[PATCH net-next v2 2/3] vhost_net: Call peek_len when using xdp

[PATCH net-next v2 2/3] vhost_net: Call peek_len when using xdp

[Yunjian Wang]

If TUN supports AF_XDP TX zero-copy, the XDP program will enqueue
packets to the XDP ring and wake up the vhost worker. This requires
the vhost worker to call peek_len(), which can be used to consume
XDP descriptors.

Signed-off-by: Yunjian Wang <wangyunjian@huawei.com>
---
 drivers/vhost/net.c | 17 ++++++++++++-----
 1 file changed, 12 insertions(+), 5 deletions(-)

diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c
index f2ed7167c848..077e74421558 100644
--- a/drivers/vhost/net.c
+++ b/drivers/vhost/net.c
@@ -207,6 +207,11 @@ static int vhost_net_buf_peek_len(void *ptr)
 	return __skb_array_len_with_tag(ptr);
 }
 
+static bool vhost_sock_xdp(struct socket *sock)
+{
+	return sock_flag(sock->sk, SOCK_XDP);
+}
+
 static int vhost_net_buf_peek(struct vhost_net_virtqueue *nvq)
 {
 	struct vhost_net_buf *rxq = &nvq->rxq;
@@ -214,6 +219,13 @@ static int vhost_net_buf_peek(struct vhost_net_virtqueue *nvq)
 	if (!vhost_net_buf_is_empty(rxq))
 		goto out;
 
+	if (ptr_ring_empty(nvq->rx_ring)) {
+		struct socket *sock = vhost_vq_get_backend(&nvq->vq);
+		/* Call peek_len to consume XSK descriptors, when using xdp */
+		if (vhost_sock_xdp(sock) && sock->ops->peek_len)
+			sock->ops->peek_len(sock);
+	}
+
 	if (!vhost_net_buf_produce(nvq))
 		return 0;
 
@@ -346,11 +358,6 @@ static bool vhost_sock_zcopy(struct socket *sock)
 		sock_flag(sock->sk, SOCK_ZEROCOPY);
 }
 
-static bool vhost_sock_xdp(struct socket *sock)
-{
-	return sock_flag(sock->sk, SOCK_XDP);
-}
-
 /* In case of DMA done not in order in lower device driver for some reason.
  * upend_idx is used to track end of used idx, done_idx is used to track head
  * of used idx. Once lower device DMA done contiguously, we will signal KVM
-- 
2.41.0

Re: [PATCH net-next v2 2/3] vhost_net: Call peek_len when using xdp

[Paolo Abeni]

On Wed, 2024-02-28 at 19:05 +0800, Yunjian Wang wrote:
> If TUN supports AF_XDP TX zero-copy, the XDP program will enqueue
> packets to the XDP ring and wake up the vhost worker. This requires
> the vhost worker to call peek_len(), which can be used to consume
> XDP descriptors.
> 
> Signed-off-by: Yunjian Wang <wangyunjian@huawei.com>
> ---
>  drivers/vhost/net.c | 17 ++++++++++++-----
>  1 file changed, 12 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c
> index f2ed7167c848..077e74421558 100644
> --- a/drivers/vhost/net.c
> +++ b/drivers/vhost/net.c
> @@ -207,6 +207,11 @@ static int vhost_net_buf_peek_len(void *ptr)
>  	return __skb_array_len_with_tag(ptr);
>  }
>  
> +static bool vhost_sock_xdp(struct socket *sock)
> +{
> +	return sock_flag(sock->sk, SOCK_XDP);
> +}
> +
>  static int vhost_net_buf_peek(struct vhost_net_virtqueue *nvq)
>  {
>  	struct vhost_net_buf *rxq = &nvq->rxq;
> @@ -214,6 +219,13 @@ static int vhost_net_buf_peek(struct vhost_net_virtqueue *nvq)
>  	if (!vhost_net_buf_is_empty(rxq))
>  		goto out;
>  
> +	if (ptr_ring_empty(nvq->rx_ring)) {
> +		struct socket *sock = vhost_vq_get_backend(&nvq->vq);
> +		/* Call peek_len to consume XSK descriptors, when using xdp */
> +		if (vhost_sock_xdp(sock) && sock->ops->peek_len)
> +			sock->ops->peek_len(sock);

This really looks like a socket API misuse. Why can't you use ptr-ring
primitives to consume XSK descriptors? peek_len could be constified
some day, this code will prevent such (good) thing.

Cheers,

Paolo

RE: [PATCH net-next v2 2/3] vhost_net: Call peek_len when using xdp

[wangyunjian]

> -----Original Message-----
> From: Paolo Abeni [mailto:pabeni@redhat.com]
> Sent: Thursday, February 29, 2024 6:49 PM
> To: wangyunjian <wangyunjian@huawei.com>; mst@redhat.com;
> willemdebruijn.kernel@gmail.com; jasowang@redhat.com; kuba@kernel.org;
> bjorn@kernel.org; magnus.karlsson@intel.com; maciej.fijalkowski@intel.com;
> jonathan.lemon@gmail.com; davem@davemloft.net
> Cc: bpf@vger.kernel.org; netdev@vger.kernel.org;
> linux-kernel@vger.kernel.org; kvm@vger.kernel.org;
> virtualization@lists.linux.dev; xudingke <xudingke@huawei.com>; liwei (DT)
> <liwei395@huawei.com>
> Subject: Re: [PATCH net-next v2 2/3] vhost_net: Call peek_len when using xdp
> 
> On Wed, 2024-02-28 at 19:05 +0800, Yunjian Wang wrote:
> > If TUN supports AF_XDP TX zero-copy, the XDP program will enqueue
> > packets to the XDP ring and wake up the vhost worker. This requires
> > the vhost worker to call peek_len(), which can be used to consume XDP
> > descriptors.
> >
> > Signed-off-by: Yunjian Wang <wangyunjian@huawei.com>
> > ---
> >  drivers/vhost/net.c | 17 ++++++++++++-----
> >  1 file changed, 12 insertions(+), 5 deletions(-)
> >
> > diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c index
> > f2ed7167c848..077e74421558 100644
> > --- a/drivers/vhost/net.c
> > +++ b/drivers/vhost/net.c
> > @@ -207,6 +207,11 @@ static int vhost_net_buf_peek_len(void *ptr)
> >  	return __skb_array_len_with_tag(ptr);  }
> >
> > +static bool vhost_sock_xdp(struct socket *sock) {
> > +	return sock_flag(sock->sk, SOCK_XDP); }
> > +
> >  static int vhost_net_buf_peek(struct vhost_net_virtqueue *nvq)  {
> >  	struct vhost_net_buf *rxq = &nvq->rxq; @@ -214,6 +219,13 @@ static
> > int vhost_net_buf_peek(struct vhost_net_virtqueue *nvq)
> >  	if (!vhost_net_buf_is_empty(rxq))
> >  		goto out;
> >
> > +	if (ptr_ring_empty(nvq->rx_ring)) {
> > +		struct socket *sock = vhost_vq_get_backend(&nvq->vq);
> > +		/* Call peek_len to consume XSK descriptors, when using xdp */
> > +		if (vhost_sock_xdp(sock) && sock->ops->peek_len)
> > +			sock->ops->peek_len(sock);
> 
> This really looks like a socket API misuse. Why can't you use ptr-ring primitives
> to consume XSK descriptors? peek_len could be constified some day, this code
> will prevent such (good) thing.

Thank you for your suggestion. I will consider that with Patch 3/3.

> 
> Cheers,
> 
> Paolo