Re: [PATCH RFC 1/2] virtio-net: bql support

[Jason Wang]

This is a multi-part message in MIME format.
On 2018/12/6 äå4:17, Jason Wang wrote:
> On 2018/12/6 äå6:54, Michael S. Tsirkin wrote:
> > When use_napi is set, let's enable BQLs. Note: some of the issues are
> > similar to wifi. It's worth considering whether something similar to
> > commit 36148c2bbfbe ("mac80211: Adjust TSQ pacing shift") might be
> > benefitial.
>
>
> I've played a similar patch several days before. The tricky part is 
> the mode switching between napi and no napi. We should make sure when 
> the packet is sent and trakced by BQL, it should be consumed by BQL 
> as well. I did it by tracking it through skb->cb. And deal with the 
> freeze by reset the BQL status. Patch attached.


Add the patch.

Thanks


> But when testing with vhost-net, I don't very a stable performance, it 
> was probably because we batch the used ring updating so tx interrupt 
> may come randomly. We probably need to implement time bounded 
> coalescing mechanism which could be configured from userspace.
>
> Btw, maybe it's time just enable napi TX by default. I get ~10% TCP_RR 
> regression on machine without APICv, (haven't found time to test APICv 
> machine). But consider it was for correctness, I think it's 
> acceptable? Then we can do optimization on top?
>
>
> Thanks
>
>
> > Signed-off-by: Michael S. Tsirkin <mst@xxxxxxxxxx>
> > ---
> > Â drivers/net/virtio_net.c | 27 +++++++++++++++++++--------
> > Â 1 file changed, 19 insertions(+), 8 deletions(-)
> >
> > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> > index cecfd77c9f3c..b657bde6b94b 100644
> > --- a/drivers/net/virtio_net.c
> > +++ b/drivers/net/virtio_net.c
> > @@ -1325,7 +1325,8 @@ static int virtnet_receive(struct receive_queue 
> > *rq, int budget,
> > ÂÂÂÂÂ return stats.packets;
> > Â }
> > Â -static void free_old_xmit_skbs(struct send_queue *sq)
> > +static void free_old_xmit_skbs(struct send_queue *sq, struct 
> > netdev_queue *txq,
> > +ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ bool use_napi)
> > Â {
> > ÂÂÂÂÂ struct sk_buff *skb;
> > ÂÂÂÂÂ unsigned int len;
> > @@ -1347,6 +1348,9 @@ static void free_old_xmit_skbs(struct 
> > send_queue *sq)
> > ÂÂÂÂÂ if (!packets)
> > ÂÂÂÂÂÂÂÂÂ return;
> > Â +ÂÂÂ if (use_napi)
> > +ÂÂÂÂÂÂÂ netdev_tx_completed_queue(txq, packets, bytes);
> > +
> > ÂÂÂÂÂ u64_stats_update_begin(&sq->stats.syncp);
> > ÂÂÂÂÂ sq->stats.bytes += bytes;
> > ÂÂÂÂÂ sq->stats.packets += packets;
> > @@ -1364,7 +1368,7 @@ static void virtnet_poll_cleantx(struct 
> > receive_queue *rq)
> > ÂÂÂÂÂÂÂÂÂ return;
> > Â ÂÂÂÂÂ if (__netif_tx_trylock(txq)) {
> > -ÂÂÂÂÂÂÂ free_old_xmit_skbs(sq);
> > +ÂÂÂÂÂÂÂ free_old_xmit_skbs(sq, txq, true);
> > ÂÂÂÂÂÂÂÂÂ __netif_tx_unlock(txq);
> > ÂÂÂÂÂ }
> > Â @@ -1440,7 +1444,7 @@ static int virtnet_poll_tx(struct napi_struct 
> > *napi, int budget)
> > ÂÂÂÂÂ struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, 
> > vq2txq(sq->vq));
> > Â ÂÂÂÂÂ __netif_tx_lock(txq, raw_smp_processor_id());
> > -ÂÂÂ free_old_xmit_skbs(sq);
> > +ÂÂÂ free_old_xmit_skbs(sq, txq, true);
> > ÂÂÂÂÂ __netif_tx_unlock(txq);
> > Â ÂÂÂÂÂ virtqueue_napi_complete(napi, sq->vq, 0);
> > @@ -1505,13 +1509,15 @@ static netdev_tx_t start_xmit(struct sk_buff 
> > *skb, struct net_device *dev)
> > ÂÂÂÂÂ struct send_queue *sq = &vi->sq[qnum];
> > ÂÂÂÂÂ int err;
> > ÂÂÂÂÂ struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum);
> > -ÂÂÂ bool kick = !skb->xmit_more;
> > +ÂÂÂ bool more = skb->xmit_more;
> > ÂÂÂÂÂ bool use_napi = sq->napi.weight;
> > +ÂÂÂ unsigned int bytes = skb->len;
> > +ÂÂÂ bool kick;
> > Â ÂÂÂÂÂ /* Free up any pending old buffers before queueing new ones. */
> > -ÂÂÂ free_old_xmit_skbs(sq);
> > +ÂÂÂ free_old_xmit_skbs(sq, txq, use_napi);
> > Â -ÂÂÂ if (use_napi && kick)
> > +ÂÂÂ if (use_napi && !more)
> > ÂÂÂÂÂÂÂÂÂ virtqueue_enable_cb_delayed(sq->vq);
> > Â ÂÂÂÂÂ /* timestamp packet in software */
> > @@ -1552,7 +1558,7 @@ static netdev_tx_t start_xmit(struct sk_buff 
> > *skb, struct net_device *dev)
> > ÂÂÂÂÂÂÂÂÂ if (!use_napi &&
> > ÂÂÂÂÂÂÂÂÂÂÂÂÂ unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
> > ÂÂÂÂÂÂÂÂÂÂÂÂÂ /* More just got used, free them then recheck. */
> > -ÂÂÂÂÂÂÂÂÂÂÂ free_old_xmit_skbs(sq);
> > +ÂÂÂÂÂÂÂÂÂÂÂ free_old_xmit_skbs(sq, txq, false);
> > ÂÂÂÂÂÂÂÂÂÂÂÂÂ if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
> > ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ netif_start_subqueue(dev, qnum);
> > ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ virtqueue_disable_cb(sq->vq);
> > @@ -1560,7 +1566,12 @@ static netdev_tx_t start_xmit(struct sk_buff 
> > *skb, struct net_device *dev)
> > ÂÂÂÂÂÂÂÂÂ }
> > ÂÂÂÂÂ }
> > Â -ÂÂÂ if (kick || netif_xmit_stopped(txq)) {
> > +ÂÂÂ if (use_napi)
> > +ÂÂÂÂÂÂÂ kick = __netdev_tx_sent_queue(txq, bytes, more);
> > +ÂÂÂ else
> > +ÂÂÂÂÂÂÂ kick = !more || netif_xmit_stopped(txq);
> > +
> > +ÂÂÂ if (kick) {
> > ÂÂÂÂÂÂÂÂÂ if (virtqueue_kick_prepare(sq->vq) && 
> > virtqueue_notify(sq->vq)) {
> > ÂÂÂÂÂÂÂÂÂÂÂÂÂ u64_stats_update_begin(&sq->stats.syncp);
> > ÂÂÂÂÂÂÂÂÂÂÂÂÂ sq->stats.kicks++;
> _______________________________________________
> Virtualization mailing list
> Virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx
> https://lists.linuxfoundation.org/mailman/listinfo/virtualization