Re: [PATCH] perf cgroups: Fix perf_cgroup_switch schedule in warning

[Jiri Olsa <jolsa@redhat.com>]

On Tue, Oct 02, 2012 at 01:53:01PM +0200, Peter Zijlstra wrote:
> On Tue, 2012-10-02 at 13:42 +0200, Jiri Olsa wrote:
> > +++ b/kernel/events/core.c
> > @@ -394,7 +394,8 @@ void perf_cgroup_switch(struct task_struct *task, int mode)
> >                         }
> >  
> >                         if (mode & PERF_CGROUP_SWIN) {
> > -                               WARN_ON_ONCE(cpuctx->cgrp);
> > +                               WARN_ON_ONCE(cpuctx->cgrp && !cpuctx->ctx.is_active);
> > +
> >                                 /* set cgrp before ctxsw in to
> >                                  * allow event_filter_match() to not
> >                                  * have to pass task around 
> 
> OK, like you mentioned this is the result of multiple PMU being able to
> share a cpuctx, shouldn't we in that case avoid the second loop over the
> cpuctx as a whole?
> 
> Would something like the below do? IIRC I introduced that active_pmu for
> exactly such reasons..
> 
> ---
>  kernel/events/core.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index 7b9df35..e98f014 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -372,6 +372,8 @@ void perf_cgroup_switch(struct task_struct *task, int mode)
>  
>  	list_for_each_entry_rcu(pmu, &pmus, entry) {
>  		cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
> +		if (cpuctx->active_pmu != pmu)
> +			continue;
>  
>  		/*
>  		 * perf_cgroup_events says at least one
> 

this passed my test

jirka