perf_counter: Optimize perf_counter_alloc()'s inherit case

We don't need to add usage counts for swcounter and attr usage
models for inherited counters since the parent counter will
always have one, which suffices to generate the needed output.

This avoids up to 3 global atomic increments per inherited
counter.

LKML-Reference: <new-submission>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

kernel/perf_counter.c

@@ -1508,11 +1508,13 @@ static void free_counter(struct perf_counter *counter)
 {
 	perf_pending_sync(counter);
 
+	if (!counter->parent) {
 		atomic_dec(&nr_counters);
 		if (counter->attr.mmap)
 			atomic_dec(&nr_mmap_counters);
 		if (counter->attr.comm)
 			atomic_dec(&nr_comm_counters);
+	}
 
 	if (counter->destroy)
 		counter->destroy(counter);
@@ -3515,6 +3517,8 @@ static void sw_perf_counter_destroy(struct perf_counter *counter)
 {
 	u64 event = counter->attr.config;
 
+	WARN_ON(counter->parent);
+
 	atomic_dec(&perf_swcounter_enabled[event]);
 }
 
@@ -3551,8 +3555,10 @@ static const struct pmu *sw_perf_counter_init(struct perf_counter *counter)
 	case PERF_COUNT_SW_PAGE_FAULTS_MAJ:
 	case PERF_COUNT_SW_CONTEXT_SWITCHES:
 	case PERF_COUNT_SW_CPU_MIGRATIONS:
+		if (!counter->parent) {
 			atomic_inc(&perf_swcounter_enabled[event]);
 			counter->destroy = sw_perf_counter_destroy;
+		}
 		pmu = &perf_ops_generic;
 		break;
 	}
@@ -3663,11 +3669,13 @@ perf_counter_alloc(struct perf_counter_attr *attr,
 
 	counter->pmu = pmu;
 
+	if (!counter->parent) {
 		atomic_inc(&nr_counters);
 		if (counter->attr.mmap)
 			atomic_inc(&nr_mmap_counters);
 		if (counter->attr.comm)
 			atomic_inc(&nr_comm_counters);
+	}
 
 	return counter;
 }