From: "Mattias Rönnblom" <mattias.ronnblom@ericsson.com>
To: <jerinj@marvell.com>, <abhinandan.gujjar@intel.com>,
<erik.g.carrillo@intel.com>, <jay.jayatheerthan@intel.com>
Cc: dev@dpdk.org, jerinjacobk@gmail.com, harry.van.haaren@intel.com,
timothy.mcdaniel@intel.com, pbhagavatula@marvell.com,
hemant.agrawal@nxp.com, liangma@liangbit.com,
bruce.richardson@intel.com,
"Mattias Rönnblom" <mattias.ronnblom@ericsson.com>
Subject: [dpdk-dev] [PATCH v3 2/3] event/dsw: make use of eventdev maintenance facility
Date: Mon, 1 Nov 2021 19:40:15 +0100 [thread overview]
Message-ID: <20211101184016.8073-2-mattias.ronnblom@ericsson.com> (raw)
In-Reply-To: <20211101184016.8073-1-mattias.ronnblom@ericsson.com>
Set the RTE_EVENT_DEV_CAP_REQUIRES_MAINT flag, and perform DSW
background tasks on rte_event_maintain() calls.
PATCH v3:
- Add support for output buffer flushing by means of the new
rte_event_maintain() op parameter, and RTE_EVENT_DEV_MAINT_OP_FLUSH.
PATCH v2:
- Update DSW documentation.
RFC v2:
- Have dsw_event_maintain() occasionally flush the port output
buffers.
Signed-off-by: Mattias Rönnblom <mattias.ronnblom@ericsson.com>
Tested-by: Richard Eklycke <richard.eklycke@ericsson.com>
Tested-by: Liron Himi <lironh@marvell.com>
---
doc/guides/eventdevs/dsw.rst | 27 +++++++++++++++------------
drivers/event/dsw/dsw_evdev.c | 4 +++-
drivers/event/dsw/dsw_evdev.h | 1 +
drivers/event/dsw/dsw_event.c | 12 ++++++++++++
4 files changed, 31 insertions(+), 13 deletions(-)
diff --git a/doc/guides/eventdevs/dsw.rst b/doc/guides/eventdevs/dsw.rst
index 6653f501c1..18f7e9588f 100644
--- a/doc/guides/eventdevs/dsw.rst
+++ b/doc/guides/eventdevs/dsw.rst
@@ -40,20 +40,20 @@ Example:
Limitations
-----------
-Unattended Ports
+Port Maintenance
~~~~~~~~~~~~~~~~
-The distributed software eventdev uses an internal signaling schema
-between the ports to achieve load balancing. In order for this to
-work, the application must perform enqueue and/or dequeue operations
-on all ports.
+The distributed software eventdev uses an internal signaling scheme
+between the ports to achieve load balancing. Therefore, it sets the
+``RTE_EVENT_DEV_CAP_REQUIRES_MAINT`` flag.
-Producer-only ports which currently have no events to enqueue should
-periodically call rte_event_enqueue_burst() with a zero-sized burst.
+During periods when the application thread using a particular port is
+neither attempting to enqueue nor to dequeue events, it must
+repeatedly call rte_event_maintain() on that port.
-Ports left unattended for longer periods of time will prevent load
-balancing, and also cause traffic interruptions on the flows which
-are in the process of being migrated.
+Ports left unmaintained for long periods of time will prevent load
+balancing and cause traffic interruptions on flows which are in the
+process of being migrated.
Output Buffering
~~~~~~~~~~~~~~~~
@@ -66,8 +66,11 @@ In case no more events are enqueued on a port with buffered events,
these events will be sent after the application has performed a number
of enqueue and/or dequeue operations.
-For explicit flushing, an application may call
-rte_event_enqueue_burst() with a zero-sized burst.
+To immediately flush a port's output buffer, an application may call
+rte_event_maintain() with op set to ``RTE_EVENT_DEV_MAINT_OP_FLUSH``.
+
+Repeated calls to rte_event_maintain() will also flush the output
+buffers.
Priorities
diff --git a/drivers/event/dsw/dsw_evdev.c b/drivers/event/dsw/dsw_evdev.c
index 0652d83ad6..5ff8fcc6a9 100644
--- a/drivers/event/dsw/dsw_evdev.c
+++ b/drivers/event/dsw/dsw_evdev.c
@@ -222,7 +222,8 @@ dsw_info_get(struct rte_eventdev *dev __rte_unused,
RTE_EVENT_DEV_CAP_DISTRIBUTED_SCHED|
RTE_EVENT_DEV_CAP_NONSEQ_MODE|
RTE_EVENT_DEV_CAP_MULTIPLE_QUEUE_PORT|
- RTE_EVENT_DEV_CAP_CARRY_FLOW_ID
+ RTE_EVENT_DEV_CAP_CARRY_FLOW_ID|
+ RTE_EVENT_DEV_CAP_REQUIRES_MAINT
};
}
@@ -441,6 +442,7 @@ dsw_probe(struct rte_vdev_device *vdev)
dev->enqueue_forward_burst = dsw_event_enqueue_forward_burst;
dev->dequeue = dsw_event_dequeue;
dev->dequeue_burst = dsw_event_dequeue_burst;
+ dev->maintain = dsw_event_maintain;
if (rte_eal_process_type() != RTE_PROC_PRIMARY)
return 0;
diff --git a/drivers/event/dsw/dsw_evdev.h b/drivers/event/dsw/dsw_evdev.h
index 631daea55c..e64ae26f6e 100644
--- a/drivers/event/dsw/dsw_evdev.h
+++ b/drivers/event/dsw/dsw_evdev.h
@@ -271,6 +271,7 @@ uint16_t dsw_event_enqueue_forward_burst(void *port,
uint16_t dsw_event_dequeue(void *port, struct rte_event *ev, uint64_t wait);
uint16_t dsw_event_dequeue_burst(void *port, struct rte_event *events,
uint16_t num, uint64_t wait);
+void dsw_event_maintain(void *port, int op);
int dsw_xstats_get_names(const struct rte_eventdev *dev,
enum rte_event_dev_xstats_mode mode,
diff --git a/drivers/event/dsw/dsw_event.c b/drivers/event/dsw/dsw_event.c
index 1f09816945..c6ed470286 100644
--- a/drivers/event/dsw/dsw_event.c
+++ b/drivers/event/dsw/dsw_event.c
@@ -1400,3 +1400,15 @@ dsw_event_dequeue_burst(void *port, struct rte_event *events, uint16_t num,
return dequeued;
}
+
+void dsw_event_maintain(void *port, int op)
+{
+ struct dsw_port *source_port = port;
+ struct dsw_evdev *dsw = source_port->dsw;
+
+ dsw_port_note_op(source_port, 0);
+ dsw_port_bg_process(dsw, source_port);
+
+ if (op & RTE_EVENT_DEV_MAINT_OP_FLUSH)
+ dsw_port_flush_out_buffers(dsw, source_port);
+}
--
2.25.1
next prev parent reply other threads:[~2021-11-01 18:40 UTC|newest]
Thread overview: 40+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-04-08 17:56 [dpdk-dev] [RFC 1/3] eventdev: allow for event devices requiring maintenance Mattias Rönnblom
2020-04-08 17:56 ` [dpdk-dev] [RFC 2/3] event/dsw: make use of eventdev maintenance facility Mattias Rönnblom
2020-04-08 17:56 ` [dpdk-dev] [RFC 3/3] eventdev: allow devices requiring maintenance with adapters Mattias Rönnblom
2020-04-08 19:36 ` [dpdk-dev] [RFC 1/3] eventdev: allow for event devices requiring maintenance Jerin Jacob
2020-04-09 12:21 ` Mattias Rönnblom
2020-04-09 13:32 ` Jerin Jacob
2020-04-09 14:02 ` Mattias Rönnblom
2020-04-10 13:00 ` Jerin Jacob
2020-04-14 15:57 ` Mattias Rönnblom
2020-04-14 16:15 ` Jerin Jacob
2020-04-14 17:55 ` Mattias Rönnblom
2020-04-16 17:19 ` Jerin Jacob
2020-04-20 9:05 ` Mattias Rönnblom
2020-05-13 18:56 ` Mattias Rönnblom
2021-08-02 16:14 ` [dpdk-dev] [RFC v2 " Mattias Rönnblom
2021-08-02 16:15 ` [dpdk-dev] [RFC v2 2/3] event/dsw: make use of eventdev maintenance facility Mattias Rönnblom
2021-08-02 16:15 ` [dpdk-dev] [RFC v2 3/3] eventdev: have adapters support device maintenance Mattias Rönnblom
2021-08-03 4:39 ` [dpdk-dev] [RFC v2 1/3] eventdev: allow for event devices requiring maintenance Jerin Jacob
2021-08-03 8:26 ` Mattias Rönnblom
2021-08-03 10:06 ` [dpdk-dev] [EXT] " Jerin Jacob Kollanukkaran
2021-10-26 17:31 ` [dpdk-dev] [PATCH " Mattias Rönnblom
2021-10-26 17:31 ` [dpdk-dev] [PATCH 2/3] event/dsw: make use of eventdev maintenance facility Mattias Rönnblom
2021-10-26 17:31 ` [dpdk-dev] [PATCH 3/3] eventdev: have adapters support device maintenance Mattias Rönnblom
2021-10-29 14:38 ` [dpdk-dev] [PATCH 1/3] eventdev: allow for event devices requiring maintenance Jerin Jacob
2021-10-29 15:03 ` Mattias Rönnblom
2021-10-29 15:17 ` Jerin Jacob
2021-10-29 16:02 ` Van Haaren, Harry
2021-10-31 9:29 ` Mattias Rönnblom
2021-10-30 17:19 ` Mattias Rönnblom
2021-10-31 13:17 ` Jerin Jacob
2021-11-01 13:28 ` [dpdk-dev] [PATCH v2 " Mattias Rönnblom
2021-11-01 13:28 ` [dpdk-dev] [PATCH v2 2/3] event/dsw: make use of eventdev maintenance facility Mattias Rönnblom
2021-11-01 13:28 ` [dpdk-dev] [PATCH v2 3/3] eventdev: have adapters support device maintenance Mattias Rönnblom
2021-11-04 12:33 ` [dpdk-dev] [PATCH 1/3] eventdev: allow for event devices requiring maintenance Jerin Jacob
2021-11-01 9:26 ` Mattias Rönnblom
2021-11-01 18:40 ` [dpdk-dev] [PATCH v3 " Mattias Rönnblom
2021-11-01 18:40 ` Mattias Rönnblom [this message]
2021-11-01 18:40 ` [dpdk-dev] [PATCH v3 3/3] eventdev: have adapters support device maintenance Mattias Rönnblom
2020-04-09 13:33 ` [dpdk-dev] [RFC 1/3] eventdev: allow for event devices requiring maintenance Eads, Gage
2020-04-09 14:14 ` Mattias Rönnblom
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20211101184016.8073-2-mattias.ronnblom@ericsson.com \
--to=mattias.ronnblom@ericsson.com \
--cc=abhinandan.gujjar@intel.com \
--cc=bruce.richardson@intel.com \
--cc=dev@dpdk.org \
--cc=erik.g.carrillo@intel.com \
--cc=harry.van.haaren@intel.com \
--cc=hemant.agrawal@nxp.com \
--cc=jay.jayatheerthan@intel.com \
--cc=jerinj@marvell.com \
--cc=jerinjacobk@gmail.com \
--cc=liangma@liangbit.com \
--cc=pbhagavatula@marvell.com \
--cc=timothy.mcdaniel@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
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).