On Sat, Feb 28, 2026 at 1:03 PM Bui Quang Minh <[email protected]> wrote: > > Hi Jason, > > On 2/28/26 10:49, Jason Xing wrote: > > Hi Bui, > > > > On Fri, Feb 27, 2026 at 11:20 PM Bui Quang Minh > > <[email protected]> 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]> > >> --- > >> drivers/net/virtio_net.c | 38 ++++++++++++++++++++++++++++++-------- > >> 1 file changed, 30 insertions(+), 8 deletions(-) > >> > >> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c > >> index db88dcaefb20..494acc904b2c 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; > >> + } > >> + > >> return -ENOMEM; > >> + } else { > >> + xsk_clear_rx_need_wakeup(pool); > >> + } > >> > >> len = xsk_pool_get_rx_frame_size(pool) + vi->hdr_len; > >> > >> @@ -1588,20 +1599,21 @@ 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; > >> + struct napi_struct *napi; > >> + struct virtqueue *vq; > >> > >> if (!netif_running(dev)) > >> return -ENETDOWN; > >> @@ -1609,9 +1621,19 @@ 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) { > > Better use &? > > Sorry, I don't get your point. Can you elaborate more?
Oh, I meant using 'flag & XDP_WAKEUP_TX' is preferrable. IIUC, since this patch provides a way to enable the RX flag, for virtio_net, this flag can be set by TX and RX altogether? Please see this call trace: xsk_poll()->xsk_wakeup(xs, pool->cached_need_wakeup). Thanks, Jason
