Re: [PATCHSET 0/7] perf sched: Introduce timehist command, again (v2)

From: Ingo Molnar
Date: Wed Nov 16 2016 - 02:53:22 EST



* Namhyung Kim <namhyung@xxxxxxxxxx> wrote:

> Hello,
>
> This patchset is a rebased version of David's sched timehist work [1].
> I plan to improve perf sched command more and think that having
> timehist command before the work looks good. It seems David is busy
> these days, so I'm retrying it by myself.
>
> * changes in v2)
> - change name 'b/n time' to 'wait time' (Ingo)
> - show arrow between functions in the callchain (Ingo)
> - fix a bug in calculating initial run time
>
> This implements only basic feature and a few options. I just split
> the patch to make it easier to review and did some cosmetic changes.
> More patches will come later.
>
> The below is from the David's original description (w/ slight change):
>
> ------------------------8<-------------------------
> 'perf sched timehist' provides an analysis of scheduling events.
>
> Example usage:
> perf sched record -- sleep 1
> perf sched timehist
>
> By default it shows the individual schedule events, including the time between
> sched-in events for the task, the task scheduling delay (time between wakeup
> and actually running) and run time for the task:
>
> time cpu task name[tid/pid] wait time sch delay run time
> ------------- ---- -------------------- --------- --------- ---------
> 79371.874569 [11] gcc[31949] 0.014 0.000 1.148
> 79371.874591 [10] gcc[31951] 0.000 0.000 0.024
> 79371.874603 [10] migration/10[59] 3.350 0.004 0.011
> 79371.874604 [11] <idle> 1.148 0.000 0.035
> 79371.874723 [05] <idle> 0.016 0.000 1.383
> 79371.874746 [05] gcc[31949] 0.153 0.078 0.022
> ...
>
> Times are in msec.usec.
>
> If callchains were recorded they are appended to the line with a default stack depth of 5:
>
> 79371.874569 [11] gcc[31949] 0.014 0.000 1.148 wait_for_completion_killable <- do_fork <- sys_vfork <- stub_vfork <- __vfork
> 79371.874591 [10] gcc[31951] 0.000 0.000 0.024 __cond_resched <- _cond_resched <- wait_for_completion <- stop_one_cpu <- sched_exec
> 79371.874603 [10] migration/10[59] 3.350 0.004 0.011 smpboot_thread_fn <- kthread <- ret_from_fork
> 79371.874604 [11] <idle> 1.148 0.000 0.035 cpu_startup_entry <- start_secondary
> 79371.874723 [05] <idle> 0.016 0.000 1.383 cpu_startup_entry <- start_secondary
> 79371.874746 [05] gcc[31949] 0.153 0.078 0.022 do_wait sys_wait4 <- system_call_fastpath <- __GI___waitpid
>
> --no-call-graph can be used to not show the callchains. --max-stack is used
> to control the number of frames shown (default of 5). -x/--excl options can
> be used to collapse redundant callchains to get more relevant data on screen.
>
> Similar to perf-trace -s and -S can be used to dump a statistical summary
> without or with events (respectively). Statistics include min run time,
> average run time and max run time. Stats are also shown for run time by
> cpu.
>
> The cpu-visual option provides a visual aid for sched switches by cpu:
> ...
> 79371.874569 [11] s gcc[31949] 0.014 0.000 1.148
> 79371.874591 [10] s gcc[31951] 0.000 0.000 0.024
> 79371.874603 [10] s migration/10[59] 3.350 0.004 0.011
> 79371.874604 [11] i <idle> 1.148 0.000 0.035
> 79371.874723 [05] i <idle> 0.016 0.000 1.383
> 79371.874746 [05] s gcc[31949] 0.153 0.078 0.022
> ...

Looks great to me!

Acked-by: Ingo Molnar <mingo@xxxxxxxxxx>

Thanks,

Ingo