mbuf: replace data pointer by an offset
[dpdk.git] / lib / librte_pmd_e1000 / igb_rxtx.c
index f3a82de..41727e7 100644 (file)
@@ -1,40 +1,38 @@
 /*-
  *   BSD LICENSE
- * 
- *   Copyright(c) 2010-2013 Intel Corporation. All rights reserved.
+ *
+ *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
  *   All rights reserved.
- * 
- *   Redistribution and use in source and binary forms, with or without 
- *   modification, are permitted provided that the following conditions 
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
  *   are met:
- * 
- *     * Redistributions of source code must retain the above copyright 
+ *
+ *     * Redistributions of source code must retain the above copyright
  *       notice, this list of conditions and the following disclaimer.
- *     * Redistributions in binary form must reproduce the above copyright 
- *       notice, this list of conditions and the following disclaimer in 
- *       the documentation and/or other materials provided with the 
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
  *       distribution.
- *     * Neither the name of Intel Corporation nor the names of its 
- *       contributors may be used to endorse or promote products derived 
+ *     * Neither the name of Intel Corporation nor the names of its
+ *       contributors may be used to endorse or promote products derived
  *       from this software without specific prior written permission.
- * 
- *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 
- *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 
- *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 
- *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 
- *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 
- *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 
- *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 
- *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 
- *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 
- *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- * 
  */
 
 #include <sys/queue.h>
 
-#include <endian.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 #include "e1000/e1000_api.h"
 #include "e1000_ethdev.h"
 
+#define IGB_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)
+
 static inline struct rte_mbuf *
 rte_rxmbuf_alloc(struct rte_mempool *mp)
 {
        struct rte_mbuf *m;
 
        m = __rte_mbuf_raw_alloc(mp);
-       __rte_mbuf_sanity_check_raw(m, RTE_MBUF_PKT, 0);
+       __rte_mbuf_sanity_check_raw(m, 0);
        return (m);
 }
 
 #define RTE_MBUF_DATA_DMA_ADDR(mb) \
-       (uint64_t) ((mb)->buf_physaddr +                   \
-                       (uint64_t) ((char *)((mb)->pkt.data) -     \
-                               (char *)(mb)->buf_addr))
+       (uint64_t) ((mb)->buf_physaddr + (mb)->data_off)
 
 #define RTE_MBUF_DATA_DMA_ADDR_DEFAULT(mb) \
        (uint64_t) ((mb)->buf_physaddr + RTE_PKTMBUF_HEADROOM)
@@ -117,6 +124,7 @@ struct igb_rx_queue {
        volatile union e1000_adv_rx_desc *rx_ring; /**< RX ring virtual address. */
        uint64_t            rx_ring_phys_addr; /**< RX ring DMA address. */
        volatile uint32_t   *rdt_reg_addr; /**< RDT register address. */
+       volatile uint32_t   *rdh_reg_addr; /**< RDH register address. */
        struct igb_rx_entry *sw_ring;   /**< address of RX software ring. */
        struct rte_mbuf *pkt_first_seg; /**< First segment of current packet. */
        struct rte_mbuf *pkt_last_seg;  /**< Last segment of current packet. */
@@ -125,6 +133,7 @@ struct igb_rx_queue {
        uint16_t            nb_rx_hold; /**< number of held free RX desc. */
        uint16_t            rx_free_thresh; /**< max free RX desc to hold. */
        uint16_t            queue_id;   /**< RX queue index. */
+       uint16_t            reg_idx;    /**< RX queue register index. */
        uint8_t             port_id;    /**< Device port identifier. */
        uint8_t             pthresh;    /**< Prefetch threshold register. */
        uint8_t             hthresh;    /**< Host threshold register. */
@@ -142,13 +151,33 @@ enum igb_advctx_num {
        IGB_CTX_NUM  = 2, /**< CTX_NUM */
 };
 
+/** Offload features */
+union igb_vlan_macip {
+       uint32_t data;
+       struct {
+               uint16_t l2_l3_len; /**< 7bit L2 and 9b L3 lengths combined */
+               uint16_t vlan_tci;
+               /**< VLAN Tag Control Identifier (CPU order). */
+       } f;
+};
+
+/*
+ * Compare mask for vlan_macip_len.data,
+ * should be in sync with igb_vlan_macip.f layout.
+ * */
+#define TX_VLAN_CMP_MASK        0xFFFF0000  /**< VLAN length - 16-bits. */
+#define TX_MAC_LEN_CMP_MASK     0x0000FE00  /**< MAC length - 7-bits. */
+#define TX_IP_LEN_CMP_MASK      0x000001FF  /**< IP  length - 9-bits. */
+/** MAC+IP  length. */
+#define TX_MACIP_LEN_CMP_MASK   (TX_MAC_LEN_CMP_MASK | TX_IP_LEN_CMP_MASK)
+
 /**
  * Strucutre to check if new context need be built
  */
 struct igb_advctx_info {
        uint16_t flags;           /**< ol_flags related to context build. */
        uint32_t cmp_mask;        /**< compare mask for vlan_macip_lens */
-       union rte_vlan_macip vlan_macip_lens; /**< vlan, mac & ip length. */
+       union igb_vlan_macip vlan_macip_lens; /**< vlan, mac & ip length. */
 };
 
 /**
@@ -165,6 +194,7 @@ struct igb_tx_queue {
        uint16_t               tx_head;
        /**< Index of first used TX descriptor. */
        uint16_t               queue_id; /**< TX queue index. */
+       uint16_t               reg_idx;  /**< TX queue register index. */
        uint8_t                port_id;  /**< Device port identifier. */
        uint8_t                pthresh;  /**< Prefetch threshold register. */
        uint8_t                hthresh;  /**< Host threshold register. */
@@ -193,6 +223,11 @@ struct igb_tx_queue {
 #define rte_packet_prefetch(p) do {} while(0)
 #endif
 
+/*
+ * Macro for VMDq feature for 1 GbE NIC.
+ */
+#define E1000_VMOLR_SIZE                       (8)
+
 /*********************************************************************
  *
  *  TX function
@@ -325,6 +360,7 @@ eth_igb_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
        volatile union e1000_adv_tx_desc *txd;
        struct rte_mbuf     *tx_pkt;
        struct rte_mbuf     *m_seg;
+       union igb_vlan_macip vlan_macip_lens;
        uint64_t buf_dma_addr;
        uint32_t olinfo_status;
        uint32_t cmd_type_len;
@@ -338,7 +374,6 @@ eth_igb_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
        uint16_t tx_ol_req;
        uint32_t new_ctx = 0;
        uint32_t ctx = 0;
-       uint32_t vlan_macip_lens;
 
        txq = tx_queue;
        sw_ring = txq->sw_ring;
@@ -348,7 +383,7 @@ eth_igb_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 
        for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
                tx_pkt = *tx_pkts++;
-               pkt_len = tx_pkt->pkt.pkt_len;
+               pkt_len = tx_pkt->pkt_len;
 
                RTE_MBUF_PREFETCH_TO_FREE(txe->mbuf);
 
@@ -360,16 +395,17 @@ eth_igb_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                 * for the packet, starting from the current position (tx_id)
                 * in the ring.
                 */
-               tx_last = (uint16_t) (tx_id + tx_pkt->pkt.nb_segs - 1);
+               tx_last = (uint16_t) (tx_id + tx_pkt->nb_segs - 1);
 
                ol_flags = tx_pkt->ol_flags;
-               vlan_macip_lens = tx_pkt->pkt.vlan_macip.data;
-               tx_ol_req = (ol_flags & PKT_TX_OFFLOAD_MASK);
+               vlan_macip_lens.f.vlan_tci = tx_pkt->vlan_tci;
+               vlan_macip_lens.f.l2_l3_len = tx_pkt->l2_l3_len;
+               tx_ol_req = (uint16_t)(ol_flags & PKT_TX_OFFLOAD_MASK);
 
                /* If a Context Descriptor need be built . */
                if (tx_ol_req) {
                        ctx = what_advctx_update(txq, tx_ol_req,
-                               vlan_macip_lens);
+                               vlan_macip_lens.data);
                        /* Only allocate context descriptor if required*/
                        new_ctx = (ctx == IGB_CTX_NUM);
                        ctx = txq->ctx_curr;
@@ -485,7 +521,7 @@ eth_igb_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                                }
 
                                igbe_set_xmit_ctx(txq, ctx_txd, tx_ol_req,
-                                   vlan_macip_lens);
+                                   vlan_macip_lens.data);
 
                                txe->last_id = tx_last;
                                tx_id = txe->next_id;
@@ -510,7 +546,7 @@ eth_igb_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                        /*
                         * Set up transmit descriptor.
                         */
-                       slen = (uint16_t) m_seg->pkt.data_len;
+                       slen = (uint16_t) m_seg->data_len;
                        buf_dma_addr = RTE_MBUF_DATA_DMA_ADDR(m_seg);
                        txd->read.buffer_addr =
                                rte_cpu_to_le_64(buf_dma_addr);
@@ -521,7 +557,7 @@ eth_igb_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                        txe->last_id = tx_last;
                        tx_id = txe->next_id;
                        txe = txn;
-                       m_seg = m_seg->pkt.next;
+                       m_seg = m_seg->next;
                } while (m_seg != NULL);
 
                /*
@@ -569,15 +605,15 @@ rx_desc_hlen_type_rss_to_pkt_flags(uint32_t hl_tp_rs)
                0, 0, 0, 0,
        };
 
-       pkt_flags = (uint16_t) (hl_tp_rs & E1000_RXDADV_PKTTYPE_ETQF) ?
+       pkt_flags = (uint16_t)((hl_tp_rs & E1000_RXDADV_PKTTYPE_ETQF) ?
                                ip_pkt_etqf_map[(hl_tp_rs >> 4) & 0x07] :
-                               ip_pkt_types_map[(hl_tp_rs >> 4) & 0x0F];
+                               ip_pkt_types_map[(hl_tp_rs >> 4) & 0x0F]);
 #else
-       pkt_flags = (uint16_t) (hl_tp_rs & E1000_RXDADV_PKTTYPE_ETQF) ? 0 :
-                               ip_pkt_types_map[(hl_tp_rs >> 4) & 0x0F];
+       pkt_flags = (uint16_t)((hl_tp_rs & E1000_RXDADV_PKTTYPE_ETQF) ? 0 :
+                               ip_pkt_types_map[(hl_tp_rs >> 4) & 0x0F]);
 #endif
-       return pkt_flags | (uint16_t) (((hl_tp_rs & 0x0F) == 0) ? 0 :
-                                       PKT_RX_RSS_HASH);
+       return (uint16_t)(pkt_flags | (((hl_tp_rs & 0x0F) == 0) ?
+                                               0 : PKT_RX_RSS_HASH));
 }
 
 static inline uint16_t
@@ -586,11 +622,12 @@ rx_desc_status_to_pkt_flags(uint32_t rx_status)
        uint16_t pkt_flags;
 
        /* Check if VLAN present */
-       pkt_flags = (uint16_t) (rx_status & E1000_RXD_STAT_VP) ? PKT_RX_VLAN_PKT : 0;
+       pkt_flags = (uint16_t)((rx_status & E1000_RXD_STAT_VP) ?
+                                               PKT_RX_VLAN_PKT : 0);
 
 #if defined(RTE_LIBRTE_IEEE1588)
        if (rx_status & E1000_RXD_STAT_TMST)
-               pkt_flags = pkt_flags | PKT_RX_IEEE1588_TMST;
+               pkt_flags = (uint16_t)(pkt_flags | PKT_RX_IEEE1588_TMST);
 #endif
        return pkt_flags;
 }
@@ -735,25 +772,24 @@ eth_igb_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                 */
                pkt_len = (uint16_t) (rte_le_to_cpu_16(rxd.wb.upper.length) -
                                      rxq->crc_len);
-               rxm->pkt.data = (char*) rxm->buf_addr + RTE_PKTMBUF_HEADROOM;
-               rte_packet_prefetch(rxm->pkt.data);
-               rxm->pkt.nb_segs = 1;
-               rxm->pkt.next = NULL;
-               rxm->pkt.pkt_len = pkt_len;
-               rxm->pkt.data_len = pkt_len;
-               rxm->pkt.in_port = rxq->port_id;
-
-               rxm->pkt.hash.rss = rxd.wb.lower.hi_dword.rss;
+               rxm->data_off = RTE_PKTMBUF_HEADROOM;
+               rte_packet_prefetch((char *)rxm->buf_addr + rxm->data_off);
+               rxm->nb_segs = 1;
+               rxm->next = NULL;
+               rxm->pkt_len = pkt_len;
+               rxm->data_len = pkt_len;
+               rxm->port = rxq->port_id;
+
+               rxm->hash.rss = rxd.wb.lower.hi_dword.rss;
                hlen_type_rss = rte_le_to_cpu_32(rxd.wb.lower.lo_dword.data);
                /* Only valid if PKT_RX_VLAN_PKT set in pkt_flags */
-               rxm->pkt.vlan_macip.f.vlan_tci =
-                       rte_le_to_cpu_16(rxd.wb.upper.vlan);
+               rxm->vlan_tci = rte_le_to_cpu_16(rxd.wb.upper.vlan);
 
                pkt_flags = rx_desc_hlen_type_rss_to_pkt_flags(hlen_type_rss);
-               pkt_flags = (pkt_flags |
-                                       rx_desc_status_to_pkt_flags(staterr));
-               pkt_flags = (pkt_flags |
-                                       rx_desc_error_to_pkt_flags(staterr));
+               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));
                rxm->ol_flags = pkt_flags;
 
                /*
@@ -911,8 +947,8 @@ eth_igb_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                 * Set data length & data buffer address of mbuf.
                 */
                data_len = rte_le_to_cpu_16(rxd.wb.upper.length);
-               rxm->pkt.data_len = data_len;
-               rxm->pkt.data = (char*) rxm->buf_addr + RTE_PKTMBUF_HEADROOM;
+               rxm->data_len = data_len;
+               rxm->data_off = RTE_PKTMBUF_HEADROOM;
 
                /*
                 * If this is the first buffer of the received packet,
@@ -924,12 +960,12 @@ eth_igb_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                 */
                if (first_seg == NULL) {
                        first_seg = rxm;
-                       first_seg->pkt.pkt_len = data_len;
-                       first_seg->pkt.nb_segs = 1;
+                       first_seg->pkt_len = data_len;
+                       first_seg->nb_segs = 1;
                } else {
-                       first_seg->pkt.pkt_len += data_len;
-                       first_seg->pkt.nb_segs++;
-                       last_seg->pkt.next = rxm;
+                       first_seg->pkt_len += data_len;
+                       first_seg->nb_segs++;
+                       last_seg->next = rxm;
                }
 
                /*
@@ -952,18 +988,18 @@ eth_igb_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                 *     mbuf, subtract the length of that CRC part from the
                 *     data length of the previous mbuf.
                 */
-               rxm->pkt.next = NULL;
+               rxm->next = NULL;
                if (unlikely(rxq->crc_len > 0)) {
-                       first_seg->pkt.pkt_len -= ETHER_CRC_LEN;
+                       first_seg->pkt_len -= ETHER_CRC_LEN;
                        if (data_len <= ETHER_CRC_LEN) {
                                rte_pktmbuf_free_seg(rxm);
-                               first_seg->pkt.nb_segs--;
-                               last_seg->pkt.data_len = (uint16_t)
-                                       (last_seg->pkt.data_len -
+                               first_seg->nb_segs--;
+                               last_seg->data_len = (uint16_t)
+                                       (last_seg->data_len -
                                         (ETHER_CRC_LEN - data_len));
-                               last_seg->pkt.next = NULL;
+                               last_seg->next = NULL;
                        } else
-                               rxm->pkt.data_len =
+                               rxm->data_len =
                                        (uint16_t) (data_len - ETHER_CRC_LEN);
                }
 
@@ -976,23 +1012,25 @@ eth_igb_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                 *      - VLAN TCI, if any,
                 *      - error flags.
                 */
-               first_seg->pkt.in_port = rxq->port_id;
-               first_seg->pkt.hash.rss = rxd.wb.lower.hi_dword.rss;
+               first_seg->port = rxq->port_id;
+               first_seg->hash.rss = rxd.wb.lower.hi_dword.rss;
 
                /*
                 * The vlan_tci field is only valid when PKT_RX_VLAN_PKT is
                 * set in the pkt_flags field.
                 */
-               first_seg->pkt.vlan_macip.f.vlan_tci =
-                       rte_le_to_cpu_16(rxd.wb.upper.vlan);
+               first_seg->vlan_tci = rte_le_to_cpu_16(rxd.wb.upper.vlan);
                hlen_type_rss = rte_le_to_cpu_32(rxd.wb.lower.lo_dword.data);
                pkt_flags = rx_desc_hlen_type_rss_to_pkt_flags(hlen_type_rss);
-               pkt_flags = (pkt_flags | rx_desc_status_to_pkt_flags(staterr));
-               pkt_flags = (pkt_flags | rx_desc_error_to_pkt_flags(staterr));
+               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;
 
                /* Prefetch data of first segment, if configured to do so. */
-               rte_packet_prefetch(first_seg->pkt.data);
+               rte_packet_prefetch((char *)first_seg->buf_addr +
+                       first_seg->data_off);
 
                /*
                 * Store the mbuf address into the next entry of the array
@@ -1069,15 +1107,20 @@ ring_dma_zone_reserve(struct rte_eth_dev *dev, const char *ring_name,
        char z_name[RTE_MEMZONE_NAMESIZE];
        const struct rte_memzone *mz;
 
-       rte_snprintf(z_name, sizeof(z_name), "%s_%s_%d_%d",
+       snprintf(z_name, sizeof(z_name), "%s_%s_%d_%d",
                        dev->driver->pci_drv.name, ring_name,
                                dev->data->port_id, queue_id);
        mz = rte_memzone_lookup(z_name);
        if (mz)
                return mz;
 
+#ifdef RTE_LIBRTE_XEN_DOM0
+       return rte_memzone_reserve_bounded(z_name, ring_size,
+                       socket_id, 0, IGB_ALIGN, RTE_PGSIZE_2M);
+#else
        return rte_memzone_reserve_aligned(z_name, ring_size,
                        socket_id, 0, IGB_ALIGN);
+#endif
 }
 
 static void
@@ -1124,20 +1167,20 @@ igb_reset_tx_queue_stat(struct igb_tx_queue *txq)
 static void
 igb_reset_tx_queue(struct igb_tx_queue *txq, struct rte_eth_dev *dev)
 {
+       static const union e1000_adv_tx_desc zeroed_desc = { .read = {
+                       .buffer_addr = 0}};
        struct igb_tx_entry *txe = txq->sw_ring;
-       uint32_t size;
        uint16_t i, prev;
        struct e1000_hw *hw;
 
        hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
-       size = sizeof(union e1000_adv_tx_desc) * txq->nb_tx_desc;
        /* Zero out HW ring memory */
-       for (i = 0; i < size; i++) {
-               ((volatile char *)txq->tx_ring)[i] = 0;
+       for (i = 0; i < txq->nb_tx_desc; i++) {
+               txq->tx_ring[i] = zeroed_desc;
        }
 
        /* Initialize ring entries */
-       prev = txq->nb_tx_desc - 1;
+       prev = (uint16_t)(txq->nb_tx_desc - 1);
        for (i = 0; i < txq->nb_tx_desc; i++) {
                volatile union e1000_adv_tx_desc *txd = &(txq->tx_ring[i]);
 
@@ -1198,8 +1241,10 @@ eth_igb_tx_queue_setup(struct rte_eth_dev *dev,
                        "the TX WTHRESH value to 4, 8, or 16.\n");
 
        /* Free memory prior to re-allocation if needed */
-       if (dev->data->tx_queues[queue_idx] != NULL)
+       if (dev->data->tx_queues[queue_idx] != NULL) {
                igb_tx_queue_release(dev->data->tx_queues[queue_idx]);
+               dev->data->tx_queues[queue_idx] = NULL;
+       }
 
        /* First allocate the tx queue data structure */
        txq = rte_zmalloc("ethdev TX queue", sizeof(struct igb_tx_queue),
@@ -1224,13 +1269,20 @@ eth_igb_tx_queue_setup(struct rte_eth_dev *dev,
        txq->pthresh = tx_conf->tx_thresh.pthresh;
        txq->hthresh = tx_conf->tx_thresh.hthresh;
        txq->wthresh = tx_conf->tx_thresh.wthresh;
+       if (txq->wthresh > 0 && hw->mac.type == e1000_82576)
+               txq->wthresh = 1;
        txq->queue_id = queue_idx;
+       txq->reg_idx = (uint16_t)((RTE_ETH_DEV_SRIOV(dev).active == 0) ?
+               queue_idx : RTE_ETH_DEV_SRIOV(dev).def_pool_q_idx + queue_idx);
        txq->port_id = dev->data->port_id;
 
-       txq->tdt_reg_addr = E1000_PCI_REG_ADDR(hw, E1000_TDT(queue_idx));
+       txq->tdt_reg_addr = E1000_PCI_REG_ADDR(hw, E1000_TDT(txq->reg_idx));
+#ifndef RTE_LIBRTE_XEN_DOM0
        txq->tx_ring_phys_addr = (uint64_t) tz->phys_addr;
-       txq->tx_ring = (union e1000_adv_tx_desc *) tz->addr;
-
+#else
+       txq->tx_ring_phys_addr = rte_mem_phy2mch(tz->memseg_id, tz->phys_addr);
+#endif
+        txq->tx_ring = (union e1000_adv_tx_desc *) tz->addr;
        /* Allocate software ring */
        txq->sw_ring = rte_zmalloc("txq->sw_ring",
                                   sizeof(struct igb_tx_entry) * nb_desc,
@@ -1283,13 +1335,13 @@ eth_igb_rx_queue_release(void *rxq)
 static void
 igb_reset_rx_queue(struct igb_rx_queue *rxq)
 {
-       unsigned size;
+       static const union e1000_adv_rx_desc zeroed_desc = { .read = {
+                       .pkt_addr = 0}};
        unsigned i;
 
        /* Zero out HW ring memory */
-       size = sizeof(union e1000_adv_rx_desc) * rxq->nb_rx_desc;
-       for (i = 0; i < size; i++) {
-               ((volatile char *)rxq->rx_ring)[i] = 0;
+       for (i = 0; i < rxq->nb_rx_desc; i++) {
+               rxq->rx_ring[i] = zeroed_desc;
        }
 
        rxq->rx_tail = 0;
@@ -1338,9 +1390,13 @@ eth_igb_rx_queue_setup(struct rte_eth_dev *dev,
        rxq->pthresh = rx_conf->rx_thresh.pthresh;
        rxq->hthresh = rx_conf->rx_thresh.hthresh;
        rxq->wthresh = rx_conf->rx_thresh.wthresh;
+       if (rxq->wthresh > 0 && hw->mac.type == e1000_82576)
+               rxq->wthresh = 1;
        rxq->drop_en = rx_conf->rx_drop_en;
        rxq->rx_free_thresh = rx_conf->rx_free_thresh;
        rxq->queue_id = queue_idx;
+       rxq->reg_idx = (uint16_t)((RTE_ETH_DEV_SRIOV(dev).active == 0) ?
+               queue_idx : RTE_ETH_DEV_SRIOV(dev).def_pool_q_idx + queue_idx);
        rxq->port_id = dev->data->port_id;
        rxq->crc_len = (uint8_t) ((dev->data->dev_conf.rxmode.hw_strip_crc) ? 0 :
                                  ETHER_CRC_LEN);
@@ -1356,8 +1412,13 @@ eth_igb_rx_queue_setup(struct rte_eth_dev *dev,
                igb_rx_queue_release(rxq);
                return (-ENOMEM);
        }
-       rxq->rdt_reg_addr = E1000_PCI_REG_ADDR(hw, E1000_RDT(queue_idx));
+       rxq->rdt_reg_addr = E1000_PCI_REG_ADDR(hw, E1000_RDT(rxq->reg_idx));
+       rxq->rdh_reg_addr = E1000_PCI_REG_ADDR(hw, E1000_RDH(rxq->reg_idx));
+#ifndef RTE_LIBRTE_XEN_DOM0
        rxq->rx_ring_phys_addr = (uint64_t) rz->phys_addr;
+#else
+       rxq->rx_ring_phys_addr = rte_mem_phy2mch(rz->memseg_id, rz->phys_addr);
+#endif
        rxq->rx_ring = (union e1000_adv_rx_desc *) rz->addr;
 
        /* Allocate software ring. */
@@ -1377,6 +1438,51 @@ eth_igb_rx_queue_setup(struct rte_eth_dev *dev,
        return 0;
 }
 
+uint32_t
+eth_igb_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
+{
+#define IGB_RXQ_SCAN_INTERVAL 4
+       volatile union e1000_adv_rx_desc *rxdp;
+       struct igb_rx_queue *rxq;
+       uint32_t desc = 0;
+
+       if (rx_queue_id >= dev->data->nb_rx_queues) {
+               PMD_RX_LOG(ERR, "Invalid RX queue id=%d\n", rx_queue_id);
+               return 0;
+       }
+
+       rxq = dev->data->rx_queues[rx_queue_id];
+       rxdp = &(rxq->rx_ring[rxq->rx_tail]);
+
+       while ((desc < rxq->nb_rx_desc) &&
+               (rxdp->wb.upper.status_error & E1000_RXD_STAT_DD)) {
+               desc += IGB_RXQ_SCAN_INTERVAL;
+               rxdp += IGB_RXQ_SCAN_INTERVAL;
+               if (rxq->rx_tail + desc >= rxq->nb_rx_desc)
+                       rxdp = &(rxq->rx_ring[rxq->rx_tail +
+                               desc - rxq->nb_rx_desc]);
+       }
+
+       return 0;
+}
+
+int
+eth_igb_rx_descriptor_done(void *rx_queue, uint16_t offset)
+{
+       volatile union e1000_adv_rx_desc *rxdp;
+       struct igb_rx_queue *rxq = rx_queue;
+       uint32_t desc;
+
+       if (unlikely(offset >= rxq->nb_rx_desc))
+               return 0;
+       desc = rxq->rx_tail + offset;
+       if (desc >= rxq->nb_rx_desc)
+               desc -= rxq->nb_rx_desc;
+
+       rxdp = &rxq->rx_ring[desc];
+       return !!(rxdp->wb.upper.status_error & E1000_RXD_STAT_DD);
+}
+
 void
 igb_dev_clear_queues(struct rte_eth_dev *dev)
 {
@@ -1444,35 +1550,141 @@ igb_rss_disable(struct rte_eth_dev *dev)
 }
 
 static void
-igb_rss_configure(struct rte_eth_dev *dev)
+igb_hw_rss_hash_set(struct e1000_hw *hw, struct rte_eth_rss_conf *rss_conf)
+{
+       uint8_t  *hash_key;
+       uint32_t rss_key;
+       uint32_t mrqc;
+       uint64_t rss_hf;
+       uint16_t i;
+
+       hash_key = rss_conf->rss_key;
+       if (hash_key != NULL) {
+               /* Fill in RSS hash key */
+               for (i = 0; i < 10; i++) {
+                       rss_key  = hash_key[(i * 4)];
+                       rss_key |= hash_key[(i * 4) + 1] << 8;
+                       rss_key |= hash_key[(i * 4) + 2] << 16;
+                       rss_key |= hash_key[(i * 4) + 3] << 24;
+                       E1000_WRITE_REG_ARRAY(hw, E1000_RSSRK(0), i, rss_key);
+               }
+       }
+
+       /* Set configured hashing protocols in MRQC register */
+       rss_hf = rss_conf->rss_hf;
+       mrqc = E1000_MRQC_ENABLE_RSS_4Q; /* RSS enabled. */
+       if (rss_hf & ETH_RSS_IPV4)
+               mrqc |= E1000_MRQC_RSS_FIELD_IPV4;
+       if (rss_hf & ETH_RSS_IPV4_TCP)
+               mrqc |= E1000_MRQC_RSS_FIELD_IPV4_TCP;
+       if (rss_hf & ETH_RSS_IPV6)
+               mrqc |= E1000_MRQC_RSS_FIELD_IPV6;
+       if (rss_hf & ETH_RSS_IPV6_EX)
+               mrqc |= E1000_MRQC_RSS_FIELD_IPV6_EX;
+       if (rss_hf & ETH_RSS_IPV6_TCP)
+               mrqc |= E1000_MRQC_RSS_FIELD_IPV6_TCP;
+       if (rss_hf & ETH_RSS_IPV6_TCP_EX)
+               mrqc |= E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
+       if (rss_hf & ETH_RSS_IPV4_UDP)
+               mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
+       if (rss_hf & ETH_RSS_IPV6_UDP)
+               mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
+       if (rss_hf & ETH_RSS_IPV6_UDP_EX)
+               mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP_EX;
+       E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
+}
+
+int
+eth_igb_rss_hash_update(struct rte_eth_dev *dev,
+                       struct rte_eth_rss_conf *rss_conf)
+{
+       struct e1000_hw *hw;
+       uint32_t mrqc;
+       uint64_t rss_hf;
+
+       hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       /*
+        * Before changing anything, first check that the update RSS operation
+        * does not attempt to disable RSS, if RSS was enabled at
+        * initialization time, or does not attempt to enable RSS, if RSS was
+        * disabled at initialization time.
+        */
+       rss_hf = rss_conf->rss_hf & IGB_RSS_OFFLOAD_ALL;
+       mrqc = E1000_READ_REG(hw, E1000_MRQC);
+       if (!(mrqc & E1000_MRQC_ENABLE_MASK)) { /* RSS disabled */
+               if (rss_hf != 0) /* Enable RSS */
+                       return -(EINVAL);
+               return 0; /* Nothing to do */
+       }
+       /* RSS enabled */
+       if (rss_hf == 0) /* Disable RSS */
+               return -(EINVAL);
+       igb_hw_rss_hash_set(hw, rss_conf);
+       return 0;
+}
+
+int eth_igb_rss_hash_conf_get(struct rte_eth_dev *dev,
+                             struct rte_eth_rss_conf *rss_conf)
 {
        struct e1000_hw *hw;
        uint8_t *hash_key;
        uint32_t rss_key;
        uint32_t mrqc;
-       uint32_t shift;
-       uint16_t rss_hf;
+       uint64_t rss_hf;
        uint16_t i;
 
        hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
-
-       rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
-       if (rss_hf == 0) /* Disable RSS. */ {
-               igb_rss_disable(dev);
-               return;
+       hash_key = rss_conf->rss_key;
+       if (hash_key != NULL) {
+               /* Return RSS hash key */
+               for (i = 0; i < 10; i++) {
+                       rss_key = E1000_READ_REG_ARRAY(hw, E1000_RSSRK(0), i);
+                       hash_key[(i * 4)] = rss_key & 0x000000FF;
+                       hash_key[(i * 4) + 1] = (rss_key >> 8) & 0x000000FF;
+                       hash_key[(i * 4) + 2] = (rss_key >> 16) & 0x000000FF;
+                       hash_key[(i * 4) + 3] = (rss_key >> 24) & 0x000000FF;
+               }
        }
-       hash_key = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key;
-       if (hash_key == NULL)
-               hash_key = rss_intel_key; /* Default hash key. */
-
-       /* Fill in RSS hash key. */
-       for (i = 0; i < 10; i++) {
-               rss_key  = hash_key[(i * 4)];
-               rss_key |= hash_key[(i * 4) + 1] << 8;
-               rss_key |= hash_key[(i * 4) + 2] << 16;
-               rss_key |= hash_key[(i * 4) + 3] << 24;
-               E1000_WRITE_REG_ARRAY(hw, E1000_RSSRK(0), i, rss_key);
+
+       /* Get RSS functions configured in MRQC register */
+       mrqc = E1000_READ_REG(hw, E1000_MRQC);
+       if ((mrqc & E1000_MRQC_ENABLE_RSS_4Q) == 0) { /* RSS is disabled */
+               rss_conf->rss_hf = 0;
+               return 0;
        }
+       rss_hf = 0;
+       if (mrqc & E1000_MRQC_RSS_FIELD_IPV4)
+               rss_hf |= ETH_RSS_IPV4;
+       if (mrqc & E1000_MRQC_RSS_FIELD_IPV4_TCP)
+               rss_hf |= ETH_RSS_IPV4_TCP;
+       if (mrqc & E1000_MRQC_RSS_FIELD_IPV6)
+               rss_hf |= ETH_RSS_IPV6;
+       if (mrqc & E1000_MRQC_RSS_FIELD_IPV6_EX)
+               rss_hf |= ETH_RSS_IPV6_EX;
+       if (mrqc & E1000_MRQC_RSS_FIELD_IPV6_TCP)
+               rss_hf |= ETH_RSS_IPV6_TCP;
+       if (mrqc & E1000_MRQC_RSS_FIELD_IPV6_TCP_EX)
+               rss_hf |= ETH_RSS_IPV6_TCP_EX;
+       if (mrqc & E1000_MRQC_RSS_FIELD_IPV4_UDP)
+               rss_hf |= ETH_RSS_IPV4_UDP;
+       if (mrqc & E1000_MRQC_RSS_FIELD_IPV6_UDP)
+               rss_hf |= ETH_RSS_IPV6_UDP;
+       if (mrqc & E1000_MRQC_RSS_FIELD_IPV6_UDP_EX)
+               rss_hf |= ETH_RSS_IPV6_UDP_EX;
+       rss_conf->rss_hf = rss_hf;
+       return 0;
+}
+
+static void
+igb_rss_configure(struct rte_eth_dev *dev)
+{
+       struct rte_eth_rss_conf rss_conf;
+       struct e1000_hw *hw;
+       uint32_t shift;
+       uint16_t i;
+
+       hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 
        /* Fill in redirection table. */
        shift = (hw->mac.type == e1000_82575) ? 6 : 0;
@@ -1490,29 +1702,132 @@ igb_rss_configure(struct rte_eth_dev *dev)
                        E1000_WRITE_REG(hw, E1000_RETA(i >> 2), reta.dword);
        }
 
-       /* Set configured hashing functions in MRQC register. */
-       mrqc = E1000_MRQC_ENABLE_RSS_4Q; /* RSS enabled. */
-       if (rss_hf & ETH_RSS_IPV4)
-               mrqc |= E1000_MRQC_RSS_FIELD_IPV4;
-       if (rss_hf & ETH_RSS_IPV4_TCP)
-               mrqc |= E1000_MRQC_RSS_FIELD_IPV4_TCP;
-       if (rss_hf & ETH_RSS_IPV6)
-               mrqc |= E1000_MRQC_RSS_FIELD_IPV6;
-       if (rss_hf & ETH_RSS_IPV6_EX)
-               mrqc |= E1000_MRQC_RSS_FIELD_IPV6_EX;
-       if (rss_hf & ETH_RSS_IPV6_TCP)
-               mrqc |= E1000_MRQC_RSS_FIELD_IPV6_TCP;
-       if (rss_hf & ETH_RSS_IPV6_TCP_EX)
-               mrqc |= E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
-       if (rss_hf & ETH_RSS_IPV4_UDP)
-               mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
-       if (rss_hf & ETH_RSS_IPV6_UDP)
-               mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
-       if (rss_hf & ETH_RSS_IPV6_UDP_EX)
-               mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP_EX;
+       /*
+        * Configure the RSS key and the RSS protocols used to compute
+        * the RSS hash of input packets.
+        */
+       rss_conf = dev->data->dev_conf.rx_adv_conf.rss_conf;
+       if ((rss_conf.rss_hf & IGB_RSS_OFFLOAD_ALL) == 0) {
+               igb_rss_disable(dev);
+               return;
+       }
+       if (rss_conf.rss_key == NULL)
+               rss_conf.rss_key = rss_intel_key; /* Default hash key */
+       igb_hw_rss_hash_set(hw, &rss_conf);
+}
+
+/*
+ * Check if the mac type support VMDq or not.
+ * Return 1 if it supports, otherwise, return 0.
+ */
+static int
+igb_is_vmdq_supported(const struct rte_eth_dev *dev)
+{
+       const struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       switch (hw->mac.type) {
+       case e1000_82576:
+       case e1000_82580:
+       case e1000_i350:
+               return 1;
+       case e1000_82540:
+       case e1000_82541:
+       case e1000_82542:
+       case e1000_82543:
+       case e1000_82544:
+       case e1000_82545:
+       case e1000_82546:
+       case e1000_82547:
+       case e1000_82571:
+       case e1000_82572:
+       case e1000_82573:
+       case e1000_82574:
+       case e1000_82583:
+       case e1000_i210:
+       case e1000_i211:
+       default:
+               PMD_INIT_LOG(ERR, "Cannot support VMDq feature\n");
+               return 0;
+       }
+}
+
+static int
+igb_vmdq_rx_hw_configure(struct rte_eth_dev *dev)
+{
+       struct rte_eth_vmdq_rx_conf *cfg;
+       struct e1000_hw *hw;
+       uint32_t mrqc, vt_ctl, vmolr, rctl;
+       int i;
+
+       PMD_INIT_LOG(DEBUG, ">>");
+       hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       cfg = &dev->data->dev_conf.rx_adv_conf.vmdq_rx_conf;
+
+       /* Check if mac type can support VMDq, return value of 0 means NOT support */
+       if (igb_is_vmdq_supported(dev) == 0)
+               return -1;
+
+       igb_rss_disable(dev);
+
+       /* RCTL: eanble VLAN filter */
+       rctl = E1000_READ_REG(hw, E1000_RCTL);
+       rctl |= E1000_RCTL_VFE;
+       E1000_WRITE_REG(hw, E1000_RCTL, rctl);
+
+       /* MRQC: enable vmdq */
+       mrqc = E1000_READ_REG(hw, E1000_MRQC);
+       mrqc |= E1000_MRQC_ENABLE_VMDQ;
        E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
+
+       /* VTCTL:  pool selection according to VLAN tag */
+       vt_ctl = E1000_READ_REG(hw, E1000_VT_CTL);
+       if (cfg->enable_default_pool)
+               vt_ctl |= (cfg->default_pool << E1000_VT_CTL_DEFAULT_POOL_SHIFT);
+       vt_ctl |= E1000_VT_CTL_IGNORE_MAC;
+       E1000_WRITE_REG(hw, E1000_VT_CTL, vt_ctl);
+
+       /*
+        * VMOLR: set STRVLAN as 1 if IGMAC in VTCTL is set as 1
+        * Both 82576 and 82580 support it
+        */
+       if (hw->mac.type != e1000_i350) {
+               for (i = 0; i < E1000_VMOLR_SIZE; i++) {
+                       vmolr = E1000_READ_REG(hw, E1000_VMOLR(i));
+                       vmolr |= E1000_VMOLR_STRVLAN;
+                       E1000_WRITE_REG(hw, E1000_VMOLR(i), vmolr);
+               }
+       }
+
+       /* VFTA - enable all vlan filters */
+       for (i = 0; i < IGB_VFTA_SIZE; i++)
+               E1000_WRITE_REG(hw, (E1000_VFTA+(i*4)), UINT32_MAX);
+
+       /* VFRE: 8 pools enabling for rx, both 82576 and i350 support it */
+       if (hw->mac.type != e1000_82580)
+               E1000_WRITE_REG(hw, E1000_VFRE, E1000_MBVFICR_VFREQ_MASK);
+
+       /*
+        * RAH/RAL - allow pools to read specific mac addresses
+        * In this case, all pools should be able to read from mac addr 0
+        */
+       E1000_WRITE_REG(hw, E1000_RAH(0), (E1000_RAH_AV | UINT16_MAX));
+       E1000_WRITE_REG(hw, E1000_RAL(0), UINT32_MAX);
+
+       /* VLVF: set up filters for vlan tags as configured */
+       for (i = 0; i < cfg->nb_pool_maps; i++) {
+               /* set vlan id in VF register and set the valid bit */
+               E1000_WRITE_REG(hw, E1000_VLVF(i), (E1000_VLVF_VLANID_ENABLE | \
+                        (cfg->pool_map[i].vlan_id & ETH_VLAN_ID_MAX) | \
+                       ((cfg->pool_map[i].pools << E1000_VLVF_POOLSEL_SHIFT ) & \
+                       E1000_VLVF_POOLSEL_MASK)));
+       }
+
+       E1000_WRITE_FLUSH(hw);
+
+       return 0;
 }
 
+
 /*********************************************************************
  *
  *  Enable receive unit.
@@ -1534,7 +1849,6 @@ igb_alloc_rx_queue_mbufs(struct igb_rx_queue *rxq)
                if (mbuf == NULL) {
                        PMD_INIT_LOG(ERR, "RX mbuf alloc failed "
                                "queue_id=%hu\n", rxq->queue_id);
-                       igb_rx_queue_release(rxq);
                        return (-ENOMEM);
                }
                dma_addr =
@@ -1548,6 +1862,46 @@ igb_alloc_rx_queue_mbufs(struct igb_rx_queue *rxq)
        return 0;
 }
 
+#define E1000_MRQC_DEF_Q_SHIFT               (3)
+static int
+igb_dev_mq_rx_configure(struct rte_eth_dev *dev)
+{
+       struct e1000_hw *hw =
+               E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       uint32_t mrqc;
+
+       if (RTE_ETH_DEV_SRIOV(dev).active == ETH_8_POOLS) {
+               /*
+                * SRIOV active scheme
+                * FIXME if support RSS together with VMDq & SRIOV
+                */
+               mrqc = E1000_MRQC_ENABLE_VMDQ;
+               /* 011b Def_Q ignore, according to VT_CTL.DEF_PL */
+               mrqc |= 0x3 << E1000_MRQC_DEF_Q_SHIFT;
+               E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
+       } else if(RTE_ETH_DEV_SRIOV(dev).active == 0) {
+               /*
+                * SRIOV inactive scheme
+                */
+               switch (dev->data->dev_conf.rxmode.mq_mode) {
+                       case ETH_MQ_RX_RSS:
+                               igb_rss_configure(dev);
+                               break;
+                       case ETH_MQ_RX_VMDQ_ONLY:
+                               /*Configure general VMDQ only RX parameters*/
+                               igb_vmdq_rx_hw_configure(dev);
+                               break;
+                       case ETH_MQ_RX_NONE:
+                               /* if mq_mode is none, disable rss mode.*/
+                       default:
+                               igb_rss_disable(dev);
+                               break;
+               }
+       }
+
+       return 0;
+}
+
 int
 eth_igb_rx_init(struct rte_eth_dev *dev)
 {
@@ -1578,9 +1932,13 @@ eth_igb_rx_init(struct rte_eth_dev *dev)
        if (dev->data->dev_conf.rxmode.jumbo_frame == 1) {
                rctl |= E1000_RCTL_LPE;
 
-               /* Set maximum packet length. */
+               /*
+                * Set maximum packet length by default, and might be updated
+                * together with enabling/disabling dual VLAN.
+                */
                E1000_WRITE_REG(hw, E1000_RLPML,
-                               dev->data->dev_conf.rxmode.max_rx_pkt_len);
+                       dev->data->dev_conf.rxmode.max_rx_pkt_len +
+                                               VLAN_TAG_SIZE);
        } else
                rctl &= ~E1000_RCTL_LPE;
 
@@ -1607,20 +1965,19 @@ eth_igb_rx_init(struct rte_eth_dev *dev)
                                                        0 : ETHER_CRC_LEN);
 
                bus_addr = rxq->rx_ring_phys_addr;
-               E1000_WRITE_REG(hw, E1000_RDLEN(i),
+               E1000_WRITE_REG(hw, E1000_RDLEN(rxq->reg_idx),
                                rxq->nb_rx_desc *
                                sizeof(union e1000_adv_rx_desc));
-               E1000_WRITE_REG(hw, E1000_RDBAH(i),
+               E1000_WRITE_REG(hw, E1000_RDBAH(rxq->reg_idx),
                                (uint32_t)(bus_addr >> 32));
-               E1000_WRITE_REG(hw, E1000_RDBAL(i), (uint32_t)bus_addr);
+               E1000_WRITE_REG(hw, E1000_RDBAL(rxq->reg_idx), (uint32_t)bus_addr);
 
                srrctl = E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
 
                /*
                 * Configure RX buffer size.
                 */
-               mbp_priv = (struct rte_pktmbuf_pool_private *)
-                       ((char *)rxq->mb_pool + sizeof(struct rte_mempool));
+               mbp_priv = rte_mempool_get_priv(rxq->mb_pool);
                buf_size = (uint16_t) (mbp_priv->mbuf_data_room_size -
                                       RTE_PKTMBUF_HEADROOM);
                if (buf_size >= 1024) {
@@ -1637,8 +1994,9 @@ eth_igb_rx_init(struct rte_eth_dev *dev)
                                                E1000_SRRCTL_BSIZEPKT_MASK) <<
                                               E1000_SRRCTL_BSIZEPKT_SHIFT);
 
-                       if (dev->data->dev_conf.rxmode.max_rx_pkt_len + VLAN_TAG_SIZE
-                                       > buf_size){
+                       /* It adds dual VLAN length for supporting dual VLAN */
+                       if ((dev->data->dev_conf.rxmode.max_rx_pkt_len +
+                                               2 * VLAN_TAG_SIZE) > buf_size){
                                dev->rx_pkt_burst = eth_igb_recv_scattered_pkts;
                                dev->data->scattered_rx = 1;
                        }
@@ -1656,16 +2014,21 @@ eth_igb_rx_init(struct rte_eth_dev *dev)
                if (rxq->drop_en)
                        srrctl |= E1000_SRRCTL_DROP_EN;
 
-               E1000_WRITE_REG(hw, E1000_SRRCTL(i), srrctl);
+               E1000_WRITE_REG(hw, E1000_SRRCTL(rxq->reg_idx), srrctl);
 
                /* Enable this RX queue. */
-               rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(i));
+               rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(rxq->reg_idx));
                rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
                rxdctl &= 0xFFF00000;
                rxdctl |= (rxq->pthresh & 0x1F);
                rxdctl |= ((rxq->hthresh & 0x1F) << 8);
                rxdctl |= ((rxq->wthresh & 0x1F) << 16);
-               E1000_WRITE_REG(hw, E1000_RXDCTL(i), rxdctl);
+               E1000_WRITE_REG(hw, E1000_RXDCTL(rxq->reg_idx), rxdctl);
+       }
+
+       if (dev->data->dev_conf.rxmode.enable_scatter) {
+               dev->rx_pkt_burst = eth_igb_recv_scattered_pkts;
+               dev->data->scattered_rx = 1;
        }
 
        /*
@@ -1692,10 +2055,10 @@ eth_igb_rx_init(struct rte_eth_dev *dev)
        /*
         * Configure RSS if device configured with multiple RX queues.
         */
-       if (dev->data->nb_rx_queues > 1)
-               igb_rss_configure(dev);
-       else
-               igb_rss_disable(dev);
+       igb_dev_mq_rx_configure(dev);
+
+       /* Update the rctl since igb_dev_mq_rx_configure may change its value */
+       rctl |= E1000_READ_REG(hw, E1000_RCTL);
 
        /*
         * Setup the Checksum Register.
@@ -1715,25 +2078,33 @@ eth_igb_rx_init(struct rte_eth_dev *dev)
        if (dev->data->dev_conf.rxmode.hw_strip_crc) {
                rctl |= E1000_RCTL_SECRC; /* Strip Ethernet CRC. */
 
-               /* set STRCRC bit in all queues for Powerville/Springville */
-               if (hw->mac.type == e1000_i350 || hw->mac.type == e1000_i210) {
+               /* set STRCRC bit in all queues */
+               if (hw->mac.type == e1000_i350 ||
+                   hw->mac.type == e1000_i210 ||
+                   hw->mac.type == e1000_i211 ||
+                   hw->mac.type == e1000_i354) {
                        for (i = 0; i < dev->data->nb_rx_queues; i++) {
+                               rxq = dev->data->rx_queues[i];
                                uint32_t dvmolr = E1000_READ_REG(hw,
-                                       E1000_DVMOLR(i));
+                                       E1000_DVMOLR(rxq->reg_idx));
                                dvmolr |= E1000_DVMOLR_STRCRC;
-                               E1000_WRITE_REG(hw, E1000_DVMOLR(i), dvmolr);
+                               E1000_WRITE_REG(hw, E1000_DVMOLR(rxq->reg_idx), dvmolr);
                        }
                }
        } else {
                rctl &= ~E1000_RCTL_SECRC; /* Do not Strip Ethernet CRC. */
 
-               /* clear STRCRC bit in all queues for Powerville/Springville */
-               if (hw->mac.type == e1000_i350 || hw->mac.type == e1000_i210) {
+               /* clear STRCRC bit in all queues */
+               if (hw->mac.type == e1000_i350 ||
+                   hw->mac.type == e1000_i210 ||
+                   hw->mac.type == e1000_i211 ||
+                   hw->mac.type == e1000_i354) {
                        for (i = 0; i < dev->data->nb_rx_queues; i++) {
+                               rxq = dev->data->rx_queues[i];
                                uint32_t dvmolr = E1000_READ_REG(hw,
-                                       E1000_DVMOLR(i));
+                                       E1000_DVMOLR(rxq->reg_idx));
                                dvmolr &= ~E1000_DVMOLR_STRCRC;
-                               E1000_WRITE_REG(hw, E1000_DVMOLR(i), dvmolr);
+                               E1000_WRITE_REG(hw, E1000_DVMOLR(rxq->reg_idx), dvmolr);
                        }
                }
        }
@@ -1744,7 +2115,8 @@ eth_igb_rx_init(struct rte_eth_dev *dev)
                (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
 
        /* Make sure VLAN Filters are off. */
-       rctl &= ~E1000_RCTL_VFE;
+       if (dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_VMDQ_ONLY)
+               rctl &= ~E1000_RCTL_VFE;
        /* Don't store bad packets. */
        rctl &= ~E1000_RCTL_SBP;
 
@@ -1757,8 +2129,8 @@ eth_igb_rx_init(struct rte_eth_dev *dev)
         */
        for (i = 0; i < dev->data->nb_rx_queues; i++) {
                rxq = dev->data->rx_queues[i];
-               E1000_WRITE_REG(hw, E1000_RDH(i), 0);
-               E1000_WRITE_REG(hw, E1000_RDT(i), rxq->nb_rx_desc - 1);
+               E1000_WRITE_REG(hw, E1000_RDH(rxq->reg_idx), 0);
+               E1000_WRITE_REG(hw, E1000_RDT(rxq->reg_idx), rxq->nb_rx_desc - 1);
        }
 
        return 0;
@@ -1786,24 +2158,24 @@ eth_igb_tx_init(struct rte_eth_dev *dev)
                txq = dev->data->tx_queues[i];
                bus_addr = txq->tx_ring_phys_addr;
 
-               E1000_WRITE_REG(hw, E1000_TDLEN(i),
+               E1000_WRITE_REG(hw, E1000_TDLEN(txq->reg_idx),
                                txq->nb_tx_desc *
                                sizeof(union e1000_adv_tx_desc));
-               E1000_WRITE_REG(hw, E1000_TDBAH(i),
+               E1000_WRITE_REG(hw, E1000_TDBAH(txq->reg_idx),
                                (uint32_t)(bus_addr >> 32));
-               E1000_WRITE_REG(hw, E1000_TDBAL(i), (uint32_t)bus_addr);
+               E1000_WRITE_REG(hw, E1000_TDBAL(txq->reg_idx), (uint32_t)bus_addr);
 
                /* Setup the HW Tx Head and Tail descriptor pointers. */
-               E1000_WRITE_REG(hw, E1000_TDT(i), 0);
-               E1000_WRITE_REG(hw, E1000_TDH(i), 0);
+               E1000_WRITE_REG(hw, E1000_TDT(txq->reg_idx), 0);
+               E1000_WRITE_REG(hw, E1000_TDH(txq->reg_idx), 0);
 
                /* Setup Transmit threshold registers. */
-               txdctl = E1000_READ_REG(hw, E1000_TXDCTL(i));
+               txdctl = E1000_READ_REG(hw, E1000_TXDCTL(txq->reg_idx));
                txdctl |= txq->pthresh & 0x1F;
                txdctl |= ((txq->hthresh & 0x1F) << 8);
                txdctl |= ((txq->wthresh & 0x1F) << 16);
                txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
-               E1000_WRITE_REG(hw, E1000_TXDCTL(i), txdctl);
+               E1000_WRITE_REG(hw, E1000_TXDCTL(txq->reg_idx), txdctl);
        }
 
        /* Program the Transmit Control Register. */
@@ -1837,6 +2209,11 @@ eth_igbvf_rx_init(struct rte_eth_dev *dev)
 
        hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 
+       /* setup MTU */
+       e1000_rlpml_set_vf(hw,
+               (uint16_t)(dev->data->dev_conf.rxmode.max_rx_pkt_len +
+               VLAN_TAG_SIZE));
+
        /* Configure and enable each RX queue. */
        rctl_bsize = 0;
        dev->rx_pkt_burst = eth_igb_recv_pkts;
@@ -1864,8 +2241,7 @@ eth_igbvf_rx_init(struct rte_eth_dev *dev)
                /*
                 * Configure RX buffer size.
                 */
-               mbp_priv = (struct rte_pktmbuf_pool_private *)
-                       ((char *)rxq->mb_pool + sizeof(struct rte_mempool));
+               mbp_priv = rte_mempool_get_priv(rxq->mb_pool);
                buf_size = (uint16_t) (mbp_priv->mbuf_data_room_size -
                                       RTE_PKTMBUF_HEADROOM);
                if (buf_size >= 1024) {
@@ -1882,7 +2258,9 @@ eth_igbvf_rx_init(struct rte_eth_dev *dev)
                                                E1000_SRRCTL_BSIZEPKT_MASK) <<
                                               E1000_SRRCTL_BSIZEPKT_SHIFT);
 
-                       if (dev->data->dev_conf.rxmode.max_rx_pkt_len > buf_size){
+                       /* It adds dual VLAN length for supporting dual VLAN */
+                       if ((dev->data->dev_conf.rxmode.max_rx_pkt_len +
+                                               2 * VLAN_TAG_SIZE) > buf_size){
                                dev->rx_pkt_burst = eth_igb_recv_scattered_pkts;
                                dev->data->scattered_rx = 1;
                        }
@@ -1908,10 +2286,10 @@ eth_igbvf_rx_init(struct rte_eth_dev *dev)
                rxdctl &= 0xFFF00000;
                rxdctl |= (rxq->pthresh & 0x1F);
                rxdctl |= ((rxq->hthresh & 0x1F) << 8);
-               if (hw->mac.type == e1000_82576) {
-                       /* 
+               if (hw->mac.type == e1000_vfadapt) {
+                       /*
                         * Workaround of 82576 VF Erratum
-                        * force set WTHRESH to 1 
+                        * force set WTHRESH to 1
                         * to avoid Write-Back not triggered sometimes
                         */
                        rxdctl |= 0x10000;
@@ -1922,6 +2300,11 @@ eth_igbvf_rx_init(struct rte_eth_dev *dev)
                E1000_WRITE_REG(hw, E1000_RXDCTL(i), rxdctl);
        }
 
+       if (dev->data->dev_conf.rxmode.enable_scatter) {
+               dev->rx_pkt_burst = eth_igb_recv_scattered_pkts;
+               dev->data->scattered_rx = 1;
+       }
+
        /*
         * Setup the HW Rx Head and Tail Descriptor Pointers.
         * This needs to be done after enable.
@@ -1972,12 +2355,12 @@ eth_igbvf_tx_init(struct rte_eth_dev *dev)
                txdctl |= txq->pthresh & 0x1F;
                txdctl |= ((txq->hthresh & 0x1F) << 8);
                if (hw->mac.type == e1000_82576) {
-                       /* 
+                       /*
                         * Workaround of 82576 VF Erratum
-                        * force set WTHRESH to 1 
+                        * force set WTHRESH to 1
                         * to avoid Write-Back not triggered sometimes
                         */
-                       txdctl |= 0x10000; 
+                       txdctl |= 0x10000;
                        PMD_INIT_LOG(DEBUG, "Force set TX WTHRESH to 1 !\n");
                }
                else