Re: Drop "KVM: TDX: Handle TDG.VP.VMCALL<GetTdVmCallInfo> hypercall"

[Paolo Bonzini <pbonzini@redhat.com>]

On 4/16/25 05:12, Edgecombe, Rick P wrote:
> Hi,
> 
> We should consider dropping/reverting "KVM: TDX: Handle
> TDG.VP.VMCALL<GetTdVmCallInfo> hypercall" from the base TDX merge. The reason is
> because Xiaoyao noticed that the GHCI spec[0] implies that KVM should only
> return success for that TDVMCALL if *all* TDVMCALLs are supported, but KVM does
> that despite skipping implementing a few. On investigation there are also no
> users except its selftest[1], and the spec is ambiguous on the right way to
> handle the case in question.
> 
> The spec talks about VMMs not supporting all TDVMCALLs, but doesn't say how to
> enumerate this (i.e. it doesn't says what the GetTdVmCallInfo VMMs should do
> instead of succeed). It actually doesn't cover how to handle if the guest calls
> an unsupported TDVMCALL either. Historically, KVM patches have returned
> TDVMCALL_STATUS_INVALID_OPERAND for any unknown TDVMCALL, as a reasonable
> interpretation of the ambiguous spec. So the spec needs to get clarified in this
> whole area.

It does, but I think we should just implement the remaining TDVMCALLs 
before 6.16 is out, which is in a while.  All that is left is really the 
userspace TDVMCALLs GetQuote, ReportFatalError and 
SetupEventNotifyInterrupt.

For Instruction.PCONFIG and for VE.RequestMMIO a dummy implementation is 
valid and returning success is probably even better than invalid-operand.

Paolo

> Since there are no real callers let's just drop GetTdVmCallInfo for now. We can
> add it back when the GHCI folks amend the spec to close the ambiguities. As a
> bonus we can save some code for the merge.
> 
> We dropped the patch internally and did some testing. Also, Binbin and I
> searched the guest code for any rare callers. Everything seems fine to drop it.
> 
> If we want to leave it in, it's probably not a disaster. We'll just slightly
> diverge from the spec. It may not be a problem depending on how the ambiguity
> resolves in future spec updates.
> 
> Thanks,
> 
> Rick
> 
> [0]
> https://www.intel.com/content/www/us/en/content-details/726790/guest-host-communication-interface-ghci-for-intel-trust-domain-extensions-intel-tdx.html
> [1]
> https://lore.kernel.org/kvm/20250414214801.2693294-14-sagis@google.com/