panic in 8.3-PRERELEASE

[Konstantin Belousov]

On Wed, Feb 22, 2012 at 11:29:40AM -0500, Rick Macklem wrote:
> Hiroki Sato wrote:
> > Hi,
> > 
> > Just a report, but I got the following panic on an NFS server running
> > 8.3-PRERELEASE:
> > 
> > ----(from here)----
> > pool.allbsd.org dumped core - see /var/crash/vmcore.0
> > 
> > Tue Feb 21 10:59:44 JST 2012
> > 
> > FreeBSD pool.allbsd.org 8.3-PRERELEASE FreeBSD 8.3-PRERELEASE #7: Thu
> > Feb 16 19:29:19 JST 2012 hrs at pool.allbsd.org:/usr/obj/usr/src/sys/POOL
> > amd64
> > 
> > panic: Assertion lock == sq->sq_lock failed at
> > /usr/src/sys/kern/subr_sleepqueue.c:335
> > 
> Oops, I didn't know that mixing msleep() and tsleep() calls on the same
> event wasn't allowed.
> There are two places in the code where it did a:
>   mtx_unlock();
>   tsleep();
> left over from the days when it was written for OpenBSD.
This sequence allows to lost the wakeup which is happen right after
cache unlock (together with clearing the RC_WANTED flag) but before
the thread enters sleep state. The tsleep has a timeout so thread should
recover in 10 seconds, but still.

Anyway, you should use consistent outer lock for the same wchan, i.e.
no lock (tsleep) or mtx (msleep), but not mix them.
> 
> I don't think the mix would actually break anything, except that the
> MPASS() assertion fails, but I've cc'd jhb@ since he seems to have been
> the author of the sleep() stuff.
> 
> Anyhow, please try the attached patch which replaces the mtx_unlock(); tsleep(); with
> msleep()s using PDROP. If the attachment gets lost, the patch is also here:
>   http://people.freebsd.org/~rmacklem/tsleep.patch
> 
> Thanks for reporting this, rick
> ps: Is mtx_lock() now preferred over msleep()?
What do you mean ?
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 196 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-stable/attachments/20120222/9fd1d5e3/attachment.pgp