Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Fri, 2 Oct 2020 16:30:20 +0200
From:   Peter Zijlstra <peterz@infradead.org>
To:     Valentin Schneider <valentin.schneider@arm.com>
Cc:     Sebastian Andrzej Siewior <bigeasy@linutronix.de>,
        tglx@linutronix.de, mingo@kernel.org, linux-kernel@vger.kernel.org,
        qais.yousef@arm.com, swood@redhat.com, juri.lelli@redhat.com,
        vincent.guittot@linaro.org, dietmar.eggemann@arm.com,
        rostedt@goodmis.org, bsegall@google.com, mgorman@suse.de,
        bristot@redhat.com, vincent.donnefort@arm.com
Subject: Re: [PATCH 0/9] sched: Migrate disable support
Message-ID: <20201002143020.GA2628@hirez.programming.kicks-ass.net>
References: <20200921163557.234036895@infradead.org>
 <20200925181727.ryuacrsipyaz65z7@linutronix.de>
 <jhjzh5dmz3r.mognet@arm.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <jhjzh5dmz3r.mognet@arm.com>
Precedence: bulk

On Fri, Sep 25, 2020 at 08:32:24PM +0100, Valentin Schneider wrote:

> The IRC handout so far is:

I meant to go post a new version today; but of course cleanup took
longer than expected.

In any case, there's a slightly updated version here:

  git://git.kernel.org/pub/scm/linux/kernel/git/peterz/queue.git sched/wip2

it has push/pull balancer changes that sorta work. There's still
something funny with the NO_RT_PUSH_IPI case (and concequently also the
deadline code, which is an exact mirror of that code).

(it looks like it ends up migrating things in circles due to the for
loop in pull_rt_task)

> As for your splat, I think this is what I was worrying about wrt
> suppressing callbacks in the switch but not preventing them from being
> queued. Perhaps the below is "better" than what I previously sent.
> 
> Technically should be doable with a cpu_active() check instead given this
> all gets flipped in sched_cpu_deactivate(), but at least this makes it
> obvious that PUSH suppresses any other callback.
> 
> ---
> diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
> index 50aac5b6db26..40d78a20fbcb 100644
> --- a/kernel/sched/sched.h
> +++ b/kernel/sched/sched.h
> @@ -1403,7 +1403,7 @@ queue_balance_callback(struct rq *rq,
>  {
>         lockdep_assert_held(&rq->lock);
> 
> -	if (unlikely(head->next))
> +	if (unlikely(head->next) || (rq->balance_flags & BALANCE_PUSH))
>                 return;

Yeah, I like that. Let me go add it.