Re: [PATCH v3 3/4] perf: Remove perf_swevent_get_recursion_context() from perf_pending_task().

[Frederic Weisbecker]

Le Fri, Mar 22, 2024 at 07:48:23AM +0100, Sebastian Andrzej Siewior a écrit :
> perf_swevent_get_recursion_context() is supposed to avoid recursion.
> This requires to remain on the same CPU in order to decrement/ increment
> the same counter. This is done by using preempt_disable(). Having
> preemption disabled while sending a signal leads to locking problems on
> PREEMPT_RT because sighand, a spinlock_t, becomes a sleeping lock.
> 
> This callback runs in task context and currently delivers only a signal
> to "itself". Any kind of recusrion protection in this context is not
> required.
> 
> Remove recursion protection in perf_pending_task().
> 
> Tested-by: Marco Elver <elver@xxxxxxxxxx>
> Tested-by: Arnaldo Carvalho de Melo <acme@xxxxxxxxxx>
> Reported-by: Arnaldo Carvalho de Melo <acme@xxxxxxxxxx>
> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx>
> ---
>  kernel/events/core.c | 12 ------------
>  1 file changed, 12 deletions(-)
> 
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index e0b2da8de485f..5400f7ed2f98b 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -6785,14 +6785,6 @@ static void perf_pending_irq(struct irq_work *entry)
>  static void perf_pending_task(struct callback_head *head)
>  {
>  	struct perf_event *event = container_of(head, struct perf_event, pending_task);
> -	int rctx;
> -
> -	/*
> -	 * If we 'fail' here, that's OK, it means recursion is already disabled
> -	 * and we won't recurse 'further'.
> -	 */
> -	preempt_disable_notrace();
> -	rctx = perf_swevent_get_recursion_context();
>  
>  	if (event->pending_work) {
>  		event->pending_work = 0;
> @@ -6800,10 +6792,6 @@ static void perf_pending_task(struct callback_head *head)
>  		local_dec(&event->ctx->nr_pending);
>  	}
>  
> -	if (rctx >= 0)
> -		perf_swevent_put_recursion_context(rctx);
> -	preempt_enable_notrace();

Well, if a software event happens during perf_sigtrap(), the task work
may be requeued endlessly and the task may get stuck in task_work_run()...

> -
>  	put_event(event);
>  }
>  
> -- 
> 2.43.0
> 
>