Date:   Sat, 11 Feb 2023 18:01:19 +0000
From:   Qais Yousef <qyousef@layalina.io>
To:     Dietmar Eggemann <dietmar.eggemann@arm.com>
Cc:     Vincent Guittot <vincent.guittot@linaro.org>,
        Ingo Molnar <mingo@kernel.org>,
        Peter Zijlstra <peterz@infradead.org>,
        linux-kernel@vger.kernel.org, Lukasz Luba <lukasz.luba@arm.com>,
        Wei Wang <wvw@google.com>, Xuewen Yan <xuewen.yan94@gmail.com>,
        Hank <han.lin@mediatek.com>,
        Jonathan JMChen <Jonathan.JMChen@mediatek.com>
Subject: Re: [PATCH v2 2/3] sched/uclamp: Ignore (util == 0) optimization in
 feec() when p_util_max = 0
Message-ID: <20230211180119.4mbfn7j3skvoasop@airbuntu>
References: <20230205224318.2035646-1-qyousef@layalina.io>
 <20230205224318.2035646-3-qyousef@layalina.io>
 <CAKfTPtCf4+orEhYmFBg+tsMH7e5sV5zJZ1k8apjZkYwSAxAu+w@mail.gmail.com>
 <d7f29def-7004-b62d-9266-3ad997111e2f@arm.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <d7f29def-7004-b62d-9266-3ad997111e2f@arm.com>
Precedence: bulk

On 02/08/23 12:52, Dietmar Eggemann wrote:
> On 07/02/2023 11:04, Vincent Guittot wrote:
> > On Sun, 5 Feb 2023 at 23:43, Qais Yousef <qyousef@layalina.io> wrote:
> >>
> >> find_energy_efficient_cpu() bails out early if effective util of the
> >> task is 0. When uclamp is being used, this could lead to wrong decisions
> >> when uclamp_max is set to 0. Cater for that.
> 
> IMHO this needs a little bit more explanation. Someone could argue that
> 'util > 0, uclamp_min=0, uclamp_max=0' is a valid setup for a task which
> should let it appear as a task with 0 util (capped to 0).

You want me to explain the purpose of the optimization then?

The optimization skips energy calculation when util is 0 because the delta will
be 0. But when uclamp_max = 0 util is not really 0 - consequently  the delta
will not be 0.

Would such an explanation clarify things better?

> 
> >> Fixes: d81304bc6193 ("sched/uclamp: Cater for uclamp in find_energy_efficient_cpu()'s early exit condition")
> >> Signed-off-by: Qais Yousef <qyousef@layalina.io>
> >> ---
> >>  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 7a21ee74139f..a8c3d92ff3f6 100644
> >> --- a/kernel/sched/fair.c
> >> +++ b/kernel/sched/fair.c
> >> @@ -7374,7 +7374,7 @@ static int find_energy_efficient_cpu(struct task_struct *p, int prev_cpu)
> >>         target = prev_cpu;
> >>
> >>         sync_entity_load_avg(&p->se);
> >> -       if (!uclamp_task_util(p, p_util_min, p_util_max))
> >> +       if (!uclamp_task_util(p, p_util_min, p_util_max) && p_util_max != 0)
> > 
> > The below should do the same without testing twice p_util_max:
> > uclamp_task_util(p, p_util_min, ULONG_MAX)
> 
> Since uclamp_task_util() is only used here and we don't want to test for
> capping to 0 anymore, why not just get rid of this function and use:
> 
>   !(task_util_est(p) || p_util_min)

That would be better, yes!

Question for you and Vincent. Do we really want this optimization? I started
with removing it - then erred on the conservative side and kept it.

I don't know how often we hit this case and I didn't see any benchmark run to
be able to verify anything when I looked at the history.

It seems helpful in theory - but why we save something if we ignore 0 but not
1 which I suspect will not produce a significant delta either.

I don't mind keeping it - but I think worth thinking if it is really adding
much.


Cheers

--
Qais Yousef