]> git.ipfire.org Git - people/arne_f/kernel.git/commitdiff
KVM: x86: nVMX: fix x2APIC VTPR read intercept
authorMarc Orr <marcorr@google.com>
Tue, 2 Apr 2019 06:56:00 +0000 (23:56 -0700)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 17 Apr 2019 06:38:55 +0000 (08:38 +0200)
commit c73f4c998e1fd4249b9edfa39e23f4fda2b9b041 upstream.

Referring to the "VIRTUALIZING MSR-BASED APIC ACCESSES" chapter of the
SDM, when "virtualize x2APIC mode" is 1 and "APIC-register
virtualization" is 0, a RDMSR of 808H should return the VTPR from the
virtual APIC page.

However, for nested, KVM currently fails to disable the read intercept
for this MSR. This means that a RDMSR exit takes precedence over
"virtualize x2APIC mode", and KVM passes through L1's TPR to L2,
instead of sourcing the value from L2's virtual APIC page.

This patch fixes the issue by disabling the read intercept, in VMCS02,
for the VTPR when "APIC-register virtualization" is 0.

The issue described above and fix prescribed here, were verified with
a related patch in kvm-unit-tests titled "Test VMX's virtualize x2APIC
mode w/ nested".

Signed-off-by: Marc Orr <marcorr@google.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Fixes: c992384bde84f ("KVM: vmx: speed up MSR bitmap merge")
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
arch/x86/kvm/vmx.c

index 440b9587bedb1a431a6c095e1d1627f0f05678c5..6b6bcafd1d2c868d61a31c05c10c71cf2dfb90a2 100644 (file)
@@ -11660,7 +11660,7 @@ static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
                nested_vmx_disable_intercept_for_msr(
                        msr_bitmap_l1, msr_bitmap_l0,
                        X2APIC_MSR(APIC_TASKPRI),
-                       MSR_TYPE_W);
+                       MSR_TYPE_R | MSR_TYPE_W);
 
                if (nested_cpu_has_vid(vmcs12)) {
                        nested_vmx_disable_intercept_for_msr(