commit 5d27c23df09b702868d9a3bff86ec6abd22963ac
author Peter Zijlstra <a.p.zijlstra@chello.nl> Thu Dec 17 13:16:32 2009 +0100
committer Ingo Molnar <mingo@elte.hu> Thu Dec 17 13:21:36 2009 +0100
tree ce29a1253087067b3a0e4239d0936eede870ab0b
parent 06d65bda75341485d32f33da474b0664819ad497

perf events: Dont report side-band events on each cpu for per-task-per-cpu events

Acme noticed that his FORK/MMAP numbers were inflated by about
the same factor as his cpu-count.

This led to the discovery of a few more sites that need to
respect the event->cpu filter.

Reported-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <20091217121830.215333434@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

kernel/perf_event.c

diff --git a/kernel/perf_event.c b/kernel/perf_event.c
index 8ab8698..03cc061 100644
--- a/kernel/perf_event.c
+++ b/kernel/perf_event.c
@@ -1381,6 +1381,9 @@
 		if (event->state != PERF_EVENT_STATE_ACTIVE)
 			continue;
 
+		if (event->cpu != -1 && event->cpu != smp_processor_id())
+			continue;
+
 		hwc = &event->hw;
 
 		interrupts = hwc->interrupts;
@@ -3265,6 +3268,9 @@
 
 static int perf_event_task_match(struct perf_event *event)
 {
+	if (event->cpu != -1 && event->cpu != smp_processor_id())
+		return 0;
+
 	if (event->attr.comm || event->attr.mmap || event->attr.task)
 		return 1;
 
@@ -3290,12 +3296,11 @@
 	rcu_read_lock();
 	cpuctx = &get_cpu_var(perf_cpu_context);
 	perf_event_task_ctx(&cpuctx->ctx, task_event);
-	put_cpu_var(perf_cpu_context);
-
 	if (!ctx)
 		ctx = rcu_dereference(task_event->task->perf_event_ctxp);
 	if (ctx)
 		perf_event_task_ctx(ctx, task_event);
+	put_cpu_var(perf_cpu_context);
 	rcu_read_unlock();
 }
 
@@ -3372,6 +3377,9 @@
 
 static int perf_event_comm_match(struct perf_event *event)
 {
+	if (event->cpu != -1 && event->cpu != smp_processor_id())
+		return 0;
+
 	if (event->attr.comm)
 		return 1;
 
@@ -3408,15 +3416,10 @@
 	rcu_read_lock();
 	cpuctx = &get_cpu_var(perf_cpu_context);
 	perf_event_comm_ctx(&cpuctx->ctx, comm_event);
-	put_cpu_var(perf_cpu_context);
-
-	/*
-	 * doesn't really matter which of the child contexts the
-	 * events ends up in.
-	 */
 	ctx = rcu_dereference(current->perf_event_ctxp);
 	if (ctx)
 		perf_event_comm_ctx(ctx, comm_event);
+	put_cpu_var(perf_cpu_context);
 	rcu_read_unlock();
 }
 
@@ -3491,6 +3494,9 @@
 static int perf_event_mmap_match(struct perf_event *event,
 				   struct perf_mmap_event *mmap_event)
 {
+	if (event->cpu != -1 && event->cpu != smp_processor_id())
+		return 0;
+
 	if (event->attr.mmap)
 		return 1;
 
@@ -3564,15 +3570,10 @@
 	rcu_read_lock();
 	cpuctx = &get_cpu_var(perf_cpu_context);
 	perf_event_mmap_ctx(&cpuctx->ctx, mmap_event);
-	put_cpu_var(perf_cpu_context);
-
-	/*
-	 * doesn't really matter which of the child contexts the
-	 * events ends up in.
-	 */
 	ctx = rcu_dereference(current->perf_event_ctxp);
 	if (ctx)
 		perf_event_mmap_ctx(ctx, mmap_event);
+	put_cpu_var(perf_cpu_context);
 	rcu_read_unlock();
 
 	kfree(buf);
@@ -3863,6 +3864,9 @@
 				struct perf_sample_data *data,
 				struct pt_regs *regs)
 {
+	if (event->cpu != -1 && event->cpu != smp_processor_id())
+		return 0;
+
 	if (!perf_swevent_is_counting(event))
 		return 0;