linux-kernel - Re: [PATCH v3] lib/raid6: Reduce high latency by using migrate instead of preempt

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: <YbzUVgPH9KDAMpA1@linutronix.de>
Date:   Fri, 17 Dec 2021 19:17:58 +0100
From:   Sebastian Andrzej Siewior <bigeasy@...utronix.de>
To:     Song Liu <song@...nel.org>
Cc:     Yajun Deng <yajun.deng@...ux.dev>, masahiroy@...nel.org,
        williams@...hat.com, Paul Menzel <pmenzel@...gen.mpg.de>,
        open list <linux-kernel@...r.kernel.org>,
        linux-rt-users@...r.kernel.org,
        linux-raid <linux-raid@...r.kernel.org>, stable@...r.kernel.org,
        Thomas Gleixner <tglx@...utronix.de>
Subject: Re: [PATCH v3] lib/raid6: Reduce high latency by using migrate
 instead of preempt

On 2021-12-17 09:25:25 [-0800], Song Liu wrote:
> > The delay is a jiffy so it depends on CONFIG_HZ. You do benchmark for
> > the best algorithm and if you get preempted during that period then your
> > results may be wrong and you make a bad selection.
> 
> With current code, the delay _should be_ 16 jiffies. However, the experiment
> hits way longer latencies. I agree this may cause inaccurate benchmark results
> and thus suboptimal RAID algorithm.

Everything less than CONFIG_PREEMPT does not have an explicit
requirement for preemption so higher latencies are not unusual. *If*
this is a problem on <= PREEMPT_VOLUNTARY then a cond_resched() between
loops would be the usual thing to do. But only *if* it is a real problem
which I doubt. It is not a preemtible kernel after all…

> I guess the key question is whether long latency at module loading time matters.
> If that doesn't matter, we should just drop this.

Correct. And should this be problematic on PREEMPT_RT then I would
restrict CONFIG_RAID6_PQ_BENCHMARK to !PREEMPT_RT.

> Thanks,
> Song

Sebastian