Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 23 Oct 2014 16:32:52 -0400
From:      John Baldwin <jhb@freebsd.org>
To:        Adrian Chadd <adrian@freebsd.org>
Cc:        "svn-src-head@freebsd.org" <svn-src-head@freebsd.org>, "svn-src-all@freebsd.org" <svn-src-all@freebsd.org>, "src-committers@freebsd.org" <src-committers@freebsd.org>, Edward Tomasz =?utf-8?q?Napiera=C5=82a?= <trasz@freebsd.org>, Mateusz Guzik <mjg@freebsd.org>
Subject:   Re: svn commit: r273549 - head/sys/kern
Message-ID:  <201410231632.53172.jhb@freebsd.org>
In-Reply-To: <CAJ-Vmo=teX0ggxDkjErA%2B_TwddedOsyeYXmkXv5pgB44d7ikBA@mail.gmail.com>
References:  <201410231535.s9NFZlbn002656@svn.freebsd.org> <20E7F407-9DC5-4A59-BC1C-7B868C31EF52@FreeBSD.org> <CAJ-Vmo=teX0ggxDkjErA%2B_TwddedOsyeYXmkXv5pgB44d7ikBA@mail.gmail.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On Thursday, October 23, 2014 3:39:39 pm Adrian Chadd wrote:
> Please back this out; it looks like the lock is protecting sf_si.

The followup fix should be fine.  The lock does indeed protect sf_si, but t=
he value
can only transition from non-NULL to NULL at this point, so if it is =3D=3D=
 NULL without
the lock, it is safe to assume it has already been cleared.

> -adrian
>=20
> On 23 October 2014 11:45, Edward Tomasz Napiera=C5=82a <trasz@freebsd.org=
> wrote:
> > Dnia 23 pa=C5=BA 2014 o godz. 20:38 John Baldwin <jhb@freebsd.org> napi=
sa=C5=82(a):
> >
> >>> On Thursday, October 23, 2014 11:35:47 am Mateusz Guzik wrote:
> >>> Author: mjg
> >>> Date: Thu Oct 23 15:35:47 2014
> >>> New Revision: 273549
> >>> URL: https://svnweb.freebsd.org/changeset/base/273549
> >>>
> >>> Log:
> >>>  Avoid taking the lock in selfdfree when not needed.
> >>>
> >>> Modified:
> >>>  head/sys/kern/sys_generic.c
> >>>
> >>> Modified: head/sys/kern/sys_generic.c
> >> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D
> >>> --- head/sys/kern/sys_generic.c    Thu Oct 23 15:16:40 2014    (r2735=
48)
> >>> +++ head/sys/kern/sys_generic.c    Thu Oct 23 15:35:47 2014    (r2735=
49)
> >>> @@ -1600,10 +1600,11 @@ static void
> >>> selfdfree(struct seltd *stp, struct selfd *sfp)
> >>> {
> >>>    STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
> >>> -    mtx_lock(sfp->sf_mtx);
> >>> -    if (sfp->sf_si)
> >>> +    if (sfp->sf_si !=3D NULL) {
> >>> +        mtx_lock(sfp->sf_mtx);
> >>>        TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
> >>> -    mtx_unlock(sfp->sf_mtx);
> >>> +        mtx_unlock(sfp->sf_mtx);
> >>> +    }
> >>>    uma_zfree(selfd_zone, sfp);
> >>
> >> How do you ensure that the value you read for sf_si here is up to date=
?  In
> >> particular, if a thread is selecting on multiple fds and one awakens i=
t,
> >> another fd can invoke selwakeup() while the thread is in seltdclear().
> >> In that case, you might see a stale value of sf_si and not realize it =
is
> >> cleared by the selwakeup() after you get the lock and you will invoke
> >> TAILQ_REMOVE an extra time.
> >
> > FWIW, I've just hit a panic in selfdfree().
> >
> >
>=20
>=20

=2D-=20
John Baldwin



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201410231632.53172.jhb>