Re: [PATCH] sched/fair: vruntime should normalize when switching from fair

From: Dietmar Eggemann
Date: Fri Sep 28 2018 - 13:12:04 EST


On 09/28/2018 02:43 AM, Wanpeng Li wrote:
On Thu, 27 Sep 2018 at 21:23, Dietmar Eggemann <dietmar.eggemann@xxxxxxx> wrote:

On 09/27/2018 03:19 AM, Wanpeng Li wrote:
On Thu, 27 Sep 2018 at 06:38, Dietmar Eggemann <dietmar.eggemann@xxxxxxx> wrote:

Hi,

On 09/26/2018 11:50 AM, Wanpeng Li wrote:
Hi Dietmar,
On Tue, 28 Aug 2018 at 22:55, Dietmar Eggemann <dietmar.eggemann@xxxxxxx> wrote:

On 08/27/2018 12:14 PM, Peter Zijlstra wrote:
On Fri, Aug 24, 2018 at 02:24:48PM -0700, Steve Muckle wrote:
On 08/24/2018 02:47 AM, Peter Zijlstra wrote:
On 08/17/2018 11:27 AM, Steve Muckle wrote:

[...]

- later, when the prio is deboosted and the task is moved back
to the fair class, the fair rq's min_vruntime is added to
the task's vruntime, even though it wasn't subtracted earlier.

Could you point out when the fair rq's min_vruntime is added to the
task's vruntime in your *later* scenario? attach_task_cfs_rq will not
do that the same reason as detach_task_cfs_rq. fair task's
sched_remote_wakeup is false which results in vruntime will not be
renormalized in enqueue_entity.

The cfs_rq->min_vruntime is still added to the se->vruntime in
enqueue_task_fair().

I understand what your patch done,

It's not my patch ;-) I just helped to find out under which
circumstances this issue can happen.

On your CPU4:
scheduler_ipi()
-> sched_ttwu_pending()
-> ttwu_do_activate() => p->sched_remote_wakeup should be
false, so ENQUEUE_WAKEUP is set, ENQUEUE_MIGRATED is not
-> ttwu_activate()
-> activate_task()
-> enqueue_task()
-> enqueue_task_fair()
-> enqueue_entity()
bool renorm = !(flags &
ENQUEUE_WAKEUP) || (flags & ENQUEUE_MIGRATE)
so renorm is false in enqueue_entity(), why you mentioned that the
cfs_rq->min_vruntime is still added to the se->vruntime in
enqueue_task_fair()?

Maybe this is a misunderstanding on my side but didn't you asked me to
'... Could you point out when the fair rq's min_vruntime is added to the
task's vruntime in your *later* scenario? ...'

Yeah, if the calltrace above and my analysis is correct, then the fair
rq's min_vruntime will not be added to the task's vruntime in your
*later* scenario, which means that your patch is not necessary.

Ah, ok, both, the ENQUEUE_WAKEUP and the ENQUEUE_MIGRATED are not set in the enqueue_entity() call so renorm is 1 (flags is 0xe).