Hey Rusty et al,
I've got automatic backports of the virtio modules[1] working back to
2.6.18. Everything seems okay except that for any kernel with the older
NAPI api, performance is extremely bad. I get about 1gbit on TX with
2.6.24 but I get somewhere around 20mbit on 2.6.22.
I had to refactor the NAPI poll function heavily to get it working with
the old api so my suspicion is that I'm doing something wrong. I'd
appreciate if ya'll could take a look at it. I've included a patch that
only has the differences of the backported version and the latest
version. Pay careful attention to the COMPAT_napi blocks as these are
what would be used on older kernels.
It's a bit ugly because it's automatically generated by an awk script.
[1] http://hg.codemonkey.ws/virtio-ext-modules
Thanks,
Anthony Liguori
--- /home/anthony/git/kvm/drivers/net/virtio_net.c 2008-01-04 12:06:27.000000000 -0600
+++ virtio_net.c 2008-01-04 16:02:17.000000000 -0600
@@ -38,7 +38,9 @@
struct virtio_device *vdev;
struct virtqueue *rvq, *svq;
struct net_device *dev;
+#ifndef COMPAT_napi
struct napi_struct napi;
+#endif
struct tasklet_struct xmit_free;
@@ -74,8 +76,12 @@
while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
pr_debug("Sent skb %p\n", skb);
__skb_unlink(skb, &vi->send);
+#ifndef COMPAT_net_stats
vi->dev->stats.tx_bytes += len;
+#endif
+#ifndef COMPAT_net_stats
vi->dev->stats.tx_packets++;
+#endif
kfree_skb(skb);
i++;
}
@@ -108,7 +114,9 @@
if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
pr_debug("%s: short packet %i\n", dev->name, len);
+#ifndef COMPAT_net_stats
dev->stats.rx_length_errors++;
+#endif
goto drop;
}
len -= sizeof(struct virtio_net_hdr);
@@ -118,9 +126,14 @@
skb->protocol = eth_type_trans(skb, dev);
pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
ntohs(skb->protocol), skb->len, skb->pkt_type);
+#ifndef COMPAT_net_stats
dev->stats.rx_bytes += skb->len;
+#endif
+#ifndef COMPAT_net_stats
dev->stats.rx_packets++;
+#endif
+#ifndef COMPAT_csum_offset
if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
pr_debug("Needs csum!\n");
skb->ip_summed = CHECKSUM_PARTIAL;
@@ -135,6 +148,7 @@
goto frame_err;
}
}
+#endif
if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
pr_debug("GSO!\n");
@@ -175,7 +189,9 @@
return;
frame_err:
+#ifndef COMPAT_net_stats
dev->stats.rx_frame_errors++;
+#endif
drop:
dev_kfree_skb(skb);
}
@@ -215,17 +231,40 @@
struct virtnet_info *vi = rvq->vdev->priv;
/* Suppress further interrupts. */
rvq->vq_ops->disable_cb(rvq);
+#ifdef COMPAT_napi
+ vi->rvq->vq_ops->enable_cb(vi->rvq);
+ if (netif_rx_schedule_prep(vi->dev)) {
+ vi->rvq->vq_ops->disable_cb(vi->rvq);
+ __netif_rx_schedule(vi->dev);
+ } else
+ vi->rvq->vq_ops->enable_cb(vi->rvq);
+#else
netif_rx_schedule(vi->dev, &vi->napi);
+#endif
}
+#ifdef COMPAT_napi
+static int virtnet_poll(struct net_device *dev, int *budget)
+#else
static int virtnet_poll(struct napi_struct *napi, int budget)
+#endif
{
+#ifdef COMPAT_napi
+ struct virtnet_info *vi = netdev_priv(dev);
+ int max_received = min(dev->quota, *budget);
+ bool no_work;
+#else
struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
+#endif
struct sk_buff *skb = NULL;
unsigned int len, received = 0;
again:
+#ifdef COMPAT_napi
+ while (received < max_received &&
+#else
while (received < budget &&
+#endif
(skb = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
__skb_unlink(skb, &vi->recv);
receive_skb(vi->dev, skb, len);
@@ -239,6 +278,27 @@
try_fill_recv(vi);
/* Out of packets? */
+#ifdef COMPAT_napi
+ if (skb) {
+ *budget -= received;
+ dev->quota -= received;
+ return 1;
+ }
+
+ no_work = vi->rvq->vq_ops->enable_cb(vi->rvq);
+ netif_rx_complete(vi->dev);
+
+ if (!no_work && netif_rx_reschedule(vi->dev, received)) {
+ vi->rvq->vq_ops->disable_cb(vi->rvq);
+ skb = NULL;
+ goto again;
+ }
+
+ dev->quota -= received;
+ *budget -= received;
+
+ return 0;
+#else
if (received < budget) {
netif_rx_complete(vi->dev, napi);
if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq))
@@ -247,9 +307,14 @@
}
return received;
+#endif
}
+#ifdef COMPAT_hrtimer_func
+static int kick_xmit(struct hrtimer *t)
+#else
static enum hrtimer_restart kick_xmit(struct hrtimer *t)
+#endif
{
struct virtnet_info *vi = container_of(t,struct virtnet_info,tx_timer);
@@ -278,6 +343,7 @@
/* Encode metadata header at front. */
hdr = skb_vnet_hdr(skb);
+#ifndef COMPAT_csum_offset
if (skb->ip_summed == CHECKSUM_PARTIAL) {
hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
hdr->csum_start = skb->csum_start - skb_headroom(skb);
@@ -286,6 +352,7 @@
hdr->flags = 0;
hdr->csum_offset = hdr->csum_start = 0;
}
+#endif
if (skb_is_gso(skb)) {
hdr->gso_size = skb_shinfo(skb)->gso_size;
@@ -356,13 +423,26 @@
if (vi->num == 0)
return -ENOMEM;
+#ifdef COMPAT_napi
+ netif_poll_enable(dev);
+#else
napi_enable(&vi->napi);
+#endif
/* If all buffers were filled by other side before we napi_enabled, we
* won't get another interrupt, so process any outstanding packets
* now. virtnet_poll wants re-enable the queue, so we disable here. */
vi->rvq->vq_ops->disable_cb(vi->rvq);
+#ifdef COMPAT_napi
+ vi->rvq->vq_ops->enable_cb(vi->rvq);
+ if (netif_rx_schedule_prep(vi->dev)) {
+ vi->rvq->vq_ops->disable_cb(vi->rvq);
+ __netif_rx_schedule(vi->dev);
+ } else
+ vi->rvq->vq_ops->enable_cb(vi->rvq);
+#else
netif_rx_schedule(vi->dev, &vi->napi);
+#endif
return 0;
}
@@ -372,7 +452,11 @@
struct virtnet_info *vi = netdev_priv(dev);
struct sk_buff *skb;
+#ifdef COMPAT_napi
+ netif_poll_disable(dev);
+#else
napi_disable(&vi->napi);
+#endif
/* networking core has neutered skb_xmit_done/skb_recv_done, so don't
* worry about races vs. get(). */
@@ -431,7 +515,12 @@
/* Set up our device-specific information */
vi = netdev_priv(dev);
+#ifdef COMPAT_napi
+ dev->poll = virtnet_poll;
+ dev->weight = 64;
+#else
netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
+#endif
vi->dev = dev;
vi->vdev = vdev;
memset(&vi->tx_timer, 0, sizeof(vi->tx_timer));
-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2005.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
kvm-devel mailing list
kvm-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/kvm-devel
