VMX's virtualization of SGX adds a lovely dependency on the guest's
supported xcr0, which is calculated in kvm_update_cpuid(). VMX must
toggled its interception of SGX instructions based on the supported
xcr0, i.e. kvm_x86_ops->cpuid_update() is certainly the correct
location for the dependent code.
kvm_update_cpuid() was originally added by commit 2acf923e38fb ("KVM:
VMX: Enable XSAVE/XRSTOR for guest"). There is no indication that its
placement after kvm_x86_ops->cpuid_update() was anything more than a
"new function at the end" decision.
Inspection of the current code reveals no dependency on kvm_x86_ops's
cpuid_update() in kvm_update_cpuid() or any of its helpers.
- SVM's sole update is to conditionally clear X86_FEATURE_X2APIC.
X86_FEATURE_X2APIC is only consumed by kvm_apic_set_state(), which
is already called immediately prior to kvm_x86_ops->cpuid_update().
- VMX updates only nested VMX MSRs, allowed FEATURE_CONTROL bits,
and VMCS fields, e.g. secondary execution controls, none of which
should bleed back into kvm_update_cpuid() barring an egregious
dependency bug somewhere else.
Signed-off-by: Sean Christopherson <sean.j.christopherson@xxxxxxxxx>
---
arch/x86/kvm/cpuid.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)
diff --git a/arch/x86/kvm/cpuid.c b/arch/x86/kvm/cpuid.c
index 70e488951f25..4c235af5318c 100644
--- a/arch/x86/kvm/cpuid.c
+++ b/arch/x86/kvm/cpuid.c
@@ -222,8 +222,9 @@ int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
vcpu->arch.cpuid_nent = cpuid->nent;
cpuid_fix_nx_cap(vcpu);
kvm_apic_set_version(vcpu);
- kvm_x86_ops->cpuid_update(vcpu);
r = kvm_update_cpuid(vcpu);
+ if (!r)
+ kvm_x86_ops->cpuid_update(vcpu);
out:
vfree(cpuid_entries);
@@ -245,8 +246,9 @@ int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
goto out;
vcpu->arch.cpuid_nent = cpuid->nent;
kvm_apic_set_version(vcpu);
- kvm_x86_ops->cpuid_update(vcpu);
r = kvm_update_cpuid(vcpu);
+ if (!r)
+ kvm_x86_ops->cpuid_update(vcpu);
out:
return r;
}
--
2.22.0
