linux-kernel - Re: [PATCH v2 4/4] sched/fair: Make feec() consider uclamp restrictions

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <f15b639d-edb2-4b9d-8983-5589590ac41e@arm.com>
Date:   Tue, 10 Dec 2019 19:23:28 +0100
From:   Dietmar Eggemann <dietmar.eggemann@....com>
To:     Valentin Schneider <valentin.schneider@....com>,
        linux-kernel@...r.kernel.org
Cc:     peterz@...radead.org, mingo@...nel.org, vincent.guittot@...aro.org,
        patrick.bellasi@...bug.net, qperret@...gle.com,
        qais.yousef@....com, morten.rasmussen@....com
Subject: Re: [PATCH v2 4/4] sched/fair: Make feec() consider uclamp
 restrictions

On 03/12/2019 16:59, Valentin Schneider wrote:

Could you replace feec (find_energy_efficient_cpu) with EAS wakeup path
in the subject line? The term EAS is described in
Documentation/scheduler/sched-energy.rst so its probably easier to match
the patch to functionality.

> We've just made task_fits_capacity() uclamp-aware, and
> find_energy_efficient_cpu() needs to go through the same treatment.
> Things are somewhat different here however - we can't directly use
> the now uclamp-aware task_fits_capacity(). Consider the following setup:
> 
>   rq.cpu_capacity_orig = 512
>   rq.util_avg = 200
>   rq.uclamp.max = 768
> 
>   p.util_est = 600
>   p.uclamp.max = 256
> 
>   (p not yet enqueued on rq)
> 
> Using task_fits_capacity() here would tell us that p fits on the above CPU.
> However, enqueuing p on that CPU *will* cause it to become overutilized
> since rq clamp values are max-aggregated, so we'd remain with

I assume it doesn't harm to explicitly mention that this rq.uclamp.max =
768 value comes from another task already enqueued on a cfs_rq of this
rq. This gives same substance to the term max-aggregated here.

>   rq.uclamp.max = 768
> 
> Thus we could reach a high enough frequency to reach beyond 0.8 * 512
> utilization (== overutilized). What feec() needs here is

s/feec()/find_energy_efficient_cpu() ?

> uclamp_rq_util_with() which lets us peek at the future utilization
> landscape, including rq-wide uclamp values.
> 
> Make find_energy_efficient_cpu() use uclamp_rq_util_with() for its
> fits_capacity() check. This is in line with what compute_energy() ends up
> using for estimating utilization.

This is also aligned with schedutil_cpu_util() (you do mention this in
the code later in this patch.

[...]