Re: [PATCH v2 5/6] KVM: x86: Track available/dirty register masks as "unsigned long" values

[Sean Christopherson]

On Wed, Apr 15, 2026, Xiaoyao Li wrote:
> On 4/15/2026 7:29 PM, Xiaoyao Li wrote:
> > > Note that updating only the "available" masks is wrong, as TDX needs
> > > to marshall written registers back to their correct location.
> > 
> > In what case it needs to marshall written registers back? Can you
> > elaborate?
> 
> Sorry that I asked a silly question. I mistakenly thought TDVMCALL for
> instructions (like, CPUID, RDMSR, WRMSR) use the same output register as the
> x86 instructions. After checking the TDX GHCI, obviously I'm wrong.
> 
> But I don't understand what's is wrong regarding "updating only the
> "available" masks". Or what is missing for "marshall written registers back
> to their correct location"?

KVM would need to also update the "dirty" mask, so that TDX knows it (a) needs
to propagate state back to the GHCI, and (b) that all registers it expects to be
dirty are indeed marked dirty.