KVM: Fix race between APIC TMR and IRR
authorAvi Kivity <avi@redhat.com>
Tue, 29 Dec 2009 10:42:16 +0000 (12:42 +0200)
committerMarcelo Tosatti <mtosatti@redhat.com>
Mon, 25 Jan 2010 14:26:36 +0000 (12:26 -0200)
When we queue an interrupt to the local apic, we set the IRR before the TMR.
The vcpu can pick up the IRR and inject the interrupt before setting the TMR,
and perhaps even EOI it, causing incorrect behaviour.

The race is really insignificant since it can only occur on the first
interrupt (usually following interrupts will not change TMR), but it's better
closed than open.

Fixed by reordering setting the TMR vs IRR.

Cc: stable@kernel.org
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
arch/x86/kvm/lapic.c

index 3063a0c4858b462a23dd8e2b7d6b81342a203fc1..ba8c045da7820fceea3286b2a60c1378fe28265d 100644 (file)
@@ -373,6 +373,12 @@ static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
                if (unlikely(!apic_enabled(apic)))
                        break;
 
+               if (trig_mode) {
+                       apic_debug("level trig mode for vector %d", vector);
+                       apic_set_vector(vector, apic->regs + APIC_TMR);
+               } else
+                       apic_clear_vector(vector, apic->regs + APIC_TMR);
+
                result = !apic_test_and_set_irr(vector, apic);
                trace_kvm_apic_accept_irq(vcpu->vcpu_id, delivery_mode,
                                          trig_mode, vector, !result);
@@ -383,11 +389,6 @@ static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
                        break;
                }
 
-               if (trig_mode) {
-                       apic_debug("level trig mode for vector %d", vector);
-                       apic_set_vector(vector, apic->regs + APIC_TMR);
-               } else
-                       apic_clear_vector(vector, apic->regs + APIC_TMR);
                kvm_vcpu_kick(vcpu);
                break;