2014-12-16 15:58:59

by Xunlei Pang

Subject: [PATCH] sched/fair: Fix the dealing with decay_count in __synchronize_entity_decay()

In __synchronize_entity_decay(), if "decays" happens to be zero,
se->avg.decay_count will not be zeroed, holding the positive value
assigned when dequeued last time.

This is problematic in the following case:
If this runnable task is CFS-balanced to other CPUs soon afterwards,
migrate_task_rq_fair() will treat it as a blocked task due to its
non-zero decay_count, thereby adding its load to cfs_rq->removed_load
wrongly.

Thus, we must zero se->avg.decay_count in this case as well.

Signed-off-by: Xunlei Pang <[email protected]>
---
kernel/sched/fair.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index df2cdf7..ea517cd 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -2574,11 +2574,11 @@ static inline u64 __synchronize_entity_decay(struct sched_entity *se)
u64 decays = atomic64_read(&cfs_rq->decay_counter);

decays -= se->avg.decay_count;
+ se->avg.decay_count = 0;
if (!decays)
return 0;

se->avg.load_avg_contrib = decay_load(se->avg.load_avg_contrib, decays);
- se->avg.decay_count = 0;

return decays;
}
--
1.9.1

2014-12-16 18:09:51

by Benjamin Segall

Subject: Re: [PATCH] sched/fair: Fix the dealing with decay_count in __synchronize_entity_decay()

Xunlei Pang <[email protected]> writes:

> In __synchronize_entity_decay(), if "decays" happens to be zero,
> se->avg.decay_count will not be zeroed, holding the positive value
> assigned when dequeued last time.
>
> This is problematic in the following case:
> If this runnable task is CFS-balanced to other CPUs soon afterwards,
> migrate_task_rq_fair() will treat it as a blocked task due to its
> non-zero decay_count, thereby adding its load to cfs_rq->removed_load
> wrongly.
>
> Thus, we must zero se->avg.decay_count in this case as well.

Yep. We probably didn't notice this because migrate will happen to clear
this by doing decay_count = -__synch(), but you can hit this via
switched_from_fair or just on enqueue followed by a load-balance
migration that thinks decay_counter was zero due to being on_rq.

>
> Signed-off-by: Xunlei Pang <[email protected]>

Reviewed-by: Ben Segall <[email protected]>

> ---
> kernel/sched/fair.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index df2cdf7..ea517cd 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -2574,11 +2574,11 @@ static inline u64 __synchronize_entity_decay(struct sched_entity *se)
> u64 decays = atomic64_read(&cfs_rq->decay_counter);
>
> decays -= se->avg.decay_count;
> + se->avg.decay_count = 0;
> if (!decays)
> return 0;
>
> se->avg.load_avg_contrib = decay_load(se->avg.load_avg_contrib, decays);
> - se->avg.decay_count = 0;
>
> return decays;
> }