Re: [PATCH] perf/core: Update cgroup time with descendants

[Jiri Olsa]

On Tue, Sep 19, 2017 at 10:02:50AM +0800, Lin Xiulei wrote:
> Add steps of reproducing
> 
> 1
> ```
> mkdir /sys/fs/cgroup/perf_event/root1
> mkdir /sys/fs/cgroup/perf_event/root1/leaf1
> ```
> 2  $pid is any running process
> ```
> echo $pid > /sys/fs/cgroup/perf_event/root1/cgroup.procs
> ```
> 
> 3
> ```
> perf stat -e cycles -G root1 -a -I 1000
> ```
> 
> You shall get 0 cycles from it, which should have been cycles of $pid
> 
> 2017-09-14 15:03 GMT+08:00  <linxiulei@xxxxxxxxx>:
> > From: "leilei.lin" <leilei.lin@xxxxxxxxxxxxxxx>
> >
> > This fix updating cgroup time when event is being scheduled in
> > by cgroup descendants
> >
> > Signed-off-by: leilei.lin <leilei.lin@xxxxxxxxxxxxxxx>
> > ---
> >  kernel/events/core.c | 4 +++-
> >  1 file changed, 3 insertions(+), 1 deletion(-)
> >
> > diff --git a/kernel/events/core.c b/kernel/events/core.c
> > index 3e691b7..6f6bd1e 100644
> > --- a/kernel/events/core.c
> > +++ b/kernel/events/core.c
> > @@ -662,7 +662,9 @@ static inline void update_cgrp_time_from_event(struct perf_event *event)
> >         /*
> >          * Do not update time when cgroup is not active
> >          */
> > -       if (cgrp == event->cgrp)
> > +       if (cgrp == event->cgrp ||
> > +               cgroup_is_descendant(cgrp->css.cgroup,
> > +                       event->cgrp->css.cgroup))
> >                 __update_cgrp_time(event->cgrp);

you no longer need the 'cgrp == event->cgrp', it's part
of the cgroup_is_descendant call

other than that it looks ok to me

Reviewed-and-tested-by: Jiri Olsa <jolsa@xxxxxxxxxx>

thanks,
jirka