Re: [BUG] perf/core: Task stuck on global_ctx_data_rwsem

[Namhyung Kim <namhyung@kernel.org>]

On Wed, Jan 07, 2026 at 10:16:52AM +0100, Peter Zijlstra wrote:
> On Tue, Jan 06, 2026 at 02:34:40PM -0800, Namhyung Kim wrote:
> > Hello,
> > 
> > On Mon, Dec 22, 2025 at 03:36:53PM -0800, Namhyung Kim wrote:
> > > On Mon, Dec 22, 2025 at 03:34:23PM -0800, Namhyung Kim wrote:
> > > > Hello,
> > > > 
> > > > I got a report that a task is stuck in perf_event_exit_task() waiting
> > > > for global_ctx_data_rwsem.  On large systems, it'd have performance
> > > > issues when it grabs the lock to iterate all threads in the system to
> > > > allocate the context data.  And it'd block task exit path which is
> > > > problematic especially under memory pressure.
> > > > 
> > > >   perf_event_open
> > > >     perf_event_alloc
> > > >       attach_perf_ctx_data
> > > >         attach_global_ctx_data
> > > >           percpu_down_write (global_ctx_data_rwsem)
> > > >             for_each_process_thread
> > > >               alloc_task_ctx_data
> > > >                                                do_exit
> > > >                                                  perf_event_exit_task
> > > >                                                    percpu_down_read (global_ctx_data_rwsem)
> > > > 
> > > > I think attach_global_ctx_data() should skip tasks with PF_EXITING and
> > > > it'd be nice if perf_event_exit_task() could release the ctx_data
> > > > unconditionally.  But I'm not sure how to synchronize them properly.
> > > > 
> > > > Any thoughts?
> > 
> > I'm curious if this makes any sense..  I feel like it needs to check the
> > flag again before allocation.
> > 
> > Thanks,
> > Namhyung
> > 
> > 
> > diff --git a/kernel/events/core.c b/kernel/events/core.c
> > index 376fb07d869b8b50..2a8847e95d7eb698 100644
> > --- a/kernel/events/core.c
> > +++ b/kernel/events/core.c
> > @@ -5469,6 +5469,8 @@ attach_global_ctx_data(struct kmem_cache *ctx_cache)
> >  	/* Allocate everything */
> >  	scoped_guard (rcu) {
> >  		for_each_process_thread(g, p) {
> > +			if (p->flags & PF_EXITING)
> > +				continue;
> >  			cd = rcu_dereference(p->perf_ctx_data);
> >  			if (cd && !cd->global) {
> >  				cd->global = 1;
> 
> I suppose this makes sense.
> 
> > @@ -14563,7 +14565,6 @@ void perf_event_exit_task(struct task_struct *task)
> >  	/*
> >  	 * Detach the perf_ctx_data for the system-wide event.
> >  	 */
> > -	guard(percpu_read)(&global_ctx_data_rwsem);
> >  	detach_task_ctx_data(task);
> >  }
> 
> This would need a comment; something like:
> 
> 	/*
> 	 * This can be done without holding global_ctx_data_rwsem
> 	 * because this is done after setting PF_EXITING such that
> 	 * attach_global_ctx_data() will skip over this task.
> 	 */
> 	WARN_ON_ONCE(!(task->flags & PF_EXITING))
> 
> But yes, I suppose this can do. The question is however, how do you get
> into this predicament to begin with? Are you creating and destroying a
> lot of global LBR events or something?

I think it's just because there are too many tasks in the system like
O(100K).  And any thread going to exit needs to wait for
attach_global_ctx_data() to finish the iteration over every task.

> 
> Would it make sense to delay detach_global_ctx_data() for a second or
> so? That is, what is your event creation pattern?

I don't think it has a special pattern, but I'm curious how we can
handle a race like below.

  attach_global_ctx_data
    check p->flags & PF_EXITING
                                              do_exit
    (preemption)                                set PF_EXITING
                                                detach_task_ctx_data()
    check p->perf_ctx_data
    attach_task_ctx_data()   ---> memory leak

Thanks,
Namhyung