ARM: 6902/1: perf: Remove erroneous check on active_events
authorMark Rutland <mark.rutland@arm.com>
Tue, 17 May 2011 08:15:38 +0000 (09:15 +0100)
committerRussell King <rmk+kernel@arm.linux.org.uk>
Fri, 20 May 2011 21:39:17 +0000 (22:39 +0100)
When initialising a PMU, there is a check to protect against races with
other CPUs filling all of the available event slots. Since armpmu_add
checks that an event can be scheduled, we do not need to do this at
initialisation time. Furthermore the current code is broken because it
assumes that atomic_inc_not_zero will unconditionally increment
active_counts and then tries to decrement it again on failure.

This patch removes the broken, redundant code.

Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Will Deacon <will.deacon@arm.com>
Cc: Jamie Iles <jamie@jamieiles.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
arch/arm/kernel/perf_event.c

index 139e3c8273696198fa0995a761a5368875961b51..d53c0abc4dd3aaa375595fcd4d54527f013b41f8 100644 (file)
@@ -560,11 +560,6 @@ static int armpmu_event_init(struct perf_event *event)
        event->destroy = hw_perf_event_destroy;
 
        if (!atomic_inc_not_zero(&active_events)) {
-               if (atomic_read(&active_events) > armpmu->num_events) {
-                       atomic_dec(&active_events);
-                       return -ENOSPC;
-               }
-
                mutex_lock(&pmu_reserve_mutex);
                if (atomic_read(&active_events) == 0) {
                        err = armpmu_reserve_hardware();