[PATCH v3] mm: Make snapshotting pages for stable writes a per-bio operation

[Darrick J. Wong]

On Wed, Apr 03, 2013 at 04:42:44PM +0200, Jan Kara wrote:
> On Wed 03-04-13 15:20:19, Mel Gorman wrote:
> > On Tue, Apr 02, 2013 at 10:01:43AM -0700, Darrick J. Wong wrote:
> > > Hi,
> > > 
> > > A couple of weeks have gone by without further comments about this patch.
> > > 
> > > Are you interested in the minor cleanups and added comments, or is the v2 patch
> > > in -next good enough?
> > > 
> > > Apparently Mel Gorman's interested in this patchset too.  Mel: Most of stable
> > > pages part 2 are already in upstream for 3.9... except this piece.  Are you
> > > interested in having this piece in 3.9 also?  Or is 3.10 good enough for
> > > everyone?
> > > 
> > 
> > My understanding is that it only affects ARM and DEBUG_VM so there is a
> > relatively small chance of this generating spurious bug reports.  However,
> > 3.9 is still far enough away that I see no good reason to delay this patch
> > until 3.10 either.
>   No, actually with direct IO, anything that needs stable pages is going to
> blow up quickly because pages attached to bio needn't be from page cache. So
> I think it should better make it into 3.9.

Hmm.  The previous version of this patch has been hanging around in -next for a
few weeks without problems (afaik).  With just a raw 3.9-rc[56] I haven't been
able to produce a failed checksum or kernel crash when running with O_DIRECT,
either with the write-after-checksum reproducer or even a simple dd
oflag=direct.  But maybe I've gotten lucky on x86?

So... Andrew: Would you like to pick up the patch with more descriptive
comments?  And, is it too late to push it for 3.9?  Jan seems to think we might
have a bug (though I haven't encountered it).

I'll resend the patch just in case it got eaten.

--D
> 
> 									Honza
> -- 
> Jan Kara <jack at suse.cz>
> SUSE Labs, CR