On 14/11/19 21:15, Suravee Suthikulpanit wrote:
> + if (svm_get_enable_apicv(svm->vcpu.kvm))
> + svm_request_update_avic(&svm->vcpu, true);
> + if (kvm_vcpu_apicv_active(vcpu))
> + svm_request_update_avic(vcpu, false);
In both cases, "if (avic)" is the right condition to test to ensure that
the bit is kept in sync with whether we are or not in the interrupt
window. kvm_request_apicv_update will check whether APICv is already
active or not.
In fact, the condition can be moved to svm_request_update_avic. Putting
everything together, we get:
diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 375352d08545..7c316c4716f9 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -878,6 +878,7 @@ enum kvm_irqchip_mode {
#define APICV_INHIBIT_REASON_DISABLE 0
#define APICV_INHIBIT_REASON_HYPERV 1
#define APICV_INHIBIT_REASON_NESTED 2
+#define APICV_INHIBIT_REASON_IRQWIN 3
struct kvm_arch {
unsigned long n_used_mmu_pages;
diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index af90f83d7123..6d300c16d756 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -387,6 +387,7 @@ struct amd_svm_iommu_ir {
static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
static void svm_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa);
static void svm_complete_interrupts(struct vcpu_svm *svm);
+static void svm_toggle_avic_for_irq_window(struct kvm_vcpu *vcpu, bool activate);
static inline void avic_post_state_restore(struct kvm_vcpu *vcpu);
static int nested_svm_exit_handled(struct vcpu_svm *svm);
@@ -4461,6 +4462,14 @@ static int interrupt_window_interception(struct vcpu_svm *svm)
{
kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
svm_clear_vintr(svm);
+
+ /*
+ * For AVIC, the only reason to end up here is ExtINTs.
+ * In this case AVIC was temporarily disabled for
+ * requesting the IRQ window and we have to re-enable it.
+ */
+ svm_toggle_avic_for_irq_window(&svm->vcpu, true);
+
svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
mark_dirty(svm->vmcb, VMCB_INTR);
++svm->vcpu.stat.irq_window_exits;
@@ -5164,6 +5173,17 @@ static void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
{
}
+static void svm_toggle_avic_for_irq_window(struct kvm_vcpu *vcpu, bool activate)
+{
+ if (!avic || !lapic_in_kernel(vcpu))
+ return;
+
+ srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
+ kvm_request_apicv_update(vcpu->kvm, activate,
+ APICV_INHIBIT_REASON_IRQWIN);
+ vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
+}
+
static int svm_set_pi_irte_mode(struct kvm_vcpu *vcpu, bool activate)
{
int ret = 0;
@@ -5504,9 +5524,6 @@ static void enable_irq_window(struct kvm_vcpu *vcpu)
{
struct vcpu_svm *svm = to_svm(vcpu);
- if (kvm_vcpu_apicv_active(vcpu))
- return;
-
/*
* In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
* 1, because that's a separate STGI/VMRUN intercept. The next time we
@@ -5516,6 +5533,13 @@ static void enable_irq_window(struct kvm_vcpu *vcpu)
* window under the assumption that the hardware will set the GIF.
*/
if ((vgif_enabled(svm) || gif_set(svm)) && nested_svm_intr(svm)) {
+ /*
+ * IRQ window is not needed when AVIC is enabled,
+ * unless we have pending ExtINT since it cannot be injected
+ * via AVIC. In such case, we need to temporarily disable AVIC,
+ * and fallback to injecting IRQ via V_IRQ.
+ */
+ svm_toggle_avic_for_irq_window(vcpu, false);
svm_set_vintr(svm);
svm_inject_irq(svm, 0x0);
}
@@ -7328,7 +7352,8 @@ static bool svm_check_apicv_inhibit_reasons(ulong bit)
{
ulong supported = BIT(APICV_INHIBIT_REASON_DISABLE) |
BIT(APICV_INHIBIT_REASON_HYPERV) |
- BIT(APICV_INHIBIT_REASON_NESTED);
+ BIT(APICV_INHIBIT_REASON_NESTED) |
+ BIT(APICV_INHIBIT_REASON_IRQWIN);
return supported & BIT(bit);
}
