On Tue, Nov 14, 2023 at 7:31 PM Xuan Zhuo <[email protected]> wrote:
>
> If the xsk is enabling, the xsk tx will share the send queue.
> But the xsk requires that the send queue use the premapped mode.
> So the send queue must support premapped mode.
>
> Signed-off-by: Xuan Zhuo <[email protected]>
> ---
> drivers/net/virtio/main.c | 114 ++++++++++++++++++++++++++++----
> drivers/net/virtio/virtio_net.h | 2 +
> 2 files changed, 102 insertions(+), 14 deletions(-)
>
> diff --git a/drivers/net/virtio/main.c b/drivers/net/virtio/main.c
> index 57cbd449637a..0766fed28afe 100644
> --- a/drivers/net/virtio/main.c
> +++ b/drivers/net/virtio/main.c
> @@ -167,13 +167,46 @@ static struct xdp_frame *ptr_to_xdp(void *ptr)
> return (struct xdp_frame *)((unsigned long)ptr & ~VIRTIO_XDP_FLAG);
> }
>
> +static void virtnet_sq_unmap(struct virtnet_sq *sq)
> +{
> + struct scatterlist *sg;
> +
> + for (sg = sq->sg; ; sg = sg_next(sg)) {
> + virtqueue_dma_unmap_single_attrs(sq->vq,
> + sg->dma_address,
> + sg->length,
> + DMA_TO_DEVICE, 0);
> +
> + if (sg_is_last(sg)) {
> + /* clear the end mark setted by virtio core. */
> + sg_unmark_end(sg);
> + break;
> + }
> + }
> +
> + /* mark the last one of sq->sg[] as the end for reuse by
> + * virtqueue_get_buf_ctx_dma().
> + */
> + sg_mark_end(&sq->sg[ARRAY_SIZE(sq->sg) - 1]);
This API seems to be odd: the flag is set to be core but clear by driver.
Can we simply follow the add API that returns the #sgs?
> +}
> +
> static void __free_old_xmit(struct virtnet_sq *sq, bool in_napi,
> u64 *bytes, u64 *packets)
> {
> + struct scatterlist *sg = NULL;
> unsigned int len;
> void *ptr;
>
> - while ((ptr = virtqueue_get_buf(sq->vq, &len)) != NULL) {
> + /* clear the sg table */
> + if (sq->do_dma) {
> + sg_init_table(sq->sg, ARRAY_SIZE(sq->sg));
> + sg = sq->sg;
> + }
> +
> + while ((ptr = virtqueue_get_buf_ctx_dma(sq->vq, &len, sg, NULL)) !=
> NULL) {
> + if (sq->do_dma)
> + virtnet_sq_unmap(sq);
> +
> if (!is_xdp_frame(ptr)) {
> struct sk_buff *skb = ptr;
>
> @@ -567,22 +600,61 @@ static void *virtnet_rq_alloc(struct virtnet_rq *rq,
> u32 size, gfp_t gfp)
> return buf;
> }
>
> -static void virtnet_rq_set_premapped(struct virtnet_info *vi)
> +static void virtnet_set_premapped(struct virtnet_info *vi)
> {
> int i;
>
> - /* disable for big mode */
> - if (!vi->mergeable_rx_bufs && vi->big_packets)
> - return;
> -
> for (i = 0; i < vi->max_queue_pairs; i++) {
> - if (virtqueue_set_dma_premapped(vi->rq[i].vq))
> - continue;
> + if (!virtqueue_set_dma_premapped(vi->sq[i].vq))
> + vi->sq[i].do_dma = true;
>
> - vi->rq[i].do_dma = true;
> + /* disable for big mode */
> + if (vi->mergeable_rx_bufs || !vi->big_packets) {
> + if (!virtqueue_set_dma_premapped(vi->rq[i].vq))
> + vi->rq[i].do_dma = true;
> + }
> }
> }
>
> +static int virtnet_sq_map_sg(struct virtnet_sq *sq)
> +{
> + struct scatterlist *sg;
> +
> + for (sg = sq->sg; sg; sg = sg_next(sg)) {
> + sg->dma_address = virtqueue_dma_map_single_attrs(sq->vq,
> sg_virt(sg),
> + sg->length,
> +
> DMA_TO_DEVICE, 0);
> + if (virtqueue_dma_mapping_error(sq->vq, sg->dma_address))
> + goto err;
> + }
> +
> + return 0;
> +
> +err:
> + if (sg != sq->sg) {
> + sg_mark_end(sg - 1);
> + virtnet_sq_unmap(sq);
> + }
> + return -ENOMEM;
> +}
> +
> +static int virtnet_add_outbuf(struct virtnet_sq *sq, u32 num, void *data)
> +{
> + int ret;
> +
> + if (sq->do_dma) {
> + ret = virtnet_sq_map_sg(sq);
> + if (ret)
> + return -ENOMEM;
> + }
> +
> + ret = virtqueue_add_outbuf(sq->vq, sq->sg, num, data, GFP_ATOMIC);
> + if (ret && sq->do_dma)
> + virtnet_sq_unmap(sq);
> +
> + return ret;
> +}
> +
> static void free_old_xmit(struct virtnet_sq *sq, bool in_napi)
> {
> u64 bytes, packets = 0;
> @@ -686,8 +758,7 @@ static int __virtnet_xdp_xmit_one(struct virtnet_info *vi,
> skb_frag_size(frag), skb_frag_off(frag));
> }
>
> - err = virtqueue_add_outbuf(sq->vq, sq->sg, nr_frags + 1,
> - xdp_to_ptr(xdpf), GFP_ATOMIC);
> + err = virtnet_add_outbuf(sq, nr_frags + 1, xdp_to_ptr(xdpf));
> if (unlikely(err))
> return -ENOSPC; /* Caller handle free/refcnt */
>
> @@ -2126,7 +2197,7 @@ static int xmit_skb(struct virtnet_sq *sq, struct
> sk_buff *skb)
> return num_sg;
> num_sg++;
> }
> - return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
> + return virtnet_add_outbuf(sq, num_sg, skb);
> }
>
> static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
> @@ -3852,10 +3923,25 @@ static void virtnet_rq_free_unused_buf(struct
> virtqueue *vq, void *buf)
>
> static void virtnet_sq_free_unused_bufs(struct virtqueue *vq)
> {
> + struct virtnet_info *vi = vq->vdev->priv;
> + struct scatterlist *sg = NULL;
> + struct virtnet_sq *sq;
> + int i = vq2txq(vq);
> void *buf;
>
> - while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
> + sq = &vi->sq[i];
> +
> + if (sq->do_dma) {
> + sg_init_table(sq->sg, ARRAY_SIZE(sq->sg));
Btw, if we only care about dma_addr and length, using sg seems to be
expensive here.
Thanks
> + sg = sq->sg;
> + }
> +
> + while ((buf = virtqueue_detach_unused_buf_premapped(vq, sg)) != NULL)
> {
> + if (sq->do_dma)
> + virtnet_sq_unmap(sq);
> +
> virtnet_sq_free_unused_buf(vq, buf);
> + }
> }
>
> static void virtnet_rq_free_unused_bufs(struct virtqueue *vq)
> @@ -4052,7 +4138,7 @@ static int init_vqs(struct virtnet_info *vi)
> if (ret)
> goto err_free;
>
> - virtnet_rq_set_premapped(vi);
> + virtnet_set_premapped(vi);
>
> cpus_read_lock();
> virtnet_set_affinity(vi);
> diff --git a/drivers/net/virtio/virtio_net.h b/drivers/net/virtio/virtio_net.h
> index ebf9f344648a..870855a829a0 100644
> --- a/drivers/net/virtio/virtio_net.h
> +++ b/drivers/net/virtio/virtio_net.h
> @@ -67,6 +67,8 @@ struct virtnet_sq {
>
> /* Record whether sq is in reset state. */
> bool reset;
> +
> + bool do_dma;
> };
>
> /* Internal representation of a receive virtqueue */
> --
> 2.32.0.3.g01195cf9f
>
