Re: [Xen-devel] Regressions in v3.3-rc1 introduced by "xen/granttable: Grant tables V2 implementation"

[Ian Campbell <Ian.Campbell@citrix.com>]

On Wed, 2012-01-25 at 20:50 +0000, Konrad Rzeszutek Wilk wrote:
> On Wed, Jan 25, 2012 at 03:23:28PM +0000, Paul Durrant wrote:
> > > -----Original Message-----
> > > From: Paul Durrant
> > > Sent: 25 January 2012 07:17
> > > To: Ian Campbell; Konrad Rzeszutek Wilk
> > > Cc: annie.li@oracle.com; xen-devel@lists.xensource.com; linux-
> > > kernel@vger.kernel.org
> > > Subject: RE: [Xen-devel] Regressions in v3.3-rc1 introduced by
> > > "xen/granttable: Grant tables V2 implementation"
> > > 
> > > > -----Original Message-----
> > > >
> > > > On HVM for gnttab_shared we make an add_to_physmap call with
> > > > XENMAPSPACE_grant_table (in gnttab_map). I don't see any support call
> > > > to do something similar for the status array though and I don't see a
> > > > XENMAPSPACE_* specifically for that case either in the hypervisor either.
> > > >
> > > 
> > > There is no map-space for the status pages. To map them you use the
> > > standard map space but OR a bit (top bit IIRC) into in the idx. It's a bit icky,
> > > but that's how it is.
> > > 
> > 
> > I fixed a bug in xen-unstable a few weeks back that prevented
> mapping of the status frames so I guess the bug is possibly due to
> trying to use gnttab 2 on 4.1, where this bug would hit, but failing
> to check the return code from the status mapping hypercall?
> 
> changeset:   24428:5b4b7e565ab8
> user:        Paul Durrant <paul.durrant@citrix.com>
> date:        Sun Dec 18 14:39:14 2011 +0000
> summary:     Fix grant_table v2 status mapping.
> 
> changeset:   24427:931bf1105730
> user:        Paul Durrant <paul.durrant@citrix.com>
> date:        Sun Dec 18 14:38:32 2011 +0000
> summary:     Allow VMs to query their own grant table version.
> 
> 
> Looks like c/s 24427 needs to be in the hypervior. Hm, but that exists
> in 4.1 too as 23204. Will try back-porting cs 24428 in Xen 4.1 and seeing
> if that does it.

I think this bug is a red herring -- the current code in Linux will
never get anywhere near triggering it because it never even tries to map
the status pages into the p2m.

IOW the call to XENMEM_add_to_physmap in
drivers/xen/grant-table.c:gnttab_map() needs to be done for the status
pages too or there is no chance of v2 grant tables working for HVM
guests.

I think the best bet is to disable v2 for HVM guests for now (something
like your second patch) and implement it for 3.4.

Ian.