mbuf: replace data pointer by an offset
[dpdk.git] / lib / librte_pmd_e1000 / em_rxtx.c
index c44b910..ed3a6fc 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>
@@ -75,6 +73,7 @@
 #include "e1000_logs.h"
 #include "e1000/e1000_api.h"
 #include "e1000_ethdev.h"
+#include "e1000/e1000_osdep.h"
 
 #define        E1000_TXD_VLAN_SHIFT    16
 
@@ -86,13 +85,12 @@ 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)
@@ -121,6 +119,7 @@ struct em_rx_queue {
        volatile struct e1000_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 em_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. */
@@ -144,13 +143,34 @@ enum {
        EM_CTX_NUM  = 1, /**< CTX NUM */
 };
 
+/** Offload features */
+union em_vlan_macip {
+       uint32_t data;
+       struct {
+               uint16_t l3_len:9; /**< L3 (IP) Header Length. */
+               uint16_t l2_len:7; /**< L2 (MAC) Header Length. */
+               uint16_t vlan_tci;
+               /**< VLAN Tag Control Identifier (CPU order). */
+       } f;
+};
+
+/*
+ * Compare mask for vlan_macip_len.data,
+ * should be in sync with em_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)
+
 /**
  * Structure to check if new context need be built
  */
 struct em_ctx_info {
-       uint16_t flags;               /**< ol_flags related to context build. */
-       uint32_t cmp_mask;            /**< compare mask */
-       union rte_vlan_macip hdrlen;  /**< L2 and L3 header lenghts */
+       uint16_t flags;              /**< ol_flags related to context build. */
+       uint32_t cmp_mask;           /**< compare mask */
+       union em_vlan_macip hdrlen;  /**< L2 and L3 header lenghts */
 };
 
 /**
@@ -219,7 +239,7 @@ static inline void
 em_set_xmit_ctx(struct em_tx_queue* txq,
                volatile struct e1000_context_desc *ctx_txd,
                uint16_t flags,
-               union rte_vlan_macip hdrlen)
+               union em_vlan_macip hdrlen)
 {
        uint32_t cmp_mask, cmd_len;
        uint16_t ipcse, l2len;
@@ -229,19 +249,20 @@ em_set_xmit_ctx(struct em_tx_queue* txq,
        cmd_len = E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_C;
 
        l2len = hdrlen.f.l2_len;
-       ipcse = l2len + hdrlen.f.l3_len;
+       ipcse = (uint16_t)(l2len + hdrlen.f.l3_len);
 
        /* setup IPCS* fields */
-       ctx.lower_setup.ip_fields.ipcss = l2len;
-       ctx.lower_setup.ip_fields.ipcso =l2len +
-               offsetof(struct ipv4_hdr, hdr_checksum);
+       ctx.lower_setup.ip_fields.ipcss = (uint8_t)l2len;
+       ctx.lower_setup.ip_fields.ipcso = (uint8_t)(l2len +
+                       offsetof(struct ipv4_hdr, hdr_checksum));
 
        /*
         * When doing checksum or TCP segmentation with IPv6 headers,
         * IPCSE field should be set t0 0.
         */
        if (flags & PKT_TX_IP_CKSUM) {
-               ctx.lower_setup.ip_fields.ipcse = rte_cpu_to_le_16(ipcse - 1);
+               ctx.lower_setup.ip_fields.ipcse =
+                       (uint16_t)rte_cpu_to_le_16(ipcse - 1);
                cmd_len |= E1000_TXD_CMD_IP;
                cmp_mask |= TX_MACIP_LEN_CMP_MASK;
        } else {
@@ -249,18 +270,18 @@ em_set_xmit_ctx(struct em_tx_queue* txq,
        }
 
        /* setup TUCS* fields */
-       ctx.upper_setup.tcp_fields.tucss = ipcse;
+       ctx.upper_setup.tcp_fields.tucss = (uint8_t)ipcse;
        ctx.upper_setup.tcp_fields.tucse = 0;
 
        switch (flags & PKT_TX_L4_MASK) {
        case PKT_TX_UDP_CKSUM:
-               ctx.upper_setup.tcp_fields.tucso = ipcse +
-                       offsetof(struct udp_hdr, dgram_cksum);
+               ctx.upper_setup.tcp_fields.tucso = (uint8_t)(ipcse +
+                               offsetof(struct udp_hdr, dgram_cksum));
                cmp_mask |= TX_MACIP_LEN_CMP_MASK;
                break;
        case PKT_TX_TCP_CKSUM:
-               ctx.upper_setup.tcp_fields.tucso = ipcse +
-                       offsetof(struct tcp_hdr, cksum);
+               ctx.upper_setup.tcp_fields.tucso = (uint8_t)(ipcse +
+                               offsetof(struct tcp_hdr, cksum));
                cmd_len |= E1000_TXD_CMD_TCP;
                cmp_mask |= TX_MACIP_LEN_CMP_MASK;
                break;
@@ -284,7 +305,7 @@ em_set_xmit_ctx(struct em_tx_queue* txq,
  */
 static inline uint32_t
 what_ctx_update(struct em_tx_queue *txq, uint16_t flags,
-               union rte_vlan_macip hdrlen)
+               union em_vlan_macip hdrlen)
 {
        /* If match with the current context */
        if (likely (txq->ctx_cache.flags == flags &&
@@ -308,9 +329,9 @@ em_xmit_cleanup(struct em_tx_queue *txq)
        uint16_t nb_tx_to_clean;
 
        /* Determine the last descriptor needing to be cleaned */
-       desc_to_clean_to = last_desc_cleaned + txq->tx_rs_thresh;
+       desc_to_clean_to = (uint16_t)(last_desc_cleaned + txq->tx_rs_thresh);
        if (desc_to_clean_to >= nb_tx_desc)
-               desc_to_clean_to = desc_to_clean_to - nb_tx_desc;
+               desc_to_clean_to = (uint16_t)(desc_to_clean_to - nb_tx_desc);
 
        /* Check to make sure the last descriptor to clean is done */
        desc_to_clean_to = sw_ring[desc_to_clean_to].last_id;
@@ -327,10 +348,11 @@ em_xmit_cleanup(struct em_tx_queue *txq)
 
        /* Figure out how many descriptors will be cleaned */
        if (last_desc_cleaned > desc_to_clean_to)
-               nb_tx_to_clean = ((nb_tx_desc - last_desc_cleaned) +
-                       desc_to_clean_to);
+               nb_tx_to_clean = (uint16_t)((nb_tx_desc - last_desc_cleaned) +
+                                                       desc_to_clean_to);
        else
-               nb_tx_to_clean = desc_to_clean_to - last_desc_cleaned;
+               nb_tx_to_clean = (uint16_t)(desc_to_clean_to -
+                                               last_desc_cleaned);
 
        PMD_TX_FREE_LOG(DEBUG,
                        "Cleaning %4u TX descriptors: %4u to %4u "
@@ -348,7 +370,7 @@ em_xmit_cleanup(struct em_tx_queue *txq)
 
        /* Update the txq to reflect the last descriptor that was cleaned */
        txq->last_desc_cleaned = desc_to_clean_to;
-       txq->nb_tx_free += nb_tx_to_clean;
+       txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + nb_tx_to_clean);
 
        /* No Error */
        return (0);
@@ -389,7 +411,7 @@ eth_em_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
        uint16_t tx_ol_req;
        uint32_t ctx;
        uint32_t new_ctx;
-       union rte_vlan_macip hdrlen;
+       union em_vlan_macip hdrlen;
 
        txq = tx_queue;
        sw_ring = txq->sw_ring;
@@ -416,9 +438,12 @@ eth_em_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_IP_CKSUM | PKT_TX_L4_MASK);
+               tx_ol_req = (uint16_t)(ol_flags & (PKT_TX_IP_CKSUM |
+                                                       PKT_TX_L4_MASK));
                if (tx_ol_req) {
-                       hdrlen = tx_pkt->pkt.vlan_macip;
+                       hdrlen.f.vlan_tci = tx_pkt->vlan_tci;
+                       hdrlen.f.l2_len = tx_pkt->l2_len;
+                       hdrlen.f.l3_len = tx_pkt->l3_len;
                        /* If new context to be built or reuse the exist ctx. */
                        ctx = what_ctx_update(txq, tx_ol_req, hdrlen);
 
@@ -431,7 +456,7 @@ eth_em_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                 * This will always be the number of segments + the number of
                 * Context descriptors required to transmit the packet
                 */
-               nb_used = tx_pkt->pkt.nb_segs + new_ctx;
+               nb_used = (uint16_t)(tx_pkt->nb_segs + new_ctx);
 
                /*
                 * The number of descriptors that must be allocated for a
@@ -451,7 +476,7 @@ eth_em_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                        " tx_first=%u tx_last=%u\n",
                        (unsigned) txq->port_id,
                        (unsigned) txq->queue_id,
-                       (unsigned) tx_pkt->pkt.pkt_len,
+                       (unsigned) tx_pkt->pkt_len,
                        (unsigned) tx_id,
                        (unsigned) tx_last);
 
@@ -513,8 +538,7 @@ eth_em_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                /* Set VLAN Tag offload fields. */
                if (ol_flags & PKT_TX_VLAN_PKT) {
                        cmd_type_len |= E1000_TXD_CMD_VLE;
-                       popts_spec = tx_pkt->pkt.vlan_macip.f.vlan_tci <<
-                               E1000_TXD_VLAN_SHIFT;
+                       popts_spec = tx_pkt->vlan_tci << E1000_TXD_VLAN_SHIFT;
                }
 
                if (tx_ol_req) {
@@ -563,7 +587,7 @@ eth_em_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                        /*
                         * Set up Transmit Data Descriptor.
                         */
-                       slen = m_seg->pkt.data_len;
+                       slen = m_seg->data_len;
                        buf_dma_addr = RTE_MBUF_DATA_DMA_ADDR(m_seg);
 
                        txd->buffer_addr = rte_cpu_to_le_64(buf_dma_addr);
@@ -573,15 +597,15 @@ eth_em_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);
 
                /*
                 * The last packet data descriptor needs End Of Packet (EOP)
                 */
                cmd_type_len |= E1000_TXD_CMD_EOP;
-               txq->nb_tx_used += nb_used;
-               txq->nb_tx_free -= nb_used;
+               txq->nb_tx_used = (uint16_t)(txq->nb_tx_used + nb_used);
+               txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_used);
 
                /* Set RS bit only on threshold packets' last descriptor */
                if (txq->nb_tx_used >= txq->tx_rs_thresh) {
@@ -624,8 +648,8 @@ 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);
 
        return pkt_flags;
 }
@@ -768,19 +792,20 @@ eth_em_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                 */
                pkt_len = (uint16_t) (rte_le_to_cpu_16(rxd.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->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->ol_flags = rx_desc_status_to_pkt_flags(status);
-               rxm->ol_flags |= rx_desc_error_to_pkt_flags(rxd.errors);
+               rxm->ol_flags = (uint16_t)(rxm->ol_flags |
+                               rx_desc_error_to_pkt_flags(rxd.errors));
 
                /* Only valid if PKT_RX_VLAN_PKT set in pkt_flags */
-               rxm->pkt.vlan_macip.f.vlan_tci = rte_le_to_cpu_16(rxd.special);
+               rxm->vlan_tci = rte_le_to_cpu_16(rxd.special);
 
                /*
                 * Store the mbuf address into the next entry of the array
@@ -936,8 +961,8 @@ eth_em_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.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,
@@ -949,12 +974,12 @@ eth_em_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;
                }
 
                /*
@@ -977,18 +1002,18 @@ eth_em_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);
                }
 
@@ -999,16 +1024,18 @@ eth_em_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                 *      - IP checksum flag,
                 *      - error flags.
                 */
-               first_seg->pkt.in_port = rxq->port_id;
+               first_seg->port = rxq->port_id;
 
                first_seg->ol_flags = rx_desc_status_to_pkt_flags(status);
-               first_seg->ol_flags |= rx_desc_error_to_pkt_flags(rxd.errors);
+               first_seg->ol_flags = (uint16_t)(first_seg->ol_flags |
+                                       rx_desc_error_to_pkt_flags(rxd.errors));
 
                /* Only valid if PKT_RX_VLAN_PKT set in pkt_flags */
-               rxm->pkt.vlan_macip.f.vlan_tci = rte_le_to_cpu_16(rxd.special);
+               rxm->vlan_tci = rte_le_to_cpu_16(rxd.special);
 
                /* 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
@@ -1088,14 +1115,19 @@ ring_dma_zone_reserve(struct rte_eth_dev *dev, const char *ring_name,
        const struct rte_memzone *mz;
        char z_name[RTE_MEMZONE_NAMESIZE];
 
-       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);
 
        if ((mz = rte_memzone_lookup(z_name)) != 0)
                return (mz);
 
-       return rte_memzone_reserve(z_name, (uint64_t) ring_size, socket_id, 0);
+#ifdef RTE_LIBRTE_XEN_DOM0
+       return rte_memzone_reserve_bounded(z_name, ring_size,
+                       socket_id, 0, CACHE_LINE_SIZE, RTE_PGSIZE_2M);
+#else
+       return rte_memzone_reserve(z_name, ring_size, socket_id, 0);
+#endif
 }
 
 static void
@@ -1192,28 +1224,27 @@ eth_em_tx_queue_setup(struct rte_eth_dev *dev,
 
        tx_free_thresh = tx_conf->tx_free_thresh;
        if (tx_free_thresh == 0)
-               tx_free_thresh = RTE_MIN(nb_desc / 4, DEFAULT_TX_FREE_THRESH);
+               tx_free_thresh = (uint16_t)RTE_MIN(nb_desc / 4,
+                                       DEFAULT_TX_FREE_THRESH);
 
        tx_rs_thresh = tx_conf->tx_rs_thresh;
        if (tx_rs_thresh == 0)
-               tx_rs_thresh = RTE_MIN(tx_free_thresh, DEFAULT_TX_RS_THRESH);
+               tx_rs_thresh = (uint16_t)RTE_MIN(tx_free_thresh,
+                                       DEFAULT_TX_RS_THRESH);
 
        if (tx_free_thresh >= (nb_desc - 3)) {
-               RTE_LOG(ERR, PMD,
-                       "tx_free_thresh must be less than the "
-                       "number of TX descriptors minus 3. "
-                       "(tx_free_thresh=%u port=%d queue=%d)\n",
-                       tx_free_thresh, dev->data->port_id, queue_idx);
+               RTE_LOG(ERR, PMD, "tx_free_thresh must be less than the "
+                       "number of TX descriptors minus 3. (tx_free_thresh=%u "
+                       "port=%d queue=%d)\n", (unsigned int)tx_free_thresh,
+                               (int)dev->data->port_id, (int)queue_idx);
                return -(EINVAL);
        }
        if (tx_rs_thresh > tx_free_thresh) {
-               RTE_LOG(ERR, PMD,
-                       "tx_rs_thresh must be less than or equal to "
-                       "tx_free_thresh. "
-                       "(tx_free_thresh=%u tx_rs_thresh=%u "
-                       "port=%d queue=%d)\n",
-                       tx_free_thresh, tx_rs_thresh, dev->data->port_id,
-                       queue_idx);
+               RTE_LOG(ERR, PMD, "tx_rs_thresh must be less than or equal to "
+                       "tx_free_thresh. (tx_free_thresh=%u tx_rs_thresh=%u "
+                       "port=%d queue=%d)\n", (unsigned int)tx_free_thresh,
+                       (unsigned int)tx_rs_thresh, (int)dev->data->port_id,
+                                                       (int)queue_idx);
                return -(EINVAL);
        }
 
@@ -1224,11 +1255,10 @@ eth_em_tx_queue_setup(struct rte_eth_dev *dev,
         * accumulates WTHRESH descriptors.
         */
        if (tx_conf->tx_thresh.wthresh != 0 && tx_rs_thresh != 1) {
-               RTE_LOG(ERR, PMD,
-                       "TX WTHRESH must be set to 0 if "
-                       "tx_rs_thresh is greater than 1. "
-                       "(tx_rs_thresh=%u port=%d queue=%d)\n",
-                       tx_rs_thresh, dev->data->port_id, queue_idx);
+               RTE_LOG(ERR, PMD, "TX WTHRESH must be set to 0 if "
+                       "tx_rs_thresh is greater than 1. (tx_rs_thresh=%u "
+                       "port=%d queue=%d)\n", (unsigned int)tx_rs_thresh,
+                               (int)dev->data->port_id, (int)queue_idx);
                return -(EINVAL);
        }
 
@@ -1271,7 +1301,11 @@ eth_em_tx_queue_setup(struct rte_eth_dev *dev,
        txq->port_id = dev->data->port_id;
 
        txq->tdt_reg_addr = E1000_PCI_REG_ADDR(hw, E1000_TDT(queue_idx));
+#ifndef RTE_LIBRTE_XEN_DOM0
        txq->tx_ring_phys_addr = (uint64_t) tz->phys_addr;
+#else
+       txq->tx_ring_phys_addr = rte_mem_phy2mch(tz->memseg_id, tz->phys_addr);
+#endif
        txq->tx_ring = (struct e1000_data_desc *) tz->addr;
 
        PMD_INIT_LOG(DEBUG, "sw_ring=%p hw_ring=%p dma_addr=0x%"PRIx64"\n",
@@ -1395,7 +1429,12 @@ eth_em_rx_queue_setup(struct rte_eth_dev *dev,
                                0 : ETHER_CRC_LEN);
 
        rxq->rdt_reg_addr = E1000_PCI_REG_ADDR(hw, E1000_RDT(queue_idx));
+       rxq->rdh_reg_addr = E1000_PCI_REG_ADDR(hw, E1000_RDH(queue_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 = (struct e1000_rx_desc *) rz->addr;
 
        PMD_INIT_LOG(DEBUG, "sw_ring=%p hw_ring=%p dma_addr=0x%"PRIx64"\n",
@@ -1407,6 +1446,51 @@ eth_em_rx_queue_setup(struct rte_eth_dev *dev,
        return (0);
 }
 
+uint32_t
+eth_em_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
+{
+#define EM_RXQ_SCAN_INTERVAL 4
+       volatile struct e1000_rx_desc *rxdp;
+       struct em_rx_queue *rxq;
+       uint32_t desc = 0;
+
+       if (rx_queue_id >= dev->data->nb_rx_queues) {
+               PMD_RX_LOG(DEBUG,"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->status & E1000_RXD_STAT_DD)) {
+               desc += EM_RXQ_SCAN_INTERVAL;
+               rxdp += EM_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 desc;
+}
+
+int
+eth_em_rx_descriptor_done(void *rx_queue, uint16_t offset)
+{
+       volatile struct e1000_rx_desc *rxdp;
+       struct em_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->status & E1000_RXD_STAT_DD);
+}
+
 void
 em_dev_clear_queues(struct rte_eth_dev *dev)
 {
@@ -1436,7 +1520,7 @@ em_dev_clear_queues(struct rte_eth_dev *dev)
  * Returns (BSIZE | BSEX | FLXBUF) fields of RCTL register.
  */
 static uint32_t
-em_rctl_bsize(enum e1000_mac_type hwtyp, uint32_t *bufsz)
+em_rctl_bsize(__rte_unused enum e1000_mac_type hwtyp, uint32_t *bufsz)
 {
        /*
         * For BSIZE & BSEX all configurable sizes are:
@@ -1518,7 +1602,6 @@ em_alloc_rx_queue_mbufs(struct em_rx_queue *rxq)
                if (mbuf == NULL) {
                        PMD_INIT_LOG(ERR, "RX mbuf alloc failed "
                                "queue_id=%hu\n", rxq->queue_id);
-                       em_rx_queue_release(rxq);
                        return (-ENOMEM);
                }
 
@@ -1641,7 +1724,7 @@ eth_em_rx_init(struct rte_eth_dev *dev)
                 * limit for packet length, jumbo frame of any size
                 * can be accepted, thus we have to enable scattered
                 * rx if jumbo frames are enabled (or if buffer size
-                * is too small to accomodate non-jumbo packets)
+                * is too small to accommodate non-jumbo packets)
                 * to avoid splitting packets that don't fit into
                 * one buffer.
                 */
@@ -1653,6 +1736,11 @@ eth_em_rx_init(struct rte_eth_dev *dev)
                }
        }
 
+       if (dev->data->dev_conf.rxmode.enable_scatter) {
+               dev->rx_pkt_burst = eth_em_recv_scattered_pkts;
+               dev->data->scattered_rx = 1;
+       }
+
        /*
         * Setup the Checksum Register.
         * Receive Full-Packet Checksum Offload is mutually exclusive with RSS.