net/txgbe: support Rx
authorJiawen Wu <jiawenwu@trustnetic.com>
Mon, 19 Oct 2020 08:53:44 +0000 (16:53 +0800)
committerFerruh Yigit <ferruh.yigit@intel.com>
Tue, 3 Nov 2020 22:24:27 +0000 (23:24 +0100)
Fill receive functions and define receive descriptor.

Signed-off-by: Jiawen Wu <jiawenwu@trustnetic.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
doc/guides/nics/features/txgbe.ini
doc/guides/nics/txgbe.rst
drivers/net/txgbe/base/txgbe_type.h
drivers/net/txgbe/txgbe_ethdev.c
drivers/net/txgbe/txgbe_ethdev.h
drivers/net/txgbe/txgbe_rxtx.c
drivers/net/txgbe/txgbe_rxtx.h

index f77afa0..c8cd58c 100644 (file)
@@ -10,6 +10,7 @@ Link status event    = Y
 Queue start/stop     = Y
 Jumbo frame          = Y
 Scattered Rx         = Y
+LRO                  = Y
 TSO                  = Y
 Unicast MAC filter   = Y
 Multicast MAC filter = Y
index 0bec2ca..717c540 100644 (file)
@@ -17,7 +17,8 @@ Features
 - TSO offload
 - Jumbo frames
 - Link state information
-- Scattered and gather for TX
+- Scattered and gather for TX and RX
+- LRO
 
 Prerequisites
 -------------
index 5237200..622c5c6 100644 (file)
@@ -38,6 +38,8 @@
 #define TXGBE_PHYSICAL_LAYER_10BASE_T          0x08000
 #define TXGBE_PHYSICAL_LAYER_2500BASE_KX       0x10000
 
+#define TXGBE_ATR_HASH_MASK                    0x7fff
+
 enum txgbe_eeprom_type {
        txgbe_eeprom_unknown = 0,
        txgbe_eeprom_spi,
index 7f32ddf..28bbd82 100644 (file)
@@ -116,6 +116,7 @@ eth_txgbe_dev_init(struct rte_eth_dev *eth_dev, void *init_params __rte_unused)
        PMD_INIT_FUNC_TRACE();
 
        eth_dev->dev_ops = &txgbe_eth_dev_ops;
+       eth_dev->rx_pkt_burst = &txgbe_recv_pkts;
        eth_dev->tx_pkt_burst = &txgbe_xmit_pkts;
        eth_dev->tx_pkt_prepare = &txgbe_prep_pkts;
 
@@ -686,6 +687,18 @@ txgbe_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
        return 0;
 }
 
+const uint32_t *
+txgbe_dev_supported_ptypes_get(struct rte_eth_dev *dev)
+{
+       if (dev->rx_pkt_burst == txgbe_recv_pkts ||
+           dev->rx_pkt_burst == txgbe_recv_pkts_lro_single_alloc ||
+           dev->rx_pkt_burst == txgbe_recv_pkts_lro_bulk_alloc ||
+           dev->rx_pkt_burst == txgbe_recv_pkts_bulk_alloc)
+               return txgbe_get_supported_ptypes();
+
+       return NULL;
+}
+
 void
 txgbe_dev_setup_link_alarm_handler(void *param)
 {
@@ -1349,6 +1362,7 @@ static const struct eth_dev_ops txgbe_eth_dev_ops = {
        .dev_infos_get              = txgbe_dev_info_get,
        .dev_set_link_up            = txgbe_dev_set_link_up,
        .dev_set_link_down          = txgbe_dev_set_link_down,
+       .dev_supported_ptypes_get   = txgbe_dev_supported_ptypes_get,
        .rx_queue_start             = txgbe_dev_rx_queue_start,
        .rx_queue_stop              = txgbe_dev_rx_queue_stop,
        .tx_queue_start             = txgbe_dev_tx_queue_start,
index 12d1761..c01f312 100644 (file)
@@ -113,6 +113,17 @@ int txgbe_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id);
 
 int txgbe_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id);
 
+uint16_t txgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
+               uint16_t nb_pkts);
+
+uint16_t txgbe_recv_pkts_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
+                                   uint16_t nb_pkts);
+
+uint16_t txgbe_recv_pkts_lro_single_alloc(void *rx_queue,
+               struct rte_mbuf **rx_pkts, uint16_t nb_pkts);
+uint16_t txgbe_recv_pkts_lro_bulk_alloc(void *rx_queue,
+               struct rte_mbuf **rx_pkts, uint16_t nb_pkts);
+
 uint16_t txgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                uint16_t nb_pkts);
 
@@ -146,6 +157,7 @@ txgbe_dev_link_update_share(struct rte_eth_dev *dev,
 #define TXGBE_DEFAULT_TX_HTHRESH      0
 #define TXGBE_DEFAULT_TX_WTHRESH      0
 
+const uint32_t *txgbe_dev_supported_ptypes_get(struct rte_eth_dev *dev);
 int txgbe_dev_set_mc_addr_list(struct rte_eth_dev *dev,
                                      struct rte_ether_addr *mc_addr_set,
                                      uint32_t nb_mc_addr);
index fd5708c..82d0104 100644 (file)
@@ -8,6 +8,10 @@
 #include <stdlib.h>
 #include <string.h>
 #include <errno.h>
+#include <stdint.h>
+#include <stdarg.h>
+#include <unistd.h>
+#include <inttypes.h>
 
 #include <rte_common.h>
 #include <rte_cycles.h>
 #include <rte_ethdev.h>
 #include <rte_ethdev_driver.h>
 #include <rte_memzone.h>
+#include <rte_atomic.h>
 #include <rte_mempool.h>
 #include <rte_malloc.h>
 #include <rte_mbuf.h>
+#include <rte_ether.h>
+#include <rte_prefetch.h>
+#include <rte_udp.h>
+#include <rte_tcp.h>
+#include <rte_sctp.h>
+#include <rte_string_fns.h>
+#include <rte_errno.h>
 #include <rte_ip.h>
 #include <rte_net.h>
 
@@ -42,6 +54,11 @@ static const u64 TXGBE_TX_OFFLOAD_MASK = (PKT_TX_IP_CKSUM |
 #define TXGBE_TX_OFFLOAD_NOTSUP_MASK \
                (PKT_TX_OFFLOAD_MASK ^ TXGBE_TX_OFFLOAD_MASK)
 
+/*
+ * Prefetch a cache line into all cache levels.
+ */
+#define rte_txgbe_prefetch(p)   rte_prefetch0(p)
+
 static int
 txgbe_is_vf(struct rte_eth_dev *dev)
 {
@@ -985,6 +1002,862 @@ txgbe_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
        return i;
 }
 
+/*********************************************************************
+ *
+ *  RX functions
+ *
+ **********************************************************************/
+/* @note: fix txgbe_dev_supported_ptypes_get() if any change here. */
+static inline uint32_t
+txgbe_rxd_pkt_info_to_pkt_type(uint32_t pkt_info, uint16_t ptid_mask)
+{
+       uint16_t ptid = TXGBE_RXD_PTID(pkt_info);
+
+       ptid &= ptid_mask;
+
+       return txgbe_decode_ptype(ptid);
+}
+
+static inline uint64_t
+txgbe_rxd_pkt_info_to_pkt_flags(uint32_t pkt_info)
+{
+       static uint64_t ip_rss_types_map[16] __rte_cache_aligned = {
+               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,
+               0, 0, 0,  PKT_RX_FDIR,
+       };
+
+       return ip_rss_types_map[TXGBE_RXD_RSSTYPE(pkt_info)];
+}
+
+static inline uint64_t
+rx_desc_status_to_pkt_flags(uint32_t rx_status, uint64_t vlan_flags)
+{
+       uint64_t pkt_flags;
+
+       /*
+        * Check if VLAN present only.
+        * Do not check whether L3/L4 rx checksum done by NIC or not,
+        * That can be found from rte_eth_rxmode.offloads flag
+        */
+       pkt_flags = (rx_status & TXGBE_RXD_STAT_VLAN &&
+                    vlan_flags & PKT_RX_VLAN_STRIPPED)
+                   ? vlan_flags : 0;
+
+       return pkt_flags;
+}
+
+static inline uint64_t
+rx_desc_error_to_pkt_flags(uint32_t rx_status)
+{
+       uint64_t pkt_flags = 0;
+
+       /* checksum offload can't be disabled */
+       if (rx_status & TXGBE_RXD_STAT_IPCS) {
+               pkt_flags |= (rx_status & TXGBE_RXD_ERR_IPCS
+                               ? PKT_RX_IP_CKSUM_BAD : PKT_RX_IP_CKSUM_GOOD);
+       }
+
+       if (rx_status & TXGBE_RXD_STAT_L4CS) {
+               pkt_flags |= (rx_status & TXGBE_RXD_ERR_L4CS
+                               ? PKT_RX_L4_CKSUM_BAD : PKT_RX_L4_CKSUM_GOOD);
+       }
+
+       if (rx_status & TXGBE_RXD_STAT_EIPCS &&
+           rx_status & TXGBE_RXD_ERR_EIPCS) {
+               pkt_flags |= PKT_RX_EIP_CKSUM_BAD;
+       }
+
+       return pkt_flags;
+}
+
+/*
+ * LOOK_AHEAD defines how many desc statuses to check beyond the
+ * current descriptor.
+ * It must be a pound define for optimal performance.
+ * Do not change the value of LOOK_AHEAD, as the txgbe_rx_scan_hw_ring
+ * function only works with LOOK_AHEAD=8.
+ */
+#define LOOK_AHEAD 8
+#if (LOOK_AHEAD != 8)
+#error "PMD TXGBE: LOOK_AHEAD must be 8\n"
+#endif
+static inline int
+txgbe_rx_scan_hw_ring(struct txgbe_rx_queue *rxq)
+{
+       volatile struct txgbe_rx_desc *rxdp;
+       struct txgbe_rx_entry *rxep;
+       struct rte_mbuf *mb;
+       uint16_t pkt_len;
+       uint64_t pkt_flags;
+       int nb_dd;
+       uint32_t s[LOOK_AHEAD];
+       uint32_t pkt_info[LOOK_AHEAD];
+       int i, j, nb_rx = 0;
+       uint32_t status;
+
+       /* get references to current descriptor and S/W ring entry */
+       rxdp = &rxq->rx_ring[rxq->rx_tail];
+       rxep = &rxq->sw_ring[rxq->rx_tail];
+
+       status = rxdp->qw1.lo.status;
+       /* check to make sure there is at least 1 packet to receive */
+       if (!(status & rte_cpu_to_le_32(TXGBE_RXD_STAT_DD)))
+               return 0;
+
+       /*
+        * Scan LOOK_AHEAD descriptors at a time to determine which descriptors
+        * reference packets that are ready to be received.
+        */
+       for (i = 0; i < RTE_PMD_TXGBE_RX_MAX_BURST;
+            i += LOOK_AHEAD, rxdp += LOOK_AHEAD, rxep += LOOK_AHEAD) {
+               /* Read desc statuses backwards to avoid race condition */
+               for (j = 0; j < LOOK_AHEAD; j++)
+                       s[j] = rte_le_to_cpu_32(rxdp[j].qw1.lo.status);
+
+               rte_smp_rmb();
+
+               /* Compute how many status bits were set */
+               for (nb_dd = 0; nb_dd < LOOK_AHEAD &&
+                               (s[nb_dd] & TXGBE_RXD_STAT_DD); nb_dd++)
+                       ;
+
+               for (j = 0; j < nb_dd; j++)
+                       pkt_info[j] = rte_le_to_cpu_32(rxdp[j].qw0.dw0);
+
+               nb_rx += nb_dd;
+
+               /* Translate descriptor info to mbuf format */
+               for (j = 0; j < nb_dd; ++j) {
+                       mb = rxep[j].mbuf;
+                       pkt_len = rte_le_to_cpu_16(rxdp[j].qw1.hi.len) -
+                                 rxq->crc_len;
+                       mb->data_len = pkt_len;
+                       mb->pkt_len = pkt_len;
+                       mb->vlan_tci = rte_le_to_cpu_16(rxdp[j].qw1.hi.tag);
+
+                       /* convert descriptor fields to rte mbuf flags */
+                       pkt_flags = rx_desc_status_to_pkt_flags(s[j],
+                                       rxq->vlan_flags);
+                       pkt_flags |= rx_desc_error_to_pkt_flags(s[j]);
+                       pkt_flags |=
+                               txgbe_rxd_pkt_info_to_pkt_flags(pkt_info[j]);
+                       mb->ol_flags = pkt_flags;
+                       mb->packet_type =
+                               txgbe_rxd_pkt_info_to_pkt_type(pkt_info[j],
+                               rxq->pkt_type_mask);
+
+                       if (likely(pkt_flags & PKT_RX_RSS_HASH))
+                               mb->hash.rss =
+                                       rte_le_to_cpu_32(rxdp[j].qw0.dw1);
+                       else if (pkt_flags & PKT_RX_FDIR) {
+                               mb->hash.fdir.hash =
+                                       rte_le_to_cpu_16(rxdp[j].qw0.hi.csum) &
+                                       TXGBE_ATR_HASH_MASK;
+                               mb->hash.fdir.id =
+                                       rte_le_to_cpu_16(rxdp[j].qw0.hi.ipid);
+                       }
+               }
+
+               /* Move mbuf pointers from the S/W ring to the stage */
+               for (j = 0; j < LOOK_AHEAD; ++j)
+                       rxq->rx_stage[i + j] = rxep[j].mbuf;
+
+               /* stop if all requested packets could not be received */
+               if (nb_dd != LOOK_AHEAD)
+                       break;
+       }
+
+       /* clear software ring entries so we can cleanup correctly */
+       for (i = 0; i < nb_rx; ++i)
+               rxq->sw_ring[rxq->rx_tail + i].mbuf = NULL;
+
+       return nb_rx;
+}
+
+static inline int
+txgbe_rx_alloc_bufs(struct txgbe_rx_queue *rxq, bool reset_mbuf)
+{
+       volatile struct txgbe_rx_desc *rxdp;
+       struct txgbe_rx_entry *rxep;
+       struct rte_mbuf *mb;
+       uint16_t alloc_idx;
+       __le64 dma_addr;
+       int diag, i;
+
+       /* allocate buffers in bulk directly into the S/W ring */
+       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);
+       if (unlikely(diag != 0))
+               return -ENOMEM;
+
+       rxdp = &rxq->rx_ring[alloc_idx];
+       for (i = 0; i < rxq->rx_free_thresh; ++i) {
+               /* populate the static rte mbuf fields */
+               mb = rxep[i].mbuf;
+               if (reset_mbuf)
+                       mb->port = rxq->port_id;
+
+               rte_mbuf_refcnt_set(mb, 1);
+               mb->data_off = RTE_PKTMBUF_HEADROOM;
+
+               /* populate the descriptors */
+               dma_addr = rte_cpu_to_le_64(rte_mbuf_data_iova_default(mb));
+               TXGBE_RXD_HDRADDR(&rxdp[i], 0);
+               TXGBE_RXD_PKTADDR(&rxdp[i], dma_addr);
+       }
+
+       /* update state of internal queue structure */
+       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 = rxq->rx_free_thresh - 1;
+
+       /* no errors */
+       return 0;
+}
+
+static inline uint16_t
+txgbe_rx_fill_from_stage(struct txgbe_rx_queue *rxq, struct rte_mbuf **rx_pkts,
+                        uint16_t nb_pkts)
+{
+       struct rte_mbuf **stage = &rxq->rx_stage[rxq->rx_next_avail];
+       int i;
+
+       /* how many packets are ready to return? */
+       nb_pkts = (uint16_t)RTE_MIN(nb_pkts, rxq->rx_nb_avail);
+
+       /* copy mbuf pointers to the application's packet list */
+       for (i = 0; i < nb_pkts; ++i)
+               rx_pkts[i] = stage[i];
+
+       /* update internal queue state */
+       rxq->rx_nb_avail = (uint16_t)(rxq->rx_nb_avail - nb_pkts);
+       rxq->rx_next_avail = (uint16_t)(rxq->rx_next_avail + nb_pkts);
+
+       return nb_pkts;
+}
+
+static inline uint16_t
+txgbe_rx_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
+            uint16_t nb_pkts)
+{
+       struct txgbe_rx_queue *rxq = (struct txgbe_rx_queue *)rx_queue;
+       struct rte_eth_dev *dev = &rte_eth_devices[rxq->port_id];
+       uint16_t nb_rx = 0;
+
+       /* Any previously recv'd pkts will be returned from the Rx stage */
+       if (rxq->rx_nb_avail)
+               return txgbe_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
+
+       /* Scan the H/W ring for packets to receive */
+       nb_rx = (uint16_t)txgbe_rx_scan_hw_ring(rxq);
+
+       /* update internal queue state */
+       rxq->rx_next_avail = 0;
+       rxq->rx_nb_avail = nb_rx;
+       rxq->rx_tail = (uint16_t)(rxq->rx_tail + nb_rx);
+
+       /* if required, allocate new buffers to replenish descriptors */
+       if (rxq->rx_tail > rxq->rx_free_trigger) {
+               uint16_t cur_free_trigger = rxq->rx_free_trigger;
+
+               if (txgbe_rx_alloc_bufs(rxq, true) != 0) {
+                       int i, j;
+
+                       PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
+                                  "queue_id=%u", (uint16_t)rxq->port_id,
+                                  (uint16_t)rxq->queue_id);
+
+                       dev->data->rx_mbuf_alloc_failed +=
+                               rxq->rx_free_thresh;
+
+                       /*
+                        * Need to rewind any previous receives if we cannot
+                        * allocate new buffers to replenish the old ones.
+                        */
+                       rxq->rx_nb_avail = 0;
+                       rxq->rx_tail = (uint16_t)(rxq->rx_tail - nb_rx);
+                       for (i = 0, j = rxq->rx_tail; i < nb_rx; ++i, ++j)
+                               rxq->sw_ring[j].mbuf = rxq->rx_stage[i];
+
+                       return 0;
+               }
+
+               /* update tail pointer */
+               rte_wmb();
+               txgbe_set32_relaxed(rxq->rdt_reg_addr, cur_free_trigger);
+       }
+
+       if (rxq->rx_tail >= rxq->nb_rx_desc)
+               rxq->rx_tail = 0;
+
+       /* received any packets this loop? */
+       if (rxq->rx_nb_avail)
+               return txgbe_rx_fill_from_stage(rxq, rx_pkts, nb_pkts);
+
+       return 0;
+}
+
+/* split requests into chunks of size RTE_PMD_TXGBE_RX_MAX_BURST */
+uint16_t
+txgbe_recv_pkts_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
+                          uint16_t nb_pkts)
+{
+       uint16_t nb_rx;
+
+       if (unlikely(nb_pkts == 0))
+               return 0;
+
+       if (likely(nb_pkts <= RTE_PMD_TXGBE_RX_MAX_BURST))
+               return txgbe_rx_recv_pkts(rx_queue, rx_pkts, nb_pkts);
+
+       /* request is relatively large, chunk it up */
+       nb_rx = 0;
+       while (nb_pkts) {
+               uint16_t ret, n;
+
+               n = (uint16_t)RTE_MIN(nb_pkts, RTE_PMD_TXGBE_RX_MAX_BURST);
+               ret = txgbe_rx_recv_pkts(rx_queue, &rx_pkts[nb_rx], n);
+               nb_rx = (uint16_t)(nb_rx + ret);
+               nb_pkts = (uint16_t)(nb_pkts - ret);
+               if (ret < n)
+                       break;
+       }
+
+       return nb_rx;
+}
+
+uint16_t
+txgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
+               uint16_t nb_pkts)
+{
+       struct txgbe_rx_queue *rxq;
+       volatile struct txgbe_rx_desc *rx_ring;
+       volatile struct txgbe_rx_desc *rxdp;
+       struct txgbe_rx_entry *sw_ring;
+       struct txgbe_rx_entry *rxe;
+       struct rte_mbuf *rxm;
+       struct rte_mbuf *nmb;
+       struct txgbe_rx_desc rxd;
+       uint64_t dma_addr;
+       uint32_t staterr;
+       uint32_t pkt_info;
+       uint16_t pkt_len;
+       uint16_t rx_id;
+       uint16_t nb_rx;
+       uint16_t nb_hold;
+       uint64_t pkt_flags;
+
+       nb_rx = 0;
+       nb_hold = 0;
+       rxq = rx_queue;
+       rx_id = rxq->rx_tail;
+       rx_ring = rxq->rx_ring;
+       sw_ring = rxq->sw_ring;
+       struct rte_eth_dev *dev = &rte_eth_devices[rxq->port_id];
+       while (nb_rx < nb_pkts) {
+               /*
+                * The order of operations here is important as the DD status
+                * bit must not be read after any other descriptor fields.
+                * rx_ring and rxdp are pointing to volatile data so the order
+                * of accesses cannot be reordered by the compiler. If they were
+                * not volatile, they could be reordered which could lead to
+                * using invalid descriptor fields when read from rxd.
+                */
+               rxdp = &rx_ring[rx_id];
+               staterr = rxdp->qw1.lo.status;
+               if (!(staterr & rte_cpu_to_le_32(TXGBE_RXD_STAT_DD)))
+                       break;
+               rxd = *rxdp;
+
+               /*
+                * End of packet.
+                *
+                * If the TXGBE_RXD_STAT_EOP flag is not set, the RX packet
+                * is likely to be invalid and to be dropped by the various
+                * validation checks performed by the network stack.
+                *
+                * Allocate a new mbuf to replenish the RX ring descriptor.
+                * If the allocation fails:
+                *    - arrange for that RX descriptor to be the first one
+                *      being parsed the next time the receive function is
+                *      invoked [on the same queue].
+                *
+                *    - Stop parsing the RX ring and return immediately.
+                *
+                * This policy do not drop the packet received in the RX
+                * descriptor for which the allocation of a new mbuf failed.
+                * Thus, it allows that packet to be later retrieved if
+                * mbuf have been freed in the mean time.
+                * As a side effect, holding RX descriptors instead of
+                * systematically giving them back to the NIC may lead to
+                * RX ring exhaustion situations.
+                * However, the NIC can gracefully prevent such situations
+                * to happen by sending specific "back-pressure" flow control
+                * frames to its peer(s).
+                */
+               PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_id=%u "
+                          "ext_err_stat=0x%08x pkt_len=%u",
+                          (uint16_t)rxq->port_id, (uint16_t)rxq->queue_id,
+                          (uint16_t)rx_id, (uint32_t)staterr,
+                          (uint16_t)rte_le_to_cpu_16(rxd.qw1.hi.len));
+
+               nmb = rte_mbuf_raw_alloc(rxq->mb_pool);
+               if (nmb == NULL) {
+                       PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
+                                  "queue_id=%u", (uint16_t)rxq->port_id,
+                                  (uint16_t)rxq->queue_id);
+                       dev->data->rx_mbuf_alloc_failed++;
+                       break;
+               }
+
+               nb_hold++;
+               rxe = &sw_ring[rx_id];
+               rx_id++;
+               if (rx_id == rxq->nb_rx_desc)
+                       rx_id = 0;
+
+               /* Prefetch next mbuf while processing current one. */
+               rte_txgbe_prefetch(sw_ring[rx_id].mbuf);
+
+               /*
+                * When next RX descriptor is on a cache-line boundary,
+                * prefetch the next 4 RX descriptors and the next 8 pointers
+                * to mbufs.
+                */
+               if ((rx_id & 0x3) == 0) {
+                       rte_txgbe_prefetch(&rx_ring[rx_id]);
+                       rte_txgbe_prefetch(&sw_ring[rx_id]);
+               }
+
+               rxm = rxe->mbuf;
+               rxe->mbuf = nmb;
+               dma_addr = rte_cpu_to_le_64(rte_mbuf_data_iova_default(nmb));
+               TXGBE_RXD_HDRADDR(rxdp, 0);
+               TXGBE_RXD_PKTADDR(rxdp, dma_addr);
+
+               /*
+                * Initialize the returned mbuf.
+                * 1) setup generic mbuf fields:
+                *    - number of segments,
+                *    - next segment,
+                *    - packet length,
+                *    - RX port identifier.
+                * 2) integrate hardware offload data, if any:
+                *    - RSS flag & hash,
+                *    - IP checksum flag,
+                *    - VLAN TCI, if any,
+                *    - error flags.
+                */
+               pkt_len = (uint16_t)(rte_le_to_cpu_16(rxd.qw1.hi.len) -
+                                     rxq->crc_len);
+               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;
+
+               pkt_info = rte_le_to_cpu_32(rxd.qw0.dw0);
+               /* Only valid if PKT_RX_VLAN set in pkt_flags */
+               rxm->vlan_tci = rte_le_to_cpu_16(rxd.qw1.hi.tag);
+
+               pkt_flags = rx_desc_status_to_pkt_flags(staterr,
+                                       rxq->vlan_flags);
+               pkt_flags |= rx_desc_error_to_pkt_flags(staterr);
+               pkt_flags |= txgbe_rxd_pkt_info_to_pkt_flags(pkt_info);
+               rxm->ol_flags = pkt_flags;
+               rxm->packet_type = txgbe_rxd_pkt_info_to_pkt_type(pkt_info,
+                                                      rxq->pkt_type_mask);
+
+               if (likely(pkt_flags & PKT_RX_RSS_HASH)) {
+                       rxm->hash.rss = rte_le_to_cpu_32(rxd.qw0.dw1);
+               } else if (pkt_flags & PKT_RX_FDIR) {
+                       rxm->hash.fdir.hash =
+                               rte_le_to_cpu_16(rxd.qw0.hi.csum) &
+                               TXGBE_ATR_HASH_MASK;
+                       rxm->hash.fdir.id = rte_le_to_cpu_16(rxd.qw0.hi.ipid);
+               }
+               /*
+                * Store the mbuf address into the next entry of the array
+                * of returned packets.
+                */
+               rx_pkts[nb_rx++] = rxm;
+       }
+       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 situation from the
+        * hardware point of view...
+        */
+       nb_hold = (uint16_t)(nb_hold + rxq->nb_rx_hold);
+       if (nb_hold > rxq->rx_free_thresh) {
+               PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
+                          "nb_hold=%u nb_rx=%u",
+                          (uint16_t)rxq->port_id, (uint16_t)rxq->queue_id,
+                          (uint16_t)rx_id, (uint16_t)nb_hold,
+                          (uint16_t)nb_rx);
+               rx_id = (uint16_t)((rx_id == 0) ?
+                               (rxq->nb_rx_desc - 1) : (rx_id - 1));
+               txgbe_set32(rxq->rdt_reg_addr, rx_id);
+               nb_hold = 0;
+       }
+       rxq->nb_rx_hold = nb_hold;
+       return nb_rx;
+}
+
+/**
+ * txgbe_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
+ * @rxq Pointer to the Rx queue
+ */
+static inline void
+txgbe_fill_cluster_head_buf(struct rte_mbuf *head, struct txgbe_rx_desc *desc,
+               struct txgbe_rx_queue *rxq, uint32_t staterr)
+{
+       uint32_t pkt_info;
+       uint64_t pkt_flags;
+
+       head->port = rxq->port_id;
+
+       /* The vlan_tci field is only valid when PKT_RX_VLAN is
+        * set in the pkt_flags field.
+        */
+       head->vlan_tci = rte_le_to_cpu_16(desc->qw1.hi.tag);
+       pkt_info = rte_le_to_cpu_32(desc->qw0.dw0);
+       pkt_flags = rx_desc_status_to_pkt_flags(staterr, rxq->vlan_flags);
+       pkt_flags |= rx_desc_error_to_pkt_flags(staterr);
+       pkt_flags |= txgbe_rxd_pkt_info_to_pkt_flags(pkt_info);
+       head->ol_flags = pkt_flags;
+       head->packet_type = txgbe_rxd_pkt_info_to_pkt_type(pkt_info,
+                                               rxq->pkt_type_mask);
+
+       if (likely(pkt_flags & PKT_RX_RSS_HASH)) {
+               head->hash.rss = rte_le_to_cpu_32(desc->qw0.dw1);
+       } else if (pkt_flags & PKT_RX_FDIR) {
+               head->hash.fdir.hash = rte_le_to_cpu_16(desc->qw0.hi.csum)
+                               & TXGBE_ATR_HASH_MASK;
+               head->hash.fdir.id = rte_le_to_cpu_16(desc->qw0.hi.ipid);
+       }
+}
+
+/**
+ * txgbe_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 txgbe_rsc_entry's that will hold the relevant RSC info.
+ *
+ * We use the same logic as in Linux and in FreeBSD txgbe 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
+txgbe_recv_pkts_lro(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts,
+                   bool bulk_alloc)
+{
+       struct txgbe_rx_queue *rxq = rx_queue;
+       struct rte_eth_dev *dev = &rte_eth_devices[rxq->port_id];
+       volatile struct txgbe_rx_desc *rx_ring = rxq->rx_ring;
+       struct txgbe_rx_entry *sw_ring = rxq->sw_ring;
+       struct txgbe_scattered_rx_entry *sw_sc_ring = rxq->sw_sc_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 txgbe_rx_entry *rxe;
+               struct txgbe_scattered_rx_entry *sc_entry;
+               struct txgbe_scattered_rx_entry *next_sc_entry = NULL;
+               struct txgbe_rx_entry *next_rxe = NULL;
+               struct rte_mbuf *first_seg;
+               struct rte_mbuf *rxm;
+               struct rte_mbuf *nmb = NULL;
+               struct txgbe_rx_desc rxd;
+               uint16_t data_len;
+               uint16_t next_id;
+               volatile struct txgbe_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 txgbe PMD.
+                *
+                * TODO:
+                *    - Get rid of "volatile" 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->qw1.lo.status);
+
+               if (!(staterr & TXGBE_RXD_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.qw1.hi.len));
+
+               if (!bulk_alloc) {
+                       nmb = rte_mbuf_raw_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);
+
+                               dev->data->rx_mbuf_alloc_failed++;
+                               break;
+                       }
+               } else if (nb_hold > rxq->rx_free_thresh) {
+                       uint16_t next_rdt = rxq->rx_free_trigger;
+
+                       if (!txgbe_rx_alloc_bufs(rxq, false)) {
+                               rte_wmb();
+                               txgbe_set32_relaxed(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);
+
+                               dev->data->rx_mbuf_alloc_failed++;
+                               break;
+                       }
+               }
+
+               nb_hold++;
+               rxe = &sw_ring[rx_id];
+               eop = staterr & TXGBE_RXD_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_txgbe_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_txgbe_prefetch(&rx_ring[next_id]);
+                       rte_txgbe_prefetch(&sw_ring[next_id]);
+               }
+
+               rxm = rxe->mbuf;
+
+               if (!bulk_alloc) {
+                       __le64 dma =
+                         rte_cpu_to_le_64(rte_mbuf_data_iova_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;
+                       TXGBE_RXD_HDRADDR(rxdp, 0);
+                       TXGBE_RXD_PKTADDR(rxdp, dma);
+               } else {
+                       rxe->mbuf = NULL;
+               }
+
+               /*
+                * Set data length & data buffer address of mbuf.
+                */
+               data_len = rte_le_to_cpu_16(rxd.qw1.hi.len);
+               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 (TXGBE_RXD_RSCCNT(rxd.qw0.dw0))
+                               nextp_id = TXGBE_RXD_NEXTP(staterr);
+                       else
+                               nextp_id = next_id;
+
+                       next_sc_entry = &sw_sc_ring[nextp_id];
+                       next_rxe = &sw_ring[nextp_id];
+                       rte_txgbe_prefetch(next_rxe);
+               }
+
+               sc_entry = &sw_sc_ring[rx_id];
+               first_seg = sc_entry->fbuf;
+               sc_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_sc_ring and continue to parse the RX ring.
+                */
+               if (!eop && next_rxe) {
+                       rxm->next = next_rxe->mbuf;
+                       next_sc_entry->fbuf = first_seg;
+                       goto next_desc;
+               }
+
+               /* Initialize the first mbuf of the returned packet */
+               txgbe_fill_cluster_head_buf(first_seg, &rxd, rxq, staterr);
+
+               /*
+                * Deal with the case, when HW CRC srip is disabled.
+                * That can't happen when LRO is enabled, but still could
+                * happen for scattered RX mode.
+                */
+               first_seg->pkt_len -= rxq->crc_len;
+               if (unlikely(rxm->data_len <= rxq->crc_len)) {
+                       struct rte_mbuf *lp;
+
+                       for (lp = first_seg; lp->next != rxm; lp = lp->next)
+                               ;
+
+                       first_seg->nb_segs--;
+                       lp->data_len -= rxq->crc_len - rxm->data_len;
+                       lp->next = NULL;
+                       rte_pktmbuf_free_seg(rxm);
+               } else {
+                       rxm->data_len -= rxq->crc_len;
+               }
+
+               /* 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 situation 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();
+               txgbe_set32_relaxed(rxq->rdt_reg_addr, prev_id);
+               nb_hold = 0;
+       }
+
+       rxq->nb_rx_hold = nb_hold;
+       return nb_rx;
+}
+
+uint16_t
+txgbe_recv_pkts_lro_single_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
+                                uint16_t nb_pkts)
+{
+       return txgbe_recv_pkts_lro(rx_queue, rx_pkts, nb_pkts, false);
+}
+
+uint16_t
+txgbe_recv_pkts_lro_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
+                              uint16_t nb_pkts)
+{
+       return txgbe_recv_pkts_lro(rx_queue, rx_pkts, nb_pkts, true);
+}
+
 uint64_t
 txgbe_get_rx_queue_offloads(struct rte_eth_dev *dev __rte_unused)
 {
@@ -1608,12 +2481,6 @@ txgbe_dev_free_queues(struct rte_eth_dev *dev)
        dev->data->nb_tx_queues = 0;
 }
 
-void __rte_cold
-txgbe_set_rx_function(struct rte_eth_dev *dev)
-{
-       RTE_SET_USED(dev);
-}
-
 static int __rte_cold
 txgbe_alloc_rx_queue_mbufs(struct txgbe_rx_queue *rxq)
 {
@@ -1788,6 +2655,71 @@ txgbe_set_rsc(struct rte_eth_dev *dev)
        return 0;
 }
 
+void __rte_cold
+txgbe_set_rx_function(struct rte_eth_dev *dev)
+{
+       struct txgbe_adapter *adapter = TXGBE_DEV_ADAPTER(dev);
+
+       /*
+        * Initialize the appropriate LRO callback.
+        *
+        * If all queues satisfy the bulk allocation preconditions
+        * (adapter->rx_bulk_alloc_allowed is TRUE) then we may use
+        * bulk allocation. Otherwise use a single allocation version.
+        */
+       if (dev->data->lro) {
+               if (adapter->rx_bulk_alloc_allowed) {
+                       PMD_INIT_LOG(DEBUG, "LRO is requested. Using a bulk "
+                                          "allocation version");
+                       dev->rx_pkt_burst = txgbe_recv_pkts_lro_bulk_alloc;
+               } else {
+                       PMD_INIT_LOG(DEBUG, "LRO is requested. Using a single "
+                                          "allocation version");
+                       dev->rx_pkt_burst = txgbe_recv_pkts_lro_single_alloc;
+               }
+       } else if (dev->data->scattered_rx) {
+               /*
+                * Set the non-LRO scattered callback: there are bulk and
+                * single allocation versions.
+                */
+               if (adapter->rx_bulk_alloc_allowed) {
+                       PMD_INIT_LOG(DEBUG, "Using a Scattered with bulk "
+                                          "allocation callback (port=%d).",
+                                    dev->data->port_id);
+                       dev->rx_pkt_burst = txgbe_recv_pkts_lro_bulk_alloc;
+               } else {
+                       PMD_INIT_LOG(DEBUG, "Using Regular (non-vector, "
+                                           "single allocation) "
+                                           "Scattered Rx callback "
+                                           "(port=%d).",
+                                    dev->data->port_id);
+
+                       dev->rx_pkt_burst = txgbe_recv_pkts_lro_single_alloc;
+               }
+       /*
+        * Below we set "simple" callbacks according to port/queues parameters.
+        * If parameters allow we are going to choose between the following
+        * callbacks:
+        *    - Bulk Allocation
+        *    - Single buffer allocation (the simplest one)
+        */
+       } else if (adapter->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 = txgbe_recv_pkts_bulk_alloc;
+       } else {
+               PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are not "
+                                   "satisfied, or Scattered Rx is requested "
+                                   "(port=%d).",
+                            dev->data->port_id);
+
+               dev->rx_pkt_burst = txgbe_recv_pkts;
+       }
+}
+
 /*
  * Initializes Receive Unit.
  */
index b61382f..624eb73 100644 (file)
@@ -50,6 +50,100 @@ struct txgbe_rx_desc {
 #define TXGBE_RXD_HDRADDR(rxd, v)  \
        (((volatile __le64 *)(rxd))[1] = cpu_to_le64(v))
 
+/* @txgbe_rx_desc.dw0 */
+#define TXGBE_RXD_RSSTYPE(dw)      RS(dw, 0, 0xF)
+#define   TXGBE_RSSTYPE_NONE       0
+#define   TXGBE_RSSTYPE_IPV4TCP    1
+#define   TXGBE_RSSTYPE_IPV4       2
+#define   TXGBE_RSSTYPE_IPV6TCP    3
+#define   TXGBE_RSSTYPE_IPV4SCTP   4
+#define   TXGBE_RSSTYPE_IPV6       5
+#define   TXGBE_RSSTYPE_IPV6SCTP   6
+#define   TXGBE_RSSTYPE_IPV4UDP    7
+#define   TXGBE_RSSTYPE_IPV6UDP    8
+#define   TXGBE_RSSTYPE_FDIR       15
+#define TXGBE_RXD_SECTYPE(dw)      RS(dw, 4, 0x3)
+#define TXGBE_RXD_SECTYPE_NONE     LS(0, 4, 0x3)
+#define TXGBE_RXD_SECTYPE_LINKSEC  LS(1, 4, 0x3)
+#define TXGBE_RXD_SECTYPE_IPSECESP LS(2, 4, 0x3)
+#define TXGBE_RXD_SECTYPE_IPSECAH  LS(3, 4, 0x3)
+#define TXGBE_RXD_TPIDSEL(dw)      RS(dw, 6, 0x7)
+#define TXGBE_RXD_PTID(dw)         RS(dw, 9, 0xFF)
+#define TXGBE_RXD_RSCCNT(dw)       RS(dw, 17, 0xF)
+#define TXGBE_RXD_HDRLEN(dw)       RS(dw, 21, 0x3FF)
+#define TXGBE_RXD_SPH              MS(31, 0x1)
+
+/* @txgbe_rx_desc.dw1 */
+/** bit 0-31, as rss hash when  **/
+#define TXGBE_RXD_RSSHASH(rxd)     ((rxd)->qw0.dw1)
+
+/** bit 0-31, as ip csum when  **/
+#define TXGBE_RXD_IPID(rxd)        ((rxd)->qw0.hi.ipid)
+#define TXGBE_RXD_CSUM(rxd)        ((rxd)->qw0.hi.csum)
+
+/** bit 0-31, as fdir id when  **/
+#define TXGBE_RXD_FDIRID(rxd)      ((rxd)->qw0.hi.dw1)
+
+/* @txgbe_rx_desc.dw2 */
+#define TXGBE_RXD_STATUS(rxd)      ((rxd)->qw1.lo.status)
+/** bit 0-1 **/
+#define TXGBE_RXD_STAT_DD          MS(0, 0x1) /* Descriptor Done */
+#define TXGBE_RXD_STAT_EOP         MS(1, 0x1) /* End of Packet */
+/** bit 2-31, when EOP=0 **/
+#define TXGBE_RXD_NEXTP_RESV(v)    LS(v, 2, 0x3)
+#define TXGBE_RXD_NEXTP(dw)        RS(dw, 4, 0xFFFF) /* Next Descriptor */
+/** bit 2-31, when EOP=1 **/
+#define TXGBE_RXD_PKT_CLS_MASK     MS(2, 0x7) /* Packet Class */
+#define TXGBE_RXD_PKT_CLS_TC_RSS   LS(0, 2, 0x7) /* RSS Hash */
+#define TXGBE_RXD_PKT_CLS_FLM      LS(1, 2, 0x7) /* FDir Match */
+#define TXGBE_RXD_PKT_CLS_SYN      LS(2, 2, 0x7) /* TCP Sync */
+#define TXGBE_RXD_PKT_CLS_5TUPLE   LS(3, 2, 0x7) /* 5 Tuple */
+#define TXGBE_RXD_PKT_CLS_ETF      LS(4, 2, 0x7) /* Ethertype Filter */
+#define TXGBE_RXD_STAT_VLAN        MS(5, 0x1) /* IEEE VLAN Packet */
+#define TXGBE_RXD_STAT_UDPCS       MS(6, 0x1) /* UDP xsum calculated */
+#define TXGBE_RXD_STAT_L4CS        MS(7, 0x1) /* L4 xsum calculated */
+#define TXGBE_RXD_STAT_IPCS        MS(8, 0x1) /* IP xsum calculated */
+#define TXGBE_RXD_STAT_PIF         MS(9, 0x1) /* Non-unicast address */
+#define TXGBE_RXD_STAT_EIPCS       MS(10, 0x1) /* Encap IP xsum calculated */
+#define TXGBE_RXD_STAT_VEXT        MS(11, 0x1) /* Multi-VLAN */
+#define TXGBE_RXD_STAT_IPV6EX      MS(12, 0x1) /* IPv6 with option header */
+#define TXGBE_RXD_STAT_LLINT       MS(13, 0x1) /* Pkt caused LLI */
+#define TXGBE_RXD_STAT_1588        MS(14, 0x1) /* IEEE1588 Time Stamp */
+#define TXGBE_RXD_STAT_SECP        MS(15, 0x1) /* Security Processing */
+#define TXGBE_RXD_STAT_LB          MS(16, 0x1) /* Loopback Status */
+/*** bit 17-30, when PTYPE=IP ***/
+#define TXGBE_RXD_STAT_BMC         MS(17, 0x1) /* PTYPE=IP, BMC status */
+#define TXGBE_RXD_ERR_FDIR_LEN     MS(20, 0x1) /* FDIR Length error */
+#define TXGBE_RXD_ERR_FDIR_DROP    MS(21, 0x1) /* FDIR Drop error */
+#define TXGBE_RXD_ERR_FDIR_COLL    MS(22, 0x1) /* FDIR Collision error */
+#define TXGBE_RXD_ERR_HBO          MS(23, 0x1) /* Header Buffer Overflow */
+#define TXGBE_RXD_ERR_EIPCS        MS(26, 0x1) /* Encap IP header error */
+#define TXGBE_RXD_ERR_SECERR       MS(27, 0x1) /* macsec or ipsec error */
+#define TXGBE_RXD_ERR_RXE          MS(29, 0x1) /* Any MAC Error */
+#define TXGBE_RXD_ERR_L4CS         MS(30, 0x1) /* TCP/UDP xsum error */
+#define TXGBE_RXD_ERR_IPCS         MS(31, 0x1) /* IP xsum error */
+#define TXGBE_RXD_ERR_CSUM(dw)     RS(dw, 30, 0x3)
+/*** bit 17-30, when PTYPE=FCOE ***/
+#define TXGBE_RXD_STAT_FCOEFS      MS(17, 0x1) /* PTYPE=FCOE, FCoE EOF/SOF */
+#define TXGBE_RXD_FCSTAT_MASK      MS(18, 0x3) /* FCoE Pkt Stat */
+#define TXGBE_RXD_FCSTAT_NOMTCH    LS(0, 18, 0x3) /* No Ctxt Match */
+#define TXGBE_RXD_FCSTAT_NODDP     LS(1, 18, 0x3) /* Ctxt w/o DDP */
+#define TXGBE_RXD_FCSTAT_FCPRSP    LS(2, 18, 0x3) /* Recv. FCP_RSP */
+#define TXGBE_RXD_FCSTAT_DDP       LS(3, 18, 0x3) /* Ctxt w/ DDP */
+#define TXGBE_RXD_FCERR_MASK       MS(20, 0x7) /* FCERR */
+#define TXGBE_RXD_FCERR_0          LS(0, 20, 0x7)
+#define TXGBE_RXD_FCERR_1          LS(1, 20, 0x7)
+#define TXGBE_RXD_FCERR_2          LS(2, 20, 0x7)
+#define TXGBE_RXD_FCERR_3          LS(3, 20, 0x7)
+#define TXGBE_RXD_FCERR_4          LS(4, 20, 0x7)
+#define TXGBE_RXD_FCERR_5          LS(5, 20, 0x7)
+#define TXGBE_RXD_FCERR_6          LS(6, 20, 0x7)
+#define TXGBE_RXD_FCERR_7          LS(7, 20, 0x7)
+
+/* @txgbe_rx_desc.dw3 */
+#define TXGBE_RXD_LENGTH(rxd)           ((rxd)->qw1.hi.len)
+#define TXGBE_RXD_VLAN(rxd)             ((rxd)->qw1.hi.tag)
+
 /*****************************************************************************
  * Transmit Descriptor
  *****************************************************************************/
@@ -145,6 +239,8 @@ struct txgbe_tx_desc {
 #define RX_RING_SZ ((TXGBE_RING_DESC_MAX + RTE_PMD_TXGBE_RX_MAX_BURST) * \
                    sizeof(struct txgbe_rx_desc))
 
+#define rte_packet_prefetch(p)  rte_prefetch1(p)
+
 #define RTE_TXGBE_REGISTER_POLL_WAIT_10_MS  10
 #define RTE_TXGBE_WAIT_100_US               100
 
@@ -206,6 +302,8 @@ struct txgbe_rx_queue {
        uint8_t             crc_len;  /**< 0 if CRC stripped, 4 otherwise. */
        uint8_t             drop_en;  /**< If not 0, set SRRCTL.Drop_En. */
        uint8_t             rx_deferred_start; /**< not in global dev start. */
+       /** flags to set in mbuf when a vlan is detected. */
+       uint64_t            vlan_flags;
        uint64_t            offloads; /**< Rx offloads with DEV_RX_OFFLOAD_* */
        /** need to alloc dummy mbuf, for wraparound when scanning hw ring */
        struct rte_mbuf fake_mbuf;