net/bnx2x: reserve enough headroom for mbuf prepend

When allocating a new mbuf for Rx, the value of m->data_off should be
reset to its default value (RTE_PKTMBUF_HEADROOM), instead of reusing
the previous undefined value, which could cause the packet to have a
too small or too high headroom.

Signed-off-by: Yangchao Zhou <zhouyates@gmail.com>
Acked-by: Harish Patil <harish.patil@cavium.com>

diff --git a/drivers/net/bnx2x/bnx2x_rxtx.c b/drivers/net/bnx2x/bnx2x_rxtx.c
index 6be7277..1ab1abe 100644 (file)
--- a/drivers/net/bnx2x/bnx2x_rxtx.c
+++ b/drivers/net/bnx2x/bnx2x_rxtx.c
@@ -141,7 +141,8 @@ bnx2x_dev_rx_queue_setup(struct rte_eth_dev *dev,
                        return -ENOMEM;
                }
                rxq->sw_ring[idx] = mbuf;
-               rxq->rx_ring[idx] = mbuf->buf_iova;
+               rxq->rx_ring[idx] =
+                       rte_cpu_to_le_64(rte_mbuf_data_iova_default(mbuf));
        }
        rxq->pkt_first_seg = NULL;
        rxq->pkt_last_seg = NULL;
@@ -401,7 +402,8 @@ bnx2x_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
 
                rx_mb = rxq->sw_ring[bd_cons];
                rxq->sw_ring[bd_cons] = new_mb;
-               rxq->rx_ring[bd_prod] = new_mb->buf_iova;
+               rxq->rx_ring[bd_prod] =
+                       rte_cpu_to_le_64(rte_mbuf_data_iova_default(new_mb));
 
                rx_pref = NEXT_RX_BD(bd_cons) & MAX_RX_BD(rxq);
                rte_prefetch0(rxq->sw_ring[rx_pref]);
@@ -410,7 +412,7 @@ bnx2x_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
                        rte_prefetch0(&rxq->sw_ring[rx_pref]);
                }
 
-               rx_mb->data_off = pad;
+               rx_mb->data_off = pad + RTE_PKTMBUF_HEADROOM;
                rx_mb->nb_segs = 1;
                rx_mb->next = NULL;
                rx_mb->pkt_len = rx_mb->data_len = len;