linux-kernel - Re: [PATCH v2 1/3] sched/fair: Add asymmetric CPU capacity wakeup scan

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: <20200124151134.GB221730@google.com>
Date:   Fri, 24 Jan 2020 15:11:34 +0000
From:   Quentin Perret <qperret@...gle.com>
To:     Valentin Schneider <valentin.schneider@....com>
Cc:     linux-kernel@...r.kernel.org, mingo@...hat.com,
        peterz@...radead.org, vincent.guittot@...aro.org,
        dietmar.eggemann@....com, morten.rasmussen@....com,
        adharmap@...eaurora.org
Subject: Re: [PATCH v2 1/3] sched/fair: Add asymmetric CPU capacity wakeup
 scan

On Friday 24 Jan 2020 at 14:14:47 (+0000), Valentin Schneider wrote:
> If we fail to find a big enough CPU, we'll just fallback to the rest of
> select_idle_sibling() which will pick an idle CPU, just without caring
> about capacity.
> 
> Now an alternative here would be to:
> - return the first idle CPU on which the task fits (what the above does)
> - else, return the biggest idle CPU we found (this could e.g. still steer
>   the task towards a medium on a tri-capacity system)

Sounds reasonable to me.

> I think what we were trying to go with here is to not entirely hijack
> select_idle_sibling(). If we go with the above alternative, topologies
> with sched_asym_cpucapacity enabled would only ever see
> select_idle_capacity() and not the rest of select_idle_sibling(). Not sure
> if it's a bad thing or not, but it's something to ponder over.

Right, I would think your suggestion above is a pretty sensible policy
for asymmetric systems, and I don't think the rest of
select_idle_sibling() will do a much better job on such systems at
finding an idle CPU than select_idle_capacity() would do, but I see
your point.

Now, not having to re-iterate over the CPUs again might keep the wakeup
latency a bit lower -- perhaps something noticeable with hackbench ?
Worth a try.

In any case, no strong opinion. With that missing call to
sync_entity_load_avg() fixed, the patch looks pretty decent to me.

Thanks,
Quentin