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;
Message-ID: <44ca58cf-3daa-bec5-2c52-eb5a6dd305b9@arm.com>
Date:   Wed, 6 Apr 2022 19:25:46 +0200
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.5.0
From:   Dietmar Eggemann <dietmar.eggemann@arm.com>
Subject: Re: [PATCH] sched/fair: Return the busiest group only if imbalance is
 meaningful
To:     Vincent Guittot <vincent.guittot@linaro.org>,
        =?UTF-8?B?5b2t5b+X5Yia?= <zgpeng.linux@gmail.com>
Cc:     mingo@redhat.com, peterz@infradead.org, juri.lelli@redhat.com,
        rostedt@goodmis.org, Benjamin Segall <bsegall@google.com>,
        mgorman@suse.de, bristot@redhat.com, linux-kernel@vger.kernel.org
References: <1649244204-23471-1-git-send-email-zgpeng@tencent.com>
 <CAKfTPtDh7qFpdjsEBvT34WEgW8uGbdwsZnyXjq4KpAcOse9Pew@mail.gmail.com>
 <CAE5vP3=ZPGV=PuYb-WJsoQ8tX4yDAjajT_WRU+7gAiW54XgX_g@mail.gmail.com>
 <CAKfTPtA8VjZ-F3vdiKJtpij6PUxu_OR7-54cG6NcS_K7guHF6g@mail.gmail.com>
Content-Language: en-US
In-Reply-To: <CAKfTPtA8VjZ-F3vdiKJtpij6PUxu_OR7-54cG6NcS_K7guHF6g@mail.gmail.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 06/04/2022 17:33, Vincent Guittot wrote:
> On Wed, 6 Apr 2022 at 17:07, 彭志刚 <zgpeng.linux@gmail.com> wrote:

[...]

> I think that this case should be covered by an additional test in
> calculate_imbalance because we should not try to pull load in groupA
> if it's already above  the average load. Something like below should
> cover this
> 
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -9469,6 +9469,16 @@ static inline void calculate_imbalance(struct
> lb_env *env, struct sd_lb_stats *s
> 
>                 sds->avg_load = (sds->total_load * SCHED_CAPACITY_SCALE) /
>                                 sds->total_capacity;
> +
> +               /*
> +                * Don't pull any tasks if this group is already above the
> +                * domain average load.
> +                */
> +               if (local->avg_load >= sds->avg_load) {
> +                       env->imbalance = 0;
> +                       return;
> +               }
> +

LGTM. Like for the `local->group_type == group_overloaded` case in
find_busiest_group() where we force `goto out_balanced`.

[...]