Re: [RFC PATCH 3/4] KVM: TDX: Exit to userspace for GetTdVmCallInfo

[Xiaoyao Li]

On 6/11/2025 9:37 AM, Binbin Wu wrote:
> On 6/10/2025 5:16 PM, Xiaoyao Li wrote:
> > On 6/10/2025 10:14 AM, Binbin Wu wrote:
> > > Exit to userspace for TDG.VP.VMCALL<GetTdVmCallInfo> via a new KVM exit
> > > reason to allow userspace to provide information about the support of
> > > TDVMCALLs when r12 is 1 for the TDVMCALLs beyond the GHCI base API.
> > >
> > > GHCI spec defines the GHCI base TDVMCALLs: <GetTdVmCallInfo>, <MapGPA>,
> > > <ReportFatalError>, <Instruction.CPUID>, <#VE.RequestMMIO>,
> > > <Instruction.HLT>, <Instruction.IO>, <Instruction.RDMSR> and
> > > <Instruction.WRMSR>. They must be supported by VMM to support TDX 
> > > guests.
> > >
> > > For GetTdVmCallInfo
> > > - When leaf (r12) to enumerate TDVMCALL functionality is set to 0,
> > >    successful execution indicates all GHCI base TDVMCALLs listed 
> > > above are
> > >    supported.
> > >
> > >    Update the KVM TDX document with the set of the GHCI base APIs.
> > >
> > > - When leaf (r12) to enumerate TDVMCALL functionality is set to 1, it
> > >    indicates the TDX guest is querying the supported TDVMCALLs beyond
> > >    the GHCI base TDVMCALLs.
> > >    Exit to userspace to let userspace set the TDVMCALL sub-function 
> > > bit(s)
> > >    accordingly to the leaf outputs.  KVM could set the TDVMCALL bit(s)
> > >    supported by itself when the TDVMCALLs don't need support from 
> > > userspace
> > >    after returning from userspace and before entering guest. 
> > > Currently, no
> > >    such TDVMCALLs implemented, KVM just sets the values returned from
> > >    userspace.
> > >
> > >    A new KVM exit reason KVM_EXIT_TDX_GET_TDVMCALL_INFO and its 
> > > structure
> > >    are added. Userspace is required to handle the exit reason as the 
> > > initial
> > >    support for TDX.
> >
> > It doesn't look like a good and correct design.
> >
> > Consider the case that userspace supports SetupEventNotifyInterrupt 
> > and returns bit 1 of leaf_output[0] as 1 to KVM, and KVM returns it to 
> > TD guest for TDVMCALL_GET_TD_VM_CALL_INFO. So TD guest treats it as 
> > SetupEventNotifyInterrupt is support. But when TD guest issues this 
> > TDVMCALL, KVM doesn't support the exit to userspace for this specific 
> > leaf and userspace doesn't have chance to handle it.
> Previously, I also had the idea of setting the information based on 
> userspace's
> opt-ins.
>
> But for simplicity, this patch set doesn't adopt the opt-in mechanism 
> for KVM
> exit reasons due to TDVMCALLs.
>
> To resolve the issue you mentions that userspace could set a bit that KVM
> doesn't support the exit to userspace, KVM could mask off the bit(s) not
> supported by KVM before returning back to guest.

silently mask off the value provided from userspace is not a stable ABI 
from userspace perspective. A kvm cap to report what value is 
allowed/supported is still helpful.

> If we agree that it's the right time to have the opt-in, we could go the 
> opt-in
> way and KVM could set the information based on userspace's opt-ins without
> exit to userspace for GetTdVmCallInfo.

Let's see what Paolo and Sean will say.