From: Chengming Zhou <zhouchengming@bytedance.com>
To: Dietmar Eggemann <dietmar.eggemann@arm.com>,
mingo@redhat.com, peterz@infradead.org,
vincent.guittot@linaro.org, rostedt@goodmis.org,
bsegall@google.com, vschneid@redhat.com
Cc: linux-kernel@vger.kernel.org
Subject: Re: [External] Re: [PATCH v2 07/10] sched/fair: use update_load_avg() to attach/detach entity load_avg
Date: Wed, 20 Jul 2022 21:43:59 +0800 [thread overview]
Message-ID: <952fc252-30e1-3629-42c4-98a55ada12f9@bytedance.com> (raw)
In-Reply-To: <17c9af40-bf53-be3c-c678-159a8ab8964a@arm.com>
On 2022/7/19 23:02, Dietmar Eggemann wrote:
> On 15/07/2022 18:21, Chengming Zhou wrote:
>> On 2022/7/15 19:18, Dietmar Eggemann wrote:
>>> On 13/07/2022 06:04, Chengming Zhou wrote:
>>>> Since update_load_avg() support DO_ATTACH and DO_DETACH now, we can
>>>> use update_load_avg() to implement attach/detach entity load_avg.
>>>>
>>>> Another advantage of using update_load_avg() is that it will check
>>>> last_update_time before attach or detach, instead of unconditional
>>>> attach/detach in the current code.
>>>>
>>>> This way can avoid some corner problematic cases of load tracking,
>>>> like twice attach problem, detach unattached NEW task problem.
>>>
>>> This explanation is somewhat hard to follow for me. Since both issues
>>> have been fixed already (you mention this further below) you're saying
>>> that with you change you don't reintroduce them?
>>
>> Sorry for this not very clear explanation.
>>
>> Yes, both issues have been fixed already, what I want to say is that bugfix
>> brings its own problem and limitation mentioned below.
>>
>> So I want to use another way to solve these problems better.
>
> [...]
>
>>>> These problems have been fixed in commit 7dc603c9028e
>>>> ("sched/fair: Fix PELT integrity for new tasks"), which also
>>>> bring its own problems.
>>>>
>>>> First, it add a new task state TASK_NEW and an unnessary limitation
>>>> that we would fail when change the cgroup of TASK_NEW tasks.
>>
>> This is the limitation that bugfix has brought.
>>
>> We can't change cgroup or switch to fair for task with last_update_time=0
>> if we don't have conditional detach/attach.
>>
>> So we have to:
>>
>> 1. !fair task also need to set last_update_time.
>> 2. cpu_cgroup_can_attach() have to wait for TASK_NEW to fully attached.
>
> I see.
>
> `cgroup_migrate_execute() -> cpu_cgroup_[can|]_attach()` has to wait for
> `wake_up_new_task() -> WRITE_ONCE(p->__state, TASK_RUNNING)`.
>
> Just to understand this change better: IMHO, this is still the case for
> fair tasks, right?
Yes, I think so. We could delete this limitation when we have conditional
detach/attach, since nothing will be detached when last_update_time==0.
Thanks!
>
> `wake_up_new_task() -> post_init_entity_util_avg() ->
> attach_entity_cfs_rq()` has to happen before the fair task can move
> between taskgroups in `cgroup_migrate_execute() -> cpu_cgroup_attach()
> -> sched_move_task() -> sched_change_group() -> task_change_group_fair()`.
>
> [...]
next prev parent reply other threads:[~2022-07-20 13:44 UTC|newest]
Thread overview: 37+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-07-13 4:04 [PATCH v2 00/10] sched: task load tracking optimization and cleanup Chengming Zhou
2022-07-13 4:04 ` [PATCH v2 01/10] sched/fair: combine detach into dequeue when migrating task Chengming Zhou
2022-07-13 4:04 ` [PATCH v2 02/10] sched/fair: update comments in enqueue/dequeue_entity() Chengming Zhou
2022-07-13 4:04 ` [PATCH v2 03/10] sched/fair: maintain task se depth in set_task_rq() Chengming Zhou
2022-07-14 12:30 ` Dietmar Eggemann
2022-07-14 13:03 ` [External] " Chengming Zhou
2022-07-18 7:16 ` Vincent Guittot
2022-07-13 4:04 ` [PATCH v2 04/10] sched/fair: remove redundant cpu_cgrp_subsys->fork() Chengming Zhou
2022-07-14 12:31 ` Dietmar Eggemann
2022-07-14 13:06 ` [External] " Chengming Zhou
2022-07-13 4:04 ` [PATCH v2 05/10] sched/fair: reset sched_avg last_update_time before set_task_rq() Chengming Zhou
2022-07-14 12:31 ` Dietmar Eggemann
2022-07-19 8:49 ` Vincent Guittot
2022-07-13 4:04 ` [PATCH v2 06/10] sched/fair: delete superfluous SKIP_AGE_LOAD Chengming Zhou
2022-07-14 12:33 ` Dietmar Eggemann
2022-07-14 13:24 ` [External] " Chengming Zhou
2022-07-13 4:04 ` [PATCH v2 07/10] sched/fair: use update_load_avg() to attach/detach entity load_avg Chengming Zhou
2022-07-15 11:18 ` Dietmar Eggemann
2022-07-15 16:21 ` [External] " Chengming Zhou
2022-07-19 10:29 ` Vincent Guittot
2022-07-20 13:40 ` Chengming Zhou
2022-07-20 15:34 ` Vincent Guittot
2022-07-21 13:56 ` Chengming Zhou
2022-07-21 14:13 ` Vincent Guittot
2022-07-19 15:02 ` Dietmar Eggemann
2022-07-20 13:43 ` Chengming Zhou [this message]
2022-07-13 4:04 ` [PATCH v2 08/10] sched/fair: fix load tracking for new forked !fair task Chengming Zhou
2022-07-19 12:35 ` Vincent Guittot
2022-07-20 13:48 ` [External] " Chengming Zhou
2022-07-13 4:04 ` [PATCH v2 09/10] sched/fair: stop load tracking when task switched_from_fair() Chengming Zhou
2022-07-14 12:33 ` Dietmar Eggemann
2022-07-14 13:43 ` [External] " Chengming Zhou
2022-07-15 11:15 ` Dietmar Eggemann
2022-07-15 16:35 ` Chengming Zhou
2022-07-19 13:20 ` Vincent Guittot
2022-07-27 10:55 ` Chengming Zhou
2022-07-13 4:04 ` [PATCH v2 10/10] sched/fair: delete superfluous set_task_rq_fair() Chengming Zhou
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=952fc252-30e1-3629-42c4-98a55ada12f9@bytedance.com \
--to=zhouchengming@bytedance.com \
--cc=bsegall@google.com \
--cc=dietmar.eggemann@arm.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@redhat.com \
--cc=peterz@infradead.org \
--cc=rostedt@goodmis.org \
--cc=vincent.guittot@linaro.org \
--cc=vschneid@redhat.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®