Re: [PATCH] virtio-net: fix qemu set used ring flag even vhost started

[Michael Tokarev]

02.04.2024 07:51, Yajun Wu:
> When vhost-user or vhost-kernel is handling virtio net datapath, qemu
> should not touch used ring.
>
> But with vhost-user socket reconnect scenario, in a very rare case (has
> pending kick event). VRING_USED_F_NO_NOTIFY is set by qemu in
> following code path:
...
> This issue causes guest kernel stop kicking device and traffic stop.
>
> Add vhost_started check in virtio_net_handle_tx_bh to fix this wrong
> VRING_USED_F_NO_NOTIFY set.

This seems to be qemu-stable material, even if the case when the issue
happens is "very rare", -- is it not?

Thanks,

/mjt

> Signed-off-by: Yajun Wu <yajunw@nvidia.com>
> Reviewed-by: Jiri Pirko <jiri@nvidia.com>
> ---
>   hw/net/virtio-net.c | 4 ++++
>   1 file changed, 4 insertions(+)
>
> diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c
> index a6ff000cd9..8035e01fdf 100644
> --- a/hw/net/virtio-net.c
> +++ b/hw/net/virtio-net.c
> @@ -2865,6 +2865,10 @@ static void virtio_net_handle_tx_bh(VirtIODevice *vdev, 
> VirtQueue *vq)
>       VirtIONet *n = VIRTIO_NET(vdev);
>       VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
>   
> +    if (n->vhost_started) {
> +        return;
> +    }
> +
>       if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) {
>           virtio_net_drop_tx_queue_data(vdev, vq);
>           return;