Every virtqueue_add must eventually be followed by virtqueue_kick for to
properly notify the peer that new buffers have been put into the queue.
This is currently missing for virtio-net and may result in non-working
network when the host has depleted the rx buffers and waits for new
buffers. Depending on the host it may busy poll on the virtio queue or
wait for the kick. Qemu does the latter and may break.
Signed-off-by: Christian Speich <c.spe...@avm.de>
---
drivers/virtio/virtio_net.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/drivers/virtio/virtio_net.c b/drivers/virtio/virtio_net.c
index 0e5367a085e..71e4d251396 100644
--- a/drivers/virtio/virtio_net.c
+++ b/drivers/virtio/virtio_net.c
@@ -128,6 +128,7 @@ static int virtio_net_free_pkt(struct udevice *dev, uchar
*packet, int length)
/* Put the buffer back to the rx ring */
virtqueue_add(priv->rx_vq, sgs, 0, 1);
+ virtqueue_kick(priv->rx_vq);
return 0;
}
--
2.43.0
base-commit: 3532f1f5edfc97c9dcea723cdeb732eda44bc669
branch: virtio-net-kick
