xfs
[Top] [All Lists]

Re: lockdep: inconsistent lock state

To: Christian Kujau <lists@xxxxxxxxxxxxxxx>
Subject: Re: lockdep: inconsistent lock state
From: Dave Chinner <david@xxxxxxxxxxxxx>
Date: Mon, 28 Dec 2009 08:27:00 +1100
Cc: xfs@xxxxxxxxxxx, LKML <linux-kernel@xxxxxxxxxxxxxxx>
In-reply-to: <alpine.DEB.2.01.0912261255380.3483@xxxxxxxxxxxxxxxxxx>
References: <alpine.DEB.2.01.0912261255380.3483@xxxxxxxxxxxxxxxxxx>
User-agent: Mutt/1.5.18 (2008-05-17)
On Sat, Dec 26, 2009 at 01:07:49PM -0800, Christian Kujau wrote:
> Hi,
> 
> during tests with bonnie++ on an XFS filesystem, the warning below is 
> issued. From the message's timestamps, the warning occurs during the 
> "Create files in sequential order" phase. I've put a few more details and 
> the config here: http://nerdbynature.de/bits/2.6.33-rc2/xfs_ilock
> 
> Something similar has been reported in http://lkml.org/lkml/2009/2/20/269 
> (for 2.6.29-rc5) and a fix for this false positive made its way into 
> mainline as ed93ec3907f063268ced18728d0653f6199d100c - so I take it this 
> is a different issue then?

This is the usual false positive that is detected - XFS takes locks in
reclaim that it also takes in non-reclaim paths. The reclaim path
from kswapd inverts lock ordering and so we get this report. This
case has never been a deadlock case because an inode in reclaim
cannot be referenced by any other path, so once again it is a
false positive....

Cheers,

Dave.
-- 
Dave Chinner
david@xxxxxxxxxxxxx

<Prev in Thread] Current Thread [Next in Thread>