perf: Use jump_labels to optimize the scheduler hooks

Trades a call + conditional + ret for an unconditional jmp.

Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <20101014203625.501657727@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

include/linux/perf_event.h

@@ -487,6 +487,7 @@ struct perf_guest_info_callbacks {
 #include <linux/ftrace.h>
 #include <linux/cpu.h>
 #include <linux/irq_work.h>
+#include <linux/jump_label_ref.h>
 #include <asm/atomic.h>
 #include <asm/local.h>
 
@@ -895,8 +896,30 @@ extern void perf_pmu_unregister(struct pmu *pmu);
 
 extern int perf_num_counters(void);
 extern const char *perf_pmu_name(void);
-extern void perf_event_task_sched_in(struct task_struct *task);
+extern void __perf_event_task_sched_in(struct task_struct *task);
-extern void perf_event_task_sched_out(struct task_struct *task, struct task_struct *next);
+extern void __perf_event_task_sched_out(struct task_struct *task, struct task_struct *next);
+
+extern atomic_t perf_task_events;
+
+static inline void perf_event_task_sched_in(struct task_struct *task)
+{
+	JUMP_LABEL(&perf_task_events, have_events);
+	return;
+
+have_events:
+	__perf_event_task_sched_in(task);
+}
+
+static inline
+void perf_event_task_sched_out(struct task_struct *task, struct task_struct *next)
+{
+	JUMP_LABEL(&perf_task_events, have_events);
+	return;
+
+have_events:
+	__perf_event_task_sched_out(task, next);
+}
+
 extern int perf_event_init_task(struct task_struct *child);
 extern void perf_event_exit_task(struct task_struct *child);
 extern void perf_event_free_task(struct task_struct *task);

kernel/perf_event.c

@@ -34,7 +34,7 @@
 
 #include <asm/irq_regs.h>
 
-static atomic_t nr_events __read_mostly;
+atomic_t perf_task_events __read_mostly;
 static atomic_t nr_mmap_events __read_mostly;
 static atomic_t nr_comm_events __read_mostly;
 static atomic_t nr_task_events __read_mostly;
@@ -1311,8 +1311,8 @@ void perf_event_context_sched_out(struct task_struct *task, int ctxn,
  * accessing the event control register. If a NMI hits, then it will
  * not restart the event.
  */
-void perf_event_task_sched_out(struct task_struct *task,
+void __perf_event_task_sched_out(struct task_struct *task,
-			       struct task_struct *next)
+				 struct task_struct *next)
 {
 	int ctxn;
 
@@ -1337,14 +1337,6 @@ static void task_ctx_sched_out(struct perf_event_context *ctx,
 	cpuctx->task_ctx = NULL;
 }
 
-/*
- * Called with IRQs disabled
- */
-static void __perf_event_task_sched_out(struct perf_event_context *ctx)
-{
-	task_ctx_sched_out(ctx, EVENT_ALL);
-}
-
 /*
  * Called with IRQs disabled
  */
@@ -1494,7 +1486,7 @@ void perf_event_context_sched_in(struct perf_event_context *ctx)
  * accessing the event control register. If a NMI hits, then it will
  * keep the event running.
  */
-void perf_event_task_sched_in(struct task_struct *task)
+void __perf_event_task_sched_in(struct task_struct *task)
 {
 	struct perf_event_context *ctx;
 	int ctxn;
@@ -2216,7 +2208,8 @@ static void free_event(struct perf_event *event)
 	irq_work_sync(&event->pending);
 
 	if (!event->parent) {
-		atomic_dec(&nr_events);
+		if (event->attach_state & PERF_ATTACH_TASK)
+			jump_label_dec(&perf_task_events);
 		if (event->attr.mmap || event->attr.mmap_data)
 			atomic_dec(&nr_mmap_events);
 		if (event->attr.comm)
@@ -5354,7 +5347,8 @@ done:
 	event->pmu = pmu;
 
 	if (!event->parent) {
-		atomic_inc(&nr_events);
+		if (event->attach_state & PERF_ATTACH_TASK)
+			jump_label_inc(&perf_task_events);
 		if (event->attr.mmap || event->attr.mmap_data)
 			atomic_inc(&nr_mmap_events);
 		if (event->attr.comm)
@@ -5849,7 +5843,7 @@ static void perf_event_exit_task_context(struct task_struct *child, int ctxn)
 	 * our context.
 	 */
 	child_ctx = child->perf_event_ctxp[ctxn];
-	__perf_event_task_sched_out(child_ctx);
+	task_ctx_sched_out(child_ctx, EVENT_ALL);
 
 	/*
 	 * Take the context lock here so that if find_get_context is