[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <5c58161b-d935-21f1-1002-c8c3b85705e7@huawei.com>
Date: Thu, 9 Mar 2023 17:43:56 +0800
From: Zhang Qiao <zhangqiao22@...wei.com>
To: Dietmar Eggemann <dietmar.eggemann@....com>
CC: <mingo@...hat.com>, <peterz@...radead.org>,
<juri.lelli@...hat.com>, <vincent.guittot@...aro.org>,
<rostedt@...dmis.org>, <bsegall@...gle.com>, <mgorman@...e.de>,
<bristot@...hat.com>, <vschneid@...hat.com>,
<linux-kernel@...r.kernel.org>, <rkagan@...zon.de>
Subject: Re: [PATCH v2] sched/fair: sanitize vruntime of entity being migrated
Hi,
在 2023/3/7 20:45, Dietmar Eggemann 写道:
> On 06/03/2023 14:24, Zhang Qiao wrote:
>> Commit 829c1651e9c4 ("sched/fair: sanitize vruntime of
>> entity being placed") fix an overflowing bug, but ignore
>> a case that se->exec_start is reset after a migration.
>>
>> For fixing this case, we reset the vruntime of a long
>> sleeping task in migrate_task_rq_fair().
>>
>> Fixes: 829c1651e9c4 ("sched/fair: sanitize vruntime of entity being placed")
>> Suggested-by: Vincent Guittot <vincent.guittot@...aro.org>
>> Signed-off-by: Zhang Qiao <zhangqiao22@...wei.com>
>
> [...]
>
>> @@ -7635,7 +7653,23 @@ static void migrate_task_rq_fair(struct task_struct *p, int new_cpu)
>> if (READ_ONCE(p->__state) == TASK_WAKING) {
>> struct cfs_rq *cfs_rq = cfs_rq_of(se);
>>
>> - se->vruntime -= u64_u32_load(cfs_rq->min_vruntime);
>> + /*
>> + * We determine whether a task sleeps for long by checking
>> + * se->exec_start, and if it is, we sanitize its vruntime at
>> + * place_entity(). However, after a migration, this detection
>> + * method fails due to se->exec_start being reset.
>> + *
>> + * For fixing this case, we add the same check here. For a task
>> + * which has slept for a long time, its vruntime should be reset
>> + * to cfs_rq->min_vruntime with a sleep credit. Because waking
>> + * task's vruntime will be added to cfs_rq->min_vruntime when
>> Isn't this the other way around? `vruntime += min_vruntime`
Yes, you're right, we can refer to:
enqueue_entity()
...
if (renorm && !curr) {
se->vruntime += cfs_rq->min_vruntime;
...
>
>> + * enqueue, we only need to reset the se->vruntime of waking task
>> + * to a credit here.
>
> You not reset it to credit, you subtract the credit from vruntime ?
>
> I assume this is done to have sleeper credit accounted on both
> (se->vruntime and vruntime) for `se->vruntime =
> max_vruntime(se->vruntime, vruntime)` in place_entity() since
> entity_is_long_sleep(se)=false for a remove wakeup since `se->exec_start=0`.
>
>
>> + */
>> + if (entity_is_long_sleep(se))
>> + se->vruntime = -sched_sleeper_credit(se);
We subtract the credit here on the originating CPU since the long
sleeping task which migrates will go through:
place_entity()
else
se->vruntime = max_vruntime(se->vruntime, vruntime (1));
and not the `if (entity_is_long_sleep(se))` path. And sleeper credit is
also subtracted from vruntime (1) before in place_entity().
IOW, We do the same thing in advance in migrate_task_rq_fair().
For the long sleeping task, se->vruntime is equal to vruntime(1) in place_entity().
Thanks.
ZhangQiao.
>> + else
>> + se->vruntime -= u64_u32_load(cfs_rq->min_vruntime);
>
> Not sure I understand this part.
> Don't we have to do `vruntime -= min_vruntime` here for long sleeping
> task as well?
>
> Since we always do the `vruntime += min_vruntime` on the new CPU for a
> remote wakeup.
>
> [...]
>
> .
>
Powered by blists - more mailing lists