Skip site navigation (1)Skip section navigation (2)
Date:      Sat, 12 Aug 2017 09:23:11 -0600
From:      Warner Losh <imp@bsdimp.com>
To:        Gautam <list@execve.net>
Cc:        freebsd-hackers <freebsd-hackers@freebsd.org>, "gautam@execve.net" <gautam@execve.net>
Subject:   Re: Freebsd 11.0 - system freeze on intensive I/O
Message-ID:  <CANCZdfr2h=PgVRxhX7-DXZhjkvEEMgV7pwhV5xSj%2B5Efe8ua%2BA@mail.gmail.com>
In-Reply-To: <CAD9pFYAyLpffrFUPOJya%2BG5cvkMNduY-j4M8PED9vS21RaDVOA@mail.gmail.com>
References:  <CAD9pFYAAy8jKcABNTun-Y3knrbUreQ7q1-CCj=323HgLMC5e6g@mail.gmail.com> <598ADB65.8060701@grosbein.net> <CALH631mEEcUwceh1UYhpF664SoEA4P%2BU87VEhrY-vnZrrQ%2Bw8g@mail.gmail.com> <B0B159BA-3C4A-4A5C-B3EB-CBF2CF251274@dsl-only.net> <CAD9pFYCxBO18GmRE6=-i==j3puJqgQ7bvPk7dmP%2B=T2hgbaAcw@mail.gmail.com> <CAD9pFYAyLpffrFUPOJya%2BG5cvkMNduY-j4M8PED9vS21RaDVOA@mail.gmail.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, Aug 11, 2017 at 11:17 PM, Gautam <list@execve.net> wrote:

> Hi,
>
> On Wed, Aug 9, 2017 at 10:23 PM, Gautam <list@execve.net> wrote:
>
> > Hi,
> >
> > I just want to comment, that when I tested it there
> > was absolutely no need for any swap usage. I tried the test on a laptop
> > with 8GB RAM and
> > did it from the console mode. But it does seem that there might be some
> > request (read/write) to swap which
> > =E2=80=8Bcauses a lock up to the system.=E2=80=8B
> >
> > =E2=80=8BOne update on this one. I installed a custom stable/11 kernel =
and world
> (sched =3D 4BSD was the only change), and the problem
> is no longer seen. From my untrained eyes, it looks like some kind of swa=
p
> request starvation causing a hang when ULE is in use.=E2=80=8B
>

I did some patches a few years ago that separated out geom requests into
their own uma map, if you wanted and hacked the swap pager to use that to
ensure that it always had BIOs available to do it's I/O. Maybe I should
dust them off and see if they help you. They didn't seem to help me too
much, but didn't hurt too much either.

There's still the issue of some drivers not pre-allocating requests, which
is what makes the deadlock possible in the first place.

Warner



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CANCZdfr2h=PgVRxhX7-DXZhjkvEEMgV7pwhV5xSj%2B5Efe8ua%2BA>