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;
Date:   Tue, 1 Jun 2021 09:07:41 +0000
From:   Quentin Perret <qperret@google.com>
To:     Viresh Kumar <viresh.kumar@linaro.org>
Cc:     Vincent Donnefort <vincent.donnefort@arm.com>,
        Peter Zijlstra <peterz@infradead.org>, rjw@rjwysocki.net,
        vincent.guittot@linaro.org, linux-kernel@vger.kernel.org,
        ionela.voinescu@arm.com, lukasz.luba@arm.com,
        dietmar.eggemann@arm.com
Subject: Re: [PATCH v2 3/3] PM / EM: Skip inefficient OPPs
Message-ID: <YLX43QT+5r/4zHoP@google.com>
References: <1621616064-340235-1-git-send-email-vincent.donnefort@arm.com>
 <1621616064-340235-4-git-send-email-vincent.donnefort@arm.com>
 <YKy516IKPfsZj3cY@hirez.programming.kicks-ass.net>
 <20210528050934.muji5bv7ed4k4t6j@vireshk-i7>
 <20210601084725.GA223449@e120877-lin.cambridge.arm.com>
 <20210601085628.75atoc4e34uttqqw@vireshk-i7>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20210601085628.75atoc4e34uttqqw@vireshk-i7>
Precedence: bulk

On Tuesday 01 Jun 2021 at 14:26:28 (+0530), Viresh Kumar wrote:
> On 01-06-21, 09:47, Vincent Donnefort wrote:
> > Seems like no one has been really convinced about the arguments in favor of
> > keeping inefficiencies into EM :) Let me then give a shot with marking the OPPs
> > for the next version.
> 
> Right, I think this is what you should do:
> 
> - Add another flag for OPP entries, and mark them inefficient.
> 
> - Whoever traverses the list to find the next frequency (cpufreq here), checks
>   that flag somehow (or replicates that to its own table) and get the right
>   frequency out.

Just to reiterate here what was discussed on IRC the other day, I still
feel that the choice of an efficient OPP or not is a policy decision,
and should be left to the governor.

It's not obvious to me that the userspace govenor for instance wants any
of this. Same thing with e.g. the powersave governor if the lowest OPPs
are inefficient (yes skipping them will not impact energy, but it will
impact instantaneous power).

So if we're going to move that logic to the cpufreq core, then we'll
probably want two separate APIs and make sure to use the effiency-aware
one is used only from the places where that makes sense.

Thanks,
Quentin