add prefix to cache line macros
[dpdk.git] / lib / librte_pmd_ixgbe / ixgbe_rxtx.c
index eec1458..8559ef6 100644 (file)
@@ -2,6 +2,7 @@
  *   BSD LICENSE
  *
  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
+ *   Copyright 2014 6WIND S.A.
  *   All rights reserved.
  *
  *   Redistribution and use in source and binary forms, with or without
                ETH_RSS_IPV6_UDP | \
                ETH_RSS_IPV6_UDP_EX)
 
+/* Bit Mask to indicate what bits required for building TX context */
+#define IXGBE_TX_OFFLOAD_MASK (                         \
+               PKT_TX_VLAN_PKT |                \
+               PKT_TX_IP_CKSUM |                \
+               PKT_TX_L4_MASK |                 \
+               PKT_TX_TCP_SEG)
+
 static inline struct rte_mbuf *
 rte_rxmbuf_alloc(struct rte_mempool *mp)
 {
@@ -142,13 +150,10 @@ ixgbe_tx_free_bufs(struct igb_tx_queue *txq)
         */
        txep = &(txq->sw_ring[txq->tx_next_dd - (txq->tx_rs_thresh - 1)]);
 
-       /* prefetch the mbufs that are about to be freed */
-       for (i = 0; i < txq->tx_rs_thresh; ++i)
-               rte_prefetch0((txep + i)->mbuf);
-
        /* free buffers one at a time */
        if ((txq->txq_flags & (uint32_t)ETH_TXQ_FLAGS_NOREFCOUNT) != 0) {
                for (i = 0; i < txq->tx_rs_thresh; ++i, ++txep) {
+                       txep->mbuf->next = NULL;
                        rte_mempool_put(txep->mbuf->pool, txep->mbuf);
                        txep->mbuf = NULL;
                }
@@ -186,6 +191,7 @@ tx4(volatile union ixgbe_adv_tx_desc *txdp, struct rte_mbuf **pkts)
                                ((uint32_t)DCMD_DTYP_FLAGS | pkt_len);
                txdp->read.olinfo_status =
                                (pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
+               rte_prefetch0(&(*pkts)->pool);
        }
 }
 
@@ -205,6 +211,7 @@ tx1(volatile union ixgbe_adv_tx_desc *txdp, struct rte_mbuf **pkts)
                        ((uint32_t)DCMD_DTYP_FLAGS | pkt_len);
        txdp->read.olinfo_status =
                        (pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
+       rte_prefetch0(&(*pkts)->pool);
 }
 
 /*
@@ -358,59 +365,84 @@ ixgbe_xmit_pkts_simple(void *tx_queue, struct rte_mbuf **tx_pkts,
 static inline void
 ixgbe_set_xmit_ctx(struct igb_tx_queue* txq,
                volatile struct ixgbe_adv_tx_context_desc *ctx_txd,
-               uint16_t ol_flags, uint32_t vlan_macip_lens)
+               uint64_t ol_flags, union ixgbe_tx_offload tx_offload)
 {
        uint32_t type_tucmd_mlhl;
-       uint32_t mss_l4len_idx;
+       uint32_t mss_l4len_idx = 0;
        uint32_t ctx_idx;
-       uint32_t cmp_mask;
+       uint32_t vlan_macip_lens;
+       union ixgbe_tx_offload tx_offload_mask;
 
        ctx_idx = txq->ctx_curr;
-       cmp_mask = 0;
+       tx_offload_mask.data = 0;
        type_tucmd_mlhl = 0;
 
+       /* Specify which HW CTX to upload. */
+       mss_l4len_idx |= (ctx_idx << IXGBE_ADVTXD_IDX_SHIFT);
+
        if (ol_flags & PKT_TX_VLAN_PKT) {
-               cmp_mask |= TX_VLAN_CMP_MASK;
+               tx_offload_mask.vlan_tci = ~0;
        }
 
-       if (ol_flags & PKT_TX_IP_CKSUM) {
-               type_tucmd_mlhl = IXGBE_ADVTXD_TUCMD_IPV4;
-               cmp_mask |= TX_MAC_LEN_CMP_MASK;
-       }
+       /* check if TCP segmentation required for this packet */
+       if (ol_flags & PKT_TX_TCP_SEG) {
+               /* implies IP cksum and TCP cksum */
+               type_tucmd_mlhl = IXGBE_ADVTXD_TUCMD_IPV4 |
+                       IXGBE_ADVTXD_TUCMD_L4T_TCP |
+                       IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
+
+               tx_offload_mask.l2_len = ~0;
+               tx_offload_mask.l3_len = ~0;
+               tx_offload_mask.l4_len = ~0;
+               tx_offload_mask.tso_segsz = ~0;
+               mss_l4len_idx |= tx_offload.tso_segsz << IXGBE_ADVTXD_MSS_SHIFT;
+               mss_l4len_idx |= tx_offload.l4_len << IXGBE_ADVTXD_L4LEN_SHIFT;
+       } else { /* no TSO, check if hardware checksum is needed */
+               if (ol_flags & PKT_TX_IP_CKSUM) {
+                       type_tucmd_mlhl = IXGBE_ADVTXD_TUCMD_IPV4;
+                       tx_offload_mask.l2_len = ~0;
+                       tx_offload_mask.l3_len = ~0;
+               }
 
-       /* Specify which HW CTX to upload. */
-       mss_l4len_idx = (ctx_idx << IXGBE_ADVTXD_IDX_SHIFT);
-       switch (ol_flags & PKT_TX_L4_MASK) {
-       case PKT_TX_UDP_CKSUM:
-               type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_UDP |
+               switch (ol_flags & PKT_TX_L4_MASK) {
+               case PKT_TX_UDP_CKSUM:
+                       type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_UDP |
                                IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
-               mss_l4len_idx |= sizeof(struct udp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
-               cmp_mask |= TX_MACIP_LEN_CMP_MASK;
-               break;
-       case PKT_TX_TCP_CKSUM:
-               type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP |
+                       mss_l4len_idx |= sizeof(struct udp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
+                       tx_offload_mask.l2_len = ~0;
+                       tx_offload_mask.l3_len = ~0;
+                       break;
+               case PKT_TX_TCP_CKSUM:
+                       type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP |
                                IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
-               mss_l4len_idx |= sizeof(struct tcp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
-               cmp_mask |= TX_MACIP_LEN_CMP_MASK;
-               break;
-       case PKT_TX_SCTP_CKSUM:
-               type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_SCTP |
+                       mss_l4len_idx |= sizeof(struct tcp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
+                       tx_offload_mask.l2_len = ~0;
+                       tx_offload_mask.l3_len = ~0;
+                       tx_offload_mask.l4_len = ~0;
+                       break;
+               case PKT_TX_SCTP_CKSUM:
+                       type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_SCTP |
                                IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
-               mss_l4len_idx |= sizeof(struct sctp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
-               cmp_mask |= TX_MACIP_LEN_CMP_MASK;
-               break;
-       default:
-               type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_RSV |
+                       mss_l4len_idx |= sizeof(struct sctp_hdr) << IXGBE_ADVTXD_L4LEN_SHIFT;
+                       tx_offload_mask.l2_len = ~0;
+                       tx_offload_mask.l3_len = ~0;
+                       break;
+               default:
+                       type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_RSV |
                                IXGBE_ADVTXD_DTYP_CTXT | IXGBE_ADVTXD_DCMD_DEXT;
-               break;
+                       break;
+               }
        }
 
        txq->ctx_cache[ctx_idx].flags = ol_flags;
-       txq->ctx_cache[ctx_idx].cmp_mask = cmp_mask;
-       txq->ctx_cache[ctx_idx].vlan_macip_lens.data =
-               vlan_macip_lens & cmp_mask;
+       txq->ctx_cache[ctx_idx].tx_offload.data  =
+               tx_offload_mask.data & tx_offload.data;
+       txq->ctx_cache[ctx_idx].tx_offload_mask    = tx_offload_mask;
 
        ctx_txd->type_tucmd_mlhl = rte_cpu_to_le_32(type_tucmd_mlhl);
+       vlan_macip_lens = tx_offload.l3_len;
+       vlan_macip_lens |= (tx_offload.l2_len << IXGBE_ADVTXD_MACLEN_SHIFT);
+       vlan_macip_lens |= ((uint32_t)tx_offload.vlan_tci << IXGBE_ADVTXD_VLAN_SHIFT);
        ctx_txd->vlan_macip_lens = rte_cpu_to_le_32(vlan_macip_lens);
        ctx_txd->mss_l4len_idx   = rte_cpu_to_le_32(mss_l4len_idx);
        ctx_txd->seqnum_seed     = 0;
@@ -421,21 +453,21 @@ ixgbe_set_xmit_ctx(struct igb_tx_queue* txq,
  * or create a new context descriptor.
  */
 static inline uint32_t
-what_advctx_update(struct igb_tx_queue *txq, uint16_t flags,
-               uint32_t vlan_macip_lens)
+what_advctx_update(struct igb_tx_queue *txq, uint64_t flags,
+               union ixgbe_tx_offload tx_offload)
 {
        /* If match with the current used context */
        if (likely((txq->ctx_cache[txq->ctx_curr].flags == flags) &&
-               (txq->ctx_cache[txq->ctx_curr].vlan_macip_lens.data ==
-               (txq->ctx_cache[txq->ctx_curr].cmp_mask & vlan_macip_lens)))) {
+               (txq->ctx_cache[txq->ctx_curr].tx_offload.data ==
+               (txq->ctx_cache[txq->ctx_curr].tx_offload_mask.data & tx_offload.data)))) {
                        return txq->ctx_curr;
        }
 
        /* What if match with the next context  */
        txq->ctx_curr ^= 1;
        if (likely((txq->ctx_cache[txq->ctx_curr].flags == flags) &&
-               (txq->ctx_cache[txq->ctx_curr].vlan_macip_lens.data ==
-               (txq->ctx_cache[txq->ctx_curr].cmp_mask & vlan_macip_lens)))) {
+               (txq->ctx_cache[txq->ctx_curr].tx_offload.data ==
+               (txq->ctx_cache[txq->ctx_curr].tx_offload_mask.data & tx_offload.data)))) {
                        return txq->ctx_curr;
        }
 
@@ -444,22 +476,27 @@ what_advctx_update(struct igb_tx_queue *txq, uint16_t flags,
 }
 
 static inline uint32_t
-tx_desc_cksum_flags_to_olinfo(uint16_t ol_flags)
+tx_desc_cksum_flags_to_olinfo(uint64_t ol_flags)
 {
-       static const uint32_t l4_olinfo[2] = {0, IXGBE_ADVTXD_POPTS_TXSM};
-       static const uint32_t l3_olinfo[2] = {0, IXGBE_ADVTXD_POPTS_IXSM};
-       uint32_t tmp;
-
-       tmp  = l4_olinfo[(ol_flags & PKT_TX_L4_MASK)  != PKT_TX_L4_NO_CKSUM];
-       tmp |= l3_olinfo[(ol_flags & PKT_TX_IP_CKSUM) != 0];
+       uint32_t tmp = 0;
+       if ((ol_flags & PKT_TX_L4_MASK) != PKT_TX_L4_NO_CKSUM)
+               tmp |= IXGBE_ADVTXD_POPTS_TXSM;
+       if (ol_flags & PKT_TX_IP_CKSUM)
+               tmp |= IXGBE_ADVTXD_POPTS_IXSM;
+       if (ol_flags & PKT_TX_TCP_SEG)
+               tmp |= IXGBE_ADVTXD_POPTS_TXSM;
        return tmp;
 }
 
 static inline uint32_t
-tx_desc_vlan_flags_to_cmdtype(uint16_t ol_flags)
+tx_desc_ol_flags_to_cmdtype(uint64_t ol_flags)
 {
-       static const uint32_t vlan_cmd[2] = {0, IXGBE_ADVTXD_DCMD_VLE};
-       return vlan_cmd[(ol_flags & PKT_TX_VLAN_PKT) != 0];
+       uint32_t cmdtype = 0;
+       if (ol_flags & PKT_TX_VLAN_PKT)
+               cmdtype |= IXGBE_ADVTXD_DCMD_VLE;
+       if (ol_flags & PKT_TX_TCP_SEG)
+               cmdtype |= IXGBE_ADVTXD_DCMD_TSE;
+       return cmdtype;
 }
 
 /* Default RS bit threshold values */
@@ -545,15 +582,15 @@ ixgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
        uint32_t cmd_type_len;
        uint32_t pkt_len;
        uint16_t slen;
-       uint16_t ol_flags;
+       uint64_t ol_flags;
        uint16_t tx_id;
        uint16_t tx_last;
        uint16_t nb_tx;
        uint16_t nb_used;
-       uint16_t tx_ol_req;
-       uint32_t vlan_macip_lens;
+       uint64_t tx_ol_req;
        uint32_t ctx = 0;
        uint32_t new_ctx;
+       union ixgbe_tx_offload tx_offload = { .data = 0 };
 
        txq = tx_queue;
        sw_ring = txq->sw_ring;
@@ -566,27 +603,32 @@ ixgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                ixgbe_xmit_cleanup(txq);
        }
 
+       rte_prefetch0(&txe->mbuf->pool);
+
        /* TX loop */
        for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
                new_ctx = 0;
                tx_pkt = *tx_pkts++;
                pkt_len = tx_pkt->pkt_len;
 
-               RTE_MBUF_PREFETCH_TO_FREE(txe->mbuf);
-
                /*
                 * Determine how many (if any) context descriptors
                 * are needed for offload functionality.
                 */
                ol_flags = tx_pkt->ol_flags;
-               vlan_macip_lens = tx_pkt->vlan_macip.data;
 
                /* If hardware offload required */
-               tx_ol_req = (uint16_t)(ol_flags & PKT_TX_OFFLOAD_MASK);
+               tx_ol_req = ol_flags & IXGBE_TX_OFFLOAD_MASK;
                if (tx_ol_req) {
+                       tx_offload.l2_len = tx_pkt->l2_len;
+                       tx_offload.l3_len = tx_pkt->l3_len;
+                       tx_offload.l4_len = tx_pkt->l4_len;
+                       tx_offload.vlan_tci = tx_pkt->vlan_tci;
+                       tx_offload.tso_segsz = tx_pkt->tso_segsz;
+
                        /* If new context need be built or reuse the exist ctx. */
                        ctx = what_advctx_update(txq, tx_ol_req,
-                               vlan_macip_lens);
+                               tx_offload);
                        /* Only allocate context descriptor if required*/
                        new_ctx = (ctx == IXGBE_CTX_NUM);
                        ctx = txq->ctx_curr;
@@ -614,7 +656,7 @@ ixgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                        tx_last = (uint16_t) (tx_last - txq->nb_tx_desc);
 
                PMD_TX_LOG(DEBUG, "port_id=%u queue_id=%u pktlen=%u"
-                          " tx_first=%u tx_last=%u\n",
+                          " tx_first=%u tx_last=%u",
                           (unsigned) txq->port_id,
                           (unsigned) txq->queue_id,
                           (unsigned) pkt_len,
@@ -701,13 +743,22 @@ ixgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                 */
                cmd_type_len = IXGBE_ADVTXD_DTYP_DATA |
                        IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
-               olinfo_status = (pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
+
 #ifdef RTE_LIBRTE_IEEE1588
                if (ol_flags & PKT_TX_IEEE1588_TMST)
                        cmd_type_len |= IXGBE_ADVTXD_MAC_1588;
 #endif
 
+               olinfo_status = 0;
                if (tx_ol_req) {
+
+                       if (ol_flags & PKT_TX_TCP_SEG) {
+                               /* when TSO is on, paylen in descriptor is the
+                                * not the packet len but the tcp payload len */
+                               pkt_len -= (tx_offload.l2_len +
+                                       tx_offload.l3_len + tx_offload.l4_len);
+                       }
+
                        /*
                         * Setup the TX Advanced Context Descriptor if required
                         */
@@ -720,7 +771,7 @@ ixgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                                    &txr[tx_id];
 
                                txn = &sw_ring[txe->next_id];
-                               RTE_MBUF_PREFETCH_TO_FREE(txn->mbuf);
+                               rte_prefetch0(&txn->mbuf->pool);
 
                                if (txe->mbuf != NULL) {
                                        rte_pktmbuf_free_seg(txe->mbuf);
@@ -728,7 +779,7 @@ ixgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                                }
 
                                ixgbe_set_xmit_ctx(txq, ctx_txd, tx_ol_req,
-                                   vlan_macip_lens);
+                                       tx_offload);
 
                                txe->last_id = tx_last;
                                tx_id = txe->next_id;
@@ -740,15 +791,18 @@ ixgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                         * This path will go through
                         * whatever new/reuse the context descriptor
                         */
-                       cmd_type_len  |= tx_desc_vlan_flags_to_cmdtype(ol_flags);
+                       cmd_type_len  |= tx_desc_ol_flags_to_cmdtype(ol_flags);
                        olinfo_status |= tx_desc_cksum_flags_to_olinfo(ol_flags);
                        olinfo_status |= ctx << IXGBE_ADVTXD_IDX_SHIFT;
                }
 
+               olinfo_status |= (pkt_len << IXGBE_ADVTXD_PAYLEN_SHIFT);
+
                m_seg = tx_pkt;
                do {
                        txd = &txr[tx_id];
                        txn = &sw_ring[txe->next_id];
+                       rte_prefetch0(&txn->mbuf->pool);
 
                        if (txe->mbuf != NULL)
                                rte_pktmbuf_free_seg(txe->mbuf);
@@ -812,19 +866,19 @@ end_of_tx:
  *  RX functions
  *
  **********************************************************************/
-static inline uint16_t
+static inline uint64_t
 rx_desc_hlen_type_rss_to_pkt_flags(uint32_t hl_tp_rs)
 {
-       uint16_t pkt_flags;
+       uint64_t pkt_flags;
 
-       static uint16_t ip_pkt_types_map[16] = {
+       static uint64_t ip_pkt_types_map[16] = {
                0, PKT_RX_IPV4_HDR, PKT_RX_IPV4_HDR_EXT, PKT_RX_IPV4_HDR_EXT,
                PKT_RX_IPV6_HDR, 0, 0, 0,
                PKT_RX_IPV6_HDR_EXT, 0, 0, 0,
                PKT_RX_IPV6_HDR_EXT, 0, 0, 0,
        };
 
-       static uint16_t ip_rss_types_map[16] = {
+       static uint64_t ip_rss_types_map[16] = {
                0, PKT_RX_RSS_HASH, PKT_RX_RSS_HASH, PKT_RX_RSS_HASH,
                0, PKT_RX_RSS_HASH, 0, PKT_RX_RSS_HASH,
                PKT_RX_RSS_HASH, 0, 0, 0,
@@ -832,50 +886,49 @@ rx_desc_hlen_type_rss_to_pkt_flags(uint32_t hl_tp_rs)
        };
 
 #ifdef RTE_LIBRTE_IEEE1588
-       static uint32_t ip_pkt_etqf_map[8] = {
+       static uint64_t ip_pkt_etqf_map[8] = {
                0, 0, 0, PKT_RX_IEEE1588_PTP,
                0, 0, 0, 0,
        };
 
-       pkt_flags = (uint16_t) ((hl_tp_rs & IXGBE_RXDADV_PKTTYPE_ETQF) ?
-                               ip_pkt_etqf_map[(hl_tp_rs >> 4) & 0x07] :
-                               ip_pkt_types_map[(hl_tp_rs >> 4) & 0x0F]);
+       pkt_flags = (hl_tp_rs & IXGBE_RXDADV_PKTTYPE_ETQF) ?
+                       ip_pkt_etqf_map[(hl_tp_rs >> 4) & 0x07] :
+                       ip_pkt_types_map[(hl_tp_rs >> 4) & 0x0F];
 #else
-       pkt_flags = (uint16_t) ((hl_tp_rs & IXGBE_RXDADV_PKTTYPE_ETQF) ? 0 :
-                               ip_pkt_types_map[(hl_tp_rs >> 4) & 0x0F]);
+       pkt_flags = (hl_tp_rs & IXGBE_RXDADV_PKTTYPE_ETQF) ? 0 :
+                       ip_pkt_types_map[(hl_tp_rs >> 4) & 0x0F];
 
 #endif
-       return (uint16_t)(pkt_flags | ip_rss_types_map[hl_tp_rs & 0xF]);
+       return pkt_flags | ip_rss_types_map[hl_tp_rs & 0xF];
 }
 
-static inline uint16_t
+static inline uint64_t
 rx_desc_status_to_pkt_flags(uint32_t rx_status)
 {
-       uint16_t pkt_flags;
+       uint64_t pkt_flags;
 
        /*
         * Check if VLAN present only.
         * Do not check whether L3/L4 rx checksum done by NIC or not,
         * That can be found from rte_eth_rxmode.hw_ip_checksum flag
         */
-       pkt_flags = (uint16_t)((rx_status & IXGBE_RXD_STAT_VP) ?
-                                               PKT_RX_VLAN_PKT : 0);
+       pkt_flags = (rx_status & IXGBE_RXD_STAT_VP) ?  PKT_RX_VLAN_PKT : 0;
 
 #ifdef RTE_LIBRTE_IEEE1588
        if (rx_status & IXGBE_RXD_STAT_TMST)
-               pkt_flags = (uint16_t)(pkt_flags | PKT_RX_IEEE1588_TMST);
+               pkt_flags = pkt_flags | PKT_RX_IEEE1588_TMST;
 #endif
        return pkt_flags;
 }
 
-static inline uint16_t
+static inline uint64_t
 rx_desc_error_to_pkt_flags(uint32_t rx_status)
 {
        /*
         * Bit 31: IPE, IPv4 checksum error
         * Bit 30: L4I, L4I integrity error
         */
-       static uint16_t error_to_pkt_flags_map[4] = {
+       static uint64_t error_to_pkt_flags_map[4] = {
                0,  PKT_RX_L4_CKSUM_BAD, PKT_RX_IP_CKSUM_BAD,
                PKT_RX_IP_CKSUM_BAD | PKT_RX_L4_CKSUM_BAD
        };
@@ -902,6 +955,7 @@ ixgbe_rx_scan_hw_ring(struct igb_rx_queue *rxq)
        struct igb_rx_entry *rxep;
        struct rte_mbuf *mb;
        uint16_t pkt_len;
+       uint64_t pkt_flags;
        int s[LOOK_AHEAD], nb_dd;
        int i, j, nb_rx = 0;
 
@@ -935,21 +989,28 @@ ixgbe_rx_scan_hw_ring(struct igb_rx_queue *rxq)
                /* Translate descriptor info to mbuf format */
                for (j = 0; j < nb_dd; ++j) {
                        mb = rxep[j].mbuf;
-                       pkt_len = (uint16_t)(rxdp[j].wb.upper.length -
-                                                       rxq->crc_len);
+                       pkt_len = (uint16_t)(rxdp[j].wb.upper.length - rxq->crc_len);
                        mb->data_len = pkt_len;
                        mb->pkt_len = pkt_len;
-                       mb->vlan_macip.f.vlan_tci = rxdp[j].wb.upper.vlan;
-                       mb->hash.rss = rxdp[j].wb.lower.hi_dword.rss;
+                       mb->vlan_tci = rxdp[j].wb.upper.vlan;
+                       mb->vlan_tci = rte_le_to_cpu_16(rxdp[j].wb.upper.vlan);
 
                        /* convert descriptor fields to rte mbuf flags */
-                       mb->ol_flags  = rx_desc_hlen_type_rss_to_pkt_flags(
+                       pkt_flags  = rx_desc_hlen_type_rss_to_pkt_flags(
                                        rxdp[j].wb.lower.lo_dword.data);
                        /* reuse status field from scan list */
-                       mb->ol_flags = (uint16_t)(mb->ol_flags |
-                                       rx_desc_status_to_pkt_flags(s[j]));
-                       mb->ol_flags = (uint16_t)(mb->ol_flags |
-                                       rx_desc_error_to_pkt_flags(s[j]));
+                       pkt_flags |= rx_desc_status_to_pkt_flags(s[j]);
+                       pkt_flags |= rx_desc_error_to_pkt_flags(s[j]);
+                       mb->ol_flags = pkt_flags;
+
+                       if (likely(pkt_flags & PKT_RX_RSS_HASH))
+                               mb->hash.rss = rxdp[j].wb.lower.hi_dword.rss;
+                       else if (pkt_flags & PKT_RX_FDIR) {
+                               mb->hash.fdir.hash =
+                                       (uint16_t)((rxdp[j].wb.lower.hi_dword.csum_ip.csum)
+                                               & IXGBE_ATR_HASH_MASK);
+                               mb->hash.fdir.id = rxdp[j].wb.lower.hi_dword.csum_ip.ip_id;
+                       }
                }
 
                /* Move mbuf pointers from the S/W ring to the stage */
@@ -996,7 +1057,7 @@ ixgbe_rx_alloc_bufs(struct igb_rx_queue *rxq)
                mb = rxep[i].mbuf;
                rte_mbuf_refcnt_set(mb, 1);
                mb->next = NULL;
-               mb->data = (char *)mb->buf_addr + RTE_PKTMBUF_HEADROOM;
+               mb->data_off = RTE_PKTMBUF_HEADROOM;
                mb->nb_segs = 1;
                mb->port = rxq->port_id;
 
@@ -1065,7 +1126,7 @@ rx_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                if (ixgbe_rx_alloc_bufs(rxq) != 0) {
                        int i, j;
                        PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
-                                  "queue_id=%u\n", (unsigned) rxq->port_id,
+                                  "queue_id=%u", (unsigned) rxq->port_id,
                                   (unsigned) rxq->queue_id);
 
                        rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed +=
@@ -1142,7 +1203,7 @@ ixgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
        uint16_t rx_id;
        uint16_t nb_rx;
        uint16_t nb_hold;
-       uint16_t pkt_flags;
+       uint64_t pkt_flags;
 
        nb_rx = 0;
        nb_hold = 0;
@@ -1192,7 +1253,7 @@ ixgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                 * frames to its peer(s).
                 */
                PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_id=%u "
-                          "ext_err_stat=0x%08x pkt_len=%u\n",
+                          "ext_err_stat=0x%08x pkt_len=%u",
                           (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
                           (unsigned) rx_id, (unsigned) staterr,
                           (unsigned) rte_le_to_cpu_16(rxd.wb.upper.length));
@@ -1200,7 +1261,7 @@ ixgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                nmb = rte_rxmbuf_alloc(rxq->mb_pool);
                if (nmb == NULL) {
                        PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
-                                  "queue_id=%u\n", (unsigned) rxq->port_id,
+                                  "queue_id=%u", (unsigned) rxq->port_id,
                                   (unsigned) rxq->queue_id);
                        rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed++;
                        break;
@@ -1247,8 +1308,8 @@ ixgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                 */
                pkt_len = (uint16_t) (rte_le_to_cpu_16(rxd.wb.upper.length) -
                                      rxq->crc_len);
-               rxm->data = (char*) rxm->buf_addr + RTE_PKTMBUF_HEADROOM;
-               rte_packet_prefetch(rxm->data);
+               rxm->data_off = RTE_PKTMBUF_HEADROOM;
+               rte_packet_prefetch((char *)rxm->buf_addr + rxm->data_off);
                rxm->nb_segs = 1;
                rxm->next = NULL;
                rxm->pkt_len = pkt_len;
@@ -1257,14 +1318,11 @@ ixgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
 
                hlen_type_rss = rte_le_to_cpu_32(rxd.wb.lower.lo_dword.data);
                /* Only valid if PKT_RX_VLAN_PKT set in pkt_flags */
-               rxm->vlan_macip.f.vlan_tci =
-                       rte_le_to_cpu_16(rxd.wb.upper.vlan);
+               rxm->vlan_tci = rte_le_to_cpu_16(rxd.wb.upper.vlan);
 
                pkt_flags = rx_desc_hlen_type_rss_to_pkt_flags(hlen_type_rss);
-               pkt_flags = (uint16_t)(pkt_flags |
-                               rx_desc_status_to_pkt_flags(staterr));
-               pkt_flags = (uint16_t)(pkt_flags |
-                               rx_desc_error_to_pkt_flags(staterr));
+               pkt_flags = pkt_flags | rx_desc_status_to_pkt_flags(staterr);
+               pkt_flags = pkt_flags | rx_desc_error_to_pkt_flags(staterr);
                rxm->ol_flags = pkt_flags;
 
                if (likely(pkt_flags & PKT_RX_RSS_HASH))
@@ -1295,7 +1353,7 @@ ixgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
        nb_hold = (uint16_t) (nb_hold + rxq->nb_rx_hold);
        if (nb_hold > rxq->rx_free_thresh) {
                PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
-                          "nb_hold=%u nb_rx=%u\n",
+                          "nb_hold=%u nb_rx=%u",
                           (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
                           (unsigned) rx_id, (unsigned) nb_hold,
                           (unsigned) nb_rx);
@@ -1329,7 +1387,7 @@ ixgbe_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
        uint16_t nb_rx;
        uint16_t nb_hold;
        uint16_t data_len;
-       uint16_t pkt_flags;
+       uint64_t pkt_flags;
 
        nb_rx = 0;
        nb_hold = 0;
@@ -1382,8 +1440,8 @@ ixgbe_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                 * to happen by sending specific "back-pressure" flow control
                 * frames to its peer(s).
                 */
-               PMD_RX_LOG(DEBUG, "\nport_id=%u queue_id=%u rx_id=%u "
-                          "staterr=0x%x data_len=%u\n",
+               PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_id=%u "
+                          "staterr=0x%x data_len=%u",
                           (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
                           (unsigned) rx_id, (unsigned) staterr,
                           (unsigned) rte_le_to_cpu_16(rxd.wb.upper.length));
@@ -1391,7 +1449,7 @@ ixgbe_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                nmb = rte_rxmbuf_alloc(rxq->mb_pool);
                if (nmb == NULL) {
                        PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
-                                  "queue_id=%u\n", (unsigned) rxq->port_id,
+                                  "queue_id=%u", (unsigned) rxq->port_id,
                                   (unsigned) rxq->queue_id);
                        rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed++;
                        break;
@@ -1431,7 +1489,7 @@ ixgbe_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                 */
                data_len = rte_le_to_cpu_16(rxd.wb.upper.length);
                rxm->data_len = data_len;
-               rxm->data = (char*) rxm->buf_addr + RTE_PKTMBUF_HEADROOM;
+               rxm->data_off = RTE_PKTMBUF_HEADROOM;
 
                /*
                 * If this is the first buffer of the received packet,
@@ -1502,13 +1560,12 @@ ixgbe_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                 * The vlan_tci field is only valid when PKT_RX_VLAN_PKT is
                 * set in the pkt_flags field.
                 */
-               first_seg->vlan_macip.f.vlan_tci =
-                               rte_le_to_cpu_16(rxd.wb.upper.vlan);
+               first_seg->vlan_tci = rte_le_to_cpu_16(rxd.wb.upper.vlan);
                hlen_type_rss = rte_le_to_cpu_32(rxd.wb.lower.lo_dword.data);
                pkt_flags = rx_desc_hlen_type_rss_to_pkt_flags(hlen_type_rss);
-               pkt_flags = (uint16_t)(pkt_flags |
+               pkt_flags = (pkt_flags |
                                rx_desc_status_to_pkt_flags(staterr));
-               pkt_flags = (uint16_t)(pkt_flags |
+               pkt_flags = (pkt_flags |
                                rx_desc_error_to_pkt_flags(staterr));
                first_seg->ol_flags = pkt_flags;
 
@@ -1523,7 +1580,8 @@ ixgbe_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                }
 
                /* Prefetch data of first segment, if configured to do so. */
-               rte_packet_prefetch(first_seg->data);
+               rte_packet_prefetch((char *)first_seg->buf_addr +
+                       first_seg->data_off);
 
                /*
                 * Store the mbuf address into the next entry of the array
@@ -1560,7 +1618,7 @@ ixgbe_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
        nb_hold = (uint16_t) (nb_hold + rxq->nb_rx_hold);
        if (nb_hold > rxq->rx_free_thresh) {
                PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
-                          "nb_hold=%u nb_rx=%u\n",
+                          "nb_hold=%u nb_rx=%u",
                           (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
                           (unsigned) rx_id, (unsigned) nb_hold,
                           (unsigned) nb_rx);
@@ -1764,33 +1822,36 @@ ixgbe_dev_tx_queue_setup(struct rte_eth_dev *dev,
        tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
                        tx_conf->tx_free_thresh : DEFAULT_TX_FREE_THRESH);
        if (tx_rs_thresh >= (nb_desc - 2)) {
-               RTE_LOG(ERR, PMD, "tx_rs_thresh must be less than the number "
-                       "of TX descriptors minus 2. (tx_rs_thresh=%u port=%d "
-                               "queue=%d)\n", (unsigned int)tx_rs_thresh,
-                               (int)dev->data->port_id, (int)queue_idx);
+               PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the number "
+                            "of TX descriptors minus 2. (tx_rs_thresh=%u "
+                            "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
+                            (int)dev->data->port_id, (int)queue_idx);
                return -(EINVAL);
        }
        if (tx_free_thresh >= (nb_desc - 3)) {
-               RTE_LOG(ERR, PMD, "tx_rs_thresh must be less than the "
-                       "tx_free_thresh must be less than the number of TX "
-                       "descriptors minus 3. (tx_free_thresh=%u port=%d "
-                               "queue=%d)\n", (unsigned int)tx_free_thresh,
-                               (int)dev->data->port_id, (int)queue_idx);
+               PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the "
+                            "tx_free_thresh must be less than the number of "
+                            "TX descriptors minus 3. (tx_free_thresh=%u "
+                            "port=%d queue=%d)",
+                            (unsigned int)tx_free_thresh,
+                            (int)dev->data->port_id, (int)queue_idx);
                return -(EINVAL);
        }
        if (tx_rs_thresh > tx_free_thresh) {
-               RTE_LOG(ERR, PMD, "tx_rs_thresh must be less than or equal to "
-                       "tx_free_thresh. (tx_free_thresh=%u tx_rs_thresh=%u "
-                       "port=%d queue=%d)\n", (unsigned int)tx_free_thresh,
-                       (unsigned int)tx_rs_thresh, (int)dev->data->port_id,
-                                                       (int)queue_idx);
+               PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than or equal to "
+                            "tx_free_thresh. (tx_free_thresh=%u "
+                            "tx_rs_thresh=%u port=%d queue=%d)",
+                            (unsigned int)tx_free_thresh,
+                            (unsigned int)tx_rs_thresh,
+                            (int)dev->data->port_id,
+                            (int)queue_idx);
                return -(EINVAL);
        }
        if ((nb_desc % tx_rs_thresh) != 0) {
-               RTE_LOG(ERR, PMD, "tx_rs_thresh must be a divisor of the "
-                       "number of TX descriptors. (tx_rs_thresh=%u port=%d "
-                               "queue=%d)\n", (unsigned int)tx_rs_thresh,
-                               (int)dev->data->port_id, (int)queue_idx);
+               PMD_INIT_LOG(ERR, "tx_rs_thresh must be a divisor of the "
+                            "number of TX descriptors. (tx_rs_thresh=%u "
+                            "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
+                            (int)dev->data->port_id, (int)queue_idx);
                return -(EINVAL);
        }
 
@@ -1801,10 +1862,10 @@ ixgbe_dev_tx_queue_setup(struct rte_eth_dev *dev,
         * accumulates WTHRESH descriptors.
         */
        if ((tx_rs_thresh > 1) && (tx_conf->tx_thresh.wthresh != 0)) {
-               RTE_LOG(ERR, PMD, "TX WTHRESH must be set to 0 if "
-                       "tx_rs_thresh is greater than 1. (tx_rs_thresh=%u "
-                       "port=%d queue=%d)\n", (unsigned int)tx_rs_thresh,
-                               (int)dev->data->port_id, (int)queue_idx);
+               PMD_INIT_LOG(ERR, "TX WTHRESH must be set to 0 if "
+                            "tx_rs_thresh is greater than 1. (tx_rs_thresh=%u "
+                            "port=%d queue=%d)", (unsigned int)tx_rs_thresh,
+                            (int)dev->data->port_id, (int)queue_idx);
                return -(EINVAL);
        }
 
@@ -1816,7 +1877,7 @@ ixgbe_dev_tx_queue_setup(struct rte_eth_dev *dev,
 
        /* First allocate the tx queue data structure */
        txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct igb_tx_queue),
-                                CACHE_LINE_SIZE, socket_id);
+                                RTE_CACHE_LINE_SIZE, socket_id);
        if (txq == NULL)
                return (-ENOMEM);
 
@@ -1845,7 +1906,7 @@ ixgbe_dev_tx_queue_setup(struct rte_eth_dev *dev,
        txq->port_id = dev->data->port_id;
        txq->txq_flags = tx_conf->txq_flags;
        txq->ops = &def_txq_ops;
-       txq->start_tx_per_q = tx_conf->start_tx_per_q;
+       txq->tx_deferred_start = tx_conf->tx_deferred_start;
 
        /*
         * Modification to set VFTDT for virtual function if vf is detected
@@ -1864,31 +1925,37 @@ ixgbe_dev_tx_queue_setup(struct rte_eth_dev *dev,
        /* Allocate software ring */
        txq->sw_ring = rte_zmalloc_socket("txq->sw_ring",
                                sizeof(struct igb_tx_entry) * nb_desc,
-                               CACHE_LINE_SIZE, socket_id);
+                               RTE_CACHE_LINE_SIZE, socket_id);
        if (txq->sw_ring == NULL) {
                ixgbe_tx_queue_release(txq);
                return (-ENOMEM);
        }
-       PMD_INIT_LOG(DEBUG, "sw_ring=%p hw_ring=%p dma_addr=0x%"PRIx64"\n",
+       PMD_INIT_LOG(DEBUG, "sw_ring=%p hw_ring=%p dma_addr=0x%"PRIx64,
                     txq->sw_ring, txq->tx_ring, txq->tx_ring_phys_addr);
 
        /* Use a simple Tx queue (no offloads, no multi segs) if possible */
        if (((txq->txq_flags & IXGBE_SIMPLE_FLAGS) == IXGBE_SIMPLE_FLAGS) &&
            (txq->tx_rs_thresh >= RTE_PMD_IXGBE_TX_MAX_BURST)) {
-               PMD_INIT_LOG(INFO, "Using simple tx code path\n");
+               PMD_INIT_LOG(INFO, "Using simple tx code path");
 #ifdef RTE_IXGBE_INC_VECTOR
                if (txq->tx_rs_thresh <= RTE_IXGBE_TX_MAX_FREE_BUF_SZ &&
-                   ixgbe_txq_vec_setup(txq, socket_id) == 0) {
-                       PMD_INIT_LOG(INFO, "Vector tx enabled.\n");
+                   ixgbe_txq_vec_setup(txq) == 0) {
+                       PMD_INIT_LOG(INFO, "Vector tx enabled.");
                        dev->tx_pkt_burst = ixgbe_xmit_pkts_vec;
                }
                else
 #endif
                        dev->tx_pkt_burst = ixgbe_xmit_pkts_simple;
        } else {
-               PMD_INIT_LOG(INFO, "Using full-featured tx code path\n");
-               PMD_INIT_LOG(INFO, " - txq_flags = %lx [IXGBE_SIMPLE_FLAGS=%lx]\n", (long unsigned)txq->txq_flags, (long unsigned)IXGBE_SIMPLE_FLAGS);
-               PMD_INIT_LOG(INFO, " - tx_rs_thresh = %lu [RTE_PMD_IXGBE_TX_MAX_BURST=%lu]\n", (long unsigned)txq->tx_rs_thresh, (long unsigned)RTE_PMD_IXGBE_TX_MAX_BURST);
+               PMD_INIT_LOG(INFO, "Using full-featured tx code path");
+               PMD_INIT_LOG(INFO, " - txq_flags = %lx "
+                            "[IXGBE_SIMPLE_FLAGS=%lx]",
+                            (long unsigned)txq->txq_flags,
+                            (long unsigned)IXGBE_SIMPLE_FLAGS);
+               PMD_INIT_LOG(INFO, " - tx_rs_thresh = %lu "
+                            "[RTE_PMD_IXGBE_TX_MAX_BURST=%lu]",
+                            (long unsigned)txq->tx_rs_thresh,
+                            (long unsigned)RTE_PMD_IXGBE_TX_MAX_BURST);
                dev->tx_pkt_burst = ixgbe_xmit_pkts;
        }
 
@@ -1968,15 +2035,34 @@ check_rx_burst_bulk_alloc_preconditions(__rte_unused struct igb_rx_queue *rxq)
         * outside of this function.
         */
 #ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
-       if (! (rxq->rx_free_thresh >= RTE_PMD_IXGBE_RX_MAX_BURST))
+       if (!(rxq->rx_free_thresh >= RTE_PMD_IXGBE_RX_MAX_BURST)) {
+               PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
+                            "rxq->rx_free_thresh=%d, "
+                            "RTE_PMD_IXGBE_RX_MAX_BURST=%d",
+                            rxq->rx_free_thresh, RTE_PMD_IXGBE_RX_MAX_BURST);
                ret = -EINVAL;
-       else if (! (rxq->rx_free_thresh < rxq->nb_rx_desc))
+       } else if (!(rxq->rx_free_thresh < rxq->nb_rx_desc)) {
+               PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
+                            "rxq->rx_free_thresh=%d, "
+                            "rxq->nb_rx_desc=%d",
+                            rxq->rx_free_thresh, rxq->nb_rx_desc);
                ret = -EINVAL;
-       else if (! ((rxq->nb_rx_desc % rxq->rx_free_thresh) == 0))
+       } else if (!((rxq->nb_rx_desc % rxq->rx_free_thresh) == 0)) {
+               PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
+                            "rxq->nb_rx_desc=%d, "
+                            "rxq->rx_free_thresh=%d",
+                            rxq->nb_rx_desc, rxq->rx_free_thresh);
                ret = -EINVAL;
-       else if (! (rxq->nb_rx_desc <
-              (IXGBE_MAX_RING_DESC - RTE_PMD_IXGBE_RX_MAX_BURST)))
+       } else if (!(rxq->nb_rx_desc <
+              (IXGBE_MAX_RING_DESC - RTE_PMD_IXGBE_RX_MAX_BURST))) {
+               PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
+                            "rxq->nb_rx_desc=%d, "
+                            "IXGBE_MAX_RING_DESC=%d, "
+                            "RTE_PMD_IXGBE_RX_MAX_BURST=%d",
+                            rxq->nb_rx_desc, IXGBE_MAX_RING_DESC,
+                            RTE_PMD_IXGBE_RX_MAX_BURST);
                ret = -EINVAL;
+       }
 #else
        ret = -EINVAL;
 #endif
@@ -2077,7 +2163,7 @@ ixgbe_dev_rx_queue_setup(struct rte_eth_dev *dev,
 
        /* First allocate the rx queue data structure */
        rxq = rte_zmalloc_socket("ethdev RX queue", sizeof(struct igb_rx_queue),
-                                CACHE_LINE_SIZE, socket_id);
+                                RTE_CACHE_LINE_SIZE, socket_id);
        if (rxq == NULL)
                return (-ENOMEM);
        rxq->mb_pool = mp;
@@ -2090,7 +2176,7 @@ ixgbe_dev_rx_queue_setup(struct rte_eth_dev *dev,
        rxq->crc_len = (uint8_t) ((dev->data->dev_conf.rxmode.hw_strip_crc) ?
                                                        0 : ETHER_CRC_LEN);
        rxq->drop_en = rx_conf->rx_drop_en;
-       rxq->start_rx_per_q = rx_conf->start_rx_per_q;
+       rxq->rx_deferred_start = rx_conf->rx_deferred_start;
 
        /*
         * Allocate RX ring hardware descriptors. A memzone large enough to
@@ -2143,12 +2229,12 @@ ixgbe_dev_rx_queue_setup(struct rte_eth_dev *dev,
 #endif
        rxq->sw_ring = rte_zmalloc_socket("rxq->sw_ring",
                                          sizeof(struct igb_rx_entry) * len,
-                                         CACHE_LINE_SIZE, socket_id);
+                                         RTE_CACHE_LINE_SIZE, socket_id);
        if (rxq->sw_ring == NULL) {
                ixgbe_rx_queue_release(rxq);
                return (-ENOMEM);
        }
-       PMD_INIT_LOG(DEBUG, "sw_ring=%p hw_ring=%p dma_addr=0x%"PRIx64"\n",
+       PMD_INIT_LOG(DEBUG, "sw_ring=%p hw_ring=%p dma_addr=0x%"PRIx64,
                     rxq->sw_ring, rxq->rx_ring, rxq->rx_ring_phys_addr);
 
        /*
@@ -2157,19 +2243,21 @@ ixgbe_dev_rx_queue_setup(struct rte_eth_dev *dev,
         */
        use_def_burst_func = check_rx_burst_bulk_alloc_preconditions(rxq);
 
+#ifdef RTE_IXGBE_INC_VECTOR
+       ixgbe_rxq_vec_setup(rxq);
+#endif
        /* Check if pre-conditions are satisfied, and no Scattered Rx */
        if (!use_def_burst_func && !dev->data->scattered_rx) {
 #ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
                PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
                             "satisfied. Rx Burst Bulk Alloc function will be "
-                            "used on port=%d, queue=%d.\n",
+                            "used on port=%d, queue=%d.",
                             rxq->port_id, rxq->queue_id);
                dev->rx_pkt_burst = ixgbe_recv_pkts_bulk_alloc;
 #ifdef RTE_IXGBE_INC_VECTOR
                if (!ixgbe_rx_vec_condition_check(dev)) {
                        PMD_INIT_LOG(INFO, "Vector rx enabled, please make "
-                                    "sure RX burst size no less than 32.\n");
-                       ixgbe_rxq_vec_setup(rxq, socket_id);
+                                    "sure RX burst size no less than 32.");
                        dev->rx_pkt_burst = ixgbe_recv_pkts_vec;
                }
 #endif
@@ -2178,7 +2266,7 @@ ixgbe_dev_rx_queue_setup(struct rte_eth_dev *dev,
                PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions "
                             "are not satisfied, Scattered Rx is requested, "
                             "or RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC is not "
-                            "enabled (port=%d, queue=%d).\n",
+                            "enabled (port=%d, queue=%d).",
                             rxq->port_id, rxq->queue_id);
        }
        dev->data->rx_queues[queue_idx] = rxq;
@@ -2197,7 +2285,7 @@ ixgbe_dev_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
        uint32_t desc = 0;
 
        if (rx_queue_id >= dev->data->nb_rx_queues) {
-               PMD_RX_LOG(ERR, "Invalid RX queue id=%d\n", rx_queue_id);
+               PMD_RX_LOG(ERR, "Invalid RX queue id=%d", rx_queue_id);
                return 0;
        }
 
@@ -2835,6 +2923,8 @@ ixgbe_dcb_hw_arbite_rx_config(struct ixgbe_hw *hw, uint16_t *refill,
                break;
        case ixgbe_mac_82599EB:
        case ixgbe_mac_X540:
+       case ixgbe_mac_X550:
+       case ixgbe_mac_X550EM_x:
                ixgbe_dcb_config_rx_arbiter_82599(hw, refill, max, bwg_id,
                                                  tsa, map);
                break;
@@ -2854,6 +2944,8 @@ ixgbe_dcb_hw_arbite_tx_config(struct ixgbe_hw *hw, uint16_t *refill, uint16_t *m
                break;
        case ixgbe_mac_82599EB:
        case ixgbe_mac_X540:
+       case ixgbe_mac_X550:
+       case ixgbe_mac_X550EM_x:
                ixgbe_dcb_config_tx_desc_arbiter_82599(hw, refill, max, bwg_id,tsa);
                ixgbe_dcb_config_tx_data_arbiter_82599(hw, refill, max, bwg_id,tsa, map);
                break;
@@ -2913,7 +3005,7 @@ ixgbe_dcb_hw_configure(struct rte_eth_dev *dev,
                ixgbe_dcb_rx_hw_config(hw, dcb_config);
                break;
        default:
-               PMD_INIT_LOG(ERR, "Incorrect DCB RX mode configuration\n");
+               PMD_INIT_LOG(ERR, "Incorrect DCB RX mode configuration");
                break;
        }
        switch (dev->data->dev_conf.txmode.mq_mode) {
@@ -2935,7 +3027,7 @@ ixgbe_dcb_hw_configure(struct rte_eth_dev *dev,
                ixgbe_dcb_tx_hw_config(hw, dcb_config);
                break;
        default:
-               PMD_INIT_LOG(ERR, "Incorrect DCB TX mode configuration\n");
+               PMD_INIT_LOG(ERR, "Incorrect DCB TX mode configuration");
                break;
        }
 
@@ -3083,6 +3175,7 @@ ixgbe_vmdq_rx_hw_configure(struct rte_eth_dev *dev)
        struct ixgbe_hw *hw;
        enum rte_eth_nb_pools num_pools;
        uint32_t mrqc, vt_ctl, vlanctrl;
+       uint32_t vmolr = 0;
        int i;
 
        PMD_INIT_FUNC_TRACE();
@@ -3105,6 +3198,11 @@ ixgbe_vmdq_rx_hw_configure(struct rte_eth_dev *dev)
 
        IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vt_ctl);
 
+       for (i = 0; i < (int)num_pools; i++) {
+               vmolr = ixgbe_convert_vm_rx_mask_to_val(cfg->rx_mode, vmolr);
+               IXGBE_WRITE_REG(hw, IXGBE_VMOLR(i), vmolr);
+       }
+
        /* VLNCTRL: enable vlan filtering and allow all vlan tags through */
        vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
        vlanctrl |= IXGBE_VLNCTRL_VFE ; /* enable vlan filters */
@@ -3206,14 +3304,14 @@ ixgbe_alloc_rx_queue_mbufs(struct igb_rx_queue *rxq)
                volatile union ixgbe_adv_rx_desc *rxd;
                struct rte_mbuf *mbuf = rte_rxmbuf_alloc(rxq->mb_pool);
                if (mbuf == NULL) {
-                       PMD_INIT_LOG(ERR, "RX mbuf alloc failed queue_id=%u\n",
+                       PMD_INIT_LOG(ERR, "RX mbuf alloc failed queue_id=%u",
                                     (unsigned) rxq->queue_id);
                        return (-ENOMEM);
                }
 
                rte_mbuf_refcnt_set(mbuf, 1);
                mbuf->next = NULL;
-               mbuf->data = (char *)mbuf->buf_addr + RTE_PKTMBUF_HEADROOM;
+               mbuf->data_off = RTE_PKTMBUF_HEADROOM;
                mbuf->nb_segs = 1;
                mbuf->port = rxq->port_id;
 
@@ -3277,7 +3375,7 @@ ixgbe_dev_mq_rx_configure(struct rte_eth_dev *dev)
                        IXGBE_WRITE_REG(hw, IXGBE_MRQC, IXGBE_MRQC_VMDQRT8TCEN);
                        break;
                default:
-                       RTE_LOG(ERR, PMD, "invalid pool number in IOV mode\n");
+                       PMD_INIT_LOG(ERR, "invalid pool number in IOV mode");
                }
        }
 
@@ -3330,7 +3428,7 @@ ixgbe_dev_mq_tx_configure(struct rte_eth_dev *dev)
                        break;
                default:
                        mtqc = IXGBE_MTQC_64Q_1PB;
-                       RTE_LOG(ERR, PMD, "invalid pool number in IOV mode\n");
+                       PMD_INIT_LOG(ERR, "invalid pool number in IOV mode");
                }
                IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
        }
@@ -3480,13 +3578,25 @@ ixgbe_dev_rx_init(struct rte_eth_dev *dev)
                /* It adds dual VLAN length for supporting dual VLAN */
                if ((dev->data->dev_conf.rxmode.max_rx_pkt_len +
                                2 * IXGBE_VLAN_TAG_SIZE) > buf_size){
+                       if (!dev->data->scattered_rx)
+                               PMD_INIT_LOG(DEBUG, "forcing scatter mode");
                        dev->data->scattered_rx = 1;
+#ifdef RTE_IXGBE_INC_VECTOR
+                       dev->rx_pkt_burst = ixgbe_recv_scattered_pkts_vec;
+#else
                        dev->rx_pkt_burst = ixgbe_recv_scattered_pkts;
+#endif
                }
        }
 
        if (dev->data->dev_conf.rxmode.enable_scatter) {
+               if (!dev->data->scattered_rx)
+                       PMD_INIT_LOG(DEBUG, "forcing scatter mode");
+#ifdef RTE_IXGBE_INC_VECTOR
+               dev->rx_pkt_burst = ixgbe_recv_scattered_pkts_vec;
+#else
                dev->rx_pkt_burst = ixgbe_recv_scattered_pkts;
+#endif
                dev->data->scattered_rx = 1;
        }
 
@@ -3538,9 +3648,10 @@ ixgbe_dev_tx_init(struct rte_eth_dev *dev)
        PMD_INIT_FUNC_TRACE();
        hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 
-       /* Enable TX CRC (checksum offload requirement) */
+       /* Enable TX CRC (checksum offload requirement) and hw padding
+        * (TSO requirement) */
        hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
-       hlreg0 |= IXGBE_HLREG0_TXCRCEN;
+       hlreg0 |= (IXGBE_HLREG0_TXCRCEN | IXGBE_HLREG0_TXPADEN);
        IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
 
        /* Setup the Base and Length of the Tx Descriptor Rings */
@@ -3573,6 +3684,8 @@ ixgbe_dev_tx_init(struct rte_eth_dev *dev)
 
                        case ixgbe_mac_82599EB:
                        case ixgbe_mac_X540:
+                       case ixgbe_mac_X550:
+                       case ixgbe_mac_X550EM_x:
                        default:
                                txctrl = IXGBE_READ_REG(hw,
                                                IXGBE_DCA_TXCTRL_82599(txq->reg_idx));
@@ -3593,12 +3706,12 @@ ixgbe_dev_tx_init(struct rte_eth_dev *dev)
 static inline void
 ixgbe_setup_loopback_link_82599(struct ixgbe_hw *hw)
 {
-       DEBUGFUNC("ixgbe_setup_loopback_link_82599");
+       PMD_INIT_FUNC_TRACE();
 
        if (ixgbe_verify_lesm_fw_enabled_82599(hw)) {
                if (hw->mac.ops.acquire_swfw_sync(hw, IXGBE_GSSR_MAC_CSR_SM) !=
                                IXGBE_SUCCESS) {
-                       PMD_INIT_LOG(ERR, "Could not enable loopback mode\n");
+                       PMD_INIT_LOG(ERR, "Could not enable loopback mode");
                        /* ignore error */
                        return;
                }
@@ -3650,13 +3763,13 @@ ixgbe_dev_rxtx_start(struct rte_eth_dev *dev)
 
        for (i = 0; i < dev->data->nb_tx_queues; i++) {
                txq = dev->data->tx_queues[i];
-               if (!txq->start_tx_per_q)
+               if (!txq->tx_deferred_start)
                        ixgbe_dev_tx_queue_start(dev, i);
        }
 
        for (i = 0; i < dev->data->nb_rx_queues; i++) {
                rxq = dev->data->rx_queues[i];
-               if (!rxq->start_rx_per_q)
+               if (!rxq->rx_deferred_start)
                        ixgbe_dev_rx_queue_start(dev, i);
        }
 
@@ -3693,9 +3806,8 @@ ixgbe_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 
                /* Allocate buffers for descriptor rings */
                if (ixgbe_alloc_rx_queue_mbufs(rxq) != 0) {
-                       PMD_INIT_LOG(ERR,
-                               "Could not alloc mbuf for queue:%d\n",
-                               rx_queue_id);
+                       PMD_INIT_LOG(ERR, "Could not alloc mbuf for queue:%d",
+                                    rx_queue_id);
                        return -1;
                }
                rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
@@ -3709,8 +3821,8 @@ ixgbe_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
                        rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
                } while (--poll_ms && !(rxdctl & IXGBE_RXDCTL_ENABLE));
                if (!poll_ms)
-                       PMD_INIT_LOG(ERR, "Could not enable "
-                                    "Rx Queue %d\n", rx_queue_id);
+                       PMD_INIT_LOG(ERR, "Could not enable Rx Queue %d",
+                                    rx_queue_id);
                rte_wmb();
                IXGBE_WRITE_REG(hw, IXGBE_RDH(rxq->reg_idx), 0);
                IXGBE_WRITE_REG(hw, IXGBE_RDT(rxq->reg_idx), rxq->nb_rx_desc - 1);
@@ -3748,8 +3860,8 @@ ixgbe_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
                        rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxq->reg_idx));
                } while (--poll_ms && (rxdctl | IXGBE_RXDCTL_ENABLE));
                if (!poll_ms)
-                       PMD_INIT_LOG(ERR, "Could not disable "
-                                    "Rx Queue %d\n", rx_queue_id);
+                       PMD_INIT_LOG(ERR, "Could not disable Rx Queue %d",
+                                    rx_queue_id);
 
                rte_delay_us(RTE_IXGBE_WAIT_100_US);
 
@@ -3792,7 +3904,7 @@ ixgbe_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
                        } while (--poll_ms && !(txdctl & IXGBE_TXDCTL_ENABLE));
                        if (!poll_ms)
                                PMD_INIT_LOG(ERR, "Could not enable "
-                                            "Tx Queue %d\n", tx_queue_id);
+                                            "Tx Queue %d", tx_queue_id);
                }
                rte_wmb();
                IXGBE_WRITE_REG(hw, IXGBE_TDH(txq->reg_idx), 0);
@@ -3832,9 +3944,8 @@ ixgbe_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
                                                IXGBE_TDT(txq->reg_idx));
                        } while (--poll_ms && (txtdh != txtdt));
                        if (!poll_ms)
-                               PMD_INIT_LOG(ERR,
-                               "Tx Queue %d is not empty when stopping.\n",
-                               tx_queue_id);
+                               PMD_INIT_LOG(ERR, "Tx Queue %d is not empty "
+                                            "when stopping.", tx_queue_id);
                }
 
                txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txq->reg_idx));
@@ -3851,7 +3962,7 @@ ixgbe_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
                        } while (--poll_ms && (txdctl | IXGBE_TXDCTL_ENABLE));
                        if (!poll_ms)
                                PMD_INIT_LOG(ERR, "Could not disable "
-                                            "Tx Queue %d\n", tx_queue_id);
+                                            "Tx Queue %d", tx_queue_id);
                }
 
                if (txq->ops != NULL) {
@@ -3973,13 +4084,25 @@ ixgbevf_dev_rx_init(struct rte_eth_dev *dev)
                /* It adds dual VLAN length for supporting dual VLAN */
                if ((dev->data->dev_conf.rxmode.max_rx_pkt_len +
                                2 * IXGBE_VLAN_TAG_SIZE) > buf_size) {
+                       if (!dev->data->scattered_rx)
+                               PMD_INIT_LOG(DEBUG, "forcing scatter mode");
                        dev->data->scattered_rx = 1;
+#ifdef RTE_IXGBE_INC_VECTOR
+                       dev->rx_pkt_burst = ixgbe_recv_scattered_pkts_vec;
+#else
                        dev->rx_pkt_burst = ixgbe_recv_scattered_pkts;
+#endif
                }
        }
 
        if (dev->data->dev_conf.rxmode.enable_scatter) {
+               if (!dev->data->scattered_rx)
+                       PMD_INIT_LOG(DEBUG, "forcing scatter mode");
+#ifdef RTE_IXGBE_INC_VECTOR
+               dev->rx_pkt_burst = ixgbe_recv_scattered_pkts_vec;
+#else
                dev->rx_pkt_burst = ixgbe_recv_scattered_pkts;
+#endif
                dev->data->scattered_rx = 1;
        }
 
@@ -4067,8 +4190,7 @@ ixgbevf_dev_rxtx_start(struct rte_eth_dev *dev)
                        txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(i));
                } while (--poll_ms && !(txdctl & IXGBE_TXDCTL_ENABLE));
                if (!poll_ms)
-                       PMD_INIT_LOG(ERR, "Could not enable "
-                                        "Tx Queue %d\n", i);
+                       PMD_INIT_LOG(ERR, "Could not enable Tx Queue %d", i);
        }
        for (i = 0; i < dev->data->nb_rx_queues; i++) {
 
@@ -4085,8 +4207,7 @@ ixgbevf_dev_rxtx_start(struct rte_eth_dev *dev)
                        rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(i));
                } while (--poll_ms && !(rxdctl & IXGBE_RXDCTL_ENABLE));
                if (!poll_ms)
-                       PMD_INIT_LOG(ERR, "Could not enable "
-                                        "Rx Queue %d\n", i);
+                       PMD_INIT_LOG(ERR, "Could not enable Rx Queue %d", i);
                rte_wmb();
                IXGBE_WRITE_REG(hw, IXGBE_VFRDT(i), rxq->nb_rx_desc - 1);