diff mbox

[dpdk-dev,v8] ether: add support for vtune task tracing

Message ID 20170922145229.28463-1-ilia.kurakin@intel.com (mailing list archive)
State Accepted, archived
Headers show

Checks

Context Check Description
ci/checkpatch warning coding style issues
ci/Intel-compilation success Compilation OK

Commit Message

ilia.kurakin@intel.com Sept. 22, 2017, 2:52 p.m. UTC
From: Ilia Kurakin <ilia.kurakin@intel.com>

The patch simplifies DPDK applications analysis for developers which use
Intel® VTune Amplifier.

The empty cycles are such iterations that yielded no RX packets. As far as
DPDK is running in poll mode, wasting cycles is equal to wasting CPU time.
Tracing such iterations can identify that device is underutilized. Tracing
empty cycles becomes even more critical if a system uses a lot of Ethernet
ports.

The patch gives possibility to analyze empty cycles without changing
application code. All needs to be done is just to reconfigure and rebuild
the DPDK itself with CONFIG_RTE_ETHDEV_PROFILE_ITT_WASTED_RX_ITERATIONS
enbled. The important thing here is that this does not affect DPDK code.
The profiling code is not being compiled if user does not specify config
flag.

The patch provides common way to inject RX queues profiling and VTune
specific implementation.

Signed-off-by: Ilia Kurakin <ilia.kurakin@intel.com>

---

-V2 change:
    ITT tasks collection is moved to rx callback

-V3 change:
    rte_ethdev_profile.c created, all profile specific code moved there.

    Added generic profile function

-V4 change:
    checkpatch issues fixed

    Added documentation topic

-V5 change:
    Documentation fixes

-V6 change:
    Documentation changes are moved to different patch

-V7 change:
    Documentation and code changes are merged to a single patch.
    Added detailed description.

-V8 change:
    Added return code handling for Rx queues profiling init.


 config/common_base                    |   1 +
 doc/guides/prog_guide/profile_app.rst |  37 +++++++-
 lib/librte_ether/Makefile             |   1 +
 lib/librte_ether/ethdev_profile.c     | 166 ++++++++++++++++++++++++++++++++++
 lib/librte_ether/ethdev_profile.h     |  56 ++++++++++++
 lib/librte_ether/rte_ethdev.c         |  11 +++
 6 files changed, 271 insertions(+), 1 deletion(-)
 create mode 100644 lib/librte_ether/ethdev_profile.c
 create mode 100644 lib/librte_ether/ethdev_profile.h

Comments

Thomas Monjalon Sept. 22, 2017, 5:04 p.m. UTC | #1
22/09/2017 16:52, ilia.kurakin@intel.com:
> From: Ilia Kurakin <ilia.kurakin@intel.com>
> 
> The patch simplifies DPDK applications analysis for developers which use
> Intel® VTune Amplifier.
> 
> The empty cycles are such iterations that yielded no RX packets. As far as
> DPDK is running in poll mode, wasting cycles is equal to wasting CPU time.
> Tracing such iterations can identify that device is underutilized. Tracing
> empty cycles becomes even more critical if a system uses a lot of Ethernet
> ports.
> 
> The patch gives possibility to analyze empty cycles without changing
> application code. All needs to be done is just to reconfigure and rebuild
> the DPDK itself with CONFIG_RTE_ETHDEV_PROFILE_ITT_WASTED_RX_ITERATIONS
> enbled. The important thing here is that this does not affect DPDK code.
> The profiling code is not being compiled if user does not specify config
> flag.
> 
> The patch provides common way to inject RX queues profiling and VTune
> specific implementation.
> 
> Signed-off-by: Ilia Kurakin <ilia.kurakin@intel.com>

Applied with previous Ack from Jerin, thanks
diff mbox

Patch

diff --git a/config/common_base b/config/common_base
index 5e97a08..12f6be9 100644
--- a/config/common_base
+++ b/config/common_base
@@ -136,6 +136,7 @@  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
 
 #
 # Turn off Tx preparation stage
diff --git a/doc/guides/prog_guide/profile_app.rst b/doc/guides/prog_guide/profile_app.rst
index 54b546a..ca1c91f 100644
--- a/doc/guides/prog_guide/profile_app.rst
+++ b/doc/guides/prog_guide/profile_app.rst
@@ -39,7 +39,8 @@  Profiling on x86
 ----------------
 
 Intel processors provide performance counters to monitor events.
-Some tools provided by Intel, such as VTune, can be used to profile and benchmark an application.
+Some tools provided by Intel, such as Intel® VTune™ Amplifier, can be used
+to profile and benchmark an application.
 See the *VTune Performance Analyzer Essentials* publication from Intel Press for more information.
 
 For a DPDK application, this can be done in a Linux* application environment only.
@@ -59,6 +60,40 @@  Refer to the
 for details about application profiling.
 
 
+Empty cycles tracing
+~~~~~~~~~~~~~~~~~~~~
+
+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.
+
+
 Profiling on ARM64
 ------------------
 
diff --git a/lib/librte_ether/Makefile b/lib/librte_ether/Makefile
index 27d9766..5f65d05 100644
--- a/lib/librte_ether/Makefile
+++ b/lib/librte_ether/Makefile
@@ -46,6 +46,7 @@  LIBABIVER := 6
 SRCS-y += rte_ethdev.c
 SRCS-y += rte_flow.c
 SRCS-y += rte_tm.c
+SRCS-y += ethdev_profile.c
 
 #
 # Export include files
diff --git a/lib/librte_ether/ethdev_profile.c b/lib/librte_ether/ethdev_profile.c
new file mode 100644
index 0000000..13c48a8
--- /dev/null
+++ b/lib/librte_ether/ethdev_profile.c
@@ -0,0 +1,166 @@ 
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2010-2017 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *     * Neither the name of Intel Corporation nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <rte_config.h>
+
+#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.
+ */
+#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];
+
+
+/**
+ * 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.
+ */
+static uint16_t
+collect_itt_rx_burst_cb(uint8_t port_id, 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.
+ * 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.
+ *
+ * @return
+ *  - On success, zero.
+ *  - On failure, a negative value.
+ */
+static inline int
+itt_profile_rx_init(uint8_t port_id, char *port_name, 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)) {
+			return -rte_errno;
+		}
+	}
+
+	return 0;
+}
+#endif /* RTE_ETHDEV_PROFILE_ITT_WASTED_RX_ITERATIONS */
+
+int
+__rte_eth_profile_rx_init(__rte_unused uint8_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);
+#endif
+	return 0;
+}
diff --git a/lib/librte_ether/ethdev_profile.h b/lib/librte_ether/ethdev_profile.h
new file mode 100644
index 0000000..32f103e
--- /dev/null
+++ b/lib/librte_ether/ethdev_profile.h
@@ -0,0 +1,56 @@ 
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2010-2017 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *     * Neither the name of Intel Corporation nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef _RTE_ETHDEV_PROFILE_H_
+#define _RTE_ETHDEV_PROFILE_H_
+
+#include "rte_ethdev.h"
+
+/**
+ * Initialization of profiling RX queues for the Ethernet device.
+ * Implementation of this function depends on chosen profiling method,
+ * defined in configs.
+ *
+ * @param port_id
+ *  The port identifier of the Ethernet device.
+ * @param dev
+ *  Pointer to struct rte_eth_dev corresponding to given port_id.
+ *
+ * @return
+ *  - On success, zero.
+ *  - On failure, a negative value.
+ */
+int
+__rte_eth_profile_rx_init(uint8_t port_id, struct rte_eth_dev *dev);
+
+#endif
diff --git a/lib/librte_ether/rte_ethdev.c b/lib/librte_ether/rte_ethdev.c
index a88916f..1849a3b 100644
--- a/lib/librte_ether/rte_ethdev.c
+++ b/lib/librte_ether/rte_ethdev.c
@@ -67,6 +67,7 @@ 
 
 #include "rte_ether.h"
 #include "rte_ethdev.h"
+#include "ethdev_profile.h"
 
 static const char *MZ_RTE_ETH_DEV_DATA = "rte_eth_dev_data";
 struct rte_eth_dev rte_eth_devices[RTE_MAX_ETHPORTS];
@@ -819,6 +820,16 @@  rte_eth_dev_configure(uint8_t port_id, uint16_t nb_rx_q, uint16_t nb_tx_q,
 		return diag;
 	}
 
+	/* Initialize Rx profiling if enabled at compilation time. */
+	diag = __rte_eth_profile_rx_init(port_id, dev);
+	if (diag != 0) {
+		RTE_PMD_DEBUG_TRACE("port%d __rte_eth_profile_rx_init = %d\n",
+				port_id, diag);
+		rte_eth_dev_rx_queue_config(dev, 0);
+		rte_eth_dev_tx_queue_config(dev, 0);
+		return diag;
+	}
+
 	return 0;
 }