Re: [PATCH RFC v2 3/4] virtio_net: move tx vq operation under tx queue lock

On Tue, Apr 13, 2021 at 10:03 AM Michael S. Tsirkin [off-list ref] wrote:

On Tue, Apr 13, 2021 at 04:54:42PM +0800, Jason Wang wrote:

quoted

在 2021/4/13 下午1:47, Michael S. Tsirkin 写道:

quoted

It's unsafe to operate a vq from multiple threads.
Unfortunately this is exactly what we do when invoking
clean tx poll from rx napi.

Actually, the issue goes back to the napi-tx even without the
opportunistic cleaning from the receive interrupt, I think? That races
with processing the vq in start_xmit.

quoted

quoted

As a fix move everything that deals with the vq to under tx lock.

If the above is correct:

Fixes: b92f1e6751a6 ("virtio-net: transmit napi")

quoted

quoted

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
---
  drivers/net/virtio_net.c | 22 +++++++++++++++++++++-
  1 file changed, 21 insertions(+), 1 deletion(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 16d5abed582c..460ccdbb840e 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c

@@ -1505,6 +1505,8 @@ static int virtnet_poll_tx(struct napi_struct *napi, int budget)
    struct virtnet_info *vi = sq->vq->vdev->priv;
    unsigned int index = vq2txq(sq->vq);
    struct netdev_queue *txq;
+   int opaque;

nit: virtqueue_napi_complete also stores as int opaque, but
virtqueue_enable_cb_prepare actually returns, and virtqueue_poll
expects, an unsigned int. In the end, conversion works correctly. But
cleaner to use the real type.

quoted

quoted

+   bool done;
    if (unlikely(is_xdp_raw_buffer_queue(vi, index))) {
            /* We don't need to enable cb for XDP */

@@ -1514,10 +1516,28 @@ static int virtnet_poll_tx(struct napi_struct *napi, int budget)
    txq = netdev_get_tx_queue(vi->dev, index);
    __netif_tx_lock(txq, raw_smp_processor_id());
+   virtqueue_disable_cb(sq->vq);
    free_old_xmit_skbs(sq, true);
+
+   opaque = virtqueue_enable_cb_prepare(sq->vq);
+
+   done = napi_complete_done(napi, 0);
+
+   if (!done)
+           virtqueue_disable_cb(sq->vq);
+
    __netif_tx_unlock(txq);
-   virtqueue_napi_complete(napi, sq->vq, 0);

So I wonder why not simply move __netif_tx_unlock() after
virtqueue_napi_complete()?

Thanks

Because that calls tx poll which also takes tx lock internally ...

which tx poll?