KVM: x86: check the kvm_cpu_get_interrupt result before using it
authorMaxim Levitsky <mlevitsk@redhat.com>
Wed, 26 Jul 2023 13:59:45 +0000 (16:59 +0300)
committerPaolo Bonzini <pbonzini@redhat.com>
Sat, 29 Jul 2023 15:05:25 +0000 (11:05 -0400)
The code was blindly assuming that kvm_cpu_get_interrupt never returns -1
when there is a pending interrupt.

While this should be true, a bug in KVM can still cause this.

If -1 is returned, the code before this patch was converting it to 0xFF,
and 0xFF interrupt was injected to the guest, which results in an issue
which was hard to debug.

Add WARN_ON_ONCE to catch this case and skip the injection
if this happens again.

Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20230726135945.260841-4-mlevitsk@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
arch/x86/kvm/x86.c

index a6b9bea62fb8ac4498cccc7686196f6b63a8c1e9..00b87fcf6da4af23a5673444483e9ad9e03c15b4 100644 (file)
@@ -10203,9 +10203,13 @@ static int kvm_check_and_inject_events(struct kvm_vcpu *vcpu,
                if (r < 0)
                        goto out;
                if (r) {
-                       kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
-                       static_call(kvm_x86_inject_irq)(vcpu, false);
-                       WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
+                       int irq = kvm_cpu_get_interrupt(vcpu);
+
+                       if (!WARN_ON_ONCE(irq == -1)) {
+                               kvm_queue_interrupt(vcpu, irq, false);
+                               static_call(kvm_x86_inject_irq)(vcpu, false);
+                               WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
+                       }
                }
                if (kvm_cpu_has_injectable_intr(vcpu))
                        static_call(kvm_x86_enable_irq_window)(vcpu);