From: Byungchul Park <[email protected]>
change from v2 to v3
* rebase to tip git
change from v1 to v2
* use #ifdef CONFIG_SMP to load tracking code
* make commit message compact which made confused
----->8-----
>From 31c4509d09cd43bf52c12ef85716aa4c3b729b57 Mon Sep 17 00:00:00 2001
From: Byungchul Park <[email protected]>
Date: Wed, 12 Aug 2015 15:51:21 +0900
Subject: [PATCH v3] sched: sync with the prev cfs when changing cgroup within
a cpu
current code seems to be wrong with cfs_rq's avg loads when changing
a task's cgroup(=cfs_rq) to another. i tested with "echo pid > cgroup" and
found that e.g. cfs_rq->avg.load_avg became larger and larger whenever i
changed a cgroup to another again and again.
we have to sync se's average load with both *prev* cfs_rq and next cfs_rq
when changing its group.
Signed-off-by: Byungchul Park <[email protected]>
---
kernel/sched/fair.c | 24 ++++++++++++++++++++++--
1 file changed, 22 insertions(+), 2 deletions(-)
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 2a33d7b..25028b9 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -8017,8 +8017,28 @@ static void task_move_group_fair(struct task_struct *p, int queued)
if (!queued && (!se->sum_exec_runtime || p->state == TASK_WAKING))
queued = 1;
- if (!queued)
- se->vruntime -= cfs_rq_of(se)->min_vruntime;
+ if (!queued) {
+ cfs_rq = cfs_rq_of(se);
+ se->vruntime -= cfs_rq->min_vruntime;
+
+#ifdef CONFIG_SMP
+ /*
+ * synchronize task with its prev cfs_rq
+ * remove our load when we leave
+ */
+ __update_load_avg(cfs_rq->avg.last_update_time, cpu_of(rq_of(cfs_rq)), &se->avg,
+ se->on_rq * scale_load_down(se->load.weight), cfs_rq->curr == se, NULL);
+
+ cfs_rq->avg.load_avg =
+ max_t(long, cfs_rq->avg.load_avg - se->avg.load_avg, 0);
+ cfs_rq->avg.load_sum =
+ max_t(s64, cfs_rq->avg.load_sum - se->avg.load_sum, 0);
+ cfs_rq->avg.util_avg =
+ max_t(long, cfs_rq->avg.util_avg - se->avg.util_avg, 0);
+ cfs_rq->avg.util_sum =
+ max_t(s32, cfs_rq->avg.util_sum - se->avg.util_sum, 0);
+#endif
+ }
set_task_rq(p, task_cpu(p));
se->depth = se->parent ? se->parent->depth + 1 : 0;
if (!queued) {
--
1.7.9.5
On Wed, Aug 12, 2015 at 04:05:15PM +0900, [email protected] wrote:
>
> current code seems to be wrong with cfs_rq's avg loads when changing
> a task's cgroup(=cfs_rq) to another. i tested with "echo pid > cgroup" and
> found that e.g. cfs_rq->avg.load_avg became larger and larger whenever i
> changed a cgroup to another again and again.
>
> we have to sync se's average load with both *prev* cfs_rq and next cfs_rq
> when changing its group.
>
I think you are right in general, it has nothing to do with
migrate_task_rq_fair() and the load should be carried to new cfs_rq.
But whether it is queued or not should be doing the same thing, so what about:
/* if queued, the previous dequeue_task() should have already done it */
if (!queued)
__upload_load_avg()
remove_from_old_cfs_rq();
/* virtually catch up because of the skew between cfs_rqs */
task's last_update_time = cfs_rq's last_update_time;
add_to_new_cfs_rq();
Thanks,
Yuyang
On Wed, Aug 12, 2015 at 09:05:48AM +0800, Yuyang Du wrote:
> On Wed, Aug 12, 2015 at 04:05:15PM +0900, [email protected] wrote:
> >
> > current code seems to be wrong with cfs_rq's avg loads when changing
> > a task's cgroup(=cfs_rq) to another. i tested with "echo pid > cgroup" and
> > found that e.g. cfs_rq->avg.load_avg became larger and larger whenever i
> > changed a cgroup to another again and again.
> >
> > we have to sync se's average load with both *prev* cfs_rq and next cfs_rq
> > when changing its group.
> >
>
> I think you are right in general, it has nothing to do with
> migrate_task_rq_fair() and the load should be carried to new cfs_rq.
>
> But whether it is queued or not should be doing the same thing, so what about:
hello yuyang,
i worked with code based on torvalds git, but now.. in tip git where sched
code is renewed much, i think you are right. i will fix it.
thank you much. :)
Byungchul
>
> /* if queued, the previous dequeue_task() should have already done it */
> if (!queued)
> __upload_load_avg()
>
> remove_from_old_cfs_rq();
>
> /* virtually catch up because of the skew between cfs_rqs */
> task's last_update_time = cfs_rq's last_update_time;
> add_to_new_cfs_rq();
>
> Thanks,
> Yuyang
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/