Re: [PATCH 2/2] perf sched timehist: Fix invalid runtime in the idle hist

From: Namhyung Kim
Date: Tue Dec 13 2016 - 05:56:04 EST


On Tue, Dec 13, 2016 at 07:32:36AM -0300, Arnaldo Carvalho de Melo wrote:
> Em Tue, Dec 13, 2016 at 05:06:32PM +0900, Namhyung Kim escreveu:
> > When --idle-hist option is used, run/wait time and sched delay value
> > should be shown for idle task only. But due to internal accounting, a
> > last thread has same value of next idle task's and it was shown.
> >
> > In the below example, firefox after idle task has same run time of
> > idle task's.
>
> Same as last message, is this something I can fold into a patch yet in
> my perf/core branch?

Yep,

Fixes: 78de3657008b ("perf sched timehist: Add -I/--idle-hist option")

I guess it makes a conflict on the callchain code. In case it doesn't
look obvious to resolve, please let me know..

Thanks,
Namhyung


>
> - Arnaldo
>
> > $ perf sched timehist --idle-hist
> > Samples do not have callchains.
> > time cpu task name wait time sch delay run time
> > [tid/pid] (msec) (msec) (msec)
> > --------------- ------ -------------------- --------- --------- ---------
> > 197731.754335 [0002] firefox[17773/17739] 0.000 0.000 0.000
> > 197731.754486 [0001] sleep[27470] 0.000 0.000 0.000
> > 197731.754903 [0002] <idle> 0.047 0.000 0.567
> > 197731.754981 [0002] firefox[17773/17739] 0.000 0.000 0.567
> > 197731.755922 [0002] <idle> 0.078 0.000 0.941
> > 197731.755994 [0002] firefox[17773/17739] 0.000 0.000 0.941
> > <SNIP>
> >
> > Signed-off-by: Namhyung Kim <namhyung@xxxxxxxxxx>
> > ---
> > tools/perf/builtin-sched.c | 9 +++++++++
> > 1 file changed, 9 insertions(+)
> >
> > diff --git a/tools/perf/builtin-sched.c b/tools/perf/builtin-sched.c
> > index 405a91d0515f..64a0959bccd7 100644
> > --- a/tools/perf/builtin-sched.c
> > +++ b/tools/perf/builtin-sched.c
> > @@ -2439,6 +2439,15 @@ static int timehist_sched_change_event(struct perf_tool *tool,
> > goto out;
> >
> > timehist_update_runtime_stats(last_tr, t, tprev);
> > + /*
> > + * remove delta time of last thread as it's not updated
> > + * and otherwise it will show an invalid value next
> > + * time. we only care total run time and run stat.
> > + */
> > + last_tr->dt_run = 0;
> > + last_tr->dt_wait = 0;
> > + last_tr->dt_delay = 0;
> > +
> > if (itr->cursor.nr)
> > callchain_append(&itr->callchain, &itr->cursor,
> > t - tprev);
> > --
> > 2.10.2