[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAEXW_YQVu9rN04voYUSWiRrzNxa+nQyNMaTyreOuBn96B8yNSg@mail.gmail.com>
Date: Fri, 18 Jun 2021 20:33:27 -0400
From: Joel Fernandes <joel@...lfernandes.org>
To: Peter Zijlstra <peterz@...radead.org>
Cc: LKML <linux-kernel@...r.kernel.org>,
Dietmar Eggemann <dietmar.eggemann@....com>,
Ingo Molnar <mingo@...hat.com>, Josh Don <joshdon@...gle.com>,
Juri Lelli <juri.lelli@...hat.com>,
Luis Chamberlain <mcgrof@...nel.org>,
Mel Gorman <mgorman@...e.de>,
Qais Yousef <qais.yousef@....com>,
"Rafael J. Wysocki" <rjw@...ysocki.net>,
Vincent Guittot <vincent.guittot@...aro.org>,
Viresh Kumar <viresh.kumar@...aro.org>,
Steven Rostedt <rostedt@...dmis.org>
Subject: Re: [RFC] schedutil: Fix iowait boost issues for slow I/O devices
On Fri, Jun 18, 2021 at 4:34 AM Peter Zijlstra <peterz@...radead.org> wrote:
>
>
> Aside from everything else:
>
> On Fri, Jun 18, 2021 at 12:06:39AM -0400, Joel Fernandes (Google) wrote:
>
> > +unsigned int sysctl_iowait_reset_ticks = 20;
> > +unsigned int sysctl_iowait_apply_ticks = 10;
>
> Never expose HZ to userspace. we even have USER_HZ (no new usage please)
> that fixed existing problems with that.
>
> If you expose a knob, make it in sane time units and convert internally
> to whatever you need. That said; less is more with knobs.
Sorry, you are right and good idea. I will look into making such a
modification while I wait for comments from others as well on rest of
the patch.
thanks,
-Joel
Powered by blists - more mailing lists