Re: [PATCH v11 39/40] virtio_net: support tx queue resize

[Xuan Zhuo <xuanzhuo@linux.alibaba.com>]

On Mon, 18 Jul 2022 16:57:53 +0800, Jason Wang <jasowang@redhat.com> wrote:
> On Fri, Jul 15, 2022 at 4:32 PM Xuan Zhuo <xuanzhuo@linux.alibaba.com> wrote:
> >
> > On Fri, 8 Jul 2022 14:23:57 +0800, Jason Wang <jasowang@redhat.com> wrote:
> > > On Tue, Jul 5, 2022 at 10:01 AM Xuan Zhuo <xuanzhuo@linux.alibaba.com> wrote:
> > > >
> > > > On Mon, 4 Jul 2022 11:45:52 +0800, Jason Wang <jasowang@redhat.com> wrote:
> > > > >
> > > > > 在 2022/6/29 14:56, Xuan Zhuo 写道:
> > > > > > This patch implements the resize function of the tx queues.
> > > > > > Based on this function, it is possible to modify the ring num of the
> > > > > > queue.
> > > > > >
> > > > > > Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
> > > > > > ---
> > > > > >   drivers/net/virtio_net.c | 48 ++++++++++++++++++++++++++++++++++++++++
> > > > > >   1 file changed, 48 insertions(+)
> > > > > >
> > > > > > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> > > > > > index 6ab16fd193e5..fd358462f802 100644
> > > > > > --- a/drivers/net/virtio_net.c
> > > > > > +++ b/drivers/net/virtio_net.c
> > > > > > @@ -135,6 +135,9 @@ struct send_queue {
> > > > > >     struct virtnet_sq_stats stats;
> > > > > >
> > > > > >     struct napi_struct napi;
> > > > > > +
> > > > > > +   /* Record whether sq is in reset state. */
> > > > > > +   bool reset;
> > > > > >   };
> > > > > >
> > > > > >   /* Internal representation of a receive virtqueue */
> > > > > > @@ -279,6 +282,7 @@ struct padded_vnet_hdr {
> > > > > >   };
> > > > > >
> > > > > >   static void virtnet_rq_free_unused_buf(struct virtqueue *vq, void *buf);
> > > > > > +static void virtnet_sq_free_unused_buf(struct virtqueue *vq, void *buf);
> > > > > >
> > > > > >   static bool is_xdp_frame(void *ptr)
> > > > > >   {
> > > > > > @@ -1603,6 +1607,11 @@ static void virtnet_poll_cleantx(struct receive_queue *rq)
> > > > > >             return;
> > > > > >
> > > > > >     if (__netif_tx_trylock(txq)) {
> > > > > > +           if (READ_ONCE(sq->reset)) {
> > > > > > +                   __netif_tx_unlock(txq);
> > > > > > +                   return;
> > > > > > +           }
> > > > > > +
> > > > > >             do {
> > > > > >                     virtqueue_disable_cb(sq->vq);
> > > > > >                     free_old_xmit_skbs(sq, true);
> > > > > > @@ -1868,6 +1877,45 @@ static int virtnet_rx_resize(struct virtnet_info *vi,
> > > > > >     return err;
> > > > > >   }
> > > > > >
> > > > > > +static int virtnet_tx_resize(struct virtnet_info *vi,
> > > > > > +                        struct send_queue *sq, u32 ring_num)
> > > > > > +{
> > > > > > +   struct netdev_queue *txq;
> > > > > > +   int err, qindex;
> > > > > > +
> > > > > > +   qindex = sq - vi->sq;
> > > > > > +
> > > > > > +   virtnet_napi_tx_disable(&sq->napi);
> > > > > > +
> > > > > > +   txq = netdev_get_tx_queue(vi->dev, qindex);
> > > > > > +
> > > > > > +   /* 1. wait all ximt complete
> > > > > > +    * 2. fix the race of netif_stop_subqueue() vs netif_start_subqueue()
> > > > > > +    */
> > > > > > +   __netif_tx_lock_bh(txq);
> > > > > > +
> > > > > > +   /* Prevent rx poll from accessing sq. */
> > > > > > +   WRITE_ONCE(sq->reset, true);
> > > > >
> > > > >
> > > > > Can we simply disable RX NAPI here?
> > > >
> > > > Disable rx napi is indeed a simple solution. But I hope that when dealing with
> > > > tx, it will not affect rx.
> > >
> > > Ok, but I think we've already synchronized with tx lock here, isn't it?
> >
> > Yes, do you have any questions about WRITE_ONCE()? There is a set false operation
> > later, I did not use lock there, so I used WRITE/READ_ONCE
> > uniformly.
>
> I mean, since we've already used tx locks somewhere, we'd better use
> them here as well at least as a start.


OK. next version will fix.

Thanks.

+static int virtnet_tx_resize(struct virtnet_info *vi,
+			     struct send_queue *sq, u32 ring_num)
+{
+	struct netdev_queue *txq;
+	int err, qindex;
+
+	qindex = sq - vi->sq;
+
+	virtnet_napi_tx_disable(&sq->napi);
+
+	txq = netdev_get_tx_queue(vi->dev, qindex);
+
+	/* 1. wait all ximt complete
+	 * 2. fix the race of netif_stop_subqueue() vs netif_start_subqueue()
+	 */
+	__netif_tx_lock_bh(txq);
+
+	sq->reset = true;
+
+	/* Prevent the upper layer from trying to send packets. */
+	netif_stop_subqueue(vi->dev, qindex);
+
+	__netif_tx_unlock_bh(txq);
+
+	err = virtqueue_resize(sq->vq, ring_num, virtnet_sq_free_unused_buf);
+	if (err)
+		netdev_err(vi->dev, "resize tx fail: tx queue index: %d err: %d\n", qindex, err);
+
+	__netif_tx_lock_bh(txq);
+	sq->reset = false;
+	netif_tx_wake_queue(txq);
+	__netif_tx_unlock_bh(txq);
+
+	virtnet_napi_tx_enable(vi, sq->vq, &sq->napi);
+	return err;
+}


>
> Thanks
>
> >
> > Thanks.
> >
> > >
> > > Thanks
> > >
> > > >
> > > > Thanks.
> > > >
> > > >
> > > > >
> > > > > Thanks
> > > > >
> > > > >
> > > > > > +
> > > > > > +   /* Prevent the upper layer from trying to send packets. */
> > > > > > +   netif_stop_subqueue(vi->dev, qindex);
> > > > > > +
> > > > > > +   __netif_tx_unlock_bh(txq);
> > > > > > +
> > > > > > +   err = virtqueue_resize(sq->vq, ring_num, virtnet_sq_free_unused_buf);
> > > > > > +   if (err)
> > > > > > +           netdev_err(vi->dev, "resize tx fail: tx queue index: %d err: %d\n", qindex, err);
> > > > > > +
> > > > > > +   /* Memory barrier before set reset and start subqueue. */
> > > > > > +   smp_mb();
> > > > > > +
> > > > > > +   WRITE_ONCE(sq->reset, false);
> > > > > > +   netif_tx_wake_queue(txq);
> > > > > > +
> > > > > > +   virtnet_napi_tx_enable(vi, sq->vq, &sq->napi);
> > > > > > +   return err;
> > > > > > +}
> > > > > > +
> > > > > >   /*
> > > > > >    * Send command via the control virtqueue and check status.  Commands
> > > > > >    * supported by the hypervisor, as indicated by feature bits, should
> > > > >
> > > >
> > >
> >
>