Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Fri, 8 Nov 2019 13:24:47 +0100
From:   Peter Zijlstra <peterz@infradead.org>
To:     Quentin Perret <qperret@google.com>
Cc:     Kirill Tkhai <ktkhai@virtuozzo.com>, linux-kernel@vger.kernel.org,
        aaron.lwe@gmail.com, valentin.schneider@arm.com, mingo@kernel.org,
        pauld@redhat.com, jdesfossez@digitalocean.com,
        naravamudan@digitalocean.com, vincent.guittot@linaro.org,
        dietmar.eggemann@arm.com, juri.lelli@redhat.com,
        rostedt@goodmis.org, bsegall@google.com, mgorman@suse.de,
        kernel-team@android.com, john.stultz@linaro.org
Subject: Re: NULL pointer dereference in pick_next_task_fair
Message-ID: <20191108122447.GQ5671@hirez.programming.kicks-ass.net>
References: <33643a5b-1b83-8605-2347-acd1aea04f93@virtuozzo.com>
 <20191106165437.GX4114@hirez.programming.kicks-ass.net>
 <20191106172737.GM5671@hirez.programming.kicks-ass.net>
 <831c2cd4-40a4-31b2-c0aa-b5f579e770d6@virtuozzo.com>
 <20191107132628.GZ4114@hirez.programming.kicks-ass.net>
 <20191107153848.GA31774@google.com>
 <20191107184356.GF4114@hirez.programming.kicks-ass.net>
 <20191107192907.GA30258@worktop.programming.kicks-ass.net>
 <20191108110212.GA204618@google.com>
 <20191108120034.GK4131@hirez.programming.kicks-ass.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20191108120034.GK4131@hirez.programming.kicks-ass.net>
User-Agent: Mutt/1.10.1 (2018-07-13)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Fri, Nov 08, 2019 at 01:00:34PM +0100, Peter Zijlstra wrote:
> > That would remove one call site to newidle_balance() too, which I think
> > is good. Hackbench probably won't like that, though.
> 
> Yeah, that fast path really is important. I've got a few patches pending
> there, fixing a few things and that gets me 2% extra on a sched-yield
> benchmark.

That is, the fast path also allows a cpu-cgroup optimization that wins
something in the order of 3% for cgroup workloads.

The cgroup optimization is basically that when we schedule from
fair->fair, we can avoid having to put/set the whole cgroup hierarchy
but only have to update the part that changed.

Couple that with the set_next_buddy() from dequeue_task_fair(), which
results in the next task being more likely to be from the same cgroup,
and you've got a win.