Skip to contentSkip to conversation

Optimize searching for highest IRR

PATCH Applied Gleb Natapov 2 participants 2 messages Jun 11, 2009 – Jun 14, 2009

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 Optimize searching for highest IRR

Most of the time IRR is empty, so instead of scanning the whole IRR on each VM entry keep a variable that tells us if IRR is not empty. IRR will have to be scanned twice on each IRQ delivery, but this is much more rare than VM entry.

v2:
The only difference from v1 is the comment describing possible race and how it is solved. The race is not created by the patch BTW.

arch/x86/kvm/lapic.c +20 -3
index 44f20cd..38a7fa0 100644
@@ -165,29 +165,45 @@ static int find_highest_vector(void *bitmap)
 
 static inline int apic_test_and_set_irr(int vec, struct kvm_lapic *apic)
 {
+	apic->irr_pending = true;
 	return apic_test_and_set_vector(vec, apic->regs + APIC_IRR);
 }
 
-static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
+static inline int apic_search_irr(struct kvm_lapic *apic)
 {
-	apic_clear_vector(vec, apic->regs + APIC_IRR);
+	return find_highest_vector(apic->regs + APIC_IRR);
 }
 
 static inline int apic_find_highest_irr(struct kvm_lapic *apic)
 {
 	int result;
 
-	result = find_highest_vector(apic->regs + APIC_IRR);
+	if (!apic->irr_pending)
+		return -1;
+
+	result = apic_search_irr(apic);
 	ASSERT(result == -1 || result >= 16);
 
 	return result;
 }
 
+static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
+{
+	apic->irr_pending = false;
+	apic_clear_vector(vec, apic->regs + APIC_IRR);
+	if (apic_search_irr(apic) != -1)
+		apic->irr_pending = true;
+}
+
 int kvm_lapic_find_highest_irr(struct kvm_vcpu *vcpu)
 {
 	struct kvm_lapic *apic = vcpu->arch.apic;
 	int highest_irr;
 
+	/* This may race with setting of irr in __apic_accept_irq() and
+	   value returned may be wrong, but kvm_vcpu_kick() in __apic_accept_irq
+	   will cause vmexit immediately and the value will be recalculated
+	   on the next vmentry. */
 	if (!apic)
 		return 0;
 	highest_irr = apic_find_highest_irr(apic);
@@ -842,6 +858,7 @@ void kvm_lapic_reset(struct kvm_vcpu *vcpu)
 		apic_set_reg(apic, APIC_ISR + 0x10 * i, 0);
 		apic_set_reg(apic, APIC_TMR + 0x10 * i, 0);
 	}
+	apic->irr_pending = false;
 	update_divide_count(apic);
 	atomic_set(&apic->lapic_timer.pending, 0);
 	if (vcpu->vcpu_id == 0)
arch/x86/kvm/lapic.h +1 -0
index a587f83..3f3ecc6 100644
@@ -12,6 +12,7 @@ struct kvm_lapic {
 	struct kvm_timer lapic_timer;
 	u32 divide_count;
 	struct kvm_vcpu *vcpu;
+	bool irr_pending;
 	struct page *regs_page;
 	void *regs;
 	gpa_t vapic_addr;

--
Gleb.

Applied
#1 lore raw
Quoted 6 lines · Gleb Natapov

Most of the time IRR is empty, so instead of scanning the whole IRR on each VM entry keep a variable that tells us if IRR is not empty. IRR will have to be scanned twice on each IRQ delivery, but this is much more rare than VM entry.

Applied, thanks.

Signature · 1 line
error compiling committee.c: too many arguments to function