Re: -mm: xfs lockdep warning

[Christoph Hellwig <hch@xxxxxxxxxxxxx>]

On Fri, Sep 17, 2010 at 10:52:27AM +1000, Dave Chinner wrote:
> Christoph, this implies an inode that has been marked for reclaim
> that has not passed through xfs_fs_evict_inode() after being
> initialised. If it went through the eviction process, the iolock
> would have been re-initialised to a different context. Can you think
> of any path that can get here without going through ->evict? I can't
> off the top of my head...

I think this could happen if the init_inode_always during
re-initialization of an inode in reclaim fails in iget.  I have a patch
to add that I'll run through xfsqa.  It should only happen very rarely.