Currently on SVM, the kvm_request_apicv_update calls the 'pre_update_apicv_exec_ctrl' without doing any synchronization and that function toggles the APIC_ACCESS_PAGE_PRIVATE_MEMSLOT. If there is a mismatch between that memslot state and the AVIC state, while a vCPU is in guest mode, an APIC mmio write can be lost: For example: VCPU0: enable the APIC_ACCESS_PAGE_PRIVATE_MEMSLOT VCPU1: write to an APIC mmio register. Since AVIC is still disabled on VCPU1, the access will not be intercepted by it, and neither will it cause MMIO fault, but rather it will just update the dummy page mapped into the APIC_ACCESS_PAGE_PRIVATE_MEMSLOT. Fix that by blocking guest entries while we update the memslot. Signed-off-by: Maxim Levitsky <mlevitsk@xxxxxxxxxx> --- arch/x86/kvm/x86.c | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c index 9af2fbbe0521..6f0d9c231249 100644 --- a/arch/x86/kvm/x86.c +++ b/arch/x86/kvm/x86.c @@ -9231,6 +9231,8 @@ void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit) if (!!old == !!new) return; + kvm_block_guest_entries(kvm); + trace_kvm_apicv_update_request(activate, bit); if (kvm_x86_ops.pre_update_apicv_exec_ctrl) static_call(kvm_x86_pre_update_apicv_exec_ctrl)(kvm, activate); @@ -9243,6 +9245,8 @@ void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit) except = kvm_get_running_vcpu(); kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE, except); + + kvm_allow_guest_entries(kvm); if (except) kvm_vcpu_update_apicv(except); } -- 2.26.3
