[PATCH v2 3/4] sched/fair: Check for CFS tasks before detach_one_task()

[Valentin Schneider]

detach_one_task() (and the CFS load-balancer in general) can only ever
pull CFS tasks, so we should check if there's any before invoking the
cpu stopper.

Likewise, when the cpu stopper is already up, we can bail out slightly
earlier when we see there's no CFS task to detach.

Check for CFS tasks at the top of need_active_balance(), and add a
similar check in active_load_balance_cpu_stop().

Signed-off-by: Valentin Schneider <valentin.schneider@xxxxxxx>
---
Note that this only touches the active load balance paths. IMO
load_balance() could go through a similar treatment
(s/rq.nr_running/rq.cfs.h_nr_running/ before going to detach_tasks())
but that conflicts with Vincent's rework, so I've not included it
here.
---
 kernel/sched/fair.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 751f41085f47..8f5f6cad5008 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -8776,6 +8776,10 @@ static inline enum alb_status active_load_balance(struct lb_env *env)
 
 	raw_spin_lock_irqsave(&busiest->lock, flags);
 
+	/* Is there actually anything to pull? */
+	if (busiest->cfs.h_nr_running < 1)
+		goto unlock;
+
 	/*
 	 * Don't kick the active_load_balance_cpu_stop, if the curr task on
 	 * busiest CPU can't be moved to dst_cpu:
@@ -9142,8 +9146,8 @@ static int active_load_balance_cpu_stop(void *data)
 		     !busiest_rq->active_balance))
 		goto out_unlock;
 
-	/* Is there any task to move? */
-	if (busiest_rq->nr_running <= 1)
+	/* Is there any CFS task to move? */
+	if (busiest_rq->cfs.h_nr_running < 1)
 		goto out_unlock;
 
 	/*
-- 
2.22.0