Re: 2.6.39-rc3, 2.6.39-rc4: XFS lockup - regression since 2.6.38

From: Dave Chinner <david@fromorbit.com>
To: Markus Trippelsdorf <markus@trippelsdorf.de>
Cc: "James Bottomley" <James.Bottomley@HansenPartnership.com>,
	"Dave Chinner" <dchinner@redhat.com>,
	linux-kernel@vger.kernel.org, xfs@oss.sgi.com,
	"Christoph Hellwig" <hch@infradead.org>,
	xfs-masters@oss.sgi.com,
	"Bruno Prémont" <bonbons@linux-vserver.org>,
	"Alex Elder" <aelder@sgi.com>
Subject: Re: 2.6.39-rc3, 2.6.39-rc4: XFS lockup - regression since 2.6.38
Date: Mon, 2 May 2011 22:40:07 +1000	[thread overview]
Message-ID: <20110502124007.GC2978@dastard> (raw)
In-Reply-To: <20110502061528.GA22538@x4.trippels.de>

On Mon, May 02, 2011 at 08:15:28AM +0200, Markus Trippelsdorf wrote:
> On 2011.04.30 at 16:18 +0200, Bruno Prémont wrote:
> > On Fri, 29 April 2011 Bruno Prémont wrote:
> > > On Fri, 29 April 2011 Markus Trippelsdorf wrote:
> > > > On 2011.04.29 at 11:19 +1000, Dave Chinner wrote:
> > > > > OK, so the common elements here appears to be root filesystems
> > > > > with small log sizes, which means they are tail pushing all the
> > > > > time metadata operations are in progress. Definitely seems like a
> > > > > race in the AIL workqueue trigger mechanism. I'll see if I can
> > > > > reproduce this and cook up a patch to fix it.
> > > > 
> > > > Hmm, I'm wondering if this issue is somehow related to the hrtimer bug,
> > > > that Thomas Gleixner fixed yesterday:
> > > > http://git.us.kernel.org/?p=linux/kernel/git/tip/linux-2.6-tip.git;a=commit;h=ce31332d3c77532d6ea97ddcb475a2b02dd358b4
> > > > http://thread.gmane.org/gmane.linux.kernel.mm/61909/
> > > > 
> > > > It also looks similar to the issue that James Bottomley reported
> > > > earlier: http://thread.gmane.org/gmane.linux.kernel.mm/62185/ 
> > > 
> > > I'm going to see, I've applied Thomas' fix on the box seeing XFS freeze (without
> > > other changes to kernel).
> > > Going to run that kernel for the week-end and beyond if it survives to see what
> > > happens.
> > 
> > Happened again (after a few hours of uptime), so it definitely is not
> > caused by hrtimer bug that Thomas Gleixner fixed.
> 
> I've enabled lock debugging and this is what happened after a few hours
> uptime. (I can't tell if this is a false positive):
> 
> =======================================================
> [ INFO: possible circular locking dependency detected ]
> 2.6.39-rc5-00130-g3fd9952 #10
> -------------------------------------------------------
> kio_file/7364 is trying to acquire lock:
>  (&sb->s_type->i_mutex_key#5/2){+.+...}, at: [<ffffffff81151e4e>] generic_file_splice_write+0xce/0x180
> 
> but task is already holding lock:
>  (xfs_iolock_active){++++++}, at: [<ffffffff811d7d05>] xfs_ilock+0x125/0x1f0
> 
> which lock already depends on the new lock.

Known problem. Been broken for ages, yet I only first saw a lockdep
report for this about a week ago on a 2.6.32 kernel....

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs