linux-kernel - Re: [External] Re: [PATCH 3/8] sched/fair: remove redundant cpu_cgrp

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <e93b5c38-0291-d6e4-811a-9aed3d6e0f87@linux.dev>
Date:   Tue, 12 Jul 2022 00:25:05 +0800
From:   Chengming Zhou <zhouchengming@...edance.com>
To:     Peter Zijlstra <peterz@...radead.org>,
        Chengming Zhou <zhouchengming@...edance.com>
Cc:     mingo@...hat.com, vincent.guittot@...aro.org,
        dietmar.eggemann@....com, rostedt@...dmis.org, bsegall@...gle.com,
        vschneid@...hat.com, linux-kernel@...r.kernel.org
Subject: Re: [External] Re: [PATCH 3/8] sched/fair: remove redundant
 cpu_cgrp_subsys->fork()

On 2022/7/11 21:53, Peter Zijlstra wrote:
> On Mon, Jul 11, 2022 at 09:02:07PM +0800, Chengming Zhou wrote:
>> On 2022/7/11 15:35, Peter Zijlstra wrote:
>>> On Sat, Jul 09, 2022 at 11:13:48PM +0800, Chengming Zhou wrote:
>>>> We use cpu_cgrp_subsys->fork() to set task group for the new fair task
>>>> in cgroup_post_fork().
>>>>
>>>> Since commit b1e8206582f9 ("sched: Fix yet more sched_fork() races")
>>>> has already set task group for the new fair task in sched_cgroup_fork(),
>>>> so cpu_cgrp_subsys->fork() can be removed.
>>>>
>>>>   cgroup_can_fork()	--> pin parent's sched_task_group
>>>>   sched_cgroup_fork()
>>>>     __set_task_cpu	--> set task group
>>>>   cgroup_post_fork()
>>>>     ss->fork() := cpu_cgroup_fork()	--> set again
>>>>
>>>> After this patch's change, task_change_group_fair() only need to
>>>> care about task cgroup migration, make the code much simplier.
>>>
>>> This:
>>>
>>>> This patch also move the task se depth setting to set_task_rq(), which
>>>> will set correct depth for the new task se in sched_cgroup_fork().
>>>>
>>>> The se depth setting in attach_entity_cfs_rq() is removed since
>>>> set_task_rq() is a better place to do this when task moves across
>>>> CPUs/groups.
>>>
>>> really should have been it's own patch. And this actually scares me. Did
>>> you test with priority inheritance bumping the task to FIFO while things
>>> change?
>>>
>>> This has nothing to do with fork().
>>
>> Ok, will put this in another patch, so this patch still need this line:
>>
>>   p->se.depth = tg->se[cpu] ? tg->se[cpu]->depth + 1 : 0;
>>
>> in set_task_rq() to set depth for new forked task.
> 
> That would suggest you ordered your patches wrong.

Yeah, I understand now, should put this se depth change before..

> 
>> I didn't test with "priority inheritance bumping the task to FIFO" case,
>> do you mean the rt_mutex_setprio() bump a fair task to FIFO?
>>
>> Sorry, I don't get how removing depth setting in attach_entity_cfs_rq()
>> affect that. Could you explain more so I can test it?
> 
> Well, if you look at the commit that introduced that code:
> 
>   eb7a59b2c888 ("sched/fair: Reset se-depth when task switched to FAIR")
> 
> then it's clear that the original problem was the task temporarily not
> being in the fair class. The most common way for that to be so is
> through PI.

Thanks for the hint, I read that commit, should be no problem if we have
se depth setting in set_task_rq().

That problem was we only maintain task se depth in task_move_group_fair(),
if a !fair task move task group, its se depth will not be updated, so
that commit fix the problem by update se depth in switched_to_fair().

In this patch we maintain se depth setting in set_task_rq(), which will be
called when CPU/cgroup change, so its depth will always be correct. So it's
ok to remove task se depth setting in attach_entity_cfs_rq().

Thanks.


>