From: ilia.kurakin@intel.com
To: thomas@monjalon.net, john.mcnamara@intel.com
Cc: dev@dpdk.org, roman.p.khatko@intel.com,
Ilia Kurakin <ilia.kurakin@intel.com>
Subject: [dpdk-dev] [PATCH] ethdev: change vtune profiling approach
Date: Tue, 3 Jul 2018 16:11:28 +0300 [thread overview]
Message-ID: <1530623488-69056-1-git-send-email-ilia.kurakin@intel.com> (raw)
From: Ilia Kurakin <ilia.kurakin@intel.com>
The patch changes rx_burst profiling approach:
1. VTune's instrumentation is removed
2. empty hook callback for profiling is added
This way all VTune-specific logic moves to the VTune side. Hook is enabled
only when CONFIG_RTE_ETHDEV_PROFILE_WITH_VTUNE option is turned on. VTune uses
this hook to attach to the polling cycle. It is not possible to attach to the
rx_burst directly, as it is inline.
Signed-off-by: Ilia Kurakin <ilia.kurakin@intel.com>
---
config/common_base | 2 +-
doc/guides/prog_guide/profile_app.rst | 34 ++---------
lib/librte_ethdev/ethdev_profile.c | 103 +++++-----------------------------
lib/librte_ethdev/ethdev_profile.h | 6 +-
lib/librte_ethdev/rte_ethdev.c | 4 +-
5 files changed, 25 insertions(+), 124 deletions(-)
diff --git a/config/common_base b/config/common_base
index 721e59b..2bdd895 100644
--- a/config/common_base
+++ b/config/common_base
@@ -128,7 +128,7 @@ CONFIG_RTE_MAX_QUEUES_PER_PORT=1024
CONFIG_RTE_LIBRTE_IEEE1588=n
CONFIG_RTE_ETHDEV_QUEUE_STAT_CNTRS=16
CONFIG_RTE_ETHDEV_RXTX_CALLBACKS=y
-CONFIG_RTE_ETHDEV_PROFILE_ITT_WASTED_RX_ITERATIONS=n
+CONFIG_RTE_ETHDEV_PROFILE_WITH_VTUNE=n
#
# Turn off Tx preparation stage
diff --git a/doc/guides/prog_guide/profile_app.rst b/doc/guides/prog_guide/profile_app.rst
index 1106216..02f0561 100644
--- a/doc/guides/prog_guide/profile_app.rst
+++ b/doc/guides/prog_guide/profile_app.rst
@@ -33,38 +33,12 @@ Refer to the
for details about application profiling.
-Empty cycles tracing
+Profiling with VTune
~~~~~~~~~~~~~~~~~~~~
-Iterations that yielded no RX packets (empty cycles, wasted iterations) can
-be analyzed using VTune Amplifier. This profiling employs the
-`Instrumentation and Tracing Technology (ITT) API
-<https://software.intel.com/en-us/node/544195>`_
-feature of VTune Amplifier and requires only reconfiguring the DPDK library,
-no changes in a DPDK application are needed.
-
-To trace wasted iterations on RX queues, first reconfigure DPDK with
-``CONFIG_RTE_ETHDEV_RXTX_CALLBACKS`` and
-``CONFIG_RTE_ETHDEV_PROFILE_ITT_WASTED_RX_ITERATIONS`` enabled.
-
-Then rebuild DPDK, specifying paths to the ITT header and library, which can
-be found in any VTune Amplifier distribution in the *include* and *lib*
-directories respectively:
-
-.. code-block:: console
-
- make EXTRA_CFLAGS=-I<path to ittnotify.h> \
- EXTRA_LDLIBS="-L<path to libittnotify.a> -littnotify"
-
-Finally, to see wasted iterations in your performance analysis results,
-select the *"Analyze user tasks, events, and counters"* checkbox in the
-*"Analysis Type"* tab when configuring analysis via VTune Amplifier GUI.
-Alternatively, when running VTune Amplifier via command line, specify
-``-knob enable-user-tasks=true`` option.
-
-Collected regions of wasted iterations will be marked on VTune Amplifier's
-timeline as ITT tasks. These ITT tasks have predefined names, containing
-Ethernet device and RX queue identifiers.
+To allow VTune attaching to the DPDK application, reconfigure and recompile
+the DPDK with ``CONFIG_RTE_ETHDEV_RXTX_CALLBACKS`` and
+``CONFIG_RTE_ETHDEV_PROFILE_WITH_VTUNE`` enabled.
Profiling on ARM64
diff --git a/lib/librte_ethdev/ethdev_profile.c b/lib/librte_ethdev/ethdev_profile.c
index 0d1dcda..a3c303f 100644
--- a/lib/librte_ethdev/ethdev_profile.c
+++ b/lib/librte_ethdev/ethdev_profile.c
@@ -1,87 +1,33 @@
/* SPDX-License-Identifier: BSD-3-Clause
- * Copyright(c) 2010-2017 Intel Corporation
+ * Copyright(c) 2010-2018 Intel Corporation
*/
#include "ethdev_profile.h"
/**
- * This conditional block enables RX queues profiling by tracking wasted
- * iterations, i.e. iterations which yielded no RX packets. Profiling is
- * performed using the Instrumentation and Tracing Technology (ITT) API,
- * employed by the Intel (R) VTune (TM) Amplifier.
+ * This conditional block enables Ethernet device profiling with
+ * Intel (R) VTune (TM) Amplifier.
*/
-#ifdef RTE_ETHDEV_PROFILE_ITT_WASTED_RX_ITERATIONS
-
-#include <ittnotify.h>
-
-#define ITT_MAX_NAME_LEN (100)
-
-/**
- * Auxiliary ITT structure belonging to Ethernet device and using to:
- * - track RX queue state to determine whether it is wasting loop iterations
- * - begin or end ITT task using task domain and task name (handle)
- */
-struct itt_profile_rx_data {
- /**
- * ITT domains for each queue.
- */
- __itt_domain *domains[RTE_MAX_QUEUES_PER_PORT];
- /**
- * ITT task names for each queue.
- */
- __itt_string_handle *handles[RTE_MAX_QUEUES_PER_PORT];
- /**
- * Flags indicating the queues state. Possible values:
- * 1 - queue is wasting iterations,
- * 0 - otherwise.
- */
- uint8_t queue_state[RTE_MAX_QUEUES_PER_PORT];
-};
-
-/**
- * The pool of *itt_profile_rx_data* structures.
- */
-struct itt_profile_rx_data itt_rx_data[RTE_MAX_ETHPORTS];
-
+#ifdef RTE_ETHDEV_PROFILE_WITH_VTUNE
/**
- * This callback function manages ITT tasks collection on given port and queue.
- * It must be registered with rte_eth_add_rx_callback() to be called from
- * rte_eth_rx_burst(). To find more comments see rte_rx_callback_fn function
- * type declaration.
+ * Hook callback to trace rte_eth_rx_burst() calls.
*/
-static uint16_t
-collect_itt_rx_burst_cb(uint16_t port_id, uint16_t queue_id,
+uint16_t
+profile_hook_rx_burst_cb(
+ __rte_unused uint16_t port_id, __rte_unused uint16_t queue_id,
__rte_unused struct rte_mbuf *pkts[], uint16_t nb_pkts,
__rte_unused uint16_t max_pkts, __rte_unused void *user_param)
{
- if (unlikely(nb_pkts == 0)) {
- if (!itt_rx_data[port_id].queue_state[queue_id]) {
- __itt_task_begin(
- itt_rx_data[port_id].domains[queue_id],
- __itt_null, __itt_null,
- itt_rx_data[port_id].handles[queue_id]);
- itt_rx_data[port_id].queue_state[queue_id] = 1;
- }
- } else {
- if (unlikely(itt_rx_data[port_id].queue_state[queue_id])) {
- __itt_task_end(
- itt_rx_data[port_id].domains[queue_id]);
- itt_rx_data[port_id].queue_state[queue_id] = 0;
- }
- }
return nb_pkts;
}
/**
- * Initialization of itt_profile_rx_data for a given Ethernet device.
+ * Setting profiling rx callback for a given Ethernet device.
* This function must be invoked when ethernet device is being configured.
- * Result will be stored in the global array *itt_rx_data*.
*
* @param port_id
* The port identifier of the Ethernet device.
- * @param port_name
- * The name of the Ethernet device.
* @param rx_queue_num
* The number of RX queues on specified port.
*
@@ -90,46 +36,27 @@ collect_itt_rx_burst_cb(uint16_t port_id, uint16_t queue_id,
* - On failure, a negative value.
*/
static inline int
-itt_profile_rx_init(uint16_t port_id, char *port_name, uint8_t rx_queue_num)
+vtune_profile_rx_init(uint16_t port_id, uint8_t rx_queue_num)
{
uint16_t q_id;
for (q_id = 0; q_id < rx_queue_num; ++q_id) {
- char domain_name[ITT_MAX_NAME_LEN];
-
- snprintf(domain_name, sizeof(domain_name),
- "RXBurst.WastedIterations.Port_%s.Queue_%d",
- port_name, q_id);
- itt_rx_data[port_id].domains[q_id]
- = __itt_domain_create(domain_name);
-
- char task_name[ITT_MAX_NAME_LEN];
-
- snprintf(task_name, sizeof(task_name),
- "port id: %d; queue id: %d",
- port_id, q_id);
- itt_rx_data[port_id].handles[q_id]
- = __itt_string_handle_create(task_name);
-
- itt_rx_data[port_id].queue_state[q_id] = 0;
-
if (!rte_eth_add_rx_callback(
- port_id, q_id, collect_itt_rx_burst_cb, NULL)) {
+ port_id, q_id, profile_hook_rx_burst_cb, NULL)) {
return -rte_errno;
}
}
return 0;
}
-#endif /* RTE_ETHDEV_PROFILE_ITT_WASTED_RX_ITERATIONS */
+#endif /* RTE_ETHDEV_PROFILE_WITH_VTUNE */
int
-__rte_eth_profile_rx_init(__rte_unused uint16_t port_id,
+__rte_eth_dev_profile_init(__rte_unused uint16_t port_id,
__rte_unused struct rte_eth_dev *dev)
{
-#ifdef RTE_ETHDEV_PROFILE_ITT_WASTED_RX_ITERATIONS
- return itt_profile_rx_init(
- port_id, dev->data->name, dev->data->nb_rx_queues);
+#ifdef RTE_ETHDEV_PROFILE_WITH_VTUNE
+ return vtune_profile_rx_init(port_id, dev->data->nb_rx_queues);
#endif
return 0;
}
diff --git a/lib/librte_ethdev/ethdev_profile.h b/lib/librte_ethdev/ethdev_profile.h
index e5ea368..65031e6 100644
--- a/lib/librte_ethdev/ethdev_profile.h
+++ b/lib/librte_ethdev/ethdev_profile.h
@@ -1,5 +1,5 @@
/* SPDX-License-Identifier: BSD-3-Clause
- * Copyright(c) 2010-2017 Intel Corporation
+ * Copyright(c) 2010-2018 Intel Corporation
*/
#ifndef _RTE_ETHDEV_PROFILE_H_
@@ -8,7 +8,7 @@
#include "rte_ethdev.h"
/**
- * Initialization of profiling RX queues for the Ethernet device.
+ * Initialization of the Ethernet device profiling.
* Implementation of this function depends on chosen profiling method,
* defined in configs.
*
@@ -22,6 +22,6 @@
* - On failure, a negative value.
*/
int
-__rte_eth_profile_rx_init(uint16_t port_id, struct rte_eth_dev *dev);
+__rte_eth_dev_profile_init(uint16_t port_id, struct rte_eth_dev *dev);
#endif
diff --git a/lib/librte_ethdev/rte_ethdev.c b/lib/librte_ethdev/rte_ethdev.c
index a9977df..acba712 100644
--- a/lib/librte_ethdev/rte_ethdev.c
+++ b/lib/librte_ethdev/rte_ethdev.c
@@ -1225,9 +1225,9 @@ rte_eth_dev_configure(uint16_t port_id, uint16_t nb_rx_q, uint16_t nb_tx_q,
}
/* Initialize Rx profiling if enabled at compilation time. */
- diag = __rte_eth_profile_rx_init(port_id, dev);
+ diag = __rte_eth_dev_profile_init(port_id, dev);
if (diag != 0) {
- RTE_PMD_DEBUG_TRACE("port%d __rte_eth_profile_rx_init = %d\n",
+ RTE_PMD_DEBUG_TRACE("port%d __rte_eth_dev_profile_init = %d\n",
port_id, diag);
rte_eth_dev_rx_queue_config(dev, 0);
rte_eth_dev_tx_queue_config(dev, 0);
--
2.7.4
next reply other threads:[~2018-07-03 13:11 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-07-03 13:11 ilia.kurakin [this message]
2018-07-19 12:21 ` [dpdk-dev] [PATCH v2] " ilia.kurakin
2018-07-24 13:17 ` Wiles, Keith
2018-08-22 17:13 ` Ferruh Yigit
2018-08-23 9:35 ` Ferruh Yigit
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=1530623488-69056-1-git-send-email-ilia.kurakin@intel.com \
--to=ilia.kurakin@intel.com \
--cc=dev@dpdk.org \
--cc=john.mcnamara@intel.com \
--cc=roman.p.khatko@intel.com \
--cc=thomas@monjalon.net \
/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).