linux perf 环境部署和基本测试(基于Ubuntu20.04）

1,linux 安装perf

sudo apt-ge install linux-tools-common

sudo apt-get install linux-tools-$(uname -r) linux-tools-generic -y

2 补充安装

sudo apt-get  install python3-q-text-as-data 

3，perf常用命令

lark@ubuntu:~$ perf

 usage: perf [--version] [--help] [OPTIONS] COMMAND [ARGS]

 The most commonly used perf commands are:
   annotate        Read perf.data (created by perf record) and display annotated code
   archive         Create archive with object files with build-ids found in perf.data file
   bench           General framework for benchmark suites
   buildid-cache   Manage build-id cache.
   buildid-list    List the buildids in a perf.data file
   c2c             Shared Data C2C/HITM Analyzer.
   config          Get and set variables in a configuration file.
   daemon          Run record sessions on background
   data            Data file related processing
   diff            Read perf.data files and display the differential profile
   evlist          List the event names in a perf.data file
   ftrace          simple wrapper for kernel's ftrace functionality
   inject          Filter to augment the events stream with additional information
   iostat          Show I/O performance metrics
   kallsyms        Searches running kernel for symbols
   kmem            Tool to trace/measure kernel memory properties
   kvm             Tool to trace/measure kvm guest os
   list            List all symbolic event types
   lock            Analyze lock events
   mem             Profile memory accesses
   record          Run a command and record its profile into perf.data
   report          Read perf.data (created by perf record) and display the profile
   sched           Tool to trace/measure scheduler properties (latencies)
   script          Read perf.data (created by perf record) and display trace output
   stat            Run a command and gather performance counter statistics
   test            Runs sanity tests.
   timechart       Tool to visualize total system behavior during a workload
   top             System profiling tool.
   version         display the version of perf binary
   probe           Define new dynamic tracepoints
   trace           strace inspired tool

 See 'perf help COMMAND' for more information on a specific command.

举例代码main.c

#include <stdio.h>
#include <stdlib.h>

void long_test()
{
    int i, j;
    while(1)
	{
		;
	}
}

void foo2()
{
    int i;
    for (i = 0; i < 100; i++) long_test();
}

void foo1()
{
    int i;
    for (i = 0; i < 1000; i++) long_test();
}

int main(void) {
    foo1();
    foo2();
}

编译：gcc main.c -o main

运行 ：./main

4， perf top命令 实时查看进程CPU和调用堆栈

 lark@ubuntu:~$ sudo perf top -a

5，构建火焰图

下载库安装

git clone https://github.com/brendangregg/FlameGraph.git

 执行文件生成火焰图

ps -axf | grep main
sudo perf record -p 2915  -g -- sleep 10 //采样10s

//采样的数据画成火焰图
sudo perf script -i perf.data &> perf.unfold
sudo ./FlameGraph/stackcollapse-perf.pl perf.unfold &> perf.folded
sudo ./FlameGraph/flamegraph.pl perf.folded > perf.svg

google浏览器打开 火焰图。

perf record 命令可以统计每个调用栈出现的百分比

lark@ubuntu:~/test$ sudo perf report -n --stdio
# To display the perf.data header info, please use --header/--header-only options.
#
#
# Total Lost Samples: 0
#
# Samples: 20K of event 'cpu-clock:pppH'
# Event count (approx.): 5069000000
#
# Children      Self       Samples  Command  Shared Object      Symbol                             
# ........  ........  ............  .......  .................  ...................................
#
   100.00%     0.00%             0  main     libc-2.31.so       [.] __libc_start_main
            |
            ---__libc_start_main
               main
               foo1
               long_test

   100.00%     0.00%             0  main     main               [.] main
            |
            ---main
               foo1
 

上面可以看到main->foo1的栈占用率 100%。