On Mon, 8 Jul 2024 16:08:44 +0800, Jason Wang <jasowang@xxxxxxxxxx> wrote:
> On Mon, Jul 8, 2024 at 3:47 PM Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx> wrote:
> >
> > On Mon, 8 Jul 2024 15:00:50 +0800, Jason Wang <jasowang@xxxxxxxxxx> wrote:
> > > On Fri, Jul 5, 2024 at 3:38 PM Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx> wrote:
> > > >
> > > > In the process:
> > > > 1. We may need to copy data to create skb for XDP_PASS.
> > > > 2. We may need to call xsk_buff_free() to release the buffer.
> > > > 3. The handle for xdp_buff is difference from the buffer.
> > > >
> > > > If we pushed this logic into existing receive handle(merge and small),
> > > > we would have to maintain code scattered inside merge and small (and big).
> > > > So I think it is a good choice for us to put the xsk code into an
> > > > independent function.
> > > >
> > > > Signed-off-by: Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx>
> > > > ---
> > > >
> > > > v7:
> > > > 1. rename xdp_construct_skb to xsk_construct_skb
> > > > 2. refactor virtnet_receive()
> > > >
> > > > drivers/net/virtio_net.c | 176 +++++++++++++++++++++++++++++++++++++--
> > > > 1 file changed, 168 insertions(+), 8 deletions(-)
> > > >
> > > > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> > > > index 2b27f5ada64a..64d8cd481890 100644
> > > > --- a/drivers/net/virtio_net.c
> > > > +++ b/drivers/net/virtio_net.c
> > > > @@ -498,6 +498,12 @@ struct virtio_net_common_hdr {
> > > > };
> > > >
> > > > static void virtnet_sq_free_unused_buf(struct virtqueue *vq, void *buf);
> > > > +static int virtnet_xdp_handler(struct bpf_prog *xdp_prog, struct xdp_buff *xdp,
> > > > + struct net_device *dev,
> > > > + unsigned int *xdp_xmit,
> > > > + struct virtnet_rq_stats *stats);
> > > > +static void virtnet_receive_done(struct virtnet_info *vi, struct receive_queue *rq,
> > > > + struct sk_buff *skb, u8 flags);
> > > >
> > > > static bool is_xdp_frame(void *ptr)
> > > > {
> > > > @@ -1062,6 +1068,124 @@ static void sg_fill_dma(struct scatterlist *sg, dma_addr_t addr, u32 len)
> > > > sg->length = len;
> > > > }
> > > >
> > > > +static struct xdp_buff *buf_to_xdp(struct virtnet_info *vi,
> > > > + struct receive_queue *rq, void *buf, u32 len)
> > > > +{
> > > > + struct xdp_buff *xdp;
> > > > + u32 bufsize;
> > > > +
> > > > + xdp = (struct xdp_buff *)buf;
> > > > +
> > > > + bufsize = xsk_pool_get_rx_frame_size(rq->xsk_pool) + vi->hdr_len;
> > > > +
> > > > + if (unlikely(len > bufsize)) {
> > > > + pr_debug("%s: rx error: len %u exceeds truesize %u\n",
> > > > + vi->dev->name, len, bufsize);
> > > > + DEV_STATS_INC(vi->dev, rx_length_errors);
> > > > + xsk_buff_free(xdp);
> > > > + return NULL;
> > > > + }
> > > > +
> > > > + xsk_buff_set_size(xdp, len);
> > > > + xsk_buff_dma_sync_for_cpu(xdp);
> > > > +
> > > > + return xdp;
> > > > +}
> > > > +
> > > > +static struct sk_buff *xsk_construct_skb(struct receive_queue *rq,
> > > > + struct xdp_buff *xdp)
> > > > +{
> > > > + unsigned int metasize = xdp->data - xdp->data_meta;
> > > > + struct sk_buff *skb;
> > > > + unsigned int size;
> > > > +
> > > > + size = xdp->data_end - xdp->data_hard_start;
> > > > + skb = napi_alloc_skb(&rq->napi, size);
> > > > + if (unlikely(!skb)) {
> > > > + xsk_buff_free(xdp);
> > > > + return NULL;
> > > > + }
> > > > +
> > > > + skb_reserve(skb, xdp->data_meta - xdp->data_hard_start);
> > > > +
> > > > + size = xdp->data_end - xdp->data_meta;
> > > > + memcpy(__skb_put(skb, size), xdp->data_meta, size);
> > > > +
> > > > + if (metasize) {
> > > > + __skb_pull(skb, metasize);
> > > > + skb_metadata_set(skb, metasize);
> > > > + }
> > > > +
> > > > + xsk_buff_free(xdp);
> > > > +
> > > > + return skb;
> > > > +}
> > > > +
> > > > +static struct sk_buff *virtnet_receive_xsk_small(struct net_device *dev, struct virtnet_info *vi,
> > > > + struct receive_queue *rq, struct xdp_buff *xdp,
> > > > + unsigned int *xdp_xmit,
> > > > + struct virtnet_rq_stats *stats)
> > > > +{
> > > > + struct bpf_prog *prog;
> > > > + u32 ret;
> > > > +
> > > > + ret = XDP_PASS;
> > > > + rcu_read_lock();
> > > > + prog = rcu_dereference(rq->xdp_prog);
> > > > + if (prog)
> > > > + ret = virtnet_xdp_handler(prog, xdp, dev, xdp_xmit, stats);
> > > > + rcu_read_unlock();
> > > > +
> > > > + switch (ret) {
> > > > + case XDP_PASS:
> > > > + return xsk_construct_skb(rq, xdp);
> > > > +
> > > > + case XDP_TX:
> > > > + case XDP_REDIRECT:
> > > > + return NULL;
> > > > +
> > > > + default:
> > > > + /* drop packet */
> > > > + xsk_buff_free(xdp);
> > > > + u64_stats_inc(&stats->drops);
> > > > + return NULL;
> > > > + }
> > > > +}
> > > > +
> > > > +static void virtnet_receive_xsk_buf(struct virtnet_info *vi, struct receive_queue *rq,
> > > > + void *buf, u32 len,
> > > > + unsigned int *xdp_xmit,
> > > > + struct virtnet_rq_stats *stats)
> > > > +{
> > > > + struct net_device *dev = vi->dev;
> > > > + struct sk_buff *skb = NULL;
> > > > + struct xdp_buff *xdp;
> > > > + u8 flags;
> > > > +
> > > > + len -= vi->hdr_len;
> > > > +
> > > > + u64_stats_add(&stats->bytes, len);
> > > > +
> > > > + xdp = buf_to_xdp(vi, rq, buf, len);
> > > > + if (!xdp)
> > > > + return;
> > > > +
> > > > + if (unlikely(len < ETH_HLEN)) {
> > > > + pr_debug("%s: short packet %i\n", dev->name, len);
> > > > + DEV_STATS_INC(dev, rx_length_errors);
> > > > + xsk_buff_free(xdp);
> > > > + return;
> > > > + }
> > > > +
> > > > + flags = ((struct virtio_net_common_hdr *)(xdp->data - vi->hdr_len))->hdr.flags;
> > > > +
> > > > + if (!vi->mergeable_rx_bufs)
> > > > + skb = virtnet_receive_xsk_small(dev, vi, rq, xdp, xdp_xmit, stats);
> > >
> > > I wonder if we add the mergeable support in the next patch would it be
> > > better to re-order the patch? For example, the xsk binding needs to be
> > > moved to the last patch, otherwise we break xsk with a mergeable
> > > buffer here?
> >
> > If you worry that the user works with this commit, I want to say you do not
> > worry.
> >
> > Because the flags NETDEV_XDP_ACT_XSK_ZEROCOPY is not added. I plan to add that
> > after the tx is completed.
>
> Ok, this is something I missed, it would be better to mention it
> somewhere (or it is already there but I miss it).
OK. I will add it to next version cover.
Thanks.
>
> >
> > I do test by adding this flags locally.
> >
> > Thanks.
>
> Acked-by: Jason Wang <jasowang@xxxxxxxxxx>
>
> Thanks
>
> >
> > >
> > > Or anything I missed here?
> > >
> > > Thanks
> > >
> >
>
