Re: [PATCH-v2 0/6] vhost/scsi: Add T10 PI SGL passthrough support

["Michael S. Tsirkin" <mst@redhat.com>]

On Tue, Jun 10, 2014 at 10:39:17AM -0700, Nicholas A. Bellinger wrote:
> On Tue, 2014-06-10 at 16:02 +0300, Michael S. Tsirkin wrote:
> > On Tue, Jun 10, 2014 at 09:52:17PM +1000, Stephen Rothwell wrote:
> > > Hi Michael,
> > > 
> > > On Tue, 10 Jun 2014 12:42:54 +0300 "Michael S. Tsirkin" <mst@redhat.com> wrote:
> > > >
> > > > So I see two options:
> > > > - I go ahead with my changes and you with yours and let Linus resolve
> > > >   the conflict.  This means bisect build will be broken since the
> > > >   breakage will likely not be noticed until after the merge.
> > > 
> > > Well, since the resolution is known, the one who submits their tree
> > > later should tell Linus (as suggested by Nicholas).  That is part of
> > > the point of the linux-next tree ... and therefore there would be no
> > > bisect problem.
> > > 
> > > > > Stephen (CC'ed) has included a fix in today's linux-next for the merge
> > > > > conflict here:
> > > > > 
> > > > > https://lkml.org/lkml/2014/6/10/3
> > > > > 
> > > > > Please confirm, as it will be a pointer to Linus within the
> > > > > target-pending/for-next PULL request.
> > > > 
> > > > Yes but this does mean people trying to bisect will
> > > > hit build breakages, not nice.
> > > 
> > > Not necessarily.
> > > 
> > > -- 
> > > Cheers,
> > > Stephen Rothwell                    sfr@canb.auug.org.au
> > 
> > 
> > I don't see how that's possible.
> > Here's a point you might have missed.
> > Nicholas's patch isn't just introducing a merge conflict.
> > It is also buggy.
> > Replacing bit access with has_feature silently fixes the bug.
> > 
> > So if we want to avoid bisect breakage target tree will
> > have to be rebased.
> > 
> > And if doing that anyway, I don't see any reason not
> > to merge everything through the vhost tree, esp
> > since I already put the patches there. Less work for
> > everyone involved.
> > 
> 
> The problem is with Sagi's recent changes wrt to including T10 PI bytes
> into expected data transfer length in target-core, you'll end up
> introducing a different bug into your tree..  ;)

I thought you wanted to fix it after -rc1 anyway?

> Why don't I simply add Stephen's patch to use vhost_has_feature() in
> target-pending/for-next, and we just make sure that the vhost PULL
> request goes out after target-pending..?
> 
> --nab

I can drop the PI feature bit in rc1. You will apply Sagi's changes and
then enable the feature in rc2?