Re: [PATCH 4/4] kvm: svm: Support KVM_SEV_SNP_PAGE_TYPE_VMSA at SNP_LAUNCH_UPDATE

On Wed, Jun 17, 2026, J=C3=B6rg R=C3=B6del wrote:
> On Tue, Jun 16, 2026 at 10:55:28AM -0700, Sean Christopherson wrote:
> > Isn't this essentially the same thing as hot-plugging vCPUs after launc=
h?  I
> > have yet to review it in depth (sorry Jethro), but it looks a *lot* sim=
pler.
>=20
> Replacing the VMSA after launch (as supported with the AP_CREATE GHCB cal=
l) is
> different because this has no influence on the launch measurement.
>=20
> The point of providing an initial VMSA is to get a predictable launch
> measurement which is independent of the number of VCPUs the guest has.
>=20
> With the current code KVM will create its own VMSA for each created VCPU =
and
> measure it into the guests initial image. This makes predicting the initi=
al
> launch measurement difficult (as it depends on KVM internals) and fragile=
 because
> KVM-internal changes always carry a risk to change the launch measurement

The same holds true for userspace.

> (which has happened a couple of times already).

Examples?  The SEV features thing jumps to mind, but I don't recall any oth=
ers
off the top of my head.