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;
MIME-Version: 1.0
References: <20230112162426.217522-1-bristot@kernel.org> <Y8UQdKx+004a28fL@hirez.programming.kicks-ass.net>
 <Y8USikoCYTBl7VFr@gmail.com> <Y8UYtEepRT3xC8O/@gmail.com>
In-Reply-To: <Y8UYtEepRT3xC8O/@gmail.com>
From:   Vincent Guittot <vincent.guittot@linaro.org>
Date:   Mon, 16 Jan 2023 10:51:24 +0100
Message-ID: <CAKfTPtCBn20ECopZoT5iC0by8SgpRN_xi4Q0Q8Uc8VQ6BR=u=Q@mail.gmail.com>
Subject: Re: [PATCH] sched/idle: Make idle poll dynamic per-cpu
To:     Ingo Molnar <mingo@kernel.org>
Cc:     Peter Zijlstra <peterz@infradead.org>,
        Daniel Bristot de Oliveira <bristot@kernel.org>,
        linux-kernel@vger.kernel.org, Ingo Molnar <mingo@redhat.com>,
        Juri Lelli <juri.lelli@redhat.com>,
        Dietmar Eggemann <dietmar.eggemann@arm.com>,
        Steven Rostedt <rostedt@goodmis.org>,
        Ben Segall <bsegall@google.com>, Mel Gorman <mgorman@suse.de>,
        Daniel Bristot de Oliveira <bristot@redhat.com>,
        Valentin Schneider <vschneid@redhat.com>,
        Joe Mario <jmario@redhat.com>,
        "Rafael J. Wysocki" <rjw@rjwysocki.net>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Mon, 16 Jan 2023 at 10:28, Ingo Molnar <mingo@kernel.org> wrote:
>
>
> * Ingo Molnar <mingo@kernel.org> wrote:
>
> > > Urgh, can we please make this a cpuidle governor thing or so? So that
> > > we don't need to invent new interfaces and such.
> >
> > I think the desired property here would be to make this interface on top
> > of pretty much any governor. Ie. have a governor, but also a way to drop
> > any CPU into idle-poll, overriding that.
>
> ... with the goal of having the best governor for power efficiency by
> default - but also the ability to drop a handful of CPUs into the highest
> performance / lowest latency idle mode.
>
> It's a special kind of nested policy, for workload exceptions.

User can set per cpu latency constraint with
/sys/devices/system/cpu/cpu*/power/pm_qos_resume_latency_us
Which is then used by cpuidle governor when selecting an idle state.
The cpuidle governor should then select the idle state that matches
with the wakeup latency for those CPUs but select the most power
efficient for others. Setting a low value should filter all idle
states except the polling one

Regards
Vincent
>
> Thanks,
>
>         Ingo