1.. SPDX-License-Identifier: BSD-3-Clause 2 Copyright(c) 2010-2014 Intel Corporation. 3 4Profile Your Application 5======================== 6 7The following sections describe methods of profiling DPDK applications on 8different architectures. 9 10 11Profiling on x86 12---------------- 13 14Intel processors provide performance counters to monitor events. 15Some tools provided by Intel, such as Intel® VTune™ Amplifier, can be used 16to profile and benchmark an application. 17See the *VTune Performance Analyzer Essentials* publication from Intel Press for more information. 18 19For a DPDK application, this can be done in a Linux* application environment only. 20 21The main situations that should be monitored through event counters are: 22 23* Cache misses 24 25* Branch mis-predicts 26 27* DTLB misses 28 29* Long latency instructions and exceptions 30 31Refer to the 32`Intel Performance Analysis Guide <http://software.intel.com/sites/products/collateral/hpc/vtune/performance_analysis_guide.pdf>`_ 33for details about application profiling. 34 35 36Empty cycles tracing 37~~~~~~~~~~~~~~~~~~~~ 38 39Iterations that yielded no RX packets (empty cycles, wasted iterations) can 40be analyzed using VTune Amplifier. This profiling employs the 41`Instrumentation and Tracing Technology (ITT) API 42<https://software.intel.com/en-us/node/544195>`_ 43feature of VTune Amplifier and requires only reconfiguring the DPDK library, 44no changes in a DPDK application are needed. 45 46To trace wasted iterations on RX queues, first reconfigure DPDK with 47``CONFIG_RTE_ETHDEV_RXTX_CALLBACKS`` and 48``CONFIG_RTE_ETHDEV_PROFILE_ITT_WASTED_RX_ITERATIONS`` enabled. 49 50Then rebuild DPDK, specifying paths to the ITT header and library, which can 51be found in any VTune Amplifier distribution in the *include* and *lib* 52directories respectively: 53 54.. code-block:: console 55 56 make EXTRA_CFLAGS=-I<path to ittnotify.h> \ 57 EXTRA_LDLIBS="-L<path to libittnotify.a> -littnotify" 58 59Finally, to see wasted iterations in your performance analysis results, 60select the *"Analyze user tasks, events, and counters"* checkbox in the 61*"Analysis Type"* tab when configuring analysis via VTune Amplifier GUI. 62Alternatively, when running VTune Amplifier via command line, specify 63``-knob enable-user-tasks=true`` option. 64 65Collected regions of wasted iterations will be marked on VTune Amplifier's 66timeline as ITT tasks. These ITT tasks have predefined names, containing 67Ethernet device and RX queue identifiers. 68 69 70Profiling on ARM64 71------------------ 72 73Using Linux perf 74~~~~~~~~~~~~~~~~ 75 76The ARM64 architecture provide performance counters to monitor events. The 77Linux ``perf`` tool can be used to profile and benchmark an application. In 78addition to the standard events, ``perf`` can be used to profile arm64 79specific PMU (Performance Monitor Unit) events through raw events (``-e`` 80``-rXX``). 81 82For more derails refer to the 83`ARM64 specific PMU events enumeration <http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.100095_0002_04_en/way1382543438508.html>`_. 84 85 86High-resolution cycle counter 87~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 88 89The default ``cntvct_el0`` based ``rte_rdtsc()`` provides a portable means to 90get a wall clock counter in user space. Typically it runs at <= 100MHz. 91 92The alternative method to enable ``rte_rdtsc()`` for a high resolution wall 93clock counter is through the armv8 PMU subsystem. The PMU cycle counter runs 94at CPU frequency. However, access to the PMU cycle counter from user space is 95not enabled by default in the arm64 linux kernel. It is possible to enable 96cycle counter for user space access by configuring the PMU from the privileged 97mode (kernel space). 98 99By default the ``rte_rdtsc()`` implementation uses a portable ``cntvct_el0`` 100scheme. Application can choose the PMU based implementation with 101``CONFIG_RTE_ARM_EAL_RDTSC_USE_PMU``. 102 103The example below shows the steps to configure the PMU based cycle counter on 104an armv8 machine. 105 106.. code-block:: console 107 108 git clone https://github.com/jerinjacobk/armv8_pmu_cycle_counter_el0 109 cd armv8_pmu_cycle_counter_el0 110 make 111 sudo insmod pmu_el0_cycle_counter.ko 112 cd $DPDK_DIR 113 make config T=arm64-armv8a-linuxapp-gcc 114 echo "CONFIG_RTE_ARM_EAL_RDTSC_USE_PMU=y" >> build/.config 115 make 116 117.. warning:: 118 119 The PMU based scheme is useful for high accuracy performance profiling with 120 ``rte_rdtsc()``. However, this method can not be used in conjunction with 121 Linux userspace profiling tools like ``perf`` as this scheme alters the PMU 122 registers state. 123