cvm guest vsm: implement missing register handling

[sluck-msft]

Adds handling for several VSM-related synthetic registers that were previously unimplemented (or incompletely implemented) for hardware-isolated CVMs, so that a guest VSM (VTL 1) running inside a CVM can query partition/VP VSM state. Also fixes a bug in HvEnableVpVtl when the caller targets itself via HV_VP_INDEX_SELF.

Tested: via private updates to opentmk, booting partner VHD

Fixes #3797, fixes #3798

[Copilot]

Pull request overview

Implements missing Hyper-V VSM synthetic register handling for hardware-isolated CVMs so a guest VSM (VTL1) can query partition/VP VSM state, and fixes HvEnableVpVtl handling when the target VP is specified as HV_VP_INDEX_SELF.

Changes:

• Add VsmPartitionConfig, VsmPartitionStatus, and VsmVpStatus synthetic register handling for hardware-isolated CVMs, plus extend VsmCapabilities reporting.
• Fix HvEnableVpVtl to correctly resolve HV_VP_INDEX_SELF to the current VP index before indexing/dispatch.
• Extend CPUID privilege exposure for CVMs to advertise VP register access and add small utilities for VtlSet.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
vm/hv1/hv1_structs/src/vtl_array.rs	Adds VtlSet::with_vtl and From<VtlSet> for u16 for easier VTL-set bitfield construction.
openhcl/virt_mshv_vtl/src/processor/hardware_cvm/mod.rs	Implements CVM-specific synthetic VSM register reads and fixes HV_VP_INDEX_SELF handling in HvEnableVpVtl.
openhcl/virt_mshv_vtl/src/lib.rs	Tracks whether the partition has VSM privilege and adds per-VP MBEC state tracking.
openhcl/virt_mshv_vtl/src/cvm_cpuid/tdx.rs	Advertises access_vp_registers in the CVM CPUID privilege leaf for TDX.
openhcl/virt_mshv_vtl/src/cvm_cpuid/snp.rs	Advertises access_vp_registers in the CVM CPUID privilege leaf for SNP.

[github-actions]

33 Petri tests failed (4 unstable)

[Steven Malis (smalis-msft)]

Is this different from CvmVtl1State's mbec_enabled?

[sluck-msft]

CvmVtl1State is per-partition, no? This one is per-vp and controlled by a different register. Although, this is making me think I'm missing some validation later on in consuming this variable, so I should look into that.

[Steven Malis (smalis-msft)]

It is per partition but I believe we're enforcing that every VP must match already, see my other comment.

[sluck-msft]

Actually I think it can be different on initialization. Mbec enabled can be false for the vp and true for the partition. But when it's configured for the vp, the new mbec enabled value should be the same as on the partition.

[Steven Malis (smalis-msft)]

I don't think this is necessary, since we know that every vp must match the original vtl1 config, checked 10 lines above.

[Copilot]

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

Suppressed comments (2)

vm/hv1/hv1_structs/src/vtl_array.rs:230

• New VtlSet::with_vtl() and From<VtlSet> for u16 are introduced but not covered by the existing unit tests in this module. Adding a small test helps ensure the builder-style API and bit conversion stay consistent with Vtl bit positions.

impl From<VtlSet> for u16 {
    fn from(set: VtlSet) -> Self {
        set.bits.into_inner()
    }
}

openhcl/virt_mshv_vtl/src/processor/hardware_cvm/mod.rs:2203

• cvm_cpuid_result() no longer clears HV_CPUID_FUNCTION_MS_HV_FEATURES privileges when guest VSM is revoked. After revoke_guest_vsm() transitions to GuestVsmState::NotPlatformSupported, the guest can still observe access_vsm/access_vp_registers set in the features CPUID leaf, which regresses the previous behavior and makes revocation incomplete.

            CpuidFunction(hvdef::HV_CPUID_FUNCTION_MS_HV_ENLIGHTENMENT_INFORMATION) => {
                // If VSM has been revoked (or just isn't available) then don't
                // recommend the use of TLB flush hypercalls. They are only needed
                // for synchronization between VTLs, and the non-hypercall direct
                // path is always more efficient.
                if matches!(
                    *self.cvm_partition().guest_vsm.read(),
                    GuestVsmState::NotPlatformSupported
                ) {
                    [eax, ebx, ecx, edx] =
                        hvdef::HvEnlightenmentInformation::from_cpuid([eax, ebx, ecx, edx])
                            .with_use_hypercall_for_remote_flush_and_local_flush_entire(false)
                            .into_cpuid();
                }
            }

            _ => {}

[Copilot]

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated no new comments.

Suppressed comments (1)

openhcl/virt_mshv_vtl/src/processor/hardware_cvm/mod.rs:383

• VsmVpStatus is implemented in get_vp_register(), but validate_register_access() doesn’t apply the same access_vsm_privilege() gate as the other VSM-related registers. As a result, a VTL0 caller without access_vsm privilege can still read VsmVpStatus (while VsmPartitionStatus/VsmCapabilities are denied), which looks like an unintended access-control inconsistency.

            HvX64RegisterName::VsmPartitionStatus | HvX64RegisterName::VsmCapabilities => {