On Mon, 18 Jul 2022 16:57:53 +0800, Jason Wang [off-list ref] wrote:

On Fri, Jul 15, 2022 at 4:32 PM Xuan Zhuo [off-list ref] wrote:

quoted

On Fri, 8 Jul 2022 14:23:57 +0800, Jason Wang [off-list ref] wrote:

quoted

On Tue, Jul 5, 2022 at 10:01 AM Xuan Zhuo [off-list ref] wrote:

quoted

On Mon, 4 Jul 2022 11:45:52 +0800, Jason Wang [off-list ref] wrote:

quoted

在 2022/6/29 14:56, Xuan Zhuo 写道:

quoted

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

quoted

Thanks.

quoted

Thanks

quoted

Thanks.

quoted

Thanks

quoted

+
+   /* 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