kvm tools: Respect ISR status in virtio header

Inject IRQ to guest only when ISR status is low which means
guest has read ISR status and device has cleared this bit as
the side effect of this reading.

This reduces a lot of unnecessary IRQ inject from device to
guest.

Netpef test shows this patch changes:

the host to guest bandwidth
from 2866.27 Mbps (cpu 33.96%) to 5548.87 Mbps (cpu 53.87%),

the guest to host bandwitdth
form 1408.86 Mbps (cpu 99.9%) to 1301.29 Mbps (cpu 99.9%).

The bottleneck of the guest to host bandwidth is guest cpu power.

Signed-off-by: Asias He <asias.hejun@gmail.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>

diff --git a/tools/kvm/include/kvm/virtio.h b/tools/kvm/include/kvm/virtio.h
index e8df8eb47577a2c6a782319036e511779fcaed01..7f92dead43b46d002a9ab70a687b45867860b058 100644 (file)
--- a/tools/kvm/include/kvm/virtio.h
+++ b/tools/kvm/include/kvm/virtio.h
@@ -8,6 +8,9 @@
 
 #include "kvm/kvm.h"
 
+#define VIRTIO_IRQ_LOW         0
+#define VIRTIO_IRQ_HIGH                1
+
 struct virt_queue {
        struct vring    vring;
        u32             pfn;
@@ -37,4 +40,6 @@ struct vring_used_elem *virt_queue__set_used_elem(struct virt_queue *queue, u32
 
 u16 virt_queue__get_iov(struct virt_queue *queue, struct iovec iov[], u16 *out, u16 *in, struct kvm *kvm);
 
+void virt_queue__trigger_irq(struct virt_queue *vq, int irq, u8 *isr, struct kvm *kvm);
+
 #endif /* KVM__VIRTIO_H */

diff --git a/tools/kvm/virtio/core.c b/tools/kvm/virtio/core.c
index 18d2c413a4644ead9f1ba881a5997318007b21aa..07349848f56709ba05bf315a640f2b60cf91c7cc 100644 (file)
--- a/tools/kvm/virtio/core.c
+++ b/tools/kvm/virtio/core.c
@@ -57,3 +57,11 @@ u16 virt_queue__get_iov(struct virt_queue *queue, struct iovec iov[], u16 *out,
 
        return head;
 }
+
+void virt_queue__trigger_irq(struct virt_queue *vq, int irq, u8 *isr, struct kvm *kvm)
+{
+       if (*isr == VIRTIO_IRQ_LOW) {
+               *isr = VIRTIO_IRQ_HIGH;
+               kvm__irq_line(kvm, irq, VIRTIO_IRQ_HIGH);
+       }
+}

diff --git a/tools/kvm/virtio/net.c b/tools/kvm/virtio/net.c
index df69ab30d9504155423fb3f41b23f227f9784653..0189f7d04ddc5a02f2139c1830919cf058a3a52c 100644 (file)
--- a/tools/kvm/virtio/net.c
+++ b/tools/kvm/virtio/net.c
@@ -35,6 +35,7 @@ struct net_device {
        u32                             guest_features;
        u16                             config_vector;
        u8                              status;
+       u8                              isr;
        u16                             queue_selector;
 
        pthread_t                       io_rx_thread;
@@ -88,8 +89,9 @@ static void *virtio_net_rx_thread(void *p)
                        head    = virt_queue__get_iov(vq, iov, &out, &in, self);
                        len     = readv(net_device.tap_fd, iov, in);
                        virt_queue__set_used_elem(vq, head, len);
+
                        /* We should interrupt guest right now, otherwise latency is huge. */
-                       kvm__irq_line(self, VIRTIO_NET_IRQ, 1);
+                       virt_queue__trigger_irq(vq, VIRTIO_NET_IRQ, &net_device.isr, self);
                }
 
        }
@@ -123,7 +125,8 @@ static void *virtio_net_tx_thread(void *p)
                        virt_queue__set_used_elem(vq, head, len);
                }
 
-               kvm__irq_line(self, VIRTIO_NET_IRQ, 1);
+               virt_queue__trigger_irq(vq, VIRTIO_NET_IRQ, &net_device.isr, self);
+
        }
 
        pthread_exit(NULL);
@@ -175,8 +178,9 @@ static bool virtio_net_pci_io_in(struct kvm *self, u16 port, void *data, int siz
                ioport__write8(data, net_device.status);
                break;
        case VIRTIO_PCI_ISR:
-               ioport__write8(data, 0x1);
-               kvm__irq_line(self, VIRTIO_NET_IRQ, 0);
+               ioport__write8(data, net_device.isr);
+               kvm__irq_line(self, VIRTIO_NET_IRQ, VIRTIO_IRQ_LOW);
+               net_device.isr = VIRTIO_IRQ_LOW;
                break;
        case VIRTIO_MSI_CONFIG_VECTOR:
                ioport__write16(data, net_device.config_vector);