virtio-net: Introduce a new bottom half packet TX

Based on a patch from Mark McLoughlin, this patch introduces a new
bottom half packet transmitter that avoids the latency imposed by
the tx_timer approach.  Rather than scheduling a timer when a TX
packet comes in, schedule a bottom half to be run from the iothread.
The bottom half handler first attempts to flush the queue with
notification disabled (this is where we could race with a guest
without txburst).  If we flush a full burst, reschedule immediately.
If we send short of a full burst, try to re-enable notification.
To avoid a race with TXs that may have occurred, we must then
flush again.  If we find some packets to send, the guest it probably
active, so we can reschedule again.

tx_timer and tx_bh are mutually exclusive, so we can re-use the
tx_waiting flag to indicate one or the other needs to be setup.
This allows us to seamlessly migrate between timer and bh TX
handling.

The bottom half handler becomes the new default and we add a new
tx= option to virtio-net-pci.  Usage:

-device virtio-net-pci,tx=timer # select timer mitigation vs "bh"

Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

hw/s390-virtio-bus.c

@@ -332,6 +332,7 @@ static VirtIOS390DeviceInfo s390_virtio_net = {
                            net.txtimer, TX_TIMER_INTERVAL),
         DEFINE_PROP_INT32("x-txburst", VirtIOS390Device,
                           net.txburst, TX_BURST),
+        DEFINE_PROP_STRING("tx", VirtIOS390Device, net.tx),
         DEFINE_PROP_END_OF_LIST(),
     },
 };

hw/syborg_virtio.c

@@ -300,6 +300,7 @@ static SysBusDeviceInfo syborg_virtio_net_info = {
                            net.txtimer, TX_TIMER_INTERVAL),
         DEFINE_PROP_INT32("x-txburst", SyborgVirtIOProxy,
                           net.txburst, TX_BURST),
+        DEFINE_PROP_STRING("tx", SyborgVirtIOProxy, net.tx),
         DEFINE_PROP_END_OF_LIST(),
     }
 };

hw/virtio-net.c

@@ -36,6 +36,7 @@ typedef struct VirtIONet
     VirtQueue *ctrl_vq;
     NICState *nic;
     QEMUTimer *tx_timer;
+    QEMUBH *tx_bh;
     uint32_t tx_timeout;
     int32_t tx_burst;
     int tx_waiting;
@@ -700,7 +701,7 @@ static int32_t virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq)
     return num_packets;
 }
 
-static void virtio_net_handle_tx(VirtIODevice *vdev, VirtQueue *vq)
+static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
 {
     VirtIONet *n = to_virtio_net(vdev);
 
@@ -717,6 +718,18 @@ static void virtio_net_handle_tx(VirtIODevice *vdev, VirtQueue *vq)
     }
 }
 
+static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
+{
+    VirtIONet *n = to_virtio_net(vdev);
+
+    if (unlikely(n->tx_waiting)) {
+        return;
+    }
+    virtio_queue_set_notification(vq, 0);
+    qemu_bh_schedule(n->tx_bh);
+    n->tx_waiting = 1;
+}
+
 static void virtio_net_tx_timer(void *opaque)
 {
     VirtIONet *n = opaque;
@@ -731,6 +744,41 @@ static void virtio_net_tx_timer(void *opaque)
     virtio_net_flush_tx(n, n->tx_vq);
 }
 
+static void virtio_net_tx_bh(void *opaque)
+{
+    VirtIONet *n = opaque;
+    int32_t ret;
+
+    n->tx_waiting = 0;
+
+    /* Just in case the driver is not ready on more */
+    if (unlikely(!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)))
+        return;
+
+    ret = virtio_net_flush_tx(n, n->tx_vq);
+    if (ret == -EBUSY) {
+        return; /* Notification re-enable handled by tx_complete */
+    }
+
+    /* If we flush a full burst of packets, assume there are
+     * more coming and immediately reschedule */
+    if (ret >= n->tx_burst) {
+        qemu_bh_schedule(n->tx_bh);
+        n->tx_waiting = 1;
+        return;
+    }
+
+    /* If less than a full burst, re-enable notification and flush
+     * anything that may have come in while we weren't looking.  If
+     * we find something, assume the guest is still active and reschedule */
+    virtio_queue_set_notification(n->tx_vq, 1);
+    if (virtio_net_flush_tx(n, n->tx_vq) > 0) {
+        virtio_queue_set_notification(n->tx_vq, 0);
+        qemu_bh_schedule(n->tx_bh);
+        n->tx_waiting = 1;
+    }
+}
+
 static void virtio_net_save(QEMUFile *f, void *opaque)
 {
     VirtIONet *n = opaque;
@@ -850,8 +898,12 @@ static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
     n->mac_table.first_multi = i;
 
     if (n->tx_waiting) {
+        if (n->tx_timer) {
             qemu_mod_timer(n->tx_timer,
                            qemu_get_clock(vm_clock) + n->tx_timeout);
+        } else {
+            qemu_bh_schedule(n->tx_bh);
+        }
     }
     return 0;
 }
@@ -929,7 +981,22 @@ VirtIODevice *virtio_net_init(DeviceState *dev, NICConf *conf,
     n->vdev.reset = virtio_net_reset;
     n->vdev.set_status = virtio_net_set_status;
     n->rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
-    n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx);
+
+    if (net->tx && strcmp(net->tx, "timer") && strcmp(net->tx, "bh")) {
+        fprintf(stderr, "virtio-net: "
+                "Unknown option tx=%s, valid options: \"timer\" \"bh\"\n",
+                net->tx);
+        fprintf(stderr, "Defaulting to \"bh\"\n");
+    }
+
+    if (net->tx && !strcmp(net->tx, "timer")) {
+        n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx_timer);
+        n->tx_timer = qemu_new_timer(vm_clock, virtio_net_tx_timer, n);
+        n->tx_timeout = net->txtimer;
+    } else {
+        n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx_bh);
+        n->tx_bh = qemu_bh_new(virtio_net_tx_bh, n);
+    }
     n->ctrl_vq = virtio_add_queue(&n->vdev, 64, virtio_net_handle_ctrl);
     qemu_macaddr_default_if_unset(&conf->macaddr);
     memcpy(&n->mac[0], &conf->macaddr, sizeof(n->mac));
@@ -939,9 +1006,7 @@ VirtIODevice *virtio_net_init(DeviceState *dev, NICConf *conf,
 
     qemu_format_nic_info_str(&n->nic->nc, conf->macaddr.a);
 
-    n->tx_timer = qemu_new_timer(vm_clock, virtio_net_tx_timer, n);
     n->tx_waiting = 0;
-    n->tx_timeout = net->txtimer;
     n->tx_burst = net->txburst;
     n->mergeable_rx_bufs = 0;
     n->promisc = 1; /* for compatibility */
@@ -974,8 +1039,12 @@ void virtio_net_exit(VirtIODevice *vdev)
     qemu_free(n->mac_table.macs);
     qemu_free(n->vlans);
 
+    if (n->tx_timer) {
         qemu_del_timer(n->tx_timer);
         qemu_free_timer(n->tx_timer);
+    } else {
+        qemu_bh_delete(n->tx_bh);
+    }
 
     virtio_cleanup(&n->vdev);
     qemu_del_vlan_client(&n->nic->nc);

hw/virtio-net.h

@@ -60,6 +60,7 @@ typedef struct virtio_net_conf
 {
     uint32_t txtimer;
     int32_t txburst;
+    char *tx;
 } virtio_net_conf;
 
 /* Maximum packet size we can receive from tap device: header + 64k */

hw/virtio-pci.c

@@ -695,6 +695,7 @@ static PCIDeviceInfo virtio_info[] = {
                                net.txtimer, TX_TIMER_INTERVAL),
             DEFINE_PROP_INT32("x-txburst", VirtIOPCIProxy,
                               net.txburst, TX_BURST),
+            DEFINE_PROP_STRING("tx", VirtIOPCIProxy, net.tx),
             DEFINE_PROP_END_OF_LIST(),
         },
         .qdev.reset = virtio_pci_reset,