Re: [PATCH v2 RESEND] perf/core: Fix missing read event generation on task exit

[Peter Zijlstra <peterz@infradead.org>]

On Thu, Oct 09, 2025 at 12:35:30AM +0800, Thaumy Cheng wrote:

> @@ -13995,13 +13997,16 @@ static void sync_child_event(struct perf_event *child_event)
>  }
> 
>  static void
> -perf_event_exit_event(struct perf_event *event,
> -		      struct perf_event_context *ctx, bool revoke)
> +perf_event_detach_event(struct perf_event *event,
> +		      struct perf_event_context *ctx, bool revoke, bool exit)
>  {
>  	struct perf_event *parent_event = event->parent;
> -	unsigned long detach_flags = DETACH_EXIT;
> +	unsigned long detach_flags = 0;
>  	unsigned int attach_state;
> 
> +	if (exit)
> +		detach_flags |= DETACH_EXIT;
> +

Urgh, at that point just pass in the right DETACH_flag directly,
something like:

perf_event_detach_event(struct perf_event *event,
			struct perf_event_context *ctx,
			unsigned long detach_flags);

	perf_event_detach_event(event, ctx, DETACH_EXIT);

	perf_event_detach_event(event, ctx, DETACH_REVOKE);



>  	if (parent_event) {
>  		/*
>  		 * Do not destroy the 'original' grouping; because of the
> @@ -14077,6 +14082,17 @@ static void perf_event_exit_task_context(struct task_struct *task, bool exit)
>  	 */
>  	mutex_lock(&ctx->mutex);
> 
> +	/*
> +	 * Report the task dead after unscheduling the events so that we
> +	 * won't get any samples after PERF_RECORD_EXIT. We can however still
> +	 * get a few PERF_RECORD_READ events.
> +	 */
> +	if (exit)
> +		perf_event_task(task, ctx, 0);
> +
> +	list_for_each_entry_safe(child_event, next, &ctx->event_list, event_entry)
> +		perf_event_detach_event(child_event, ctx, false, exit);
> +
>  	/*
>  	 * In a single ctx::lock section, de-schedule the events and detach the
>  	 * context from the task such that we cannot ever get it scheduled back

This can't be right; read the comment you moved and then look where we
unschedule the event.