Re: [Xen-devel] grant table interface addition?

[Kieran Mansley <kmansley@xxxxxxxxxxxxxx>]

On Thu, 2008-10-30 at 15:58 +0000, Keir Fraser wrote:
> On 30/10/08 15:53, "Jan Beulich" <jbeulich@xxxxxxxxxx> wrote:
> 
> >>>> Keir Fraser <keir.fraser@xxxxxxxxxxxxx> 30.10.08 15:15 >>>
> >> On 30/10/08 14:06, "Jan Beulich" <jbeulich@xxxxxxxxxx> wrote:
> >>> As to PAGE_IO, wouldn't it be appropriate to set this flag on pte-s 
> >>> created
> >>> with GNTMAP_device_map? Or is the name misleading me? I'm perhaps
> >>> confused by the fact that the use of this flag is not tied to granting 
> >>> iomem
> >>> pages...
> >> 
> >> The idea was that this would grant-map pages into an iommu. Really it's not
> >> actually been used, but GNTMAP_device_map doesn't cause mappings to appear
> >> in host page tables.
> > 
> > It actually is being used in Solarflare's acceleration backend - what you 
> > say
> > makes me think this could be wrong then.
> 
> I suppose it's useful if you only want the machine address of the granted
> page, and don't care about getting a host mapping.

Yes, that's exactly what we use it for.  The guest passes pages to dom0
that it would like to have added to the buffer table (effectively the
iommu) on the NIC.  dom0 doesn't need to map these pages, but it needs
the MFN to pass to the NIC, and it uses the GNTMAP_device_map to do
this.  I can't remember off the top of my head if performing this grant
has other beneficial side effects, but it wouldn't surprise me if it
did.

Kieran


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel