>
> Call xsk_ring_prod__submit() before kick_tx() so that the kernel
> consumer sees the updated state of Tx ring. Otherwise, Tx packets are
> stuck in the ring until the next call to af_xdp_tx_zc().
>
> Fixes: d8a210774e1d ("net/af_xdp: support unaligned umem chunks")
> Cc: sta...@dpdk.org
>
> Signed-off-by: Baruch Siach <bar...@tkos.co.il>
Thanks for respinning. I tested it out and it looks good to me.
Acked-by: Ciara Loftus <ciara.lof...@intel.com>
> --
> v2:
>
> Don't call xsk_ring_prod__submit() when kick_tx() is only used to
> drain the completion queue (Ciara Loftus)
> ---
> drivers/net/af_xdp/rte_eth_af_xdp.c | 4 +---
> 1 file changed, 1 insertion(+), 3 deletions(-)
>
> diff --git a/drivers/net/af_xdp/rte_eth_af_xdp.c
> b/drivers/net/af_xdp/rte_eth_af_xdp.c
> index 74ffa4511284..9bea0a895a3e 100644
> --- a/drivers/net/af_xdp/rte_eth_af_xdp.c
> +++ b/drivers/net/af_xdp/rte_eth_af_xdp.c
> @@ -527,7 +527,6 @@ af_xdp_tx_zc(void *queue, struct rte_mbuf **bufs,
> uint16_t nb_pkts)
>
> if (!xsk_ring_prod__reserve(&txq->tx, 1, &idx_tx)) {
> rte_pktmbuf_free(local_mbuf);
> - kick_tx(txq, cq);
> goto out;
> }
>
> @@ -551,10 +550,9 @@ af_xdp_tx_zc(void *queue, struct rte_mbuf **bufs,
> uint16_t nb_pkts)
> tx_bytes += mbuf->pkt_len;
> }
>
> - kick_tx(txq, cq);
> -
> out:
> xsk_ring_prod__submit(&txq->tx, count);
> + kick_tx(txq, cq);
>
> txq->stats.tx_pkts += count;
> txq->stats.tx_bytes += tx_bytes;
> --
> 2.32.0
