Re: [Xen-devel] [PATCH 34/34] RFC x86: introduce directio virt cap

[Wei Liu <wei.liu2@xxxxxxxxxx>]

On Tue, Aug 21, 2018 at 12:40:22PM +0200, Roger Pau Monné wrote:
> On Tue, Aug 21, 2018 at 11:25:58AM +0100, Wei Liu wrote:
> > On Tue, Aug 21, 2018 at 10:32:18AM +0200, Roger Pau Monné wrote:
> > > On Fri, Aug 17, 2018 at 04:12:52PM +0100, Wei Liu wrote:
> > > > hvm_directio is set when iommu is enabled, but in fact iommu is not
> > > > tied to HVM. In order to not break existing tools, expose a new flag
> > > > directio for (iommu_enabled && !hvm_enabled).
> > > > 
> > > > RFC This doesn't build at the moment. Do we care about that flag being
> > > > inaccurate?
> > > 
> > > I think there is no hardware out there with an IOMMU that don't have
> > > virtualization extensions (ie: having VTd but not VTx for example),
> > > but maybe I'm wrong.
> > 
> > The question is whether it makes sense to expose the name "hvm_directio"
> > at all when you can't run an HVM guest in the first place.
> > 
> > Also iommu isn't an HVM only feature, PV guests can also make use of it
> > if I understand correctly, hence the suggestion of "directio".
> 
> Right, I see your point.
> 
> Could we remove this hvm_directio artifact and just expose an iommu
> capability?
> 
> Since this is a sysctl I think we can change the interface without
> issues, so I would just
> s/XEN_SYSCTL_PHYSCAP_hvm_directio/XEN_SYSCTL_PHYSCAP_iommu/.
> 

I agree and am very tempted at the moment. But let's wait to see if
there is objection.

Wei.

> Roger.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel