Fix two bugs introduced by a semi-recent fix for AMD's Page Real Mode. Patch 1 was tested against the syzkaller testcase that exposed the bug[*]. Patch 2 was testing by enabling the VMware backdoor in L1 to turn on #GP interception, hacking L0 KVM to passthrough MSR_IA32_VMX_BASIC to L1 and then writing the MSR from L2 while in Real Mode, thus forcing L0 KVM to emulate the WRMSR from L2 and synthesiz a #GP VM-Exit into L1. Confirmed that bad behavior in L1 with an assertion: diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c index bcac3efcde41..ef3bb5ab9654 100644 --- a/arch/x86/kvm/vmx/vmx.c +++ b/arch/x86/kvm/vmx/vmx.c @@ -5178,6 +5178,9 @@ static int handle_exception_nmi(struct kvm_vcpu *vcpu) vect_info = vmx->idt_vectoring_info; intr_info = vmx_get_intr_info(vcpu); + WARN_ONCE((intr_info & INTR_INFO_DELIVER_CODE_MASK) && !is_protmode(vcpu), + "Exception VM-Exit shouldn't report error code when CPU is in Real Mode"); + /* * Machine checks are handled by handle_exception_irqoff(), or by * vmx_vcpu_run() if a #MC occurs on VM-Entry. NMIs are handled by [*] https://lkml.kernel.org/r/ZBNrWZQhMX8AHzWM%40google.com Sean Christopherson (2): KVM: x86: Clear "has_error_code", not "error_code", for RM exception injection KVM: nVMX: Do not report error code when synthesizing VM-Exit from Real Mode arch/x86/kvm/vmx/nested.c | 7 ++++++- arch/x86/kvm/x86.c | 11 +++++++++-- 2 files changed, 15 insertions(+), 3 deletions(-) base-commit: 45dd9bc75d9adc9483f0c7d662ba6e73ed698a0b -- 2.40.0.rc2.332.ga46443480c-goog