在 2021/6/17 下午2:37, Xuan Zhuo 写道:
On Thu, 17 Jun 2021 14:03:29 +0800, Jason Wang <jasowang@xxxxxxxxxx> wrote:
在 2021/6/17 下午1:53, Xuan Zhuo 写道:
On Thu, 17 Jun 2021 11:23:52 +0800, Jason Wang <jasowang@xxxxxxxxxx> wrote:
在 2021/6/10 下午4:22, Xuan Zhuo 写道:
Compared to the case of xsk tx, the case of xsk zc rx is more
complicated.
When we process the buf received by vq, we may encounter ordinary
buffers, or xsk buffers. What makes the situation more complicated is
that in the case of mergeable, when num_buffer > 1, we may still
encounter the case where xsk buffer is mixed with ordinary buffer.
Another thing that makes the situation more complicated is that when we
get an xsk buffer from vq, the xsk bound to this xsk buffer may have
been unbound.
Signed-off-by: Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx>
This is somehow similar to the case of tx where we don't have per vq reset.
[...]
- if (vi->mergeable_rx_bufs)
+ if (is_xsk_ctx(ctx))
+ skb = receive_xsk(dev, vi, rq, buf, len, xdp_xmit, stats);
+ else if (vi->mergeable_rx_bufs)
skb = receive_mergeable(dev, vi, rq, buf, ctx, len, xdp_xmit,
stats);
else if (vi->big_packets)
@@ -1175,6 +1296,14 @@ static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq,
int err;
bool oom;
+ /* Because virtio-net does not yet support flow direct,
Note that this is not the case any more. RSS has been supported by
virtio spec and qemu/vhost/tap now. We just need some work on the
virtio-net driver part (e.g the ethool interface).
Oh, are there any plans? Who is doing this work, can I help?
Qemu and spec has support RSS.
TAP support is ready via steering eBPF program, you can try to play it
with current qemu master.
The only thing missed is the Linux driver, I think Yuri or Andrew is
working on this.
I feel that in the case of xsk, the flow director is more appropriate.
Users may still want to allocate packets to a certain channel based on
information such as port/ip/tcp/udp, and then xsk will process them.
I will try to push the flow director to the spec.
That would be fine. For the backend implementation, it could still be
implemented via steering eBPF.
Thanks
Thanks.
Thanks
Thanks.
Thanks
