Re: [PATCH] x86/tdx: Fix zero-extension for CPUID emulation

[Edgecombe, Rick P]

On Tue, 2026-05-12 at 15:37 -0700, Dave Hansen wrote:
> On 5/12/26 15:24, Edgecombe, Rick P wrote:
> > On the other hand, the #VE handler is supposed to do the emulation of the
> > instruction, with the help of the TDVMCALL, so maybe the correctness should be
> > in the guest... Hmm...
> 
> Maybe we should just change the GHCI spec.
> 
> What if we said:
> 
>  | Operand 	       | ... |
>  | R12 (lower 32 bits) | EAX |
>  | R13 (lower 32 bits) | EBX |
>  | R14 (lower 32 bits) | ECX |
>  | R15 (lower 32 bits) | EDX |
> 
> Then said the upper 32 bits are undefined. Then the kernel *must* mask
> them to be correct. Then we don't have to do any checking at all and
> there's no ambiguity about what the VMM is allowed to do or what chaos
> it might cause.

Hmm, let me check. It intersects with the other guests/hosts, but hard to see
how the other ones could be out of spec and not be buggy.