Performance optimization analysis tool | perf

perf brief introduction

Perf(Performance Event) yes Linux 2.6.31 Built in performance analysis tool , Compared with others Prof The biggest advantage of the tool lies in its connection with Linux Kernel Close combination , It can analyze the performance of kernel and even hardware . Through it , Applications can take advantage of PMU（Performance Monitor Unit）、tracepoint The performance of the kernel and the special counters , It can not only analyze the performance problems of specified applications , You can also analyze the performance problems of the kernel , Thus, we can fully understand the performance bottlenecks in the application .

perf Compile configuration and installation

stay Linux Compile and install in the kernel source code :

#cd Root directory of kernel source code
#cd tools/perf/
#make && make install
#cp perf target_filesystem/bin/

Kernel configuration requirements

1. performance Options
   General Setup ->
   Kernel Performance Events And Counters ->
   [*]Kernel performance events and counters
   Kernel configuration options : PERF_EVENTS [=y]
2. profile Options
   General Setup ->
   [*]Profiling support
   Kernel configuration options : PROFILING [=y]
3. drivers Options
   Device Drivers->
   Performance monitor support->
   [*]ARM PMU framework
   Kernel configuration options ：ARM PMU framework[=y]
4. Device tree pmu Node requirements
   edit ft2000plus.dts, stay / Add under node pmu node ：
   pmu {
   compatible = “arm,armv8-pmuv3”;
   interrupts = <1 7 4>;
   };
   explain ：PMU(Performance Monitoring Unit) It's all CPU Manufacturer with CPU Hardware provided , It allows software to target certain CPU Hardware events ( Such as cache miss, branch-misses, instructions) Set up counter, And count the number of this event , When the number reaches counter After value , The interrupt . Software inspects by capturing these interrupts CPU usage .

perf Common commands

perf stat

perf stat Is used to run instructions , And analyze the statistical results , The information of the whole life cycle of application can be completely counted .
The command format is ：perf stat [ options] [command]
tat means statistic or counting.
In fact, it's statistics 、 Count .

Common options ：
• -a: counting for entire system.
• -I: Print count deltas every N milliseconds (minimum: 1ms)
• -C: Appoint core.

perf top

Display the occupation of each function cpu The proportion .

perf record

perf record It is used to record more granular statistical information , And save to file . For example, you can record statistics at the level of a single function ,
And use perf report To display the statistics .
The command format is ：perf record [options] [command]
or perf record [options] – command [options]

• -F： Sampling frequency , adopt -F To specify the
• -c ：Event period to sample.
• -g： Record function call stack

perf report

perf report Yes, it will perf record The generated file is parsed and displayed . The result is similar to perf top Real time display of .
The command format is ：perf report [options]
to perf record Generated perf.data Data display .

Some commands used for debugging in the project

1、
perf stat -r 5 -e cache-misses,cache-references,instructions,cycles,L1-dcache-stores,L1-dcache-store-misses -C 8

2、
perf record -e cache-misses -C 8
3、
perf top -C 8
perf top -g -C 8

Reference link

Video link ： https://www.bilibili.com/video/BV1hK4y1A7U4?spm_id_from=333.337.search-card.all.click&vd_source=9a8a120f260b2f6555d727c9d48cf27d
perf Related information ：
link ：https://pan.baidu.com/s/15ER6N3azPm0cv1i0mG1QkQ
Extraction code ：lbqq