linux-kernel - Re: lost some call trace for sleep function

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date:   Thu, 26 Oct 2017 14:32:54 -0300
From:   Arnaldo Carvalho de Melo <acme@...nel.org>
To:     yuzhoujian <ufo19890607@...il.com>
Cc:     peterz@...radead.org, mingo@...hat.com,
        alexander.shishkin@...ux.intel.com, jolsa@...nel.org,
        dsahern@...il.com, namhyung@...nel.org, milian.wolff@...b.com,
        arnaldo.melo@...il.com, yuzhoujian@...ichuxing.com,
        adrian.hunter@...el.com, wangnan0@...wei.com,
        linux-perf-users@...r.kernel.org, linux-kernel@...r.kernel.org,
        acme@...hat.com
Subject: Re: lost some call trace for sleep function

Em Thu, Oct 26, 2017 at 06:24:56PM +0800, yuzhoujian escreveu:
> Hi, all.
> I find a strange problem. Perf cannot record call stack which contains sleep functions.
> The last function of the call trace is always "__GI___libc_nanosleep" for each sample.
> one of the sample for perf script is below:

How are you recording it? Please state the exact command line you use for 'record'.

Here are some attempts at doing that on a fedora 26 x86_64 system:

[acme@...et linux]$ uname -a
Linux jouet 4.14.0-rc3+ #1 SMP Fri Oct 13 12:21:12 -03 2017 x86_64 x86_64 x86_64 GNU/Linux

# perf trace -e nanosleep --max-stack=10 sleep 1
     0.649 (1000.121 ms): sleep/9566 nanosleep(rqtp: 0x7ffe56769570                                        ) = 0
                                       __nanosleep_nocancel (/usr/lib64/libc-2.25.so)
                                       rpl_nanosleep (/usr/bin/sleep)
                                       xnanosleep (/usr/bin/sleep)
                                       main (/usr/bin/sleep)
                                       __libc_start_main (/usr/lib64/libc-2.25.so)
                                       _start (/usr/bin/sleep)
#

Which is equivalent to:

# perf record -e syscalls:sys_enter_nanosleep/call-graph=dwarf,max-stack=10/ sleep 1
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.027 MB perf.data (1 samples) ]
# perf script
sleep  9629 [001] 210689.400780: syscalls:sys_enter_nanosleep: rqtp: 0x7ffd6a99b180, rmtp: 0x00000000
                   d4420 __nanosleep_nocancel (/usr/lib64/libc-2.25.so)
                    46c6 rpl_nanosleep (/usr/bin/sleep)
                    449f xnanosleep (/usr/bin/sleep)
                    1773 main (/usr/bin/sleep)
                   20509 __libc_start_main (/usr/lib64/libc-2.25.so)
                    1869 _start (/usr/bin/sleep)

# 

But why are you trying to sample CPU cycles used on a function that sleeps?

- Arnaldo
 
> test_sleep 12275 185233.961287:          1 cycles:ppp: 
>         ffffffff8100add0 intel_bts_enable_local ([kernel.kallsyms])
>         ffffffff81008f20 intel_pmu_enable_all ([kernel.kallsyms])
>         ffffffff810057ec x86_pmu_enable ([kernel.kallsyms])
>         ffffffff81173e57 perf_pmu_enable ([kernel.kallsyms])
>         ffffffff81175404 __perf_event_task_sched_in ([kernel.kallsyms])
>         ffffffff810c1aa8 finish_task_switch ([kernel.kallsyms])
>         ffffffff81690e00 __schedule ([kernel.kallsyms])
>         ffffffff81691409 schedule ([kernel.kallsyms])
>         ffffffff816902d6 do_nanosleep ([kernel.kallsyms])
>         ffffffff810b747b hrtimer_nanosleep ([kernel.kallsyms])
>         ffffffff810b75be sys_nanosleep ([kernel.kallsyms])
>         ffffffff8169c749 system_call_fastpath ([kernel.kallsyms])
> 		   bf190 __GI___libc_nanosleep (/usr/lib64/libc-2.17.so)
> 
> Below is the source code of test_sleep:
> 
> void f2()
> {
>         sleep(1);
> }
> void f1()
> {
>         f2();
> }
> int main()
> {
>   	while(1)
>      		f1();
>    	return 0;
> }
> 
> I think the right call stack should contain the __sleep function in glibc, just as follow
> 
> test_sleep 12275 185233.961287:          1 cycles:ppp: 
>         ffffffff8100add0 intel_bts_enable_local ([kernel.kallsyms])
>         ffffffff81008f20 intel_pmu_enable_all ([kernel.kallsyms])
>         ffffffff810057ec x86_pmu_enable ([kernel.kallsyms])
>         ffffffff81173e57 perf_pmu_enable ([kernel.kallsyms])
>         ffffffff81175404 __perf_event_task_sched_in ([kernel.kallsyms])
>         ffffffff810c1aa8 finish_task_switch ([kernel.kallsyms])
>         ffffffff81690e00 __schedule ([kernel.kallsyms])
>         ffffffff81691409 schedule ([kernel.kallsyms])
>         ffffffff816902d6 do_nanosleep ([kernel.kallsyms])
>         ffffffff810b747b hrtimer_nanosleep ([kernel.kallsyms])
>         ffffffff810b75be sys_nanosleep ([kernel.kallsyms])
>         ffffffff8169c749 system_call_fastpath ([kernel.kallsyms])
> 		   bf190 __GI___libc_nanosleep (/usr/lib64/libc-2.17.so)
> 		   bef70 __sleep (/usr/lib64/libc-2.17.so)
>         	   5a1 f2 (/home/test_sleep)
>         	   5c1 f1 (/home/test_sleep)
>         	   5d1 main (/home/test_sleep)
>         	   21c05 __libc_start_main (/usr/lib64/libc-2.17.so)
> 
> Is it a bug for perf record ??