Re: [PATCH v3a 6/7] perf: Rename perf_event_exit_task(.child)

[Ravi Bangoria]

Hi Peter,

> The task passed to perf_event_exit_task() is not a child, it is
> current. Fix this confusing naming, since much of the rest of the code
> also relies on it being current.
> 
> Specifically, both exec() and exit() callers use it with current as
> the argument.

When perf_event_exit_task_context() gets called by perf_event_free_task():

1) task_ctx_sched_out(ctx) function should be avoided because the 'ctx'
   argument is of the (half baked)child task whereas task_ctx_sched_out()
   expects 'ctx' to be the context of 'current'.
2) Similarly, 'task' argument != 'current'.

--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -13573,7 +13573,8 @@ static void perf_event_exit_task_context(struct task_struct *task, bool exit)
 	 * in.
 	 */
 	raw_spin_lock_irq(&ctx->lock);
-	task_ctx_sched_out(ctx, NULL, EVENT_ALL);
+	if (exit)
+		task_ctx_sched_out(ctx, NULL, EVENT_ALL);
 
 	/*
 	 * Now that the context is inactive, destroy the task <-> ctx relation
@@ -13582,7 +13583,7 @@ static void perf_event_exit_task_context(struct task_struct *task, bool exit)
 	RCU_INIT_POINTER(task->perf_event_ctxp, NULL);
 	put_ctx(ctx); /* cannot be last */
 	WRITE_ONCE(ctx->task, TASK_TOMBSTONE);
-	put_task_struct(current); /* cannot be last */
+	put_task_struct(task); /* cannot be last */
 
 	clone_ctx = unclone_ctx(ctx);
 	raw_spin_unlock_irq(&ctx->lock);

Thanks,
Ravi