virtio-net: batch stats updating
authorJason Wang <jasowang@redhat.com>
Fri, 17 Feb 2017 03:33:09 +0000 (11:33 +0800)
committerDavid S. Miller <davem@davemloft.net>
Sun, 19 Feb 2017 23:15:23 +0000 (18:15 -0500)
We already have counters for sent/recv packets and sent/recv bytes.
Doing a batched update to reduce the number of
u64_stats_update_begin/end().

Take care not to bother with stats update when called
speculatively.

Cc: Willem de Bruijn <willemb@google.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/virtio_net.c

index 11e28530c83c9fa6162725419f4c219ea82b7136..05a83dbc910d04387ae90ca9a0da299d59967810 100644 (file)
@@ -719,13 +719,13 @@ xdp_xmit:
        return NULL;
 }
 
-static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
-                       void *buf, unsigned int len)
+static int receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
+                      void *buf, unsigned int len)
 {
        struct net_device *dev = vi->dev;
-       struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
        struct sk_buff *skb;
        struct virtio_net_hdr_mrg_rxbuf *hdr;
+       int ret;
 
        if (unlikely(len < vi->hdr_len + ETH_HLEN)) {
                pr_debug("%s: short packet %i\n", dev->name, len);
@@ -739,7 +739,7 @@ static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
                } else {
                        dev_kfree_skb(buf);
                }
-               return;
+               return 0;
        }
 
        if (vi->mergeable_rx_bufs)
@@ -750,14 +750,11 @@ static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
                skb = receive_small(dev, vi, rq, buf, len);
 
        if (unlikely(!skb))
-               return;
+               return 0;
 
        hdr = skb_vnet_hdr(skb);
 
-       u64_stats_update_begin(&stats->rx_syncp);
-       stats->rx_bytes += skb->len;
-       stats->rx_packets++;
-       u64_stats_update_end(&stats->rx_syncp);
+       ret = skb->len;
 
        if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID)
                skb->ip_summed = CHECKSUM_UNNECESSARY;
@@ -775,11 +772,12 @@ static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
                 ntohs(skb->protocol), skb->len, skb->pkt_type);
 
        napi_gro_receive(&rq->napi, skb);
-       return;
+       return ret;
 
 frame_err:
        dev->stats.rx_frame_errors++;
        dev_kfree_skb(skb);
+       return 0;
 }
 
 static unsigned int virtnet_get_headroom(struct virtnet_info *vi)
@@ -994,12 +992,13 @@ static void refill_work(struct work_struct *work)
 static int virtnet_receive(struct receive_queue *rq, int budget)
 {
        struct virtnet_info *vi = rq->vq->vdev->priv;
-       unsigned int len, received = 0;
+       unsigned int len, received = 0, bytes = 0;
        void *buf;
+       struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
 
        while (received < budget &&
               (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
-               receive_buf(vi, rq, buf, len);
+               bytes += receive_buf(vi, rq, buf, len);
                received++;
        }
 
@@ -1008,6 +1007,11 @@ static int virtnet_receive(struct receive_queue *rq, int budget)
                        schedule_delayed_work(&vi->refill, 0);
        }
 
+       u64_stats_update_begin(&stats->rx_syncp);
+       stats->rx_bytes += bytes;
+       stats->rx_packets += received;
+       u64_stats_update_end(&stats->rx_syncp);
+
        return received;
 }
 
@@ -1056,17 +1060,28 @@ static void free_old_xmit_skbs(struct send_queue *sq)
        unsigned int len;
        struct virtnet_info *vi = sq->vq->vdev->priv;
        struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
+       unsigned int packets = 0;
+       unsigned int bytes = 0;
 
        while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
                pr_debug("Sent skb %p\n", skb);
 
-               u64_stats_update_begin(&stats->tx_syncp);
-               stats->tx_bytes += skb->len;
-               stats->tx_packets++;
-               u64_stats_update_end(&stats->tx_syncp);
+               bytes += skb->len;
+               packets++;
 
                dev_kfree_skb_any(skb);
        }
+
+       /* Avoid overhead when no packets have been processed
+        * happens when called speculatively from start_xmit.
+        */
+       if (!packets)
+               return;
+
+       u64_stats_update_begin(&stats->tx_syncp);
+       stats->tx_bytes += bytes;
+       stats->tx_packets += packets;
+       u64_stats_update_end(&stats->tx_syncp);
 }
 
 static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)