net: add rte prefix to ether defines
[dpdk.git] / drivers / net / i40e / i40e_rxtx.c
index a35a9d3..ee87651 100644 (file)
@@ -17,7 +17,7 @@
 #include <rte_mbuf.h>
 #include <rte_malloc.h>
 #include <rte_ether.h>
-#include <rte_ethdev.h>
+#include <rte_ethdev_driver.h>
 #include <rte_tcp.h>
 #include <rte_sctp.h>
 #include <rte_udp.h>
@@ -40,9 +40,6 @@
 /* Base address of the HW descriptor ring should be 128B aligned. */
 #define I40E_RING_BASE_ALIGN   128
 
-#define I40E_SIMPLE_FLAGS ((uint32_t)ETH_TXQ_FLAGS_NOMULTSEGS | \
-                                       ETH_TXQ_FLAGS_NOOFFLOADS)
-
 #define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS)
 
 #ifdef RTE_LIBRTE_IEEE1588
                PKT_TX_OUTER_IP_CKSUM)
 
 #define I40E_TX_OFFLOAD_MASK (  \
+               PKT_TX_OUTER_IPV4 |     \
+               PKT_TX_OUTER_IPV6 |     \
+               PKT_TX_IPV4 |           \
+               PKT_TX_IPV6 |           \
                PKT_TX_IP_CKSUM |       \
                PKT_TX_L4_MASK |        \
                PKT_TX_OUTER_IP_CKSUM | \
@@ -86,7 +87,8 @@ i40e_rxd_to_vlan_tci(struct rte_mbuf *mb, volatile union i40e_rx_desc *rxdp)
 #ifndef RTE_LIBRTE_I40E_16BYTE_RX_DESC
        if (rte_le_to_cpu_16(rxdp->wb.qword2.ext_status) &
                (1 << I40E_RX_DESC_EXT_STATUS_L2TAG2P_SHIFT)) {
-               mb->ol_flags |= PKT_RX_QINQ_STRIPPED;
+               mb->ol_flags |= PKT_RX_QINQ_STRIPPED | PKT_RX_QINQ |
+                       PKT_RX_VLAN_STRIPPED | PKT_RX_VLAN;
                mb->vlan_tci_outer = mb->vlan_tci;
                mb->vlan_tci = rte_le_to_cpu_16(rxdp->wb.qword2.l2tag2_2);
                PMD_RX_LOG(DEBUG, "Descriptor l2tag2_1: %u, l2tag2_2: %u",
@@ -887,17 +889,17 @@ i40e_recv_scattered_pkts(void *rx_queue,
                 */
                rxm->next = NULL;
                if (unlikely(rxq->crc_len > 0)) {
-                       first_seg->pkt_len -= ETHER_CRC_LEN;
-                       if (rx_packet_len <= ETHER_CRC_LEN) {
+                       first_seg->pkt_len -= RTE_ETHER_CRC_LEN;
+                       if (rx_packet_len <= RTE_ETHER_CRC_LEN) {
                                rte_pktmbuf_free_seg(rxm);
                                first_seg->nb_segs--;
                                last_seg->data_len =
                                        (uint16_t)(last_seg->data_len -
-                                       (ETHER_CRC_LEN - rx_packet_len));
+                                       (RTE_ETHER_CRC_LEN - rx_packet_len));
                                last_seg->next = NULL;
                        } else
                                rxm->data_len = (uint16_t)(rx_packet_len -
-                                                               ETHER_CRC_LEN);
+                                                       RTE_ETHER_CRC_LEN);
                }
 
                first_seg->port = rxq->port_id;
@@ -1240,7 +1242,7 @@ i40e_tx_free_bufs(struct i40e_tx_queue *txq)
        for (i = 0; i < txq->tx_rs_thresh; i++)
                rte_prefetch0((txep + i)->mbuf);
 
-       if (txq->txq_flags & (uint32_t)ETH_TXQ_FLAGS_NOREFCOUNT) {
+       if (txq->offloads & DEV_TX_OFFLOAD_MBUF_FAST_FREE) {
                for (i = 0; i < txq->tx_rs_thresh; ++i, ++txep) {
                        rte_mempool_put(txep->mbuf->pool, txep->mbuf);
                        txep->mbuf = NULL;
@@ -1442,13 +1444,15 @@ i40e_prep_pkts(__rte_unused void *tx_queue, struct rte_mbuf **tx_pkts,
 
                /* Check for m->nb_segs to not exceed the limits. */
                if (!(ol_flags & PKT_TX_TCP_SEG)) {
-                       if (m->nb_segs > I40E_TX_MAX_SEG ||
-                           m->nb_segs > I40E_TX_MAX_MTU_SEG) {
+                       if (m->nb_segs > I40E_TX_MAX_MTU_SEG ||
+                           m->pkt_len > I40E_FRAME_SIZE_MAX) {
                                rte_errno = -EINVAL;
                                return i;
                        }
-               } else if ((m->tso_segsz < I40E_MIN_TSO_MSS) ||
-                               (m->tso_segsz > I40E_MAX_TSO_MSS)) {
+               } else if (m->nb_segs > I40E_TX_MAX_SEG ||
+                          m->tso_segsz < I40E_MIN_TSO_MSS ||
+                          m->tso_segsz > I40E_MAX_TSO_MSS ||
+                          m->pkt_len > I40E_TSO_FRAME_SIZE_MAX) {
                        /* MSS outside the range (256B - 9674B) are considered
                         * malicious
                         */
@@ -1461,6 +1465,12 @@ i40e_prep_pkts(__rte_unused void *tx_queue, struct rte_mbuf **tx_pkts,
                        return i;
                }
 
+               /* check the size of packet */
+               if (m->pkt_len < I40E_TX_MIN_PKT_LEN) {
+                       rte_errno = -EINVAL;
+                       return i;
+               }
+
 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
                ret = rte_validate_tx_offload(m);
                if (ret != 0) {
@@ -1527,38 +1537,36 @@ int
 i40e_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 {
        struct i40e_rx_queue *rxq;
-       int err = -1;
+       int err;
        struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 
        PMD_INIT_FUNC_TRACE();
 
-       if (rx_queue_id < dev->data->nb_rx_queues) {
-               rxq = dev->data->rx_queues[rx_queue_id];
-
-               err = i40e_alloc_rx_queue_mbufs(rxq);
-               if (err) {
-                       PMD_DRV_LOG(ERR, "Failed to allocate RX queue mbuf");
-                       return err;
-               }
+       rxq = dev->data->rx_queues[rx_queue_id];
 
-               rte_wmb();
+       err = i40e_alloc_rx_queue_mbufs(rxq);
+       if (err) {
+               PMD_DRV_LOG(ERR, "Failed to allocate RX queue mbuf");
+               return err;
+       }
 
-               /* Init the RX tail regieter. */
-               I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
+       rte_wmb();
 
-               err = i40e_switch_rx_queue(hw, rxq->reg_idx, TRUE);
+       /* Init the RX tail regieter. */
+       I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
 
-               if (err) {
-                       PMD_DRV_LOG(ERR, "Failed to switch RX queue %u on",
-                                   rx_queue_id);
+       err = i40e_switch_rx_queue(hw, rxq->reg_idx, TRUE);
+       if (err) {
+               PMD_DRV_LOG(ERR, "Failed to switch RX queue %u on",
+                           rx_queue_id);
 
-                       i40e_rx_queue_release_mbufs(rxq);
-                       i40e_reset_rx_queue(rxq);
-               } else
-                       dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
+               i40e_rx_queue_release_mbufs(rxq);
+               i40e_reset_rx_queue(rxq);
+               return err;
        }
+       dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 
-       return err;
+       return 0;
 }
 
 int
@@ -1568,24 +1576,21 @@ i40e_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
        int err;
        struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 
-       if (rx_queue_id < dev->data->nb_rx_queues) {
-               rxq = dev->data->rx_queues[rx_queue_id];
-
-               /*
-               * rx_queue_id is queue id application refers to, while
-               * rxq->reg_idx is the real queue index.
-               */
-               err = i40e_switch_rx_queue(hw, rxq->reg_idx, FALSE);
+       rxq = dev->data->rx_queues[rx_queue_id];
 
-               if (err) {
-                       PMD_DRV_LOG(ERR, "Failed to switch RX queue %u off",
-                                   rx_queue_id);
-                       return err;
-               }
-               i40e_rx_queue_release_mbufs(rxq);
-               i40e_reset_rx_queue(rxq);
-               dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
+       /*
+        * rx_queue_id is queue id application refers to, while
+        * rxq->reg_idx is the real queue index.
+        */
+       err = i40e_switch_rx_queue(hw, rxq->reg_idx, FALSE);
+       if (err) {
+               PMD_DRV_LOG(ERR, "Failed to switch RX queue %u off",
+                           rx_queue_id);
+               return err;
        }
+       i40e_rx_queue_release_mbufs(rxq);
+       i40e_reset_rx_queue(rxq);
+       dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 
        return 0;
 }
@@ -1593,28 +1598,27 @@ i40e_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 int
 i40e_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 {
-       int err = -1;
+       int err;
        struct i40e_tx_queue *txq;
        struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 
        PMD_INIT_FUNC_TRACE();
 
-       if (tx_queue_id < dev->data->nb_tx_queues) {
-               txq = dev->data->tx_queues[tx_queue_id];
-
-               /*
-               * tx_queue_id is queue id application refers to, while
-               * rxq->reg_idx is the real queue index.
-               */
-               err = i40e_switch_tx_queue(hw, txq->reg_idx, TRUE);
-               if (err)
-                       PMD_DRV_LOG(ERR, "Failed to switch TX queue %u on",
-                                   tx_queue_id);
-               else
-                       dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
+       txq = dev->data->tx_queues[tx_queue_id];
+
+       /*
+        * tx_queue_id is queue id application refers to, while
+        * rxq->reg_idx is the real queue index.
+        */
+       err = i40e_switch_tx_queue(hw, txq->reg_idx, TRUE);
+       if (err) {
+               PMD_DRV_LOG(ERR, "Failed to switch TX queue %u on",
+                           tx_queue_id);
+               return err;
        }
+       dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 
-       return err;
+       return 0;
 }
 
 int
@@ -1624,26 +1628,23 @@ i40e_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
        int err;
        struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 
-       if (tx_queue_id < dev->data->nb_tx_queues) {
-               txq = dev->data->tx_queues[tx_queue_id];
+       txq = dev->data->tx_queues[tx_queue_id];
 
-               /*
-               * tx_queue_id is queue id application refers to, while
-               * txq->reg_idx is the real queue index.
-               */
-               err = i40e_switch_tx_queue(hw, txq->reg_idx, FALSE);
-
-               if (err) {
-                       PMD_DRV_LOG(ERR, "Failed to switch TX queue %u of",
-                                   tx_queue_id);
-                       return err;
-               }
-
-               i40e_tx_queue_release_mbufs(txq);
-               i40e_reset_tx_queue(txq);
-               dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
+       /*
+        * tx_queue_id is queue id application refers to, while
+        * txq->reg_idx is the real queue index.
+        */
+       err = i40e_switch_tx_queue(hw, txq->reg_idx, FALSE);
+       if (err) {
+               PMD_DRV_LOG(ERR, "Failed to switch TX queue %u of",
+                           tx_queue_id);
+               return err;
        }
 
+       i40e_tx_queue_release_mbufs(txq);
+       i40e_reset_tx_queue(txq);
+       dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
+
        return 0;
 }
 
@@ -1685,11 +1686,87 @@ i40e_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 #endif
            dev->rx_pkt_burst == i40e_recv_scattered_pkts ||
            dev->rx_pkt_burst == i40e_recv_scattered_pkts_vec ||
-           dev->rx_pkt_burst == i40e_recv_pkts_vec)
+           dev->rx_pkt_burst == i40e_recv_pkts_vec ||
+           dev->rx_pkt_burst == i40e_recv_scattered_pkts_vec_avx2 ||
+           dev->rx_pkt_burst == i40e_recv_pkts_vec_avx2)
                return ptypes;
        return NULL;
 }
 
+static int
+i40e_dev_first_queue(uint16_t idx, void **queues, int num)
+{
+       uint16_t i;
+
+       for (i = 0; i < num; i++) {
+               if (i != idx && queues[i])
+                       return 0;
+       }
+
+       return 1;
+}
+
+static int
+i40e_dev_rx_queue_setup_runtime(struct rte_eth_dev *dev,
+                               struct i40e_rx_queue *rxq)
+{
+       struct i40e_adapter *ad =
+               I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
+       int use_def_burst_func =
+               check_rx_burst_bulk_alloc_preconditions(rxq);
+       uint16_t buf_size =
+               (uint16_t)(rte_pktmbuf_data_room_size(rxq->mp) -
+                          RTE_PKTMBUF_HEADROOM);
+       int use_scattered_rx =
+               (rxq->max_pkt_len > buf_size);
+
+       if (i40e_rx_queue_init(rxq) != I40E_SUCCESS) {
+               PMD_DRV_LOG(ERR,
+                           "Failed to do RX queue initialization");
+               return -EINVAL;
+       }
+
+       if (i40e_dev_first_queue(rxq->queue_id,
+                                dev->data->rx_queues,
+                                dev->data->nb_rx_queues)) {
+               /**
+                * If it is the first queue to setup,
+                * set all flags to default and call
+                * i40e_set_rx_function.
+                */
+               ad->rx_bulk_alloc_allowed = true;
+               ad->rx_vec_allowed = true;
+               dev->data->scattered_rx = use_scattered_rx;
+               if (use_def_burst_func)
+                       ad->rx_bulk_alloc_allowed = false;
+               i40e_set_rx_function(dev);
+               return 0;
+       } else if (ad->rx_vec_allowed && !rte_is_power_of_2(rxq->nb_rx_desc)) {
+               PMD_DRV_LOG(ERR, "Vector mode is allowed, but descriptor"
+                           " number %d of queue %d isn't power of 2",
+                           rxq->nb_rx_desc, rxq->queue_id);
+               return -EINVAL;
+       }
+
+       /* check bulk alloc conflict */
+       if (ad->rx_bulk_alloc_allowed && use_def_burst_func) {
+               PMD_DRV_LOG(ERR, "Can't use default burst.");
+               return -EINVAL;
+       }
+       /* check scatterred conflict */
+       if (!dev->data->scattered_rx && use_scattered_rx) {
+               PMD_DRV_LOG(ERR, "Scattered rx is required.");
+               return -EINVAL;
+       }
+       /* check vector conflict */
+       if (ad->rx_vec_allowed && i40e_rxq_vec_setup(rxq)) {
+               PMD_DRV_LOG(ERR, "Failed vector rx setup.");
+               return -EINVAL;
+       }
+
+       return 0;
+}
+
 int
 i40e_dev_rx_queue_setup(struct rte_eth_dev *dev,
                        uint16_t queue_idx,
@@ -1710,6 +1787,9 @@ i40e_dev_rx_queue_setup(struct rte_eth_dev *dev,
        uint16_t len, i;
        uint16_t reg_idx, base, bsf, tc_mapping;
        int q_offset, use_def_burst_func = 1;
+       uint64_t offloads;
+
+       offloads = rx_conf->offloads | dev->data->dev_conf.rxmode.offloads;
 
        if (hw->mac.type == I40E_MAC_VF || hw->mac.type == I40E_MAC_X722_VF) {
                vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
@@ -1758,11 +1838,14 @@ i40e_dev_rx_queue_setup(struct rte_eth_dev *dev,
        rxq->queue_id = queue_idx;
        rxq->reg_idx = reg_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);
+       if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_KEEP_CRC)
+               rxq->crc_len = RTE_ETHER_CRC_LEN;
+       else
+               rxq->crc_len = 0;
        rxq->drop_en = rx_conf->rx_drop_en;
        rxq->vsi = vsi;
        rxq->rx_deferred_start = rx_conf->rx_deferred_start;
+       rxq->offloads = offloads;
 
        /* Allocate the maximun number of RX ring hardware descriptor. */
        len = I40E_MAX_RING_DESC;
@@ -1806,25 +1889,6 @@ i40e_dev_rx_queue_setup(struct rte_eth_dev *dev,
 
        i40e_reset_rx_queue(rxq);
        rxq->q_set = TRUE;
-       dev->data->rx_queues[queue_idx] = rxq;
-
-       use_def_burst_func = check_rx_burst_bulk_alloc_preconditions(rxq);
-
-       if (!use_def_burst_func) {
-#ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
-               PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
-                            "satisfied. Rx Burst Bulk Alloc function will be "
-                            "used on port=%d, queue=%d.",
-                            rxq->port_id, rxq->queue_id);
-#endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
-       } else {
-               PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
-                            "not satisfied, Scattered Rx is requested, "
-                            "or RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC is "
-                            "not enabled on port=%d, queue=%d.",
-                            rxq->port_id, rxq->queue_id);
-               ad->rx_bulk_alloc_allowed = false;
-       }
 
        for (i = 0; i < I40E_MAX_TRAFFIC_CLASS; i++) {
                if (!(vsi->enabled_tc & (1 << i)))
@@ -1839,6 +1903,34 @@ i40e_dev_rx_queue_setup(struct rte_eth_dev *dev,
                        rxq->dcb_tc = i;
        }
 
+       if (dev->data->dev_started) {
+               if (i40e_dev_rx_queue_setup_runtime(dev, rxq)) {
+                       i40e_dev_rx_queue_release(rxq);
+                       return -EINVAL;
+               }
+       } else {
+               use_def_burst_func =
+                       check_rx_burst_bulk_alloc_preconditions(rxq);
+               if (!use_def_burst_func) {
+#ifdef RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC
+                       PMD_INIT_LOG(DEBUG,
+                         "Rx Burst Bulk Alloc Preconditions are "
+                         "satisfied. Rx Burst Bulk Alloc function will be "
+                         "used on port=%d, queue=%d.",
+                         rxq->port_id, rxq->queue_id);
+#endif /* RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC */
+               } else {
+                       PMD_INIT_LOG(DEBUG,
+                         "Rx Burst Bulk Alloc Preconditions are "
+                         "not satisfied, Scattered Rx is requested, "
+                         "or RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC is "
+                         "not enabled on port=%d, queue=%d.",
+                         rxq->port_id, rxq->queue_id);
+                       ad->rx_bulk_alloc_allowed = false;
+               }
+       }
+
+       dev->data->rx_queues[queue_idx] = rxq;
        return 0;
 }
 
@@ -1970,6 +2062,52 @@ i40e_dev_tx_descriptor_status(void *tx_queue, uint16_t offset)
        return RTE_ETH_TX_DESC_FULL;
 }
 
+static int
+i40e_dev_tx_queue_setup_runtime(struct rte_eth_dev *dev,
+                               struct i40e_tx_queue *txq)
+{
+       struct i40e_adapter *ad =
+               I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
+
+       if (i40e_tx_queue_init(txq) != I40E_SUCCESS) {
+               PMD_DRV_LOG(ERR,
+                           "Failed to do TX queue initialization");
+               return -EINVAL;
+       }
+
+       if (i40e_dev_first_queue(txq->queue_id,
+                                dev->data->tx_queues,
+                                dev->data->nb_tx_queues)) {
+               /**
+                * If it is the first queue to setup,
+                * set all flags and call
+                * i40e_set_tx_function.
+                */
+               i40e_set_tx_function_flag(dev, txq);
+               i40e_set_tx_function(dev);
+               return 0;
+       }
+
+       /* check vector conflict */
+       if (ad->tx_vec_allowed) {
+               if (txq->tx_rs_thresh > RTE_I40E_TX_MAX_FREE_BUF_SZ ||
+                   i40e_txq_vec_setup(txq)) {
+                       PMD_DRV_LOG(ERR, "Failed vector tx setup.");
+                       return -EINVAL;
+               }
+       }
+       /* check simple tx conflict */
+       if (ad->tx_simple_allowed) {
+               if ((txq->offloads & ~DEV_TX_OFFLOAD_MBUF_FAST_FREE) != 0 ||
+                               txq->tx_rs_thresh < RTE_PMD_I40E_TX_MAX_BURST) {
+                       PMD_DRV_LOG(ERR, "No-simple tx is required.");
+                       return -EINVAL;
+               }
+       }
+
+       return 0;
+}
+
 int
 i40e_dev_tx_queue_setup(struct rte_eth_dev *dev,
                        uint16_t queue_idx,
@@ -1987,6 +2125,9 @@ i40e_dev_tx_queue_setup(struct rte_eth_dev *dev,
        uint16_t tx_rs_thresh, tx_free_thresh;
        uint16_t reg_idx, i, base, bsf, tc_mapping;
        int q_offset;
+       uint64_t offloads;
+
+       offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
 
        if (hw->mac.type == I40E_MAC_VF || hw->mac.type == I40E_MAC_X722_VF) {
                vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
@@ -2028,15 +2169,30 @@ i40e_dev_tx_queue_setup(struct rte_eth_dev *dev,
         *  - tx_rs_thresh must be a divisor of the ring size.
         *  - tx_free_thresh must be greater than 0.
         *  - tx_free_thresh must be less than the size of the ring minus 3.
+        *  - tx_free_thresh + tx_rs_thresh must not exceed nb_desc.
         *
         * One descriptor in the TX ring is used as a sentinel to avoid a H/W
         * race condition, hence the maximum threshold constraints. When set
         * to zero use default values.
         */
-       tx_rs_thresh = (uint16_t)((tx_conf->tx_rs_thresh) ?
-               tx_conf->tx_rs_thresh : DEFAULT_TX_RS_THRESH);
        tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
                tx_conf->tx_free_thresh : DEFAULT_TX_FREE_THRESH);
+       /* force tx_rs_thresh to adapt an aggresive tx_free_thresh */
+       tx_rs_thresh = (DEFAULT_TX_RS_THRESH + tx_free_thresh > nb_desc) ?
+               nb_desc - tx_free_thresh : DEFAULT_TX_RS_THRESH;
+       if (tx_conf->tx_rs_thresh > 0)
+               tx_rs_thresh = tx_conf->tx_rs_thresh;
+       if (tx_rs_thresh + tx_free_thresh > nb_desc) {
+               PMD_INIT_LOG(ERR, "tx_rs_thresh + tx_free_thresh must not "
+                               "exceed nb_desc. (tx_rs_thresh=%u "
+                               "tx_free_thresh=%u nb_desc=%u port=%d queue=%d)",
+                               (unsigned int)tx_rs_thresh,
+                               (unsigned int)tx_free_thresh,
+                               (unsigned int)nb_desc,
+                               (int)dev->data->port_id,
+                               (int)queue_idx);
+               return I40E_ERR_PARAM;
+       }
        if (tx_rs_thresh >= (nb_desc - 2)) {
                PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the "
                             "number of TX descriptors minus 2. "
@@ -2121,7 +2277,7 @@ i40e_dev_tx_queue_setup(struct rte_eth_dev *dev,
        txq->queue_id = queue_idx;
        txq->reg_idx = reg_idx;
        txq->port_id = dev->data->port_id;
-       txq->txq_flags = tx_conf->txq_flags;
+       txq->offloads = offloads;
        txq->vsi = vsi;
        txq->tx_deferred_start = tx_conf->tx_deferred_start;
 
@@ -2142,10 +2298,6 @@ i40e_dev_tx_queue_setup(struct rte_eth_dev *dev,
 
        i40e_reset_tx_queue(txq);
        txq->q_set = TRUE;
-       dev->data->tx_queues[queue_idx] = txq;
-
-       /* Use a simple TX queue without offloads or multi segs if possible */
-       i40e_set_tx_function_flag(dev, txq);
 
        for (i = 0; i < I40E_MAX_TRAFFIC_CLASS; i++) {
                if (!(vsi->enabled_tc & (1 << i)))
@@ -2160,6 +2312,20 @@ i40e_dev_tx_queue_setup(struct rte_eth_dev *dev,
                        txq->dcb_tc = i;
        }
 
+       if (dev->data->dev_started) {
+               if (i40e_dev_tx_queue_setup_runtime(dev, txq)) {
+                       i40e_dev_tx_queue_release(txq);
+                       return -EINVAL;
+               }
+       } else {
+               /**
+                * Use a simple TX queue without offloads or
+                * multi segs if possible
+                */
+               i40e_set_tx_function_flag(dev, txq);
+       }
+       dev->data->tx_queues[queue_idx] = txq;
+
        return 0;
 }
 
@@ -2187,8 +2353,8 @@ i40e_memzone_reserve(const char *name, uint32_t len, int socket_id)
        if (mz)
                return mz;
 
-       mz = rte_memzone_reserve_aligned(name, len,
-                                        socket_id, 0, I40E_RING_BASE_ALIGN);
+       mz = rte_memzone_reserve_aligned(name, len, socket_id,
+                       RTE_MEMZONE_IOVA_CONTIG, I40E_RING_BASE_ALIGN);
        return mz;
 }
 
@@ -2272,18 +2438,19 @@ i40e_tx_queue_release_mbufs(struct i40e_tx_queue *txq)
        struct rte_eth_dev *dev;
        uint16_t i;
 
-       dev = &rte_eth_devices[txq->port_id];
-
        if (!txq || !txq->sw_ring) {
-               PMD_DRV_LOG(DEBUG, "Pointer to rxq or sw_ring is NULL");
+               PMD_DRV_LOG(DEBUG, "Pointer to txq or sw_ring is NULL");
                return;
        }
 
+       dev = &rte_eth_devices[txq->port_id];
+
        /**
         *  vPMD tx will not set sw_ring's mbuf to NULL after free,
         *  so need to free remains more carefully.
         */
-       if (dev->tx_pkt_burst == i40e_xmit_pkts_vec) {
+       if (dev->tx_pkt_burst == i40e_xmit_pkts_vec_avx2 ||
+                       dev->tx_pkt_burst == i40e_xmit_pkts_vec) {
                i = txq->tx_next_dd - txq->tx_rs_thresh + 1;
                if (txq->tx_tail < i) {
                        for (; i < txq->nb_tx_desc; i++) {
@@ -2466,24 +2633,24 @@ i40e_rx_queue_config(struct i40e_rx_queue *rxq)
 
        len = hw->func_caps.rx_buf_chain_len * rxq->rx_buf_len;
        rxq->max_pkt_len = RTE_MIN(len, data->dev_conf.rxmode.max_rx_pkt_len);
-       if (data->dev_conf.rxmode.jumbo_frame == 1) {
-               if (rxq->max_pkt_len <= ETHER_MAX_LEN ||
+       if (data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME) {
+               if (rxq->max_pkt_len <= RTE_ETHER_MAX_LEN ||
                        rxq->max_pkt_len > I40E_FRAME_SIZE_MAX) {
                        PMD_DRV_LOG(ERR, "maximum packet length must "
                                    "be larger than %u and smaller than %u,"
                                    "as jumbo frame is enabled",
-                                   (uint32_t)ETHER_MAX_LEN,
+                                   (uint32_t)RTE_ETHER_MAX_LEN,
                                    (uint32_t)I40E_FRAME_SIZE_MAX);
                        return I40E_ERR_CONFIG;
                }
        } else {
-               if (rxq->max_pkt_len < ETHER_MIN_LEN ||
-                       rxq->max_pkt_len > ETHER_MAX_LEN) {
+               if (rxq->max_pkt_len < RTE_ETHER_MIN_LEN ||
+                       rxq->max_pkt_len > RTE_ETHER_MAX_LEN) {
                        PMD_DRV_LOG(ERR, "maximum packet length must be "
                                    "larger than %u and smaller than %u, "
                                    "as jumbo frame is disabled",
-                                   (uint32_t)ETHER_MIN_LEN,
-                                   (uint32_t)ETHER_MAX_LEN);
+                                   (uint32_t)RTE_ETHER_MIN_LEN,
+                                   (uint32_t)RTE_ETHER_MAX_LEN);
                        return I40E_ERR_CONFIG;
                }
        }
@@ -2556,9 +2723,8 @@ i40e_rx_queue_init(struct i40e_rx_queue *rxq)
                RTE_PKTMBUF_HEADROOM);
 
        /* Check if scattered RX needs to be used. */
-       if ((rxq->max_pkt_len + 2 * I40E_VLAN_TAG_SIZE) > buf_size) {
+       if (rxq->max_pkt_len > buf_size)
                dev_data->scattered_rx = 1;
-       }
 
        /* Init the RX tail regieter. */
        I40E_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
@@ -2601,7 +2767,6 @@ i40e_dev_free_queues(struct rte_eth_dev *dev)
                i40e_dev_rx_queue_release(dev->data->rx_queues[i]);
                dev->data->rx_queues[i] = NULL;
        }
-       dev->data->nb_rx_queues = 0;
 
        for (i = 0; i < dev->data->nb_tx_queues; i++) {
                if (!dev->data->tx_queues[i])
@@ -2609,7 +2774,6 @@ i40e_dev_free_queues(struct rte_eth_dev *dev)
                i40e_dev_tx_queue_release(dev->data->tx_queues[i]);
                dev->data->tx_queues[i] = NULL;
        }
-       dev->data->nb_tx_queues = 0;
 }
 
 #define I40E_FDIR_NUM_TX_DESC  I40E_MIN_RING_DESC
@@ -2716,6 +2880,7 @@ i40e_fdir_setup_rx_resources(struct i40e_pf *pf)
        rxq->vsi = pf->fdir.fdir_vsi;
 
        rxq->rx_ring_phys_addr = rz->iova;
+       memset(rz->addr, 0, I40E_FDIR_NUM_RX_DESC * sizeof(union i40e_rx_desc));
        rxq->rx_ring = (union i40e_rx_desc *)rz->addr;
 
        /*
@@ -2743,6 +2908,7 @@ i40e_rxq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
        qinfo->conf.rx_free_thresh = rxq->rx_free_thresh;
        qinfo->conf.rx_drop_en = rxq->drop_en;
        qinfo->conf.rx_deferred_start = rxq->rx_deferred_start;
+       qinfo->conf.offloads = rxq->offloads;
 }
 
 void
@@ -2761,8 +2927,37 @@ i40e_txq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
 
        qinfo->conf.tx_free_thresh = txq->tx_free_thresh;
        qinfo->conf.tx_rs_thresh = txq->tx_rs_thresh;
-       qinfo->conf.txq_flags = txq->txq_flags;
        qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
+       qinfo->conf.offloads = txq->offloads;
+}
+
+static eth_rx_burst_t
+i40e_get_latest_rx_vec(bool scatter)
+{
+#ifdef RTE_ARCH_X86
+       if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2))
+               return scatter ? i40e_recv_scattered_pkts_vec_avx2 :
+                                i40e_recv_pkts_vec_avx2;
+#endif
+       return scatter ? i40e_recv_scattered_pkts_vec :
+                        i40e_recv_pkts_vec;
+}
+
+static eth_rx_burst_t
+i40e_get_recommend_rx_vec(bool scatter)
+{
+#ifdef RTE_ARCH_X86
+       /*
+        * since AVX frequency can be different to base frequency, limit
+        * use of AVX2 version to later plaforms, not all those that could
+        * theoretically run it.
+        */
+       if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F))
+               return scatter ? i40e_recv_scattered_pkts_vec_avx2 :
+                                i40e_recv_pkts_vec_avx2;
+#endif
+       return scatter ? i40e_recv_scattered_pkts_vec :
+                        i40e_recv_pkts_vec;
 }
 
 void __attribute__((cold))
@@ -2796,36 +2991,17 @@ i40e_set_rx_function(struct rte_eth_dev *dev)
                }
        }
 
-       if (dev->data->scattered_rx) {
-               /* Set the non-LRO scattered callback: there are Vector and
-                * single allocation versions.
-                */
-               if (ad->rx_vec_allowed) {
-                       PMD_INIT_LOG(DEBUG, "Using Vector Scattered Rx "
-                                           "callback (port=%d).",
-                                    dev->data->port_id);
-
-                       dev->rx_pkt_burst = i40e_recv_scattered_pkts_vec;
-               } else {
-                       PMD_INIT_LOG(DEBUG, "Using a Scattered with bulk "
-                                          "allocation callback (port=%d).",
-                                    dev->data->port_id);
-                       dev->rx_pkt_burst = i40e_recv_scattered_pkts;
-               }
-       /* If parameters allow we are going to choose between the following
-        * callbacks:
-        *    - Vector
-        *    - Bulk Allocation
-        *    - Single buffer allocation (the simplest one)
-        */
-       } else if (ad->rx_vec_allowed) {
-               PMD_INIT_LOG(DEBUG, "Vector rx enabled, please make sure RX "
-                                   "burst size no less than %d (port=%d).",
-                            RTE_I40E_DESCS_PER_LOOP,
-                            dev->data->port_id);
-
-               dev->rx_pkt_burst = i40e_recv_pkts_vec;
-       } else if (ad->rx_bulk_alloc_allowed) {
+       if (ad->rx_vec_allowed) {
+               /* Vec Rx path */
+               PMD_INIT_LOG(DEBUG, "Vector Rx path will be used on port=%d.",
+                               dev->data->port_id);
+               if (ad->use_latest_vec)
+                       dev->rx_pkt_burst =
+                       i40e_get_latest_rx_vec(dev->data->scattered_rx);
+               else
+                       dev->rx_pkt_burst =
+                       i40e_get_recommend_rx_vec(dev->data->scattered_rx);
+       } else if (!dev->data->scattered_rx && ad->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.",
@@ -2833,19 +3009,21 @@ i40e_set_rx_function(struct rte_eth_dev *dev)
 
                dev->rx_pkt_burst = i40e_recv_pkts_bulk_alloc;
        } else {
-               PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are not "
-                                   "satisfied, or Scattered Rx is requested "
-                                   "(port=%d).",
+               /* Simple Rx Path. */
+               PMD_INIT_LOG(DEBUG, "Simple Rx path will be used on port=%d.",
                             dev->data->port_id);
-
-               dev->rx_pkt_burst = i40e_recv_pkts;
+               dev->rx_pkt_burst = dev->data->scattered_rx ?
+                                       i40e_recv_scattered_pkts :
+                                       i40e_recv_pkts;
        }
 
        /* Propagate information about RX function choice through all queues. */
        if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
                rx_using_sse =
                        (dev->rx_pkt_burst == i40e_recv_scattered_pkts_vec ||
-                        dev->rx_pkt_burst == i40e_recv_pkts_vec);
+                        dev->rx_pkt_burst == i40e_recv_pkts_vec ||
+                        dev->rx_pkt_burst == i40e_recv_scattered_pkts_vec_avx2 ||
+                        dev->rx_pkt_burst == i40e_recv_pkts_vec_avx2);
 
                for (i = 0; i < dev->data->nb_rx_queues; i++) {
                        struct i40e_rx_queue *rxq = dev->data->rx_queues[i];
@@ -2862,19 +3040,49 @@ i40e_set_tx_function_flag(struct rte_eth_dev *dev, struct i40e_tx_queue *txq)
        struct i40e_adapter *ad =
                I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
 
-       /* Use a simple Tx queue (no offloads, no multi segs) if possible */
-       if (((txq->txq_flags & I40E_SIMPLE_FLAGS) == I40E_SIMPLE_FLAGS)
-                       && (txq->tx_rs_thresh >= RTE_PMD_I40E_TX_MAX_BURST)) {
-               if (txq->tx_rs_thresh <= RTE_I40E_TX_MAX_FREE_BUF_SZ) {
-                       PMD_INIT_LOG(DEBUG, "Vector tx"
-                                    " can be enabled on this txq.");
+       /* Use a simple Tx queue if possible (only fast free is allowed) */
+       ad->tx_simple_allowed =
+               (txq->offloads ==
+                (txq->offloads & DEV_TX_OFFLOAD_MBUF_FAST_FREE) &&
+                txq->tx_rs_thresh >= RTE_PMD_I40E_TX_MAX_BURST);
+       ad->tx_vec_allowed = (ad->tx_simple_allowed &&
+                       txq->tx_rs_thresh <= RTE_I40E_TX_MAX_FREE_BUF_SZ);
+
+       if (ad->tx_vec_allowed)
+               PMD_INIT_LOG(DEBUG, "Vector Tx can be enabled on Tx queue %u.",
+                               txq->queue_id);
+       else if (ad->tx_simple_allowed)
+               PMD_INIT_LOG(DEBUG, "Simple Tx can be enabled on Tx queue %u.",
+                               txq->queue_id);
+       else
+               PMD_INIT_LOG(DEBUG,
+                               "Neither simple nor vector Tx enabled on Tx queue %u\n",
+                               txq->queue_id);
+}
 
-               } else {
-                       ad->tx_vec_allowed = false;
-               }
-       } else {
-               ad->tx_simple_allowed = false;
-       }
+static eth_tx_burst_t
+i40e_get_latest_tx_vec(void)
+{
+#ifdef RTE_ARCH_X86
+       if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2))
+               return i40e_xmit_pkts_vec_avx2;
+#endif
+       return i40e_xmit_pkts_vec;
+}
+
+static eth_tx_burst_t
+i40e_get_recommend_tx_vec(void)
+{
+#ifdef RTE_ARCH_X86
+       /*
+        * since AVX frequency can be different to base frequency, limit
+        * use of AVX2 version to later plaforms, not all those that could
+        * theoretically run it.
+        */
+       if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F))
+               return i40e_xmit_pkts_vec_avx2;
+#endif
+       return i40e_xmit_pkts_vec;
 }
 
 void __attribute__((cold))
@@ -2901,7 +3109,12 @@ i40e_set_tx_function(struct rte_eth_dev *dev)
        if (ad->tx_simple_allowed) {
                if (ad->tx_vec_allowed) {
                        PMD_INIT_LOG(DEBUG, "Vector tx finally be used.");
-                       dev->tx_pkt_burst = i40e_xmit_pkts_vec;
+                       if (ad->use_latest_vec)
+                               dev->tx_pkt_burst =
+                                       i40e_get_latest_tx_vec();
+                       else
+                               dev->tx_pkt_burst =
+                                       i40e_get_recommend_tx_vec();
                } else {
                        PMD_INIT_LOG(DEBUG, "Simple tx finally be used.");
                        dev->tx_pkt_burst = i40e_xmit_pkts_simple;
@@ -2983,14 +3196,14 @@ i40e_set_default_pctype_table(struct rte_eth_dev *dev)
        }
 }
 
-/* Stubs needed for linkage when CONFIG_RTE_I40E_INC_VECTOR is set to 'n' */
-int __attribute__((weak))
+/* Stubs needed for linkage when CONFIG_RTE_LIBRTE_I40E_INC_VECTOR is set to 'n' */
+__rte_weak int
 i40e_rx_vec_dev_conf_condition_check(struct rte_eth_dev __rte_unused *dev)
 {
        return -1;
 }
 
-uint16_t __attribute__((weak))
+__rte_weak uint16_t
 i40e_recv_pkts_vec(
        void __rte_unused *rx_queue,
        struct rte_mbuf __rte_unused **rx_pkts,
@@ -2999,7 +3212,7 @@ i40e_recv_pkts_vec(
        return 0;
 }
 
-uint16_t __attribute__((weak))
+__rte_weak uint16_t
 i40e_recv_scattered_pkts_vec(
        void __rte_unused *rx_queue,
        struct rte_mbuf __rte_unused **rx_pkts,
@@ -3008,28 +3221,52 @@ i40e_recv_scattered_pkts_vec(
        return 0;
 }
 
-int __attribute__((weak))
+__rte_weak uint16_t
+i40e_recv_pkts_vec_avx2(void __rte_unused *rx_queue,
+                       struct rte_mbuf __rte_unused **rx_pkts,
+                       uint16_t __rte_unused nb_pkts)
+{
+       return 0;
+}
+
+__rte_weak uint16_t
+i40e_recv_scattered_pkts_vec_avx2(void __rte_unused *rx_queue,
+                       struct rte_mbuf __rte_unused **rx_pkts,
+                       uint16_t __rte_unused nb_pkts)
+{
+       return 0;
+}
+
+__rte_weak int
 i40e_rxq_vec_setup(struct i40e_rx_queue __rte_unused *rxq)
 {
        return -1;
 }
 
-int __attribute__((weak))
+__rte_weak int
 i40e_txq_vec_setup(struct i40e_tx_queue __rte_unused *txq)
 {
        return -1;
 }
 
-void __attribute__((weak))
+__rte_weak void
 i40e_rx_queue_release_mbufs_vec(struct i40e_rx_queue __rte_unused*rxq)
 {
        return;
 }
 
-uint16_t __attribute__((weak))
+__rte_weak uint16_t
 i40e_xmit_fixed_burst_vec(void __rte_unused * tx_queue,
                          struct rte_mbuf __rte_unused **tx_pkts,
                          uint16_t __rte_unused nb_pkts)
 {
        return 0;
 }
+
+__rte_weak uint16_t
+i40e_xmit_pkts_vec_avx2(void __rte_unused * tx_queue,
+                         struct rte_mbuf __rte_unused **tx_pkts,
+                         uint16_t __rte_unused nb_pkts)
+{
+       return 0;
+}