Re: [RFC v2] perf: Rewrite core context handling

[Peter Zijlstra <peterz@infradead.org>]

On Mon, Jun 13, 2022 at 04:35:11PM +0200, Peter Zijlstra wrote:

Another one of those lockdep splats:

> @@ -12147,42 +12256,37 @@ SYSCALL_DEFINE5(perf_event_open,
>  	if (pmu->task_ctx_nr == perf_sw_context)
>  		event->event_caps |= PERF_EV_CAP_SOFTWARE;
>  
> -	if (group_leader) {
> -		if (is_software_event(event) &&
> -		    !in_software_context(group_leader)) {
> -			/*
> -			 * If the event is a sw event, but the group_leader
> -			 * is on hw context.
> -			 *
> -			 * Allow the addition of software events to hw
> -			 * groups, this is safe because software events
> -			 * never fail to schedule.
> -			 */
> -			pmu = group_leader->ctx->pmu;
> -		} else if (!is_software_event(event) &&
> -			   is_software_event(group_leader) &&
> -			   (group_leader->group_caps & PERF_EV_CAP_SOFTWARE)) {
> -			/*
> -			 * In case the group is a pure software group, and we
> -			 * try to add a hardware event, move the whole group to
> -			 * the hardware context.
> -			 */
> -			move_group = 1;
> -		}
> -	}
> -
>  	/*
>  	 * Get the target context (task or percpu):
>  	 */
> -	ctx = find_get_context(pmu, task, event);
> +	ctx = find_get_context(task, event);
>  	if (IS_ERR(ctx)) {
>  		err = PTR_ERR(ctx);
>  		goto err_alloc;
>  	}
>  
> -	/*
> -	 * Look up the group leader (we will attach this event to it):
> -	 */
> +	mutex_lock(&ctx->mutex);
> +
> +	if (ctx->task == TASK_TOMBSTONE) {
> +		err = -ESRCH;
> +		goto err_locked;
> +	}
> +
> +	if (!task) {
> +		/*
> +		 * Check if the @cpu we're creating an event for is online.
> +		 *
> +		 * We use the perf_cpu_context::ctx::mutex to serialize against
> +		 * the hotplug notifiers. See perf_event_{init,exit}_cpu().
> +		 */
> +		struct perf_cpu_context *cpuctx = per_cpu_ptr(&cpu_context, event->cpu);
> +
> +		if (!cpuctx->online) {
> +			err = -ENODEV;
> +			goto err_locked;
> +		}
> +	}
> +
>  	if (group_leader) {
>  		err = -EINVAL;
>  


pulling up the ctx->mutex makes things simpler, but also violates the
locking order vs exec_update_lock.

Pull that lock up as well...

---
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -12254,13 +12254,29 @@ SYSCALL_DEFINE5(perf_event_open,
 	if (pmu->task_ctx_nr == perf_sw_context)
 		event->event_caps |= PERF_EV_CAP_SOFTWARE;
 
+	if (task) {
+		err = down_read_interruptible(&task->signal->exec_update_lock);
+		if (err)
+			goto err_alloc;
+
+		/*
+		 * We must hold exec_update_lock across this and any potential
+		 * perf_install_in_context() call for this new event to
+		 * serialize against exec() altering our credentials (and the
+		 * perf_event_exit_task() that could imply).
+		 */
+		err = -EACCES;
+		if (!perf_check_permission(&attr, task))
+			goto err_cred;
+	}
+
 	/*
 	 * Get the target context (task or percpu):
 	 */
 	ctx = find_get_context(task, event);
 	if (IS_ERR(ctx)) {
 		err = PTR_ERR(ctx);
-		goto err_alloc;
+		goto err_cred;
 	}
 
 	mutex_lock(&ctx->mutex);
@@ -12358,58 +12374,14 @@ SYSCALL_DEFINE5(perf_event_open,
 			goto err_context;
 	}
 
-	event_file = anon_inode_getfile("[perf_event]", &perf_fops, event, f_flags);
-	if (IS_ERR(event_file)) {
-		err = PTR_ERR(event_file);
-		event_file = NULL;
-		goto err_context;
-	}
-
-	if (task) {
-		err = down_read_interruptible(&task->signal->exec_update_lock);
-		if (err)
-			goto err_file;
-
-		/*
-		 * We must hold exec_update_lock across this and any potential
-		 * perf_install_in_context() call for this new event to
-		 * serialize against exec() altering our credentials (and the
-		 * perf_event_exit_task() that could imply).
-		 */
-		err = -EACCES;
-		if (!perf_check_permission(&attr, task))
-			goto err_cred;
-	}
-
-	if (ctx->task == TASK_TOMBSTONE) {
-		err = -ESRCH;
-		goto err_locked;
-	}
-
 	if (!perf_event_validate_size(event)) {
 		err = -E2BIG;
-		goto err_locked;
-	}
-
-	if (!task) {
-		/*
-		 * Check if the @cpu we're creating an event for is online.
-		 *
-		 * We use the perf_cpu_context::ctx::mutex to serialize against
-		 * the hotplug notifiers. See perf_event_{init,exit}_cpu().
-		 */
-		struct perf_cpu_context *cpuctx =
-			container_of(ctx, struct perf_cpu_context, ctx);
-
-		if (!cpuctx->online) {
-			err = -ENODEV;
-			goto err_locked;
-		}
+		goto err_context;
 	}
 
 	if (perf_need_aux_event(event) && !perf_get_aux_event(event, group_leader)) {
 		err = -EINVAL;
-		goto err_locked;
+		goto err_context;
 	}
 
 	/*
@@ -12418,11 +12390,18 @@ SYSCALL_DEFINE5(perf_event_open,
 	 */
 	if (!exclusive_event_installable(event, ctx)) {
 		err = -EBUSY;
-		goto err_cred;
+		goto err_context;
 	}
 
 	WARN_ON_ONCE(ctx->parent_ctx);
 
+	event_file = anon_inode_getfile("[perf_event]", &perf_fops, event, f_flags);
+	if (IS_ERR(event_file)) {
+		err = PTR_ERR(event_file);
+		event_file = NULL;
+		goto err_context;
+	}
+
 	/*
 	 * This is the point on no return; we cannot fail hereafter. This is
 	 * where we start modifying current state.
@@ -12500,17 +12479,15 @@ SYSCALL_DEFINE5(perf_event_open,
 	fd_install(event_fd, event_file);
 	return event_fd;
 
-err_cred:
-	if (task)
-		up_read(&task->signal->exec_update_lock);
-err_file:
-	fput(event_file);
 err_context:
 	/* event->pmu_ctx freed by free_event() */
 err_locked:
 	mutex_unlock(&ctx->mutex);
 	perf_unpin_context(ctx);
 	put_ctx(ctx);
+err_cred:
+	if (task)
+		up_read(&task->signal->exec_update_lock);
 err_alloc:
 	/*
 	 * If event_file is set, the fput() above will have called ->release()