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;
References: <20200624154422.29166-1-vincent.guittot@linaro.org>
User-agent: mu4e 0.9.17; emacs 26.3
From:   Valentin Schneider <valentin.schneider@arm.com>
To:     Vincent Guittot <vincent.guittot@linaro.org>
Cc:     mingo@redhat.com, peterz@infradead.org, juri.lelli@redhat.com,
        dietmar.eggemann@arm.com, rostedt@goodmis.org, bsegall@google.com,
        mgorman@suse.de, linux-kernel@vger.kernel.org,
        rong.a.chen@intel.com, pauld@redhat.com, hdanton@sina.com
Subject: Re: [PATCH] sched/cfs: change initial value of runnable_avg
In-reply-to: <20200624154422.29166-1-vincent.guittot@linaro.org>
Date:   Wed, 24 Jun 2020 17:32:12 +0100
Message-ID: <jhj8sgcs9v7.mognet@arm.com>
MIME-Version: 1.0
Content-Type: text/plain
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


On 24/06/20 16:44, Vincent Guittot wrote:
> Some performance regression on reaim benchmark have been raised with
>   commit 070f5e860ee2 ("sched/fair: Take into account runnable_avg to classify group")
>
> The problem comes from the init value of runnable_avg which is initialized
> with max value. This can be a problem if the newly forked task is finally
> a short task because the group of CPUs is wrongly set to overloaded and
> tasks are pulled less agressively.
>
> Set initial value of runnable_avg equals to util_avg to reflect that there
> is no waiting time so far.
>
> Fixes: 070f5e860ee2 ("sched/fair: Take into account runnable_avg to classify group")
> Reported-by: kernel test robot <rong.a.chen@intel.com>
> Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
> ---
>  kernel/sched/fair.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 0424a0af5f87..45e467bf42fc 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -806,7 +806,7 @@ void post_init_entity_util_avg(struct task_struct *p)
>  		}
>  	}
>  
> -	sa->runnable_avg = cpu_scale;
> +	sa->runnable_avg = sa->util_avg;

IIRC we didn't go for this initially because hackbench behaved slightly
worse with it. Did we end up re-evaluating this? Also, how does this reaim
benchmark behave with it? I *think* the table from that regression thread
says it behaves better, but I had a hard time parsing it (seems like it got
damaged by line wrapping)

Conceptually I'm all for it, so as long as the tests back it up:
Reviewed-by: Valentin Schneider <valentin.schneider@arm.com>

>  
>  	if (p->sched_class != &fair_sched_class) {
>  		/*