From mboxrd@z Thu Jan 1 00:00:00 1970 From: Dave Chinner Subject: Re: 2.6.27-rc6: lockdep warning: iprune_mutex at shrink_icache_memory+0x38/0x1a8 Date: Tue, 16 Sep 2008 17:03:17 +1000 Message-ID: <20080916070317.GB5811@disturbed> References: <20080913233138.GA19576@orion> <20080916025204.GL5811@disturbed> <7iduc45t9dvo0396fm78d8uat84uurh131@4ax.com> Mime-Version: 1.0 Return-path: Content-Disposition: inline In-Reply-To: <7iduc45t9dvo0396fm78d8uat84uurh131-e09XROE/p8c@public.gmane.org> Sender: kernel-testers-owner-u79uwXL29TY76Z2rM5mHXA@public.gmane.org List-ID: Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: Grant Coady Cc: Alexander Beregalov , rjw-KKrjLPT3xs0@public.gmane.org, linux-kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, kernel-testers-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, linux-fsdevel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, xfs-VZNHf3L845pBDgjK7y7TUQ@public.gmane.org On Tue, Sep 16, 2008 at 02:31:05PM +1000, Grant Coady wrote: > On Tue, 16 Sep 2008 12:52:04 +1000, Dave Chinner wrote: > > >On Sun, Sep 14, 2008 at 03:31:38AM +0400, Alexander Beregalov wrote: > >> Hi > >> > >> [ INFO: possible circular locking dependency detected ] > >> 2.6.27-rc6-00034-gd1c6d2e #3 > >> ------------------------------------------------------- > >> nfsd/1766 is trying to acquire lock: > >> (iprune_mutex){--..}, at: [] shrink_icache_memory+0x38/0x1a8 > >> > >> but task is already holding lock: > >> (&(&ip->i_iolock)->mr_lock){----}, at: [] > >> xfs_ilock+0xa2/0xd6 > >> > >> > >> I read files through nfs and saw delay for few seconds. > >> System is x86_32, nfs, xfs. > >> The last working kernel is 2.6.27-rc5, > >> I do not know yet is it reproducible or not. > > > > > > > >We need a FAQ for this one. It's a false positive. Google for an > >explanation - I've explained it 4 or 5 times in the past year and > >asked that the lockdep folk invent a special annotation for the > >iprune_mutex (or memory reclaim) because of the way it can cause > >recursion into the filesystem and hence invert lock orders without > >causing deadlocks..... > > Yeah, but a 30 second dreadlock? It's a long wait wondering what's > gone down or not ;) The delay will be probably due to how slow the system can be when it runs out of memory, not from the lockdep report. Cheers, Dave. -- Dave Chinner david-FqsqvQoI3Ljby3iVrkZq2A@public.gmane.org