On Mon, Mar 02, 2026 at 11:41:58PM +0700, Bui Quang Minh wrote:
> When XDP_USE_NEED_WAKEUP is used and the fill ring is empty so no buffer
> is allocated on RX side, allow RX NAPI to be descheduled. This avoids
> wasting CPU cycles on polling. Users will be notified and they need to
> make a wakeup call after refilling the ring.
>
> Signed-off-by: Bui Quang Minh <[email protected]>
> ---
> Changes in v2:
> - Fix the flag check in virtnet_xsk_wakeup
> - Link to v1:
> https://lore.kernel.org/netdev/[email protected]/
> ---
> drivers/net/virtio_net.c | 35 +++++++++++++++++++++++++++--------
> 1 file changed, 27 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> index db88dcaefb20..3614002fc87c 100644
> --- a/drivers/net/virtio_net.c
> +++ b/drivers/net/virtio_net.c
> @@ -1454,8 +1454,19 @@ static int virtnet_add_recvbuf_xsk(struct virtnet_info
> *vi, struct receive_queue
> xsk_buffs = rq->xsk_buffs;
>
> num = xsk_buff_alloc_batch(pool, xsk_buffs, rq->vq->num_free);
> - if (!num)
> + if (!num) {
> + if (xsk_uses_need_wakeup(pool)) {
> + xsk_set_rx_need_wakeup(pool);
> + /* Return 0 instead of -ENOMEM so that NAPI is
> + * descheduled.
> + */
> + return 0;
With this change, try_fill_recv() can return true even though zero RX
buffers were posted in need_wakeup mode, which contradicts its
documentation ("Returns false if we couldn't fill entirely (OOM)").
Methinks we need to at least update the comment near try_fill_recv.
> + }
> +
> return -ENOMEM;
> + } else {
> + xsk_clear_rx_need_wakeup(pool);
> + }
>
> len = xsk_pool_get_rx_frame_size(pool) + vi->hdr_len;
>
> @@ -1588,20 +1599,19 @@ static bool virtnet_xsk_xmit(struct send_queue *sq,
> struct xsk_buff_pool *pool,
> return sent;
> }
>
> -static void xsk_wakeup(struct send_queue *sq)
> +static void xsk_wakeup(struct napi_struct *napi, struct virtqueue *vq)
> {
> - if (napi_if_scheduled_mark_missed(&sq->napi))
> + if (napi_if_scheduled_mark_missed(napi))
> return;
>
> local_bh_disable();
> - virtqueue_napi_schedule(&sq->napi, sq->vq);
> + virtqueue_napi_schedule(napi, vq);
> local_bh_enable();
> }
>
> static int virtnet_xsk_wakeup(struct net_device *dev, u32 qid, u32 flag)
> {
> struct virtnet_info *vi = netdev_priv(dev);
> - struct send_queue *sq;
>
> if (!netif_running(dev))
> return -ENETDOWN;
> @@ -1609,9 +1619,18 @@ static int virtnet_xsk_wakeup(struct net_device *dev,
> u32 qid, u32 flag)
> if (qid >= vi->curr_queue_pairs)
> return -EINVAL;
>
> - sq = &vi->sq[qid];
> + if (flag & XDP_WAKEUP_TX) {
> + struct send_queue *sq = &vi->sq[qid];
> +
> + xsk_wakeup(&sq->napi, sq->vq);
> + }
> +
> + if (flag & XDP_WAKEUP_RX) {
> + struct receive_queue *rq = &vi->rq[qid];
> +
> + xsk_wakeup(&rq->napi, rq->vq);
> + }
>
> - xsk_wakeup(sq);
> return 0;
> }
>
> @@ -1623,7 +1642,7 @@ static void virtnet_xsk_completed(struct send_queue
> *sq, int num)
> * wakeup the tx napi to consume the xsk tx queue, because the tx
> * interrupt may not be triggered.
> */
> - xsk_wakeup(sq);
> + xsk_wakeup(&sq->napi, sq->vq);
> }
>
> static int __virtnet_xdp_xmit_one(struct virtnet_info *vi,
> --
> 2.43.0
