Subject: RE: nfsd lockups with xfs during SPEC SFS testing
From: Steve Lord <lord@sgi.com>
To: "HABBINGA,ERIK   ""(HP-Loveland,ex1)" <erik_habbinga@hp.com>
Cc: "'linux-xfs@oss.sgi.com'" <linux-xfs@oss.sgi.com>
In-Reply-To: <F341E03C8ED6D311805E00902761278C0C35DD17@xfc04.fc.hp.com>
References: <F341E03C8ED6D311805E00902761278C0C35DD17@xfc04.fc.hp.com>
Content-Type: text/plain
Content-Transfer-Encoding: 7bit
Date: 01 Feb 2002 11:34:52 -0600
Message-Id: <1012584892.26363.391.camel@jen.americas.sgi.com>
Mime-Version: 1.0
Sender: owner-linux-xfs@oss.sgi.com
Precedence: bulk

On Thu, 2002-01-31 at 16:38, HABBINGA,ERIK (HP-Loveland,ex1) wrote:
> Steve,
>    A co-worker of mine sent me a patch containing the CVS bits at 3:21pm
> Mountain time 1/30/02, he started the CVS download at 2:37pm Mountain time
> 1/30/02 and ended it at 2:38pm Mountain time 1/30/02.  I've been working on
> a patch to remove the BKL from the nfsd process, and have been seeing
> lockups in the xfs code.  I saw these lockups with XFS CVS downloads from
> 1/10/02 and 1/18/02.  I finally started running SPEC tests with out my
> nfsd-BKL-removal patch and still got lockups in the XFS code.  So, I don't
> think this is a regression.
> 
> I ran another test this morning, and got a different profile of lockups.
> I've attached the decoded output from alt-sysrq.  kupdated is locked up in
> xlog_grant_log_space, and all the nfsd processes are locked up either in:
> 
> - fh_lock (all the nfsd3_proc_create->stext_lock->__down_failed->__down
> cases)
> - nfsd_sync (the nfsd_commit->stext_lock->__down_failed->__down cases)
> - pagebuf_grab_lock (the
> _pagebuf_find_lockable_buffer->stext_lock->__down_failed->__down cases)
> - lock_wait (the xfs_access->mraccessf->lock_wait cases)
> - xlog_grant_log_space
> 
> I'll help anyway I can to track this problem down.
> 
> Erik
> 

Are you using anything unusual as the block device here? I would be
suspicious of it not coming back with I/O completions. Basically
all the places threads are waiting are places we wait blocked for
a read or a write to complete. If you have other filesystems on
the same device, can you do I/O to them?

Steve
 
-- 

Steve Lord                                      voice: +1-651-683-3511
Principal Engineer, Filesystem Software         email: lord@sgi.com