]> git.ipfire.org Git - thirdparty/linux.git/commitdiff
KVM: x86: Move "ack" phase of local APIC IRQ delivery to separate API
authorSean Christopherson <seanjc@google.com>
Fri, 6 Sep 2024 04:34:07 +0000 (21:34 -0700)
committerSean Christopherson <seanjc@google.com>
Tue, 10 Sep 2024 03:14:57 +0000 (20:14 -0700)
Split the "ack" phase, i.e. the movement of an interrupt from IRR=>ISR,
out of kvm_get_apic_interrupt() and into a separate API so that nested
VMX can acknowledge a specific interrupt _after_ emulating a VM-Exit from
L2 to L1.

To correctly emulate nested posted interrupts while APICv is active, KVM
must:

  1. find the highest pending interrupt.
  2. check if that IRQ is L2's notification vector
  3. emulate VM-Exit if the IRQ is NOT the notification vector
  4. ACK the IRQ in L1 _after_ VM-Exit

When APICv is active, the process of moving the IRQ from the IRR to the
ISR also requires a VMWRITE to update vmcs01.GUEST_INTERRUPT_STATUS.SVI,
and so acknowledging the interrupt before switching to vmcs01 would result
in marking the IRQ as in-service in the wrong VMCS.

KVM currently fudges around this issue by doing kvm_get_apic_interrupt()
smack dab in the middle of emulating VM-Exit, but that hack doesn't play
nice with nested posted interrupts, as notification vector IRQs don't
trigger a VM-Exit in the first place.

Cc: Nathan Chancellor <nathan@kernel.org>
Link: https://lore.kernel.org/r/20240906043413.1049633-2-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
arch/x86/kvm/lapic.c
arch/x86/kvm/lapic.h

index 5bb481aefcbcdc4c393ea0de917dc4ffc576002f..c1b2a615acfd6b299dd884c0db056fc72fb8a3f9 100644 (file)
@@ -2922,14 +2922,13 @@ void kvm_inject_apic_timer_irqs(struct kvm_vcpu *vcpu)
        }
 }
 
-int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
+void kvm_apic_ack_interrupt(struct kvm_vcpu *vcpu, int vector)
 {
-       int vector = kvm_apic_has_interrupt(vcpu);
        struct kvm_lapic *apic = vcpu->arch.apic;
        u32 ppr;
 
-       if (vector == -1)
-               return -1;
+       if (WARN_ON_ONCE(vector < 0 || !apic))
+               return;
 
        /*
         * We get here even with APIC virtualization enabled, if doing
@@ -2957,6 +2956,16 @@ int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
                __apic_update_ppr(apic, &ppr);
        }
 
+}
+EXPORT_SYMBOL_GPL(kvm_apic_ack_interrupt);
+
+int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu)
+{
+       int vector = kvm_apic_has_interrupt(vcpu);
+
+       if (vector != -1)
+               kvm_apic_ack_interrupt(vcpu, vector);
+
        return vector;
 }
 
index 7ef8ae73e82d3990fd659d6bb51e34a79a8ba488..db80a2965b9c554cef02418c41902ef89cfecaf1 100644 (file)
@@ -88,6 +88,7 @@ int kvm_create_lapic(struct kvm_vcpu *vcpu);
 void kvm_free_lapic(struct kvm_vcpu *vcpu);
 
 int kvm_apic_has_interrupt(struct kvm_vcpu *vcpu);
+void kvm_apic_ack_interrupt(struct kvm_vcpu *vcpu, int vector);
 int kvm_apic_accept_pic_intr(struct kvm_vcpu *vcpu);
 int kvm_get_apic_interrupt(struct kvm_vcpu *vcpu);
 int kvm_apic_accept_events(struct kvm_vcpu *vcpu);