On Tue, Mar 29, 2022 at 12:27:51PM +0200, Ivan Vecera wrote:
> Function ice_tx_xsk_pool() used to get XSK buffer pool associated
> with XDP Tx queue returns NULL when number of ordinary Tx queues
> is not equal to num_possible_cpus().
>
> The function computes XDP Tx queue ID as an expression
> `ring->q_index - vsi->num_xdp_txq` but this is wrong because
> XDP Tx queues are placed after ordinary ones so the correct
> formula is `ring->q_index - vsi->alloc_txq`.
>
> Prior commit 792b2086584f ("ice: fix vsi->txq_map sizing") number
> of XDP Tx queues was equal to number of ordinary Tx queues so
> the bug in mentioned function was hidden.
>
> Reproducer:
> host# ethtool -L ens7f0 combined 1
> host# ./xdpsock -i ens7f0 -q 0 -t -N
> samples/bpf/xdpsock_user.c:kick_tx:794: errno: 6/"No such device or address"
>
> sock0@ens7f0:0 txonly xdp-drv
> pps pkts 0.00
> rx 0 0
> tx 0 0
>
> Fixes: 2d4238f55697 ("ice: Add support for AF_XDP")
> Fixes: 792b2086584f ("ice: fix vsi->txq_map sizing")
> Signed-off-by: Ivan Vecera <ivecera@xxxxxxxxxx>
Thanks for this fix! I did exactly the same patch yesterday and it's
already applied to bpf tree:
https://lore.kernel.org/bpf/20220328142123.170157-5-maciej.fijalkowski@xxxxxxxxx/T/#u
Maciej
> ---
> drivers/net/ethernet/intel/ice/ice.h | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/net/ethernet/intel/ice/ice.h b/drivers/net/ethernet/intel/ice/ice.h
> index b0b27bfcd7a2..d4f1874df7d0 100644
> --- a/drivers/net/ethernet/intel/ice/ice.h
> +++ b/drivers/net/ethernet/intel/ice/ice.h
> @@ -710,7 +710,7 @@ static inline struct xsk_buff_pool *ice_tx_xsk_pool(struct ice_tx_ring *ring)
> struct ice_vsi *vsi = ring->vsi;
> u16 qid;
>
> - qid = ring->q_index - vsi->num_xdp_txq;
> + qid = ring->q_index - vsi->alloc_txq;
>
> if (!ice_is_xdp_ena_vsi(vsi) || !test_bit(qid, vsi->af_xdp_zc_qps))
> return NULL;
> --
> 2.34.1
>
