perfcounters: add task migrations counter

Impact: add new feature, new sw counter

Add a counter that counts the number of cross-CPU migrations a
task is suffering.

Signed-off-by: Ingo Molnar <mingo@elte.hu>

4 files changed, 61 insertions, 6 deletions

diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h
index d2a16563415..f30486fc55d 100644
--- a/include/linux/perf_counter.h
+++ b/include/linux/perf_counter.h
@@ -42,6 +42,8 @@ enum hw_event_types {
 	PERF_COUNT_BRANCH_INSTRUCTIONS	=  4,
 	PERF_COUNT_BRANCH_MISSES	=  5,
 
+	PERF_HW_EVENTS_MAX		=  6,
+
 	/*
 	 * Special "software" counters provided by the kernel, even if
 	 * the hardware does not support performance counters. These
@@ -50,11 +52,11 @@ enum hw_event_types {
 	 */
 	PERF_COUNT_CPU_CLOCK		= -1,
 	PERF_COUNT_TASK_CLOCK		= -2,
-	/*
-	 * Future software events:
-	 */
 	PERF_COUNT_PAGE_FAULTS		= -3,
 	PERF_COUNT_CONTEXT_SWITCHES	= -4,
+	PERF_COUNT_CPU_MIGRATIONS	= -5,
+
+	PERF_SW_EVENTS_MIN		= -6,
 };
 
 /*
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 4c530278391..2e15be8fc79 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1014,6 +1014,8 @@ struct sched_entity {
 	u64			last_wakeup;
 	u64			avg_overlap;
 
+	u64			nr_migrations;
+
 #ifdef CONFIG_SCHEDSTATS
 	u64			wait_start;
 	u64			wait_max;
@@ -1029,7 +1031,6 @@ struct sched_entity {
 	u64			exec_max;
 	u64			slice_max;
 
-	u64			nr_migrations;
 	u64			nr_migrations_cold;
 	u64			nr_failed_migrations_affine;
 	u64			nr_failed_migrations_running;
diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c
index 09287091c52..fb11e351e44 100644
--- a/kernel/perf_counter.c
+++ b/kernel/perf_counter.c
@@ -936,6 +936,52 @@ static const struct hw_perf_counter_ops perf_ops_context_switches = {
 	.hw_perf_counter_read		= context_switches_perf_counter_read,
 };
 
+static inline u64 get_cpu_migrations(void)
+{
+	return current->se.nr_migrations;
+}
+
+static void cpu_migrations_perf_counter_update(struct perf_counter *counter)
+{
+	u64 prev, now;
+	s64 delta;
+
+	prev = atomic64_read(&counter->hw.prev_count);
+	now = get_cpu_migrations();
+
+	atomic64_set(&counter->hw.prev_count, now);
+
+	delta = now - prev;
+	if (WARN_ON_ONCE(delta < 0))
+		delta = 0;
+
+	atomic64_add(delta, &counter->count);
+}
+
+static void cpu_migrations_perf_counter_read(struct perf_counter *counter)
+{
+	cpu_migrations_perf_counter_update(counter);
+}
+
+static void cpu_migrations_perf_counter_enable(struct perf_counter *counter)
+{
+	/*
+	 * se.nr_migrations is a per-task value already,
+	 * so we dont have to clear it on switch-in.
+	 */
+}
+
+static void cpu_migrations_perf_counter_disable(struct perf_counter *counter)
+{
+	cpu_migrations_perf_counter_update(counter);
+}
+
+static const struct hw_perf_counter_ops perf_ops_cpu_migrations = {
+	.hw_perf_counter_enable		= cpu_migrations_perf_counter_enable,
+	.hw_perf_counter_disable	= cpu_migrations_perf_counter_disable,
+	.hw_perf_counter_read		= cpu_migrations_perf_counter_read,
+};
+
 static const struct hw_perf_counter_ops *
 sw_perf_counter_init(struct perf_counter *counter)
 {
@@ -951,6 +997,9 @@ sw_perf_counter_init(struct perf_counter *counter)
 	case PERF_COUNT_CONTEXT_SWITCHES:
 		hw_ops = &perf_ops_context_switches;
 		break;
+	case PERF_COUNT_CPU_MIGRATIONS:
+		hw_ops = &perf_ops_cpu_migrations;
+		break;
 	default:
 		break;
 	}
diff --git a/kernel/sched.c b/kernel/sched.c
index 5c3f4106314..382cfdb5e38 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -1852,12 +1852,14 @@ void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
 		p->se.sleep_start -= clock_offset;
 	if (p->se.block_start)
 		p->se.block_start -= clock_offset;
+#endif
 	if (old_cpu != new_cpu) {
-		schedstat_inc(p, se.nr_migrations);
+		p->se.nr_migrations++;
+#ifdef CONFIG_SCHEDSTATS
 		if (task_hot(p, old_rq->clock, NULL))
 			schedstat_inc(p, se.nr_forced2_migrations);
-	}
 #endif
+	}
 	p->se.vruntime -= old_cfsrq->min_vruntime -
 					 new_cfsrq->min_vruntime;
 
@@ -2375,6 +2377,7 @@ static void __sched_fork(struct task_struct *p)
 	p->se.exec_start		= 0;
 	p->se.sum_exec_runtime		= 0;
 	p->se.prev_sum_exec_runtime	= 0;
+	p->se.nr_migrations		= 0;
 	p->se.last_wakeup		= 0;
 	p->se.avg_overlap		= 0;