KVM: Intelligent device lookup on I/O bus

Currently the method of dealing with an IO operation on a bus (PIO/MMIO)
is to call the read or write callback for each device registered
on the bus until we find a device which handles it.

Since the number of devices on a bus can be significant due to ioeventfds
and coalesced MMIO zones, this leads to a lot of overhead on each IO
operation.

Instead of registering devices, we now register ranges which points to
a device. Lookup is done using an efficient bsearch instead of a linear
search.

Performance test was conducted by comparing exit count per second with
200 ioeventfds created on one byte and the guest is trying to access a
different byte continuously (triggering usermode exits).
Before the patch the guest has achieved 259k exits per second, after the
patch the guest does 274k exits per second.

Cc: Avi Kivity <avi@redhat.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

4 files changed, 101 insertions, 23 deletions

diff --git a/arch/x86/kvm/i8254.c b/arch/x86/kvm/i8254.c
index efad72385058..76e3f1cd0369 100644
--- a/arch/x86/kvm/i8254.c
+++ b/arch/x86/kvm/i8254.c
@@ -713,14 +713,16 @@ struct kvm_pit *kvm_create_pit(struct kvm *kvm, u32 flags)
 	kvm_register_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
 
 	kvm_iodevice_init(&pit->dev, &pit_dev_ops);
-	ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS, &pit->dev);
+	ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS, KVM_PIT_BASE_ADDRESS,
+				      KVM_PIT_MEM_LENGTH, &pit->dev);
 	if (ret < 0)
 		goto fail;
 
 	if (flags & KVM_PIT_SPEAKER_DUMMY) {
 		kvm_iodevice_init(&pit->speaker_dev, &speaker_dev_ops);
 		ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS,
-						&pit->speaker_dev);
+					      KVM_SPEAKER_BASE_ADDRESS, 4,
+					      &pit->speaker_dev);
 		if (ret < 0)
 			goto fail_unregister;
 	}
diff --git a/arch/x86/kvm/i8259.c b/arch/x86/kvm/i8259.c
index 19fe855e7953..6b869ce0cc19 100644
--- a/arch/x86/kvm/i8259.c
+++ b/arch/x86/kvm/i8259.c
@@ -459,15 +459,9 @@ static int picdev_in_range(gpa_t addr)
 	}
 }
 
-static inline struct kvm_pic *to_pic(struct kvm_io_device *dev)
-{
-	return container_of(dev, struct kvm_pic, dev);
-}
-
-static int picdev_write(struct kvm_io_device *this,
+static int picdev_write(struct kvm_pic *s,
 			 gpa_t addr, int len, const void *val)
 {
-	struct kvm_pic *s = to_pic(this);
 	unsigned char data = *(unsigned char *)val;
 	if (!picdev_in_range(addr))
 		return -EOPNOTSUPP;
@@ -494,10 +488,9 @@ static int picdev_write(struct kvm_io_device *this,
 	return 0;
 }
 
-static int picdev_read(struct kvm_io_device *this,
+static int picdev_read(struct kvm_pic *s,
 		       gpa_t addr, int len, void *val)
 {
-	struct kvm_pic *s = to_pic(this);
 	unsigned char data = 0;
 	if (!picdev_in_range(addr))
 		return -EOPNOTSUPP;
@@ -525,6 +518,48 @@ static int picdev_read(struct kvm_io_device *this,
 	return 0;
 }
 
+static int picdev_master_write(struct kvm_io_device *dev,
+			       gpa_t addr, int len, const void *val)
+{
+	return picdev_write(container_of(dev, struct kvm_pic, dev_master),
+			    addr, len, val);
+}
+
+static int picdev_master_read(struct kvm_io_device *dev,
+			      gpa_t addr, int len, void *val)
+{
+	return picdev_read(container_of(dev, struct kvm_pic, dev_master),
+			    addr, len, val);
+}
+
+static int picdev_slave_write(struct kvm_io_device *dev,
+			      gpa_t addr, int len, const void *val)
+{
+	return picdev_write(container_of(dev, struct kvm_pic, dev_slave),
+			    addr, len, val);
+}
+
+static int picdev_slave_read(struct kvm_io_device *dev,
+			     gpa_t addr, int len, void *val)
+{
+	return picdev_read(container_of(dev, struct kvm_pic, dev_slave),
+			    addr, len, val);
+}
+
+static int picdev_eclr_write(struct kvm_io_device *dev,
+			     gpa_t addr, int len, const void *val)
+{
+	return picdev_write(container_of(dev, struct kvm_pic, dev_eclr),
+			    addr, len, val);
+}
+
+static int picdev_eclr_read(struct kvm_io_device *dev,
+			    gpa_t addr, int len, void *val)
+{
+	return picdev_read(container_of(dev, struct kvm_pic, dev_eclr),
+			    addr, len, val);
+}
+
 /*
  * callback when PIC0 irq status changed
  */
@@ -537,9 +572,19 @@ static void pic_irq_request(struct kvm *kvm, int level)
 	s->output = level;
 }
 
-static const struct kvm_io_device_ops picdev_ops = {
-	.read     = picdev_read,
-	.write    = picdev_write,
+static const struct kvm_io_device_ops picdev_master_ops = {
+	.read     = picdev_master_read,
+	.write    = picdev_master_write,
+};
+
+static const struct kvm_io_device_ops picdev_slave_ops = {
+	.read     = picdev_slave_read,
+	.write    = picdev_slave_write,
+};
+
+static const struct kvm_io_device_ops picdev_eclr_ops = {
+	.read     = picdev_eclr_read,
+	.write    = picdev_eclr_write,
 };
 
 struct kvm_pic *kvm_create_pic(struct kvm *kvm)
@@ -560,16 +605,39 @@ struct kvm_pic *kvm_create_pic(struct kvm *kvm)
 	/*
 	 * Initialize PIO device
 	 */
-	kvm_iodevice_init(&s->dev, &picdev_ops);
+	kvm_iodevice_init(&s->dev_master, &picdev_master_ops);
+	kvm_iodevice_init(&s->dev_slave, &picdev_slave_ops);
+	kvm_iodevice_init(&s->dev_eclr, &picdev_eclr_ops);
 	mutex_lock(&kvm->slots_lock);
-	ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS, &s->dev);
+	ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS, 0x20, 2,
+				      &s->dev_master);
+	if (ret < 0)
+		goto fail_unlock;
+
+	ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS, 0xa0, 2, &s->dev_slave);
+	if (ret < 0)
+		goto fail_unreg_2;
+
+	ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS, 0x4d0, 2, &s->dev_eclr);
+	if (ret < 0)
+		goto fail_unreg_1;
+
 	mutex_unlock(&kvm->slots_lock);
-	if (ret < 0) {
-		kfree(s);
-		return NULL;
-	}
 
 	return s;
+
+fail_unreg_1:
+	kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &s->dev_slave);
+
+fail_unreg_2:
+	kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &s->dev_master);
+
+fail_unlock:
+	mutex_unlock(&kvm->slots_lock);
+
+	kfree(s);
+
+	return NULL;
 }
 
 void kvm_destroy_pic(struct kvm *kvm)
@@ -577,7 +645,9 @@ void kvm_destroy_pic(struct kvm *kvm)
 	struct kvm_pic *vpic = kvm->arch.vpic;
 
 	if (vpic) {
-		kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &vpic->dev);
+		kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &vpic->dev_master);
+		kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &vpic->dev_slave);
+		kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &vpic->dev_eclr);
 		kvm->arch.vpic = NULL;
 		kfree(vpic);
 	}
diff --git a/arch/x86/kvm/irq.h b/arch/x86/kvm/irq.h
index 53e2d084bffb..2086f2bfba33 100644
--- a/arch/x86/kvm/irq.h
+++ b/arch/x86/kvm/irq.h
@@ -66,7 +66,9 @@ struct kvm_pic {
 	struct kvm *kvm;
 	struct kvm_kpic_state pics[2]; /* 0 is master pic, 1 is slave pic */
 	int output;		/* intr from master PIC */
-	struct kvm_io_device dev;
+	struct kvm_io_device dev_master;
+	struct kvm_io_device dev_slave;
+	struct kvm_io_device dev_eclr;
 	void (*ack_notifier)(void *opaque, int irq);
 	unsigned long irq_states[16];
 };
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 6cb353c83a12..d28dff749dfd 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -3562,7 +3562,11 @@ long kvm_arch_vm_ioctl(struct file *filp,
 			if (r) {
 				mutex_lock(&kvm->slots_lock);
 				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
-							  &vpic->dev);
+							  &vpic->dev_master);
+				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
+							  &vpic->dev_slave);
+				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
+							  &vpic->dev_eclr);
 				mutex_unlock(&kvm->slots_lock);
 				kfree(vpic);
 				goto create_irqchip_unlock;