Re: inconsistent lock state message on 2.6.36.4 stable release

[Mel Gorman <mel@csn.ul.ie>]

On Wed, Mar 02, 2011 at 04:22:53PM -0700, Nadolski, Edmund wrote:
> > 
> > On Wed, Mar 02, 2011 at 03:25:15PM -0500, Greg KH wrote:
> > > On Wed, Mar 02, 2011 at 01:19:48PM -0700, Nadolski, Edmund wrote:
> > > > With v2.6.36.4 stable release and CONFIG_PROVE_LOCKING set, I
> > notice
> > > > the below message during boot (full dmesg attached):
> > >
> > > 2.6.36 is now "end of life" so there's really no need to test, debug,
> > or
> > > report problems here as there's nothing we can do about it.
> > >
> > > > Per git bisect, it seems to have originated with this commit:
> > > >
> > > > $ git bisect good
> > > > c04eb9683fbb6374275309b859fcbf02e1db2c78 is the first bad commit
> > > > commit c04eb9683fbb6374275309b859fcbf02e1db2c78
> > > > Author: Mel Gorman <mel@csn.ul.ie>
> > > > Date:   Thu Jan 13 15:45:41 2011 -0800
> > > >
> > > >     mm: page allocator: adjust the per-cpu counter threshold when
> > > > memory is low
> > > >
> > > >     commit 88f5acf88ae6a9778f6d25d0d5d7ec2d57764a97 upstream.
> > > >
> > > > The message goes away after I do a git revert on this commit. I've
> > > > also seen this message on 2.6.37.2, but not on 2.6.38-rc7.
> > >
> > > Care to figure out what commit fixed this so that we can backport it
> > to
> > > the .37-stable kernel?
> > >
> > > git bisect can be used to do this as well.
> > >
> > 
> > It looks like the call to get_online_cpus() that is causing the
> > problem. If
> > so, I believe it to be fixed by the commit [b44129b3: mm: vmstat: use a
> > single setter function and callback for adjusting percpu thresholds].
> > Edmund, care to test with that patch applied?
> 
> This patch looks good, I've applied it onto 2.6.37.2 and I no longer see the message.
> 

Good stuff, thanks.

I'll send the patch to the original reviewers and cc stable to ask for
it to be picked up.

-- 
Mel Gorman
SUSE Labs