perf/core: Remove confusing comment and move put_ctx()
authorPeter Zijlstra <peterz@infradead.org>
Thu, 16 Feb 2017 09:28:37 +0000 (10:28 +0100)
committerIngo Molnar <mingo@kernel.org>
Fri, 24 Feb 2017 07:56:32 +0000 (08:56 +0100)
Since commit:

  321027c1fe77 ("perf/core: Fix concurrent sys_perf_event_open() vs. 'move_group' race")

... the code looks like (assuming move_group==1):

  gctx = __perf_event_ctx_lock_double(group_leader, ctx);

  perf_remove_from_context(group_leader, 0);
  list_for_each_entry(sibling, &group_leader->sibling_list, group_entry) {
perf_remove_from_context(sibling, 0);
put_ctx(gctx);
  }

  /* ... */

  /* misleading comment about how this is the last reference */
  put_ctx(gctx);

  perf_event_ctx_unlock(group_leader, gctx);

What that 'last' put_ctx() does is drop @group_leader's reference on
gctx after having dropped all its potential sibling references.

But the thing is that __perf_event_ctx_lock_double() returns with a
reference _and_ a held lock, and perf_event_ctx_unlock() unlocks that
lock and drops that reference. Therefore that put_ctx() cannot be the
'last' of anything, nor is there an unbalance in puts.

To reduce confusion, remove the comment and place the put_ctx() next
to the remove_from_context() call.

Reported-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
kernel/events/core.c

index 77a932b..94d7b9a 100644 (file)
@@ -9955,6 +9955,7 @@ SYSCALL_DEFINE5(perf_event_open,
                 * of swizzling perf_event::ctx.
                 */
                perf_remove_from_context(group_leader, 0);
+               put_ctx(gctx);
 
                list_for_each_entry(sibling, &group_leader->sibling_list,
                                    group_entry) {
@@ -9993,13 +9994,6 @@ SYSCALL_DEFINE5(perf_event_open,
                perf_event__state_init(group_leader);
                perf_install_in_context(ctx, group_leader, group_leader->cpu);
                get_ctx(ctx);
-
-               /*
-                * Now that all events are installed in @ctx, nothing
-                * references @gctx anymore, so drop the last reference we have
-                * on it.
-                */
-               put_ctx(gctx);
        }
 
        /*