[PATCH] sched/eevdf: clear buddies for preempt_short
From: Vincent Guittot
Date: Fri Apr 10 2026 - 09:23:45 EST
next buddy should not prevent shorter slice preemption. Don't take buddy
into account when checking if shorter slice entity can preempt and clear it
if the entity with a shorter slice can preempt current.
Test on snapdragon rb5:
hackbench -T -p -l 16000000 -g 2 1> /dev/null &
hackbench runs in cgroup /test-A
cyclictest -t 1 -i 2777 -D 63 --policy=fair --mlock -h 20000 -q
cyclictest runs in cgroup /test-B
tip/sched/core tip/sched/core +this patch
cyclictest slice (ms) (default)2.8 8 8
hackbench slice (ms) (default)2.8 20 20
Total Samples | 22679 22595 22686
Average (us) | 84 94(-12%) 59( 37%)
Median (P50) (us) | 56 56( 0%) 56( 0%)
90th Percentile (us) | 64 65(- 2%) 63( 3%)
99th Percentile (us) | 1047 1273(-22%) 74( 94%)
99.9th Percentile (us) | 2431 4751(-95%) 663( 86%)
Maximum (us) | 4694 8655(-84%) 3934( 55%)
Signed-off-by: Vincent Guittot <vincent.guittot@xxxxxxxxxx>
---
kernel/sched/fair.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 597ce5b718d2..0c6b9541d864 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -1116,7 +1116,7 @@ static struct sched_entity *__pick_eevdf(struct cfs_rq *cfs_rq, bool protect)
/*
* Picking the ->next buddy will affect latency but not fairness.
*/
- if (sched_feat(PICK_BUDDY) &&
+ if (sched_feat(PICK_BUDDY) && protect &&
cfs_rq->next && entity_eligible(cfs_rq, cfs_rq->next)) {
/* ->next will never be delayed */
WARN_ON_ONCE(cfs_rq->next->sched_delayed);
@@ -9110,8 +9110,10 @@ static void wakeup_preempt_fair(struct rq *rq, struct task_struct *p, int wake_f
return;
preempt:
- if (preempt_action == PREEMPT_WAKEUP_SHORT)
+ if (preempt_action == PREEMPT_WAKEUP_SHORT) {
cancel_protect_slice(se);
+ clear_buddies(cfs_rq, se);
+ }
resched_curr_lazy(rq);
}
--
2.43.0