sched: disable forced preemption by default

Implement feature bit to disable forced preemption. This way
it can be checked whether a workload is overscheduling or not.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

kernel/sched.c

@@ -444,13 +444,15 @@ enum {
 	SCHED_FEAT_START_DEBIT		= 2,
 	SCHED_FEAT_TREE_AVG             = 4,
 	SCHED_FEAT_APPROX_AVG           = 8,
+	SCHED_FEAT_WAKEUP_PREEMPT	= 16,
 };
 
 const_debug unsigned int sysctl_sched_features =
 		SCHED_FEAT_NEW_FAIR_SLEEPERS	*1 |
 		SCHED_FEAT_START_DEBIT		*1 |
 		SCHED_FEAT_TREE_AVG		*0 |
-		SCHED_FEAT_APPROX_AVG		*0;
+		SCHED_FEAT_APPROX_AVG		*0 |
+		SCHED_FEAT_WAKEUP_PREEMPT	*1;
 
 #define sched_feat(x) (sysctl_sched_features & SCHED_FEAT_##x)
 

kernel/sched_fair.c

@@ -626,7 +626,7 @@ static void entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
 	 */
 	update_curr(cfs_rq);
 
-	if (cfs_rq->nr_running > 1)
+	if (cfs_rq->nr_running > 1 || !sched_feat(WAKEUP_PREEMPT))
 		check_preempt_tick(cfs_rq, curr);
 }
 
@@ -828,6 +828,7 @@ static void check_preempt_wakeup(struct rq *rq, struct task_struct *p)
 		return;
 	}
 
+	if (sched_feat(WAKEUP_PREEMPT)) {
 		while (!is_same_group(se, pse)) {
 			se = parent_entity(se);
 			pse = parent_entity(pse);
@@ -840,6 +841,7 @@ static void check_preempt_wakeup(struct rq *rq, struct task_struct *p)
 
 		if (delta > gran)
 			resched_task(curr);
+	}
 }
 
 static struct task_struct *pick_next_task_fair(struct rq *rq)