ns, conf->vector_mp,
> +				    rx_queue_id, dev_info->dev->data->port_id);
> +		rx_adapter->ena_vector = 1;
> +		rx_adapter->vector_tmo_ticks =
> +			rx_adapter->vector_tmo_ticks ?
> +				      RTE_MIN(queue_info->vector_data
> +							.vector_timeout_ticks >>
> +						1,
> +					rx_adapter->vector_tmo_ticks) :
> +				queue_info->vector_data.vector_timeout_ticks >>
> +					1;
> +	}
> +
>  	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
>  	if (rxa_polled_queue(dev_info, rx_queue_id)) {
>  		rx_adapter->num_rx_polled += !pollq;
> @@ -1921,42 +1939,6 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
>  	}
>  }
> 
> -static void
> -rxa_sw_event_vector_configure(
> -	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
> -	int rx_queue_id,
> -	const struct rte_event_eth_rx_adapter_event_vector_config *config)
> -{
> -	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
> -	struct eth_rx_queue_info *queue_info;
> -	struct rte_event *qi_ev;
> -
> -	if (rx_queue_id == -1) {
> -		uint16_t nb_rx_queues;
> -		uint16_t i;
> -
> -		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
> -		for (i = 0; i < nb_rx_queues; i++)
> -			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
> -						      config);
> -		return;
> -	}
> -
> -	queue_info = &dev_info->rx_queue[rx_queue_id];
> -	qi_ev = (struct rte_event *)&queue_info->event;
> -	queue_info->ena_vector = 1;
> -	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
> -	rxa_set_vector_data(queue_info, config->vector_sz,
> -			    config->vector_timeout_ns, config->vector_mp,
> -			    rx_queue_id, dev_info->dev->data->port_id);
> -	rx_adapter->ena_vector = 1;
> -	rx_adapter->vector_tmo_ticks =
> -		rx_adapter->vector_tmo_ticks ?
> -			      RTE_MIN(config->vector_timeout_ns >> 1,
> -				rx_adapter->vector_tmo_ticks) :
> -			      config->vector_timeout_ns >> 1;
> -}
> -
>  static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
>  		uint16_t eth_dev_id,
>  		int rx_queue_id,
> @@ -2271,6 +2253,7 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
>  	struct rte_event_eth_rx_adapter *rx_adapter;
>  	struct rte_eventdev *dev;
>  	struct eth_device_info *dev_info;
> +	struct rte_event_eth_rx_adapter_vector_limits limits;
> 
>  	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
>  	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
> @@ -2298,13 +2281,46 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
>  		return -EINVAL;
>  	}
> 
> -	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
> -	    (queue_conf->rx_queue_flags &
> -	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
> -		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
> -				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
> -				 eth_dev_id, id);
> -		return -EINVAL;
> +	if (queue_conf->rx_queue_flags &
> +	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
> +
> +		if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0) {
> +			RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
> +					 " eth port: %" PRIu16
> +					 " adapter id: %" PRIu8,
> +					 eth_dev_id, id);
> +			return -EINVAL;
> +		}
> +
> +		ret = rte_event_eth_rx_adapter_vector_limits_get(
> +			rx_adapter->eventdev_id, eth_dev_id, &limits);
> +		if (ret < 0) {
> +			RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
> +					 " eth port: %" PRIu16
> +					 " adapter id: %" PRIu8,
> +					 eth_dev_id, id);
> +			return -EINVAL;
> +		}
> +		if (queue_conf->vector_sz < limits.min_sz ||
> +		    queue_conf->vector_sz > limits.max_sz ||
> +		    queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
> +		    queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
> +		    queue_conf->vector_mp == NULL) {
> +			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> +					 " eth port: %" PRIu16
> +					 " adapter id: %" PRIu8,
> +					 eth_dev_id, id);
> +			return -EINVAL;
> +		}
> +		if (queue_conf->vector_mp->elt_size <
> +		    (sizeof(struct rte_event_vector) +
> +		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
> +			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> +					 " eth port: %" PRIu16
> +					 " adapter id: %" PRIu8,
> +					 eth_dev_id, id);
> +			return -EINVAL;
> +		}
>  	}
> 
>  	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
> @@ -2500,83 +2516,6 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
>  	return ret;
>  }
> 
> -int
> -rte_event_eth_rx_adapter_queue_event_vector_config(
> -	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
> -	struct rte_event_eth_rx_adapter_event_vector_config *config)
> -{
> -	struct rte_event_eth_rx_adapter_vector_limits limits;
> -	struct rte_event_eth_rx_adapter *rx_adapter;
> -	struct rte_eventdev *dev;
> -	uint32_t cap;
> -	int ret;
> -
> -	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
> -	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
> -
> -	rx_adapter = rxa_id_to_adapter(id);
> -	if ((rx_adapter == NULL) || (config == NULL))
> -		return -EINVAL;
> -
> -	dev = &rte_eventdevs[rx_adapter->eventdev_id];
> -	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
> -						eth_dev_id, &cap);
> -	if (ret) {
> -		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
> -				 "eth port %" PRIu16,
> -				 id, eth_dev_id);
> -		return ret;
> -	}
> -
> -	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
> -		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
> -				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
> -				 eth_dev_id, id);
> -		return -EINVAL;
> -	}
> -
> -	ret = rte_event_eth_rx_adapter_vector_limits_get(
> -		rx_adapter->eventdev_id, eth_dev_id, &limits);
> -	if (ret) {
> -		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
> -				 "eth port %" PRIu16,
> -				 rx_adapter->eventdev_id, eth_dev_id);
> -		return ret;
> -	}
> -
> -	if (config->vector_sz < limits.min_sz ||
> -	    config->vector_sz > limits.max_sz ||
> -	    config->vector_timeout_ns < limits.min_timeout_ns ||
> -	    config->vector_timeout_ns > limits.max_timeout_ns ||
> -	    config->vector_mp == NULL) {
> -		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> -				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
> -				 eth_dev_id, id);
> -		return -EINVAL;
> -	}
> -	if (config->vector_mp->elt_size <
> -	    (sizeof(struct rte_event_vector) +
> -	     (sizeof(uintptr_t) * config->vector_sz))) {
> -		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> -				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
> -				 eth_dev_id, id);
> -		return -EINVAL;
> -	}
> -
> -	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
> -		RTE_FUNC_PTR_OR_ERR_RET(
> -			*dev->dev_ops->eth_rx_adapter_event_vector_config,
> -			-ENOTSUP);
> -		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
> -			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
> -	} else {
> -		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
> -					      rx_queue_id, config);
> -	}
> -
> -	return ret;
> -}
> -
>  int
>  rte_event_eth_rx_adapter_vector_limits_get(
>  	uint8_t dev_id, uint16_t eth_port_id,
> diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> index 7407cde00..3f8b36229 100644
> --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
> @@ -171,9 +171,6 @@ struct rte_event_eth_rx_adapter_queue_conf {
>  	 * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
>  	 * enqueued event.
>  	 */
> -};
> -
> -struct rte_event_eth_rx_adapter_event_vector_config {
>  	uint16_t vector_sz;
>  	/**<
>  	 * Indicates the maximum number for mbufs to combine and form a vector.
> @@ -548,30 +545,6 @@ int rte_event_eth_rx_adapter_vector_limits_get(
>  	uint8_t dev_id, uint16_t eth_port_id,
>  	struct rte_event_eth_rx_adapter_vector_limits *limits);
> 
> -/**
> - * Configure event vectorization for a given ethernet device queue, that has
> - * been added to a event eth Rx adapter.
> - *
> - * @param id
> - *  The identifier of the ethernet Rx event adapter.
> - *
> - * @param eth_dev_id
> - *  The identifier of the ethernet device.
> - *
> - * @param rx_queue_id
> - *  Ethernet device receive queue index.
> - *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
> - *  are configured with event vectorization.
> - *
> - * @return
> - *  - 0: Success, Receive queue configured correctly.
> - *  - <0: Error code on failure.
> - */
> -__rte_experimental
> -int rte_event_eth_rx_adapter_queue_event_vector_config(
> -	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
> -	struct rte_event_eth_rx_adapter_event_vector_config *config);
> -
>  #ifdef __cplusplus
>  }
>  #endif
> diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
> index 902df0ae3..34c1c830e 100644
> --- a/lib/librte_eventdev/version.map
> +++ b/lib/librte_eventdev/version.map
> @@ -142,7 +142,6 @@ EXPERIMENTAL {
>  	#added in 21.05
>  	rte_event_vector_pool_create;
>  	rte_event_eth_rx_adapter_vector_limits_get;
> -	rte_event_eth_rx_adapter_queue_event_vector_config;
>  };
> 
>  INTERNAL {
> --
> 2.17.1

Looks good. Thanks for all your effort and help!

Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v10 0/8] Introduce event vectorization
  2021-03-30  8:22               ` [dpdk-dev] [PATCH v9 0/8] Introduce event vectorization pbhagavatula
                                   ` (7 preceding siblings ...)
  2021-03-30  8:22                 ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v9 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
@ 2021-03-31  9:29                 ` pbhagavatula
  2021-03-31  9:29                   ` [dpdk-dev] [PATCH v10 1/8] eventdev: introduce event vector capability pbhagavatula
                                     ` (8 more replies)
  8 siblings, 9 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-31  9:29 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

In traditional event programming model, events are identified by a
flow-id and a uintptr_t. The flow-id uniquely identifies a given event
and determines the order of scheduling based on schedule type, the
uintptr_t holds a single object.

Event devices also support burst mode with configurable dequeue depth,
i.e. each dequeue call would return multiple events and each event
might be at a different stage of the pipeline.
Having a burst of events belonging to different stages in a dequeue
burst is not only difficult to vectorize but also increases the scheduler
overhead and application overhead of pipelining events further.
Using event vectors we see a performance gain of ~742.3% as shown in [1].

By introducing event vectorization, each event will be capable of holding
multiple uintptr_t of the same flow thereby allowing applications
to vectorize their pipeline and reduce the complexity of pipelining
events across multiple stages. This also reduces the complexity of handling
enqueue and dequeue on an event device.

Since event devices are transparent to the events they are scheduling
so the event producers such as eth_rx_adapter, crypto_adapter , etc..
are responsible for vectorizing the buffers of the same flow into a single
event.

The series also breaks ABI in the patch [8/8] which is targetted to the
v21.11 release.

The dpdk-test-eventdev application has been updated with options to test
multiple vector sizes and timeouts.

[1]
As for performance improvement, with a ARM Cortex-A72 equivalent processer,
software event device (--vdev=event_sw0), single worker core, single stage
and using one service core for Rx adapter, Tx adapter, Scheduling.

Without this patchset applied:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
         --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
         --stlist=a --wlcores=20
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    5.071 mpps

With the patchset applied and Without event vectorization:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
         --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
         --stlist=a --wlcores=20
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    5.123 mpps

With event vectorization:
    ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
        --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
        --stlist=a --wlcores=20 --enable_vector --nb_eth_queues 1
        --vector_size 256
    Port[0] using Rx adapter[0] configured
    Port[0] using Tx adapter[0] Configured
    42.715 mpps

Having dedicated service cores for each Rx queues and tweaking the vector,
dequeue burst size would further improve performance.

API usage is shown below:

Configuration:

	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

	vector_pool = rte_event_vector_pool_create("vector_pool",
			nb_elem, 0, vector_size, socket_id);

	rte_event_eth_rx_adapter_create(id, event_id, &adptr_conf);
	rte_event_eth_rx_adapter_queue_add(id, eth_id, -1, &queue_conf);
	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
		vec_conf.vector_sz = vector_size;
		vec_conf.vector_timeout_ns = vector_tmo_nsec;
		vec_conf.vector_mp = vector_pool;
		rte_event_eth_rx_adapter_queue_event_vector_config(id,
				eth_id, -1, &vec_conf);
	}

Fastpath:

	num = rte_event_dequeue_burst(event_id, port_id, &ev, 1, 0);
	if (!num)
		continue;

	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
		switch (ev.event_type) {
		case RTE_EVENT_TYPE_ETHDEV_VECTOR:
		case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
			struct rte_mbuf **mbufs;

			mbufs = ev.vector_ev->mbufs;
			for (i = 0; i < ev.vector_ev->nb_elem; i++)
				//Process mbufs.
			break;
		case ...
		}
	}
	...

v10 Changes:
- Update Rx adapter documentation with flow identifier bitfield format. (Jay)

v9 Changes:
- Update Rx adapter documentation w.r.t SW event vectorizations. (Jay)
- Push partial vectors to event device on queue delete. (Jay)

v8 Changes:
- Fix incorrect shift for vector timeout interval.(Jay)
- Code reallocation.(Jay)

v7 Changes:
- More doxygen fixes.(Jay)
- Reduce code duplication in 4/8.(Jay)

v6 Changes:
- Make rte_errno sign consistant.(Jay)
- Gramatical and doxygen fixes. (Jay)

v5 Changes:
- Make `rte_event_vector_pool_create non-inline` to ease ABI stability.(Ray)
- Move `rte_event_eth_rx_adapter_queue_event_vector_config` and
  `rte_event_eth_rx_adapter_vector_limits_get` implementation to the patch
  where they are initially defined.(Ray)
- Multiple gramatical and style fixes.(Jerin)
- Add missing release notes.(Jerin)

v4 Changes:
- Fix missing event vector structure in event structure.(Jay)

v3 Changes:
- Fix unintended formatting changes.

v2 Changes:
- Multiple gramatical and style fixes.(Jerin)
- Add parameter to define vector size in power of 2. (Jerin)
- Redo patch series w/o breaking ABI till the last patch.(David)
- Add deprication notice to announce ABI break in 21.11.(David)
- Add vector limits validation to app/test-eventdev.

Pavan Nikhilesh (8):
  eventdev: introduce event vector capability
  eventdev: introduce event vector Rx capability
  eventdev: introduce event vector Tx capability
  eventdev: add Rx adapter event vector support
  eventdev: add Tx adapter event vector support
  app/eventdev: add event vector mode in pipeline test
  doc: announce event Rx adapter config changes
  eventdev: simplify Rx adapter event vector config

 app/test-eventdev/evt_common.h                |   4 +
 app/test-eventdev/evt_options.c               |  52 +++
 app/test-eventdev/evt_options.h               |   4 +
 app/test-eventdev/test_pipeline_atq.c         | 310 +++++++++++++++--
 app/test-eventdev/test_pipeline_common.c      | 105 +++++-
 app/test-eventdev/test_pipeline_common.h      |  18 +
 app/test-eventdev/test_pipeline_queue.c       | 320 +++++++++++++++--
 .../prog_guide/event_ethernet_rx_adapter.rst  |  57 +++
 .../prog_guide/event_ethernet_tx_adapter.rst  |  12 +
 doc/guides/prog_guide/eventdev.rst            |  36 +-
 doc/guides/rel_notes/deprecation.rst          |   9 +
 doc/guides/rel_notes/release_21_05.rst        |   8 +
 doc/guides/tools/testeventdev.rst             |  45 ++-
 lib/librte_eventdev/eventdev_pmd.h            |  31 +-
 .../rte_event_eth_rx_adapter.c                | 327 +++++++++++++++++-
 .../rte_event_eth_rx_adapter.h                |  78 +++++
 .../rte_event_eth_tx_adapter.c                |  66 +++-
 lib/librte_eventdev/rte_eventdev.c            |  53 ++-
 lib/librte_eventdev/rte_eventdev.h            | 114 +++++-
 lib/librte_eventdev/version.map               |   4 +
 20 files changed, 1562 insertions(+), 91 deletions(-)

--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v10 1/8] eventdev: introduce event vector capability
  2021-03-31  9:29                 ` [dpdk-dev] [PATCH v10 0/8] Introduce event vectorization pbhagavatula
@ 2021-03-31  9:29                   ` pbhagavatula
  2021-03-31  9:29                   ` [dpdk-dev] [PATCH v10 2/8] eventdev: introduce event vector Rx capability pbhagavatula
                                     ` (7 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-31  9:29 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce rte_event_vector datastructure which is capable of holding
multiple uintptr_t of the same flow thereby allowing applications
to vectorize their pipeline and reducing the complexity of pipelining
the events across multiple stages.
This approach also reduces the scheduling overhead on a event device.

Add a event vector mempool create handler to create mempools based on
the best mempool ops available on a given platform.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
Acked-by: Ray Kinsella <mdr@ashroe.eu>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
---
 doc/guides/prog_guide/eventdev.rst     | 36 ++++++++++-
 doc/guides/rel_notes/release_21_05.rst |  8 +++
 lib/librte_eventdev/rte_eventdev.c     | 42 +++++++++++++
 lib/librte_eventdev/rte_eventdev.h     | 82 +++++++++++++++++++++++++-
 lib/librte_eventdev/version.map        |  3 +
 5 files changed, 168 insertions(+), 3 deletions(-)

diff --git a/doc/guides/prog_guide/eventdev.rst b/doc/guides/prog_guide/eventdev.rst
index ccde086f6..fda9c3743 100644
--- a/doc/guides/prog_guide/eventdev.rst
+++ b/doc/guides/prog_guide/eventdev.rst
@@ -63,13 +63,45 @@ the actual event being scheduled is. The payload is a union of the following:
 * ``uint64_t u64``
 * ``void *event_ptr``
 * ``struct rte_mbuf *mbuf``
+* ``struct rte_event_vector *vec``
 
-These three items in a union occupy the same 64 bits at the end of the rte_event
+These four items in a union occupy the same 64 bits at the end of the rte_event
 structure. The application can utilize the 64 bits directly by accessing the
-u64 variable, while the event_ptr and mbuf are provided as convenience
+u64 variable, while the event_ptr, mbuf, vec are provided as a convenience
 variables.  For example the mbuf pointer in the union can used to schedule a
 DPDK packet.
 
+Event Vector
+~~~~~~~~~~~~
+
+The rte_event_vector struct contains a vector of elements defined by the event
+type specified in the ``rte_event``. The event_vector structure contains the
+following data:
+
+* ``nb_elem`` - The number of elements held within the vector.
+
+Similar to ``rte_event`` the payload of event vector is also a union, allowing
+flexibility in what the actual vector is.
+
+* ``struct rte_mbuf *mbufs[0]`` - An array of mbufs.
+* ``void *ptrs[0]`` - An array of pointers.
+* ``uint64_t *u64s[0]`` - An array of uint64_t elements.
+
+The size of the event vector is related to the total number of elements it is
+configured to hold, this is achieved by making `rte_event_vector` a variable
+length structure.
+A helper function is provided to create a mempool that holds event vector, which
+takes name of the pool, total number of required ``rte_event_vector``,
+cache size, number of elements in each ``rte_event_vector`` and socket id.
+
+.. code-block:: c
+
+        rte_event_vector_pool_create("vector_pool", nb_event_vectors, cache_sz,
+                                     nb_elements_per_vector, socket_id);
+
+The function ``rte_event_vector_pool_create`` creates mempool with the best
+platform mempool ops.
+
 Queues
 ~~~~~~
 
diff --git a/doc/guides/rel_notes/release_21_05.rst b/doc/guides/rel_notes/release_21_05.rst
index e2b0886a9..4cbf2769f 100644
--- a/doc/guides/rel_notes/release_21_05.rst
+++ b/doc/guides/rel_notes/release_21_05.rst
@@ -106,6 +106,14 @@ New Features
   * Added support for periodic timer mode in eventdev timer adapter.
   * Added support for periodic timer mode in octeontx2 event device driver.
 
+* **Add Event device vector capability.**
+
+  * Added ``rte_event_vector`` data structure which is capable of holding
+    multiple ``uintptr_t`` of the same flow thereby allowing applications
+    to vectorize their pipelines and also reduce the complexity of pipelining
+    the events across multiple stages.
+  * This also reduces the scheduling overhead on a event device.
+
 
 Removed Items
 -------------
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index b57363f80..be0499c52 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -1266,6 +1266,48 @@ int rte_event_dev_selftest(uint8_t dev_id)
 	return -ENOTSUP;
 }
 
+struct rte_mempool *
+rte_event_vector_pool_create(const char *name, unsigned int n,
+			     unsigned int cache_size, uint16_t nb_elem,
+			     int socket_id)
+{
+	const char *mp_ops_name;
+	struct rte_mempool *mp;
+	unsigned int elt_sz;
+	int ret;
+
+	if (!nb_elem) {
+		RTE_LOG(ERR, EVENTDEV,
+			"Invalid number of elements=%d requested\n", nb_elem);
+		rte_errno = EINVAL;
+		return NULL;
+	}
+
+	elt_sz =
+		sizeof(struct rte_event_vector) + (nb_elem * sizeof(uintptr_t));
+	mp = rte_mempool_create_empty(name, n, elt_sz, cache_size, 0, socket_id,
+				      0);
+	if (mp == NULL)
+		return NULL;
+
+	mp_ops_name = rte_mbuf_best_mempool_ops();
+	ret = rte_mempool_set_ops_byname(mp, mp_ops_name, NULL);
+	if (ret != 0) {
+		RTE_LOG(ERR, EVENTDEV, "error setting mempool handler\n");
+		goto err;
+	}
+
+	ret = rte_mempool_populate_default(mp);
+	if (ret < 0)
+		goto err;
+
+	return mp;
+err:
+	rte_mempool_free(mp);
+	rte_errno = -ret;
+	return NULL;
+}
+
 int
 rte_event_dev_start(uint8_t dev_id)
 {
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index 9fc39e9ca..dee468ed0 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -212,8 +212,10 @@ extern "C" {
 
 #include <rte_common.h>
 #include <rte_config.h>
-#include <rte_memory.h>
 #include <rte_errno.h>
+#include <rte_mbuf_pool_ops.h>
+#include <rte_memory.h>
+#include <rte_mempool.h>
 
 #include "rte_eventdev_trace_fp.h"
 
@@ -913,6 +915,31 @@ rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
 int
 rte_event_dev_close(uint8_t dev_id);
 
+/**
+ * Event vector structure.
+ */
+struct rte_event_vector {
+	uint64_t nb_elem : 16;
+	/**< Number of elements in this event vector. */
+	uint64_t rsvd : 48;
+	/**< Reserved for future use */
+	uint64_t impl_opaque;
+	/**< Implementation specific opaque value.
+	 * An implementation may use this field to hold implementation specific
+	 * value to share between dequeue and enqueue operation.
+	 * The application should not modify this field.
+	 */
+	union {
+		struct rte_mbuf *mbufs[0];
+		void *ptrs[0];
+		uint64_t *u64s[0];
+	} __rte_aligned(16);
+	/**< Start of the vector array union. Depending upon the event type the
+	 * vector array can be an array of mbufs or pointers or opaque u64
+	 * values.
+	 */
+};
+
 /* Scheduler type definitions */
 #define RTE_SCHED_TYPE_ORDERED          0
 /**< Ordered scheduling
@@ -986,6 +1013,21 @@ rte_event_dev_close(uint8_t dev_id);
  */
 #define RTE_EVENT_TYPE_ETH_RX_ADAPTER   0x4
 /**< The event generated from event eth Rx adapter */
+#define RTE_EVENT_TYPE_VECTOR           0x8
+/**< Indicates that event is a vector.
+ * All vector event types should be a logical OR of EVENT_TYPE_VECTOR.
+ * This simplifies the pipeline design as one can split processing the events
+ * between vector events and normal event across event types.
+ * Example:
+ *	if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
+ *		// Classify and handle vector event.
+ *	} else {
+ *		// Classify and handle event.
+ *	}
+ */
+#define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
+/**< The event vector generated from cpu for pipelining. */
+
 #define RTE_EVENT_TYPE_MAX              0x10
 /**< Maximum number of event types */
 
@@ -1108,6 +1150,8 @@ struct rte_event {
 		/**< Opaque event pointer */
 		struct rte_mbuf *mbuf;
 		/**< mbuf pointer if dequeued event is associated with mbuf */
+		struct rte_event_vector *vec;
+		/**< Event vector pointer. */
 	};
 };
 
@@ -2026,6 +2070,42 @@ rte_event_dev_xstats_reset(uint8_t dev_id,
  */
 int rte_event_dev_selftest(uint8_t dev_id);
 
+/**
+ * Get the memory required per event vector based on the number of elements per
+ * vector.
+ * This should be used to create the mempool that holds the event vectors.
+ *
+ * @param name
+ *   The name of the vector pool.
+ * @param n
+ *   The number of elements in the mbuf pool.
+ * @param cache_size
+ *   Size of the per-core object cache. See rte_mempool_create() for
+ *   details.
+ * @param nb_elem
+ *   The number of elements that a single event vector should be able to hold.
+ * @param socket_id
+ *   The socket identifier where the memory should be allocated. The
+ *   value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
+ *   reserved zone
+ *
+ * @return
+ *   The pointer to the newly allocated mempool, on success. NULL on error
+ *   with rte_errno set appropriately. Possible rte_errno values include:
+ *    - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
+ *    - E_RTE_SECONDARY - function was called from a secondary process instance
+ *    - EINVAL - cache size provided is too large, or priv_size is not aligned.
+ *    - ENOSPC - the maximum number of memzones has already been allocated
+ *    - EEXIST - a memzone with the same name already exists
+ *    - ENOMEM - no appropriate memory area found in which to create memzone
+ *    - ENAMETOOLONG - mempool name requested is too long.
+ */
+__rte_experimental
+struct rte_mempool *
+rte_event_vector_pool_create(const char *name, unsigned int n,
+			     unsigned int cache_size, uint16_t nb_elem,
+			     int socket_id);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index 3e5c09cfd..a070ef56e 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -138,6 +138,9 @@ EXPERIMENTAL {
 	__rte_eventdev_trace_port_setup;
 	# added in 20.11
 	rte_event_pmd_pci_probe_named;
+
+	#added in 21.05
+	rte_event_vector_pool_create;
 };
 
 INTERNAL {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v10 2/8] eventdev: introduce event vector Rx capability
  2021-03-31  9:29                 ` [dpdk-dev] [PATCH v10 0/8] Introduce event vectorization pbhagavatula
  2021-03-31  9:29                   ` [dpdk-dev] [PATCH v10 1/8] eventdev: introduce event vector capability pbhagavatula
@ 2021-03-31  9:29                   ` pbhagavatula
  2021-03-31  9:29                   ` [dpdk-dev] [PATCH v10 3/8] eventdev: introduce event vector Tx capability pbhagavatula
                                     ` (6 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-31  9:29 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce event ethernet Rx adapter event vector capability.

If an event eth Rx adapter has the capability of
RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR then a given Rx queue
can be configured to enable event vectorization by passing the
flag RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR to
rte_event_eth_rx_adapter_queue_conf::rx_queue_flags while configuring
Rx adapter through rte_event_eth_rx_adapter_queue_add().

The max vector size, vector timeout define the vector size and
mempool used for allocating vector event are configured through
rte_event_eth_rx_adapter_queue_add. The element size of the element
in the vector pool should be equal to
    sizeof(struct rte_event_vector) + (vector_sz * sizeof(uintptr_t))

Application can use `rte_event_vector_pool_create` to create the
vector mempool used for
rte_event_eth_rx_adapter_queue_conf::vector_mp.

The Rx adapter would be responsible for vectorizing the mbufs
based on the flow, the vector limits configured by the application
and add the vector event of mbufs to the event queue set via
rte_event_eth_rx_adapter_queue_conf::ev::queue_id.
It should also mark rte_event_vector::union_valid and fill
rte_event_vector::port, rte_event_vector::queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
Acked-by: Ray Kinsella <mdr@ashroe.eu>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
---
 .../prog_guide/event_ethernet_rx_adapter.rst  |  38 ++++++
 lib/librte_eventdev/eventdev_pmd.h            |  53 ++++++++
 .../rte_event_eth_rx_adapter.c                | 114 ++++++++++++++++++
 .../rte_event_eth_rx_adapter.h                | 105 ++++++++++++++++
 lib/librte_eventdev/rte_eventdev.h            |  30 ++++-
 lib/librte_eventdev/version.map               |   2 +
 6 files changed, 340 insertions(+), 2 deletions(-)

diff --git a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
index cb44ce0e4..5eefef355 100644
--- a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
@@ -186,3 +186,41 @@ the event buffer fill level is low. The
 ``rte_event_eth_rx_adapter_cb_register()`` function allow the application
 to register a callback that selects which packets to enqueue to the event
 device.
+
+Rx event vectorization
+~~~~~~~~~~~~~~~~~~~~~~
+
+The event devices, ethernet device pairs which support the capability
+``RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR`` can aggregate packets based on
+flow characteristics and generate a ``rte_event`` containing ``rte_event_vector``
+whose event type is either ``RTE_EVENT_TYPE_ETHDEV_VECTOR`` or
+``RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR``.
+The aggregation size and timeout are configurable at a queue level and the
+maximum, minimum vector sizes and timeouts vary based on the device capability
+and can be queried using ``rte_event_eth_rx_adapter_vector_limits_get``.
+The Rx adapter additionally might include useful data such as ethernet device
+port and queue identifier in the ``rte_event_vector::port`` and
+``rte_event_vector::queue`` and mark ``rte_event_vector::attr_valid`` as true.
+
+A loop processing ``rte_event_vector`` containing mbufs is shown below.
+
+.. code-block:: c
+
+        event = rte_event_dequeue_burst(event_dev, event_port, &event,
+                                        1, 0);
+        if (!event)
+                continue;
+
+        switch (ev.event_type) {
+        case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
+        case RTE_EVENT_TYPE_ETHDEV_VECTOR:
+                struct rte_mbufs **mbufs;
+
+                mbufs = (struct rte_mbufs **)ev[i].vec->mbufs;
+                for (i = 0; i < ev.vec->nb_elem; i++) {
+                        /* Process each mbuf. */
+                }
+        break;
+        case ...
+        ...
+        }
diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 7eb9a7739..9297f1433 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -645,6 +645,53 @@ typedef int (*eventdev_eth_rx_adapter_stats_reset)
  */
 typedef int (*eventdev_selftest)(void);
 
+struct rte_event_eth_rx_adapter_vector_limits;
+/**
+ * Get event vector limits for a given event, ethernet device pair.
+ *
+ * @param dev
+ *   Event device pointer
+ *
+ * @param eth_dev
+ *   Ethernet device pointer
+ *
+ * @param[out] limits
+ *   Pointer to the limits structure to be filled.
+ *
+ * @return
+ *   - 0: Success.
+ *   - <0: Error code returned by the driver function.
+ */
+typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
+	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
+	struct rte_event_eth_rx_adapter_vector_limits *limits);
+
+struct rte_event_eth_rx_adapter_event_vector_config;
+/**
+ * Enable event vector on an given Rx queue of a ethernet devices belonging to
+ * the Rx adapter.
+ *
+ * @param dev
+ *   Event device pointer
+ *
+ * @param eth_dev
+ *   Ethernet device pointer
+ *
+ * @param rx_queue_id
+ *   The Rx queue identifier
+ *
+ * @param config
+ *   Pointer to the event vector configuration structure.
+ *
+ * @return
+ *   - 0: Success.
+ *   - <0: Error code returned by the driver function.
+ */
+typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
+	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
+	int32_t rx_queue_id,
+	const struct rte_event_eth_rx_adapter_event_vector_config *config);
+
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;
 
@@ -1067,6 +1114,12 @@ struct rte_eventdev_ops {
 	/**< Get ethernet Rx stats */
 	eventdev_eth_rx_adapter_stats_reset eth_rx_adapter_stats_reset;
 	/**< Reset ethernet Rx stats */
+	eventdev_eth_rx_adapter_vector_limits_get_t
+		eth_rx_adapter_vector_limits_get;
+	/**< Get event vector limits for the Rx adapter */
+	eventdev_eth_rx_adapter_event_vector_config_t
+		eth_rx_adapter_event_vector_config;
+	/**< Configure Rx adapter with event vector */
 
 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index d8c635e99..ac8ba5bf0 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -2263,6 +2263,120 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }
 
+int
+rte_event_eth_rx_adapter_queue_event_vector_config(
+	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
+	struct rte_event_eth_rx_adapter_event_vector_config *config)
+{
+	struct rte_event_eth_rx_adapter_vector_limits limits;
+	struct rte_event_eth_rx_adapter *rx_adapter;
+	struct rte_eventdev *dev;
+	uint32_t cap;
+	int ret;
+
+	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
+
+	rx_adapter = rxa_id_to_adapter(id);
+	if ((rx_adapter == NULL) || (config == NULL))
+		return -EINVAL;
+
+	dev = &rte_eventdevs[rx_adapter->eventdev_id];
+	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
+						eth_dev_id, &cap);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
+				 "eth port %" PRIu16,
+				 id, eth_dev_id);
+		return ret;
+	}
+
+	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
+		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
+	ret = rte_event_eth_rx_adapter_vector_limits_get(
+		rx_adapter->eventdev_id, eth_dev_id, &limits);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
+				 "eth port %" PRIu16,
+				 rx_adapter->eventdev_id, eth_dev_id);
+		return ret;
+	}
+
+	if (config->vector_sz < limits.min_sz ||
+	    config->vector_sz > limits.max_sz ||
+	    config->vector_timeout_ns < limits.min_timeout_ns ||
+	    config->vector_timeout_ns > limits.max_timeout_ns ||
+	    config->vector_mp == NULL) {
+		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+	if (config->vector_mp->elt_size <
+	    (sizeof(struct rte_event_vector) +
+	     (sizeof(uintptr_t) * config->vector_sz))) {
+		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
+	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
+		RTE_FUNC_PTR_OR_ERR_RET(
+			*dev->dev_ops->eth_rx_adapter_event_vector_config,
+			-ENOTSUP);
+		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
+			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
+	} else {
+		ret = -ENOTSUP;
+	}
+
+	return ret;
+}
+
+int
+rte_event_eth_rx_adapter_vector_limits_get(
+	uint8_t dev_id, uint16_t eth_port_id,
+	struct rte_event_eth_rx_adapter_vector_limits *limits)
+{
+	struct rte_eventdev *dev;
+	uint32_t cap;
+	int ret;
+
+	RTE_EVENTDEV_VALID_DEVID_OR_ERR_RET(dev_id, -EINVAL);
+	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_port_id, -EINVAL);
+
+	if (limits == NULL)
+		return -EINVAL;
+
+	dev = &rte_eventdevs[dev_id];
+
+	ret = rte_event_eth_rx_adapter_caps_get(dev_id, eth_port_id, &cap);
+	if (ret) {
+		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
+				 "eth port %" PRIu16,
+				 dev_id, eth_port_id);
+		return ret;
+	}
+
+	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
+		RTE_FUNC_PTR_OR_ERR_RET(
+			*dev->dev_ops->eth_rx_adapter_vector_limits_get,
+			-ENOTSUP);
+		ret = dev->dev_ops->eth_rx_adapter_vector_limits_get(
+			dev, &rte_eth_devices[eth_port_id], limits);
+	} else {
+		ret = -ENOTSUP;
+	}
+
+	return ret;
+}
+
 int
 rte_event_eth_rx_adapter_start(uint8_t id)
 {
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
index 21bb1e54c..7407cde00 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
@@ -92,6 +92,10 @@ extern "C" {
 /**< This flag indicates the flow identifier is valid
  * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
  */
+#define RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR	0x2
+/**< This flag indicates that mbufs arriving on the queue need to be vectorized
+ * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+ */
 
 /**
  * Adapter configuration structure that the adapter configuration callback
@@ -169,6 +173,36 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 */
 };
 
+struct rte_event_eth_rx_adapter_event_vector_config {
+	uint16_t vector_sz;
+	/**<
+	 * Indicates the maximum number for mbufs to combine and form a vector.
+	 * Should be within
+	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_sz
+	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_sz
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+	 */
+	uint64_t vector_timeout_ns;
+	/**<
+	 * Indicates the maximum number of nanoseconds to wait for receiving
+	 * mbufs. Should be within vectorization limits of the
+	 * adapter
+	 * @see rte_event_eth_rx_adapter_vector_limits::min_vector_ns
+	 * @see rte_event_eth_rx_adapter_vector_limits::max_vector_ns
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
+	 */
+	struct rte_mempool *vector_mp;
+	/**<
+	 * Indicates the mempool that should be used for allocating
+	 * rte_event_vector container.
+	 * Should be created by using `rte_event_vector_pool_create`.
+	 * Valid when RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR flag is set in
+	 * @see rte_event_eth_rx_adapter_queue_conf::rx_queue_flags.
+	 */
+};
+
 /**
  * A structure used to retrieve statistics for an eth rx adapter instance.
  */
@@ -199,6 +233,32 @@ struct rte_event_eth_rx_adapter_stats {
 	/**< Received packet count for interrupt mode Rx queues */
 };
 
+/**
+ * A structure used to retrieve eth rx adapter vector limits.
+ */
+struct rte_event_eth_rx_adapter_vector_limits {
+	uint16_t min_sz;
+	/**< Minimum vector limit configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_sz
+	 */
+	uint16_t max_sz;
+	/**< Maximum vector limit configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_sz
+	 */
+	uint8_t log2_sz;
+	/**< True if the size configured should be in log2.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_sz
+	 */
+	uint64_t min_timeout_ns;
+	/**< Minimum vector timeout configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_timeout_ns
+	 */
+	uint64_t max_timeout_ns;
+	/**< Maximum vector timeout configurable.
+	 * @see rte_event_eth_rx_adapter_event_vector_config::vector_timeout_ns
+	 */
+};
+
 /**
  *
  * Callback function invoked by the SW adapter before it continues
@@ -467,6 +527,51 @@ int rte_event_eth_rx_adapter_cb_register(uint8_t id, uint16_t eth_dev_id,
 					 rte_event_eth_rx_adapter_cb_fn cb_fn,
 					 void *cb_arg);
 
+/**
+ * Retrieve vector limits for a given event dev and eth dev pair.
+ * @see rte_event_eth_rx_adapter_vector_limits
+ *
+ * @param dev_id
+ *  Event device identifier.
+ * @param eth_port_id
+ *  Port identifier of the ethernet device.
+ * @param [out] limits
+ *  A pointer to rte_event_eth_rx_adapter_vector_limits structure that has to
+ * be filled.
+ *
+ * @return
+ *  - 0: Success.
+ *  - <0: Error code on failure.
+ */
+__rte_experimental
+int rte_event_eth_rx_adapter_vector_limits_get(
+	uint8_t dev_id, uint16_t eth_port_id,
+	struct rte_event_eth_rx_adapter_vector_limits *limits);
+
+/**
+ * Configure event vectorization for a given ethernet device queue, that has
+ * been added to a event eth Rx adapter.
+ *
+ * @param id
+ *  The identifier of the ethernet Rx event adapter.
+ *
+ * @param eth_dev_id
+ *  The identifier of the ethernet device.
+ *
+ * @param rx_queue_id
+ *  Ethernet device receive queue index.
+ *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
+ *  are configured with event vectorization.
+ *
+ * @return
+ *  - 0: Success, Receive queue configured correctly.
+ *  - <0: Error code on failure.
+ */
+__rte_experimental
+int rte_event_eth_rx_adapter_queue_event_vector_config(
+	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
+	struct rte_event_eth_rx_adapter_event_vector_config *config);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index dee468ed0..936c97cc2 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -919,10 +919,28 @@ rte_event_dev_close(uint8_t dev_id);
  * Event vector structure.
  */
 struct rte_event_vector {
-	uint64_t nb_elem : 16;
+	uint16_t nb_elem;
 	/**< Number of elements in this event vector. */
-	uint64_t rsvd : 48;
+	uint16_t rsvd : 15;
 	/**< Reserved for future use */
+	uint16_t attr_valid : 1;
+	/**< Indicates that the below union attributes have valid information.
+	 */
+	union {
+		/* Used by Rx adapter.
+		 * Indicates that all the elements in this vector belong to the
+		 * same port and queue pair when originating from Rx adapter,
+		 * valid only when event type is ETHDEV_VECTOR or
+		 * ETH_RX_ADAPTER_VECTOR.
+		 */
+		struct {
+			uint16_t port;
+			/* Ethernet device port id. */
+			uint16_t queue;
+			/* Ethernet device queue id. */
+		};
+	};
+	/**< Union to hold common attributes of the vector array. */
 	uint64_t impl_opaque;
 	/**< Implementation specific opaque value.
 	 * An implementation may use this field to hold implementation specific
@@ -1025,8 +1043,14 @@ struct rte_event_vector {
  *		// Classify and handle event.
  *	}
  */
+#define RTE_EVENT_TYPE_ETHDEV_VECTOR                                           \
+	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETHDEV)
+/**< The event vector generated from ethdev subsystem */
 #define RTE_EVENT_TYPE_CPU_VECTOR (RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_CPU)
 /**< The event vector generated from cpu for pipelining. */
+#define RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR                                   \
+	(RTE_EVENT_TYPE_VECTOR | RTE_EVENT_TYPE_ETH_RX_ADAPTER)
+/**< The event vector generated from eth Rx adapter. */
 
 #define RTE_EVENT_TYPE_MAX              0x10
 /**< Maximum number of event types */
@@ -1171,6 +1195,8 @@ struct rte_event {
  * @see struct rte_event_eth_rx_adapter_queue_conf::ev
  * @see struct rte_event_eth_rx_adapter_queue_conf::rx_queue_flags
  */
+#define RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR	0x8
+/**< Adapter supports event vectorization per ethdev. */
 
 /**
  * Retrieve the event device's ethdev Rx adapter capabilities for the
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index a070ef56e..902df0ae3 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -141,6 +141,8 @@ EXPERIMENTAL {
 
 	#added in 21.05
 	rte_event_vector_pool_create;
+	rte_event_eth_rx_adapter_vector_limits_get;
+	rte_event_eth_rx_adapter_queue_event_vector_config;
 };
 
 INTERNAL {
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v10 3/8] eventdev: introduce event vector Tx capability
  2021-03-31  9:29                 ` [dpdk-dev] [PATCH v10 0/8] Introduce event vectorization pbhagavatula
  2021-03-31  9:29                   ` [dpdk-dev] [PATCH v10 1/8] eventdev: introduce event vector capability pbhagavatula
  2021-03-31  9:29                   ` [dpdk-dev] [PATCH v10 2/8] eventdev: introduce event vector Rx capability pbhagavatula
@ 2021-03-31  9:29                   ` pbhagavatula
  2021-03-31  9:29                   ` [dpdk-dev] [PATCH v10 4/8] eventdev: add Rx adapter event vector support pbhagavatula
                                     ` (5 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-31  9:29 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Introduce event vector transmit capability for event eth
tx adapter.

The capability indicates that the Tx adapter is capable of
transmitting event vectors.
When rte_event_vector::union_valid is set, the Tx adapter should
transmit all the packets to the rte_event_vector::port using the
rte_event_vector::queue.
If rte_event_vector::union_valid is not set then the Tx adapter
should peek into each mbuf to get the destination port and queue
pair.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
---
 doc/guides/prog_guide/event_ethernet_tx_adapter.rst | 12 ++++++++++++
 lib/librte_eventdev/rte_eventdev.h                  |  8 +++++++-
 2 files changed, 19 insertions(+), 1 deletion(-)

diff --git a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
index a8c13e136..f80d22621 100644
--- a/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_tx_adapter.rst
@@ -164,3 +164,15 @@ The  ``rte_event_eth_tx_adapter_stats_get()`` function reports counters defined
 in struct ``rte_event_eth_tx_adapter_stats``. The counter values are the sum of
 the counts from the eventdev PMD callback if the callback is supported, and
 the counts maintained by the service function, if one exists.
+
+Tx event vectorization
+~~~~~~~~~~~~~~~~~~~~~~
+
+The event device, ethernet device pairs which support the capability
+``RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR`` can process event vector of mbufs.
+Additionally, application can provide a hint to the Tx adapter that all the
+mbufs are destined to the same ethernet port and queue by setting the bit
+``rte_event_vector::attr_valid`` and filling `rte_event_vector::port`` and
+``rte_event_vector::queue``.
+If ``rte_event_vector::attr_valid`` is not set then the Tx adapter should peek
+into each mbuf and transmit them to the requested ethernet port and queue pair.
diff --git a/lib/librte_eventdev/rte_eventdev.h b/lib/librte_eventdev/rte_eventdev.h
index 936c97cc2..5f1f544cc 100644
--- a/lib/librte_eventdev/rte_eventdev.h
+++ b/lib/librte_eventdev/rte_eventdev.h
@@ -927,11 +927,13 @@ struct rte_event_vector {
 	/**< Indicates that the below union attributes have valid information.
 	 */
 	union {
-		/* Used by Rx adapter.
+		/* Used by Rx/Tx adapter.
 		 * Indicates that all the elements in this vector belong to the
 		 * same port and queue pair when originating from Rx adapter,
 		 * valid only when event type is ETHDEV_VECTOR or
 		 * ETH_RX_ADAPTER_VECTOR.
+		 * Can also be used to indicate the Tx adapter the destination
+		 * port and queue of the mbufs in the vector
 		 */
 		struct {
 			uint16_t port;
@@ -1296,6 +1298,10 @@ rte_event_crypto_adapter_caps_get(uint8_t dev_id, uint8_t cdev_id,
 #define RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT	0x1
 /**< This flag is sent when the PMD supports a packet transmit callback
  */
+#define RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR	0x2
+/**< Indicates that the Tx adapter is capable of handling event vector of
+ * mbufs.
+ */
 
 /**
  * Retrieve the event device's eth Tx adapter capabilities
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v10 4/8] eventdev: add Rx adapter event vector support
  2021-03-31  9:29                 ` [dpdk-dev] [PATCH v10 0/8] Introduce event vectorization pbhagavatula
                                     ` (2 preceding siblings ...)
  2021-03-31  9:29                   ` [dpdk-dev] [PATCH v10 3/8] eventdev: introduce event vector Tx capability pbhagavatula
@ 2021-03-31  9:29                   ` pbhagavatula
  2021-03-31  9:29                   ` [dpdk-dev] [PATCH v10 5/8] eventdev: add Tx " pbhagavatula
                                     ` (4 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-31  9:29 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support for event eth Rx adapter, the implementation
creates vector flows based on port and queue identifier of the received
mbufs.
The flow id for SW Rx event vectorization will use 12-bits of queue
identifier and 8-bits port identifier when custom flow id is not set
for simplicity.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>.
---
 .../prog_guide/event_ethernet_rx_adapter.rst  |  19 ++
 lib/librte_eventdev/eventdev_pmd.h            |   7 +-
 .../rte_event_eth_rx_adapter.c                | 278 ++++++++++++++++--
 lib/librte_eventdev/rte_eventdev.c            |   6 +-
 4 files changed, 286 insertions(+), 24 deletions(-)

diff --git a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
index 5eefef355..c01e5a966 100644
--- a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
@@ -224,3 +224,22 @@ A loop processing ``rte_event_vector`` containing mbufs is shown below.
         case ...
         ...
         }
+
+Rx event vectorization for SW Rx adapter
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+For SW based event vectorization, i.e., when the
+``RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT`` is not set in the adapter's
+capabilities flags for a particular ethernet device, the service function
+creates a single event vector flow for all the mbufs arriving on the given
+Rx queue.
+The 20-bit event flow identifier is set to 12-bits of Rx queue identifier
+and 8-bits of ethernet device identifier.
+Flow identifier is formatted as follows:
+
+.. code-block:: console
+
+    19      12,11            0
+    +---------+--------------+
+    | port_id |   queue_id   |
+    +---------+--------------+
diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 9297f1433..0f724ac85 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -69,9 +69,10 @@ extern "C" {
 	} \
 } while (0)
 
-#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP \
-		((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) | \
-			(RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ))
+#define RTE_EVENT_ETH_RX_ADAPTER_SW_CAP                                        \
+	((RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID) |                     \
+	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) |                         \
+	 (RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR))
 
 #define RTE_EVENT_CRYPTO_ADAPTER_SW_CAP \
 		RTE_EVENT_CRYPTO_ADAPTER_CAP_SESSION_PRIVATE_DATA
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index ac8ba5bf0..e273b3acf 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -26,6 +26,10 @@
 #define BATCH_SIZE		32
 #define BLOCK_CNT_THRESHOLD	10
 #define ETH_EVENT_BUFFER_SIZE	(4*BATCH_SIZE)
+#define MAX_VECTOR_SIZE		1024
+#define MIN_VECTOR_SIZE		4
+#define MAX_VECTOR_NS		1E9
+#define MIN_VECTOR_NS		1E5
 
 #define ETH_RX_ADAPTER_SERVICE_NAME_LEN	32
 #define ETH_RX_ADAPTER_MEM_NAME_LEN	32
@@ -59,6 +63,20 @@ struct eth_rx_poll_entry {
 	uint16_t eth_rx_qid;
 };
 
+struct eth_rx_vector_data {
+	TAILQ_ENTRY(eth_rx_vector_data) next;
+	uint16_t port;
+	uint16_t queue;
+	uint16_t max_vector_count;
+	uint64_t event;
+	uint64_t ts;
+	uint64_t vector_timeout_ticks;
+	struct rte_mempool *vector_pool;
+	struct rte_event_vector *vector_ev;
+} __rte_cache_aligned;
+
+TAILQ_HEAD(eth_rx_vector_data_list, eth_rx_vector_data);
+
 /* Instance per adapter */
 struct rte_eth_event_enqueue_buffer {
 	/* Count of events in this buffer */
@@ -92,6 +110,14 @@ struct rte_event_eth_rx_adapter {
 	uint32_t wrr_pos;
 	/* Event burst buffer */
 	struct rte_eth_event_enqueue_buffer event_enqueue_buffer;
+	/* Vector enable flag */
+	uint8_t ena_vector;
+	/* Timestamp of previous vector expiry list traversal */
+	uint64_t prev_expiry_ts;
+	/* Minimum ticks to wait before traversing expiry list */
+	uint64_t vector_tmo_ticks;
+	/* vector list */
+	struct eth_rx_vector_data_list vector_list;
 	/* Per adapter stats */
 	struct rte_event_eth_rx_adapter_stats stats;
 	/* Block count, counts up to BLOCK_CNT_THRESHOLD */
@@ -198,9 +224,11 @@ struct eth_device_info {
 struct eth_rx_queue_info {
 	int queue_enabled;	/* True if added */
 	int intr_enabled;
+	uint8_t ena_vector;
 	uint16_t wt;		/* Polling weight */
 	uint32_t flow_id_mask;	/* Set to ~0 if app provides flow id else 0 */
 	uint64_t event;
+	struct eth_rx_vector_data vector_data;
 };
 
 static struct rte_event_eth_rx_adapter **event_eth_rx_adapter;
@@ -722,6 +750,9 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
 	    &rx_adapter->event_enqueue_buffer;
 	struct rte_event_eth_rx_adapter_stats *stats = &rx_adapter->stats;
 
+	if (!buf->count)
+		return 0;
+
 	uint16_t n = rte_event_enqueue_new_burst(rx_adapter->eventdev_id,
 					rx_adapter->event_port_id,
 					buf->events,
@@ -742,6 +773,77 @@ rxa_flush_event_buffer(struct rte_event_eth_rx_adapter *rx_adapter)
 	return n;
 }
 
+static inline void
+rxa_init_vector(struct rte_event_eth_rx_adapter *rx_adapter,
+		struct eth_rx_vector_data *vec)
+{
+	vec->vector_ev->nb_elem = 0;
+	vec->vector_ev->port = vec->port;
+	vec->vector_ev->queue = vec->queue;
+	vec->vector_ev->attr_valid = true;
+	TAILQ_INSERT_TAIL(&rx_adapter->vector_list, vec, next);
+}
+
+static inline uint16_t
+rxa_create_event_vector(struct rte_event_eth_rx_adapter *rx_adapter,
+			struct eth_rx_queue_info *queue_info,
+			struct rte_eth_event_enqueue_buffer *buf,
+			struct rte_mbuf **mbufs, uint16_t num)
+{
+	struct rte_event *ev = &buf->events[buf->count];
+	struct eth_rx_vector_data *vec;
+	uint16_t filled, space, sz;
+
+	filled = 0;
+	vec = &queue_info->vector_data;
+
+	if (vec->vector_ev == NULL) {
+		if (rte_mempool_get(vec->vector_pool,
+				    (void **)&vec->vector_ev) < 0) {
+			rte_pktmbuf_free_bulk(mbufs, num);
+			return 0;
+		}
+		rxa_init_vector(rx_adapter, vec);
+	}
+	while (num) {
+		if (vec->vector_ev->nb_elem == vec->max_vector_count) {
+			/* Event ready. */
+			ev->event = vec->event;
+			ev->vec = vec->vector_ev;
+			ev++;
+			filled++;
+			vec->vector_ev = NULL;
+			TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+			if (rte_mempool_get(vec->vector_pool,
+					    (void **)&vec->vector_ev) < 0) {
+				rte_pktmbuf_free_bulk(mbufs, num);
+				return 0;
+			}
+			rxa_init_vector(rx_adapter, vec);
+		}
+
+		space = vec->max_vector_count - vec->vector_ev->nb_elem;
+		sz = num > space ? space : num;
+		memcpy(vec->vector_ev->mbufs + vec->vector_ev->nb_elem, mbufs,
+		       sizeof(void *) * sz);
+		vec->vector_ev->nb_elem += sz;
+		num -= sz;
+		mbufs += sz;
+		vec->ts = rte_rdtsc();
+	}
+
+	if (vec->vector_ev->nb_elem == vec->max_vector_count) {
+		ev->event = vec->event;
+		ev->vec = vec->vector_ev;
+		ev++;
+		filled++;
+		vec->vector_ev = NULL;
+		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+	}
+
+	return filled;
+}
+
 static inline void
 rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
@@ -766,29 +868,33 @@ rxa_buffer_mbufs(struct rte_event_eth_rx_adapter *rx_adapter,
 	uint16_t nb_cb;
 	uint16_t dropped;
 
-	/* 0xffff ffff if PKT_RX_RSS_HASH is set, otherwise 0 */
-	rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
-	do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
-
-	for (i = 0; i < num; i++) {
-		m = mbufs[i];
-
-		rss = do_rss ?
-			rxa_do_softrss(m, rx_adapter->rss_key_be) :
-			m->hash.rss;
-		ev->event = event;
-		ev->flow_id = (rss & ~flow_id_mask) |
-				(ev->flow_id & flow_id_mask);
-		ev->mbuf = m;
-		ev++;
+	if (!eth_rx_queue_info->ena_vector) {
+		/* 0xffff ffff if PKT_RX_RSS_HASH is set, otherwise 0 */
+		rss_mask = ~(((m->ol_flags & PKT_RX_RSS_HASH) != 0) - 1);
+		do_rss = !rss_mask && !eth_rx_queue_info->flow_id_mask;
+		for (i = 0; i < num; i++) {
+			m = mbufs[i];
+
+			rss = do_rss ? rxa_do_softrss(m, rx_adapter->rss_key_be)
+				     : m->hash.rss;
+			ev->event = event;
+			ev->flow_id = (rss & ~flow_id_mask) |
+				      (ev->flow_id & flow_id_mask);
+			ev->mbuf = m;
+			ev++;
+		}
+	} else {
+		num = rxa_create_event_vector(rx_adapter, eth_rx_queue_info,
+					      buf, mbufs, num);
 	}
 
-	if (dev_info->cb_fn) {
+	if (num && dev_info->cb_fn) {
 
 		dropped = 0;
 		nb_cb = dev_info->cb_fn(eth_dev_id, rx_queue_id,
-					ETH_EVENT_BUFFER_SIZE, buf->count, ev,
-					num, dev_info->cb_arg, &dropped);
+					ETH_EVENT_BUFFER_SIZE, buf->count,
+					&buf->events[buf->count], num,
+					dev_info->cb_arg, &dropped);
 		if (unlikely(nb_cb > num))
 			RTE_EDEV_LOG_ERR("Rx CB returned %d (> %d) events",
 				nb_cb, num);
@@ -1124,6 +1230,30 @@ rxa_poll(struct rte_event_eth_rx_adapter *rx_adapter)
 	return nb_rx;
 }
 
+static void
+rxa_vector_expire(struct eth_rx_vector_data *vec, void *arg)
+{
+	struct rte_event_eth_rx_adapter *rx_adapter = arg;
+	struct rte_eth_event_enqueue_buffer *buf =
+		&rx_adapter->event_enqueue_buffer;
+	struct rte_event *ev;
+
+	if (buf->count)
+		rxa_flush_event_buffer(rx_adapter);
+
+	if (vec->vector_ev->nb_elem == 0)
+		return;
+	ev = &buf->events[buf->count];
+
+	/* Event ready. */
+	ev->event = vec->event;
+	ev->vec = vec->vector_ev;
+	buf->count++;
+
+	vec->vector_ev = NULL;
+	vec->ts = 0;
+}
+
 static int
 rxa_service_func(void *args)
 {
@@ -1137,6 +1267,24 @@ rxa_service_func(void *args)
 		return 0;
 	}
 
+	if (rx_adapter->ena_vector) {
+		if ((rte_rdtsc() - rx_adapter->prev_expiry_ts) >=
+		    rx_adapter->vector_tmo_ticks) {
+			struct eth_rx_vector_data *vec;
+
+			TAILQ_FOREACH(vec, &rx_adapter->vector_list, next) {
+				uint64_t elapsed_time = rte_rdtsc() - vec->ts;
+
+				if (elapsed_time >= vec->vector_timeout_ticks) {
+					rxa_vector_expire(vec, rx_adapter);
+					TAILQ_REMOVE(&rx_adapter->vector_list,
+						     vec, next);
+				}
+			}
+			rx_adapter->prev_expiry_ts = rte_rdtsc();
+		}
+	}
+
 	stats = &rx_adapter->stats;
 	stats->rx_packets += rxa_intr_ring_dequeue(rx_adapter);
 	stats->rx_packets += rxa_poll(rx_adapter);
@@ -1640,11 +1788,35 @@ rxa_update_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }
 
+static void
+rxa_set_vector_data(struct eth_rx_queue_info *queue_info, uint16_t vector_count,
+		    uint64_t vector_ns, struct rte_mempool *mp, int32_t qid,
+		    uint16_t port_id)
+{
+#define NSEC2TICK(__ns, __freq) (((__ns) * (__freq)) / 1E9)
+	struct eth_rx_vector_data *vector_data;
+	uint32_t flow_id;
+
+	vector_data = &queue_info->vector_data;
+	vector_data->max_vector_count = vector_count;
+	vector_data->port = port_id;
+	vector_data->queue = qid;
+	vector_data->vector_pool = mp;
+	vector_data->vector_timeout_ticks =
+		NSEC2TICK(vector_ns, rte_get_timer_hz());
+	vector_data->ts = 0;
+	flow_id = queue_info->event & 0xFFFFF;
+	flow_id =
+		flow_id == 0 ? (qid & 0xFFF) | (port_id & 0xFF) << 12 : flow_id;
+	vector_data->event = (queue_info->event & ~0xFFFFF) | flow_id;
+}
+
 static void
 rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
 	struct eth_device_info *dev_info,
 	int32_t rx_queue_id)
 {
+	struct eth_rx_vector_data *vec;
 	int pollq;
 	int intrq;
 	int sintrq;
@@ -1663,6 +1835,14 @@ rxa_sw_del(struct rte_event_eth_rx_adapter *rx_adapter,
 		return;
 	}
 
+	/* Push all the partial event vectors to event device. */
+	TAILQ_FOREACH(vec, &rx_adapter->vector_list, next) {
+		if (vec->queue != rx_queue_id)
+			continue;
+		rxa_vector_expire(vec, rx_adapter);
+		TAILQ_REMOVE(&rx_adapter->vector_list, vec, next);
+	}
+
 	pollq = rxa_polled_queue(dev_info, rx_queue_id);
 	intrq = rxa_intr_queue(dev_info, rx_queue_id);
 	sintrq = rxa_shared_intr(dev_info, rx_queue_id);
@@ -1741,6 +1921,42 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }
 
+static void
+rxa_sw_event_vector_configure(
+	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
+	int rx_queue_id,
+	const struct rte_event_eth_rx_adapter_event_vector_config *config)
+{
+	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
+	struct eth_rx_queue_info *queue_info;
+	struct rte_event *qi_ev;
+
+	if (rx_queue_id == -1) {
+		uint16_t nb_rx_queues;
+		uint16_t i;
+
+		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
+		for (i = 0; i < nb_rx_queues; i++)
+			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
+						      config);
+		return;
+	}
+
+	queue_info = &dev_info->rx_queue[rx_queue_id];
+	qi_ev = (struct rte_event *)&queue_info->event;
+	queue_info->ena_vector = 1;
+	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+	rxa_set_vector_data(queue_info, config->vector_sz,
+			    config->vector_timeout_ns, config->vector_mp,
+			    rx_queue_id, dev_info->dev->data->port_id);
+	rx_adapter->ena_vector = 1;
+	rx_adapter->vector_tmo_ticks =
+		rx_adapter->vector_tmo_ticks ?
+			      RTE_MIN(config->vector_timeout_ns >> 1,
+				rx_adapter->vector_tmo_ticks) :
+			      config->vector_timeout_ns >> 1;
+}
+
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -1967,6 +2183,7 @@ rte_event_eth_rx_adapter_create_ext(uint8_t id, uint8_t dev_id,
 	rx_adapter->conf_cb = conf_cb;
 	rx_adapter->conf_arg = conf_arg;
 	rx_adapter->id = id;
+	TAILQ_INIT(&rx_adapter->vector_list);
 	strcpy(rx_adapter->mem_name, mem_name);
 	rx_adapter->eth_devices = rte_zmalloc_socket(rx_adapter->mem_name,
 					RTE_MAX_ETHPORTS *
@@ -2081,6 +2298,15 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}
 
+	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
+	    (queue_conf->rx_queue_flags &
+	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
+		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
+				 eth_dev_id, id);
+		return -EINVAL;
+	}
+
 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
 		(rx_queue_id != -1)) {
 		RTE_EDEV_LOG_ERR("Rx queues can only be connected to single "
@@ -2143,6 +2369,17 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	return 0;
 }
 
+static int
+rxa_sw_vector_limits(struct rte_event_eth_rx_adapter_vector_limits *limits)
+{
+	limits->max_sz = MAX_VECTOR_SIZE;
+	limits->min_sz = MIN_VECTOR_SIZE;
+	limits->max_timeout_ns = MAX_VECTOR_NS;
+	limits->min_timeout_ns = MIN_VECTOR_NS;
+
+	return 0;
+}
+
 int
 rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 				int32_t rx_queue_id)
@@ -2333,7 +2570,8 @@ rte_event_eth_rx_adapter_queue_event_vector_config(
 		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
 			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
 	} else {
-		ret = -ENOTSUP;
+		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
+					      rx_queue_id, config);
 	}
 
 	return ret;
@@ -2371,7 +2609,7 @@ rte_event_eth_rx_adapter_vector_limits_get(
 		ret = dev->dev_ops->eth_rx_adapter_vector_limits_get(
 			dev, &rte_eth_devices[eth_port_id], limits);
 	} else {
-		ret = -ENOTSUP;
+		ret = rxa_sw_vector_limits(limits);
 	}
 
 	return ret;
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index be0499c52..62824654b 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -122,7 +122,11 @@ rte_event_eth_rx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
 
 	if (caps == NULL)
 		return -EINVAL;
-	*caps = 0;
+
+	if (dev->dev_ops->eth_rx_adapter_caps_get == NULL)
+		*caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
+	else
+		*caps = 0;
 
 	return dev->dev_ops->eth_rx_adapter_caps_get ?
 				(*dev->dev_ops->eth_rx_adapter_caps_get)(dev,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v10 5/8] eventdev: add Tx adapter event vector support
  2021-03-31  9:29                 ` [dpdk-dev] [PATCH v10 0/8] Introduce event vectorization pbhagavatula
                                     ` (3 preceding siblings ...)
  2021-03-31  9:29                   ` [dpdk-dev] [PATCH v10 4/8] eventdev: add Rx adapter event vector support pbhagavatula
@ 2021-03-31  9:29                   ` pbhagavatula
  2021-03-31  9:30                   ` [dpdk-dev] [PATCH v10 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
                                     ` (3 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-31  9:29 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support for event eth Tx adapter, the implementation
receives events from the single linked queue and based on
rte_event_vector::attr_valid transmits the vector of mbufs to a given
port, queue pair.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
---
 .../rte_event_eth_tx_adapter.c                | 66 ++++++++++++++++---
 lib/librte_eventdev/rte_eventdev.c            |  5 +-
 2 files changed, 60 insertions(+), 11 deletions(-)

diff --git a/lib/librte_eventdev/rte_event_eth_tx_adapter.c b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
index 5b4c42dcf..db260bfb6 100644
--- a/lib/librte_eventdev/rte_event_eth_tx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_tx_adapter.c
@@ -510,6 +510,47 @@ txa_service_buffer_retry(struct rte_mbuf **pkts, uint16_t unsent,
 	stats->tx_dropped += unsent - sent;
 }
 
+static uint16_t
+txa_process_event_vector(struct txa_service_data *txa,
+			 struct rte_event_vector *vec)
+{
+	struct txa_service_queue_info *tqi;
+	uint16_t port, queue, nb_tx = 0;
+	struct rte_mbuf **mbufs;
+	int i;
+
+	mbufs = (struct rte_mbuf **)vec->mbufs;
+	if (vec->attr_valid) {
+		port = vec->port;
+		queue = vec->queue;
+		tqi = txa_service_queue(txa, port, queue);
+		if (unlikely(tqi == NULL || !tqi->added)) {
+			rte_pktmbuf_free_bulk(mbufs, vec->nb_elem);
+			rte_mempool_put(rte_mempool_from_obj(vec), vec);
+			return 0;
+		}
+		for (i = 0; i < vec->nb_elem; i++) {
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
+						   mbufs[i]);
+		}
+	} else {
+		for (i = 0; i < vec->nb_elem; i++) {
+			port = mbufs[i]->port;
+			queue = rte_event_eth_tx_adapter_txq_get(mbufs[i]);
+			tqi = txa_service_queue(txa, port, queue);
+			if (unlikely(tqi == NULL || !tqi->added)) {
+				rte_pktmbuf_free(mbufs[i]);
+				continue;
+			}
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf,
+						   mbufs[i]);
+		}
+	}
+	rte_mempool_put(rte_mempool_from_obj(vec), vec);
+
+	return nb_tx;
+}
+
 static void
 txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
 	uint32_t n)
@@ -522,22 +563,27 @@ txa_service_tx(struct txa_service_data *txa, struct rte_event *ev,
 
 	nb_tx = 0;
 	for (i = 0; i < n; i++) {
-		struct rte_mbuf *m;
 		uint16_t port;
 		uint16_t queue;
 		struct txa_service_queue_info *tqi;
 
-		m = ev[i].mbuf;
-		port = m->port;
-		queue = rte_event_eth_tx_adapter_txq_get(m);
+		if (!(ev[i].event_type & RTE_EVENT_TYPE_VECTOR)) {
+			struct rte_mbuf *m;
 
-		tqi = txa_service_queue(txa, port, queue);
-		if (unlikely(tqi == NULL || !tqi->added)) {
-			rte_pktmbuf_free(m);
-			continue;
-		}
+			m = ev[i].mbuf;
+			port = m->port;
+			queue = rte_event_eth_tx_adapter_txq_get(m);
 
-		nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
+			tqi = txa_service_queue(txa, port, queue);
+			if (unlikely(tqi == NULL || !tqi->added)) {
+				rte_pktmbuf_free(m);
+				continue;
+			}
+
+			nb_tx += rte_eth_tx_buffer(port, queue, tqi->tx_buf, m);
+		} else {
+			nb_tx += txa_process_event_vector(txa, ev[i].vec);
+		}
 	}
 
 	stats->tx_packets += nb_tx;
diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
index 62824654b..c9bb5d227 100644
--- a/lib/librte_eventdev/rte_eventdev.c
+++ b/lib/librte_eventdev/rte_eventdev.c
@@ -196,7 +196,10 @@ rte_event_eth_tx_adapter_caps_get(uint8_t dev_id, uint16_t eth_port_id,
 	if (caps == NULL)
 		return -EINVAL;
 
-	*caps = 0;
+	if (dev->dev_ops->eth_tx_adapter_caps_get == NULL)
+		*caps = RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR;
+	else
+		*caps = 0;
 
 	return dev->dev_ops->eth_tx_adapter_caps_get ?
 			(*dev->dev_ops->eth_tx_adapter_caps_get)(dev,
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v10 6/8] app/eventdev: add event vector mode in pipeline test
  2021-03-31  9:29                 ` [dpdk-dev] [PATCH v10 0/8] Introduce event vectorization pbhagavatula
                                     ` (4 preceding siblings ...)
  2021-03-31  9:29                   ` [dpdk-dev] [PATCH v10 5/8] eventdev: add Tx " pbhagavatula
@ 2021-03-31  9:30                   ` pbhagavatula
  2021-03-31  9:30                   ` [dpdk-dev] [PATCH v10 7/8] doc: announce event Rx adapter config changes pbhagavatula
                                     ` (2 subsequent siblings)
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-31  9:30 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Add event vector support in pipeline tests. By default this mode
is disabled, it can be enabled by using the option --enable_vector.
example:
	dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
	--nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
	--wlcores=20-23  --enable_vector

Additional options to configure vector size and vector timeout are
also implemented and can be used by specifying --vector_size and
--vector_tmo_ns

This patch also adds a new option to set the number of Rx queues
configured per event eth rx adapter.
example:
	dpdk-test-eventdev -l 7-23 -s 0xff00 -- --prod_type_ethdev
	--nb_pkts=0 --verbose 2 --test=pipeline_atq --stlist=a
	--wlcores=20-23  --nb_eth_queues 4

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jerin Jacob <jerinj@marvell.com>
---
 app/test-eventdev/evt_common.h           |   4 +
 app/test-eventdev/evt_options.c          |  52 ++++
 app/test-eventdev/evt_options.h          |   4 +
 app/test-eventdev/test_pipeline_atq.c    | 310 ++++++++++++++++++++--
 app/test-eventdev/test_pipeline_common.c | 113 +++++++-
 app/test-eventdev/test_pipeline_common.h |  18 ++
 app/test-eventdev/test_pipeline_queue.c  | 320 +++++++++++++++++++++--
 doc/guides/tools/testeventdev.rst        |  45 +++-
 8 files changed, 811 insertions(+), 55 deletions(-)

diff --git a/app/test-eventdev/evt_common.h b/app/test-eventdev/evt_common.h
index a1da1cf11..0e228258e 100644
--- a/app/test-eventdev/evt_common.h
+++ b/app/test-eventdev/evt_common.h
@@ -58,16 +58,20 @@ struct evt_options {
 	uint8_t sched_type_list[EVT_MAX_STAGES];
 	uint16_t mbuf_sz;
 	uint16_t wkr_deq_dep;
+	uint16_t vector_size;
+	uint16_t eth_queues;
 	uint32_t nb_flows;
 	uint32_t tx_first;
 	uint32_t max_pkt_sz;
 	uint32_t deq_tmo_nsec;
 	uint32_t q_priority:1;
 	uint32_t fwd_latency:1;
+	uint32_t ena_vector : 1;
 	uint64_t nb_pkts;
 	uint64_t nb_timers;
 	uint64_t expiry_nsec;
 	uint64_t max_tmo_nsec;
+	uint64_t vector_tmo_nsec;
 	uint64_t timer_tick_nsec;
 	uint64_t optm_timer_tick_nsec;
 	enum evt_prod_type prod_type;
diff --git a/app/test-eventdev/evt_options.c b/app/test-eventdev/evt_options.c
index 0d04ea9f8..0d5540574 100644
--- a/app/test-eventdev/evt_options.c
+++ b/app/test-eventdev/evt_options.c
@@ -34,6 +34,9 @@ evt_options_default(struct evt_options *opt)
 	opt->max_tmo_nsec = 1E5;  /* 100000ns ~100us */
 	opt->expiry_nsec = 1E4;   /* 10000ns ~10us */
 	opt->prod_type = EVT_PROD_TYPE_SYNT;
+	opt->eth_queues = 1;
+	opt->vector_size = 64;
+	opt->vector_tmo_nsec = 100E3;
 }
 
 typedef int (*option_parser_t)(struct evt_options *opt,
@@ -257,6 +260,43 @@ evt_parse_max_pkt_sz(struct evt_options *opt, const char *arg)
 	return ret;
 }
 
+static int
+evt_parse_ena_vector(struct evt_options *opt, const char *arg __rte_unused)
+{
+	opt->ena_vector = 1;
+	return 0;
+}
+
+static int
+evt_parse_vector_size(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint16(&(opt->vector_size), arg);
+
+	return ret;
+}
+
+static int
+evt_parse_vector_tmo_ns(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint64(&(opt->vector_tmo_nsec), arg);
+
+	return ret;
+}
+
+static int
+evt_parse_eth_queues(struct evt_options *opt, const char *arg)
+{
+	int ret;
+
+	ret = parser_read_uint16(&(opt->eth_queues), arg);
+
+	return ret;
+}
+
 static void
 usage(char *program)
 {
@@ -289,6 +329,10 @@ usage(char *program)
 		"\t--expiry_nsec      : event timer expiry ns.\n"
 		"\t--mbuf_sz          : packet mbuf size.\n"
 		"\t--max_pkt_sz       : max packet size.\n"
+		"\t--nb_eth_queues    : number of ethernet Rx queues.\n"
+		"\t--enable_vector    : enable event vectorization.\n"
+		"\t--vector_size      : Max vector size.\n"
+		"\t--vector_tmo_ns    : Max vector timeout in nanoseconds\n"
 		);
 	printf("available tests:\n");
 	evt_test_dump_names();
@@ -360,6 +404,10 @@ static struct option lgopts[] = {
 	{ EVT_EXPIRY_NSEC,         1, 0, 0 },
 	{ EVT_MBUF_SZ,             1, 0, 0 },
 	{ EVT_MAX_PKT_SZ,          1, 0, 0 },
+	{ EVT_NB_ETH_QUEUES,       1, 0, 0 },
+	{ EVT_ENA_VECTOR,          0, 0, 0 },
+	{ EVT_VECTOR_SZ,           1, 0, 0 },
+	{ EVT_VECTOR_TMO,          1, 0, 0 },
 	{ EVT_HELP,                0, 0, 0 },
 	{ NULL,                    0, 0, 0 }
 };
@@ -394,6 +442,10 @@ evt_opts_parse_long(int opt_idx, struct evt_options *opt)
 		{ EVT_EXPIRY_NSEC, evt_parse_expiry_nsec},
 		{ EVT_MBUF_SZ, evt_parse_mbuf_sz},
 		{ EVT_MAX_PKT_SZ, evt_parse_max_pkt_sz},
+		{ EVT_NB_ETH_QUEUES, evt_parse_eth_queues},
+		{ EVT_ENA_VECTOR, evt_parse_ena_vector},
+		{ EVT_VECTOR_SZ, evt_parse_vector_size},
+		{ EVT_VECTOR_TMO, evt_parse_vector_tmo_ns},
 	};
 
 	for (i = 0; i < RTE_DIM(parsermap); i++) {
diff --git a/app/test-eventdev/evt_options.h b/app/test-eventdev/evt_options.h
index 748e54fae..1cea2a3e1 100644
--- a/app/test-eventdev/evt_options.h
+++ b/app/test-eventdev/evt_options.h
@@ -42,6 +42,10 @@
 #define EVT_EXPIRY_NSEC          ("expiry_nsec")
 #define EVT_MBUF_SZ              ("mbuf_sz")
 #define EVT_MAX_PKT_SZ           ("max_pkt_sz")
+#define EVT_NB_ETH_QUEUES        ("nb_eth_queues")
+#define EVT_ENA_VECTOR           ("enable_vector")
+#define EVT_VECTOR_SZ            ("vector_size")
+#define EVT_VECTOR_TMO           ("vector_tmo_ns")
 #define EVT_HELP                 ("help")
 
 void evt_options_default(struct evt_options *opt);
diff --git a/app/test-eventdev/test_pipeline_atq.c b/app/test-eventdev/test_pipeline_atq.c
index 0872b25b5..84dd4f44e 100644
--- a/app/test-eventdev/test_pipeline_atq.c
+++ b/app/test-eventdev/test_pipeline_atq.c
@@ -15,6 +15,8 @@ pipeline_atq_nb_event_queues(struct evt_options *opt)
 	return rte_eth_dev_count_avail();
 }
 
+typedef int (*pipeline_atq_worker_t)(void *arg);
+
 static __rte_noinline int
 pipeline_atq_worker_single_stage_tx(void *arg)
 {
@@ -113,6 +115,112 @@ pipeline_atq_worker_single_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_atq_worker_single_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+		vector_sz = ev.vec->nb_elem;
+		pipeline_event_tx_vector(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = ev.vec->nb_elem;
+		ev.queue_id = tx_queue[ev.vec->port];
+		ev.vec->queue = 0;
+		pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			vector_sz += ev[i].vec->nb_elem;
+			ev[i].vec->queue = 0;
+		}
+
+		pipeline_event_tx_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_single_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			ev[i].queue_id = tx_queue[ev[i].vec->port];
+			ev[i].vec->queue = 0;
+			vector_sz += ev[i].vec->nb_elem;
+			pipeline_fwd_event_vector(&ev[i],
+						  RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
 static __rte_noinline int
 pipeline_atq_worker_multi_stage_tx(void *arg)
 {
@@ -245,6 +353,147 @@ pipeline_atq_worker_multi_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.sub_event_type % nb_stages;
+
+		if (cq_id == last_queue) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+			continue;
+		}
+
+		ev.sub_event_type++;
+		pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.sub_event_type % nb_stages;
+
+		if (cq_id == last_queue) {
+			ev.queue_id = tx_queue[ev.vec->port];
+			ev.vec->queue = 0;
+			vector_sz = ev.vec->nb_elem;
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			pipeline_event_enqueue(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.sub_event_type++;
+			pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+			pipeline_event_enqueue(dev, port, &ev);
+		}
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].sub_event_type % nb_stages;
+
+			if (cq_id == last_queue) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+				continue;
+			}
+
+			ev[i].sub_event_type++;
+			pipeline_fwd_event_vector(&ev[i],
+						  sched_type_list[cq_id]);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_atq_worker_multi_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].sub_event_type % nb_stages;
+
+			if (cq_id == last_queue) {
+				vector_sz = ev[i].vec->nb_elem;
+				ev[i].queue_id = tx_queue[ev[i].vec->port];
+				ev[i].vec->queue = 0;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].sub_event_type++;
+				pipeline_fwd_event_vector(
+					&ev[i], sched_type_list[cq_id]);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
 static int
 worker_wrapper(void *arg)
 {
@@ -253,27 +502,36 @@ worker_wrapper(void *arg)
 	const bool burst = evt_has_burst_mode(w->dev_id);
 	const bool internal_port = w->t->internal_port;
 	const uint8_t nb_stages = opt->nb_stages;
-	RTE_SET_USED(opt);
+	/*vector/burst/internal_port*/
+	const pipeline_atq_worker_t
+	pipeline_atq_worker_single_stage[2][2][2] = {
+		[0][0][0] = pipeline_atq_worker_single_stage_fwd,
+		[0][0][1] = pipeline_atq_worker_single_stage_tx,
+		[0][1][0] = pipeline_atq_worker_single_stage_burst_fwd,
+		[0][1][1] = pipeline_atq_worker_single_stage_burst_tx,
+		[1][0][0] = pipeline_atq_worker_single_stage_fwd_vector,
+		[1][0][1] = pipeline_atq_worker_single_stage_tx_vector,
+		[1][1][0] = pipeline_atq_worker_single_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_atq_worker_single_stage_burst_tx_vector,
+	};
+	const pipeline_atq_worker_t
+	pipeline_atq_worker_multi_stage[2][2][2] = {
+		[0][0][0] = pipeline_atq_worker_multi_stage_fwd,
+		[0][0][1] = pipeline_atq_worker_multi_stage_tx,
+		[0][1][0] = pipeline_atq_worker_multi_stage_burst_fwd,
+		[0][1][1] = pipeline_atq_worker_multi_stage_burst_tx,
+		[1][0][0] = pipeline_atq_worker_multi_stage_fwd_vector,
+		[1][0][1] = pipeline_atq_worker_multi_stage_tx_vector,
+		[1][1][0] = pipeline_atq_worker_multi_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_atq_worker_multi_stage_burst_tx_vector,
+	};
 
-	if (nb_stages == 1) {
-		if (!burst && internal_port)
-			return pipeline_atq_worker_single_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_atq_worker_single_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_atq_worker_single_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_atq_worker_single_stage_burst_fwd(arg);
-	} else {
-		if (!burst && internal_port)
-			return pipeline_atq_worker_multi_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_atq_worker_multi_stage_fwd(arg);
-		if (burst && internal_port)
-			return pipeline_atq_worker_multi_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_atq_worker_multi_stage_burst_fwd(arg);
-	}
+	if (nb_stages == 1)
+		return (pipeline_atq_worker_single_stage[opt->ena_vector][burst]
+							[internal_port])(arg);
+	else
+		return (pipeline_atq_worker_multi_stage[opt->ena_vector][burst]
+						       [internal_port])(arg);
 
 	rte_panic("invalid worker\n");
 }
@@ -290,7 +548,7 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
 	int ret;
 	int nb_ports;
 	int nb_queues;
-	uint8_t queue;
+	uint8_t queue, is_prod;
 	uint8_t tx_evqueue_id[RTE_MAX_ETHPORTS];
 	uint8_t queue_arr[RTE_EVENT_MAX_QUEUES_PER_DEV];
 	uint8_t nb_worker_queues = 0;
@@ -330,15 +588,19 @@ pipeline_atq_eventdev_setup(struct evt_test *test, struct evt_options *opt)
 		q_conf.event_queue_cfg = RTE_EVENT_QUEUE_CFG_ALL_TYPES;
 
 		if (!t->internal_port) {
+			is_prod = false;
 			RTE_ETH_FOREACH_DEV(prod) {
 				if (queue == tx_evqueue_id[prod]) {
 					q_conf.event_queue_cfg =
 						RTE_EVENT_QUEUE_CFG_SINGLE_LINK;
-				} else {
-					queue_arr[nb_worker_queues] = queue;
-					nb_worker_queues++;
+					is_prod = true;
+					break;
 				}
 			}
+			if (!is_prod) {
+				queue_arr[nb_worker_queues] = queue;
+				nb_worker_queues++;
+			}
 		}
 
 		ret = rte_event_queue_setup(opt->dev_id, queue, &q_conf);
diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index b47d76743..d5ef90500 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -36,6 +36,12 @@ pipeline_opt_dump(struct evt_options *opt, uint8_t nb_queues)
 	evt_dump_queue_priority(opt);
 	evt_dump_sched_type_list(opt);
 	evt_dump_producer_type(opt);
+	evt_dump("nb_eth_rx_queues", "%d", opt->eth_queues);
+	evt_dump("event_vector", "%d", opt->ena_vector);
+	if (opt->ena_vector) {
+		evt_dump("vector_size", "%d", opt->vector_size);
+		evt_dump("vector_tmo_ns", "%" PRIu64 "", opt->vector_tmo_nsec);
+	}
 }
 
 static inline uint64_t
@@ -163,7 +169,7 @@ pipeline_opt_check(struct evt_options *opt, uint64_t nb_queues)
 int
 pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 {
-	uint16_t i;
+	uint16_t i, j;
 	int ret;
 	uint8_t nb_queues = 1;
 	struct test_pipeline *t = evt_test_priv(test);
@@ -210,6 +216,16 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 		if (!(caps & RTE_EVENT_ETH_TX_ADAPTER_CAP_INTERNAL_PORT))
 			t->internal_port = 0;
 
+		ret = rte_event_eth_rx_adapter_caps_get(opt->dev_id, i, &caps);
+		if (ret != 0) {
+			evt_err("failed to get event tx adapter[%d] caps", i);
+			return ret;
+		}
+
+		if (!(caps & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT))
+			local_port_conf.rxmode.offloads |=
+				DEV_RX_OFFLOAD_RSS_HASH;
+
 		ret = rte_eth_dev_info_get(i, &dev_info);
 		if (ret != 0) {
 			evt_err("Error during getting device (port %u) info: %s\n",
@@ -236,19 +252,22 @@ pipeline_ethdev_setup(struct evt_test *test, struct evt_options *opt)
 				local_port_conf.rx_adv_conf.rss_conf.rss_hf);
 		}
 
-		if (rte_eth_dev_configure(i, nb_queues, nb_queues,
-					&local_port_conf)
-				< 0) {
+		if (rte_eth_dev_configure(i, opt->eth_queues, nb_queues,
+					  &local_port_conf) < 0) {
 			evt_err("Failed to configure eth port [%d]", i);
 			return -EINVAL;
 		}
 
-		if (rte_eth_rx_queue_setup(i, 0, NB_RX_DESC,
-				rte_socket_id(), &rx_conf, t->pool) < 0) {
-			evt_err("Failed to setup eth port [%d] rx_queue: %d.",
+		for (j = 0; j < opt->eth_queues; j++) {
+			if (rte_eth_rx_queue_setup(i, j, NB_RX_DESC,
+						   rte_socket_id(), &rx_conf,
+						   t->pool) < 0) {
+				evt_err("Failed to setup eth port [%d] rx_queue: %d.",
 					i, 0);
-			return -EINVAL;
+				return -EINVAL;
+			}
 		}
+
 		if (rte_eth_tx_queue_setup(i, 0, NB_TX_DESC,
 					rte_socket_id(), NULL) < 0) {
 			evt_err("Failed to setup eth port [%d] tx_queue: %d.",
@@ -310,12 +329,27 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 {
 	int ret = 0;
 	uint16_t prod;
+	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
+	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;
 
 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
 	queue_conf.ev.sched_type = opt->sched_type_list[0];
+	if (opt->ena_vector) {
+		unsigned int nb_elem = (opt->pool_sz / opt->vector_size) << 1;
+
+		nb_elem = nb_elem ? nb_elem : 1;
+		vector_pool = rte_event_vector_pool_create(
+			"vector_pool", nb_elem, 0, opt->vector_size,
+			opt->socket_id);
+		if (vector_pool == NULL) {
+			evt_err("failed to create event vector pool");
+			return -ENOMEM;
+		}
+	}
 	RTE_ETH_FOREACH_DEV(prod) {
+		struct rte_event_eth_rx_adapter_vector_limits limits;
 		uint32_t cap;
 
 		ret = rte_event_eth_rx_adapter_caps_get(opt->dev_id,
@@ -326,6 +360,50 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 					opt->dev_id);
 			return ret;
 		}
+
+		if (opt->ena_vector) {
+			memset(&limits, 0, sizeof(limits));
+			ret = rte_event_eth_rx_adapter_vector_limits_get(
+				opt->dev_id, prod, &limits);
+			if (ret) {
+				evt_err("failed to get vector limits");
+				return ret;
+			}
+
+			if (opt->vector_size < limits.min_sz ||
+			    opt->vector_size > limits.max_sz) {
+				evt_err("Vector size [%d] not within limits max[%d] min[%d]",
+					opt->vector_size, limits.min_sz,
+					limits.max_sz);
+				return -EINVAL;
+			}
+
+			if (limits.log2_sz &&
+			    !rte_is_power_of_2(opt->vector_size)) {
+				evt_err("Vector size [%d] not power of 2",
+					opt->vector_size);
+				return -EINVAL;
+			}
+
+			if (opt->vector_tmo_nsec > limits.max_timeout_ns ||
+			    opt->vector_tmo_nsec < limits.min_timeout_ns) {
+				evt_err("Vector timeout [%" PRIu64
+					"] not within limits max[%" PRIu64
+					"] min[%" PRIu64 "]",
+					opt->vector_tmo_nsec,
+					limits.max_timeout_ns,
+					limits.min_timeout_ns);
+				return -EINVAL;
+			}
+
+			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.rx_queue_flags |=
+				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+			} else {
+				evt_err("Rx adapter doesn't support event vector");
+				return -EINVAL;
+			}
+		}
 		queue_conf.ev.queue_id = prod * stride;
 		ret = rte_event_eth_rx_adapter_create(prod, opt->dev_id,
 				&prod_conf);
@@ -340,6 +418,17 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}
 
+		if (opt->ena_vector) {
+			vec_conf.vector_sz = opt->vector_size;
+			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
+			vec_conf.vector_mp = vector_pool;
+			if (rte_event_eth_rx_adapter_queue_event_vector_config(
+				    prod, prod, -1, &vec_conf) < 0) {
+				evt_err("Failed to configure event vectorization for Rx adapter");
+				return -EINVAL;
+			}
+		}
+
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;
 
@@ -378,6 +467,14 @@ pipeline_event_tx_adapter_setup(struct evt_options *opt,
 			return ret;
 		}
 
+		if (opt->ena_vector) {
+			if (!(cap &
+			      RTE_EVENT_ETH_TX_ADAPTER_CAP_EVENT_VECTOR)) {
+				evt_err("Tx adapter doesn't support event vector");
+				return -EINVAL;
+			}
+		}
+
 		ret = rte_event_eth_tx_adapter_create(consm, opt->dev_id,
 				&port_conf);
 		if (ret) {
diff --git a/app/test-eventdev/test_pipeline_common.h b/app/test-eventdev/test_pipeline_common.h
index 6e73c6ab2..800a90616 100644
--- a/app/test-eventdev/test_pipeline_common.h
+++ b/app/test-eventdev/test_pipeline_common.h
@@ -101,6 +101,14 @@ pipeline_fwd_event(struct rte_event *ev, uint8_t sched)
 	ev->sched_type = sched;
 }
 
+static __rte_always_inline void
+pipeline_fwd_event_vector(struct rte_event *ev, uint8_t sched)
+{
+	ev->event_type = RTE_EVENT_TYPE_CPU_VECTOR;
+	ev->op = RTE_EVENT_OP_FORWARD;
+	ev->sched_type = sched;
+}
+
 static __rte_always_inline void
 pipeline_event_tx(const uint8_t dev, const uint8_t port,
 		struct rte_event * const ev)
@@ -110,6 +118,16 @@ pipeline_event_tx(const uint8_t dev, const uint8_t port,
 		rte_pause();
 }
 
+static __rte_always_inline void
+pipeline_event_tx_vector(const uint8_t dev, const uint8_t port,
+			 struct rte_event *const ev)
+{
+	ev->vec->queue = 0;
+
+	while (!rte_event_eth_tx_adapter_enqueue(dev, port, ev, 1, 0))
+		rte_pause();
+}
+
 static __rte_always_inline void
 pipeline_event_tx_burst(const uint8_t dev, const uint8_t port,
 		struct rte_event *ev, const uint16_t nb_rx)
diff --git a/app/test-eventdev/test_pipeline_queue.c b/app/test-eventdev/test_pipeline_queue.c
index 9a9febb19..f6cc3e358 100644
--- a/app/test-eventdev/test_pipeline_queue.c
+++ b/app/test-eventdev/test_pipeline_queue.c
@@ -15,6 +15,8 @@ pipeline_queue_nb_event_queues(struct evt_options *opt)
 	return (eth_count * opt->nb_stages) + eth_count;
 }
 
+typedef int (*pipeline_queue_worker_t)(void *arg);
+
 static __rte_noinline int
 pipeline_queue_worker_single_stage_tx(void *arg)
 {
@@ -126,6 +128,125 @@ pipeline_queue_worker_single_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_queue_worker_single_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		if (ev.sched_type == RTE_SCHED_TYPE_ATOMIC) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.queue_id++;
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			pipeline_event_enqueue(dev, port, &ev);
+		}
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		ev.queue_id = tx_queue[ev.vec->port];
+		ev.vec->queue = 0;
+		vector_sz = ev.vec->nb_elem;
+		pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			if (ev[i].sched_type == RTE_SCHED_TYPE_ATOMIC) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].queue_id++;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_single_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_SINGLE_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		vector_sz = 0;
+		for (i = 0; i < nb_rx; i++) {
+			ev[i].queue_id = tx_queue[ev[i].vec->port];
+			ev[i].vec->queue = 0;
+			vector_sz += ev[i].vec->nb_elem;
+			pipeline_fwd_event_vector(&ev[i],
+						  RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+		w->processed_pkts += vector_sz;
+	}
+
+	return 0;
+}
 
 static __rte_noinline int
 pipeline_queue_worker_multi_stage_tx(void *arg)
@@ -267,6 +388,151 @@ pipeline_queue_worker_multi_stage_burst_fwd(void *arg)
 	return 0;
 }
 
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.queue_id % nb_stages;
+
+		if (ev.queue_id == tx_queue[ev.vec->port]) {
+			vector_sz = ev.vec->nb_elem;
+			pipeline_event_tx_vector(dev, port, &ev);
+			w->processed_pkts += vector_sz;
+			continue;
+		}
+
+		ev.queue_id++;
+		pipeline_fwd_event_vector(&ev, cq_id != last_queue
+						       ? sched_type_list[cq_id]
+						       : RTE_SCHED_TYPE_ATOMIC);
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0);
+
+		if (!event) {
+			rte_pause();
+			continue;
+		}
+
+		cq_id = ev.queue_id % nb_stages;
+
+		if (cq_id == last_queue) {
+			vector_sz = ev.vec->nb_elem;
+			ev.queue_id = tx_queue[ev.vec->port];
+			pipeline_fwd_event_vector(&ev, RTE_SCHED_TYPE_ATOMIC);
+			w->processed_pkts += vector_sz;
+		} else {
+			ev.queue_id++;
+			pipeline_fwd_event_vector(&ev, sched_type_list[cq_id]);
+		}
+
+		pipeline_event_enqueue(dev, port, &ev);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_burst_tx_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].queue_id % nb_stages;
+
+			if (ev[i].queue_id == tx_queue[ev[i].vec->port]) {
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_event_tx_vector(dev, port, &ev[i]);
+				ev[i].op = RTE_EVENT_OP_RELEASE;
+				w->processed_pkts += vector_sz;
+				continue;
+			}
+
+			ev[i].queue_id++;
+			pipeline_fwd_event_vector(
+				&ev[i], cq_id != last_queue
+						? sched_type_list[cq_id]
+						: RTE_SCHED_TYPE_ATOMIC);
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
+static __rte_noinline int
+pipeline_queue_worker_multi_stage_burst_fwd_vector(void *arg)
+{
+	PIPELINE_WORKER_MULTI_STAGE_BURST_INIT;
+	const uint8_t *tx_queue = t->tx_evqueue_id;
+	uint16_t vector_sz;
+
+	while (!t->done) {
+		uint16_t nb_rx =
+			rte_event_dequeue_burst(dev, port, ev, BURST_SIZE, 0);
+
+		if (!nb_rx) {
+			rte_pause();
+			continue;
+		}
+
+		for (i = 0; i < nb_rx; i++) {
+			cq_id = ev[i].queue_id % nb_stages;
+
+			if (cq_id == last_queue) {
+				ev[i].queue_id = tx_queue[ev[i].vec->port];
+				vector_sz = ev[i].vec->nb_elem;
+				pipeline_fwd_event_vector(
+					&ev[i], RTE_SCHED_TYPE_ATOMIC);
+				w->processed_pkts += vector_sz;
+			} else {
+				ev[i].queue_id++;
+				pipeline_fwd_event_vector(
+					&ev[i], sched_type_list[cq_id]);
+			}
+		}
+
+		pipeline_event_enqueue_burst(dev, port, ev, nb_rx);
+	}
+
+	return 0;
+}
+
 static int
 worker_wrapper(void *arg)
 {
@@ -275,29 +541,39 @@ worker_wrapper(void *arg)
 	const bool burst = evt_has_burst_mode(w->dev_id);
 	const bool internal_port = w->t->internal_port;
 	const uint8_t nb_stages = opt->nb_stages;
-	RTE_SET_USED(opt);
-
-	if (nb_stages == 1) {
-		if (!burst && internal_port)
-			return pipeline_queue_worker_single_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_queue_worker_single_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_queue_worker_single_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_queue_worker_single_stage_burst_fwd(
-					arg);
-	} else {
-		if (!burst && internal_port)
-			return pipeline_queue_worker_multi_stage_tx(arg);
-		else if (!burst && !internal_port)
-			return pipeline_queue_worker_multi_stage_fwd(arg);
-		else if (burst && internal_port)
-			return pipeline_queue_worker_multi_stage_burst_tx(arg);
-		else if (burst && !internal_port)
-			return pipeline_queue_worker_multi_stage_burst_fwd(arg);
+	/*vector/burst/internal_port*/
+	const pipeline_queue_worker_t
+	pipeline_queue_worker_single_stage[2][2][2] = {
+		[0][0][0] = pipeline_queue_worker_single_stage_fwd,
+		[0][0][1] = pipeline_queue_worker_single_stage_tx,
+		[0][1][0] = pipeline_queue_worker_single_stage_burst_fwd,
+		[0][1][1] = pipeline_queue_worker_single_stage_burst_tx,
+		[1][0][0] = pipeline_queue_worker_single_stage_fwd_vector,
+		[1][0][1] = pipeline_queue_worker_single_stage_tx_vector,
+		[1][1][0] = pipeline_queue_worker_single_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_queue_worker_single_stage_burst_tx_vector,
+	};
+	const pipeline_queue_worker_t
+	pipeline_queue_worker_multi_stage[2][2][2] = {
+		[0][0][0] = pipeline_queue_worker_multi_stage_fwd,
+		[0][0][1] = pipeline_queue_worker_multi_stage_tx,
+		[0][1][0] = pipeline_queue_worker_multi_stage_burst_fwd,
+		[0][1][1] = pipeline_queue_worker_multi_stage_burst_tx,
+		[1][0][0] = pipeline_queue_worker_multi_stage_fwd_vector,
+		[1][0][1] = pipeline_queue_worker_multi_stage_tx_vector,
+		[1][1][0] = pipeline_queue_worker_multi_stage_burst_fwd_vector,
+		[1][1][1] = pipeline_queue_worker_multi_stage_burst_tx_vector,
+	};
+
+	if (nb_stages == 1)
+		return (pipeline_queue_worker_single_stage[opt->ena_vector]
+							  [burst]
+							  [internal_port])(arg);
+	else
+		return (pipeline_queue_worker_multi_stage[opt->ena_vector]
+							 [burst]
+							 [internal_port])(arg);
 
-	}
 	rte_panic("invalid worker\n");
 }
 
diff --git a/doc/guides/tools/testeventdev.rst b/doc/guides/tools/testeventdev.rst
index ad1788a3d..f252dc2c1 100644
--- a/doc/guides/tools/testeventdev.rst
+++ b/doc/guides/tools/testeventdev.rst
@@ -158,6 +158,26 @@ The following are the application command-line options:
        Set max packet mbuf size. Can be used configure Rx/Tx scatter gather.
        Only applicable for `pipeline_atq` and `pipeline_queue` tests.
 
+* ``--nb_eth_queues``
+
+       Configure multiple Rx queues per each ethernet port.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--enable_vector``
+
+       Enable event vector for Rx/Tx adapters.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--vector_size``
+
+       Vector size to configure for the Rx adapter.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
+* ``--vector_tmo_ns``
+
+       Vector timeout nanoseconds to be configured for the Rx adapter.
+       Only applicable for `pipeline_atq` and `pipeline_queue` tests.
+
 
 Eventdev Tests
 --------------
@@ -607,6 +627,10 @@ Supported application command line options are following::
         --worker_deq_depth
         --prod_type_ethdev
         --deq_tmo_nsec
+        --nb_eth_queues
+        --enable_vector
+        --vector_size
+        --vector_tmo_ns
 
 
 .. Note::
@@ -623,6 +647,13 @@ Example command to run pipeline queue test:
     sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
         --test=pipeline_queue --wlcore=1 --prod_type_ethdev --stlist=a
 
+Example command to run pipeline atq test with vector events:
+
+.. code-block:: console
+
+    sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
+        --test=pipeline_queue --wlcore=1 --prod_type_ethdev --stlist=a \
+        --enable_vector  --vector_size 512
 
 PIPELINE_ATQ Test
 ~~~~~~~~~~~~~~~~~~~
@@ -699,6 +730,10 @@ Supported application command line options are following::
         --worker_deq_depth
         --prod_type_ethdev
         --deq_tmo_nsec
+        --nb_eth_queues
+        --enable_vector
+        --vector_size
+        --vector_tmo_ns
 
 
 .. Note::
@@ -708,9 +743,17 @@ Supported application command line options are following::
 Example
 ^^^^^^^
 
-Example command to run pipeline queue test:
+Example command to run pipeline atq test:
 
 .. code-block:: console
 
     sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
         --test=pipeline_atq --wlcore=1 --prod_type_ethdev --stlist=a
+
+Example command to run pipeline atq test with vector events:
+
+.. code-block:: console
+
+    sudo <build_dir>/app/dpdk-test-eventdev -c 0xf -s 0x8 --vdev=event_sw0 -- \
+        --test=pipeline_atq --wlcore=1 --prod_type_ethdev --stlist=a \
+        --enable_vector  --vector_size 512
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v10 7/8] doc: announce event Rx adapter config changes
  2021-03-31  9:29                 ` [dpdk-dev] [PATCH v10 0/8] Introduce event vectorization pbhagavatula
                                     ` (5 preceding siblings ...)
  2021-03-31  9:30                   ` [dpdk-dev] [PATCH v10 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
@ 2021-03-31  9:30                   ` pbhagavatula
  2021-03-31  9:30                   ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v10 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
  2021-04-03  9:44                   ` [dpdk-dev] [PATCH v10 0/8] Introduce event vectorization Jerin Jacob
  8 siblings, 0 replies; 153+ messages in thread
From: pbhagavatula @ 2021-03-31  9:30 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

The Rx adapter event vector configuration will be merged into
Rx adapter queue configuration to simplify enabling event
vectorization.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Ray Kinsella <mdr@ashroe.eu>
Acked-by: Jerin Jacob <jerinj@marvell.com>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
---
 doc/guides/rel_notes/deprecation.rst | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst
index 64629e064..ef6d1bb18 100644
--- a/doc/guides/rel_notes/deprecation.rst
+++ b/doc/guides/rel_notes/deprecation.rst
@@ -130,3 +130,12 @@ Deprecation Notices
 * cmdline: ``cmdline`` structure will be made opaque to hide platform-specific
   content. On Linux and FreeBSD, supported prior to DPDK 20.11,
   original structure will be kept until DPDK 21.11.
+
+* eventdev: The structure ``rte_event_eth_rx_adapter_queue_conf`` will be
+  extended to include ``rte_event_eth_rx_adapter_event_vector_config`` elements
+  and the function ``rte_event_eth_rx_adapter_queue_event_vector_config`` will
+  be removed in DPDK 21.11.
+
+  An application can enable event vectorization by passing the desired vector
+  values to the function ``rte_event_eth_rx_adapter_queue_add`` using
+  the structure ``rte_event_eth_rx_adapter_queue_add``.
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [dpdk-dev v21.11] [PATCH v10 8/8] eventdev: simplify Rx adapter event vector config
  2021-03-31  9:29                 ` [dpdk-dev] [PATCH v10 0/8] Introduce event vectorization pbhagavatula
                                     ` (6 preceding siblings ...)
  2021-03-31  9:30                   ` [dpdk-dev] [PATCH v10 7/8] doc: announce event Rx adapter config changes pbhagavatula
@ 2021-03-31  9:30                   ` pbhagavatula
  2021-08-18  4:56                     ` [dpdk-dev] [PATCH v11] " pbhagavatula
  2021-04-03  9:44                   ` [dpdk-dev] [PATCH v10 0/8] Introduce event vectorization Jerin Jacob
  8 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-03-31  9:30 UTC (permalink / raw)
  To: jerinj, jay.jayatheerthan, erik.g.carrillo, abhinandan.gujjar,
	timothy.mcdaniel, hemant.agrawal, harry.van.haaren,
	mattias.ronnblom, liang.j.ma, Ray Kinsella, Neil Horman
  Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Include vector configuration into the structure
``rte_event_eth_rx_adapter_queue_conf`` used when configuring rest
of the Rx adapter ethernet device Rx queue parameters.
This simplifies event vector configuration as it avoids splitting
configuration per Rx queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
---
 app/test-eventdev/test_pipeline_common.c      |  16 +-
 lib/librte_eventdev/eventdev_pmd.h            |  29 ---
 .../rte_event_eth_rx_adapter.c                | 179 ++++++------------
 .../rte_event_eth_rx_adapter.h                |  27 ---
 lib/librte_eventdev/version.map               |   1 -
 5 files changed, 63 insertions(+), 189 deletions(-)

diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index d5ef90500..76aee254b 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -331,7 +331,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 	uint16_t prod;
 	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
-	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
@@ -397,8 +396,12 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			}

 			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.vector_sz = opt->vector_size;
+				queue_conf.vector_timeout_ns =
+					opt->vector_tmo_nsec;
 				queue_conf.rx_queue_flags |=
 				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+				queue_conf.vector_mp = vector_pool;
 			} else {
 				evt_err("Rx adapter doesn't support event vector");
 				return -EINVAL;
@@ -418,17 +421,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}

-		if (opt->ena_vector) {
-			vec_conf.vector_sz = opt->vector_size;
-			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
-			vec_conf.vector_mp = vector_pool;
-			if (rte_event_eth_rx_adapter_queue_event_vector_config(
-				    prod, prod, -1, &vec_conf) < 0) {
-				evt_err("Failed to configure event vectorization for Rx adapter");
-				return -EINVAL;
-			}
-		}
-
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;

diff --git a/lib/librte_eventdev/eventdev_pmd.h b/lib/librte_eventdev/eventdev_pmd.h
index 0f724ac85..63b3bc4b5 100644
--- a/lib/librte_eventdev/eventdev_pmd.h
+++ b/lib/librte_eventdev/eventdev_pmd.h
@@ -667,32 +667,6 @@ typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
 	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-struct rte_event_eth_rx_adapter_event_vector_config;
-/**
- * Enable event vector on an given Rx queue of a ethernet devices belonging to
- * the Rx adapter.
- *
- * @param dev
- *   Event device pointer
- *
- * @param eth_dev
- *   Ethernet device pointer
- *
- * @param rx_queue_id
- *   The Rx queue identifier
- *
- * @param config
- *   Pointer to the event vector configuration structure.
- *
- * @return
- *   - 0: Success.
- *   - <0: Error code returned by the driver function.
- */
-typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
-	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
-	int32_t rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;

@@ -1118,9 +1092,6 @@ struct rte_eventdev_ops {
 	eventdev_eth_rx_adapter_vector_limits_get_t
 		eth_rx_adapter_vector_limits_get;
 	/**< Get event vector limits for the Rx adapter */
-	eventdev_eth_rx_adapter_event_vector_config_t
-		eth_rx_adapter_event_vector_config;
-	/**< Configure Rx adapter with event vector */

 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
index e273b3acf..5cdb3c3cb 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
@@ -1896,6 +1896,24 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	} else
 		qi_ev->flow_id = 0;

+	if (conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+		queue_info->ena_vector = 1;
+		qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+		rxa_set_vector_data(queue_info, conf->vector_sz,
+				    conf->vector_timeout_ns, conf->vector_mp,
+				    rx_queue_id, dev_info->dev->data->port_id);
+		rx_adapter->ena_vector = 1;
+		rx_adapter->vector_tmo_ticks =
+			rx_adapter->vector_tmo_ticks ?
+				      RTE_MIN(queue_info->vector_data
+							.vector_timeout_ticks >>
+						1,
+					rx_adapter->vector_tmo_ticks) :
+				queue_info->vector_data.vector_timeout_ticks >>
+					1;
+	}
+
 	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
 	if (rxa_polled_queue(dev_info, rx_queue_id)) {
 		rx_adapter->num_rx_polled += !pollq;
@@ -1921,42 +1939,6 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }

-static void
-rxa_sw_event_vector_configure(
-	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
-	int rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
-	struct eth_rx_queue_info *queue_info;
-	struct rte_event *qi_ev;
-
-	if (rx_queue_id == -1) {
-		uint16_t nb_rx_queues;
-		uint16_t i;
-
-		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
-		for (i = 0; i < nb_rx_queues; i++)
-			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
-						      config);
-		return;
-	}
-
-	queue_info = &dev_info->rx_queue[rx_queue_id];
-	qi_ev = (struct rte_event *)&queue_info->event;
-	queue_info->ena_vector = 1;
-	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
-	rxa_set_vector_data(queue_info, config->vector_sz,
-			    config->vector_timeout_ns, config->vector_mp,
-			    rx_queue_id, dev_info->dev->data->port_id);
-	rx_adapter->ena_vector = 1;
-	rx_adapter->vector_tmo_ticks =
-		rx_adapter->vector_tmo_ticks ?
-			      RTE_MIN(config->vector_timeout_ns >> 1,
-				rx_adapter->vector_tmo_ticks) :
-			      config->vector_timeout_ns >> 1;
-}
-
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -2271,6 +2253,7 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	struct rte_event_eth_rx_adapter *rx_adapter;
 	struct rte_eventdev *dev;
 	struct eth_device_info *dev_info;
+	struct rte_event_eth_rx_adapter_vector_limits limits;

 	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
 	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
@@ -2298,13 +2281,46 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}

-	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
-	    (queue_conf->rx_queue_flags &
-	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
-		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
+	if (queue_conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+
+		if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0) {
+			RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+
+		ret = rte_event_eth_rx_adapter_vector_limits_get(
+			rx_adapter->eventdev_id, eth_dev_id, &limits);
+		if (ret < 0) {
+			RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_sz < limits.min_sz ||
+		    queue_conf->vector_sz > limits.max_sz ||
+		    queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
+		    queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
+		    queue_conf->vector_mp == NULL) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_mp->elt_size <
+		    (sizeof(struct rte_event_vector) +
+		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
 	}

 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
@@ -2500,83 +2516,6 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }

-int
-rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct rte_event_eth_rx_adapter_vector_limits limits;
-	struct rte_event_eth_rx_adapter *rx_adapter;
-	struct rte_eventdev *dev;
-	uint32_t cap;
-	int ret;
-
-	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
-	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
-
-	rx_adapter = rxa_id_to_adapter(id);
-	if ((rx_adapter == NULL) || (config == NULL))
-		return -EINVAL;
-
-	dev = &rte_eventdevs[rx_adapter->eventdev_id];
-	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
-						eth_dev_id, &cap);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
-				 "eth port %" PRIu16,
-				 id, eth_dev_id);
-		return ret;
-	}
-
-	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
-		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	ret = rte_event_eth_rx_adapter_vector_limits_get(
-		rx_adapter->eventdev_id, eth_dev_id, &limits);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
-				 "eth port %" PRIu16,
-				 rx_adapter->eventdev_id, eth_dev_id);
-		return ret;
-	}
-
-	if (config->vector_sz < limits.min_sz ||
-	    config->vector_sz > limits.max_sz ||
-	    config->vector_timeout_ns < limits.min_timeout_ns ||
-	    config->vector_timeout_ns > limits.max_timeout_ns ||
-	    config->vector_mp == NULL) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-	if (config->vector_mp->elt_size <
-	    (sizeof(struct rte_event_vector) +
-	     (sizeof(uintptr_t) * config->vector_sz))) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
-		RTE_FUNC_PTR_OR_ERR_RET(
-			*dev->dev_ops->eth_rx_adapter_event_vector_config,
-			-ENOTSUP);
-		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
-			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
-	} else {
-		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
-					      rx_queue_id, config);
-	}
-
-	return ret;
-}
-
 int
 rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.h b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
index 7407cde00..3f8b36229 100644
--- a/lib/librte_eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.h
@@ -171,9 +171,6 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
 	 * enqueued event.
 	 */
-};
-
-struct rte_event_eth_rx_adapter_event_vector_config {
 	uint16_t vector_sz;
 	/**<
 	 * Indicates the maximum number for mbufs to combine and form a vector.
@@ -548,30 +545,6 @@ int rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-/**
- * Configure event vectorization for a given ethernet device queue, that has
- * been added to a event eth Rx adapter.
- *
- * @param id
- *  The identifier of the ethernet Rx event adapter.
- *
- * @param eth_dev_id
- *  The identifier of the ethernet device.
- *
- * @param rx_queue_id
- *  Ethernet device receive queue index.
- *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
- *  are configured with event vectorization.
- *
- * @return
- *  - 0: Success, Receive queue configured correctly.
- *  - <0: Error code on failure.
- */
-__rte_experimental
-int rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eventdev/version.map b/lib/librte_eventdev/version.map
index 902df0ae3..34c1c830e 100644
--- a/lib/librte_eventdev/version.map
+++ b/lib/librte_eventdev/version.map
@@ -142,7 +142,6 @@ EXPERIMENTAL {
 	#added in 21.05
 	rte_event_vector_pool_create;
 	rte_event_eth_rx_adapter_vector_limits_get;
-	rte_event_eth_rx_adapter_queue_event_vector_config;
 };

 INTERNAL {
--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v10 0/8] Introduce event vectorization
  2021-03-31  9:29                 ` [dpdk-dev] [PATCH v10 0/8] Introduce event vectorization pbhagavatula
                                     ` (7 preceding siblings ...)
  2021-03-31  9:30                   ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v10 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
@ 2021-04-03  9:44                   ` Jerin Jacob
  8 siblings, 0 replies; 153+ messages in thread
From: Jerin Jacob @ 2021-04-03  9:44 UTC (permalink / raw)
  To: Pavan Nikhilesh
  Cc: Jerin Jacob, Jayatheerthan, Jay, Erik Gabriel Carrillo, Gujjar,
	Abhinandan S, McDaniel, Timothy, Hemant Agrawal, Van Haaren,
	Harry, Mattias Rönnblom, Liang Ma, dpdk-dev

On Wed, Mar 31, 2021 at 3:00 PM <pbhagavatula@marvell.com> wrote:
>
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>
> In traditional event programming model, events are identified by a
> flow-id and a uintptr_t. The flow-id uniquely identifies a given event
> and determines the order of scheduling based on schedule type, the
> uintptr_t holds a single object.
>
> Event devices also support burst mode with configurable dequeue depth,
> i.e. each dequeue call would return multiple events and each event
> might be at a different stage of the pipeline.
> Having a burst of events belonging to different stages in a dequeue
> burst is not only difficult to vectorize but also increases the scheduler
> overhead and application overhead of pipelining events further.
> Using event vectors we see a performance gain of ~742.3% as shown in [1].
>
> By introducing event vectorization, each event will be capable of holding
> multiple uintptr_t of the same flow thereby allowing applications
> to vectorize their pipeline and reduce the complexity of pipelining
> events across multiple stages. This also reduces the complexity of handling
> enqueue and dequeue on an event device.
>
> Since event devices are transparent to the events they are scheduling
> so the event producers such as eth_rx_adapter, crypto_adapter , etc..
> are responsible for vectorizing the buffers of the same flow into a single
> event.
>
> The series also breaks ABI in the patch [8/8] which is targetted to the
> v21.11 release.
>
> The dpdk-test-eventdev application has been updated with options to test
> multiple vector sizes and timeouts.
>
> [1]
> As for performance improvement, with a ARM Cortex-A72 equivalent processer,
> software event device (--vdev=event_sw0), single worker core, single stage
> and using one service core for Rx adapter, Tx adapter, Scheduling.
>
> Without this patchset applied:
>     ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
>          --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
>          --stlist=a --wlcores=20
>     Port[0] using Rx adapter[0] configured
>     Port[0] using Tx adapter[0] Configured
>     5.071 mpps
>
> With the patchset applied and Without event vectorization:
>     ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
>          --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
>          --stlist=a --wlcores=20
>     Port[0] using Rx adapter[0] configured
>     Port[0] using Tx adapter[0] Configured
>     5.123 mpps
>
> With event vectorization:
>     ./build/app/dpdk-test-eventdev -l 7-23 -s 0x700 --vdev="event_sw0" --
>         --prod_type_ethdev --nb_pkts=0 --verbose 2 --test=pipeline_queue
>         --stlist=a --wlcores=20 --enable_vector --nb_eth_queues 1
>         --vector_size 256
>     Port[0] using Rx adapter[0] configured
>     Port[0] using Tx adapter[0] Configured
>     42.715 mpps
>
> Having dedicated service cores for each Rx queues and tweaking the vector,
> dequeue burst size would further improve performance.
>
> API usage is shown below:
>
> Configuration:
>
>         struct rte_event_eth_rx_adapter_event_vector_config vec_conf;
>
>         vector_pool = rte_event_vector_pool_create("vector_pool",
>                         nb_elem, 0, vector_size, socket_id);
>
>         rte_event_eth_rx_adapter_create(id, event_id, &adptr_conf);
>         rte_event_eth_rx_adapter_queue_add(id, eth_id, -1, &queue_conf);
>         if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
>                 vec_conf.vector_sz = vector_size;
>                 vec_conf.vector_timeout_ns = vector_tmo_nsec;
>                 vec_conf.vector_mp = vector_pool;
>                 rte_event_eth_rx_adapter_queue_event_vector_config(id,
>                                 eth_id, -1, &vec_conf);
>         }
>
> Fastpath:
>
>         num = rte_event_dequeue_burst(event_id, port_id, &ev, 1, 0);
>         if (!num)
>                 continue;
>
>         if (ev.event_type & RTE_EVENT_TYPE_VECTOR) {
>                 switch (ev.event_type) {
>                 case RTE_EVENT_TYPE_ETHDEV_VECTOR:
>                 case RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR:
>                         struct rte_mbuf **mbufs;
>
>                         mbufs = ev.vector_ev->mbufs;
>                         for (i = 0; i < ev.vector_ev->nb_elem; i++)
>                                 //Process mbufs.
>                         break;
>                 case ...
>                 }
>         }
>         ...
>




Series applied to dpdk-next-net-eventdev/for-main. Thanks



> v10 Changes:
> - Update Rx adapter documentation with flow identifier bitfield format. (Jay)
>
> v9 Changes:
> - Update Rx adapter documentation w.r.t SW event vectorizations. (Jay)
> - Push partial vectors to event device on queue delete. (Jay)
>
> v8 Changes:
> - Fix incorrect shift for vector timeout interval.(Jay)
> - Code reallocation.(Jay)
>
> v7 Changes:
> - More doxygen fixes.(Jay)
> - Reduce code duplication in 4/8.(Jay)
>
> v6 Changes:
> - Make rte_errno sign consistant.(Jay)
> - Gramatical and doxygen fixes. (Jay)
>
> v5 Changes:
> - Make `rte_event_vector_pool_create non-inline` to ease ABI stability.(Ray)
> - Move `rte_event_eth_rx_adapter_queue_event_vector_config` and
>   `rte_event_eth_rx_adapter_vector_limits_get` implementation to the patch
>   where they are initially defined.(Ray)
> - Multiple gramatical and style fixes.(Jerin)
> - Add missing release notes.(Jerin)
>
> v4 Changes:
> - Fix missing event vector structure in event structure.(Jay)
>
> v3 Changes:
> - Fix unintended formatting changes.
>
> v2 Changes:
> - Multiple gramatical and style fixes.(Jerin)
> - Add parameter to define vector size in power of 2. (Jerin)
> - Redo patch series w/o breaking ABI till the last patch.(David)
> - Add deprication notice to announce ABI break in 21.11.(David)
> - Add vector limits validation to app/test-eventdev.
>
> Pavan Nikhilesh (8):
>   eventdev: introduce event vector capability
>   eventdev: introduce event vector Rx capability
>   eventdev: introduce event vector Tx capability
>   eventdev: add Rx adapter event vector support
>   eventdev: add Tx adapter event vector support
>   app/eventdev: add event vector mode in pipeline test
>   doc: announce event Rx adapter config changes
>   eventdev: simplify Rx adapter event vector config
>
>  app/test-eventdev/evt_common.h                |   4 +
>  app/test-eventdev/evt_options.c               |  52 +++
>  app/test-eventdev/evt_options.h               |   4 +
>  app/test-eventdev/test_pipeline_atq.c         | 310 +++++++++++++++--
>  app/test-eventdev/test_pipeline_common.c      | 105 +++++-
>  app/test-eventdev/test_pipeline_common.h      |  18 +
>  app/test-eventdev/test_pipeline_queue.c       | 320 +++++++++++++++--
>  .../prog_guide/event_ethernet_rx_adapter.rst  |  57 +++
>  .../prog_guide/event_ethernet_tx_adapter.rst  |  12 +
>  doc/guides/prog_guide/eventdev.rst            |  36 +-
>  doc/guides/rel_notes/deprecation.rst          |   9 +
>  doc/guides/rel_notes/release_21_05.rst        |   8 +
>  doc/guides/tools/testeventdev.rst             |  45 ++-
>  lib/librte_eventdev/eventdev_pmd.h            |  31 +-
>  .../rte_event_eth_rx_adapter.c                | 327 +++++++++++++++++-
>  .../rte_event_eth_rx_adapter.h                |  78 +++++
>  .../rte_event_eth_tx_adapter.c                |  66 +++-
>  lib/librte_eventdev/rte_eventdev.c            |  53 ++-
>  lib/librte_eventdev/rte_eventdev.h            | 114 +++++-
>  lib/librte_eventdev/version.map               |   4 +
>  20 files changed, 1562 insertions(+), 91 deletions(-)
>
> --
> 2.17.1
>

^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v11] eventdev: simplify Rx adapter event vector config
  2021-03-31  9:30                   ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v10 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
@ 2021-08-18  4:56                     ` pbhagavatula
  2021-08-18  4:59                       ` [dpdk-dev] [PATCH v12] " pbhagavatula
  0 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-08-18  4:56 UTC (permalink / raw)
  To: jerinj, Ray Kinsella, Jay Jayatheerthan; +Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Include vector configuration into the structure
``rte_event_eth_rx_adapter_queue_conf`` used when configuring rest
of the Rx adapter ethernet device Rx queue parameters.
This simplifies event vector configuration as it avoids splitting
configuration per Rx queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
Change-Id: I6bbb90c956d82a9a676ede567db3fea547276c44
---
 app/test-eventdev/test_pipeline_common.c |  16 +-
 doc/guides/rel_notes/deprecation.rst     |   9 --
 lib/eventdev/eventdev_pmd.h              |  29 ----
 lib/eventdev/rte_event_eth_rx_adapter.c  | 179 ++++++++---------------
 lib/eventdev/rte_event_eth_rx_adapter.h  |  30 ----
 lib/eventdev/version.map                 |   1 -
 6 files changed, 63 insertions(+), 201 deletions(-)

diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index 6ee530d4cd..2697547641 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -332,7 +332,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 	uint16_t prod;
 	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
-	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;
 
 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
@@ -398,8 +397,12 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			}
 
 			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.vector_sz = opt->vector_size;
+				queue_conf.vector_timeout_ns =
+					opt->vector_tmo_nsec;
 				queue_conf.rx_queue_flags |=
 				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+				queue_conf.vector_mp = vector_pool;
 			} else {
 				evt_err("Rx adapter doesn't support event vector");
 				return -EINVAL;
@@ -419,17 +422,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}
 
-		if (opt->ena_vector) {
-			vec_conf.vector_sz = opt->vector_size;
-			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
-			vec_conf.vector_mp = vector_pool;
-			if (rte_event_eth_rx_adapter_queue_event_vector_config(
-				    prod, prod, -1, &vec_conf) < 0) {
-				evt_err("Failed to configure event vectorization for Rx adapter");
-				return -EINVAL;
-			}
-		}
-
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;
 
diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst
index 76a4abfd6b..2c37d7222c 100644
--- a/doc/guides/rel_notes/deprecation.rst
+++ b/doc/guides/rel_notes/deprecation.rst
@@ -257,15 +257,6 @@ Deprecation Notices
   An 8-byte reserved field will be added to the structure ``rte_event_timer`` to
   support future extensions.
 
-* eventdev: The structure ``rte_event_eth_rx_adapter_queue_conf`` will be
-  extended to include ``rte_event_eth_rx_adapter_event_vector_config`` elements
-  and the function ``rte_event_eth_rx_adapter_queue_event_vector_config`` will
-  be removed in DPDK 21.11.
-
-  An application can enable event vectorization by passing the desired vector
-  values to the function ``rte_event_eth_rx_adapter_queue_add`` using
-  the structure ``rte_event_eth_rx_adapter_queue_add``.
-
 * eventdev: Reserved bytes of ``rte_event_crypto_request`` is a space holder
   for ``response_info``. Both should be decoupled for better clarity.
   New space for ``response_info`` can be made by changing
diff --git a/lib/eventdev/eventdev_pmd.h b/lib/eventdev/eventdev_pmd.h
index 0f724ac85d..63b3bc4b51 100644
--- a/lib/eventdev/eventdev_pmd.h
+++ b/lib/eventdev/eventdev_pmd.h
@@ -667,32 +667,6 @@ typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
 	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);
 
-struct rte_event_eth_rx_adapter_event_vector_config;
-/**
- * Enable event vector on an given Rx queue of a ethernet devices belonging to
- * the Rx adapter.
- *
- * @param dev
- *   Event device pointer
- *
- * @param eth_dev
- *   Ethernet device pointer
- *
- * @param rx_queue_id
- *   The Rx queue identifier
- *
- * @param config
- *   Pointer to the event vector configuration structure.
- *
- * @return
- *   - 0: Success.
- *   - <0: Error code returned by the driver function.
- */
-typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
-	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
-	int32_t rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;
 
@@ -1118,9 +1092,6 @@ struct rte_eventdev_ops {
 	eventdev_eth_rx_adapter_vector_limits_get_t
 		eth_rx_adapter_vector_limits_get;
 	/**< Get event vector limits for the Rx adapter */
-	eventdev_eth_rx_adapter_event_vector_config_t
-		eth_rx_adapter_event_vector_config;
-	/**< Configure Rx adapter with event vector */
 
 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/eventdev/rte_event_eth_rx_adapter.c b/lib/eventdev/rte_event_eth_rx_adapter.c
index 13dfb28401..2b2dd688fc 100644
--- a/lib/eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/eventdev/rte_event_eth_rx_adapter.c
@@ -1895,6 +1895,24 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	} else
 		qi_ev->flow_id = 0;
 
+	if (conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+		queue_info->ena_vector = 1;
+		qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+		rxa_set_vector_data(queue_info, conf->vector_sz,
+				    conf->vector_timeout_ns, conf->vector_mp,
+				    rx_queue_id, dev_info->dev->data->port_id);
+		rx_adapter->ena_vector = 1;
+		rx_adapter->vector_tmo_ticks =
+			rx_adapter->vector_tmo_ticks ?
+				      RTE_MIN(queue_info->vector_data
+							.vector_timeout_ticks >>
+						1,
+					rx_adapter->vector_tmo_ticks) :
+				queue_info->vector_data.vector_timeout_ticks >>
+					1;
+	}
+
 	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
 	if (rxa_polled_queue(dev_info, rx_queue_id)) {
 		rx_adapter->num_rx_polled += !pollq;
@@ -1920,42 +1938,6 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }
 
-static void
-rxa_sw_event_vector_configure(
-	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
-	int rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
-	struct eth_rx_queue_info *queue_info;
-	struct rte_event *qi_ev;
-
-	if (rx_queue_id == -1) {
-		uint16_t nb_rx_queues;
-		uint16_t i;
-
-		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
-		for (i = 0; i < nb_rx_queues; i++)
-			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
-						      config);
-		return;
-	}
-
-	queue_info = &dev_info->rx_queue[rx_queue_id];
-	qi_ev = (struct rte_event *)&queue_info->event;
-	queue_info->ena_vector = 1;
-	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
-	rxa_set_vector_data(queue_info, config->vector_sz,
-			    config->vector_timeout_ns, config->vector_mp,
-			    rx_queue_id, dev_info->dev->data->port_id);
-	rx_adapter->ena_vector = 1;
-	rx_adapter->vector_tmo_ticks =
-		rx_adapter->vector_tmo_ticks ?
-			      RTE_MIN(config->vector_timeout_ns >> 1,
-				rx_adapter->vector_tmo_ticks) :
-			      config->vector_timeout_ns >> 1;
-}
-
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -2270,6 +2252,7 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	struct rte_event_eth_rx_adapter *rx_adapter;
 	struct rte_eventdev *dev;
 	struct eth_device_info *dev_info;
+	struct rte_event_eth_rx_adapter_vector_limits limits;
 
 	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
 	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
@@ -2297,13 +2280,46 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}
 
-	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
-	    (queue_conf->rx_queue_flags &
-	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
-		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
+	if (queue_conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+
+		if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0) {
+			RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+
+		ret = rte_event_eth_rx_adapter_vector_limits_get(
+			rx_adapter->eventdev_id, eth_dev_id, &limits);
+		if (ret < 0) {
+			RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_sz < limits.min_sz ||
+		    queue_conf->vector_sz > limits.max_sz ||
+		    queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
+		    queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
+		    queue_conf->vector_mp == NULL) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_mp->elt_size <
+		    (sizeof(struct rte_event_vector) +
+		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
 	}
 
 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
@@ -2499,83 +2515,6 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }
 
-int
-rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct rte_event_eth_rx_adapter_vector_limits limits;
-	struct rte_event_eth_rx_adapter *rx_adapter;
-	struct rte_eventdev *dev;
-	uint32_t cap;
-	int ret;
-
-	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
-	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
-
-	rx_adapter = rxa_id_to_adapter(id);
-	if ((rx_adapter == NULL) || (config == NULL))
-		return -EINVAL;
-
-	dev = &rte_eventdevs[rx_adapter->eventdev_id];
-	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
-						eth_dev_id, &cap);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
-				 "eth port %" PRIu16,
-				 id, eth_dev_id);
-		return ret;
-	}
-
-	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
-		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	ret = rte_event_eth_rx_adapter_vector_limits_get(
-		rx_adapter->eventdev_id, eth_dev_id, &limits);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
-				 "eth port %" PRIu16,
-				 rx_adapter->eventdev_id, eth_dev_id);
-		return ret;
-	}
-
-	if (config->vector_sz < limits.min_sz ||
-	    config->vector_sz > limits.max_sz ||
-	    config->vector_timeout_ns < limits.min_timeout_ns ||
-	    config->vector_timeout_ns > limits.max_timeout_ns ||
-	    config->vector_mp == NULL) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-	if (config->vector_mp->elt_size <
-	    (sizeof(struct rte_event_vector) +
-	     (sizeof(uintptr_t) * config->vector_sz))) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
-		RTE_FUNC_PTR_OR_ERR_RET(
-			*dev->dev_ops->eth_rx_adapter_event_vector_config,
-			-ENOTSUP);
-		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
-			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
-	} else {
-		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
-					      rx_queue_id, config);
-	}
-
-	return ret;
-}
-
 int
 rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
diff --git a/lib/eventdev/rte_event_eth_rx_adapter.h b/lib/eventdev/rte_event_eth_rx_adapter.h
index 182dd2e5dd..3f8b362295 100644
--- a/lib/eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/eventdev/rte_event_eth_rx_adapter.h
@@ -171,9 +171,6 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
 	 * enqueued event.
 	 */
-};
-
-struct rte_event_eth_rx_adapter_event_vector_config {
 	uint16_t vector_sz;
 	/**<
 	 * Indicates the maximum number for mbufs to combine and form a vector.
@@ -548,33 +545,6 @@ int rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);
 
-/**
- * Configure event vectorization for a given ethernet device queue, that has
- * been added to a event eth Rx adapter.
- *
- * @param id
- *  The identifier of the ethernet Rx event adapter.
- *
- * @param eth_dev_id
- *  The identifier of the ethernet device.
- *
- * @param rx_queue_id
- *  Ethernet device receive queue index.
- *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
- *  are configured with event vectorization.
- *
- * @param config
- *  Event vector configuration structure.
- *
- * @return
- *  - 0: Success, Receive queue configured correctly.
- *  - <0: Error code on failure.
- */
-__rte_experimental
-int rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/eventdev/version.map b/lib/eventdev/version.map
index 88625621ec..cd86d2d908 100644
--- a/lib/eventdev/version.map
+++ b/lib/eventdev/version.map
@@ -142,7 +142,6 @@ EXPERIMENTAL {
 	#added in 21.05
 	rte_event_vector_pool_create;
 	rte_event_eth_rx_adapter_vector_limits_get;
-	rte_event_eth_rx_adapter_queue_event_vector_config;
 	__rte_eventdev_trace_crypto_adapter_enqueue;
 };
 
-- 
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v12] eventdev: simplify Rx adapter event vector config
  2021-08-18  4:56                     ` [dpdk-dev] [PATCH v11] " pbhagavatula
@ 2021-08-18  4:59                       ` pbhagavatula
  2021-08-18  6:57                         ` [dpdk-dev] [PATCH v13] " pbhagavatula
  0 siblings, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-08-18  4:59 UTC (permalink / raw)
  To: jerinj, Ray Kinsella, Jay Jayatheerthan; +Cc: dev, Pavan Nikhilesh

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Include vector configuration into the structure
``rte_event_eth_rx_adapter_queue_conf`` used when configuring rest
of the Rx adapter ethernet device Rx queue parameters.
This simplifies event vector configuration as it avoids splitting
configuration per Rx queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
---
 v12 Changes:
 - Remove deprication notice.
 - Remove unnecessary change Id.

 app/test-eventdev/test_pipeline_common.c |  16 +-
 doc/guides/rel_notes/deprecation.rst     |   9 --
 lib/eventdev/eventdev_pmd.h              |  29 ----
 lib/eventdev/rte_event_eth_rx_adapter.c  | 179 ++++++++---------------
 lib/eventdev/rte_event_eth_rx_adapter.h  |  30 ----
 lib/eventdev/version.map                 |   1 -
 6 files changed, 63 insertions(+), 201 deletions(-)

diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index 6ee530d4cd..2697547641 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -332,7 +332,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 	uint16_t prod;
 	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
-	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
@@ -398,8 +397,12 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			}

 			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.vector_sz = opt->vector_size;
+				queue_conf.vector_timeout_ns =
+					opt->vector_tmo_nsec;
 				queue_conf.rx_queue_flags |=
 				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+				queue_conf.vector_mp = vector_pool;
 			} else {
 				evt_err("Rx adapter doesn't support event vector");
 				return -EINVAL;
@@ -419,17 +422,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}

-		if (opt->ena_vector) {
-			vec_conf.vector_sz = opt->vector_size;
-			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
-			vec_conf.vector_mp = vector_pool;
-			if (rte_event_eth_rx_adapter_queue_event_vector_config(
-				    prod, prod, -1, &vec_conf) < 0) {
-				evt_err("Failed to configure event vectorization for Rx adapter");
-				return -EINVAL;
-			}
-		}
-
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;

diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst
index 76a4abfd6b..2c37d7222c 100644
--- a/doc/guides/rel_notes/deprecation.rst
+++ b/doc/guides/rel_notes/deprecation.rst
@@ -257,15 +257,6 @@ Deprecation Notices
   An 8-byte reserved field will be added to the structure ``rte_event_timer`` to
   support future extensions.

-* eventdev: The structure ``rte_event_eth_rx_adapter_queue_conf`` will be
-  extended to include ``rte_event_eth_rx_adapter_event_vector_config`` elements
-  and the function ``rte_event_eth_rx_adapter_queue_event_vector_config`` will
-  be removed in DPDK 21.11.
-
-  An application can enable event vectorization by passing the desired vector
-  values to the function ``rte_event_eth_rx_adapter_queue_add`` using
-  the structure ``rte_event_eth_rx_adapter_queue_add``.
-
 * eventdev: Reserved bytes of ``rte_event_crypto_request`` is a space holder
   for ``response_info``. Both should be decoupled for better clarity.
   New space for ``response_info`` can be made by changing
diff --git a/lib/eventdev/eventdev_pmd.h b/lib/eventdev/eventdev_pmd.h
index 0f724ac85d..63b3bc4b51 100644
--- a/lib/eventdev/eventdev_pmd.h
+++ b/lib/eventdev/eventdev_pmd.h
@@ -667,32 +667,6 @@ typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
 	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-struct rte_event_eth_rx_adapter_event_vector_config;
-/**
- * Enable event vector on an given Rx queue of a ethernet devices belonging to
- * the Rx adapter.
- *
- * @param dev
- *   Event device pointer
- *
- * @param eth_dev
- *   Ethernet device pointer
- *
- * @param rx_queue_id
- *   The Rx queue identifier
- *
- * @param config
- *   Pointer to the event vector configuration structure.
- *
- * @return
- *   - 0: Success.
- *   - <0: Error code returned by the driver function.
- */
-typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
-	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
-	int32_t rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;

@@ -1118,9 +1092,6 @@ struct rte_eventdev_ops {
 	eventdev_eth_rx_adapter_vector_limits_get_t
 		eth_rx_adapter_vector_limits_get;
 	/**< Get event vector limits for the Rx adapter */
-	eventdev_eth_rx_adapter_event_vector_config_t
-		eth_rx_adapter_event_vector_config;
-	/**< Configure Rx adapter with event vector */

 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/eventdev/rte_event_eth_rx_adapter.c b/lib/eventdev/rte_event_eth_rx_adapter.c
index 13dfb28401..2b2dd688fc 100644
--- a/lib/eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/eventdev/rte_event_eth_rx_adapter.c
@@ -1895,6 +1895,24 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	} else
 		qi_ev->flow_id = 0;

+	if (conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+		queue_info->ena_vector = 1;
+		qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+		rxa_set_vector_data(queue_info, conf->vector_sz,
+				    conf->vector_timeout_ns, conf->vector_mp,
+				    rx_queue_id, dev_info->dev->data->port_id);
+		rx_adapter->ena_vector = 1;
+		rx_adapter->vector_tmo_ticks =
+			rx_adapter->vector_tmo_ticks ?
+				      RTE_MIN(queue_info->vector_data
+							.vector_timeout_ticks >>
+						1,
+					rx_adapter->vector_tmo_ticks) :
+				queue_info->vector_data.vector_timeout_ticks >>
+					1;
+	}
+
 	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
 	if (rxa_polled_queue(dev_info, rx_queue_id)) {
 		rx_adapter->num_rx_polled += !pollq;
@@ -1920,42 +1938,6 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }

-static void
-rxa_sw_event_vector_configure(
-	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
-	int rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
-	struct eth_rx_queue_info *queue_info;
-	struct rte_event *qi_ev;
-
-	if (rx_queue_id == -1) {
-		uint16_t nb_rx_queues;
-		uint16_t i;
-
-		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
-		for (i = 0; i < nb_rx_queues; i++)
-			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
-						      config);
-		return;
-	}
-
-	queue_info = &dev_info->rx_queue[rx_queue_id];
-	qi_ev = (struct rte_event *)&queue_info->event;
-	queue_info->ena_vector = 1;
-	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
-	rxa_set_vector_data(queue_info, config->vector_sz,
-			    config->vector_timeout_ns, config->vector_mp,
-			    rx_queue_id, dev_info->dev->data->port_id);
-	rx_adapter->ena_vector = 1;
-	rx_adapter->vector_tmo_ticks =
-		rx_adapter->vector_tmo_ticks ?
-			      RTE_MIN(config->vector_timeout_ns >> 1,
-				rx_adapter->vector_tmo_ticks) :
-			      config->vector_timeout_ns >> 1;
-}
-
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -2270,6 +2252,7 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	struct rte_event_eth_rx_adapter *rx_adapter;
 	struct rte_eventdev *dev;
 	struct eth_device_info *dev_info;
+	struct rte_event_eth_rx_adapter_vector_limits limits;

 	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
 	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
@@ -2297,13 +2280,46 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}

-	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
-	    (queue_conf->rx_queue_flags &
-	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
-		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
+	if (queue_conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+
+		if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0) {
+			RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+
+		ret = rte_event_eth_rx_adapter_vector_limits_get(
+			rx_adapter->eventdev_id, eth_dev_id, &limits);
+		if (ret < 0) {
+			RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_sz < limits.min_sz ||
+		    queue_conf->vector_sz > limits.max_sz ||
+		    queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
+		    queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
+		    queue_conf->vector_mp == NULL) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_mp->elt_size <
+		    (sizeof(struct rte_event_vector) +
+		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
 	}

 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
@@ -2499,83 +2515,6 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }

-int
-rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct rte_event_eth_rx_adapter_vector_limits limits;
-	struct rte_event_eth_rx_adapter *rx_adapter;
-	struct rte_eventdev *dev;
-	uint32_t cap;
-	int ret;
-
-	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
-	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
-
-	rx_adapter = rxa_id_to_adapter(id);
-	if ((rx_adapter == NULL) || (config == NULL))
-		return -EINVAL;
-
-	dev = &rte_eventdevs[rx_adapter->eventdev_id];
-	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
-						eth_dev_id, &cap);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
-				 "eth port %" PRIu16,
-				 id, eth_dev_id);
-		return ret;
-	}
-
-	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
-		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	ret = rte_event_eth_rx_adapter_vector_limits_get(
-		rx_adapter->eventdev_id, eth_dev_id, &limits);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
-				 "eth port %" PRIu16,
-				 rx_adapter->eventdev_id, eth_dev_id);
-		return ret;
-	}
-
-	if (config->vector_sz < limits.min_sz ||
-	    config->vector_sz > limits.max_sz ||
-	    config->vector_timeout_ns < limits.min_timeout_ns ||
-	    config->vector_timeout_ns > limits.max_timeout_ns ||
-	    config->vector_mp == NULL) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-	if (config->vector_mp->elt_size <
-	    (sizeof(struct rte_event_vector) +
-	     (sizeof(uintptr_t) * config->vector_sz))) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
-		RTE_FUNC_PTR_OR_ERR_RET(
-			*dev->dev_ops->eth_rx_adapter_event_vector_config,
-			-ENOTSUP);
-		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
-			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
-	} else {
-		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
-					      rx_queue_id, config);
-	}
-
-	return ret;
-}
-
 int
 rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
diff --git a/lib/eventdev/rte_event_eth_rx_adapter.h b/lib/eventdev/rte_event_eth_rx_adapter.h
index 182dd2e5dd..3f8b362295 100644
--- a/lib/eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/eventdev/rte_event_eth_rx_adapter.h
@@ -171,9 +171,6 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
 	 * enqueued event.
 	 */
-};
-
-struct rte_event_eth_rx_adapter_event_vector_config {
 	uint16_t vector_sz;
 	/**<
 	 * Indicates the maximum number for mbufs to combine and form a vector.
@@ -548,33 +545,6 @@ int rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-/**
- * Configure event vectorization for a given ethernet device queue, that has
- * been added to a event eth Rx adapter.
- *
- * @param id
- *  The identifier of the ethernet Rx event adapter.
- *
- * @param eth_dev_id
- *  The identifier of the ethernet device.
- *
- * @param rx_queue_id
- *  Ethernet device receive queue index.
- *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
- *  are configured with event vectorization.
- *
- * @param config
- *  Event vector configuration structure.
- *
- * @return
- *  - 0: Success, Receive queue configured correctly.
- *  - <0: Error code on failure.
- */
-__rte_experimental
-int rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/eventdev/version.map b/lib/eventdev/version.map
index 88625621ec..cd86d2d908 100644
--- a/lib/eventdev/version.map
+++ b/lib/eventdev/version.map
@@ -142,7 +142,6 @@ EXPERIMENTAL {
 	#added in 21.05
 	rte_event_vector_pool_create;
 	rte_event_eth_rx_adapter_vector_limits_get;
-	rte_event_eth_rx_adapter_queue_event_vector_config;
 	__rte_eventdev_trace_crypto_adapter_enqueue;
 };

--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v13] eventdev: simplify Rx adapter event vector config
  2021-08-18  4:59                       ` [dpdk-dev] [PATCH v12] " pbhagavatula
@ 2021-08-18  6:57                         ` pbhagavatula
  2021-08-18  8:22                           ` Jayatheerthan, Jay
  2021-09-15 13:15                           ` [dpdk-dev] [PATCH v14] " pbhagavatula
  0 siblings, 2 replies; 153+ messages in thread
From: pbhagavatula @ 2021-08-18  6:57 UTC (permalink / raw)
  To: jerinj, Ray Kinsella, Pavan Nikhilesh, Shijith Thotton,
	Jay Jayatheerthan
  Cc: dev

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Include vector configuration into the structure
``rte_event_eth_rx_adapter_queue_conf`` that is used to configure
Rx adapter ethernet device Rx queue parameters.
This simplifies event vector configuration as it avoids splitting
configuration per Rx queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
---
 v13 Changes:
 - Fix cnxk driver compilation.
 v12 Changes:
 - Remove deprication notice.
 - Remove unnecessary change Id.

 app/test-eventdev/test_pipeline_common.c |  16 +-
 doc/guides/rel_notes/deprecation.rst     |   9 --
 drivers/event/cnxk/cn10k_eventdev.c      |  77 ----------
 drivers/event/cnxk/cnxk_eventdev_adptr.c |  41 ++++++
 lib/eventdev/eventdev_pmd.h              |  29 ----
 lib/eventdev/rte_event_eth_rx_adapter.c  | 179 ++++++++---------------
 lib/eventdev/rte_event_eth_rx_adapter.h  |  30 ----
 lib/eventdev/version.map                 |   1 -
 8 files changed, 104 insertions(+), 278 deletions(-)

diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index 6ee530d4cd..2697547641 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -332,7 +332,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 	uint16_t prod;
 	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
-	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
@@ -398,8 +397,12 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			}

 			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.vector_sz = opt->vector_size;
+				queue_conf.vector_timeout_ns =
+					opt->vector_tmo_nsec;
 				queue_conf.rx_queue_flags |=
 				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+				queue_conf.vector_mp = vector_pool;
 			} else {
 				evt_err("Rx adapter doesn't support event vector");
 				return -EINVAL;
@@ -419,17 +422,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}

-		if (opt->ena_vector) {
-			vec_conf.vector_sz = opt->vector_size;
-			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
-			vec_conf.vector_mp = vector_pool;
-			if (rte_event_eth_rx_adapter_queue_event_vector_config(
-				    prod, prod, -1, &vec_conf) < 0) {
-				evt_err("Failed to configure event vectorization for Rx adapter");
-				return -EINVAL;
-			}
-		}
-
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;

diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst
index 76a4abfd6b..2c37d7222c 100644
--- a/doc/guides/rel_notes/deprecation.rst
+++ b/doc/guides/rel_notes/deprecation.rst
@@ -257,15 +257,6 @@ Deprecation Notices
   An 8-byte reserved field will be added to the structure ``rte_event_timer`` to
   support future extensions.

-* eventdev: The structure ``rte_event_eth_rx_adapter_queue_conf`` will be
-  extended to include ``rte_event_eth_rx_adapter_event_vector_config`` elements
-  and the function ``rte_event_eth_rx_adapter_queue_event_vector_config`` will
-  be removed in DPDK 21.11.
-
-  An application can enable event vectorization by passing the desired vector
-  values to the function ``rte_event_eth_rx_adapter_queue_add`` using
-  the structure ``rte_event_eth_rx_adapter_queue_add``.
-
 * eventdev: Reserved bytes of ``rte_event_crypto_request`` is a space holder
   for ``response_info``. Both should be decoupled for better clarity.
   New space for ``response_info`` can be made by changing
diff --git a/drivers/event/cnxk/cn10k_eventdev.c b/drivers/event/cnxk/cn10k_eventdev.c
index 6f37c5bd23..160192bb27 100644
--- a/drivers/event/cnxk/cn10k_eventdev.c
+++ b/drivers/event/cnxk/cn10k_eventdev.c
@@ -696,81 +696,6 @@ cn10k_sso_rx_adapter_vector_limits(
 	return 0;
 }

-static int
-cnxk_sso_rx_adapter_vwqe_enable(struct cnxk_eth_dev *cnxk_eth_dev,
-				uint16_t port_id, uint16_t rq_id, uint16_t sz,
-				uint64_t tmo_ns, struct rte_mempool *vmp)
-{
-	struct roc_nix_rq *rq;
-
-	rq = &cnxk_eth_dev->rqs[rq_id];
-
-	if (!rq->sso_ena)
-		return -EINVAL;
-	if (rq->flow_tag_width == 0)
-		return -EINVAL;
-
-	rq->vwqe_ena = 1;
-	rq->vwqe_first_skip = 0;
-	rq->vwqe_aura_handle = roc_npa_aura_handle_to_aura(vmp->pool_id);
-	rq->vwqe_max_sz_exp = rte_log2_u32(sz);
-	rq->vwqe_wait_tmo =
-		tmo_ns /
-		((roc_nix_get_vwqe_interval(&cnxk_eth_dev->nix) + 1) * 100);
-	rq->tag_mask = (port_id & 0xF) << 20;
-	rq->tag_mask |=
-		(((port_id >> 4) & 0xF) | (RTE_EVENT_TYPE_ETHDEV_VECTOR << 4))
-		<< 24;
-
-	return roc_nix_rq_modify(&cnxk_eth_dev->nix, rq, 0);
-}
-
-static int
-cn10k_sso_rx_adapter_vector_config(
-	const struct rte_eventdev *event_dev, const struct rte_eth_dev *eth_dev,
-	int32_t rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct cnxk_eth_dev *cnxk_eth_dev;
-	struct cnxk_sso_evdev *dev;
-	int i, rc;
-
-	rc = strncmp(eth_dev->device->driver->name, "net_cn10k", 8);
-	if (rc)
-		return -EINVAL;
-
-	dev = cnxk_sso_pmd_priv(event_dev);
-	cnxk_eth_dev = cnxk_eth_pmd_priv(eth_dev);
-	if (rx_queue_id < 0) {
-		for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
-			cnxk_sso_updt_xae_cnt(dev, config->vector_mp,
-					      RTE_EVENT_TYPE_ETHDEV_VECTOR);
-			rc = cnxk_sso_xae_reconfigure(
-				(struct rte_eventdev *)(uintptr_t)event_dev);
-			rc = cnxk_sso_rx_adapter_vwqe_enable(
-				cnxk_eth_dev, eth_dev->data->port_id, i,
-				config->vector_sz, config->vector_timeout_ns,
-				config->vector_mp);
-			if (rc)
-				return -EINVAL;
-		}
-	} else {
-
-		cnxk_sso_updt_xae_cnt(dev, config->vector_mp,
-				      RTE_EVENT_TYPE_ETHDEV_VECTOR);
-		rc = cnxk_sso_xae_reconfigure(
-			(struct rte_eventdev *)(uintptr_t)event_dev);
-		rc = cnxk_sso_rx_adapter_vwqe_enable(
-			cnxk_eth_dev, eth_dev->data->port_id, rx_queue_id,
-			config->vector_sz, config->vector_timeout_ns,
-			config->vector_mp);
-		if (rc)
-			return -EINVAL;
-	}
-
-	return 0;
-}
-
 static int
 cn10k_sso_tx_adapter_caps_get(const struct rte_eventdev *dev,
 			      const struct rte_eth_dev *eth_dev, uint32_t *caps)
@@ -841,8 +766,6 @@ static struct rte_eventdev_ops cn10k_sso_dev_ops = {
 	.eth_rx_adapter_stop = cnxk_sso_rx_adapter_stop,

 	.eth_rx_adapter_vector_limits_get = cn10k_sso_rx_adapter_vector_limits,
-	.eth_rx_adapter_event_vector_config =
-		cn10k_sso_rx_adapter_vector_config,

 	.eth_tx_adapter_caps_get = cn10k_sso_tx_adapter_caps_get,
 	.eth_tx_adapter_queue_add = cn10k_sso_tx_adapter_queue_add,
diff --git a/drivers/event/cnxk/cnxk_eventdev_adptr.c b/drivers/event/cnxk/cnxk_eventdev_adptr.c
index baf2f2aa6b..80f5602286 100644
--- a/drivers/event/cnxk/cnxk_eventdev_adptr.c
+++ b/drivers/event/cnxk/cnxk_eventdev_adptr.c
@@ -156,6 +156,35 @@ cnxk_sso_rxq_disable(struct cnxk_eth_dev *cnxk_eth_dev, uint16_t rq_id)
 	return roc_nix_rq_modify(&cnxk_eth_dev->nix, rq, 0);
 }

+static int
+cnxk_sso_rx_adapter_vwqe_enable(struct cnxk_eth_dev *cnxk_eth_dev,
+				uint16_t port_id, uint16_t rq_id, uint16_t sz,
+				uint64_t tmo_ns, struct rte_mempool *vmp)
+{
+	struct roc_nix_rq *rq;
+
+	rq = &cnxk_eth_dev->rqs[rq_id];
+
+	if (!rq->sso_ena)
+		return -EINVAL;
+	if (rq->flow_tag_width == 0)
+		return -EINVAL;
+
+	rq->vwqe_ena = 1;
+	rq->vwqe_first_skip = 0;
+	rq->vwqe_aura_handle = roc_npa_aura_handle_to_aura(vmp->pool_id);
+	rq->vwqe_max_sz_exp = rte_log2_u32(sz);
+	rq->vwqe_wait_tmo =
+		tmo_ns /
+		((roc_nix_get_vwqe_interval(&cnxk_eth_dev->nix) + 1) * 100);
+	rq->tag_mask = (port_id & 0xF) << 20;
+	rq->tag_mask |=
+		(((port_id >> 4) & 0xF) | (RTE_EVENT_TYPE_ETHDEV_VECTOR << 4))
+		<< 24;
+
+	return roc_nix_rq_modify(&cnxk_eth_dev->nix, rq, 0);
+}
+
 int
 cnxk_sso_rx_adapter_queue_add(
 	const struct rte_eventdev *event_dev, const struct rte_eth_dev *eth_dev,
@@ -183,6 +212,18 @@ cnxk_sso_rx_adapter_queue_add(
 			&queue_conf->ev,
 			!!(queue_conf->rx_queue_flags &
 			   RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID));
+		if (queue_conf->rx_queue_flags &
+		    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+			cnxk_sso_updt_xae_cnt(dev, queue_conf->vector_mp,
+					      RTE_EVENT_TYPE_ETHDEV_VECTOR);
+			rc |= cnxk_sso_xae_reconfigure(
+				(struct rte_eventdev *)(uintptr_t)event_dev);
+			rc |= cnxk_sso_rx_adapter_vwqe_enable(
+				cnxk_eth_dev, port, rx_queue_id,
+				queue_conf->vector_sz,
+				queue_conf->vector_timeout_ns,
+				queue_conf->vector_mp);
+		}
 		rox_nix_fc_npa_bp_cfg(&cnxk_eth_dev->nix,
 				      rxq_sp->qconf.mp->pool_id, true,
 				      dev->force_ena_bp);
diff --git a/lib/eventdev/eventdev_pmd.h b/lib/eventdev/eventdev_pmd.h
index 0f724ac85d..63b3bc4b51 100644
--- a/lib/eventdev/eventdev_pmd.h
+++ b/lib/eventdev/eventdev_pmd.h
@@ -667,32 +667,6 @@ typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
 	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-struct rte_event_eth_rx_adapter_event_vector_config;
-/**
- * Enable event vector on an given Rx queue of a ethernet devices belonging to
- * the Rx adapter.
- *
- * @param dev
- *   Event device pointer
- *
- * @param eth_dev
- *   Ethernet device pointer
- *
- * @param rx_queue_id
- *   The Rx queue identifier
- *
- * @param config
- *   Pointer to the event vector configuration structure.
- *
- * @return
- *   - 0: Success.
- *   - <0: Error code returned by the driver function.
- */
-typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
-	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
-	int32_t rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;

@@ -1118,9 +1092,6 @@ struct rte_eventdev_ops {
 	eventdev_eth_rx_adapter_vector_limits_get_t
 		eth_rx_adapter_vector_limits_get;
 	/**< Get event vector limits for the Rx adapter */
-	eventdev_eth_rx_adapter_event_vector_config_t
-		eth_rx_adapter_event_vector_config;
-	/**< Configure Rx adapter with event vector */

 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/eventdev/rte_event_eth_rx_adapter.c b/lib/eventdev/rte_event_eth_rx_adapter.c
index 13dfb28401..2b2dd688fc 100644
--- a/lib/eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/eventdev/rte_event_eth_rx_adapter.c
@@ -1895,6 +1895,24 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	} else
 		qi_ev->flow_id = 0;

+	if (conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+		queue_info->ena_vector = 1;
+		qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+		rxa_set_vector_data(queue_info, conf->vector_sz,
+				    conf->vector_timeout_ns, conf->vector_mp,
+				    rx_queue_id, dev_info->dev->data->port_id);
+		rx_adapter->ena_vector = 1;
+		rx_adapter->vector_tmo_ticks =
+			rx_adapter->vector_tmo_ticks ?
+				      RTE_MIN(queue_info->vector_data
+							.vector_timeout_ticks >>
+						1,
+					rx_adapter->vector_tmo_ticks) :
+				queue_info->vector_data.vector_timeout_ticks >>
+					1;
+	}
+
 	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
 	if (rxa_polled_queue(dev_info, rx_queue_id)) {
 		rx_adapter->num_rx_polled += !pollq;
@@ -1920,42 +1938,6 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }

-static void
-rxa_sw_event_vector_configure(
-	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
-	int rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
-	struct eth_rx_queue_info *queue_info;
-	struct rte_event *qi_ev;
-
-	if (rx_queue_id == -1) {
-		uint16_t nb_rx_queues;
-		uint16_t i;
-
-		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
-		for (i = 0; i < nb_rx_queues; i++)
-			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
-						      config);
-		return;
-	}
-
-	queue_info = &dev_info->rx_queue[rx_queue_id];
-	qi_ev = (struct rte_event *)&queue_info->event;
-	queue_info->ena_vector = 1;
-	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
-	rxa_set_vector_data(queue_info, config->vector_sz,
-			    config->vector_timeout_ns, config->vector_mp,
-			    rx_queue_id, dev_info->dev->data->port_id);
-	rx_adapter->ena_vector = 1;
-	rx_adapter->vector_tmo_ticks =
-		rx_adapter->vector_tmo_ticks ?
-			      RTE_MIN(config->vector_timeout_ns >> 1,
-				rx_adapter->vector_tmo_ticks) :
-			      config->vector_timeout_ns >> 1;
-}
-
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -2270,6 +2252,7 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	struct rte_event_eth_rx_adapter *rx_adapter;
 	struct rte_eventdev *dev;
 	struct eth_device_info *dev_info;
+	struct rte_event_eth_rx_adapter_vector_limits limits;

 	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
 	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
@@ -2297,13 +2280,46 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}

-	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
-	    (queue_conf->rx_queue_flags &
-	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
-		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
+	if (queue_conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+
+		if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0) {
+			RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+
+		ret = rte_event_eth_rx_adapter_vector_limits_get(
+			rx_adapter->eventdev_id, eth_dev_id, &limits);
+		if (ret < 0) {
+			RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_sz < limits.min_sz ||
+		    queue_conf->vector_sz > limits.max_sz ||
+		    queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
+		    queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
+		    queue_conf->vector_mp == NULL) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_mp->elt_size <
+		    (sizeof(struct rte_event_vector) +
+		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
 	}

 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
@@ -2499,83 +2515,6 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }

-int
-rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct rte_event_eth_rx_adapter_vector_limits limits;
-	struct rte_event_eth_rx_adapter *rx_adapter;
-	struct rte_eventdev *dev;
-	uint32_t cap;
-	int ret;
-
-	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
-	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
-
-	rx_adapter = rxa_id_to_adapter(id);
-	if ((rx_adapter == NULL) || (config == NULL))
-		return -EINVAL;
-
-	dev = &rte_eventdevs[rx_adapter->eventdev_id];
-	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
-						eth_dev_id, &cap);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
-				 "eth port %" PRIu16,
-				 id, eth_dev_id);
-		return ret;
-	}
-
-	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
-		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	ret = rte_event_eth_rx_adapter_vector_limits_get(
-		rx_adapter->eventdev_id, eth_dev_id, &limits);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
-				 "eth port %" PRIu16,
-				 rx_adapter->eventdev_id, eth_dev_id);
-		return ret;
-	}
-
-	if (config->vector_sz < limits.min_sz ||
-	    config->vector_sz > limits.max_sz ||
-	    config->vector_timeout_ns < limits.min_timeout_ns ||
-	    config->vector_timeout_ns > limits.max_timeout_ns ||
-	    config->vector_mp == NULL) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-	if (config->vector_mp->elt_size <
-	    (sizeof(struct rte_event_vector) +
-	     (sizeof(uintptr_t) * config->vector_sz))) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
-		RTE_FUNC_PTR_OR_ERR_RET(
-			*dev->dev_ops->eth_rx_adapter_event_vector_config,
-			-ENOTSUP);
-		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
-			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
-	} else {
-		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
-					      rx_queue_id, config);
-	}
-
-	return ret;
-}
-
 int
 rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
diff --git a/lib/eventdev/rte_event_eth_rx_adapter.h b/lib/eventdev/rte_event_eth_rx_adapter.h
index 182dd2e5dd..3f8b362295 100644
--- a/lib/eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/eventdev/rte_event_eth_rx_adapter.h
@@ -171,9 +171,6 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
 	 * enqueued event.
 	 */
-};
-
-struct rte_event_eth_rx_adapter_event_vector_config {
 	uint16_t vector_sz;
 	/**<
 	 * Indicates the maximum number for mbufs to combine and form a vector.
@@ -548,33 +545,6 @@ int rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-/**
- * Configure event vectorization for a given ethernet device queue, that has
- * been added to a event eth Rx adapter.
- *
- * @param id
- *  The identifier of the ethernet Rx event adapter.
- *
- * @param eth_dev_id
- *  The identifier of the ethernet device.
- *
- * @param rx_queue_id
- *  Ethernet device receive queue index.
- *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
- *  are configured with event vectorization.
- *
- * @param config
- *  Event vector configuration structure.
- *
- * @return
- *  - 0: Success, Receive queue configured correctly.
- *  - <0: Error code on failure.
- */
-__rte_experimental
-int rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/eventdev/version.map b/lib/eventdev/version.map
index 88625621ec..cd86d2d908 100644
--- a/lib/eventdev/version.map
+++ b/lib/eventdev/version.map
@@ -142,7 +142,6 @@ EXPERIMENTAL {
 	#added in 21.05
 	rte_event_vector_pool_create;
 	rte_event_eth_rx_adapter_vector_limits_get;
-	rte_event_eth_rx_adapter_queue_event_vector_config;
 	__rte_eventdev_trace_crypto_adapter_enqueue;
 };

--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v13] eventdev: simplify Rx adapter event vector config
  2021-08-18  6:57                         ` [dpdk-dev] [PATCH v13] " pbhagavatula
@ 2021-08-18  8:22                           ` Jayatheerthan, Jay
  2021-08-20  7:33                             ` Naga Harish K, S V
  2021-09-15 13:15                           ` [dpdk-dev] [PATCH v14] " pbhagavatula
  1 sibling, 1 reply; 153+ messages in thread
From: Jayatheerthan, Jay @ 2021-08-18  8:22 UTC (permalink / raw)
  To: pbhagavatula, jerinj, Ray Kinsella, Shijith Thotton, Naga Harish K, S V
  Cc: dev

HI Harish,
Could you review this patch ?

-Jay


> -----Original Message-----
> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
> Sent: Wednesday, August 18, 2021 12:27 PM
> To: jerinj@marvell.com; Ray Kinsella <mdr@ashroe.eu>; Pavan Nikhilesh <pbhagavatula@marvell.com>; Shijith Thotton
> <sthotton@marvell.com>; Jayatheerthan, Jay <jay.jayatheerthan@intel.com>
> Cc: dev@dpdk.org
> Subject: [dpdk-dev] [PATCH v13] eventdev: simplify Rx adapter event vector config
> 
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> Include vector configuration into the structure
> ``rte_event_eth_rx_adapter_queue_conf`` that is used to configure
> Rx adapter ethernet device Rx queue parameters.
> This simplifies event vector configuration as it avoids splitting
> configuration per Rx queue.
> 
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
> ---
>  v13 Changes:
>  - Fix cnxk driver compilation.
>  v12 Changes:
>  - Remove deprication notice.
>  - Remove unnecessary change Id.
> 
>  app/test-eventdev/test_pipeline_common.c |  16 +-
>  doc/guides/rel_notes/deprecation.rst     |   9 --
>  drivers/event/cnxk/cn10k_eventdev.c      |  77 ----------
>  drivers/event/cnxk/cnxk_eventdev_adptr.c |  41 ++++++
>  lib/eventdev/eventdev_pmd.h              |  29 ----
>  lib/eventdev/rte_event_eth_rx_adapter.c  | 179 ++++++++---------------
>  lib/eventdev/rte_event_eth_rx_adapter.h  |  30 ----
>  lib/eventdev/version.map                 |   1 -
>  8 files changed, 104 insertions(+), 278 deletions(-)
> 
> diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
> index 6ee530d4cd..2697547641 100644
> --- a/app/test-eventdev/test_pipeline_common.c
> +++ b/app/test-eventdev/test_pipeline_common.c
> @@ -332,7 +332,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
>  	uint16_t prod;
>  	struct rte_mempool *vector_pool = NULL;
>  	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
> -	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;
> 
>  	memset(&queue_conf, 0,
>  			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
> @@ -398,8 +397,12 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
>  			}
> 
>  			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
> +				queue_conf.vector_sz = opt->vector_size;
> +				queue_conf.vector_timeout_ns =
> +					opt->vector_tmo_nsec;
>  				queue_conf.rx_queue_flags |=
>  				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
> +				queue_conf.vector_mp = vector_pool;
>  			} else {
>  				evt_err("Rx adapter doesn't support event vector");
>  				return -EINVAL;
> @@ -419,17 +422,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
>  			return ret;
>  		}
> 
> -		if (opt->ena_vector) {
> -			vec_conf.vector_sz = opt->vector_size;
> -			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
> -			vec_conf.vector_mp = vector_pool;
> -			if (rte_event_eth_rx_adapter_queue_event_vector_config(
> -				    prod, prod, -1, &vec_conf) < 0) {
> -				evt_err("Failed to configure event vectorization for Rx adapter");
> -				return -EINVAL;
> -			}
> -		}
> -
>  		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
>  			uint32_t service_id = -1U;
> 
> diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst
> index 76a4abfd6b..2c37d7222c 100644
> --- a/doc/guides/rel_notes/deprecation.rst
> +++ b/doc/guides/rel_notes/deprecation.rst
> @@ -257,15 +257,6 @@ Deprecation Notices
>    An 8-byte reserved field will be added to the structure ``rte_event_timer`` to
>    support future extensions.
> 
> -* eventdev: The structure ``rte_event_eth_rx_adapter_queue_conf`` will be
> -  extended to include ``rte_event_eth_rx_adapter_event_vector_config`` elements
> -  and the function ``rte_event_eth_rx_adapter_queue_event_vector_config`` will
> -  be removed in DPDK 21.11.
> -
> -  An application can enable event vectorization by passing the desired vector
> -  values to the function ``rte_event_eth_rx_adapter_queue_add`` using
> -  the structure ``rte_event_eth_rx_adapter_queue_add``.
> -
>  * eventdev: Reserved bytes of ``rte_event_crypto_request`` is a space holder
>    for ``response_info``. Both should be decoupled for better clarity.
>    New space for ``response_info`` can be made by changing
> diff --git a/drivers/event/cnxk/cn10k_eventdev.c b/drivers/event/cnxk/cn10k_eventdev.c
> index 6f37c5bd23..160192bb27 100644
> --- a/drivers/event/cnxk/cn10k_eventdev.c
> +++ b/drivers/event/cnxk/cn10k_eventdev.c
> @@ -696,81 +696,6 @@ cn10k_sso_rx_adapter_vector_limits(
>  	return 0;
>  }
> 
> -static int
> -cnxk_sso_rx_adapter_vwqe_enable(struct cnxk_eth_dev *cnxk_eth_dev,
> -				uint16_t port_id, uint16_t rq_id, uint16_t sz,
> -				uint64_t tmo_ns, struct rte_mempool *vmp)
> -{
> -	struct roc_nix_rq *rq;
> -
> -	rq = &cnxk_eth_dev->rqs[rq_id];
> -
> -	if (!rq->sso_ena)
> -		return -EINVAL;
> -	if (rq->flow_tag_width == 0)
> -		return -EINVAL;
> -
> -	rq->vwqe_ena = 1;
> -	rq->vwqe_first_skip = 0;
> -	rq->vwqe_aura_handle = roc_npa_aura_handle_to_aura(vmp->pool_id);
> -	rq->vwqe_max_sz_exp = rte_log2_u32(sz);
> -	rq->vwqe_wait_tmo =
> -		tmo_ns /
> -		((roc_nix_get_vwqe_interval(&cnxk_eth_dev->nix) + 1) * 100);
> -	rq->tag_mask = (port_id & 0xF) << 20;
> -	rq->tag_mask |=
> -		(((port_id >> 4) & 0xF) | (RTE_EVENT_TYPE_ETHDEV_VECTOR << 4))
> -		<< 24;
> -
> -	return roc_nix_rq_modify(&cnxk_eth_dev->nix, rq, 0);
> -}
> -
> -static int
> -cn10k_sso_rx_adapter_vector_config(
> -	const struct rte_eventdev *event_dev, const struct rte_eth_dev *eth_dev,
> -	int32_t rx_queue_id,
> -	const struct rte_event_eth_rx_adapter_event_vector_config *config)
> -{
> -	struct cnxk_eth_dev *cnxk_eth_dev;
> -	struct cnxk_sso_evdev *dev;
> -	int i, rc;
> -
> -	rc = strncmp(eth_dev->device->driver->name, "net_cn10k", 8);
> -	if (rc)
> -		return -EINVAL;
> -
> -	dev = cnxk_sso_pmd_priv(event_dev);
> -	cnxk_eth_dev = cnxk_eth_pmd_priv(eth_dev);
> -	if (rx_queue_id < 0) {
> -		for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
> -			cnxk_sso_updt_xae_cnt(dev, config->vector_mp,
> -					      RTE_EVENT_TYPE_ETHDEV_VECTOR);
> -			rc = cnxk_sso_xae_reconfigure(
> -				(struct rte_eventdev *)(uintptr_t)event_dev);
> -			rc = cnxk_sso_rx_adapter_vwqe_enable(
> -				cnxk_eth_dev, eth_dev->data->port_id, i,
> -				config->vector_sz, config->vector_timeout_ns,
> -				config->vector_mp);
> -			if (rc)
> -				return -EINVAL;
> -		}
> -	} else {
> -
> -		cnxk_sso_updt_xae_cnt(dev, config->vector_mp,
> -				      RTE_EVENT_TYPE_ETHDEV_VECTOR);
> -		rc = cnxk_sso_xae_reconfigure(
> -			(struct rte_eventdev *)(uintptr_t)event_dev);
> -		rc = cnxk_sso_rx_adapter_vwqe_enable(
> -			cnxk_eth_dev, eth_dev->data->port_id, rx_queue_id,
> -			config->vector_sz, config->vector_timeout_ns,
> -			config->vector_mp);
> -		if (rc)
> -			return -EINVAL;
> -	}
> -
> -	return 0;
> -}
> -
>  static int
>  cn10k_sso_tx_adapter_caps_get(const struct rte_eventdev *dev,
>  			      const struct rte_eth_dev *eth_dev, uint32_t *caps)
> @@ -841,8 +766,6 @@ static struct rte_eventdev_ops cn10k_sso_dev_ops = {
>  	.eth_rx_adapter_stop = cnxk_sso_rx_adapter_stop,
> 
>  	.eth_rx_adapter_vector_limits_get = cn10k_sso_rx_adapter_vector_limits,
> -	.eth_rx_adapter_event_vector_config =
> -		cn10k_sso_rx_adapter_vector_config,
> 
>  	.eth_tx_adapter_caps_get = cn10k_sso_tx_adapter_caps_get,
>  	.eth_tx_adapter_queue_add = cn10k_sso_tx_adapter_queue_add,
> diff --git a/drivers/event/cnxk/cnxk_eventdev_adptr.c b/drivers/event/cnxk/cnxk_eventdev_adptr.c
> index baf2f2aa6b..80f5602286 100644
> --- a/drivers/event/cnxk/cnxk_eventdev_adptr.c
> +++ b/drivers/event/cnxk/cnxk_eventdev_adptr.c
> @@ -156,6 +156,35 @@ cnxk_sso_rxq_disable(struct cnxk_eth_dev *cnxk_eth_dev, uint16_t rq_id)
>  	return roc_nix_rq_modify(&cnxk_eth_dev->nix, rq, 0);
>  }
> 
> +static int
> +cnxk_sso_rx_adapter_vwqe_enable(struct cnxk_eth_dev *cnxk_eth_dev,
> +				uint16_t port_id, uint16_t rq_id, uint16_t sz,
> +				uint64_t tmo_ns, struct rte_mempool *vmp)
> +{
> +	struct roc_nix_rq *rq;
> +
> +	rq = &cnxk_eth_dev->rqs[rq_id];
> +
> +	if (!rq->sso_ena)
> +		return -EINVAL;
> +	if (rq->flow_tag_width == 0)
> +		return -EINVAL;
> +
> +	rq->vwqe_ena = 1;
> +	rq->vwqe_first_skip = 0;
> +	rq->vwqe_aura_handle = roc_npa_aura_handle_to_aura(vmp->pool_id);
> +	rq->vwqe_max_sz_exp = rte_log2_u32(sz);
> +	rq->vwqe_wait_tmo =
> +		tmo_ns /
> +		((roc_nix_get_vwqe_interval(&cnxk_eth_dev->nix) + 1) * 100);
> +	rq->tag_mask = (port_id & 0xF) << 20;
> +	rq->tag_mask |=
> +		(((port_id >> 4) & 0xF) | (RTE_EVENT_TYPE_ETHDEV_VECTOR << 4))
> +		<< 24;
> +
> +	return roc_nix_rq_modify(&cnxk_eth_dev->nix, rq, 0);
> +}
> +
>  int
>  cnxk_sso_rx_adapter_queue_add(
>  	const struct rte_eventdev *event_dev, const struct rte_eth_dev *eth_dev,
> @@ -183,6 +212,18 @@ cnxk_sso_rx_adapter_queue_add(
>  			&queue_conf->ev,
>  			!!(queue_conf->rx_queue_flags &
>  			   RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID));
> +		if (queue_conf->rx_queue_flags &
> +		    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
> +			cnxk_sso_updt_xae_cnt(dev, queue_conf->vector_mp,
> +					      RTE_EVENT_TYPE_ETHDEV_VECTOR);
> +			rc |= cnxk_sso_xae_reconfigure(
> +				(struct rte_eventdev *)(uintptr_t)event_dev);
> +			rc |= cnxk_sso_rx_adapter_vwqe_enable(
> +				cnxk_eth_dev, port, rx_queue_id,
> +				queue_conf->vector_sz,
> +				queue_conf->vector_timeout_ns,
> +				queue_conf->vector_mp);
> +		}
>  		rox_nix_fc_npa_bp_cfg(&cnxk_eth_dev->nix,
>  				      rxq_sp->qconf.mp->pool_id, true,
>  				      dev->force_ena_bp);
> diff --git a/lib/eventdev/eventdev_pmd.h b/lib/eventdev/eventdev_pmd.h
> index 0f724ac85d..63b3bc4b51 100644
> --- a/lib/eventdev/eventdev_pmd.h
> +++ b/lib/eventdev/eventdev_pmd.h
> @@ -667,32 +667,6 @@ typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
>  	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
>  	struct rte_event_eth_rx_adapter_vector_limits *limits);
> 
> -struct rte_event_eth_rx_adapter_event_vector_config;
> -/**
> - * Enable event vector on an given Rx queue of a ethernet devices belonging to
> - * the Rx adapter.
> - *
> - * @param dev
> - *   Event device pointer
> - *
> - * @param eth_dev
> - *   Ethernet device pointer
> - *
> - * @param rx_queue_id
> - *   The Rx queue identifier
> - *
> - * @param config
> - *   Pointer to the event vector configuration structure.
> - *
> - * @return
> - *   - 0: Success.
> - *   - <0: Error code returned by the driver function.
> - */
> -typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
> -	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
> -	int32_t rx_queue_id,
> -	const struct rte_event_eth_rx_adapter_event_vector_config *config);
> -
>  typedef uint32_t rte_event_pmd_selftest_seqn_t;
>  extern int rte_event_pmd_selftest_seqn_dynfield_offset;
> 
> @@ -1118,9 +1092,6 @@ struct rte_eventdev_ops {
>  	eventdev_eth_rx_adapter_vector_limits_get_t
>  		eth_rx_adapter_vector_limits_get;
>  	/**< Get event vector limits for the Rx adapter */
> -	eventdev_eth_rx_adapter_event_vector_config_t
> -		eth_rx_adapter_event_vector_config;
> -	/**< Configure Rx adapter with event vector */
> 
>  	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
>  	/**< Get timer adapter capabilities */
> diff --git a/lib/eventdev/rte_event_eth_rx_adapter.c b/lib/eventdev/rte_event_eth_rx_adapter.c
> index 13dfb28401..2b2dd688fc 100644
> --- a/lib/eventdev/rte_event_eth_rx_adapter.c
> +++ b/lib/eventdev/rte_event_eth_rx_adapter.c
> @@ -1895,6 +1895,24 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
>  	} else
>  		qi_ev->flow_id = 0;
> 
> +	if (conf->rx_queue_flags &
> +	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
> +		queue_info->ena_vector = 1;
> +		qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
> +		rxa_set_vector_data(queue_info, conf->vector_sz,
> +				    conf->vector_timeout_ns, conf->vector_mp,
> +				    rx_queue_id, dev_info->dev->data->port_id);
> +		rx_adapter->ena_vector = 1;
> +		rx_adapter->vector_tmo_ticks =
> +			rx_adapter->vector_tmo_ticks ?
> +				      RTE_MIN(queue_info->vector_data
> +							.vector_timeout_ticks >>
> +						1,
> +					rx_adapter->vector_tmo_ticks) :
> +				queue_info->vector_data.vector_timeout_ticks >>
> +					1;
> +	}
> +
>  	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
>  	if (rxa_polled_queue(dev_info, rx_queue_id)) {
>  		rx_adapter->num_rx_polled += !pollq;
> @@ -1920,42 +1938,6 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
>  	}
>  }
> 
> -static void
> -rxa_sw_event_vector_configure(
> -	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
> -	int rx_queue_id,
> -	const struct rte_event_eth_rx_adapter_event_vector_config *config)
> -{
> -	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
> -	struct eth_rx_queue_info *queue_info;
> -	struct rte_event *qi_ev;
> -
> -	if (rx_queue_id == -1) {
> -		uint16_t nb_rx_queues;
> -		uint16_t i;
> -
> -		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
> -		for (i = 0; i < nb_rx_queues; i++)
> -			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
> -						      config);
> -		return;
> -	}
> -
> -	queue_info = &dev_info->rx_queue[rx_queue_id];
> -	qi_ev = (struct rte_event *)&queue_info->event;
> -	queue_info->ena_vector = 1;
> -	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
> -	rxa_set_vector_data(queue_info, config->vector_sz,
> -			    config->vector_timeout_ns, config->vector_mp,
> -			    rx_queue_id, dev_info->dev->data->port_id);
> -	rx_adapter->ena_vector = 1;
> -	rx_adapter->vector_tmo_ticks =
> -		rx_adapter->vector_tmo_ticks ?
> -			      RTE_MIN(config->vector_timeout_ns >> 1,
> -				rx_adapter->vector_tmo_ticks) :
> -			      config->vector_timeout_ns >> 1;
> -}
> -
>  static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
>  		uint16_t eth_dev_id,
>  		int rx_queue_id,
> @@ -2270,6 +2252,7 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
>  	struct rte_event_eth_rx_adapter *rx_adapter;
>  	struct rte_eventdev *dev;
>  	struct eth_device_info *dev_info;
> +	struct rte_event_eth_rx_adapter_vector_limits limits;
> 
>  	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
>  	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
> @@ -2297,13 +2280,46 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
>  		return -EINVAL;
>  	}
> 
> -	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
> -	    (queue_conf->rx_queue_flags &
> -	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
> -		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
> -				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
> -				 eth_dev_id, id);
> -		return -EINVAL;
> +	if (queue_conf->rx_queue_flags &
> +	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
> +
> +		if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0) {
> +			RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
> +					 " eth port: %" PRIu16
> +					 " adapter id: %" PRIu8,
> +					 eth_dev_id, id);
> +			return -EINVAL;
> +		}
> +
> +		ret = rte_event_eth_rx_adapter_vector_limits_get(
> +			rx_adapter->eventdev_id, eth_dev_id, &limits);
> +		if (ret < 0) {
> +			RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
> +					 " eth port: %" PRIu16
> +					 " adapter id: %" PRIu8,
> +					 eth_dev_id, id);
> +			return -EINVAL;
> +		}
> +		if (queue_conf->vector_sz < limits.min_sz ||
> +		    queue_conf->vector_sz > limits.max_sz ||
> +		    queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
> +		    queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
> +		    queue_conf->vector_mp == NULL) {
> +			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> +					 " eth port: %" PRIu16
> +					 " adapter id: %" PRIu8,
> +					 eth_dev_id, id);
> +			return -EINVAL;
> +		}
> +		if (queue_conf->vector_mp->elt_size <
> +		    (sizeof(struct rte_event_vector) +
> +		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
> +			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> +					 " eth port: %" PRIu16
> +					 " adapter id: %" PRIu8,
> +					 eth_dev_id, id);
> +			return -EINVAL;
> +		}
>  	}
> 
>  	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
> @@ -2499,83 +2515,6 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
>  	return ret;
>  }
> 
> -int
> -rte_event_eth_rx_adapter_queue_event_vector_config(
> -	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
> -	struct rte_event_eth_rx_adapter_event_vector_config *config)
> -{
> -	struct rte_event_eth_rx_adapter_vector_limits limits;
> -	struct rte_event_eth_rx_adapter *rx_adapter;
> -	struct rte_eventdev *dev;
> -	uint32_t cap;
> -	int ret;
> -
> -	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
> -	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
> -
> -	rx_adapter = rxa_id_to_adapter(id);
> -	if ((rx_adapter == NULL) || (config == NULL))
> -		return -EINVAL;
> -
> -	dev = &rte_eventdevs[rx_adapter->eventdev_id];
> -	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
> -						eth_dev_id, &cap);
> -	if (ret) {
> -		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
> -				 "eth port %" PRIu16,
> -				 id, eth_dev_id);
> -		return ret;
> -	}
> -
> -	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
> -		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
> -				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
> -				 eth_dev_id, id);
> -		return -EINVAL;
> -	}
> -
> -	ret = rte_event_eth_rx_adapter_vector_limits_get(
> -		rx_adapter->eventdev_id, eth_dev_id, &limits);
> -	if (ret) {
> -		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
> -				 "eth port %" PRIu16,
> -				 rx_adapter->eventdev_id, eth_dev_id);
> -		return ret;
> -	}
> -
> -	if (config->vector_sz < limits.min_sz ||
> -	    config->vector_sz > limits.max_sz ||
> -	    config->vector_timeout_ns < limits.min_timeout_ns ||
> -	    config->vector_timeout_ns > limits.max_timeout_ns ||
> -	    config->vector_mp == NULL) {
> -		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> -				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
> -				 eth_dev_id, id);
> -		return -EINVAL;
> -	}
> -	if (config->vector_mp->elt_size <
> -	    (sizeof(struct rte_event_vector) +
> -	     (sizeof(uintptr_t) * config->vector_sz))) {
> -		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> -				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
> -				 eth_dev_id, id);
> -		return -EINVAL;
> -	}
> -
> -	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
> -		RTE_FUNC_PTR_OR_ERR_RET(
> -			*dev->dev_ops->eth_rx_adapter_event_vector_config,
> -			-ENOTSUP);
> -		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
> -			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
> -	} else {
> -		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
> -					      rx_queue_id, config);
> -	}
> -
> -	return ret;
> -}
> -
>  int
>  rte_event_eth_rx_adapter_vector_limits_get(
>  	uint8_t dev_id, uint16_t eth_port_id,
> diff --git a/lib/eventdev/rte_event_eth_rx_adapter.h b/lib/eventdev/rte_event_eth_rx_adapter.h
> index 182dd2e5dd..3f8b362295 100644
> --- a/lib/eventdev/rte_event_eth_rx_adapter.h
> +++ b/lib/eventdev/rte_event_eth_rx_adapter.h
> @@ -171,9 +171,6 @@ struct rte_event_eth_rx_adapter_queue_conf {
>  	 * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
>  	 * enqueued event.
>  	 */
> -};
> -
> -struct rte_event_eth_rx_adapter_event_vector_config {
>  	uint16_t vector_sz;
>  	/**<
>  	 * Indicates the maximum number for mbufs to combine and form a vector.
> @@ -548,33 +545,6 @@ int rte_event_eth_rx_adapter_vector_limits_get(
>  	uint8_t dev_id, uint16_t eth_port_id,
>  	struct rte_event_eth_rx_adapter_vector_limits *limits);
> 
> -/**
> - * Configure event vectorization for a given ethernet device queue, that has
> - * been added to a event eth Rx adapter.
> - *
> - * @param id
> - *  The identifier of the ethernet Rx event adapter.
> - *
> - * @param eth_dev_id
> - *  The identifier of the ethernet device.
> - *
> - * @param rx_queue_id
> - *  Ethernet device receive queue index.
> - *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
> - *  are configured with event vectorization.
> - *
> - * @param config
> - *  Event vector configuration structure.
> - *
> - * @return
> - *  - 0: Success, Receive queue configured correctly.
> - *  - <0: Error code on failure.
> - */
> -__rte_experimental
> -int rte_event_eth_rx_adapter_queue_event_vector_config(
> -	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
> -	struct rte_event_eth_rx_adapter_event_vector_config *config);
> -
>  #ifdef __cplusplus
>  }
>  #endif
> diff --git a/lib/eventdev/version.map b/lib/eventdev/version.map
> index 88625621ec..cd86d2d908 100644
> --- a/lib/eventdev/version.map
> +++ b/lib/eventdev/version.map
> @@ -142,7 +142,6 @@ EXPERIMENTAL {
>  	#added in 21.05
>  	rte_event_vector_pool_create;
>  	rte_event_eth_rx_adapter_vector_limits_get;
> -	rte_event_eth_rx_adapter_queue_event_vector_config;
>  	__rte_eventdev_trace_crypto_adapter_enqueue;
>  };
> 
> --
> 2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v13] eventdev: simplify Rx adapter event vector config
  2021-08-18  8:22                           ` Jayatheerthan, Jay
@ 2021-08-20  7:33                             ` Naga Harish K, S V
  2021-09-07  8:30                               ` Jerin Jacob
  0 siblings, 1 reply; 153+ messages in thread
From: Naga Harish K, S V @ 2021-08-20  7:33 UTC (permalink / raw)
  To: Jayatheerthan, Jay, pbhagavatula, jerinj, Ray Kinsella, Shijith Thotton
  Cc: dev



-----Original Message-----
From: Jayatheerthan, Jay <jay.jayatheerthan@intel.com> 
Sent: Wednesday, August 18, 2021 1:53 PM
To: pbhagavatula@marvell.com; jerinj@marvell.com; Ray Kinsella <mdr@ashroe.eu>; Shijith Thotton <sthotton@marvell.com>; Naga Harish K, S V <s.v.naga.harish.k@intel.com>
Cc: dev@dpdk.org
Subject: RE: [dpdk-dev] [PATCH v13] eventdev: simplify Rx adapter event vector config

HI Harish,
Could you review this patch ?

-Jay


> -----Original Message-----
> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
> Sent: Wednesday, August 18, 2021 12:27 PM
> To: jerinj@marvell.com; Ray Kinsella <mdr@ashroe.eu>; Pavan Nikhilesh 
> <pbhagavatula@marvell.com>; Shijith Thotton <sthotton@marvell.com>; 
> Jayatheerthan, Jay <jay.jayatheerthan@intel.com>
> Cc: dev@dpdk.org
> Subject: [dpdk-dev] [PATCH v13] eventdev: simplify Rx adapter event 
> vector config
> 
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> Include vector configuration into the structure 
> ``rte_event_eth_rx_adapter_queue_conf`` that is used to configure Rx 
> adapter ethernet device Rx queue parameters.
> This simplifies event vector configuration as it avoids splitting 
> configuration per Rx queue.
> 
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
> ---
>  v13 Changes:
>  - Fix cnxk driver compilation.
>  v12 Changes:
>  - Remove deprication notice.
>  - Remove unnecessary change Id.
> 
>  app/test-eventdev/test_pipeline_common.c |  16 +-
>  doc/guides/rel_notes/deprecation.rst     |   9 --
>  drivers/event/cnxk/cn10k_eventdev.c      |  77 ----------
>  drivers/event/cnxk/cnxk_eventdev_adptr.c |  41 ++++++
>  lib/eventdev/eventdev_pmd.h              |  29 ----
>  lib/eventdev/rte_event_eth_rx_adapter.c  | 179 
> ++++++++---------------  lib/eventdev/rte_event_eth_rx_adapter.h  |  30 ----
>  lib/eventdev/version.map                 |   1 -
>  8 files changed, 104 insertions(+), 278 deletions(-)
> 
> diff --git a/app/test-eventdev/test_pipeline_common.c 
> b/app/test-eventdev/test_pipeline_common.c
> index 6ee530d4cd..2697547641 100644
> --- a/app/test-eventdev/test_pipeline_common.c
> +++ b/app/test-eventdev/test_pipeline_common.c
> @@ -332,7 +332,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
>  	uint16_t prod;
>  	struct rte_mempool *vector_pool = NULL;
>  	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
> -	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;
> 
>  	memset(&queue_conf, 0,
>  			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
> @@ -398,8 +397,12 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
>  			}
> 
>  			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
> +				queue_conf.vector_sz = opt->vector_size;
> +				queue_conf.vector_timeout_ns =
> +					opt->vector_tmo_nsec;
>  				queue_conf.rx_queue_flags |=
>  				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
> +				queue_conf.vector_mp = vector_pool;
>  			} else {
>  				evt_err("Rx adapter doesn't support event vector");
>  				return -EINVAL;
> @@ -419,17 +422,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
>  			return ret;
>  		}
> 
> -		if (opt->ena_vector) {
> -			vec_conf.vector_sz = opt->vector_size;
> -			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
> -			vec_conf.vector_mp = vector_pool;
> -			if (rte_event_eth_rx_adapter_queue_event_vector_config(
> -				    prod, prod, -1, &vec_conf) < 0) {
> -				evt_err("Failed to configure event vectorization for Rx adapter");
> -				return -EINVAL;
> -			}
> -		}
> -
>  		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
>  			uint32_t service_id = -1U;
> 
> diff --git a/doc/guides/rel_notes/deprecation.rst 
> b/doc/guides/rel_notes/deprecation.rst
> index 76a4abfd6b..2c37d7222c 100644
> --- a/doc/guides/rel_notes/deprecation.rst
> +++ b/doc/guides/rel_notes/deprecation.rst
> @@ -257,15 +257,6 @@ Deprecation Notices
>    An 8-byte reserved field will be added to the structure ``rte_event_timer`` to
>    support future extensions.
> 
> -* eventdev: The structure ``rte_event_eth_rx_adapter_queue_conf`` 
> will be
> -  extended to include 
> ``rte_event_eth_rx_adapter_event_vector_config`` elements
> -  and the function 
> ``rte_event_eth_rx_adapter_queue_event_vector_config`` will
> -  be removed in DPDK 21.11.
> -
> -  An application can enable event vectorization by passing the 
> desired vector
> -  values to the function ``rte_event_eth_rx_adapter_queue_add`` using
> -  the structure ``rte_event_eth_rx_adapter_queue_add``.
> -

The above paragraph on how to enable Rx adapter vectorization may need to be 
added in rx adapter documentation with correct structure name.


>  * eventdev: Reserved bytes of ``rte_event_crypto_request`` is a space holder
>    for ``response_info``. Both should be decoupled for better clarity.
>    New space for ``response_info`` can be made by changing diff --git 
> a/drivers/event/cnxk/cn10k_eventdev.c 
> b/drivers/event/cnxk/cn10k_eventdev.c
> index 6f37c5bd23..160192bb27 100644
> --- a/drivers/event/cnxk/cn10k_eventdev.c
> +++ b/drivers/event/cnxk/cn10k_eventdev.c
> @@ -696,81 +696,6 @@ cn10k_sso_rx_adapter_vector_limits(
>  	return 0;
>  }
> 
> -static int
> -cnxk_sso_rx_adapter_vwqe_enable(struct cnxk_eth_dev *cnxk_eth_dev,
> -				uint16_t port_id, uint16_t rq_id, uint16_t sz,
> -				uint64_t tmo_ns, struct rte_mempool *vmp)
> -{
> -	struct roc_nix_rq *rq;
> -
> -	rq = &cnxk_eth_dev->rqs[rq_id];
> -
> -	if (!rq->sso_ena)
> -		return -EINVAL;
> -	if (rq->flow_tag_width == 0)
> -		return -EINVAL;
> -
> -	rq->vwqe_ena = 1;
> -	rq->vwqe_first_skip = 0;
> -	rq->vwqe_aura_handle = roc_npa_aura_handle_to_aura(vmp->pool_id);
> -	rq->vwqe_max_sz_exp = rte_log2_u32(sz);
> -	rq->vwqe_wait_tmo =
> -		tmo_ns /
> -		((roc_nix_get_vwqe_interval(&cnxk_eth_dev->nix) + 1) * 100);
> -	rq->tag_mask = (port_id & 0xF) << 20;
> -	rq->tag_mask |=
> -		(((port_id >> 4) & 0xF) | (RTE_EVENT_TYPE_ETHDEV_VECTOR << 4))
> -		<< 24;
> -
> -	return roc_nix_rq_modify(&cnxk_eth_dev->nix, rq, 0);
> -}
> -
> -static int
> -cn10k_sso_rx_adapter_vector_config(
> -	const struct rte_eventdev *event_dev, const struct rte_eth_dev *eth_dev,
> -	int32_t rx_queue_id,
> -	const struct rte_event_eth_rx_adapter_event_vector_config *config)
> -{
> -	struct cnxk_eth_dev *cnxk_eth_dev;
> -	struct cnxk_sso_evdev *dev;
> -	int i, rc;
> -
> -	rc = strncmp(eth_dev->device->driver->name, "net_cn10k", 8);
> -	if (rc)
> -		return -EINVAL;
> -
> -	dev = cnxk_sso_pmd_priv(event_dev);
> -	cnxk_eth_dev = cnxk_eth_pmd_priv(eth_dev);
> -	if (rx_queue_id < 0) {
> -		for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
> -			cnxk_sso_updt_xae_cnt(dev, config->vector_mp,
> -					      RTE_EVENT_TYPE_ETHDEV_VECTOR);
> -			rc = cnxk_sso_xae_reconfigure(
> -				(struct rte_eventdev *)(uintptr_t)event_dev);
> -			rc = cnxk_sso_rx_adapter_vwqe_enable(
> -				cnxk_eth_dev, eth_dev->data->port_id, i,
> -				config->vector_sz, config->vector_timeout_ns,
> -				config->vector_mp);
> -			if (rc)
> -				return -EINVAL;
> -		}
> -	} else {
> -
> -		cnxk_sso_updt_xae_cnt(dev, config->vector_mp,
> -				      RTE_EVENT_TYPE_ETHDEV_VECTOR);
> -		rc = cnxk_sso_xae_reconfigure(
> -			(struct rte_eventdev *)(uintptr_t)event_dev);
> -		rc = cnxk_sso_rx_adapter_vwqe_enable(
> -			cnxk_eth_dev, eth_dev->data->port_id, rx_queue_id,
> -			config->vector_sz, config->vector_timeout_ns,
> -			config->vector_mp);
> -		if (rc)
> -			return -EINVAL;
> -	}
> -
> -	return 0;
> -}
> -
>  static int
>  cn10k_sso_tx_adapter_caps_get(const struct rte_eventdev *dev,
>  			      const struct rte_eth_dev *eth_dev, uint32_t *caps) @@ -841,8 
> +766,6 @@ static struct rte_eventdev_ops cn10k_sso_dev_ops = {
>  	.eth_rx_adapter_stop = cnxk_sso_rx_adapter_stop,
> 
>  	.eth_rx_adapter_vector_limits_get = cn10k_sso_rx_adapter_vector_limits,
> -	.eth_rx_adapter_event_vector_config =
> -		cn10k_sso_rx_adapter_vector_config,
> 
>  	.eth_tx_adapter_caps_get = cn10k_sso_tx_adapter_caps_get,
>  	.eth_tx_adapter_queue_add = cn10k_sso_tx_adapter_queue_add, diff 
> --git a/drivers/event/cnxk/cnxk_eventdev_adptr.c 
> b/drivers/event/cnxk/cnxk_eventdev_adptr.c
> index baf2f2aa6b..80f5602286 100644
> --- a/drivers/event/cnxk/cnxk_eventdev_adptr.c
> +++ b/drivers/event/cnxk/cnxk_eventdev_adptr.c
> @@ -156,6 +156,35 @@ cnxk_sso_rxq_disable(struct cnxk_eth_dev *cnxk_eth_dev, uint16_t rq_id)
>  	return roc_nix_rq_modify(&cnxk_eth_dev->nix, rq, 0);  }
> 
> +static int
> +cnxk_sso_rx_adapter_vwqe_enable(struct cnxk_eth_dev *cnxk_eth_dev,
> +				uint16_t port_id, uint16_t rq_id, uint16_t sz,
> +				uint64_t tmo_ns, struct rte_mempool *vmp) {
> +	struct roc_nix_rq *rq;
> +
> +	rq = &cnxk_eth_dev->rqs[rq_id];
> +
> +	if (!rq->sso_ena)
> +		return -EINVAL;
> +	if (rq->flow_tag_width == 0)
> +		return -EINVAL;
> +
> +	rq->vwqe_ena = 1;
> +	rq->vwqe_first_skip = 0;
> +	rq->vwqe_aura_handle = roc_npa_aura_handle_to_aura(vmp->pool_id);
> +	rq->vwqe_max_sz_exp = rte_log2_u32(sz);
> +	rq->vwqe_wait_tmo =
> +		tmo_ns /
> +		((roc_nix_get_vwqe_interval(&cnxk_eth_dev->nix) + 1) * 100);
> +	rq->tag_mask = (port_id & 0xF) << 20;
> +	rq->tag_mask |=
> +		(((port_id >> 4) & 0xF) | (RTE_EVENT_TYPE_ETHDEV_VECTOR << 4))
> +		<< 24;
> +
> +	return roc_nix_rq_modify(&cnxk_eth_dev->nix, rq, 0); }
> +
>  int
>  cnxk_sso_rx_adapter_queue_add(
>  	const struct rte_eventdev *event_dev, const struct rte_eth_dev 
> *eth_dev, @@ -183,6 +212,18 @@ cnxk_sso_rx_adapter_queue_add(
>  			&queue_conf->ev,
>  			!!(queue_conf->rx_queue_flags &
>  			   RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID));
> +		if (queue_conf->rx_queue_flags &
> +		    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
> +			cnxk_sso_updt_xae_cnt(dev, queue_conf->vector_mp,
> +					      RTE_EVENT_TYPE_ETHDEV_VECTOR);
> +			rc |= cnxk_sso_xae_reconfigure(
> +				(struct rte_eventdev *)(uintptr_t)event_dev);
> +			rc |= cnxk_sso_rx_adapter_vwqe_enable(
> +				cnxk_eth_dev, port, rx_queue_id,
> +				queue_conf->vector_sz,
> +				queue_conf->vector_timeout_ns,
> +				queue_conf->vector_mp);
> +		}
>  		rox_nix_fc_npa_bp_cfg(&cnxk_eth_dev->nix,
>  				      rxq_sp->qconf.mp->pool_id, true,
>  				      dev->force_ena_bp);
> diff --git a/lib/eventdev/eventdev_pmd.h b/lib/eventdev/eventdev_pmd.h 
> index 0f724ac85d..63b3bc4b51 100644
> --- a/lib/eventdev/eventdev_pmd.h
> +++ b/lib/eventdev/eventdev_pmd.h
> @@ -667,32 +667,6 @@ typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
>  	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
>  	struct rte_event_eth_rx_adapter_vector_limits *limits);
> 
> -struct rte_event_eth_rx_adapter_event_vector_config;
> -/**
> - * Enable event vector on an given Rx queue of a ethernet devices 
> belonging to
> - * the Rx adapter.
> - *
> - * @param dev
> - *   Event device pointer
> - *
> - * @param eth_dev
> - *   Ethernet device pointer
> - *
> - * @param rx_queue_id
> - *   The Rx queue identifier
> - *
> - * @param config
> - *   Pointer to the event vector configuration structure.
> - *
> - * @return
> - *   - 0: Success.
> - *   - <0: Error code returned by the driver function.
> - */
> -typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
> -	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
> -	int32_t rx_queue_id,
> -	const struct rte_event_eth_rx_adapter_event_vector_config *config);
> -
>  typedef uint32_t rte_event_pmd_selftest_seqn_t;  extern int 
> rte_event_pmd_selftest_seqn_dynfield_offset;
> 
> @@ -1118,9 +1092,6 @@ struct rte_eventdev_ops {
>  	eventdev_eth_rx_adapter_vector_limits_get_t
>  		eth_rx_adapter_vector_limits_get;
>  	/**< Get event vector limits for the Rx adapter */
> -	eventdev_eth_rx_adapter_event_vector_config_t
> -		eth_rx_adapter_event_vector_config;
> -	/**< Configure Rx adapter with event vector */
> 
>  	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
>  	/**< Get timer adapter capabilities */ diff --git 
> a/lib/eventdev/rte_event_eth_rx_adapter.c 
> b/lib/eventdev/rte_event_eth_rx_adapter.c
> index 13dfb28401..2b2dd688fc 100644
> --- a/lib/eventdev/rte_event_eth_rx_adapter.c
> +++ b/lib/eventdev/rte_event_eth_rx_adapter.c
> @@ -1895,6 +1895,24 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
>  	} else
>  		qi_ev->flow_id = 0;
> 
> +	if (conf->rx_queue_flags &
> +	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
> +		queue_info->ena_vector = 1;
> +		qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
> +		rxa_set_vector_data(queue_info, conf->vector_sz,
> +				    conf->vector_timeout_ns, conf->vector_mp,
> +				    rx_queue_id, dev_info->dev->data->port_id);
> +		rx_adapter->ena_vector = 1;
> +		rx_adapter->vector_tmo_ticks =
> +			rx_adapter->vector_tmo_ticks ?
> +				      RTE_MIN(queue_info->vector_data
> +							.vector_timeout_ticks >>
> +						1,
> +					rx_adapter->vector_tmo_ticks) :
> +				queue_info->vector_data.vector_timeout_ticks >>
> +					1;
> +	}
> +
>  	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
>  	if (rxa_polled_queue(dev_info, rx_queue_id)) {
>  		rx_adapter->num_rx_polled += !pollq; @@ -1920,42 +1938,6 @@ 
> rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
>  	}
>  }
> 
> -static void
> -rxa_sw_event_vector_configure(
> -	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
> -	int rx_queue_id,
> -	const struct rte_event_eth_rx_adapter_event_vector_config *config)
> -{
> -	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
> -	struct eth_rx_queue_info *queue_info;
> -	struct rte_event *qi_ev;
> -
> -	if (rx_queue_id == -1) {
> -		uint16_t nb_rx_queues;
> -		uint16_t i;
> -
> -		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
> -		for (i = 0; i < nb_rx_queues; i++)
> -			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
> -						      config);
> -		return;
> -	}
> -
> -	queue_info = &dev_info->rx_queue[rx_queue_id];
> -	qi_ev = (struct rte_event *)&queue_info->event;
> -	queue_info->ena_vector = 1;
> -	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
> -	rxa_set_vector_data(queue_info, config->vector_sz,
> -			    config->vector_timeout_ns, config->vector_mp,
> -			    rx_queue_id, dev_info->dev->data->port_id);
> -	rx_adapter->ena_vector = 1;
> -	rx_adapter->vector_tmo_ticks =
> -		rx_adapter->vector_tmo_ticks ?
> -			      RTE_MIN(config->vector_timeout_ns >> 1,
> -				rx_adapter->vector_tmo_ticks) :
> -			      config->vector_timeout_ns >> 1;
> -}
> -
>  static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
>  		uint16_t eth_dev_id,
>  		int rx_queue_id,
> @@ -2270,6 +2252,7 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
>  	struct rte_event_eth_rx_adapter *rx_adapter;
>  	struct rte_eventdev *dev;
>  	struct eth_device_info *dev_info;
> +	struct rte_event_eth_rx_adapter_vector_limits limits;
> 
>  	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
>  	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL); @@ -2297,13 
> +2280,46 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
>  		return -EINVAL;
>  	}
> 
> -	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
> -	    (queue_conf->rx_queue_flags &
> -	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
> -		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
> -				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
> -				 eth_dev_id, id);
> -		return -EINVAL;
> +	if (queue_conf->rx_queue_flags &
> +	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
> +
> +		if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0) {
> +			RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
> +					 " eth port: %" PRIu16
> +					 " adapter id: %" PRIu8,
> +					 eth_dev_id, id);
> +			return -EINVAL;
> +		}
> +
> +		ret = rte_event_eth_rx_adapter_vector_limits_get(
> +			rx_adapter->eventdev_id, eth_dev_id, &limits);
> +		if (ret < 0) {
> +			RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
> +					 " eth port: %" PRIu16
> +					 " adapter id: %" PRIu8,
> +					 eth_dev_id, id);
> +			return -EINVAL;
> +		}
> +		if (queue_conf->vector_sz < limits.min_sz ||
> +		    queue_conf->vector_sz > limits.max_sz ||
> +		    queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
> +		    queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
> +		    queue_conf->vector_mp == NULL) {
> +			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> +					 " eth port: %" PRIu16
> +					 " adapter id: %" PRIu8,
> +					 eth_dev_id, id);
> +			return -EINVAL;
> +		}
> +		if (queue_conf->vector_mp->elt_size <
> +		    (sizeof(struct rte_event_vector) +
> +		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
> +			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> +					 " eth port: %" PRIu16
> +					 " adapter id: %" PRIu8,
> +					 eth_dev_id, id);
> +			return -EINVAL;
> +		}
>  	}
> 
>  	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 && @@ 
> -2499,83 +2515,6 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
>  	return ret;
>  }
> 
> -int
> -rte_event_eth_rx_adapter_queue_event_vector_config(
> -	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
> -	struct rte_event_eth_rx_adapter_event_vector_config *config)
> -{
> -	struct rte_event_eth_rx_adapter_vector_limits limits;
> -	struct rte_event_eth_rx_adapter *rx_adapter;
> -	struct rte_eventdev *dev;
> -	uint32_t cap;
> -	int ret;
> -
> -	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
> -	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
> -
> -	rx_adapter = rxa_id_to_adapter(id);
> -	if ((rx_adapter == NULL) || (config == NULL))
> -		return -EINVAL;
> -
> -	dev = &rte_eventdevs[rx_adapter->eventdev_id];
> -	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
> -						eth_dev_id, &cap);
> -	if (ret) {
> -		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
> -				 "eth port %" PRIu16,
> -				 id, eth_dev_id);
> -		return ret;
> -	}
> -
> -	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
> -		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
> -				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
> -				 eth_dev_id, id);
> -		return -EINVAL;
> -	}
> -
> -	ret = rte_event_eth_rx_adapter_vector_limits_get(
> -		rx_adapter->eventdev_id, eth_dev_id, &limits);
> -	if (ret) {
> -		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
> -				 "eth port %" PRIu16,
> -				 rx_adapter->eventdev_id, eth_dev_id);
> -		return ret;
> -	}
> -
> -	if (config->vector_sz < limits.min_sz ||
> -	    config->vector_sz > limits.max_sz ||
> -	    config->vector_timeout_ns < limits.min_timeout_ns ||
> -	    config->vector_timeout_ns > limits.max_timeout_ns ||
> -	    config->vector_mp == NULL) {
> -		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> -				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
> -				 eth_dev_id, id);
> -		return -EINVAL;
> -	}
> -	if (config->vector_mp->elt_size <
> -	    (sizeof(struct rte_event_vector) +
> -	     (sizeof(uintptr_t) * config->vector_sz))) {
> -		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> -				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
> -				 eth_dev_id, id);
> -		return -EINVAL;
> -	}
> -
> -	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
> -		RTE_FUNC_PTR_OR_ERR_RET(
> -			*dev->dev_ops->eth_rx_adapter_event_vector_config,
> -			-ENOTSUP);
> -		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
> -			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
> -	} else {
> -		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
> -					      rx_queue_id, config);
> -	}
> -
> -	return ret;
> -}
> -
>  int
>  rte_event_eth_rx_adapter_vector_limits_get(
>  	uint8_t dev_id, uint16_t eth_port_id, diff --git 
> a/lib/eventdev/rte_event_eth_rx_adapter.h 
> b/lib/eventdev/rte_event_eth_rx_adapter.h
> index 182dd2e5dd..3f8b362295 100644
> --- a/lib/eventdev/rte_event_eth_rx_adapter.h
> +++ b/lib/eventdev/rte_event_eth_rx_adapter.h
> @@ -171,9 +171,6 @@ struct rte_event_eth_rx_adapter_queue_conf {
>  	 * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
>  	 * enqueued event.
>  	 */
> -};
> -
> -struct rte_event_eth_rx_adapter_event_vector_config {
>  	uint16_t vector_sz;
>  	/**<
>  	 * Indicates the maximum number for mbufs to combine and form a vector.
> @@ -548,33 +545,6 @@ int rte_event_eth_rx_adapter_vector_limits_get(
>  	uint8_t dev_id, uint16_t eth_port_id,
>  	struct rte_event_eth_rx_adapter_vector_limits *limits);
> 
> -/**
> - * Configure event vectorization for a given ethernet device queue, 
> that has
> - * been added to a event eth Rx adapter.
> - *
> - * @param id
> - *  The identifier of the ethernet Rx event adapter.
> - *
> - * @param eth_dev_id
> - *  The identifier of the ethernet device.
> - *
> - * @param rx_queue_id
> - *  Ethernet device receive queue index.
> - *  If rx_queue_id is -1, then all Rx queues configured for the 
> ethernet device
> - *  are configured with event vectorization.
> - *
> - * @param config
> - *  Event vector configuration structure.
> - *
> - * @return
> - *  - 0: Success, Receive queue configured correctly.
> - *  - <0: Error code on failure.
> - */
> -__rte_experimental
> -int rte_event_eth_rx_adapter_queue_event_vector_config(
> -	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
> -	struct rte_event_eth_rx_adapter_event_vector_config *config);
> -
>  #ifdef __cplusplus
>  }
>  #endif
> diff --git a/lib/eventdev/version.map b/lib/eventdev/version.map index 
> 88625621ec..cd86d2d908 100644
> --- a/lib/eventdev/version.map
> +++ b/lib/eventdev/version.map
> @@ -142,7 +142,6 @@ EXPERIMENTAL {
>  	#added in 21.05
>  	rte_event_vector_pool_create;
>  	rte_event_eth_rx_adapter_vector_limits_get;
> -	rte_event_eth_rx_adapter_queue_event_vector_config;
>  	__rte_eventdev_trace_crypto_adapter_enqueue;
>  };
> 
> --
> 2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v13] eventdev: simplify Rx adapter event vector config
  2021-08-20  7:33                             ` Naga Harish K, S V
@ 2021-09-07  8:30                               ` Jerin Jacob
  0 siblings, 0 replies; 153+ messages in thread
From: Jerin Jacob @ 2021-09-07  8:30 UTC (permalink / raw)
  To: Naga Harish K, S V
  Cc: Jayatheerthan, Jay, pbhagavatula, jerinj, Ray Kinsella,
	Shijith Thotton, dev

On Fri, Aug 20, 2021 at 1:04 PM Naga Harish K, S V
<s.v.naga.harish.k@intel.com> wrote:
>
>
>
> -----Original Message-----
> From: Jayatheerthan, Jay <jay.jayatheerthan@intel.com>
> Sent: Wednesday, August 18, 2021 1:53 PM
> To: pbhagavatula@marvell.com; jerinj@marvell.com; Ray Kinsella <mdr@ashroe.eu>; Shijith Thotton <sthotton@marvell.com>; Naga Harish K, S V <s.v.naga.harish.k@intel.com>
> Cc: dev@dpdk.org
> Subject: RE: [dpdk-dev] [PATCH v13] eventdev: simplify Rx adapter event vector config
>
> HI Harish,
> Could you review this patch ?
>
> -Jay
>
>
> > -----Original Message-----
> > From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
> > Sent: Wednesday, August 18, 2021 12:27 PM
> > To: jerinj@marvell.com; Ray Kinsella <mdr@ashroe.eu>; Pavan Nikhilesh
> > <pbhagavatula@marvell.com>; Shijith Thotton <sthotton@marvell.com>;
> > Jayatheerthan, Jay <jay.jayatheerthan@intel.com>
> > Cc: dev@dpdk.org
> > Subject: [dpdk-dev] [PATCH v13] eventdev: simplify Rx adapter event
> > vector config
> >
> > From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> >
> > Include vector configuration into the structure
> > ``rte_event_eth_rx_adapter_queue_conf`` that is used to configure Rx
> > adapter ethernet device Rx queue parameters.
> > This simplifies event vector configuration as it avoids splitting
> > configuration per Rx queue.
> >
> > Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> > Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
> > ---
> >  v13 Changes:
> >  - Fix cnxk driver compilation.
> >  v12 Changes:
> >  - Remove deprication notice.
> >  - Remove unnecessary change Id.
> >
> >  app/test-eventdev/test_pipeline_common.c |  16 +-
> >  doc/guides/rel_notes/deprecation.rst     |   9 --
> >  drivers/event/cnxk/cn10k_eventdev.c      |  77 ----------
> >  drivers/event/cnxk/cnxk_eventdev_adptr.c |  41 ++++++
> >  lib/eventdev/eventdev_pmd.h              |  29 ----
> >  lib/eventdev/rte_event_eth_rx_adapter.c  | 179
> > ++++++++---------------  lib/eventdev/rte_event_eth_rx_adapter.h  |  30 ----
> >  lib/eventdev/version.map                 |   1 -
> >  8 files changed, 104 insertions(+), 278 deletions(-)
> >
> > diff --git a/app/test-eventdev/test_pipeline_common.c
> > b/app/test-eventdev/test_pipeline_common.c
> > index 6ee530d4cd..2697547641 100644
> > --- a/app/test-eventdev/test_pipeline_common.c
> > +++ b/app/test-eventdev/test_pipeline_common.c
> > @@ -332,7 +332,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
> >       uint16_t prod;
> >       struct rte_mempool *vector_pool = NULL;
> >       struct rte_event_eth_rx_adapter_queue_conf queue_conf;
> > -     struct rte_event_eth_rx_adapter_event_vector_config vec_conf;
> >
> >       memset(&queue_conf, 0,
> >                       sizeof(struct rte_event_eth_rx_adapter_queue_conf));
> > @@ -398,8 +397,12 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
> >                       }
> >
> >                       if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
> > +                             queue_conf.vector_sz = opt->vector_size;
> > +                             queue_conf.vector_timeout_ns =
> > +                                     opt->vector_tmo_nsec;
> >                               queue_conf.rx_queue_flags |=
> >                               RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
> > +                             queue_conf.vector_mp = vector_pool;
> >                       } else {
> >                               evt_err("Rx adapter doesn't support event vector");
> >                               return -EINVAL;
> > @@ -419,17 +422,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
> >                       return ret;
> >               }
> >
> > -             if (opt->ena_vector) {
> > -                     vec_conf.vector_sz = opt->vector_size;
> > -                     vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
> > -                     vec_conf.vector_mp = vector_pool;
> > -                     if (rte_event_eth_rx_adapter_queue_event_vector_config(
> > -                                 prod, prod, -1, &vec_conf) < 0) {
> > -                             evt_err("Failed to configure event vectorization for Rx adapter");
> > -                             return -EINVAL;
> > -                     }
> > -             }
> > -
> >               if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
> >                       uint32_t service_id = -1U;
> >
> > diff --git a/doc/guides/rel_notes/deprecation.rst
> > b/doc/guides/rel_notes/deprecation.rst
> > index 76a4abfd6b..2c37d7222c 100644
> > --- a/doc/guides/rel_notes/deprecation.rst
> > +++ b/doc/guides/rel_notes/deprecation.rst
> > @@ -257,15 +257,6 @@ Deprecation Notices
> >    An 8-byte reserved field will be added to the structure ``rte_event_timer`` to
> >    support future extensions.
> >
> > -* eventdev: The structure ``rte_event_eth_rx_adapter_queue_conf``
> > will be
> > -  extended to include
> > ``rte_event_eth_rx_adapter_event_vector_config`` elements
> > -  and the function
> > ``rte_event_eth_rx_adapter_queue_event_vector_config`` will
> > -  be removed in DPDK 21.11.
> > -
> > -  An application can enable event vectorization by passing the
> > desired vector
> > -  values to the function ``rte_event_eth_rx_adapter_queue_add`` using
> > -  the structure ``rte_event_eth_rx_adapter_queue_add``.
> > -
>
> The above paragraph on how to enable Rx adapter vectorization may need to be
> added in rx adapter documentation with correct structure name.


@Pavan Nikhilesh  Could you send the next version based on @Naga
Harish K, S V  comments?
Marking as "Changes requested"



>
>
> >  * eventdev: Reserved bytes of ``rte_event_crypto_request`` is a space holder
> >    for ``response_info``. Both should be decoupled for better clarity.
> >    New space for ``response_info`` can be made by changing diff --git
> > a/drivers/event/cnxk/cn10k_eventdev.c
> > b/drivers/event/cnxk/cn10k_eventdev.c
> > index 6f37c5bd23..160192bb27 100644
> > --- a/drivers/event/cnxk/cn10k_eventdev.c
> > +++ b/drivers/event/cnxk/cn10k_eventdev.c
> > @@ -696,81 +696,6 @@ cn10k_sso_rx_adapter_vector_limits(
> >       return 0;
> >  }
> >
> > -static int
> > -cnxk_sso_rx_adapter_vwqe_enable(struct cnxk_eth_dev *cnxk_eth_dev,
> > -                             uint16_t port_id, uint16_t rq_id, uint16_t sz,
> > -                             uint64_t tmo_ns, struct rte_mempool *vmp)
> > -{
> > -     struct roc_nix_rq *rq;
> > -
> > -     rq = &cnxk_eth_dev->rqs[rq_id];
> > -
> > -     if (!rq->sso_ena)
> > -             return -EINVAL;
> > -     if (rq->flow_tag_width == 0)
> > -             return -EINVAL;
> > -
> > -     rq->vwqe_ena = 1;
> > -     rq->vwqe_first_skip = 0;
> > -     rq->vwqe_aura_handle = roc_npa_aura_handle_to_aura(vmp->pool_id);
> > -     rq->vwqe_max_sz_exp = rte_log2_u32(sz);
> > -     rq->vwqe_wait_tmo =
> > -             tmo_ns /
> > -             ((roc_nix_get_vwqe_interval(&cnxk_eth_dev->nix) + 1) * 100);
> > -     rq->tag_mask = (port_id & 0xF) << 20;
> > -     rq->tag_mask |=
> > -             (((port_id >> 4) & 0xF) | (RTE_EVENT_TYPE_ETHDEV_VECTOR << 4))
> > -             << 24;
> > -
> > -     return roc_nix_rq_modify(&cnxk_eth_dev->nix, rq, 0);
> > -}
> > -
> > -static int
> > -cn10k_sso_rx_adapter_vector_config(
> > -     const struct rte_eventdev *event_dev, const struct rte_eth_dev *eth_dev,
> > -     int32_t rx_queue_id,
> > -     const struct rte_event_eth_rx_adapter_event_vector_config *config)
> > -{
> > -     struct cnxk_eth_dev *cnxk_eth_dev;
> > -     struct cnxk_sso_evdev *dev;
> > -     int i, rc;
> > -
> > -     rc = strncmp(eth_dev->device->driver->name, "net_cn10k", 8);
> > -     if (rc)
> > -             return -EINVAL;
> > -
> > -     dev = cnxk_sso_pmd_priv(event_dev);
> > -     cnxk_eth_dev = cnxk_eth_pmd_priv(eth_dev);
> > -     if (rx_queue_id < 0) {
> > -             for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
> > -                     cnxk_sso_updt_xae_cnt(dev, config->vector_mp,
> > -                                           RTE_EVENT_TYPE_ETHDEV_VECTOR);
> > -                     rc = cnxk_sso_xae_reconfigure(
> > -                             (struct rte_eventdev *)(uintptr_t)event_dev);
> > -                     rc = cnxk_sso_rx_adapter_vwqe_enable(
> > -                             cnxk_eth_dev, eth_dev->data->port_id, i,
> > -                             config->vector_sz, config->vector_timeout_ns,
> > -                             config->vector_mp);
> > -                     if (rc)
> > -                             return -EINVAL;
> > -             }
> > -     } else {
> > -
> > -             cnxk_sso_updt_xae_cnt(dev, config->vector_mp,
> > -                                   RTE_EVENT_TYPE_ETHDEV_VECTOR);
> > -             rc = cnxk_sso_xae_reconfigure(
> > -                     (struct rte_eventdev *)(uintptr_t)event_dev);
> > -             rc = cnxk_sso_rx_adapter_vwqe_enable(
> > -                     cnxk_eth_dev, eth_dev->data->port_id, rx_queue_id,
> > -                     config->vector_sz, config->vector_timeout_ns,
> > -                     config->vector_mp);
> > -             if (rc)
> > -                     return -EINVAL;
> > -     }
> > -
> > -     return 0;
> > -}
> > -
> >  static int
> >  cn10k_sso_tx_adapter_caps_get(const struct rte_eventdev *dev,
> >                             const struct rte_eth_dev *eth_dev, uint32_t *caps) @@ -841,8
> > +766,6 @@ static struct rte_eventdev_ops cn10k_sso_dev_ops = {
> >       .eth_rx_adapter_stop = cnxk_sso_rx_adapter_stop,
> >
> >       .eth_rx_adapter_vector_limits_get = cn10k_sso_rx_adapter_vector_limits,
> > -     .eth_rx_adapter_event_vector_config =
> > -             cn10k_sso_rx_adapter_vector_config,
> >
> >       .eth_tx_adapter_caps_get = cn10k_sso_tx_adapter_caps_get,
> >       .eth_tx_adapter_queue_add = cn10k_sso_tx_adapter_queue_add, diff
> > --git a/drivers/event/cnxk/cnxk_eventdev_adptr.c
> > b/drivers/event/cnxk/cnxk_eventdev_adptr.c
> > index baf2f2aa6b..80f5602286 100644
> > --- a/drivers/event/cnxk/cnxk_eventdev_adptr.c
> > +++ b/drivers/event/cnxk/cnxk_eventdev_adptr.c
> > @@ -156,6 +156,35 @@ cnxk_sso_rxq_disable(struct cnxk_eth_dev *cnxk_eth_dev, uint16_t rq_id)
> >       return roc_nix_rq_modify(&cnxk_eth_dev->nix, rq, 0);  }
> >
> > +static int
> > +cnxk_sso_rx_adapter_vwqe_enable(struct cnxk_eth_dev *cnxk_eth_dev,
> > +                             uint16_t port_id, uint16_t rq_id, uint16_t sz,
> > +                             uint64_t tmo_ns, struct rte_mempool *vmp) {
> > +     struct roc_nix_rq *rq;
> > +
> > +     rq = &cnxk_eth_dev->rqs[rq_id];
> > +
> > +     if (!rq->sso_ena)
> > +             return -EINVAL;
> > +     if (rq->flow_tag_width == 0)
> > +             return -EINVAL;
> > +
> > +     rq->vwqe_ena = 1;
> > +     rq->vwqe_first_skip = 0;
> > +     rq->vwqe_aura_handle = roc_npa_aura_handle_to_aura(vmp->pool_id);
> > +     rq->vwqe_max_sz_exp = rte_log2_u32(sz);
> > +     rq->vwqe_wait_tmo =
> > +             tmo_ns /
> > +             ((roc_nix_get_vwqe_interval(&cnxk_eth_dev->nix) + 1) * 100);
> > +     rq->tag_mask = (port_id & 0xF) << 20;
> > +     rq->tag_mask |=
> > +             (((port_id >> 4) & 0xF) | (RTE_EVENT_TYPE_ETHDEV_VECTOR << 4))
> > +             << 24;
> > +
> > +     return roc_nix_rq_modify(&cnxk_eth_dev->nix, rq, 0); }
> > +
> >  int
> >  cnxk_sso_rx_adapter_queue_add(
> >       const struct rte_eventdev *event_dev, const struct rte_eth_dev
> > *eth_dev, @@ -183,6 +212,18 @@ cnxk_sso_rx_adapter_queue_add(
> >                       &queue_conf->ev,
> >                       !!(queue_conf->rx_queue_flags &
> >                          RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID));
> > +             if (queue_conf->rx_queue_flags &
> > +                 RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
> > +                     cnxk_sso_updt_xae_cnt(dev, queue_conf->vector_mp,
> > +                                           RTE_EVENT_TYPE_ETHDEV_VECTOR);
> > +                     rc |= cnxk_sso_xae_reconfigure(
> > +                             (struct rte_eventdev *)(uintptr_t)event_dev);
> > +                     rc |= cnxk_sso_rx_adapter_vwqe_enable(
> > +                             cnxk_eth_dev, port, rx_queue_id,
> > +                             queue_conf->vector_sz,
> > +                             queue_conf->vector_timeout_ns,
> > +                             queue_conf->vector_mp);
> > +             }
> >               rox_nix_fc_npa_bp_cfg(&cnxk_eth_dev->nix,
> >                                     rxq_sp->qconf.mp->pool_id, true,
> >                                     dev->force_ena_bp);
> > diff --git a/lib/eventdev/eventdev_pmd.h b/lib/eventdev/eventdev_pmd.h
> > index 0f724ac85d..63b3bc4b51 100644
> > --- a/lib/eventdev/eventdev_pmd.h
> > +++ b/lib/eventdev/eventdev_pmd.h
> > @@ -667,32 +667,6 @@ typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
> >       const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
> >       struct rte_event_eth_rx_adapter_vector_limits *limits);
> >
> > -struct rte_event_eth_rx_adapter_event_vector_config;
> > -/**
> > - * Enable event vector on an given Rx queue of a ethernet devices
> > belonging to
> > - * the Rx adapter.
> > - *
> > - * @param dev
> > - *   Event device pointer
> > - *
> > - * @param eth_dev
> > - *   Ethernet device pointer
> > - *
> > - * @param rx_queue_id
> > - *   The Rx queue identifier
> > - *
> > - * @param config
> > - *   Pointer to the event vector configuration structure.
> > - *
> > - * @return
> > - *   - 0: Success.
> > - *   - <0: Error code returned by the driver function.
> > - */
> > -typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
> > -     const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
> > -     int32_t rx_queue_id,
> > -     const struct rte_event_eth_rx_adapter_event_vector_config *config);
> > -
> >  typedef uint32_t rte_event_pmd_selftest_seqn_t;  extern int
> > rte_event_pmd_selftest_seqn_dynfield_offset;
> >
> > @@ -1118,9 +1092,6 @@ struct rte_eventdev_ops {
> >       eventdev_eth_rx_adapter_vector_limits_get_t
> >               eth_rx_adapter_vector_limits_get;
> >       /**< Get event vector limits for the Rx adapter */
> > -     eventdev_eth_rx_adapter_event_vector_config_t
> > -             eth_rx_adapter_event_vector_config;
> > -     /**< Configure Rx adapter with event vector */
> >
> >       eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
> >       /**< Get timer adapter capabilities */ diff --git
> > a/lib/eventdev/rte_event_eth_rx_adapter.c
> > b/lib/eventdev/rte_event_eth_rx_adapter.c
> > index 13dfb28401..2b2dd688fc 100644
> > --- a/lib/eventdev/rte_event_eth_rx_adapter.c
> > +++ b/lib/eventdev/rte_event_eth_rx_adapter.c
> > @@ -1895,6 +1895,24 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
> >       } else
> >               qi_ev->flow_id = 0;
> >
> > +     if (conf->rx_queue_flags &
> > +         RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
> > +             queue_info->ena_vector = 1;
> > +             qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
> > +             rxa_set_vector_data(queue_info, conf->vector_sz,
> > +                                 conf->vector_timeout_ns, conf->vector_mp,
> > +                                 rx_queue_id, dev_info->dev->data->port_id);
> > +             rx_adapter->ena_vector = 1;
> > +             rx_adapter->vector_tmo_ticks =
> > +                     rx_adapter->vector_tmo_ticks ?
> > +                                   RTE_MIN(queue_info->vector_data
> > +                                                     .vector_timeout_ticks >>
> > +                                             1,
> > +                                     rx_adapter->vector_tmo_ticks) :
> > +                             queue_info->vector_data.vector_timeout_ticks >>
> > +                                     1;
> > +     }
> > +
> >       rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
> >       if (rxa_polled_queue(dev_info, rx_queue_id)) {
> >               rx_adapter->num_rx_polled += !pollq; @@ -1920,42 +1938,6 @@
> > rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
> >       }
> >  }
> >
> > -static void
> > -rxa_sw_event_vector_configure(
> > -     struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
> > -     int rx_queue_id,
> > -     const struct rte_event_eth_rx_adapter_event_vector_config *config)
> > -{
> > -     struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
> > -     struct eth_rx_queue_info *queue_info;
> > -     struct rte_event *qi_ev;
> > -
> > -     if (rx_queue_id == -1) {
> > -             uint16_t nb_rx_queues;
> > -             uint16_t i;
> > -
> > -             nb_rx_queues = dev_info->dev->data->nb_rx_queues;
> > -             for (i = 0; i < nb_rx_queues; i++)
> > -                     rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
> > -                                                   config);
> > -             return;
> > -     }
> > -
> > -     queue_info = &dev_info->rx_queue[rx_queue_id];
> > -     qi_ev = (struct rte_event *)&queue_info->event;
> > -     queue_info->ena_vector = 1;
> > -     qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
> > -     rxa_set_vector_data(queue_info, config->vector_sz,
> > -                         config->vector_timeout_ns, config->vector_mp,
> > -                         rx_queue_id, dev_info->dev->data->port_id);
> > -     rx_adapter->ena_vector = 1;
> > -     rx_adapter->vector_tmo_ticks =
> > -             rx_adapter->vector_tmo_ticks ?
> > -                           RTE_MIN(config->vector_timeout_ns >> 1,
> > -                             rx_adapter->vector_tmo_ticks) :
> > -                           config->vector_timeout_ns >> 1;
> > -}
> > -
> >  static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
> >               uint16_t eth_dev_id,
> >               int rx_queue_id,
> > @@ -2270,6 +2252,7 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
> >       struct rte_event_eth_rx_adapter *rx_adapter;
> >       struct rte_eventdev *dev;
> >       struct eth_device_info *dev_info;
> > +     struct rte_event_eth_rx_adapter_vector_limits limits;
> >
> >       RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
> >       RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL); @@ -2297,13
> > +2280,46 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
> >               return -EINVAL;
> >       }
> >
> > -     if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
> > -         (queue_conf->rx_queue_flags &
> > -          RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
> > -             RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
> > -                              " eth port: %" PRIu16 " adapter id: %" PRIu8,
> > -                              eth_dev_id, id);
> > -             return -EINVAL;
> > +     if (queue_conf->rx_queue_flags &
> > +         RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
> > +
> > +             if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0) {
> > +                     RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
> > +                                      " eth port: %" PRIu16
> > +                                      " adapter id: %" PRIu8,
> > +                                      eth_dev_id, id);
> > +                     return -EINVAL;
> > +             }
> > +
> > +             ret = rte_event_eth_rx_adapter_vector_limits_get(
> > +                     rx_adapter->eventdev_id, eth_dev_id, &limits);
> > +             if (ret < 0) {
> > +                     RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
> > +                                      " eth port: %" PRIu16
> > +                                      " adapter id: %" PRIu8,
> > +                                      eth_dev_id, id);
> > +                     return -EINVAL;
> > +             }
> > +             if (queue_conf->vector_sz < limits.min_sz ||
> > +                 queue_conf->vector_sz > limits.max_sz ||
> > +                 queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
> > +                 queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
> > +                 queue_conf->vector_mp == NULL) {
> > +                     RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> > +                                      " eth port: %" PRIu16
> > +                                      " adapter id: %" PRIu8,
> > +                                      eth_dev_id, id);
> > +                     return -EINVAL;
> > +             }
> > +             if (queue_conf->vector_mp->elt_size <
> > +                 (sizeof(struct rte_event_vector) +
> > +                  (sizeof(uintptr_t) * queue_conf->vector_sz))) {
> > +                     RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> > +                                      " eth port: %" PRIu16
> > +                                      " adapter id: %" PRIu8,
> > +                                      eth_dev_id, id);
> > +                     return -EINVAL;
> > +             }
> >       }
> >
> >       if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 && @@
> > -2499,83 +2515,6 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
> >       return ret;
> >  }
> >
> > -int
> > -rte_event_eth_rx_adapter_queue_event_vector_config(
> > -     uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
> > -     struct rte_event_eth_rx_adapter_event_vector_config *config)
> > -{
> > -     struct rte_event_eth_rx_adapter_vector_limits limits;
> > -     struct rte_event_eth_rx_adapter *rx_adapter;
> > -     struct rte_eventdev *dev;
> > -     uint32_t cap;
> > -     int ret;
> > -
> > -     RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
> > -     RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
> > -
> > -     rx_adapter = rxa_id_to_adapter(id);
> > -     if ((rx_adapter == NULL) || (config == NULL))
> > -             return -EINVAL;
> > -
> > -     dev = &rte_eventdevs[rx_adapter->eventdev_id];
> > -     ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
> > -                                             eth_dev_id, &cap);
> > -     if (ret) {
> > -             RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
> > -                              "eth port %" PRIu16,
> > -                              id, eth_dev_id);
> > -             return ret;
> > -     }
> > -
> > -     if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
> > -             RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
> > -                              " eth port: %" PRIu16 " adapter id: %" PRIu8,
> > -                              eth_dev_id, id);
> > -             return -EINVAL;
> > -     }
> > -
> > -     ret = rte_event_eth_rx_adapter_vector_limits_get(
> > -             rx_adapter->eventdev_id, eth_dev_id, &limits);
> > -     if (ret) {
> > -             RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
> > -                              "eth port %" PRIu16,
> > -                              rx_adapter->eventdev_id, eth_dev_id);
> > -             return ret;
> > -     }
> > -
> > -     if (config->vector_sz < limits.min_sz ||
> > -         config->vector_sz > limits.max_sz ||
> > -         config->vector_timeout_ns < limits.min_timeout_ns ||
> > -         config->vector_timeout_ns > limits.max_timeout_ns ||
> > -         config->vector_mp == NULL) {
> > -             RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> > -                              " eth port: %" PRIu16 " adapter id: %" PRIu8,
> > -                              eth_dev_id, id);
> > -             return -EINVAL;
> > -     }
> > -     if (config->vector_mp->elt_size <
> > -         (sizeof(struct rte_event_vector) +
> > -          (sizeof(uintptr_t) * config->vector_sz))) {
> > -             RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
> > -                              " eth port: %" PRIu16 " adapter id: %" PRIu8,
> > -                              eth_dev_id, id);
> > -             return -EINVAL;
> > -     }
> > -
> > -     if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
> > -             RTE_FUNC_PTR_OR_ERR_RET(
> > -                     *dev->dev_ops->eth_rx_adapter_event_vector_config,
> > -                     -ENOTSUP);
> > -             ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
> > -                     dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
> > -     } else {
> > -             rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
> > -                                           rx_queue_id, config);
> > -     }
> > -
> > -     return ret;
> > -}
> > -
> >  int
> >  rte_event_eth_rx_adapter_vector_limits_get(
> >       uint8_t dev_id, uint16_t eth_port_id, diff --git
> > a/lib/eventdev/rte_event_eth_rx_adapter.h
> > b/lib/eventdev/rte_event_eth_rx_adapter.h
> > index 182dd2e5dd..3f8b362295 100644
> > --- a/lib/eventdev/rte_event_eth_rx_adapter.h
> > +++ b/lib/eventdev/rte_event_eth_rx_adapter.h
> > @@ -171,9 +171,6 @@ struct rte_event_eth_rx_adapter_queue_conf {
> >        * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
> >        * enqueued event.
> >        */
> > -};
> > -
> > -struct rte_event_eth_rx_adapter_event_vector_config {
> >       uint16_t vector_sz;
> >       /**<
> >        * Indicates the maximum number for mbufs to combine and form a vector.
> > @@ -548,33 +545,6 @@ int rte_event_eth_rx_adapter_vector_limits_get(
> >       uint8_t dev_id, uint16_t eth_port_id,
> >       struct rte_event_eth_rx_adapter_vector_limits *limits);
> >
> > -/**
> > - * Configure event vectorization for a given ethernet device queue,
> > that has
> > - * been added to a event eth Rx adapter.
> > - *
> > - * @param id
> > - *  The identifier of the ethernet Rx event adapter.
> > - *
> > - * @param eth_dev_id
> > - *  The identifier of the ethernet device.
> > - *
> > - * @param rx_queue_id
> > - *  Ethernet device receive queue index.
> > - *  If rx_queue_id is -1, then all Rx queues configured for the
> > ethernet device
> > - *  are configured with event vectorization.
> > - *
> > - * @param config
> > - *  Event vector configuration structure.
> > - *
> > - * @return
> > - *  - 0: Success, Receive queue configured correctly.
> > - *  - <0: Error code on failure.
> > - */
> > -__rte_experimental
> > -int rte_event_eth_rx_adapter_queue_event_vector_config(
> > -     uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
> > -     struct rte_event_eth_rx_adapter_event_vector_config *config);
> > -
> >  #ifdef __cplusplus
> >  }
> >  #endif
> > diff --git a/lib/eventdev/version.map b/lib/eventdev/version.map index
> > 88625621ec..cd86d2d908 100644
> > --- a/lib/eventdev/version.map
> > +++ b/lib/eventdev/version.map
> > @@ -142,7 +142,6 @@ EXPERIMENTAL {
> >       #added in 21.05
> >       rte_event_vector_pool_create;
> >       rte_event_eth_rx_adapter_vector_limits_get;
> > -     rte_event_eth_rx_adapter_queue_event_vector_config;
> >       __rte_eventdev_trace_crypto_adapter_enqueue;
> >  };
> >
> > --
> > 2.17.1
>

^ permalink raw reply	[flat|nested] 153+ messages in thread

* [dpdk-dev] [PATCH v14] eventdev: simplify Rx adapter event vector config
  2021-08-18  6:57                         ` [dpdk-dev] [PATCH v13] " pbhagavatula
  2021-08-18  8:22                           ` Jayatheerthan, Jay
@ 2021-09-15 13:15                           ` pbhagavatula
  2021-09-15 13:18                             ` Kinsella, Ray
  1 sibling, 1 reply; 153+ messages in thread
From: pbhagavatula @ 2021-09-15 13:15 UTC (permalink / raw)
  To: jerinj, Jay Jayatheerthan, Ray Kinsella, Pavan Nikhilesh,
	Shijith Thotton
  Cc: dev

From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Include vector configuration into the structure
``rte_event_eth_rx_adapter_queue_conf`` that is used to configure
Rx adapter ethernet device Rx queue parameters.
This simplifies event vector configuration as it avoids splitting
configuration per Rx queue.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
---
 v14 Changes:
 - Update documentation.
 v13 Changes:
 - Fix cnxk driver compilation.
 v12 Changes:
 - Remove deprication notice.
 - Remove unnecessary change Id.

 app/test-eventdev/test_pipeline_common.c      |  16 +-
 .../prog_guide/event_ethernet_rx_adapter.rst  |  11 +-
 doc/guides/rel_notes/deprecation.rst          |   9 -
 drivers/event/cnxk/cn10k_eventdev.c           |  77 --------
 drivers/event/cnxk/cnxk_eventdev_adptr.c      |  41 ++++
 lib/eventdev/eventdev_pmd.h                   |  29 ---
 lib/eventdev/rte_event_eth_rx_adapter.c       | 179 ++++++------------
 lib/eventdev/rte_event_eth_rx_adapter.h       |  30 ---
 lib/eventdev/version.map                      |   1 -
 9 files changed, 112 insertions(+), 281 deletions(-)

diff --git a/app/test-eventdev/test_pipeline_common.c b/app/test-eventdev/test_pipeline_common.c
index 6ee530d4cd..2697547641 100644
--- a/app/test-eventdev/test_pipeline_common.c
+++ b/app/test-eventdev/test_pipeline_common.c
@@ -332,7 +332,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 	uint16_t prod;
 	struct rte_mempool *vector_pool = NULL;
 	struct rte_event_eth_rx_adapter_queue_conf queue_conf;
-	struct rte_event_eth_rx_adapter_event_vector_config vec_conf;

 	memset(&queue_conf, 0,
 			sizeof(struct rte_event_eth_rx_adapter_queue_conf));
@@ -398,8 +397,12 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			}

 			if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) {
+				queue_conf.vector_sz = opt->vector_size;
+				queue_conf.vector_timeout_ns =
+					opt->vector_tmo_nsec;
 				queue_conf.rx_queue_flags |=
 				RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR;
+				queue_conf.vector_mp = vector_pool;
 			} else {
 				evt_err("Rx adapter doesn't support event vector");
 				return -EINVAL;
@@ -419,17 +422,6 @@ pipeline_event_rx_adapter_setup(struct evt_options *opt, uint8_t stride,
 			return ret;
 		}

-		if (opt->ena_vector) {
-			vec_conf.vector_sz = opt->vector_size;
-			vec_conf.vector_timeout_ns = opt->vector_tmo_nsec;
-			vec_conf.vector_mp = vector_pool;
-			if (rte_event_eth_rx_adapter_queue_event_vector_config(
-				    prod, prod, -1, &vec_conf) < 0) {
-				evt_err("Failed to configure event vectorization for Rx adapter");
-				return -EINVAL;
-			}
-		}
-
 		if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT)) {
 			uint32_t service_id = -1U;

diff --git a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
index c01e5a9666..0780b6f711 100644
--- a/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
+++ b/doc/guides/prog_guide/event_ethernet_rx_adapter.rst
@@ -195,12 +195,17 @@ The event devices, ethernet device pairs which support the capability
 flow characteristics and generate a ``rte_event`` containing ``rte_event_vector``
 whose event type is either ``RTE_EVENT_TYPE_ETHDEV_VECTOR`` or
 ``RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR``.
-The aggregation size and timeout are configurable at a queue level and the
-maximum, minimum vector sizes and timeouts vary based on the device capability
-and can be queried using ``rte_event_eth_rx_adapter_vector_limits_get``.
+The maximum, minimum vector sizes and timeouts vary based on the device
+capability and can be queried using
+``rte_event_eth_rx_adapter_vector_limits_get``.
 The Rx adapter additionally might include useful data such as ethernet device
 port and queue identifier in the ``rte_event_vector::port`` and
 ``rte_event_vector::queue`` and mark ``rte_event_vector::attr_valid`` as true.
+The aggregation size and timeout are configurable at a queue level by setting
+``rte_event_eth_rx_adapter_queue_conf::vector_sz``,
+``rte_event_eth_rx_adapter_queue_conf::vector_timeout_ns`` and
+``rte_event_eth_rx_adapter_queue_conf::vector_mp`` when adding queues using
+``rte_event_eth_rx_adapter_queue_add``.

 A loop processing ``rte_event_vector`` containing mbufs is shown below.

diff --git a/doc/guides/rel_notes/deprecation.rst b/doc/guides/rel_notes/deprecation.rst
index 59445a6f42..bf1e07c0a8 100644
--- a/doc/guides/rel_notes/deprecation.rst
+++ b/doc/guides/rel_notes/deprecation.rst
@@ -254,15 +254,6 @@ Deprecation Notices
   An 8-byte reserved field will be added to the structure ``rte_event_timer`` to
   support future extensions.

-* eventdev: The structure ``rte_event_eth_rx_adapter_queue_conf`` will be
-  extended to include ``rte_event_eth_rx_adapter_event_vector_config`` elements
-  and the function ``rte_event_eth_rx_adapter_queue_event_vector_config`` will
-  be removed in DPDK 21.11.
-
-  An application can enable event vectorization by passing the desired vector
-  values to the function ``rte_event_eth_rx_adapter_queue_add`` using
-  the structure ``rte_event_eth_rx_adapter_queue_add``.
-
 * eventdev: Reserved bytes of ``rte_event_crypto_request`` is a space holder
   for ``response_info``. Both should be decoupled for better clarity.
   New space for ``response_info`` can be made by changing
diff --git a/drivers/event/cnxk/cn10k_eventdev.c b/drivers/event/cnxk/cn10k_eventdev.c
index 8af273a01b..a8ab4747f7 100644
--- a/drivers/event/cnxk/cn10k_eventdev.c
+++ b/drivers/event/cnxk/cn10k_eventdev.c
@@ -690,81 +690,6 @@ cn10k_sso_rx_adapter_vector_limits(
 	return 0;
 }

-static int
-cnxk_sso_rx_adapter_vwqe_enable(struct cnxk_eth_dev *cnxk_eth_dev,
-				uint16_t port_id, uint16_t rq_id, uint16_t sz,
-				uint64_t tmo_ns, struct rte_mempool *vmp)
-{
-	struct roc_nix_rq *rq;
-
-	rq = &cnxk_eth_dev->rqs[rq_id];
-
-	if (!rq->sso_ena)
-		return -EINVAL;
-	if (rq->flow_tag_width == 0)
-		return -EINVAL;
-
-	rq->vwqe_ena = 1;
-	rq->vwqe_first_skip = 0;
-	rq->vwqe_aura_handle = roc_npa_aura_handle_to_aura(vmp->pool_id);
-	rq->vwqe_max_sz_exp = rte_log2_u32(sz);
-	rq->vwqe_wait_tmo =
-		tmo_ns /
-		((roc_nix_get_vwqe_interval(&cnxk_eth_dev->nix) + 1) * 100);
-	rq->tag_mask = (port_id & 0xF) << 20;
-	rq->tag_mask |=
-		(((port_id >> 4) & 0xF) | (RTE_EVENT_TYPE_ETHDEV_VECTOR << 4))
-		<< 24;
-
-	return roc_nix_rq_modify(&cnxk_eth_dev->nix, rq, 0);
-}
-
-static int
-cn10k_sso_rx_adapter_vector_config(
-	const struct rte_eventdev *event_dev, const struct rte_eth_dev *eth_dev,
-	int32_t rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct cnxk_eth_dev *cnxk_eth_dev;
-	struct cnxk_sso_evdev *dev;
-	int i, rc;
-
-	rc = strncmp(eth_dev->device->driver->name, "net_cn10k", 8);
-	if (rc)
-		return -EINVAL;
-
-	dev = cnxk_sso_pmd_priv(event_dev);
-	cnxk_eth_dev = cnxk_eth_pmd_priv(eth_dev);
-	if (rx_queue_id < 0) {
-		for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
-			cnxk_sso_updt_xae_cnt(dev, config->vector_mp,
-					      RTE_EVENT_TYPE_ETHDEV_VECTOR);
-			rc = cnxk_sso_xae_reconfigure(
-				(struct rte_eventdev *)(uintptr_t)event_dev);
-			rc = cnxk_sso_rx_adapter_vwqe_enable(
-				cnxk_eth_dev, eth_dev->data->port_id, i,
-				config->vector_sz, config->vector_timeout_ns,
-				config->vector_mp);
-			if (rc)
-				return -EINVAL;
-		}
-	} else {
-
-		cnxk_sso_updt_xae_cnt(dev, config->vector_mp,
-				      RTE_EVENT_TYPE_ETHDEV_VECTOR);
-		rc = cnxk_sso_xae_reconfigure(
-			(struct rte_eventdev *)(uintptr_t)event_dev);
-		rc = cnxk_sso_rx_adapter_vwqe_enable(
-			cnxk_eth_dev, eth_dev->data->port_id, rx_queue_id,
-			config->vector_sz, config->vector_timeout_ns,
-			config->vector_mp);
-		if (rc)
-			return -EINVAL;
-	}
-
-	return 0;
-}
-
 static int
 cn10k_sso_tx_adapter_caps_get(const struct rte_eventdev *dev,
 			      const struct rte_eth_dev *eth_dev, uint32_t *caps)
@@ -878,8 +803,6 @@ static struct rte_eventdev_ops cn10k_sso_dev_ops = {
 	.eth_rx_adapter_stop = cnxk_sso_rx_adapter_stop,

 	.eth_rx_adapter_vector_limits_get = cn10k_sso_rx_adapter_vector_limits,
-	.eth_rx_adapter_event_vector_config =
-		cn10k_sso_rx_adapter_vector_config,

 	.eth_tx_adapter_caps_get = cn10k_sso_tx_adapter_caps_get,
 	.eth_tx_adapter_queue_add = cn10k_sso_tx_adapter_queue_add,
diff --git a/drivers/event/cnxk/cnxk_eventdev_adptr.c b/drivers/event/cnxk/cnxk_eventdev_adptr.c
index baf2f2aa6b..80f5602286 100644
--- a/drivers/event/cnxk/cnxk_eventdev_adptr.c
+++ b/drivers/event/cnxk/cnxk_eventdev_adptr.c
@@ -156,6 +156,35 @@ cnxk_sso_rxq_disable(struct cnxk_eth_dev *cnxk_eth_dev, uint16_t rq_id)
 	return roc_nix_rq_modify(&cnxk_eth_dev->nix, rq, 0);
 }

+static int
+cnxk_sso_rx_adapter_vwqe_enable(struct cnxk_eth_dev *cnxk_eth_dev,
+				uint16_t port_id, uint16_t rq_id, uint16_t sz,
+				uint64_t tmo_ns, struct rte_mempool *vmp)
+{
+	struct roc_nix_rq *rq;
+
+	rq = &cnxk_eth_dev->rqs[rq_id];
+
+	if (!rq->sso_ena)
+		return -EINVAL;
+	if (rq->flow_tag_width == 0)
+		return -EINVAL;
+
+	rq->vwqe_ena = 1;
+	rq->vwqe_first_skip = 0;
+	rq->vwqe_aura_handle = roc_npa_aura_handle_to_aura(vmp->pool_id);
+	rq->vwqe_max_sz_exp = rte_log2_u32(sz);
+	rq->vwqe_wait_tmo =
+		tmo_ns /
+		((roc_nix_get_vwqe_interval(&cnxk_eth_dev->nix) + 1) * 100);
+	rq->tag_mask = (port_id & 0xF) << 20;
+	rq->tag_mask |=
+		(((port_id >> 4) & 0xF) | (RTE_EVENT_TYPE_ETHDEV_VECTOR << 4))
+		<< 24;
+
+	return roc_nix_rq_modify(&cnxk_eth_dev->nix, rq, 0);
+}
+
 int
 cnxk_sso_rx_adapter_queue_add(
 	const struct rte_eventdev *event_dev, const struct rte_eth_dev *eth_dev,
@@ -183,6 +212,18 @@ cnxk_sso_rx_adapter_queue_add(
 			&queue_conf->ev,
 			!!(queue_conf->rx_queue_flags &
 			   RTE_EVENT_ETH_RX_ADAPTER_CAP_OVERRIDE_FLOW_ID));
+		if (queue_conf->rx_queue_flags &
+		    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+			cnxk_sso_updt_xae_cnt(dev, queue_conf->vector_mp,
+					      RTE_EVENT_TYPE_ETHDEV_VECTOR);
+			rc |= cnxk_sso_xae_reconfigure(
+				(struct rte_eventdev *)(uintptr_t)event_dev);
+			rc |= cnxk_sso_rx_adapter_vwqe_enable(
+				cnxk_eth_dev, port, rx_queue_id,
+				queue_conf->vector_sz,
+				queue_conf->vector_timeout_ns,
+				queue_conf->vector_mp);
+		}
 		rox_nix_fc_npa_bp_cfg(&cnxk_eth_dev->nix,
 				      rxq_sp->qconf.mp->pool_id, true,
 				      dev->force_ena_bp);
diff --git a/lib/eventdev/eventdev_pmd.h b/lib/eventdev/eventdev_pmd.h
index 0f724ac85d..63b3bc4b51 100644
--- a/lib/eventdev/eventdev_pmd.h
+++ b/lib/eventdev/eventdev_pmd.h
@@ -667,32 +667,6 @@ typedef int (*eventdev_eth_rx_adapter_vector_limits_get_t)(
 	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-struct rte_event_eth_rx_adapter_event_vector_config;
-/**
- * Enable event vector on an given Rx queue of a ethernet devices belonging to
- * the Rx adapter.
- *
- * @param dev
- *   Event device pointer
- *
- * @param eth_dev
- *   Ethernet device pointer
- *
- * @param rx_queue_id
- *   The Rx queue identifier
- *
- * @param config
- *   Pointer to the event vector configuration structure.
- *
- * @return
- *   - 0: Success.
- *   - <0: Error code returned by the driver function.
- */
-typedef int (*eventdev_eth_rx_adapter_event_vector_config_t)(
-	const struct rte_eventdev *dev, const struct rte_eth_dev *eth_dev,
-	int32_t rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 typedef uint32_t rte_event_pmd_selftest_seqn_t;
 extern int rte_event_pmd_selftest_seqn_dynfield_offset;

@@ -1118,9 +1092,6 @@ struct rte_eventdev_ops {
 	eventdev_eth_rx_adapter_vector_limits_get_t
 		eth_rx_adapter_vector_limits_get;
 	/**< Get event vector limits for the Rx adapter */
-	eventdev_eth_rx_adapter_event_vector_config_t
-		eth_rx_adapter_event_vector_config;
-	/**< Configure Rx adapter with event vector */

 	eventdev_timer_adapter_caps_get_t timer_adapter_caps_get;
 	/**< Get timer adapter capabilities */
diff --git a/lib/eventdev/rte_event_eth_rx_adapter.c b/lib/eventdev/rte_event_eth_rx_adapter.c
index 13dfb28401..2b2dd688fc 100644
--- a/lib/eventdev/rte_event_eth_rx_adapter.c
+++ b/lib/eventdev/rte_event_eth_rx_adapter.c
@@ -1895,6 +1895,24 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	} else
 		qi_ev->flow_id = 0;

+	if (conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+		queue_info->ena_vector = 1;
+		qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
+		rxa_set_vector_data(queue_info, conf->vector_sz,
+				    conf->vector_timeout_ns, conf->vector_mp,
+				    rx_queue_id, dev_info->dev->data->port_id);
+		rx_adapter->ena_vector = 1;
+		rx_adapter->vector_tmo_ticks =
+			rx_adapter->vector_tmo_ticks ?
+				      RTE_MIN(queue_info->vector_data
+							.vector_timeout_ticks >>
+						1,
+					rx_adapter->vector_tmo_ticks) :
+				queue_info->vector_data.vector_timeout_ticks >>
+					1;
+	}
+
 	rxa_update_queue(rx_adapter, dev_info, rx_queue_id, 1);
 	if (rxa_polled_queue(dev_info, rx_queue_id)) {
 		rx_adapter->num_rx_polled += !pollq;
@@ -1920,42 +1938,6 @@ rxa_add_queue(struct rte_event_eth_rx_adapter *rx_adapter,
 	}
 }

-static void
-rxa_sw_event_vector_configure(
-	struct rte_event_eth_rx_adapter *rx_adapter, uint16_t eth_dev_id,
-	int rx_queue_id,
-	const struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct eth_device_info *dev_info = &rx_adapter->eth_devices[eth_dev_id];
-	struct eth_rx_queue_info *queue_info;
-	struct rte_event *qi_ev;
-
-	if (rx_queue_id == -1) {
-		uint16_t nb_rx_queues;
-		uint16_t i;
-
-		nb_rx_queues = dev_info->dev->data->nb_rx_queues;
-		for (i = 0; i < nb_rx_queues; i++)
-			rxa_sw_event_vector_configure(rx_adapter, eth_dev_id, i,
-						      config);
-		return;
-	}
-
-	queue_info = &dev_info->rx_queue[rx_queue_id];
-	qi_ev = (struct rte_event *)&queue_info->event;
-	queue_info->ena_vector = 1;
-	qi_ev->event_type = RTE_EVENT_TYPE_ETH_RX_ADAPTER_VECTOR;
-	rxa_set_vector_data(queue_info, config->vector_sz,
-			    config->vector_timeout_ns, config->vector_mp,
-			    rx_queue_id, dev_info->dev->data->port_id);
-	rx_adapter->ena_vector = 1;
-	rx_adapter->vector_tmo_ticks =
-		rx_adapter->vector_tmo_ticks ?
-			      RTE_MIN(config->vector_timeout_ns >> 1,
-				rx_adapter->vector_tmo_ticks) :
-			      config->vector_timeout_ns >> 1;
-}
-
 static int rxa_sw_add(struct rte_event_eth_rx_adapter *rx_adapter,
 		uint16_t eth_dev_id,
 		int rx_queue_id,
@@ -2270,6 +2252,7 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 	struct rte_event_eth_rx_adapter *rx_adapter;
 	struct rte_eventdev *dev;
 	struct eth_device_info *dev_info;
+	struct rte_event_eth_rx_adapter_vector_limits limits;

 	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
 	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
@@ -2297,13 +2280,46 @@ rte_event_eth_rx_adapter_queue_add(uint8_t id,
 		return -EINVAL;
 	}

-	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0 &&
-	    (queue_conf->rx_queue_flags &
-	     RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR)) {
-		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
+	if (queue_conf->rx_queue_flags &
+	    RTE_EVENT_ETH_RX_ADAPTER_QUEUE_EVENT_VECTOR) {
+
+		if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR) == 0) {
+			RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+
+		ret = rte_event_eth_rx_adapter_vector_limits_get(
+			rx_adapter->eventdev_id, eth_dev_id, &limits);
+		if (ret < 0) {
+			RTE_EDEV_LOG_ERR("Failed to get event device vector limits,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_sz < limits.min_sz ||
+		    queue_conf->vector_sz > limits.max_sz ||
+		    queue_conf->vector_timeout_ns < limits.min_timeout_ns ||
+		    queue_conf->vector_timeout_ns > limits.max_timeout_ns ||
+		    queue_conf->vector_mp == NULL) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
+		if (queue_conf->vector_mp->elt_size <
+		    (sizeof(struct rte_event_vector) +
+		     (sizeof(uintptr_t) * queue_conf->vector_sz))) {
+			RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
+					 " eth port: %" PRIu16
+					 " adapter id: %" PRIu8,
+					 eth_dev_id, id);
+			return -EINVAL;
+		}
 	}

 	if ((cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_MULTI_EVENTQ) == 0 &&
@@ -2499,83 +2515,6 @@ rte_event_eth_rx_adapter_queue_del(uint8_t id, uint16_t eth_dev_id,
 	return ret;
 }

-int
-rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config)
-{
-	struct rte_event_eth_rx_adapter_vector_limits limits;
-	struct rte_event_eth_rx_adapter *rx_adapter;
-	struct rte_eventdev *dev;
-	uint32_t cap;
-	int ret;
-
-	RTE_EVENT_ETH_RX_ADAPTER_ID_VALID_OR_ERR_RET(id, -EINVAL);
-	RTE_ETH_VALID_PORTID_OR_ERR_RET(eth_dev_id, -EINVAL);
-
-	rx_adapter = rxa_id_to_adapter(id);
-	if ((rx_adapter == NULL) || (config == NULL))
-		return -EINVAL;
-
-	dev = &rte_eventdevs[rx_adapter->eventdev_id];
-	ret = rte_event_eth_rx_adapter_caps_get(rx_adapter->eventdev_id,
-						eth_dev_id, &cap);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get adapter caps edev %" PRIu8
-				 "eth port %" PRIu16,
-				 id, eth_dev_id);
-		return ret;
-	}
-
-	if (!(cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_EVENT_VECTOR)) {
-		RTE_EDEV_LOG_ERR("Event vectorization is not supported,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	ret = rte_event_eth_rx_adapter_vector_limits_get(
-		rx_adapter->eventdev_id, eth_dev_id, &limits);
-	if (ret) {
-		RTE_EDEV_LOG_ERR("Failed to get vector limits edev %" PRIu8
-				 "eth port %" PRIu16,
-				 rx_adapter->eventdev_id, eth_dev_id);
-		return ret;
-	}
-
-	if (config->vector_sz < limits.min_sz ||
-	    config->vector_sz > limits.max_sz ||
-	    config->vector_timeout_ns < limits.min_timeout_ns ||
-	    config->vector_timeout_ns > limits.max_timeout_ns ||
-	    config->vector_mp == NULL) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-	if (config->vector_mp->elt_size <
-	    (sizeof(struct rte_event_vector) +
-	     (sizeof(uintptr_t) * config->vector_sz))) {
-		RTE_EDEV_LOG_ERR("Invalid event vector configuration,"
-				 " eth port: %" PRIu16 " adapter id: %" PRIu8,
-				 eth_dev_id, id);
-		return -EINVAL;
-	}
-
-	if (cap & RTE_EVENT_ETH_RX_ADAPTER_CAP_INTERNAL_PORT) {
-		RTE_FUNC_PTR_OR_ERR_RET(
-			*dev->dev_ops->eth_rx_adapter_event_vector_config,
-			-ENOTSUP);
-		ret = dev->dev_ops->eth_rx_adapter_event_vector_config(
-			dev, &rte_eth_devices[eth_dev_id], rx_queue_id, config);
-	} else {
-		rxa_sw_event_vector_configure(rx_adapter, eth_dev_id,
-					      rx_queue_id, config);
-	}
-
-	return ret;
-}
-
 int
 rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
diff --git a/lib/eventdev/rte_event_eth_rx_adapter.h b/lib/eventdev/rte_event_eth_rx_adapter.h
index 182dd2e5dd..3f8b362295 100644
--- a/lib/eventdev/rte_event_eth_rx_adapter.h
+++ b/lib/eventdev/rte_event_eth_rx_adapter.h
@@ -171,9 +171,6 @@ struct rte_event_eth_rx_adapter_queue_conf {
 	 * The event adapter sets ev.event_type to RTE_EVENT_TYPE_ETHDEV in the
 	 * enqueued event.
 	 */
-};
-
-struct rte_event_eth_rx_adapter_event_vector_config {
 	uint16_t vector_sz;
 	/**<
 	 * Indicates the maximum number for mbufs to combine and form a vector.
@@ -548,33 +545,6 @@ int rte_event_eth_rx_adapter_vector_limits_get(
 	uint8_t dev_id, uint16_t eth_port_id,
 	struct rte_event_eth_rx_adapter_vector_limits *limits);

-/**
- * Configure event vectorization for a given ethernet device queue, that has
- * been added to a event eth Rx adapter.
- *
- * @param id
- *  The identifier of the ethernet Rx event adapter.
- *
- * @param eth_dev_id
- *  The identifier of the ethernet device.
- *
- * @param rx_queue_id
- *  Ethernet device receive queue index.
- *  If rx_queue_id is -1, then all Rx queues configured for the ethernet device
- *  are configured with event vectorization.
- *
- * @param config
- *  Event vector configuration structure.
- *
- * @return
- *  - 0: Success, Receive queue configured correctly.
- *  - <0: Error code on failure.
- */
-__rte_experimental
-int rte_event_eth_rx_adapter_queue_event_vector_config(
-	uint8_t id, uint16_t eth_dev_id, int32_t rx_queue_id,
-	struct rte_event_eth_rx_adapter_event_vector_config *config);
-
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/eventdev/version.map b/lib/eventdev/version.map
index 88625621ec..cd86d2d908 100644
--- a/lib/eventdev/version.map
+++ b/lib/eventdev/version.map
@@ -142,7 +142,6 @@ EXPERIMENTAL {
 	#added in 21.05
 	rte_event_vector_pool_create;
 	rte_event_eth_rx_adapter_vector_limits_get;
-	rte_event_eth_rx_adapter_queue_event_vector_config;
 	__rte_eventdev_trace_crypto_adapter_enqueue;
 };

--
2.17.1


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v14] eventdev: simplify Rx adapter event vector config
  2021-09-15 13:15                           ` [dpdk-dev] [PATCH v14] " pbhagavatula
@ 2021-09-15 13:18                             ` Kinsella, Ray
  2021-09-16  4:28                               ` Jerin Jacob
  0 siblings, 1 reply; 153+ messages in thread
From: Kinsella, Ray @ 2021-09-15 13:18 UTC (permalink / raw)
  To: pbhagavatula, jerinj, Jay Jayatheerthan, Shijith Thotton; +Cc: dev



On 15/09/2021 14:15, pbhagavatula@marvell.com wrote:
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> 
> Include vector configuration into the structure
> ``rte_event_eth_rx_adapter_queue_conf`` that is used to configure
> Rx adapter ethernet device Rx queue parameters.
> This simplifies event vector configuration as it avoids splitting
> configuration per Rx queue.
> 
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
> ---
>  v14 Changes:
>  - Update documentation.
>  v13 Changes:
>  - Fix cnxk driver compilation.
>  v12 Changes:
>  - Remove deprication notice.
>  - Remove unnecessary change Id.
> 
>  app/test-eventdev/test_pipeline_common.c      |  16 +-
>  .../prog_guide/event_ethernet_rx_adapter.rst  |  11 +-
>  doc/guides/rel_notes/deprecation.rst          |   9 -
>  drivers/event/cnxk/cn10k_eventdev.c           |  77 --------
>  drivers/event/cnxk/cnxk_eventdev_adptr.c      |  41 ++++
>  lib/eventdev/eventdev_pmd.h                   |  29 ---
>  lib/eventdev/rte_event_eth_rx_adapter.c       | 179 ++++++------------
>  lib/eventdev/rte_event_eth_rx_adapter.h       |  30 ---
>  lib/eventdev/version.map                      |   1 -
>  9 files changed, 112 insertions(+), 281 deletions(-)

FYI - deprication is spelt deprecation 

Acked-by: Ray Kinsella <mdr@ashroe.eu>


^ permalink raw reply	[flat|nested] 153+ messages in thread

* Re: [dpdk-dev] [PATCH v14] eventdev: simplify Rx adapter event vector config
  2021-09-15 13:18                             ` Kinsella, Ray
@ 2021-09-16  4:28                               ` Jerin Jacob
  0 siblings, 0 replies; 153+ messages in thread
From: Jerin Jacob @ 2021-09-16  4:28 UTC (permalink / raw)
  To: Kinsella, Ray
  Cc: Pavan Nikhilesh, Jerin Jacob, Jay Jayatheerthan, Shijith Thotton,
	dpdk-dev

On Wed, Sep 15, 2021 at 6:48 PM Kinsella, Ray <mdr@ashroe.eu> wrote:
>
>
>
> On 15/09/2021 14:15, pbhagavatula@marvell.com wrote:
> > From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> >
> > Include vector configuration into the structure
> > ``rte_event_eth_rx_adapter_queue_conf`` that is used to configure
> > Rx adapter ethernet device Rx queue parameters.
> > This simplifies event vector configuration as it avoids splitting
> > configuration per Rx queue.
> >
> > Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> > Acked-by: Jay Jayatheerthan <jay.jayatheerthan@intel.com>
> > ---
> >  v14 Changes:
> >  - Update documentation.
> >  v13 Changes:
> >  - Fix cnxk driver compilation.
> >  v12 Changes:
> >  - Remove deprication notice.
> >  - Remove unnecessary change Id.
> >
> >  app/test-eventdev/test_pipeline_common.c      |  16 +-
> >  .../prog_guide/event_ethernet_rx_adapter.rst  |  11 +-
> >  doc/guides/rel_notes/deprecation.rst          |   9 -
> >  drivers/event/cnxk/cn10k_eventdev.c           |  77 --------
> >  drivers/event/cnxk/cnxk_eventdev_adptr.c      |  41 ++++
> >  lib/eventdev/eventdev_pmd.h                   |  29 ---
> >  lib/eventdev/rte_event_eth_rx_adapter.c       | 179 ++++++------------
> >  lib/eventdev/rte_event_eth_rx_adapter.h       |  30 ---
> >  lib/eventdev/version.map                      |   1 -
> >  9 files changed, 112 insertions(+), 281 deletions(-)
>
> FYI - deprication is spelt deprecation
>
> Acked-by: Ray Kinsella <mdr@ashroe.eu>

Acked-by: Jerin Jacob <jerinj@marvell.com>
Applied to dpdk-next-eventdev/for-main. Thanks.


>

^ permalink raw reply	[flat|nested] 153+ messages in thread

end of thread, other threads:[~2021-09-16  4:29 UTC | newest]

Thread overview: 153+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-02-20 22:09 [dpdk-dev] [PATCH 0/7] Introduce event vectorization pbhagavatula
2021-02-20 22:09 ` [dpdk-dev] [PATCH 1/7] eventdev: introduce event vector capability pbhagavatula
2021-03-08 16:49   ` Jerin Jacob
2021-02-20 22:09 ` [dpdk-dev] [PATCH 2/7] eventdev: introduce event vector Rx capability pbhagavatula
2021-03-08 17:07   ` Jerin Jacob
2021-02-20 22:09 ` [dpdk-dev] [PATCH 3/7] eventdev: introduce event vector Tx capability pbhagavatula
2021-03-08 17:09   ` Jerin Jacob
2021-02-20 22:09 ` [dpdk-dev] [PATCH 4/7] eventdev: add Rx adapter event vector support pbhagavatula
2021-03-08 17:27   ` Jerin Jacob
2021-03-16 10:41     ` Jayatheerthan, Jay
2021-02-20 22:09 ` [dpdk-dev] [PATCH 5/7] eventdev: add Tx " pbhagavatula
2021-02-20 22:09 ` [dpdk-dev] [PATCH 6/7] app/eventdev: add event vector mode in pipeline test pbhagavatula
2021-02-20 22:09 ` [dpdk-dev] [PATCH 7/7] eventdev: fix ABI breakage due to event vector pbhagavatula
2021-03-08 18:44   ` Jerin Jacob
2021-03-12 14:28     ` David Marchand
2021-03-16  5:54       ` [dpdk-dev] [EXT] " Pavan Nikhilesh Bhagavatula
2021-03-15 10:01     ` [dpdk-dev] " Kinsella, Ray
2021-03-08 16:41 ` [dpdk-dev] [PATCH 0/7] Introduce event vectorization Jerin Jacob
2021-03-16 15:48 ` [dpdk-dev] [PATCH v2 0/8] " pbhagavatula
2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 1/8] eventdev: introduce event vector capability pbhagavatula
2021-03-16 17:48     ` Jerin Jacob
2021-03-16 19:17       ` [dpdk-dev] [EXT] " Pavan Nikhilesh Bhagavatula
2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 2/8] eventdev: introduce event vector Rx capability pbhagavatula
2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 3/8] eventdev: introduce event vector Tx capability pbhagavatula
2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 4/8] eventdev: add Rx adapter event vector support pbhagavatula
2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 5/8] eventdev: add Tx " pbhagavatula
2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
2021-03-16 15:48   ` [dpdk-dev] [PATCH v2 7/8] doc: announce event Rx adapter config changes pbhagavatula
2021-03-16 15:48   ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v2 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
2021-03-16 20:01   ` [dpdk-dev] [PATCH v3 0/8] Introduce event vectorization pbhagavatula
2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 1/8] eventdev: introduce event vector capability pbhagavatula
2021-03-18  6:19       ` Jayatheerthan, Jay
2021-03-18  6:23         ` Pavan Nikhilesh Bhagavatula
2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 2/8] eventdev: introduce event vector Rx capability pbhagavatula
2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 3/8] eventdev: introduce event vector Tx capability pbhagavatula
2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 4/8] eventdev: add Rx adapter event vector support pbhagavatula
2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 5/8] eventdev: add Tx " pbhagavatula
2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
2021-03-16 20:01     ` [dpdk-dev] [PATCH v3 7/8] doc: announce event Rx adapter config changes pbhagavatula
2021-03-16 20:01     ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v3 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
2021-03-19 20:57     ` [dpdk-dev] [PATCH v4 0/8] Introduce event vectorization pbhagavatula
2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 1/8] eventdev: introduce event vector capability pbhagavatula
2021-03-22  9:06         ` Kinsella, Ray
2021-03-22  9:10           ` [dpdk-dev] [EXT] " Pavan Nikhilesh Bhagavatula
2021-03-23 11:12         ` [dpdk-dev] " Jerin Jacob
2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 2/8] eventdev: introduce event vector Rx capability pbhagavatula
2021-03-22  9:12         ` Kinsella, Ray
2021-03-22 10:07           ` [dpdk-dev] [EXT] " Pavan Nikhilesh Bhagavatula
2021-03-22 11:07             ` Kinsella, Ray
2021-03-23 16:56         ` [dpdk-dev] " Jerin Jacob
2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 3/8] eventdev: introduce event vector Tx capability pbhagavatula
2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 4/8] eventdev: add Rx adapter event vector support pbhagavatula
2021-03-23 18:30         ` Jerin Jacob
2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 5/8] eventdev: add Tx " pbhagavatula
2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
2021-03-23 18:39         ` Jerin Jacob
2021-03-19 20:57       ` [dpdk-dev] [PATCH v4 7/8] doc: announce event Rx adapter config changes pbhagavatula
2021-03-19 20:57       ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v4 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
2021-03-23 18:44       ` [dpdk-dev] [PATCH v4 0/8] Introduce event vectorization Jerin Jacob
2021-03-24  5:05       ` [dpdk-dev] [PATCH v5 " pbhagavatula
2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 1/8] eventdev: introduce event vector capability pbhagavatula
2021-03-24  6:48           ` Jayatheerthan, Jay
2021-03-24 18:20             ` Pavan Nikhilesh Bhagavatula
2021-03-24  9:16           ` Kinsella, Ray
2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 2/8] eventdev: introduce event vector Rx capability pbhagavatula
2021-03-24  9:15           ` Kinsella, Ray
2021-03-25  8:15           ` Jayatheerthan, Jay
2021-03-25  9:24             ` Pavan Nikhilesh Bhagavatula
2021-03-25  9:50               ` Jayatheerthan, Jay
2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 3/8] eventdev: introduce event vector Tx capability pbhagavatula
2021-03-25  8:16           ` Jayatheerthan, Jay
2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 4/8] eventdev: add Rx adapter event vector support pbhagavatula
2021-03-25 10:37           ` Jayatheerthan, Jay
2021-03-25 13:14             ` Pavan Nikhilesh Bhagavatula
2021-03-26  6:26               ` Jayatheerthan, Jay
2021-03-26  9:00                 ` Pavan Nikhilesh Bhagavatula
2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 5/8] eventdev: add Tx " pbhagavatula
2021-03-25 11:44           ` Jayatheerthan, Jay
2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
2021-03-24  5:05         ` [dpdk-dev] [PATCH v5 7/8] doc: announce event Rx adapter config changes pbhagavatula
2021-03-24  9:16           ` Kinsella, Ray
2021-03-24  5:05         ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v5 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
2021-03-25 12:27           ` Jayatheerthan, Jay
2021-03-25 13:55             ` Pavan Nikhilesh Bhagavatula
2021-03-26  7:09               ` Jayatheerthan, Jay
2021-03-26  9:44                 ` Pavan Nikhilesh Bhagavatula
2021-03-24  5:39         ` [dpdk-dev] [PATCH v5 0/8] Introduce event vectorization Jayatheerthan, Jay
2021-03-24  6:44           ` Pavan Nikhilesh Bhagavatula
2021-03-24  8:10             ` Jayatheerthan, Jay
2021-03-24 19:28         ` [dpdk-dev] [PATCH v6 " pbhagavatula
2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 1/8] eventdev: introduce event vector capability pbhagavatula
2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 2/8] eventdev: introduce event vector Rx capability pbhagavatula
2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 3/8] eventdev: introduce event vector Tx capability pbhagavatula
2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 4/8] eventdev: add Rx adapter event vector support pbhagavatula
2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 5/8] eventdev: add Tx " pbhagavatula
2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
2021-03-24 19:28           ` [dpdk-dev] [PATCH v6 7/8] doc: announce event Rx adapter config changes pbhagavatula
2021-03-24 19:28           ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v6 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
2021-03-25 17:10           ` [dpdk-dev] [PATCH v7 0/8] Introduce event vectorization pbhagavatula
2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 1/8] eventdev: introduce event vector capability pbhagavatula
2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 2/8] eventdev: introduce event vector Rx capability pbhagavatula
2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 3/8] eventdev: introduce event vector Tx capability pbhagavatula
2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 4/8] eventdev: add Rx adapter event vector support pbhagavatula
2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 5/8] eventdev: add Tx " pbhagavatula
2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
2021-03-25 17:10             ` [dpdk-dev] [PATCH v7 7/8] doc: announce event Rx adapter config changes pbhagavatula
2021-03-25 17:10             ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v7 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
2021-03-26 14:08             ` [dpdk-dev] [PATCH v8 0/8] Introduce event vectorization pbhagavatula
2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 1/8] eventdev: introduce event vector capability pbhagavatula
2021-03-27 12:07                 ` Jayatheerthan, Jay
2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 2/8] eventdev: introduce event vector Rx capability pbhagavatula
2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 3/8] eventdev: introduce event vector Tx capability pbhagavatula
2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 4/8] eventdev: add Rx adapter event vector support pbhagavatula
2021-03-28  8:18                 ` Jerin Jacob
2021-03-29  6:09                   ` Jayatheerthan, Jay
2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 5/8] eventdev: add Tx " pbhagavatula
2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
2021-03-26 14:08               ` [dpdk-dev] [PATCH v8 7/8] doc: announce event Rx adapter config changes pbhagavatula
2021-03-26 14:43                 ` Jerin Jacob
2021-03-27 12:07                   ` Jayatheerthan, Jay
2021-03-26 14:08               ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v8 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
2021-03-30  8:22               ` [dpdk-dev] [PATCH v9 0/8] Introduce event vectorization pbhagavatula
2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 1/8] eventdev: introduce event vector capability pbhagavatula
2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 2/8] eventdev: introduce event vector Rx capability pbhagavatula
2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 3/8] eventdev: introduce event vector Tx capability pbhagavatula
2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 4/8] eventdev: add Rx adapter event vector support pbhagavatula
2021-03-31  6:35                   ` Jayatheerthan, Jay
2021-03-31  6:40                     ` Pavan Nikhilesh Bhagavatula
2021-03-31  6:55                       ` Jayatheerthan, Jay
2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 5/8] eventdev: add Tx " pbhagavatula
2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
2021-03-30  8:22                 ` [dpdk-dev] [PATCH v9 7/8] doc: announce event Rx adapter config changes pbhagavatula
2021-03-30  8:22                 ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v9 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
2021-03-31  6:55                   ` Jayatheerthan, Jay
2021-03-31  9:29                 ` [dpdk-dev] [PATCH v10 0/8] Introduce event vectorization pbhagavatula
2021-03-31  9:29                   ` [dpdk-dev] [PATCH v10 1/8] eventdev: introduce event vector capability pbhagavatula
2021-03-31  9:29                   ` [dpdk-dev] [PATCH v10 2/8] eventdev: introduce event vector Rx capability pbhagavatula
2021-03-31  9:29                   ` [dpdk-dev] [PATCH v10 3/8] eventdev: introduce event vector Tx capability pbhagavatula
2021-03-31  9:29                   ` [dpdk-dev] [PATCH v10 4/8] eventdev: add Rx adapter event vector support pbhagavatula
2021-03-31  9:29                   ` [dpdk-dev] [PATCH v10 5/8] eventdev: add Tx " pbhagavatula
2021-03-31  9:30                   ` [dpdk-dev] [PATCH v10 6/8] app/eventdev: add event vector mode in pipeline test pbhagavatula
2021-03-31  9:30                   ` [dpdk-dev] [PATCH v10 7/8] doc: announce event Rx adapter config changes pbhagavatula
2021-03-31  9:30                   ` [dpdk-dev] [dpdk-dev v21.11] [PATCH v10 8/8] eventdev: simplify Rx adapter event vector config pbhagavatula
2021-08-18  4:56                     ` [dpdk-dev] [PATCH v11] " pbhagavatula
2021-08-18  4:59                       ` [dpdk-dev] [PATCH v12] " pbhagavatula
2021-08-18  6:57                         ` [dpdk-dev] [PATCH v13] " pbhagavatula
2021-08-18  8:22                           ` Jayatheerthan, Jay
2021-08-20  7:33                             ` Naga Harish K, S V
2021-09-07  8:30                               ` Jerin Jacob
2021-09-15 13:15                           ` [dpdk-dev] [PATCH v14] " pbhagavatula
2021-09-15 13:18                             ` Kinsella, Ray
2021-09-16  4:28                               ` Jerin Jacob
2021-04-03  9:44                   ` [dpdk-dev] [PATCH v10 0/8] Introduce event vectorization Jerin Jacob

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).