Immediately reinject #GP (if intercepted) if the VMware backdoor is
disabled and the instruction is not affected by the erratum that causes
bogus #GPs on SVM instructions. It is completely reasonable for the
guest to take a #GP(0) with EFER.SVME=1, e.g. when probing an MSR, and
attempting emulation on an unknown instruction is obviously not good.
Fixes: b3f4e11adc7d ("KVM: SVM: Add emulation support for #GP triggered by SVM
instructions")
Cc: Bandan Das <b...@redhat.com>
Cc: Maxim Levitsky <mlevi...@redhat.com>
Signed-off-by: Sean Christopherson <sea...@google.com>
---
arch/x86/kvm/svm/svm.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
index f53e6377a933..707a2f85bcc6 100644
--- a/arch/x86/kvm/svm/svm.c
+++ b/arch/x86/kvm/svm/svm.c
@@ -2263,7 +2263,8 @@ static int gp_interception(struct vcpu_svm *svm)
opcode = svm_instr_opcode(vcpu);
if (opcode == NONE_SVM_INSTR) {
- WARN_ON_ONCE(!enable_vmware_backdoor);
+ if (!enable_vmware_backdoor)
+ goto reinject;
/*
* VMware backdoor emulation on #GP interception only handles
--
2.30.0.365.g02bc693789-goog
