mbuf: add accessors for data room and private size
[dpdk.git] / lib / librte_pmd_ixgbe / ixgbe_rxtx.c
index 3a5a8ff..7f15f15 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>
@@ -70,6 +70,7 @@
 #include <rte_sctp.h>
 #include <rte_string_fns.h>
 #include <rte_errno.h>
+#include <rte_ip.h>
 
 #include "ixgbe_logs.h"
 #include "ixgbe/ixgbe_api.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)
@@ -125,9 +122,9 @@ rte_rxmbuf_alloc(struct rte_mempool *mp)
  * Return the total number of buffers freed.
  */
 static inline int __attribute__((always_inline))
-ixgbe_tx_free_bufs(struct igb_tx_queue *txq)
+ixgbe_tx_free_bufs(struct ixgbe_tx_queue *txq)
 {
-       struct igb_tx_entry *txep;
+       struct ixgbe_tx_entry *txep;
        uint32_t status;
        int i;
 
@@ -211,11 +208,11 @@ tx1(volatile union ixgbe_adv_tx_desc *txdp, struct rte_mbuf **pkts)
  * Copy mbuf pointers to the S/W ring.
  */
 static inline void
-ixgbe_tx_fill_hw_ring(struct igb_tx_queue *txq, struct rte_mbuf **pkts,
+ixgbe_tx_fill_hw_ring(struct ixgbe_tx_queue *txq, struct rte_mbuf **pkts,
                      uint16_t nb_pkts)
 {
        volatile union ixgbe_adv_tx_desc *txdp = &(txq->tx_ring[txq->tx_tail]);
-       struct igb_tx_entry *txep = &(txq->sw_ring[txq->tx_tail]);
+       struct ixgbe_tx_entry *txep = &(txq->sw_ring[txq->tx_tail]);
        const int N_PER_LOOP = 4;
        const int N_PER_LOOP_MASK = N_PER_LOOP-1;
        int mainpart, leftover;
@@ -247,7 +244,7 @@ static inline uint16_t
 tx_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
             uint16_t nb_pkts)
 {
-       struct igb_tx_queue *txq = (struct igb_tx_queue *)tx_queue;
+       struct ixgbe_tx_queue *txq = (struct ixgbe_tx_queue *)tx_queue;
        volatile union ixgbe_adv_tx_desc *tx_r = txq->tx_ring;
        uint16_t n = 0;
 
@@ -355,61 +352,86 @@ ixgbe_xmit_pkts_simple(void *tx_queue, struct rte_mbuf **tx_pkts,
 }
 
 static inline void
-ixgbe_set_xmit_ctx(struct igb_tx_queue* txq,
+ixgbe_set_xmit_ctx(struct ixgbe_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;
@@ -420,21 +442,21 @@ 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, uint64_t flags,
-               uint32_t vlan_macip_lens)
+what_advctx_update(struct ixgbe_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;
        }
 
@@ -445,20 +467,25 @@ 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 */
@@ -471,9 +498,9 @@ tx_desc_vlan_flags_to_cmdtype(uint64_t ol_flags)
 
 /* Reset transmit descriptors after they have been used */
 static inline int
-ixgbe_xmit_cleanup(struct igb_tx_queue *txq)
+ixgbe_xmit_cleanup(struct ixgbe_tx_queue *txq)
 {
-       struct igb_tx_entry *sw_ring = txq->sw_ring;
+       struct ixgbe_tx_entry *sw_ring = txq->sw_ring;
        volatile union ixgbe_adv_tx_desc *txr = txq->tx_ring;
        uint16_t last_desc_cleaned = txq->last_desc_cleaned;
        uint16_t nb_tx_desc = txq->nb_tx_desc;
@@ -532,14 +559,13 @@ uint16_t
 ixgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                uint16_t nb_pkts)
 {
-       struct igb_tx_queue *txq;
-       struct igb_tx_entry *sw_ring;
-       struct igb_tx_entry *txe, *txn;
+       struct ixgbe_tx_queue *txq;
+       struct ixgbe_tx_entry *sw_ring;
+       struct ixgbe_tx_entry *txe, *txn;
        volatile union ixgbe_adv_tx_desc *txr;
        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;
@@ -553,6 +579,7 @@ 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 = {0};
 
        txq = tx_queue;
        sw_ring = txq->sw_ring;
@@ -580,14 +607,17 @@ ixgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                ol_flags = tx_pkt->ol_flags;
 
                /* 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) {
-                       vlan_macip_lens.f.vlan_tci = tx_pkt->vlan_tci;
-                       vlan_macip_lens.f.l2_l3_len = tx_pkt->l2_l3_len;
+                       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;
@@ -702,13 +732,22 @@ 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
                         */
@@ -729,7 +768,7 @@ 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;
@@ -741,11 +780,13 @@ 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];
@@ -817,16 +858,16 @@ 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] = {
+       static const 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 uint64_t ip_rss_types_map[16] = {
+       static const 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,
@@ -834,7 +875,7 @@ 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,
        };
@@ -897,13 +938,13 @@ rx_desc_error_to_pkt_flags(uint32_t rx_status)
 #error "PMD IXGBE: LOOK_AHEAD must be 8\n"
 #endif
 static inline int
-ixgbe_rx_scan_hw_ring(struct igb_rx_queue *rxq)
+ixgbe_rx_scan_hw_ring(struct ixgbe_rx_queue *rxq)
 {
        volatile union ixgbe_adv_rx_desc *rxdp;
-       struct igb_rx_entry *rxep;
+       struct ixgbe_rx_entry *rxep;
        struct rte_mbuf *mb;
        uint16_t pkt_len;
-       uint16_t pkt_flags;
+       uint64_t pkt_flags;
        int s[LOOK_AHEAD], nb_dd;
        int i, j, nb_rx = 0;
 
@@ -981,18 +1022,17 @@ ixgbe_rx_scan_hw_ring(struct igb_rx_queue *rxq)
 }
 
 static inline int
-ixgbe_rx_alloc_bufs(struct igb_rx_queue *rxq)
+ixgbe_rx_alloc_bufs(struct ixgbe_rx_queue *rxq, bool reset_mbuf)
 {
        volatile union ixgbe_adv_rx_desc *rxdp;
-       struct igb_rx_entry *rxep;
+       struct ixgbe_rx_entry *rxep;
        struct rte_mbuf *mb;
        uint16_t alloc_idx;
-       uint64_t dma_addr;
+       __le64 dma_addr;
        int diag, i;
 
        /* allocate buffers in bulk directly into the S/W ring */
-       alloc_idx = (uint16_t)(rxq->rx_free_trigger -
-                               (rxq->rx_free_thresh - 1));
+       alloc_idx = rxq->rx_free_trigger - (rxq->rx_free_thresh - 1);
        rxep = &rxq->sw_ring[alloc_idx];
        diag = rte_mempool_get_bulk(rxq->mb_pool, (void *)rxep,
                                    rxq->rx_free_thresh);
@@ -1003,34 +1043,32 @@ ixgbe_rx_alloc_bufs(struct igb_rx_queue *rxq)
        for (i = 0; i < rxq->rx_free_thresh; ++i) {
                /* populate the static rte mbuf fields */
                mb = rxep[i].mbuf;
+               if (reset_mbuf) {
+                       mb->next = NULL;
+                       mb->nb_segs = 1;
+                       mb->port = rxq->port_id;
+               }
+
                rte_mbuf_refcnt_set(mb, 1);
-               mb->next = NULL;
                mb->data_off = RTE_PKTMBUF_HEADROOM;
-               mb->nb_segs = 1;
-               mb->port = rxq->port_id;
 
                /* populate the descriptors */
-               dma_addr = (uint64_t)mb->buf_physaddr + RTE_PKTMBUF_HEADROOM;
+               dma_addr = rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mb));
                rxdp[i].read.hdr_addr = dma_addr;
                rxdp[i].read.pkt_addr = dma_addr;
        }
 
-       /* update tail pointer */
-       rte_wmb();
-       IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, rxq->rx_free_trigger);
-
        /* update state of internal queue structure */
-       rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_trigger +
-                                               rxq->rx_free_thresh);
+       rxq->rx_free_trigger = rxq->rx_free_trigger + rxq->rx_free_thresh;
        if (rxq->rx_free_trigger >= rxq->nb_rx_desc)
-               rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
+               rxq->rx_free_trigger = rxq->rx_free_thresh - 1;
 
        /* no errors */
        return 0;
 }
 
 static inline uint16_t
-ixgbe_rx_fill_from_stage(struct igb_rx_queue *rxq, struct rte_mbuf **rx_pkts,
+ixgbe_rx_fill_from_stage(struct ixgbe_rx_queue *rxq, struct rte_mbuf **rx_pkts,
                         uint16_t nb_pkts)
 {
        struct rte_mbuf **stage = &rxq->rx_stage[rxq->rx_next_avail];
@@ -1054,7 +1092,7 @@ static inline uint16_t
 rx_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
             uint16_t nb_pkts)
 {
-       struct igb_rx_queue *rxq = (struct igb_rx_queue *)rx_queue;
+       struct ixgbe_rx_queue *rxq = (struct ixgbe_rx_queue *)rx_queue;
        uint16_t nb_rx = 0;
 
        /* Any previously recv'd pkts will be returned from the Rx stage */
@@ -1071,7 +1109,9 @@ rx_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
 
        /* if required, allocate new buffers to replenish descriptors */
        if (rxq->rx_tail > rxq->rx_free_trigger) {
-               if (ixgbe_rx_alloc_bufs(rxq) != 0) {
+               uint16_t cur_free_trigger = rxq->rx_free_trigger;
+
+               if (ixgbe_rx_alloc_bufs(rxq, true) != 0) {
                        int i, j;
                        PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
                                   "queue_id=%u", (unsigned) rxq->port_id,
@@ -1091,6 +1131,10 @@ rx_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
 
                        return 0;
                }
+
+               /* update tail pointer */
+               rte_wmb();
+               IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, cur_free_trigger);
        }
 
        if (rxq->rx_tail >= rxq->nb_rx_desc)
@@ -1104,7 +1148,7 @@ rx_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
 }
 
 /* split requests into chunks of size RTE_PMD_IXGBE_RX_MAX_BURST */
-uint16_t
+static uint16_t
 ixgbe_recv_pkts_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
                           uint16_t nb_pkts)
 {
@@ -1130,17 +1174,34 @@ ixgbe_recv_pkts_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
 
        return nb_rx;
 }
+
+#else
+
+/* Stub to avoid extra ifdefs */
+static uint16_t
+ixgbe_recv_pkts_bulk_alloc(__rte_unused void *rx_queue,
+       __rte_unused struct rte_mbuf **rx_pkts, __rte_unused uint16_t nb_pkts)
+{
+       return 0;
+}
+
+static inline int
+ixgbe_rx_alloc_bufs(__rte_unused struct ixgbe_rx_queue *rxq,
+                   __rte_unused bool reset_mbuf)
+{
+       return -ENOMEM;
+}
 #endif /* RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC */
 
 uint16_t
 ixgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                uint16_t nb_pkts)
 {
-       struct igb_rx_queue *rxq;
+       struct ixgbe_rx_queue *rxq;
        volatile union ixgbe_adv_rx_desc *rx_ring;
        volatile union ixgbe_adv_rx_desc *rxdp;
-       struct igb_rx_entry *sw_ring;
-       struct igb_rx_entry *rxe;
+       struct ixgbe_rx_entry *sw_ring;
+       struct ixgbe_rx_entry *rxe;
        struct rte_mbuf *rxm;
        struct rte_mbuf *nmb;
        union ixgbe_adv_rx_desc rxd;
@@ -1314,15 +1375,362 @@ ixgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
        return (nb_rx);
 }
 
+/**
+ * Detect an RSC descriptor.
+ */
+static inline uint32_t
+ixgbe_rsc_count(union ixgbe_adv_rx_desc *rx)
+{
+       return (rte_le_to_cpu_32(rx->wb.lower.lo_dword.data) &
+               IXGBE_RXDADV_RSCCNT_MASK) >> IXGBE_RXDADV_RSCCNT_SHIFT;
+}
+
+/**
+ * ixgbe_fill_cluster_head_buf - fill the first mbuf of the returned packet
+ *
+ * Fill the following info in the HEAD buffer of the Rx cluster:
+ *    - RX port identifier
+ *    - hardware offload data, if any:
+ *      - RSS flag & hash
+ *      - IP checksum flag
+ *      - VLAN TCI, if any
+ *      - error flags
+ * @head HEAD of the packet cluster
+ * @desc HW descriptor to get data from
+ * @port_id Port ID of the Rx queue
+ */
+static inline void
+ixgbe_fill_cluster_head_buf(
+       struct rte_mbuf *head,
+       union ixgbe_adv_rx_desc *desc,
+       uint8_t port_id,
+       uint32_t staterr)
+{
+       uint32_t hlen_type_rss;
+       uint64_t pkt_flags;
+
+       head->port = port_id;
+
+       /*
+        * The vlan_tci field is only valid when PKT_RX_VLAN_PKT is
+        * set in the pkt_flags field.
+        */
+       head->vlan_tci = rte_le_to_cpu_16(desc->wb.upper.vlan);
+       hlen_type_rss = rte_le_to_cpu_32(desc->wb.lower.lo_dword.data);
+       pkt_flags = rx_desc_hlen_type_rss_to_pkt_flags(hlen_type_rss);
+       pkt_flags |= rx_desc_status_to_pkt_flags(staterr);
+       pkt_flags |= rx_desc_error_to_pkt_flags(staterr);
+       head->ol_flags = pkt_flags;
+
+       if (likely(pkt_flags & PKT_RX_RSS_HASH))
+               head->hash.rss = rte_le_to_cpu_32(desc->wb.lower.hi_dword.rss);
+       else if (pkt_flags & PKT_RX_FDIR) {
+               head->hash.fdir.hash =
+                       rte_le_to_cpu_16(desc->wb.lower.hi_dword.csum_ip.csum)
+                                                         & IXGBE_ATR_HASH_MASK;
+               head->hash.fdir.id =
+                       rte_le_to_cpu_16(desc->wb.lower.hi_dword.csum_ip.ip_id);
+       }
+}
+
+/**
+ * ixgbe_recv_pkts_lro - receive handler for and LRO case.
+ *
+ * @rx_queue Rx queue handle
+ * @rx_pkts table of received packets
+ * @nb_pkts size of rx_pkts table
+ * @bulk_alloc if TRUE bulk allocation is used for a HW ring refilling
+ *
+ * Handles the Rx HW ring completions when RSC feature is configured. Uses an
+ * additional ring of ixgbe_rsc_entry's that will hold the relevant RSC info.
+ *
+ * We use the same logic as in Linux and in FreeBSD ixgbe drivers:
+ * 1) When non-EOP RSC completion arrives:
+ *    a) Update the HEAD of the current RSC aggregation cluster with the new
+ *       segment's data length.
+ *    b) Set the "next" pointer of the current segment to point to the segment
+ *       at the NEXTP index.
+ *    c) Pass the HEAD of RSC aggregation cluster on to the next NEXTP entry
+ *       in the sw_rsc_ring.
+ * 2) When EOP arrives we just update the cluster's total length and offload
+ *    flags and deliver the cluster up to the upper layers. In our case - put it
+ *    in the rx_pkts table.
+ *
+ * Returns the number of received packets/clusters (according to the "bulk
+ * receive" interface).
+ */
+static inline uint16_t
+ixgbe_recv_pkts_lro(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts,
+                   bool bulk_alloc)
+{
+       struct ixgbe_rx_queue *rxq = rx_queue;
+       volatile union ixgbe_adv_rx_desc *rx_ring = rxq->rx_ring;
+       struct ixgbe_rx_entry *sw_ring = rxq->sw_ring;
+       struct ixgbe_rsc_entry *sw_rsc_ring = rxq->sw_rsc_ring;
+       uint16_t rx_id = rxq->rx_tail;
+       uint16_t nb_rx = 0;
+       uint16_t nb_hold = rxq->nb_rx_hold;
+       uint16_t prev_id = rxq->rx_tail;
+
+       while (nb_rx < nb_pkts) {
+               bool eop;
+               struct ixgbe_rx_entry *rxe;
+               struct ixgbe_rsc_entry *rsc_entry;
+               struct ixgbe_rsc_entry *next_rsc_entry;
+               struct ixgbe_rx_entry *next_rxe;
+               struct rte_mbuf *first_seg;
+               struct rte_mbuf *rxm;
+               struct rte_mbuf *nmb;
+               union ixgbe_adv_rx_desc rxd;
+               uint16_t data_len;
+               uint16_t next_id;
+               volatile union ixgbe_adv_rx_desc *rxdp;
+               uint32_t staterr;
+
+next_desc:
+               /*
+                * The code in this whole file uses the volatile pointer to
+                * ensure the read ordering of the status and the rest of the
+                * descriptor fields (on the compiler level only!!!). This is so
+                * UGLY - why not to just use the compiler barrier instead? DPDK
+                * even has the rte_compiler_barrier() for that.
+                *
+                * But most importantly this is just wrong because this doesn't
+                * ensure memory ordering in a general case at all. For
+                * instance, DPDK is supposed to work on Power CPUs where
+                * compiler barrier may just not be enough!
+                *
+                * I tried to write only this function properly to have a
+                * starting point (as a part of an LRO/RSC series) but the
+                * compiler cursed at me when I tried to cast away the
+                * "volatile" from rx_ring (yes, it's volatile too!!!). So, I'm
+                * keeping it the way it is for now.
+                *
+                * The code in this file is broken in so many other places and
+                * will just not work on a big endian CPU anyway therefore the
+                * lines below will have to be revisited together with the rest
+                * of the ixgbe PMD.
+                *
+                * TODO:
+                *    - Get rid of "volatile" crap and let the compiler do its
+                *      job.
+                *    - Use the proper memory barrier (rte_rmb()) to ensure the
+                *      memory ordering below.
+                */
+               rxdp = &rx_ring[rx_id];
+               staterr = rte_le_to_cpu_32(rxdp->wb.upper.status_error);
+
+               if (!(staterr & IXGBE_RXDADV_STAT_DD))
+                       break;
+
+               rxd = *rxdp;
+
+               PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_id=%u "
+                                 "staterr=0x%x data_len=%u",
+                          rxq->port_id, rxq->queue_id, rx_id, staterr,
+                          rte_le_to_cpu_16(rxd.wb.upper.length));
+
+               if (!bulk_alloc) {
+                       nmb = rte_rxmbuf_alloc(rxq->mb_pool);
+                       if (nmb == NULL) {
+                               PMD_RX_LOG(DEBUG, "RX mbuf alloc failed "
+                                                 "port_id=%u queue_id=%u",
+                                          rxq->port_id, rxq->queue_id);
+
+                               rte_eth_devices[rxq->port_id].data->
+                                                       rx_mbuf_alloc_failed++;
+                               break;
+                       }
+               } else if (nb_hold > rxq->rx_free_thresh) {
+                       uint16_t next_rdt = rxq->rx_free_trigger;
+
+                       if (!ixgbe_rx_alloc_bufs(rxq, false)) {
+                               rte_wmb();
+                               IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr,
+                                                   next_rdt);
+                               nb_hold -= rxq->rx_free_thresh;
+                       } else {
+                               PMD_RX_LOG(DEBUG, "RX bulk alloc failed "
+                                                 "port_id=%u queue_id=%u",
+                                          rxq->port_id, rxq->queue_id);
+
+                               rte_eth_devices[rxq->port_id].data->
+                                                       rx_mbuf_alloc_failed++;
+                               break;
+                       }
+               }
+
+               nb_hold++;
+               rxe = &sw_ring[rx_id];
+               eop = staterr & IXGBE_RXDADV_STAT_EOP;
+
+               next_id = rx_id + 1;
+               if (next_id == rxq->nb_rx_desc)
+                       next_id = 0;
+
+               /* Prefetch next mbuf while processing current one. */
+               rte_ixgbe_prefetch(sw_ring[next_id].mbuf);
+
+               /*
+                * When next RX descriptor is on a cache-line boundary,
+                * prefetch the next 4 RX descriptors and the next 4 pointers
+                * to mbufs.
+                */
+               if ((next_id & 0x3) == 0) {
+                       rte_ixgbe_prefetch(&rx_ring[next_id]);
+                       rte_ixgbe_prefetch(&sw_ring[next_id]);
+               }
+
+               rxm = rxe->mbuf;
+
+               if (!bulk_alloc) {
+                       __le64 dma =
+                         rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(nmb));
+                       /*
+                        * Update RX descriptor with the physical address of the
+                        * new data buffer of the new allocated mbuf.
+                        */
+                       rxe->mbuf = nmb;
+
+                       rxm->data_off = RTE_PKTMBUF_HEADROOM;
+                       rxdp->read.hdr_addr = dma;
+                       rxdp->read.pkt_addr = dma;
+               } else
+                       rxe->mbuf = NULL;
+
+               /*
+                * Set data length & data buffer address of mbuf.
+                */
+               data_len = rte_le_to_cpu_16(rxd.wb.upper.length);
+               rxm->data_len = data_len;
+
+               if (!eop) {
+                       uint16_t nextp_id;
+                       /*
+                        * Get next descriptor index:
+                        *  - For RSC it's in the NEXTP field.
+                        *  - For a scattered packet - it's just a following
+                        *    descriptor.
+                        */
+                       if (ixgbe_rsc_count(&rxd))
+                               nextp_id =
+                                       (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
+                                                      IXGBE_RXDADV_NEXTP_SHIFT;
+                       else
+                               nextp_id = next_id;
+
+                       next_rsc_entry = &sw_rsc_ring[nextp_id];
+                       next_rxe = &sw_ring[nextp_id];
+                       rte_ixgbe_prefetch(next_rxe);
+               }
+
+               rsc_entry = &sw_rsc_ring[rx_id];
+               first_seg = rsc_entry->fbuf;
+               rsc_entry->fbuf = NULL;
+
+               /*
+                * If this is the first buffer of the received packet,
+                * set the pointer to the first mbuf of the packet and
+                * initialize its context.
+                * Otherwise, update the total length and the number of segments
+                * of the current scattered packet, and update the pointer to
+                * the last mbuf of the current packet.
+                */
+               if (first_seg == NULL) {
+                       first_seg = rxm;
+                       first_seg->pkt_len = data_len;
+                       first_seg->nb_segs = 1;
+               } else {
+                       first_seg->pkt_len += data_len;
+                       first_seg->nb_segs++;
+               }
+
+               prev_id = rx_id;
+               rx_id = next_id;
+
+               /*
+                * If this is not the last buffer of the received packet, update
+                * the pointer to the first mbuf at the NEXTP entry in the
+                * sw_rsc_ring and continue to parse the RX ring.
+                */
+               if (!eop) {
+                       rxm->next = next_rxe->mbuf;
+                       next_rsc_entry->fbuf = first_seg;
+                       goto next_desc;
+               }
+
+               /*
+                * This is the last buffer of the received packet - return
+                * the current cluster to the user.
+                */
+               rxm->next = NULL;
+
+               /* Initialize the first mbuf of the returned packet */
+               ixgbe_fill_cluster_head_buf(first_seg, &rxd, rxq->port_id,
+                                           staterr);
+
+               /* Prefetch data of first segment, if configured to do so. */
+               rte_packet_prefetch((char *)first_seg->buf_addr +
+                       first_seg->data_off);
+
+               /*
+                * Store the mbuf address into the next entry of the array
+                * of returned packets.
+                */
+               rx_pkts[nb_rx++] = first_seg;
+       }
+
+       /*
+        * Record index of the next RX descriptor to probe.
+        */
+       rxq->rx_tail = rx_id;
+
+       /*
+        * If the number of free RX descriptors is greater than the RX free
+        * threshold of the queue, advance the Receive Descriptor Tail (RDT)
+        * register.
+        * Update the RDT with the value of the last processed RX descriptor
+        * minus 1, to guarantee that the RDT register is never equal to the
+        * RDH register, which creates a "full" ring situtation from the
+        * hardware point of view...
+        */
+       if (!bulk_alloc && 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",
+                          rxq->port_id, rxq->queue_id, rx_id, nb_hold, nb_rx);
+
+               rte_wmb();
+               IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, prev_id);
+               nb_hold = 0;
+       }
+
+       rxq->nb_rx_hold = nb_hold;
+       return nb_rx;
+}
+
+uint16_t
+ixgbe_recv_pkts_lro_single_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
+                                uint16_t nb_pkts)
+{
+       return ixgbe_recv_pkts_lro(rx_queue, rx_pkts, nb_pkts, false);
+}
+
+uint16_t
+ixgbe_recv_pkts_lro_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
+                              uint16_t nb_pkts)
+{
+       return ixgbe_recv_pkts_lro(rx_queue, rx_pkts, nb_pkts, true);
+}
+
 uint16_t
 ixgbe_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                          uint16_t nb_pkts)
 {
-       struct igb_rx_queue *rxq;
+       struct ixgbe_rx_queue *rxq;
        volatile union ixgbe_adv_rx_desc *rx_ring;
        volatile union ixgbe_adv_rx_desc *rxdp;
-       struct igb_rx_entry *sw_ring;
-       struct igb_rx_entry *rxe;
+       struct ixgbe_rx_entry *sw_ring;
+       struct ixgbe_rx_entry *rxe;
        struct rte_mbuf *first_seg;
        struct rte_mbuf *last_seg;
        struct rte_mbuf *rxm;
@@ -1330,12 +1738,10 @@ ixgbe_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
        union ixgbe_adv_rx_desc rxd;
        uint64_t dma; /* Physical address of mbuf data buffer */
        uint32_t staterr;
-       uint32_t hlen_type_rss;
        uint16_t rx_id;
        uint16_t nb_rx;
        uint16_t nb_hold;
        uint16_t data_len;
-       uint16_t pkt_flags;
 
        nb_rx = 0;
        nb_hold = 0;
@@ -1493,39 +1899,9 @@ ixgbe_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                                        (uint16_t) (data_len - ETHER_CRC_LEN);
                }
 
-               /*
-                * Initialize the first mbuf of the returned packet:
-                *    - RX port identifier,
-                *    - hardware offload data, if any:
-                *      - RSS flag & hash,
-                *      - IP checksum flag,
-                *      - VLAN TCI, if any,
-                *      - error flags.
-                */
-               first_seg->port = rxq->port_id;
-
-               /*
-                * The vlan_tci field is only valid when PKT_RX_VLAN_PKT is
-                * set in the pkt_flags field.
-                */
-               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 |
-                               rx_desc_status_to_pkt_flags(staterr));
-               pkt_flags = (uint16_t)(pkt_flags |
-                               rx_desc_error_to_pkt_flags(staterr));
-               first_seg->ol_flags = pkt_flags;
-
-               if (likely(pkt_flags & PKT_RX_RSS_HASH))
-                       first_seg->hash.rss = rxd.wb.lower.hi_dword.rss;
-               else if (pkt_flags & PKT_RX_FDIR) {
-                       first_seg->hash.fdir.hash =
-                               (uint16_t)((rxd.wb.lower.hi_dword.csum_ip.csum)
-                                          & IXGBE_ATR_HASH_MASK);
-                       first_seg->hash.fdir.id =
-                               rxd.wb.lower.hi_dword.csum_ip.ip_id;
-               }
+               /* Initialize the first mbuf of the returned packet */
+               ixgbe_fill_cluster_head_buf(first_seg, &rxd, rxq->port_id,
+                                           staterr);
 
                /* Prefetch data of first segment, if configured to do so. */
                rte_packet_prefetch((char *)first_seg->buf_addr +
@@ -1634,7 +2010,7 @@ ring_dma_zone_reserve(struct rte_eth_dev *dev, const char *ring_name,
 }
 
 static void
-ixgbe_tx_queue_release_mbufs(struct igb_tx_queue *txq)
+ixgbe_tx_queue_release_mbufs(struct ixgbe_tx_queue *txq)
 {
        unsigned i;
 
@@ -1649,7 +2025,7 @@ ixgbe_tx_queue_release_mbufs(struct igb_tx_queue *txq)
 }
 
 static void
-ixgbe_tx_free_swring(struct igb_tx_queue *txq)
+ixgbe_tx_free_swring(struct ixgbe_tx_queue *txq)
 {
        if (txq != NULL &&
            txq->sw_ring != NULL)
@@ -1657,7 +2033,7 @@ ixgbe_tx_free_swring(struct igb_tx_queue *txq)
 }
 
 static void
-ixgbe_tx_queue_release(struct igb_tx_queue *txq)
+ixgbe_tx_queue_release(struct ixgbe_tx_queue *txq)
 {
        if (txq != NULL && txq->ops != NULL) {
                txq->ops->release_mbufs(txq);
@@ -1672,13 +2048,12 @@ ixgbe_dev_tx_queue_release(void *txq)
        ixgbe_tx_queue_release(txq);
 }
 
-/* (Re)set dynamic igb_tx_queue fields to defaults */
+/* (Re)set dynamic ixgbe_tx_queue fields to defaults */
 static void
-ixgbe_reset_tx_queue(struct igb_tx_queue *txq)
+ixgbe_reset_tx_queue(struct ixgbe_tx_queue *txq)
 {
-       static const union ixgbe_adv_tx_desc zeroed_desc = { .read = {
-                       .buffer_addr = 0}};
-       struct igb_tx_entry *txe = txq->sw_ring;
+       static const union ixgbe_adv_tx_desc zeroed_desc = {{0}};
+       struct ixgbe_tx_entry *txe = txq->sw_ring;
        uint16_t prev, i;
 
        /* Zero out HW ring memory */
@@ -1713,12 +2088,46 @@ ixgbe_reset_tx_queue(struct igb_tx_queue *txq)
                IXGBE_CTX_NUM * sizeof(struct ixgbe_advctx_info));
 }
 
-static struct ixgbe_txq_ops def_txq_ops = {
+static const struct ixgbe_txq_ops def_txq_ops = {
        .release_mbufs = ixgbe_tx_queue_release_mbufs,
        .free_swring = ixgbe_tx_free_swring,
        .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
+ixgbe_set_tx_function(struct rte_eth_dev *dev, struct ixgbe_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,
@@ -1727,7 +2136,7 @@ ixgbe_dev_tx_queue_setup(struct rte_eth_dev *dev,
                         const struct rte_eth_txconf *tx_conf)
 {
        const struct rte_memzone *tz;
-       struct igb_tx_queue *txq;
+       struct ixgbe_tx_queue *txq;
        struct ixgbe_hw     *hw;
        uint16_t tx_rs_thresh, tx_free_thresh;
 
@@ -1824,8 +2233,8 @@ 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);
+       txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct ixgbe_tx_queue),
+                                RTE_CACHE_LINE_SIZE, socket_id);
        if (txq == NULL)
                return (-ENOMEM);
 
@@ -1859,7 +2268,10 @@ ixgbe_dev_tx_queue_setup(struct rte_eth_dev *dev,
        /*
         * 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));
@@ -1872,8 +2284,8 @@ 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);
+                               sizeof(struct ixgbe_tx_entry) * nb_desc,
+                               RTE_CACHE_LINE_SIZE, socket_id);
        if (txq->sw_ring == NULL) {
                ixgbe_tx_queue_release(txq);
                return (-ENOMEM);
@@ -1881,31 +2293,8 @@ 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 */
+       ixgbe_set_tx_function(dev, txq);
 
        txq->ops->reset(txq);
 
@@ -1915,8 +2304,33 @@ ixgbe_dev_tx_queue_setup(struct rte_eth_dev *dev,
        return (0);
 }
 
+/**
+ * ixgbe_free_rsc_cluster - free the not-yet-completed RSC cluster
+ *
+ * The "next" pointer of the last segment of (not-yet-completed) RSC clusters
+ * in the sw_rsc_ring is not set to NULL but rather points to the next
+ * mbuf of this RSC aggregation (that has not been completed yet and still
+ * resides on the HW ring). So, instead of calling for rte_pktmbuf_free() we
+ * will just free first "nb_segs" segments of the cluster explicitly by calling
+ * an rte_pktmbuf_free_seg().
+ *
+ * @m RSC cluster head
+ */
+static void
+ixgbe_free_rsc_cluster(struct rte_mbuf *m)
+{
+       uint8_t i, nb_segs = m->nb_segs;
+       struct rte_mbuf *next_seg;
+
+       for (i = 0; i < nb_segs; i++) {
+               next_seg = m->next;
+               rte_pktmbuf_free_seg(m);
+               m = next_seg;
+       }
+}
+
 static void
-ixgbe_rx_queue_release_mbufs(struct igb_rx_queue *rxq)
+ixgbe_rx_queue_release_mbufs(struct ixgbe_rx_queue *rxq)
 {
        unsigned i;
 
@@ -1938,14 +2352,22 @@ ixgbe_rx_queue_release_mbufs(struct igb_rx_queue *rxq)
                }
 #endif
        }
+
+       if (rxq->sw_rsc_ring)
+               for (i = 0; i < rxq->nb_rx_desc; i++)
+                       if (rxq->sw_rsc_ring[i].fbuf) {
+                               ixgbe_free_rsc_cluster(rxq->sw_rsc_ring[i].fbuf);
+                               rxq->sw_rsc_ring[i].fbuf = NULL;
+                       }
 }
 
 static void
-ixgbe_rx_queue_release(struct igb_rx_queue *rxq)
+ixgbe_rx_queue_release(struct ixgbe_rx_queue *rxq)
 {
        if (rxq != NULL) {
                ixgbe_rx_queue_release_mbufs(rxq);
                rte_free(rxq->sw_ring);
+               rte_free(rxq->sw_rsc_ring);
                rte_free(rxq);
        }
 }
@@ -1966,9 +2388,9 @@ ixgbe_dev_rx_queue_release(void *rxq)
  */
 static inline int
 #ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
-check_rx_burst_bulk_alloc_preconditions(struct igb_rx_queue *rxq)
+check_rx_burst_bulk_alloc_preconditions(struct ixgbe_rx_queue *rxq)
 #else
-check_rx_burst_bulk_alloc_preconditions(__rte_unused struct igb_rx_queue *rxq)
+check_rx_burst_bulk_alloc_preconditions(__rte_unused struct ixgbe_rx_queue *rxq)
 #endif
 {
        int ret = 0;
@@ -2018,14 +2440,13 @@ check_rx_burst_bulk_alloc_preconditions(__rte_unused struct igb_rx_queue *rxq)
        return ret;
 }
 
-/* Reset dynamic igb_rx_queue fields back to defaults */
+/* Reset dynamic ixgbe_rx_queue fields back to defaults */
 static void
-ixgbe_reset_rx_queue(struct igb_rx_queue *rxq)
+ixgbe_reset_rx_queue(struct ixgbe_hw *hw, struct ixgbe_rx_queue *rxq)
 {
-       static const union ixgbe_adv_rx_desc zeroed_desc = { .read = {
-                       .pkt_addr = 0}};
+       static const union ixgbe_adv_rx_desc zeroed_desc = {{0}};
        unsigned i;
-       uint16_t len;
+       uint16_t len = rxq->nb_rx_desc;
 
        /*
         * By default, the Rx queue setup function allocates enough memory for
@@ -2037,14 +2458,9 @@ ixgbe_reset_rx_queue(struct igb_rx_queue *rxq)
         * constraints here to see if we need to zero out memory after the end
         * of the H/W descriptor ring.
         */
-#ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
-       if (check_rx_burst_bulk_alloc_preconditions(rxq) == 0)
+       if (hw->rx_bulk_alloc_allowed)
                /* zero out extra memory */
-               len = (uint16_t)(rxq->nb_rx_desc + RTE_PMD_IXGBE_RX_MAX_BURST);
-       else
-#endif
-               /* do not zero out extra memory */
-               len = rxq->nb_rx_desc;
+               len += RTE_PMD_IXGBE_RX_MAX_BURST;
 
        /*
         * Zero out HW ring memory. Zero out extra memory at the end of
@@ -2061,8 +2477,8 @@ ixgbe_reset_rx_queue(struct igb_rx_queue *rxq)
         * entries is always allocated
         */
        memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
-       for (i = 0; i < RTE_PMD_IXGBE_RX_MAX_BURST; ++i) {
-               rxq->sw_ring[rxq->nb_rx_desc + i].mbuf = &rxq->fake_mbuf;
+       for (i = rxq->nb_rx_desc; i < len; ++i) {
+               rxq->sw_ring[i].mbuf = &rxq->fake_mbuf;
        }
 
        rxq->rx_nb_avail = 0;
@@ -2073,6 +2489,7 @@ ixgbe_reset_rx_queue(struct igb_rx_queue *rxq)
        rxq->nb_rx_hold = 0;
        rxq->pkt_first_seg = NULL;
        rxq->pkt_last_seg = NULL;
+       rxq->rsc_en = 0;
 }
 
 int
@@ -2084,10 +2501,17 @@ ixgbe_dev_rx_queue_setup(struct rte_eth_dev *dev,
                         struct rte_mempool *mp)
 {
        const struct rte_memzone *rz;
-       struct igb_rx_queue *rxq;
+       struct ixgbe_rx_queue *rxq;
        struct ixgbe_hw     *hw;
-       int use_def_burst_func = 1;
        uint16_t len;
+       struct rte_eth_dev_info dev_info = { 0 };
+       struct rte_eth_rxmode *dev_rx_mode = &dev->data->dev_conf.rxmode;
+       bool rsc_requested = false;
+
+       dev->dev_ops->dev_infos_get(dev, &dev_info);
+       if ((dev_info.rx_offload_capa & DEV_RX_OFFLOAD_TCP_LRO) &&
+           dev_rx_mode->enable_lro)
+               rsc_requested = true;
 
        PMD_INIT_FUNC_TRACE();
        hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
@@ -2110,8 +2534,8 @@ 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);
+       rxq = rte_zmalloc_socket("ethdev RX queue", sizeof(struct ixgbe_rx_queue),
+                                RTE_CACHE_LINE_SIZE, socket_id);
        if (rxq == NULL)
                return (-ENOMEM);
        rxq->mb_pool = mp;
@@ -2146,7 +2570,10 @@ 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 =
@@ -2165,61 +2592,65 @@ ixgbe_dev_rx_queue_setup(struct rte_eth_dev *dev,
 #endif
        rxq->rx_ring = (union ixgbe_adv_rx_desc *) rz->addr;
 
+       /*
+        * Certain constraints must be met in order to use the bulk buffer
+        * allocation Rx burst function. If any of Rx queues doesn't meet them
+        * the feature should be disabled for the whole port.
+        */
+       if (check_rx_burst_bulk_alloc_preconditions(rxq)) {
+               PMD_INIT_LOG(DEBUG, "queue[%d] doesn't meet Rx Bulk Alloc "
+                                   "preconditions - canceling the feature for "
+                                   "the whole port[%d]",
+                            rxq->queue_id, rxq->port_id);
+               hw->rx_bulk_alloc_allowed = false;
+       }
+
        /*
         * Allocate software ring. Allow for space at the end of the
         * S/W ring to make sure look-ahead logic in bulk alloc Rx burst
         * function does not access an invalid memory region.
         */
-#ifdef RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
-       len = (uint16_t)(nb_desc + RTE_PMD_IXGBE_RX_MAX_BURST);
-#else
        len = nb_desc;
-#endif
+       if (hw->rx_bulk_alloc_allowed)
+               len += RTE_PMD_IXGBE_RX_MAX_BURST;
+
        rxq->sw_ring = rte_zmalloc_socket("rxq->sw_ring",
-                                         sizeof(struct igb_rx_entry) * len,
-                                         CACHE_LINE_SIZE, socket_id);
-       if (rxq->sw_ring == NULL) {
+                                         sizeof(struct ixgbe_rx_entry) * len,
+                                         RTE_CACHE_LINE_SIZE, socket_id);
+       if (!rxq->sw_ring) {
                ixgbe_rx_queue_release(rxq);
                return (-ENOMEM);
        }
-       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);
-
-       /*
-        * Certain constraints must be met in order to use the bulk buffer
-        * allocation Rx burst function.
-        */
-       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.",
-                            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.");
-                       dev->rx_pkt_burst = ixgbe_recv_pkts_vec;
+       if (rsc_requested) {
+               rxq->sw_rsc_ring =
+                       rte_zmalloc_socket("rxq->sw_rsc_ring",
+                                          sizeof(struct ixgbe_rsc_entry) * len,
+                                          RTE_CACHE_LINE_SIZE, socket_id);
+               if (!rxq->sw_rsc_ring) {
+                       ixgbe_rx_queue_release(rxq);
+                       return (-ENOMEM);
                }
-#endif
-#endif
-       } else {
-               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).",
-                            rxq->port_id, rxq->queue_id);
-       }
+       } else
+               rxq->sw_rsc_ring = NULL;
+
+       PMD_INIT_LOG(DEBUG, "sw_ring=%p sw_rsc_ring=%p hw_ring=%p "
+                           "dma_addr=0x%"PRIx64,
+                    rxq->sw_ring, rxq->sw_rsc_ring, rxq->rx_ring,
+                    rxq->rx_ring_phys_addr);
+
+       if (!rte_is_power_of_2(nb_desc)) {
+               PMD_INIT_LOG(DEBUG, "queue[%d] doesn't meet Vector Rx "
+                                   "preconditions - canceling the feature for "
+                                   "the whole port[%d]",
+                            rxq->queue_id, rxq->port_id);
+               hw->rx_vec_allowed = false;
+       } else
+               ixgbe_rxq_vec_setup(rxq);
+
        dev->data->rx_queues[queue_idx] = rxq;
 
-       ixgbe_reset_rx_queue(rxq);
+       ixgbe_reset_rx_queue(hw, rxq);
 
        return 0;
 }
@@ -2229,7 +2660,7 @@ ixgbe_dev_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 {
 #define IXGBE_RXQ_SCAN_INTERVAL 4
        volatile union ixgbe_adv_rx_desc *rxdp;
-       struct igb_rx_queue *rxq;
+       struct ixgbe_rx_queue *rxq;
        uint32_t desc = 0;
 
        if (rx_queue_id >= dev->data->nb_rx_queues) {
@@ -2256,7 +2687,7 @@ int
 ixgbe_dev_rx_descriptor_done(void *rx_queue, uint16_t offset)
 {
        volatile union ixgbe_adv_rx_desc *rxdp;
-       struct igb_rx_queue *rxq = rx_queue;
+       struct ixgbe_rx_queue *rxq = rx_queue;
        uint32_t desc;
 
        if (unlikely(offset >= rxq->nb_rx_desc))
@@ -2273,11 +2704,12 @@ void
 ixgbe_dev_clear_queues(struct rte_eth_dev *dev)
 {
        unsigned i;
+       struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 
        PMD_INIT_FUNC_TRACE();
 
        for (i = 0; i < dev->data->nb_tx_queues; i++) {
-               struct igb_tx_queue *txq = dev->data->tx_queues[i];
+               struct ixgbe_tx_queue *txq = dev->data->tx_queues[i];
                if (txq != NULL) {
                        txq->ops->release_mbufs(txq);
                        txq->ops->reset(txq);
@@ -2285,10 +2717,10 @@ ixgbe_dev_clear_queues(struct rte_eth_dev *dev)
        }
 
        for (i = 0; i < dev->data->nb_rx_queues; i++) {
-               struct igb_rx_queue *rxq = dev->data->rx_queues[i];
+               struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
                if (rxq != NULL) {
                        ixgbe_rx_queue_release_mbufs(rxq);
-                       ixgbe_reset_rx_queue(rxq);
+                       ixgbe_reset_rx_queue(hw, rxq);
                }
        }
 }
@@ -2367,19 +2799,19 @@ 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;
@@ -2453,19 +2885,19 @@ 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;
@@ -3123,6 +3555,7 @@ 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();
@@ -3145,6 +3578,11 @@ 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 */
@@ -3235,9 +3673,9 @@ ixgbe_vmdq_tx_hw_configure(struct ixgbe_hw *hw)
 }
 
 static int
-ixgbe_alloc_rx_queue_mbufs(struct igb_rx_queue *rxq)
+ixgbe_alloc_rx_queue_mbufs(struct ixgbe_rx_queue *rxq)
 {
-       struct igb_rx_entry *rxe = rxq->sw_ring;
+       struct ixgbe_rx_entry *rxe = rxq->sw_ring;
        uint64_t dma_addr;
        unsigned i;
 
@@ -3268,6 +3706,67 @@ 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)
 {
@@ -3300,24 +3799,25 @@ 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;
                }
        }
 
@@ -3382,6 +3882,319 @@ ixgbe_dev_mq_tx_configure(struct rte_eth_dev *dev)
        return 0;
 }
 
+/**
+ * ixgbe_get_rscctl_maxdesc - Calculate the RSCCTL[n].MAXDESC for PF
+ *
+ * Return the RSCCTL[n].MAXDESC for 82599 and x540 PF devices according to the
+ * spec rev. 3.0 chapter 8.2.3.8.13.
+ *
+ * @pool Memory pool of the Rx queue
+ */
+static inline uint32_t
+ixgbe_get_rscctl_maxdesc(struct rte_mempool *pool)
+{
+       struct rte_pktmbuf_pool_private *mp_priv = rte_mempool_get_priv(pool);
+
+       /* MAXDESC * SRRCTL.BSIZEPKT must not exceed 64 KB minus one */
+       uint16_t maxdesc =
+               IPV4_MAX_PKT_LEN /
+                       (mp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM);
+
+       if (maxdesc >= 16)
+               return IXGBE_RSCCTL_MAXDESC_16;
+       else if (maxdesc >= 8)
+               return IXGBE_RSCCTL_MAXDESC_8;
+       else if (maxdesc >= 4)
+               return IXGBE_RSCCTL_MAXDESC_4;
+       else
+               return IXGBE_RSCCTL_MAXDESC_1;
+}
+
+/**
+ * ixgbe_set_ivar - Setup the correct IVAR register for a particular MSIX
+ * interrupt
+ *
+ * (Taken from FreeBSD tree)
+ * (yes this is all very magic and confusing :)
+ *
+ * @dev port handle
+ * @entry the register array entry
+ * @vector the MSIX vector for this queue
+ * @type RX/TX/MISC
+ */
+static void
+ixgbe_set_ivar(struct rte_eth_dev *dev, u8 entry, u8 vector, s8 type)
+{
+       struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       u32 ivar, index;
+
+       vector |= IXGBE_IVAR_ALLOC_VAL;
+
+       switch (hw->mac.type) {
+
+       case ixgbe_mac_82598EB:
+               if (type == -1)
+                       entry = IXGBE_IVAR_OTHER_CAUSES_INDEX;
+               else
+                       entry += (type * 64);
+               index = (entry >> 2) & 0x1F;
+               ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
+               ivar &= ~(0xFF << (8 * (entry & 0x3)));
+               ivar |= (vector << (8 * (entry & 0x3)));
+               IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
+               break;
+
+       case ixgbe_mac_82599EB:
+       case ixgbe_mac_X540:
+               if (type == -1) { /* MISC IVAR */
+                       index = (entry & 1) * 8;
+                       ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
+                       ivar &= ~(0xFF << index);
+                       ivar |= (vector << index);
+                       IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
+               } else {        /* RX/TX IVARS */
+                       index = (16 * (entry & 1)) + (8 * type);
+                       ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(entry >> 1));
+                       ivar &= ~(0xFF << index);
+                       ivar |= (vector << index);
+                       IXGBE_WRITE_REG(hw, IXGBE_IVAR(entry >> 1), ivar);
+               }
+
+               break;
+
+       default:
+               break;
+       }
+}
+
+void ixgbe_set_rx_function(struct rte_eth_dev *dev)
+{
+       struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       /*
+        * In order to allow Vector Rx there are a few configuration
+        * conditions to be met and Rx Bulk Allocation should be allowed.
+        */
+       if (ixgbe_rx_vec_dev_conf_condition_check(dev) ||
+           !hw->rx_bulk_alloc_allowed) {
+               PMD_INIT_LOG(DEBUG, "Port[%d] doesn't meet Vector Rx "
+                                   "preconditions or RTE_IXGBE_INC_VECTOR is "
+                                   "not enabled",
+                            dev->data->port_id);
+
+               hw->rx_vec_allowed = false;
+       }
+
+       /*
+        * Initialize the appropriate LRO callback.
+        *
+        * If all queues satisfy the bulk allocation preconditions
+        * (hw->rx_bulk_alloc_allowed is TRUE) then we may use bulk allocation.
+        * Otherwise use a single allocation version.
+        */
+       if (dev->data->lro) {
+               if (hw->rx_bulk_alloc_allowed) {
+                       PMD_INIT_LOG(INFO, "LRO is requested. Using a bulk "
+                                          "allocation version");
+                       dev->rx_pkt_burst = ixgbe_recv_pkts_lro_bulk_alloc;
+               } else {
+                       PMD_INIT_LOG(INFO, "LRO is requested. Using a single "
+                                          "allocation version");
+                       dev->rx_pkt_burst = ixgbe_recv_pkts_lro_single_alloc;
+               }
+       } else if (dev->data->scattered_rx) {
+               /*
+                * Set the non-LRO scattered callback: there are Vector and
+                * single allocation versions.
+                */
+               if (hw->rx_vec_allowed) {
+                       PMD_INIT_LOG(DEBUG, "Using Vector Scattered Rx "
+                                           "callback (port=%d).",
+                                    dev->data->port_id);
+
+                       dev->rx_pkt_burst = ixgbe_recv_scattered_pkts_vec;
+               } else {
+                       PMD_INIT_LOG(DEBUG, "Using Regualr (non-vector) "
+                                           "Scattered Rx callback "
+                                           "(port=%d).",
+                                    dev->data->port_id);
+
+                       dev->rx_pkt_burst = ixgbe_recv_scattered_pkts;
+               }
+       /*
+        * Below we set "simple" callbacks according to port/queues parameters.
+        * If parameters allow we are going to choose between the following
+        * callbacks:
+        *    - Vector
+        *    - Bulk Allocation
+        *    - Single buffer allocation (the simplest one)
+        */
+       } else if (hw->rx_vec_allowed) {
+               PMD_INIT_LOG(INFO, "Vector rx enabled, please make sure RX "
+                                  "burst size no less than 32.");
+
+               dev->rx_pkt_burst = ixgbe_recv_pkts_vec;
+       } else if (hw->rx_bulk_alloc_allowed) {
+               PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
+                                   "satisfied. Rx Burst Bulk Alloc function "
+                                   "will be used on port=%d.",
+                            dev->data->port_id);
+
+               dev->rx_pkt_burst = ixgbe_recv_pkts_bulk_alloc;
+       } else {
+               PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are not "
+                                   "satisfied, or Scattered Rx is requested, "
+                                   "or RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC "
+                                   "is not enabled (port=%d).",
+                            dev->data->port_id);
+
+               dev->rx_pkt_burst = ixgbe_recv_pkts;
+       }
+}
+
+/**
+ * ixgbe_set_rsc - configure RSC related port HW registers
+ *
+ * Configures the port's RSC related registers according to the 4.6.7.2 chapter
+ * of 82599 Spec (x540 configuration is virtually the same).
+ *
+ * @dev port handle
+ *
+ * Returns 0 in case of success or a non-zero error code
+ */
+static int
+ixgbe_set_rsc(struct rte_eth_dev *dev)
+{
+       struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
+       struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct rte_eth_dev_info dev_info = { 0 };
+       bool rsc_capable = false;
+       uint16_t i;
+       uint32_t rdrxctl;
+
+       /* Sanity check */
+       dev->dev_ops->dev_infos_get(dev, &dev_info);
+       if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_TCP_LRO)
+               rsc_capable = true;
+
+       if (!rsc_capable && rx_conf->enable_lro) {
+               PMD_INIT_LOG(CRIT, "LRO is requested on HW that doesn't "
+                                  "support it");
+               return -EINVAL;
+       }
+
+       /* RSC global configuration (chapter 4.6.7.2.1 of 82599 Spec) */
+
+       if (!rx_conf->hw_strip_crc && rx_conf->enable_lro) {
+               /*
+                * According to chapter of 4.6.7.2.1 of the Spec Rev.
+                * 3.0 RSC configuration requires HW CRC stripping being
+                * enabled. If user requested both HW CRC stripping off
+                * and RSC on - return an error.
+                */
+               PMD_INIT_LOG(CRIT, "LRO can't be enabled when HW CRC "
+                                   "is disabled");
+               return -EINVAL;
+       }
+
+       /* RFCTL configuration  */
+       if (rsc_capable) {
+               uint32_t rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
+               if (rx_conf->enable_lro)
+                       /*
+                        * Since NFS packets coalescing is not supported - clear
+                        * RFCTL.NFSW_DIS and RFCTL.NFSR_DIS when RSC is
+                        * enabled.
+                        */
+                       rfctl &= ~(IXGBE_RFCTL_RSC_DIS | IXGBE_RFCTL_NFSW_DIS |
+                                  IXGBE_RFCTL_NFSR_DIS);
+               else
+                       rfctl |= IXGBE_RFCTL_RSC_DIS;
+
+               IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
+       }
+
+       /* If LRO hasn't been requested - we are done here. */
+       if (!rx_conf->enable_lro)
+               return 0;
+
+       /* Set RDRXCTL.RSCACKC bit */
+       rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
+       rdrxctl |= IXGBE_RDRXCTL_RSCACKC;
+       IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
+
+       /* Per-queue RSC configuration (chapter 4.6.7.2.2 of 82599 Spec) */
+       for (i = 0; i < dev->data->nb_rx_queues; i++) {
+               struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];
+               uint32_t srrctl =
+                       IXGBE_READ_REG(hw, IXGBE_SRRCTL(rxq->reg_idx));
+               uint32_t rscctl =
+                       IXGBE_READ_REG(hw, IXGBE_RSCCTL(rxq->reg_idx));
+               uint32_t psrtype =
+                       IXGBE_READ_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx));
+               uint32_t eitr =
+                       IXGBE_READ_REG(hw, IXGBE_EITR(rxq->reg_idx));
+
+               /*
+                * ixgbe PMD doesn't support header-split at the moment.
+                *
+                * Following the 4.6.7.2.1 chapter of the 82599/x540
+                * Spec if RSC is enabled the SRRCTL[n].BSIZEHEADER
+                * should be configured even if header split is not
+                * enabled. We will configure it 128 bytes following the
+                * recommendation in the spec.
+                */
+               srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
+               srrctl |= (128 << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
+                                           IXGBE_SRRCTL_BSIZEHDR_MASK;
+
+               /*
+                * TODO: Consider setting the Receive Descriptor Minimum
+                * Threshold Size for an RSC case. This is not an obviously
+                * beneficiary option but the one worth considering...
+                */
+
+               rscctl |= IXGBE_RSCCTL_RSCEN;
+               rscctl |= ixgbe_get_rscctl_maxdesc(rxq->mb_pool);
+               psrtype |= IXGBE_PSRTYPE_TCPHDR;
+
+               /*
+                * RSC: Set ITR interval corresponding to 2K ints/s.
+                *
+                * Full-sized RSC aggregations for a 10Gb/s link will
+                * arrive at about 20K aggregation/s rate.
+                *
+                * 2K inst/s rate will make only 10% of the
+                * aggregations to be closed due to the interrupt timer
+                * expiration for a streaming at wire-speed case.
+                *
+                * For a sparse streaming case this setting will yield
+                * at most 500us latency for a single RSC aggregation.
+                */
+               eitr &= ~IXGBE_EITR_ITR_INT_MASK;
+               eitr |= IXGBE_EITR_INTERVAL_US(500) | IXGBE_EITR_CNT_WDIS;
+
+               IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rxq->reg_idx), srrctl);
+               IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(rxq->reg_idx), rscctl);
+               IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(rxq->reg_idx), psrtype);
+               IXGBE_WRITE_REG(hw, IXGBE_EITR(rxq->reg_idx), eitr);
+
+               /*
+                * RSC requires the mapping of the queue to the
+                * interrupt vector.
+                */
+               ixgbe_set_ivar(dev, rxq->reg_idx, i, 0);
+
+               rxq->rsc_en = 1;
+       }
+
+       dev->data->lro = 1;
+
+       PMD_INIT_LOG(INFO, "enabling LRO mode");
+
+       return 0;
+}
+
 /*
  * Initializes Receive Unit.
  */
@@ -3389,8 +4202,7 @@ int
 ixgbe_dev_rx_init(struct rte_eth_dev *dev)
 {
        struct ixgbe_hw     *hw;
-       struct igb_rx_queue *rxq;
-       struct rte_pktmbuf_pool_private *mbp_priv;
+       struct ixgbe_rx_queue *rxq;
        uint64_t bus_addr;
        uint32_t rxctrl;
        uint32_t fctrl;
@@ -3401,6 +4213,8 @@ ixgbe_dev_rx_init(struct rte_eth_dev *dev)
        uint32_t rxcsum;
        uint16_t buf_size;
        uint16_t i;
+       struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
+       int rc;
 
        PMD_INIT_FUNC_TRACE();
        hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
@@ -3423,7 +4237,7 @@ ixgbe_dev_rx_init(struct rte_eth_dev *dev)
         * Configure CRC stripping, if any.
         */
        hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
-       if (dev->data->dev_conf.rxmode.hw_strip_crc)
+       if (rx_conf->hw_strip_crc)
                hlreg0 |= IXGBE_HLREG0_RXCRCSTRP;
        else
                hlreg0 &= ~IXGBE_HLREG0_RXCRCSTRP;
@@ -3431,11 +4245,11 @@ ixgbe_dev_rx_init(struct rte_eth_dev *dev)
        /*
         * Configure jumbo frame support, if any.
         */
-       if (dev->data->dev_conf.rxmode.jumbo_frame == 1) {
+       if (rx_conf->jumbo_frame == 1) {
                hlreg0 |= IXGBE_HLREG0_JUMBOEN;
                maxfrs = IXGBE_READ_REG(hw, IXGBE_MAXFRS);
                maxfrs &= 0x0000FFFF;
-               maxfrs |= (dev->data->dev_conf.rxmode.max_rx_pkt_len << 16);
+               maxfrs |= (rx_conf->max_rx_pkt_len << 16);
                IXGBE_WRITE_REG(hw, IXGBE_MAXFRS, maxfrs);
        } else
                hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
@@ -3459,9 +4273,7 @@ ixgbe_dev_rx_init(struct rte_eth_dev *dev)
                 * Reset crc_len in case it was changed after queue setup by a
                 * call to configure.
                 */
-               rxq->crc_len = (uint8_t)
-                               ((dev->data->dev_conf.rxmode.hw_strip_crc) ? 0 :
-                               ETHER_CRC_LEN);
+               rxq->crc_len = rx_conf->hw_strip_crc ? 0 : ETHER_CRC_LEN;
 
                /* Setup the Base and Length of the Rx Descriptor Rings */
                bus_addr = rxq->rx_ring_phys_addr;
@@ -3479,7 +4291,7 @@ ixgbe_dev_rx_init(struct rte_eth_dev *dev)
                /*
                 * Configure Header Split
                 */
-               if (dev->data->dev_conf.rxmode.header_split) {
+               if (rx_conf->header_split) {
                        if (hw->mac.type == ixgbe_mac_82599EB) {
                                /* Must setup the PSRTYPE register */
                                uint32_t psrtype;
@@ -3489,10 +4301,10 @@ ixgbe_dev_rx_init(struct rte_eth_dev *dev)
                                        IXGBE_PSRTYPE_IPV6HDR;
                                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;
+                       srrctl = ((rx_conf->split_hdr_size <<
+                               IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
+                               IXGBE_SRRCTL_BSIZEHDR_MASK);
+                       srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
                } else
 #endif
                        srrctl = IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
@@ -3507,40 +4319,24 @@ ixgbe_dev_rx_init(struct rte_eth_dev *dev)
                 * The value is in 1 KB resolution. Valid values can be from
                 * 1 KB to 16 KB.
                 */
-               mbp_priv = rte_mempool_get_priv(rxq->mb_pool);
-               buf_size = (uint16_t) (mbp_priv->mbuf_data_room_size -
-                                      RTE_PKTMBUF_HEADROOM);
+               buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
+                       RTE_PKTMBUF_HEADROOM);
                srrctl |= ((buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) &
                           IXGBE_SRRCTL_BSIZEPKT_MASK);
+
                IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rxq->reg_idx), srrctl);
 
                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 +
-                               2 * IXGBE_VLAN_TAG_SIZE) > buf_size){
-                       if (!dev->data->scattered_rx)
-                               PMD_INIT_LOG(DEBUG, "forcing scatter mode");
+               if (dev->data->dev_conf.rxmode.max_rx_pkt_len +
+                                           2 * IXGBE_VLAN_TAG_SIZE > buf_size)
                        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
+       if (rx_conf->enable_scatter)
                dev->data->scattered_rx = 1;
-       }
 
        /*
         * Device configured with multiple RX queues.
@@ -3554,16 +4350,17 @@ ixgbe_dev_rx_init(struct rte_eth_dev *dev)
         */
        rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
        rxcsum |= IXGBE_RXCSUM_PCSD;
-       if (dev->data->dev_conf.rxmode.hw_ip_checksum)
+       if (rx_conf->hw_ip_checksum)
                rxcsum |= IXGBE_RXCSUM_IPPCSE;
        else
                rxcsum &= ~IXGBE_RXCSUM_IPPCSE;
 
        IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
 
-       if (hw->mac.type == ixgbe_mac_82599EB) {
+       if (hw->mac.type == ixgbe_mac_82599EB ||
+           hw->mac.type == ixgbe_mac_X540) {
                rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
-               if (dev->data->dev_conf.rxmode.hw_strip_crc)
+               if (rx_conf->hw_strip_crc)
                        rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
                else
                        rdrxctl &= ~IXGBE_RDRXCTL_CRCSTRIP;
@@ -3571,6 +4368,12 @@ ixgbe_dev_rx_init(struct rte_eth_dev *dev)
                IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
        }
 
+       rc = ixgbe_set_rsc(dev);
+       if (rc)
+               return rc;
+
+       ixgbe_set_rx_function(dev);
+
        return 0;
 }
 
@@ -3581,7 +4384,7 @@ void
 ixgbe_dev_tx_init(struct rte_eth_dev *dev)
 {
        struct ixgbe_hw     *hw;
-       struct igb_tx_queue *txq;
+       struct ixgbe_tx_queue *txq;
        uint64_t bus_addr;
        uint32_t hlreg0;
        uint32_t txctrl;
@@ -3590,9 +4393,10 @@ 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 */
@@ -3672,16 +4476,17 @@ 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;
-       struct igb_tx_queue *txq;
-       struct igb_rx_queue *rxq;
+       struct ixgbe_tx_queue *txq;
+       struct ixgbe_rx_queue *rxq;
        uint32_t txdctl;
        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);
@@ -3704,14 +4509,20 @@ 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->tx_deferred_start)
-                       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->rx_deferred_start)
-                       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 */
@@ -3726,6 +4537,7 @@ 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;
 }
 
 /*
@@ -3735,7 +4547,7 @@ int
 ixgbe_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 {
        struct ixgbe_hw     *hw;
-       struct igb_rx_queue *rxq;
+       struct ixgbe_rx_queue *rxq;
        uint32_t rxdctl;
        int poll_ms;
 
@@ -3780,7 +4592,7 @@ int
 ixgbe_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 {
        struct ixgbe_hw     *hw;
-       struct igb_rx_queue *rxq;
+       struct ixgbe_rx_queue *rxq;
        uint32_t rxdctl;
        int poll_ms;
 
@@ -3807,7 +4619,7 @@ ixgbe_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
                rte_delay_us(RTE_IXGBE_WAIT_100_US);
 
                ixgbe_rx_queue_release_mbufs(rxq);
-               ixgbe_reset_rx_queue(rxq);
+               ixgbe_reset_rx_queue(hw, rxq);
        } else
                return -1;
 
@@ -3822,7 +4634,7 @@ int
 ixgbe_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 {
        struct ixgbe_hw     *hw;
-       struct igb_tx_queue *txq;
+       struct ixgbe_tx_queue *txq;
        uint32_t txdctl;
        int poll_ms;
 
@@ -3863,7 +4675,7 @@ int
 ixgbe_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 {
        struct ixgbe_hw     *hw;
-       struct igb_tx_queue *txq;
+       struct ixgbe_tx_queue *txq;
        uint32_t txdctl;
        uint32_t txtdh, txtdt;
        int poll_ms;
@@ -3923,10 +4735,9 @@ int
 ixgbevf_dev_rx_init(struct rte_eth_dev *dev)
 {
        struct ixgbe_hw     *hw;
-       struct igb_rx_queue *rxq;
-       struct rte_pktmbuf_pool_private *mbp_priv;
+       struct ixgbe_rx_queue *rxq;
        uint64_t bus_addr;
-       uint32_t srrctl;
+       uint32_t srrctl, psrtype = 0;
        uint16_t buf_size;
        uint16_t i;
        int ret;
@@ -3934,6 +4745,19 @@ 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.
@@ -3980,20 +4804,10 @@ 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;
@@ -4008,9 +4822,8 @@ ixgbevf_dev_rx_init(struct rte_eth_dev *dev)
                 * The value is in 1 KB resolution. Valid values can be from
                 * 1 KB to 16 KB.
                 */
-               mbp_priv = rte_mempool_get_priv(rxq->mb_pool);
-               buf_size = (uint16_t) (mbp_priv->mbuf_data_room_size -
-                                      RTE_PKTMBUF_HEADROOM);
+               buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
+                       RTE_PKTMBUF_HEADROOM);
                srrctl |= ((buf_size >> IXGBE_SRRCTL_BSIZEPKT_SHIFT) &
                           IXGBE_SRRCTL_BSIZEPKT_MASK);
 
@@ -4022,30 +4835,36 @@ 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;
 }
@@ -4057,7 +4876,7 @@ void
 ixgbevf_dev_tx_init(struct rte_eth_dev *dev)
 {
        struct ixgbe_hw     *hw;
-       struct igb_tx_queue *txq;
+       struct ixgbe_tx_queue *txq;
        uint64_t bus_addr;
        uint32_t txctrl;
        uint16_t i;
@@ -4098,8 +4917,8 @@ void
 ixgbevf_dev_rxtx_start(struct rte_eth_dev *dev)
 {
        struct ixgbe_hw     *hw;
-       struct igb_tx_queue *txq;
-       struct igb_rx_queue *rxq;
+       struct ixgbe_tx_queue *txq;
+       struct ixgbe_rx_queue *rxq;
        uint32_t txdctl;
        uint32_t rxdctl;
        uint16_t i;
@@ -4154,3 +4973,34 @@ ixgbevf_dev_rxtx_start(struct rte_eth_dev *dev)
 
        }
 }
+
+/* Stubs needed for linkage when CONFIG_RTE_IXGBE_INC_VECTOR is set to 'n' */
+int __attribute__((weak))
+ixgbe_rx_vec_dev_conf_condition_check(struct rte_eth_dev __rte_unused *dev)
+{
+       return -1;
+}
+
+uint16_t __attribute__((weak))
+ixgbe_recv_pkts_vec(
+       void __rte_unused *rx_queue,
+       struct rte_mbuf __rte_unused **rx_pkts,
+       uint16_t __rte_unused nb_pkts)
+{
+       return 0;
+}
+
+uint16_t __attribute__((weak))
+ixgbe_recv_scattered_pkts_vec(
+       void __rte_unused *rx_queue,
+       struct rte_mbuf __rte_unused **rx_pkts,
+       uint16_t __rte_unused nb_pkts)
+{
+       return 0;
+}
+
+int __attribute__((weak))
+ixgbe_rxq_vec_setup(struct ixgbe_rx_queue __rte_unused *rxq)
+{
+       return -1;
+}