tailq: remove unneeded inclusions
[dpdk.git] / lib / librte_pmd_ixgbe / ixgbe_rxtx.c
index 7c04977..1a68f9e 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
@@ -52,7 +53,6 @@
 #include <rte_memory.h>
 #include <rte_memzone.h>
 #include <rte_launch.h>
-#include <rte_tailq.h>
 #include <rte_eal.h>
 #include <rte_per_lcore.h>
 #include <rte_lcore.h>
 #include "ixgbe/ixgbe_common.h"
 #include "ixgbe_rxtx.h"
 
-#define IXGBE_RSS_OFFLOAD_ALL ( \
-               ETH_RSS_IPV4 | \
-               ETH_RSS_IPV4_TCP | \
-               ETH_RSS_IPV6 | \
-               ETH_RSS_IPV6_EX | \
-               ETH_RSS_IPV6_TCP | \
-               ETH_RSS_IPV6_TCP_EX | \
-               ETH_RSS_IPV4_UDP | \
-               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)
@@ -145,6 +141,7 @@ ixgbe_tx_free_bufs(struct igb_tx_queue *txq)
        /* 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;
                }
@@ -356,59 +353,84 @@ 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,
-               uint64_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;
@@ -420,20 +442,20 @@ ixgbe_set_xmit_ctx(struct igb_tx_queue* txq,
  */
 static inline uint32_t
 what_advctx_update(struct igb_tx_queue *txq, uint64_t flags,
-               uint32_t vlan_macip_lens)
+               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;
        }
 
@@ -444,20 +466,25 @@ what_advctx_update(struct igb_tx_queue *txq, uint64_t flags,
 static inline uint32_t
 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(uint64_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 */
@@ -538,7 +565,6 @@ ixgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
        volatile union ixgbe_adv_tx_desc *txd;
        struct rte_mbuf     *tx_pkt;
        struct rte_mbuf     *m_seg;
-       union ixgbe_vlan_macip vlan_macip_lens;
        uint64_t buf_dma_addr;
        uint32_t olinfo_status;
        uint32_t cmd_type_len;
@@ -552,6 +578,7 @@ ixgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
        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;
@@ -564,28 +591,32 @@ 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.f.vlan_tci = tx_pkt->vlan_tci;
-               vlan_macip_lens.f.l2_l3_len = tx_pkt->l2_l3_len;
 
                /* If hardware offload required */
-               tx_ol_req = 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.data);
+                               tx_offload);
                        /* Only allocate context descriptor if required*/
                        new_ctx = (ctx == IXGBE_CTX_NUM);
                        ctx = txq->ctx_curr;
@@ -700,13 +731,22 @@ 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
                         */
@@ -719,7 +759,7 @@ 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);
@@ -727,7 +767,7 @@ ixgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                                }
 
                                ixgbe_set_xmit_ctx(txq, ctx_txd, tx_ol_req,
-                                   vlan_macip_lens.data);
+                                       tx_offload);
 
                                txe->last_id = tx_last;
                                tx_id = txe->next_id;
@@ -739,15 +779,18 @@ 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);
@@ -814,7 +857,7 @@ end_of_tx:
 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 uint64_t ip_pkt_types_map[16] = {
                0, PKT_RX_IPV4_HDR, PKT_RX_IPV4_HDR_EXT, PKT_RX_IPV4_HDR_EXT,
@@ -831,7 +874,7 @@ 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,
        };
@@ -900,6 +943,7 @@ 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;
 
@@ -933,21 +977,28 @@ 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_tci = rxdp[j].wb.upper.vlan;
-                       mb->hash.rss = rxdp[j].wb.lower.hi_dword.rss;
+                       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 = mb->ol_flags |
-                                       rx_desc_status_to_pkt_flags(s[j]);
-                       mb->ol_flags = 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 */
@@ -1324,7 +1375,7 @@ 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;
@@ -1500,9 +1551,9 @@ ixgbe_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                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;
 
@@ -1708,6 +1759,40 @@ static struct ixgbe_txq_ops def_txq_ops = {
        .reset = ixgbe_reset_tx_queue,
 };
 
+/* Takes an ethdev and a queue and sets up the tx function to be used based on
+ * the queue parameters. Used in tx_queue_setup by primary process and then
+ * in dev_init by secondary process when attaching to an existing ethdev.
+ */
+void
+set_tx_function(struct rte_eth_dev *dev, struct igb_tx_queue *txq)
+{
+       /* 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");
+#ifdef RTE_IXGBE_INC_VECTOR
+               if (txq->tx_rs_thresh <= RTE_IXGBE_TX_MAX_FREE_BUF_SZ &&
+                               (rte_eal_process_type() != RTE_PROC_PRIMARY ||
+                                       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");
+               PMD_INIT_LOG(INFO,
+                               " - txq_flags = %lx " "[IXGBE_SIMPLE_FLAGS=%lx]",
+                               (unsigned long)txq->txq_flags,
+                               (unsigned long)IXGBE_SIMPLE_FLAGS);
+               PMD_INIT_LOG(INFO,
+                               " - tx_rs_thresh = %lu " "[RTE_PMD_IXGBE_TX_MAX_BURST=%lu]",
+                               (unsigned long)txq->tx_rs_thresh,
+                               (unsigned long)RTE_PMD_IXGBE_TX_MAX_BURST);
+               dev->tx_pkt_burst = ixgbe_xmit_pkts;
+       }
+}
+
 int
 ixgbe_dev_tx_queue_setup(struct rte_eth_dev *dev,
                         uint16_t queue_idx,
@@ -1814,7 +1899,7 @@ 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);
 
@@ -1843,12 +1928,15 @@ 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
         */
-       if (hw->mac.type == ixgbe_mac_82599_vf)
+       if (hw->mac.type == ixgbe_mac_82599_vf ||
+           hw->mac.type == ixgbe_mac_X540_vf ||
+           hw->mac.type == ixgbe_mac_X550_vf ||
+           hw->mac.type == ixgbe_mac_X550EM_x_vf)
                txq->tdt_reg_addr = IXGBE_PCI_REG_ADDR(hw, IXGBE_VFTDT(queue_idx));
        else
                txq->tdt_reg_addr = IXGBE_PCI_REG_ADDR(hw, IXGBE_TDT(txq->reg_idx));
@@ -1862,7 +1950,7 @@ 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);
@@ -1870,31 +1958,8 @@ ixgbe_dev_tx_queue_setup(struct rte_eth_dev *dev,
        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");
-#ifdef RTE_IXGBE_INC_VECTOR
-               if (txq->tx_rs_thresh <= RTE_IXGBE_TX_MAX_FREE_BUF_SZ &&
-                   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");
-               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;
-       }
+       /* set up vector or scalar TX function as appropriate */
+       set_tx_function(dev, txq);
 
        txq->ops->reset(txq);
 
@@ -1972,15 +2037,34 @@ 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
@@ -2081,7 +2165,7 @@ 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;
@@ -2094,7 +2178,7 @@ 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
@@ -2116,7 +2200,10 @@ ixgbe_dev_rx_queue_setup(struct rte_eth_dev *dev,
        /*
         * Modified to setup VFRDT for Virtual Function
         */
-       if (hw->mac.type == ixgbe_mac_82599_vf) {
+       if (hw->mac.type == ixgbe_mac_82599_vf ||
+           hw->mac.type == ixgbe_mac_X540_vf ||
+           hw->mac.type == ixgbe_mac_X550_vf ||
+           hw->mac.type == ixgbe_mac_X550EM_x_vf) {
                rxq->rdt_reg_addr =
                        IXGBE_PCI_REG_ADDR(hw, IXGBE_VFRDT(queue_idx));
                rxq->rdh_reg_addr =
@@ -2147,7 +2234,7 @@ 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);
@@ -2161,6 +2248,9 @@ 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
@@ -2170,10 +2260,10 @@ ixgbe_dev_rx_queue_setup(struct rte_eth_dev *dev,
                             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)) {
+               if (!ixgbe_rx_vec_condition_check(dev) &&
+                   (rte_is_power_of_2(nb_desc))) {
                        PMD_INIT_LOG(INFO, "Vector rx enabled, please make "
                                     "sure RX burst size no less than 32.");
-                       ixgbe_rxq_vec_setup(rxq);
                        dev->rx_pkt_burst = ixgbe_recv_pkts_vec;
                }
 #endif
@@ -2335,19 +2425,19 @@ ixgbe_hw_rss_hash_set(struct ixgbe_hw *hw, struct rte_eth_rss_conf *rss_conf)
        mrqc = IXGBE_MRQC_RSSEN; /* Enable RSS */
        if (rss_hf & ETH_RSS_IPV4)
                mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4;
-       if (rss_hf & ETH_RSS_IPV4_TCP)
+       if (rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
                mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_TCP;
        if (rss_hf & ETH_RSS_IPV6)
                mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6;
        if (rss_hf & ETH_RSS_IPV6_EX)
                mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX;
-       if (rss_hf & ETH_RSS_IPV6_TCP)
+       if (rss_hf & ETH_RSS_NONFRAG_IPV6_TCP)
                mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
        if (rss_hf & ETH_RSS_IPV6_TCP_EX)
                mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP;
-       if (rss_hf & ETH_RSS_IPV4_UDP)
+       if (rss_hf & ETH_RSS_NONFRAG_IPV4_UDP)
                mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
-       if (rss_hf & ETH_RSS_IPV6_UDP)
+       if (rss_hf & ETH_RSS_NONFRAG_IPV6_UDP)
                mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
        if (rss_hf & ETH_RSS_IPV6_UDP_EX)
                mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
@@ -2421,19 +2511,19 @@ ixgbe_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
        if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4)
                rss_hf |= ETH_RSS_IPV4;
        if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4_TCP)
-               rss_hf |= ETH_RSS_IPV4_TCP;
+               rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
        if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6)
                rss_hf |= ETH_RSS_IPV6;
        if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX)
                rss_hf |= ETH_RSS_IPV6_EX;
        if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_TCP)
-               rss_hf |= ETH_RSS_IPV6_TCP;
+               rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP;
        if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP)
                rss_hf |= ETH_RSS_IPV6_TCP_EX;
        if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV4_UDP)
-               rss_hf |= ETH_RSS_IPV4_UDP;
+               rss_hf |= ETH_RSS_NONFRAG_IPV4_UDP;
        if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_UDP)
-               rss_hf |= ETH_RSS_IPV6_UDP;
+               rss_hf |= ETH_RSS_NONFRAG_IPV6_UDP;
        if (mrqc & IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP)
                rss_hf |= ETH_RSS_IPV6_UDP_EX;
        rss_conf->rss_hf = rss_hf;
@@ -2839,6 +2929,8 @@ 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;
@@ -2858,6 +2950,8 @@ 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;
@@ -3087,6 +3181,7 @@ 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();
@@ -3109,6 +3204,11 @@ 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 */
@@ -3232,6 +3332,67 @@ ixgbe_alloc_rx_queue_mbufs(struct igb_rx_queue *rxq)
        return 0;
 }
 
+static int
+ixgbe_config_vf_rss(struct rte_eth_dev *dev)
+{
+       struct ixgbe_hw *hw;
+       uint32_t mrqc;
+
+       ixgbe_rss_configure(dev);
+
+       hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       /* MRQC: enable VF RSS */
+       mrqc = IXGBE_READ_REG(hw, IXGBE_MRQC);
+       mrqc &= ~IXGBE_MRQC_MRQE_MASK;
+       switch (RTE_ETH_DEV_SRIOV(dev).active) {
+       case ETH_64_POOLS:
+               mrqc |= IXGBE_MRQC_VMDQRSS64EN;
+               break;
+
+       case ETH_32_POOLS:
+               mrqc |= IXGBE_MRQC_VMDQRSS32EN;
+               break;
+
+       default:
+               PMD_INIT_LOG(ERR, "Invalid pool number in IOV mode with VMDQ RSS");
+               return -EINVAL;
+       }
+
+       IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
+
+       return 0;
+}
+
+static int
+ixgbe_config_vf_default(struct rte_eth_dev *dev)
+{
+       struct ixgbe_hw *hw =
+               IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       switch (RTE_ETH_DEV_SRIOV(dev).active) {
+       case ETH_64_POOLS:
+               IXGBE_WRITE_REG(hw, IXGBE_MRQC,
+                       IXGBE_MRQC_VMDQEN);
+               break;
+
+       case ETH_32_POOLS:
+               IXGBE_WRITE_REG(hw, IXGBE_MRQC,
+                       IXGBE_MRQC_VMDQRT4TCEN);
+               break;
+
+       case ETH_16_POOLS:
+               IXGBE_WRITE_REG(hw, IXGBE_MRQC,
+                       IXGBE_MRQC_VMDQRT8TCEN);
+               break;
+       default:
+               PMD_INIT_LOG(ERR,
+                       "invalid pool number in IOV mode");
+               break;
+       }
+       return 0;
+}
+
 static int
 ixgbe_dev_mq_rx_configure(struct rte_eth_dev *dev)
 {
@@ -3264,24 +3425,25 @@ ixgbe_dev_mq_rx_configure(struct rte_eth_dev *dev)
                        default: ixgbe_rss_disable(dev);
                }
        } else {
-               switch (RTE_ETH_DEV_SRIOV(dev).active) {
                /*
                 * SRIOV active scheme
-                * FIXME if support DCB/RSS together with VMDq & SRIOV
+                * Support RSS together with VMDq & SRIOV
                 */
-               case ETH_64_POOLS:
-                       IXGBE_WRITE_REG(hw, IXGBE_MRQC, IXGBE_MRQC_VMDQEN);
-                       break;
-
-               case ETH_32_POOLS:
-                       IXGBE_WRITE_REG(hw, IXGBE_MRQC, IXGBE_MRQC_VMDQRT4TCEN);
+               switch (dev->data->dev_conf.rxmode.mq_mode) {
+               case ETH_MQ_RX_RSS:
+               case ETH_MQ_RX_VMDQ_RSS:
+                       ixgbe_config_vf_rss(dev);
                        break;
 
-               case ETH_16_POOLS:
-                       IXGBE_WRITE_REG(hw, IXGBE_MRQC, IXGBE_MRQC_VMDQRT8TCEN);
-                       break;
+               /* FIXME if support DCB/RSS together with VMDq & SRIOV */
+               case ETH_MQ_RX_VMDQ_DCB:
+               case ETH_MQ_RX_VMDQ_DCB_RSS:
+                       PMD_INIT_LOG(ERR,
+                               "Could not support DCB with VMDq & SRIOV");
+                       return -1;
                default:
-                       PMD_INIT_LOG(ERR, "invalid pool number in IOV mode");
+                       ixgbe_config_vf_default(dev);
+                       break;
                }
        }
 
@@ -3454,9 +3616,9 @@ ixgbe_dev_rx_init(struct rte_eth_dev *dev)
                                IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx), psrtype);
                        }
                        srrctl = ((dev->data->dev_conf.rxmode.split_hdr_size <<
-                                  IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
-                                 IXGBE_SRRCTL_BSIZEHDR_MASK);
-                       srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
+                               IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
+                               IXGBE_SRRCTL_BSIZEHDR_MASK);
+                       srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
                } else
 #endif
                        srrctl = IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
@@ -3481,31 +3643,23 @@ ixgbe_dev_rx_init(struct rte_eth_dev *dev)
                buf_size = (uint16_t) ((srrctl & IXGBE_SRRCTL_BSIZEPKT_MASK) <<
                                       IXGBE_SRRCTL_BSIZEPKT_SHIFT);
 
-               /* It adds dual VLAN length for supporting dual VLAN */
-               if ((dev->data->dev_conf.rxmode.max_rx_pkt_len +
+               if (dev->data->dev_conf.rxmode.enable_scatter ||
+                   /* It adds dual VLAN length for supporting dual VLAN */
+                   (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;
+                       if (rte_is_power_of_2(rxq->nb_rx_desc))
+                               dev->rx_pkt_burst =
+                                       ixgbe_recv_scattered_pkts_vec;
+                       else
 #endif
+                               dev->rx_pkt_burst = ixgbe_recv_scattered_pkts;
                }
        }
 
-       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;
-       }
-
        /*
         * Device configured with multiple RX queues.
         */
@@ -3554,9 +3708,10 @@ 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 */
@@ -3589,6 +3744,8 @@ 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));
@@ -3634,7 +3791,7 @@ ixgbe_setup_loopback_link_82599(struct ixgbe_hw *hw)
 /*
  * Start Transmit and Receive Units.
  */
-void
+int
 ixgbe_dev_rxtx_start(struct rte_eth_dev *dev)
 {
        struct ixgbe_hw     *hw;
@@ -3644,6 +3801,7 @@ ixgbe_dev_rxtx_start(struct rte_eth_dev *dev)
        uint32_t dmatxctl;
        uint32_t rxctrl;
        uint16_t i;
+       int ret = 0;
 
        PMD_INIT_FUNC_TRACE();
        hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
@@ -3666,14 +3824,20 @@ 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)
-                       ixgbe_dev_tx_queue_start(dev, i);
+               if (!txq->tx_deferred_start) {
+                       ret = ixgbe_dev_tx_queue_start(dev, i);
+                       if (ret < 0)
+                               return ret;
+               }
        }
 
        for (i = 0; i < dev->data->nb_rx_queues; i++) {
                rxq = dev->data->rx_queues[i];
-               if (!rxq->start_rx_per_q)
-                       ixgbe_dev_rx_queue_start(dev, i);
+               if (!rxq->rx_deferred_start) {
+                       ret = ixgbe_dev_rx_queue_start(dev, i);
+                       if (ret < 0)
+                               return ret;
+               }
        }
 
        /* Enable Receive engine */
@@ -3688,6 +3852,7 @@ ixgbe_dev_rxtx_start(struct rte_eth_dev *dev)
                        dev->data->dev_conf.lpbk_mode == IXGBE_LPBK_82599_TX_RX)
                ixgbe_setup_loopback_link_82599(hw);
 
+       return 0;
 }
 
 /*
@@ -3888,7 +4053,7 @@ ixgbevf_dev_rx_init(struct rte_eth_dev *dev)
        struct igb_rx_queue *rxq;
        struct rte_pktmbuf_pool_private *mbp_priv;
        uint64_t bus_addr;
-       uint32_t srrctl;
+       uint32_t srrctl, psrtype = 0;
        uint16_t buf_size;
        uint16_t i;
        int ret;
@@ -3896,6 +4061,19 @@ ixgbevf_dev_rx_init(struct rte_eth_dev *dev)
        PMD_INIT_FUNC_TRACE();
        hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 
+       if (rte_is_power_of_2(dev->data->nb_rx_queues) == 0) {
+               PMD_INIT_LOG(ERR, "The number of Rx queue invalid, "
+                       "it should be power of 2");
+               return -1;
+       }
+
+       if (dev->data->nb_rx_queues > hw->mac.max_rx_queues) {
+               PMD_INIT_LOG(ERR, "The number of Rx queue invalid, "
+                       "it should be equal to or less than %d",
+                       hw->mac.max_rx_queues);
+               return -1;
+       }
+
        /*
         * When the VF driver issues a IXGBE_VF_RESET request, the PF driver
         * disables the VF receipt of packets if the PF MTU is > 1500.
@@ -3942,20 +4120,10 @@ ixgbevf_dev_rx_init(struct rte_eth_dev *dev)
                 * Configure Header Split
                 */
                if (dev->data->dev_conf.rxmode.header_split) {
-
-                       /* Must setup the PSRTYPE register */
-                       uint32_t psrtype;
-                       psrtype = IXGBE_PSRTYPE_TCPHDR |
-                               IXGBE_PSRTYPE_UDPHDR   |
-                               IXGBE_PSRTYPE_IPV4HDR  |
-                               IXGBE_PSRTYPE_IPV6HDR;
-
-                       IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE(i), psrtype);
-
                        srrctl = ((dev->data->dev_conf.rxmode.split_hdr_size <<
-                                  IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
-                                 IXGBE_SRRCTL_BSIZEHDR_MASK);
-                       srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
+                               IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
+                               IXGBE_SRRCTL_BSIZEHDR_MASK);
+                       srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
                } else
 #endif
                        srrctl = IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
@@ -3984,30 +4152,36 @@ ixgbevf_dev_rx_init(struct rte_eth_dev *dev)
                buf_size = (uint16_t) ((srrctl & IXGBE_SRRCTL_BSIZEPKT_MASK) <<
                                       IXGBE_SRRCTL_BSIZEPKT_SHIFT);
 
-               /* It adds dual VLAN length for supporting dual VLAN */
-               if ((dev->data->dev_conf.rxmode.max_rx_pkt_len +
+               if (dev->data->dev_conf.rxmode.enable_scatter ||
+                   /* It adds dual VLAN length for supporting dual VLAN */
+                   (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;
+                       if (rte_is_power_of_2(rxq->nb_rx_desc))
+                               dev->rx_pkt_burst =
+                                       ixgbe_recv_scattered_pkts_vec;
+                       else
 #endif
+                               dev->rx_pkt_burst = ixgbe_recv_scattered_pkts;
                }
        }
 
-       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;
+#ifdef RTE_HEADER_SPLIT_ENABLE
+       if (dev->data->dev_conf.rxmode.header_split)
+               /* Must setup the PSRTYPE register */
+               psrtype = IXGBE_PSRTYPE_TCPHDR |
+                       IXGBE_PSRTYPE_UDPHDR   |
+                       IXGBE_PSRTYPE_IPV4HDR  |
+                       IXGBE_PSRTYPE_IPV6HDR;
 #endif
-               dev->data->scattered_rx = 1;
-       }
+
+       /* Set RQPL for VF RSS according to max Rx queue */
+       psrtype |= (dev->data->nb_rx_queues >> 1) <<
+               IXGBE_PSRTYPE_RQPL_SHIFT;
+       IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
 
        return 0;
 }