Re: xfs i_lock vs mmap_sem lockdep trace.

[Dave Chinner <david@fromorbit.com>]

On Sun, Mar 30, 2014 at 08:20:30PM -0400, Dave Jones wrote:
> On Mon, Mar 31, 2014 at 10:43:35AM +1100, Dave Chinner wrote:
>  > On Sat, Mar 29, 2014 at 06:31:09PM -0400, Dave Jones wrote:
>  > > Not sure if I've reported this already (it looks familiar, though I've not managed
>  > > to find it in my sent mail folder).  This is rc8 + a diff to fix the stack usage reports
>  > > I was seeing (diff at http://paste.fedoraproject.org/89854/13210913/raw)
>  > > 
>  > >  ======================================================
>  > >  [ INFO: possible circular locking dependency detected ]
>  > >  3.14.0-rc8+ #153 Not tainted
>  > >  -------------------------------------------------------
>  > >  git/32710 is trying to acquire lock:
>  > >   (&(&ip->i_lock)->mr_lock){++++.+}, at: [<ffffffffc03bd782>] xfs_ilock+0x122/0x250 [xfs]
>  > >  
>  > > but task is already holding lock:
>  > >   (&mm->mmap_sem){++++++}, at: [<ffffffffae7b816a>] __do_page_fault+0x14a/0x610
>  > > 
>  > > which lock already depends on the new lock.
>  > 
>  > filldir on a directory inode vs page fault on regular file. Known
>  > issue, definitely a false positive.
> 
> ah yeah, thought it looked familiar. I think I reported this last summer.
> 
>  > We have to change locking
>  > algorithms to avoid such deficiencies of lockdep (a case of "lockdep
>  > considered harmful", perhaps?) so it's not something I'm about to
>  > rush...
> 
> Bummer, as it makes lockdep useless on my test box using xfs because it
> disables itself after hitting this very quickly.
> (I re-enabled it a couple days ago wondering why I'd left it turned off,
>  chances are it was because of this)

Yup, and seeing as SGI haven't shown any indication that they are
going to help fix it any time soon, it won't get fixed until I get
to it (hopefully) sometime soon.

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com