Re: [lockdep] possible circular locking, between &mm->mmap_sem and &dev->ev_mutex

[Nick Piggin <npiggin@suse.de>]

On Sun, Sep 14, 2008 at 12:44:42AM -0700, Andrew Morton wrote:
> On Sun, 14 Sep 2008 09:39:06 +0200 Ingo Molnar <mingo@elte.hu> wrote:
> 
> > with the new mmap_sem lockdep annotations, -tip testing found a third 
> > lockdep assert, see below. Config attached.
> > 
> > 	Ingo
> > 
> > ----------------->
> > [ 6460.634452] 
> > [ 6460.634465] =======================================================
> > [ 6460.634494] [ INFO: possible circular locking dependency detected ]
> > [ 6460.634517] 2.6.27-rc6-tip-00290-g8e229c3-dirty #1
> > [ 6460.634535] -------------------------------------------------------
> > [ 6460.634555] gdm-simple-gree/4778 is trying to acquire lock:
> > [ 6460.634574]  (&mm->mmap_sem){----}, at: [<c018fe33>] might_fault+0x36/0x73
> > [ 6460.634639] 
> > [ 6460.634645] but task is already holding lock:
> > [ 6460.634662]  (&dev->ev_mutex){--..}, at: [<c01c7a76>] inotify_read+0xd8/0x16e
> > [ 6460.634715] 
> > [ 6460.634721] which lock already depends on the new lock.
> 
> Yes, there's a thread in my intray called "inotify_read's ev_mutex vs
> do_page_fault's mmap_sem...".  It's a bit flakey-looking, but there's a
> patch in there.

This is the first one I fixed up, I think? What's flakey about it? I sent
2 versions, and the first had an obvious bug but the second worked for
me and reporter...