Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:6 as permitted sender) client-ip=2620:137:e000::3:6;
Date:   Thu, 28 Sep 2023 23:05:04 +0100
From:   Qais Yousef <qyousef@layalina.io>
To:     Vincent Guittot <vincent.guittot@linaro.org>
Cc:     "Rafael J. Wysocki" <rafael@kernel.org>,
        Viresh Kumar <viresh.kumar@linaro.org>,
        Ingo Molnar <mingo@kernel.org>,
        Peter Zijlstra <peterz@infradead.org>,
        linux-kernel@vger.kernel.org, linux-pm@vger.kernel.org,
        Dietmar Eggemann <dietmar.eggemann@arm.com>,
        Lukasz Luba <lukasz.luba@arm.com>
Subject: Re: [PATCH 2/4] sched: cpufreq: Fix apply_dvfs_headroom() escaping
 uclamp constraints
Message-ID: <20230928220504.gcft523kvt5jlfoi@airbuntu>
References: <20230829163740.uadhv2jfjuumqk3w@airbuntu>
 <CAKfTPtCP6uX79dOrzN4PxFTMBFrDAMOOrWyZrsVypUQ0RY7BAA@mail.gmail.com>
 <20230907215555.exjxho34ntkjmn6r@airbuntu>
 <CAKfTPtA8Ljy4NBqjw8Wj4pEFc-OCR55QPuwh+5GgrHN6u+ugsg@mail.gmail.com>
 <20230910174638.qe7jqq6mq36brh6o@airbuntu>
 <CAKfTPtBFAXO=CgqSJ1+y=2ppb5t4oErCtvV336fS6J2nSjBCkQ@mail.gmail.com>
 <20230916192509.bportepj7dbgp6ro@airbuntu>
 <CAKfTPtA5JqNCauG-rP3wGfq+p8EEVx9Tvwj6ksM3SYCwRmfCTg@mail.gmail.com>
 <20230924172301.7lqdcsnpqk7trtno@airbuntu>
 <ZRW86K0Y1ECd4NRG@vingu-book>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <ZRW86K0Y1ECd4NRG@vingu-book>
Precedence: bulk

On 09/28/23 19:50, Vincent Guittot wrote:

> > 
> > Yep, absolutely. It seems we agree that CFS shouldn't go above 800 if it is
> > capped even if there's headroom, but the question you have on the way it is
> 
> At least I want to ensure that cpufreq has the right information to make a
> smart decision. In the example above, it's not needed to go above 800 for
> neither cfs nor irq.

Okay you want to do even bigger rework :-) I thought I might have pushed some
boundary with the rework I had in mind hehe.

> I'm not in favor of showing all details to cpufreq because it will have to
> follow the internal changes. In instead, I was thinking of something like:
> 
> /* Function name to be changed */
> unsigned_long effective_cpu_util(int cpu, unsigned int *min, unsigned int *max)
> 
> The function returns the actual utilization of the CPU and some minimum and
> maximum limits with the possibility to have the min and/or Actual values > Max
> because the min would be a hard minimum value whereas max only a soft maximum
> value.
> 
> Min would be the minimum perf to provide to the cpu : typically DL_bw + irq
> Actual would be the actual utilization of the cpu: cfs+rt+dl+irq (after scaling
>   everything in the normal range)
> Max would be the maximum needed performance for normal work: typically the
> minimum between uclamp and capacity
> 
> Then cpufreq can use these 3 values to compute a performance level and it 
> will know up to which perf level it should go and if it is worth it.
> Something likr:

Okay thanks! I think I have better clarity now. Let me try to rework the
patches.


Cheers

--
Qais Yousef