Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20221027081630.34081-1-zhouchuyi@bytedance.com>
 <CABk29NtDny9qKZbZZ_i8Brwjtqs5GA0G4_SffzK4HzG3RrXVhQ@mail.gmail.com>
 <64d963b6-2d9c-3f93-d427-a1ff705fb65a@bytedance.com> <CABk29NsbGuMPcA8NJagMfPyHij3864F3DVQ6wpCj3UAo8tbFNQ@mail.gmail.com>
 <5af26ac9-3bdb-32d2-77a7-6cd8feca97aa@bytedance.com> <CABk29Nt38d31LnazbcSVB036jUYemC1KRpR2Dn7YgLmXkp-QBQ@mail.gmail.com>
 <8142b5db-f543-57e6-0f68-f62274c0e379@bytedance.com> <CABk29Nsnx=PfeLMEsD0qsnh5-QTHLT4xVB3HmBgGAqqmWnkmvg@mail.gmail.com>
 <0796b344-4c4d-5163-d3f2-c6bbf875b129@bytedance.com>
In-Reply-To: <0796b344-4c4d-5163-d3f2-c6bbf875b129@bytedance.com>
From:   Josh Don <joshdon@google.com>
Date:   Fri, 11 Nov 2022 11:14:54 -0800
Message-ID: <CABk29NtwCi0Jwti2XR=J4gPcnN-udV2aP_A9i4bbWmYHfwrHjA@mail.gmail.com>
Subject: Re: [PATCH] sched/fair: favor non-idle group in tick preemption
To:     Abel Wu <wuyun.abel@bytedance.com>
Cc:     Chuyi Zhou <zhouchuyi@bytedance.com>, peterz@infradead.org,
        juri.lelli@redhat.com, mingo@redhat.com,
        vincent.guittot@linaro.org, linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Thu, Nov 10, 2022 at 7:50 PM Abel Wu <wuyun.abel@bytedance.com> wrote:
>
> > By the way, I'm curious if you modified any of the sched_idle_cpu()
> > and related load balancing around idle entities given that you've made
> > it so that idle entities can have arbitrary weight (since, as I
> > described in my prior email, this can otherwise cause issues there).
>
> Being able to change idle entities' weight can bring nothing but
> convenience, because it can also be achieved by modifying all their
> siblings' weight. Which seems not a strong reason to get merged.
>
> And I'm also thinking that, although rare, a non-idle group can also
> have a weight close or even equal to 3. I guess some users who made
> this kind of setting might only want to benefit from the preemption
> at wakeup? Nevertheless this setting is supported now :)

Strongly disagree with this; part of the semantics for idle relies on
the minimum weight value. It is true that this behavior gets a little
weirder if siblings also have close to min weight, but this is an
artifact of the fact that SCHED_IDLE is built into CFS rather than
being a separate scheduling class. The minimum weight in general is
assumed by load balance, etc. for the purpose of placing non-idle
entities.

For consistency with the per-task SCHED_IDLE behavior, which
effectively makes idle tasks have the max nice value, we also need to
match the cgroup idle behavior.

I think the use case you're describing would honestly be better served
by extending SCHED_BATCH with the preemption properties and using
that, rather than try to overload SCHED_IDLE here. There's a
difference between non-interactive entities that are ok getting
aggressively preempted, vs "idle" entities that should really only be
soaking the remaining cycles on the machine.

Best,
Josh