Re: [RFC 1/4] KVM: nVMX: Don't copy L2's CET state to L1 if VM-entry didn't load it

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] KVM fails to propagate L2's modified BNDCFGS and PAT to L1 when VM_EXIT_CLEAR_BNDCFGS or VM_EXIT_LOAD_IA32_PAT is clear.
--

commit cf134f2b2249852af5c9f30ff344481f932e8567
Author: Zhao Liu <zhao1.liu@intel.com>

KVM: nVMX: Don't copy L2's CET state to L1 if VM-entry didn't load it

On a nested VM-exit that disables VM_EXIT_LOAD_CET_STATE, only copy L2's
CET state from vmcs12 to vmcs01 if VM-entry really loaded that state,
i.e. don't copy when VM-entry fails before loading guest state.

> diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c
> index 151873407abd3..35f0bf84b373a 100644
> --- a/arch/x86/kvm/vmx/nested.c
> +++ b/arch/x86/kvm/vmx/nested.c

[ ... ]

> @@ -4856,12 +4902,13 @@ static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
>  	/*
>  	 * Load CET state from host state if VM_EXIT_LOAD_CET_STATE is set.
>  	 * otherwise CET state should be retained across VM-exit, i.e.,
> -	 * guest values should be propagated from vmcs12 to vmcs01.
> +	 * guest values should be propagated from vmcs12 to vmcs01, but only if
> +	 * L2's CET state is live in hardware.
>  	 */
>  	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_CET_STATE)
>  		vmcs_write_cet_state(vcpu, vmcs12->host_s_cet, vmcs12->host_ssp,
>  				     vmcs12->host_ssp_tbl);
> -	else
> +	else if (nested_l2_state_is_live(vmcs12, VM_ENTRY_LOAD_CET_STATE, l2_state))
>  		vmcs_write_cet_state(vcpu, vmcs12->guest_s_cet, vmcs12->guest_ssp,
>  				     vmcs12->guest_ssp_tbl);
>  

[Severity: High]
This isn't a bug introduced by this patch, but this pre-existing logic error
applies to BNDCFGS and PAT in arch/x86/kvm/vmx/nested.c:load_vmcs12_host_state()
as well.

If VM_EXIT_CLEAR_BNDCFGS is false, KVM skips updating vmcs01's BNDCFGS:

    /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
    if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
        vmcs_write64(GUEST_BNDCFGS, 0);

This causes L1 to incorrectly resume with its original state rather than
retaining L2's modified state.

Similarly, if VM_EXIT_LOAD_IA32_PAT is false, KVM skips updating vmcs01's PAT:

    if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
        vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
        vcpu->arch.pat = vmcs12->host_ia32_pat;
    }

This causes L1 to incorrectly resume with its original PAT instead of retaining
L2's modified state.

In both cases, a malicious or buggy L1 hypervisor that disables these VM-exit
controls could cause L1 to resume with the wrong state (its pre-VMENTRY state
instead of L2's state). This breaks the architectural guarantee of state
preservation on VM-exit and leads to guest hypervisor state corruption.

Should we also propagate L2's modified BNDCFGS and PAT to vmcs01 when these
controls are clear, perhaps reusing the new nested_l2_state_is_live() logic?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260904023105.1167376-1-zhao1.liu@intel.com?part=1