perf: Annotate perf_event_read_group() vs perf_event_release_kernel()

Stephane reported a lockdep warning while using PERF_FORMAT_GROUP.

The issue is that perf_event_read_group() takes faults while holding
the ctx->mutex, while perf_event_release_kernel() can be called from
munmap(). Which makes for an AB-BA deadlock.

Except we can never establish the deadlock because we'll only ever
call perf_event_release_kernel() after all file descriptors are dead
so there is no concurrency possible.

Reported-by: Stephane Eranian <eranian@google.com>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

diff --git a/kernel/perf_event.c b/kernel/perf_event.c
index 49d8be5a45e3bed77d159865c9773f63a439b1ea..34659d4085c79b8cafb4bf122af6b3dfcd497a33 100644 (file)
--- a/kernel/perf_event.c
+++ b/kernel/perf_event.c
@@ -1867,7 +1867,19 @@ int perf_event_release_kernel(struct perf_event *event)
        event->state = PERF_EVENT_STATE_FREE;
 
        WARN_ON_ONCE(ctx->parent_ctx);
-       mutex_lock(&ctx->mutex);
+       /*
+        * There are two ways this annotation is useful:
+        *
+        *  1) there is a lock recursion from perf_event_exit_task
+        *     see the comment there.
+        *
+        *  2) there is a lock-inversion with mmap_sem through
+        *     perf_event_read_group(), which takes faults while
+        *     holding ctx->mutex, however this is called after
+        *     the last filedesc died, so there is no possibility
+        *     to trigger the AB-BA case.
+        */
+       mutex_lock_nested(&ctx->mutex, SINGLE_DEPTH_NESTING);
        perf_event_remove_from_context(event);
        mutex_unlock(&ctx->mutex);
 
@@ -5305,7 +5317,7 @@ void perf_event_exit_task(struct task_struct *child)
         *
         * But since its the parent context it won't be the same instance.
         */
-       mutex_lock_nested(&child_ctx->mutex, SINGLE_DEPTH_NESTING);
+       mutex_lock(&child_ctx->mutex);
 
 again:
        list_for_each_entry_safe(child_event, tmp, &child_ctx->pinned_groups,