Re: [PATCH 10/14] Nested Virtualization: svm specific implementation

[Christoph Egger <Christoph.Egger@amd.com>]

On Wednesday 18 August 2010 12:45:11 Tim Deegan wrote:
> At 08:48 +0100 on 18 Aug (1282121312), Christoph Egger wrote:
> > On Tuesday 17 August 2010 18:57:55 Tim Deegan wrote:
> > > At 16:49 +0100 on 17 Aug (1282063795), Christoph Egger wrote:
> > > > > Can you explain why we shouldn't sync the vTPR and the vlapic state
> > > > > when the guest is in nested mode?
> > > >
> > > > When the vcpu is in guest mode then
> > > > v->arch.hvm_svm.vmcb->vintr.fields.tpr represents the tpr of the l2
> > > > guest. The l2 guest is not allowed to touch the l1 guest's vTPR.
> > >
> > > Hmmm.  I'm not sure I understand exactly what the hardware vTPR should
> > > contain when the L2 is running; I'll have to think back about exactly
> > > what the sync of vtpr <-> vlapic means and whether the L2 guest can
> > > cause surprising things to happen by setting the vTPR.
> >
> > The l2 guest can confuse the l1 guest. E.g. Booting Windows 7
> > as l1 guest and the XP mode as l2 guest won't work when done
> > incorrectly.
>
> What cases will cause it to break?  I had just convinced myself that
> your patch was correct but now you've got me worried again. :)

It is correct. Don't worry. :)
I just tried to explain what exactly goes wrong when l2 guest fiddles
with l1 guest's vTPR.

Christoph

-- 
---to satisfy European Law for business letters:
Advanced Micro Devices GmbH
Einsteinring 24, 85609 Dornach b. Muenchen
Geschaeftsfuehrer: Alberto Bozzo, Andrew Bowd
Sitz: Dornach, Gemeinde Aschheim, Landkreis Muenchen
Registergericht Muenchen, HRB Nr. 43632