On Wed, Oct 15, 2014 at 03:25:29PM +0800, Jason Wang wrote:
> Orphan skb in ndo_start_xmit() breaks socket accounting and packet
> queuing. This in fact breaks lots of things such as pktgen and several
> TCP optimizations. And also make BQL can't be implemented for
> virtio-net.
>
> This patch tries to solve this issue by enabling tx interrupt. To
> avoid introducing extra spinlocks, a tx napi was scheduled to free
> those packets.
>
> More tx interrupt mitigation method could be used on top.
>
> Cc: Rusty Russell <rusty@xxxxxxxxxxxxxxx>
> Cc: Michael S. Tsirkin <mst@xxxxxxxxxx>
> Signed-off-by: Jason Wang <jasowang@xxxxxxxxxx>
> ---
> drivers/net/virtio_net.c | 125 +++++++++++++++++++++++++++++++---------------
> 1 files changed, 85 insertions(+), 40 deletions(-)
>
> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> index ccf98f9..2afc2e2 100644
> --- a/drivers/net/virtio_net.c
> +++ b/drivers/net/virtio_net.c
> @@ -72,6 +72,8 @@ struct send_queue {
>
> /* Name of the send queue: output.$index */
> char name[40];
> +
> + struct napi_struct napi;
> };
>
> /* Internal representation of a receive virtqueue */
> @@ -217,15 +219,40 @@ static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
> return p;
> }
>
> +static int free_old_xmit_skbs(struct send_queue *sq, int budget)
> +{
> + struct sk_buff *skb;
> + unsigned int len;
> + struct virtnet_info *vi = sq->vq->vdev->priv;
> + struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
> + u64 tx_bytes = 0, tx_packets = 0;
> +
> + while (tx_packets < budget &&
> + (skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
> + pr_debug("Sent skb %p\n", skb);
> +
> + tx_bytes += skb->len;
> + tx_packets++;
> +
> + dev_kfree_skb_any(skb);
> + }
> +
> + u64_stats_update_begin(&stats->tx_syncp);
> + stats->tx_bytes += tx_bytes;
> + stats->tx_packets =+ tx_packets;
> + u64_stats_update_end(&stats->tx_syncp);
> +
> + return tx_packets;
> +}
> +
> static void skb_xmit_done(struct virtqueue *vq)
> {
> struct virtnet_info *vi = vq->vdev->priv;
> + struct send_queue *sq = &vi->sq[vq2txq(vq)];
>
> - /* Suppress further interrupts. */
> - virtqueue_disable_cb(vq);
> -
> - /* We were probably waiting for more output buffers. */
> - netif_wake_subqueue(vi->dev, vq2txq(vq));
> + if (napi_schedule_prep(&sq->napi)) {
> + __napi_schedule(&sq->napi);
> + }
> }
>
> static unsigned int mergeable_ctx_to_buf_truesize(unsigned long mrg_ctx)
> @@ -774,7 +801,39 @@ again:
> return received;
> }
>
> +static int virtnet_poll_tx(struct napi_struct *napi, int budget)
> +{
> + struct send_queue *sq =
> + container_of(napi, struct send_queue, napi);
> + struct virtnet_info *vi = sq->vq->vdev->priv;
> + struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, vq2txq(sq->vq));
> + unsigned int r, sent = 0;
> +
> +again:
> + __netif_tx_lock(txq, smp_processor_id());
> + virtqueue_disable_cb(sq->vq);
> + sent += free_old_xmit_skbs(sq, budget - sent);
> +
> + if (sent < budget) {
> + r = virtqueue_enable_cb_prepare(sq->vq);
> + napi_complete(napi);
> + __netif_tx_unlock(txq);
> + if (unlikely(virtqueue_poll(sq->vq, r)) &&
So you are enabling callback on the next packet,
which is almost sure to cause an interrupt storm
on the guest.
I think it's a bad idea, this is why I used
enable_cb_delayed in my patch.
> + napi_schedule_prep(napi)) {
> + virtqueue_disable_cb(sq->vq);
> + __napi_schedule(napi);
> + goto again;
> + }
> + } else {
> + __netif_tx_unlock(txq);
> + }
> +
> + netif_wake_subqueue(vi->dev, vq2txq(sq->vq));
> + return sent;
> +}
> +
> #ifdef CONFIG_NET_RX_BUSY_POLL
> +
> /* must be called with local_bh_disable()d */
> static int virtnet_busy_poll(struct napi_struct *napi)
> {
> @@ -822,36 +881,12 @@ static int virtnet_open(struct net_device *dev)
> if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
> schedule_delayed_work(&vi->refill, 0);
> virtnet_napi_enable(&vi->rq[i]);
> + napi_enable(&vi->sq[i].napi);
> }
>
> return 0;
> }
>
> -static int free_old_xmit_skbs(struct send_queue *sq)
> -{
> - struct sk_buff *skb;
> - unsigned int len;
> - struct virtnet_info *vi = sq->vq->vdev->priv;
> - struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
> - u64 tx_bytes = 0, tx_packets = 0;
> -
> - while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
> - pr_debug("Sent skb %p\n", skb);
> -
> - tx_bytes += skb->len;
> - tx_packets++;
> -
> - dev_kfree_skb_any(skb);
> - }
> -
> - u64_stats_update_begin(&stats->tx_syncp);
> - stats->tx_bytes += tx_bytes;
> - stats->tx_packets =+ tx_packets;
> - u64_stats_update_end(&stats->tx_syncp);
> -
> - return tx_packets;
> -}
> -
So you end up moving it all anyway, why bother splitting out
minor changes in previous patches?
> static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
> {
> struct skb_vnet_hdr *hdr;
> @@ -917,6 +952,7 @@ static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
> sg_set_buf(sq->sg, hdr, hdr_len);
> num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1;
> }
> +
> return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
> }
>
> @@ -925,10 +961,11 @@ static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
> struct virtnet_info *vi = netdev_priv(dev);
> int qnum = skb_get_queue_mapping(skb);
> struct send_queue *sq = &vi->sq[qnum];
> - int err;
> + int err, qsize = virtqueue_get_vring_size(sq->vq);
>
> + virtqueue_disable_cb(sq->vq);
> /* Free up any pending old buffers before queueing new ones. */
> - free_old_xmit_skbs(sq);
> + free_old_xmit_skbs(sq, qsize);
>
> /* Try to transmit */
> err = xmit_skb(sq, skb);
> @@ -944,22 +981,20 @@ static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
> return NETDEV_TX_OK;
> }
>
> - /* Don't wait up for transmitted skbs to be freed. */
> - skb_orphan(skb);
> - nf_reset(skb);
> -
> /* Apparently nice girls don't return TX_BUSY; stop the queue
> * before it gets out of hand. Naturally, this wastes entries. */
> if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
> netif_stop_subqueue(dev, qnum);
> if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
> /* More just got used, free them then recheck. */
> - free_old_xmit_skbs(sq);
> + free_old_xmit_skbs(sq, qsize);
> if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
> netif_start_subqueue(dev, qnum);
> virtqueue_disable_cb(sq->vq);
> }
> }
> + } else if (virtqueue_enable_cb(sq->vq)) {
> + free_old_xmit_skbs(sq, qsize);
> }
>
> if (__netif_subqueue_stopped(dev, qnum) || !skb->xmit_more)
> @@ -1141,8 +1176,10 @@ static int virtnet_close(struct net_device *dev)
> /* Make sure refill_work doesn't re-enable napi! */
> cancel_delayed_work_sync(&vi->refill);
>
> - for (i = 0; i < vi->max_queue_pairs; i++)
> + for (i = 0; i < vi->max_queue_pairs; i++) {
> napi_disable(&vi->rq[i].napi);
> + napi_disable(&vi->sq[i].napi);
> + }
>
> return 0;
> }
> @@ -1461,8 +1498,10 @@ static void virtnet_free_queues(struct virtnet_info *vi)
> {
> int i;
>
> - for (i = 0; i < vi->max_queue_pairs; i++)
> + for (i = 0; i < vi->max_queue_pairs; i++) {
> netif_napi_del(&vi->rq[i].napi);
> + netif_napi_del(&vi->sq[i].napi);
> + }
>
> kfree(vi->rq);
> kfree(vi->sq);
> @@ -1616,6 +1655,8 @@ static int virtnet_alloc_queues(struct virtnet_info *vi)
> netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
> napi_weight);
> napi_hash_add(&vi->rq[i].napi);
> + netif_napi_add(vi->dev, &vi->sq[i].napi, virtnet_poll_tx,
> + napi_weight);
>
> sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
> ewma_init(&vi->rq[i].mrg_avg_pkt_len, 1, RECEIVE_AVG_WEIGHT);
> @@ -1920,8 +1961,10 @@ static int virtnet_freeze(struct virtio_device *vdev)
> if (netif_running(vi->dev)) {
> for (i = 0; i < vi->max_queue_pairs; i++) {
> napi_disable(&vi->rq[i].napi);
> + napi_disable(&vi->sq[i].napi);
> napi_hash_del(&vi->rq[i].napi);
> netif_napi_del(&vi->rq[i].napi);
> + netif_napi_del(&vi->sq[i].napi);
> }
> }
>
> @@ -1946,8 +1989,10 @@ static int virtnet_restore(struct virtio_device *vdev)
> if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
> schedule_delayed_work(&vi->refill, 0);
>
> - for (i = 0; i < vi->max_queue_pairs; i++)
> + for (i = 0; i < vi->max_queue_pairs; i++) {
> virtnet_napi_enable(&vi->rq[i]);
> + napi_enable(&vi->sq[i].napi);
> + }
> }
>
> netif_device_attach(vi->dev);
> --
> 1.7.1
_______________________________________________
Virtualization mailing list
Virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
