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: <20230530135526.2385378-1-vineeth@bitbyteword.org> <6ba5e4bc-dfb9-dccf-d075-f7bde831acf6@arm.com>
In-Reply-To: <6ba5e4bc-dfb9-dccf-d075-f7bde831acf6@arm.com>
From:   Vineeth Remanan Pillai <vineeth@bitbyteword.org>
Date:   Sun, 4 Jun 2023 21:57:59 -0400
Message-ID: <CAO7JXPgV4p3-c-ZB=Ab0rNphdg4+gb7p2KKYwV3CgnMJiYKvPQ@mail.gmail.com>
Subject: Re: [PATCH v5 1/2] sched/deadline: Fix bandwidth reclaim equation in GRUB
To:     Dietmar Eggemann <dietmar.eggemann@arm.com>
Cc:     luca.abeni@santannapisa.it, Juri Lelli <juri.lelli@redhat.com>,
        Daniel Bristot de Oliveira <bristot@redhat.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Ingo Molnar <mingo@redhat.com>,
        Vincent Guittot <vincent.guittot@linaro.org>,
        Steven Rostedt <rostedt@goodmis.org>,
        Joel Fernandes <joel@joelfernandes.org>,
        youssefesmat@google.com, Ben Segall <bsegall@google.com>,
        Mel Gorman <mgorman@suse.de>,
        Valentin Schneider <vschneid@redhat.com>,
        Jonathan Corbet <corbet@lwn.net>, linux-kernel@vger.kernel.org,
        linux-doc@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Hi Dietmar,

Sorry for my late response..

> > + * Since rq->dl.bw_ratio contains 1 / Umax multiplied by 2^RATIO_SHIFT, dl_bw
> > + * is multiped by rq->dl.bw_ratio and shifted right by RATIO_SHIFT.
>
> nit-pick:
>
> s/multiped/multiplied
>
Sorry I missed this. I am working on couple more GRUB fixes and will fix
this along with those changes.

> Not related to this patch directly but I still can't see how `GRUB-PA`
> with schedutil CPUfreq governor should work together with GRUB reclaiming.
>
> CPU frequency is influenced by Uact (rq->dl.running_bw):
>
> sugov_get_util() -> effective_cpu_util() -> cpu_bw_dl() ->
>
>       return rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT
>
> But the extra GRUB reclaim runtime is calculated based on rq->dl.max_bw
> and AFAICS, Uact is not adjusted by scaled_delta_exec?
>
I haven't had a chance to look at GRUB-PA till now and after reading
your mail, I had a quick read of GRUB-PA paper[1] :-). From what I
understood, the dea of GRUB-PA is not to reclaim to the maximum
allowable bandwidth, but to scale down the frequency to the required
utilization(running_bw). This way the tasks run longer(similar to
reclaiming) but using less power.

GRUB reclaim is calculated indirectly based on running_bw as well as
"Uinact = this_bw - running_bw". We factor in reserved and running bw
into the equation and then scale it down to max_bw. So if my
understanding is correct, GRUB-PA doesn't technically reclaim extra
processing capacity, but just allows the task to run longer at lower
frequency so as to save power. I might be wrong and not an expert
here. Luca, please correct me if I am wrong.

Thanks,
Vineeth

[1] C. Scordino, G. Lipari, A Resource Reservation Algorithm for Power-Aware
    Scheduling of Periodic and Aperiodic Real-Time Tasks, IEEE Transactions
    on Computers, December 2006.