net/nfp: fix VLAN offload flags check
[dpdk.git] / drivers / net / nfp / nfp_net.c
index a9be403..ab2d331 100644 (file)
  * Netronome vNIC DPDK Poll-Mode Driver: Main entry point
  */
 
-#include <stdio.h>
-#include <stdlib.h>
-#include <stdint.h>
-#include <unistd.h>
-#include <string.h>
-#include <sys/mman.h>
-#include <sys/socket.h>
-#include <sys/io.h>
-#include <assert.h>
-#include <time.h>
 #include <math.h>
-#include <inttypes.h>
 
 #include <rte_byteorder.h>
 #include <rte_common.h>
@@ -65,6 +54,7 @@
 #include <rte_version.h>
 #include <rte_string_fns.h>
 #include <rte_alarm.h>
+#include <rte_spinlock.h>
 
 #include "nfp_net_pmd.h"
 #include "nfp_net_logs.h"
 /* Prototypes */
 static void nfp_net_close(struct rte_eth_dev *dev);
 static int nfp_net_configure(struct rte_eth_dev *dev);
+static void nfp_net_dev_interrupt_handler(struct rte_intr_handle *handle,
+                                         void *param);
+static void nfp_net_dev_interrupt_delayed_handler(void *param);
+static int nfp_net_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
+static void nfp_net_infos_get(struct rte_eth_dev *dev,
+                             struct rte_eth_dev_info *dev_info);
 static int nfp_net_init(struct rte_eth_dev *eth_dev);
+static int nfp_net_link_update(struct rte_eth_dev *dev, int wait_to_complete);
+static void nfp_net_promisc_enable(struct rte_eth_dev *dev);
+static void nfp_net_promisc_disable(struct rte_eth_dev *dev);
 static int nfp_net_rx_fill_freelist(struct nfp_net_rxq *rxq);
 static uint32_t nfp_net_rx_queue_count(struct rte_eth_dev *dev,
                                       uint16_t queue_idx);
@@ -90,6 +89,9 @@ static int nfp_net_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
                                  uint16_t nb_desc, unsigned int socket_id,
                                  const struct rte_eth_txconf *tx_conf);
 static int nfp_net_start(struct rte_eth_dev *dev);
+static void nfp_net_stats_get(struct rte_eth_dev *dev,
+                             struct rte_eth_stats *stats);
+static void nfp_net_stats_reset(struct rte_eth_dev *dev);
 static void nfp_net_stop(struct rte_eth_dev *dev);
 static uint16_t nfp_net_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                                  uint16_t nb_pkts);
@@ -212,7 +214,7 @@ ring_dma_zone_reserve(struct rte_eth_dev *dev, const char *ring_name,
        const struct rte_memzone *mz;
 
        snprintf(z_name, sizeof(z_name), "%s_%s_%d_%d",
-                dev->driver->pci_drv.name,
+                dev->driver->pci_drv.driver.name,
                 ring_name, dev->data->port_id, queue_id);
 
        mz = rte_memzone_lookup(z_name);
@@ -223,6 +225,57 @@ ring_dma_zone_reserve(struct rte_eth_dev *dev, const char *ring_name,
                                           NFP_MEMZONE_ALIGN);
 }
 
+/*
+ * Atomically reads link status information from global structure rte_eth_dev.
+ *
+ * @param dev
+ *   - Pointer to the structure rte_eth_dev to read from.
+ *   - Pointer to the buffer to be saved with the link status.
+ *
+ * @return
+ *   - On success, zero.
+ *   - On failure, negative value.
+ */
+static inline int
+nfp_net_dev_atomic_read_link_status(struct rte_eth_dev *dev,
+                                   struct rte_eth_link *link)
+{
+       struct rte_eth_link *dst = link;
+       struct rte_eth_link *src = &dev->data->dev_link;
+
+       if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
+                               *(uint64_t *)src) == 0)
+               return -1;
+
+       return 0;
+}
+
+/*
+ * Atomically writes the link status information into global
+ * structure rte_eth_dev.
+ *
+ * @param dev
+ *   - Pointer to the structure rte_eth_dev to read from.
+ *   - Pointer to the buffer to be saved with the link status.
+ *
+ * @return
+ *   - On success, zero.
+ *   - On failure, negative value.
+ */
+static inline int
+nfp_net_dev_atomic_write_link_status(struct rte_eth_dev *dev,
+                                    struct rte_eth_link *link)
+{
+       struct rte_eth_link *dst = &dev->data->dev_link;
+       struct rte_eth_link *src = link;
+
+       if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
+                               *(uint64_t *)src) == 0)
+               return -1;
+
+       return 0;
+}
+
 static void
 nfp_net_rx_queue_release_mbufs(struct nfp_net_rxq *rxq)
 {
@@ -255,7 +308,6 @@ static void
 nfp_net_reset_rx_queue(struct nfp_net_rxq *rxq)
 {
        nfp_net_rx_queue_release_mbufs(rxq);
-       rxq->wr_p = 0;
        rxq->rd_p = 0;
        rxq->nb_rx_hold = 0;
 }
@@ -270,7 +322,7 @@ nfp_net_tx_queue_release_mbufs(struct nfp_net_txq *txq)
 
        for (i = 0; i < txq->tx_count; i++) {
                if (txq->txbufs[i].mbuf) {
-                       rte_pktmbuf_free_seg(txq->txbufs[i].mbuf);
+                       rte_pktmbuf_free(txq->txbufs[i].mbuf);
                        txq->txbufs[i].mbuf = NULL;
                }
        }
@@ -294,7 +346,6 @@ nfp_net_reset_tx_queue(struct nfp_net_txq *txq)
        nfp_net_tx_queue_release_mbufs(txq);
        txq->wr_p = 0;
        txq->rd_p = 0;
-       txq->tail = 0;
 }
 
 static int
@@ -354,6 +405,8 @@ nfp_net_reconfig(struct nfp_net_hw *hw, uint32_t ctrl, uint32_t update)
        PMD_DRV_LOG(DEBUG, "nfp_net_reconfig: ctrl=%08x update=%08x\n",
                    ctrl, update);
 
+       rte_spinlock_lock(&hw->reconfig_lock);
+
        nn_cfg_writel(hw, NFP_NET_CFG_CTRL, ctrl);
        nn_cfg_writel(hw, NFP_NET_CFG_UPDATE, update);
 
@@ -361,6 +414,8 @@ nfp_net_reconfig(struct nfp_net_hw *hw, uint32_t ctrl, uint32_t update)
 
        err = __nfp_net_reconfig(hw, update);
 
+       rte_spinlock_unlock(&hw->reconfig_lock);
+
        if (!err)
                return 0;
 
@@ -549,18 +604,8 @@ nfp_net_rx_freelist_setup(struct rte_eth_dev *dev)
 static void
 nfp_net_params_setup(struct nfp_net_hw *hw)
 {
-       uint32_t *mac_address;
-
        nn_cfg_writel(hw, NFP_NET_CFG_MTU, hw->mtu);
        nn_cfg_writel(hw, NFP_NET_CFG_FLBUFSZ, hw->flbufsz);
-
-       /* A MAC address is 8 bytes long */
-       mac_address = (uint32_t *)(hw->mac_addr);
-
-       nn_cfg_writel(hw, NFP_NET_CFG_MACADDR,
-                     rte_cpu_to_be_32(*mac_address));
-       nn_cfg_writel(hw, NFP_NET_CFG_MACADDR + 4,
-                     rte_cpu_to_be_32(*(mac_address + 4)));
 }
 
 static void
@@ -569,11 +614,64 @@ nfp_net_cfg_queue_setup(struct nfp_net_hw *hw)
        hw->qcp_cfg = hw->tx_bar + NFP_QCP_QUEUE_ADDR_SZ;
 }
 
+static void nfp_net_read_mac(struct nfp_net_hw *hw)
+{
+       uint32_t tmp;
+
+       tmp = rte_be_to_cpu_32(nn_cfg_readl(hw, NFP_NET_CFG_MACADDR));
+       memcpy(&hw->mac_addr[0], &tmp, sizeof(struct ether_addr));
+
+       tmp = rte_be_to_cpu_32(nn_cfg_readl(hw, NFP_NET_CFG_MACADDR + 4));
+       memcpy(&hw->mac_addr[4], &tmp, 2);
+}
+
+static int
+nfp_configure_rx_interrupt(struct rte_eth_dev *dev,
+                          struct rte_intr_handle *intr_handle)
+{
+       struct nfp_net_hw *hw;
+       int i;
+
+       if (!intr_handle->intr_vec) {
+               intr_handle->intr_vec =
+                       rte_zmalloc("intr_vec",
+                                   dev->data->nb_rx_queues * sizeof(int), 0);
+               if (!intr_handle->intr_vec) {
+                       PMD_INIT_LOG(ERR, "Failed to allocate %d rx_queues"
+                                    " intr_vec\n", dev->data->nb_rx_queues);
+                       return -ENOMEM;
+               }
+       }
+
+       hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       if (intr_handle->type == RTE_INTR_HANDLE_UIO) {
+               PMD_INIT_LOG(INFO, "VF: enabling RX interrupt with UIO\n");
+               /* UIO just supports one queue and no LSC*/
+               nn_cfg_writeb(hw, NFP_NET_CFG_RXR_VEC(0), 0);
+       } else {
+               PMD_INIT_LOG(INFO, "VF: enabling RX interrupt with VFIO\n");
+               for (i = 0; i < dev->data->nb_rx_queues; i++)
+                       /*
+                        * The first msix vector is reserved for non
+                        * efd interrupts
+                       */
+                       nn_cfg_writeb(hw, NFP_NET_CFG_RXR_VEC(i), i + 1);
+       }
+
+       /* Avoiding TX interrupts */
+       hw->ctrl |= NFP_NET_CFG_CTRL_MSIX_TX_OFF;
+       return 0;
+}
+
 static int
 nfp_net_start(struct rte_eth_dev *dev)
 {
+       struct rte_pci_device *pci_dev = RTE_DEV_TO_PCI(dev->device);
+       struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
        uint32_t new_ctrl, update = 0;
        struct nfp_net_hw *hw;
+       uint32_t intr_vector;
        int ret;
 
        hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
@@ -589,10 +687,39 @@ nfp_net_start(struct rte_eth_dev *dev)
        /* Enabling the required queues in the device */
        nfp_net_enable_queues(dev);
 
+       /* check and configure queue intr-vector mapping */
+       if (dev->data->dev_conf.intr_conf.rxq != 0) {
+               if (intr_handle->type == RTE_INTR_HANDLE_UIO) {
+                       /*
+                        * Better not to share LSC with RX interrupts.
+                        * Unregistering LSC interrupt handler
+                        */
+                       rte_intr_callback_unregister(&pci_dev->intr_handle,
+                               nfp_net_dev_interrupt_handler, (void *)dev);
+
+                       if (dev->data->nb_rx_queues > 1) {
+                               PMD_INIT_LOG(ERR, "PMD rx interrupt only "
+                                            "supports 1 queue with UIO");
+                               return -EIO;
+                       }
+               }
+               intr_vector = dev->data->nb_rx_queues;
+               if (rte_intr_efd_enable(intr_handle, intr_vector))
+                       return -1;
+       }
+
+       nfp_configure_rx_interrupt(dev, intr_handle);
+
+       rte_intr_enable(intr_handle);
+
        /* Enable device */
-       new_ctrl = hw->ctrl | NFP_NET_CFG_CTRL_ENABLE | NFP_NET_CFG_UPDATE_MSIX;
+       new_ctrl = hw->ctrl | NFP_NET_CFG_CTRL_ENABLE;
        update = NFP_NET_CFG_UPDATE_GEN | NFP_NET_CFG_UPDATE_RING;
 
+       /* Just configuring queues interrupts when necessary */
+       if (rte_intr_dp_is_en(intr_handle))
+               update |= NFP_NET_CFG_UPDATE_MSIX;
+
        if (hw->cap & NFP_NET_CFG_CTRL_RINGCFG)
                new_ctrl |= NFP_NET_CFG_CTRL_RINGCFG;
 
@@ -659,10 +786,12 @@ static void
 nfp_net_close(struct rte_eth_dev *dev)
 {
        struct nfp_net_hw *hw;
+       struct rte_pci_device *pci_dev;
 
        PMD_INIT_LOG(DEBUG, "Close\n");
 
        hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       pci_dev = RTE_DEV_TO_PCI(dev->device);
 
        /*
         * We assume that the DPDK application is stopping all the
@@ -671,14 +800,380 @@ nfp_net_close(struct rte_eth_dev *dev)
 
        nfp_net_stop(dev);
 
+       rte_intr_disable(&pci_dev->intr_handle);
        nn_cfg_writeb(hw, NFP_NET_CFG_LSC, 0xff);
 
+       /* unregister callback func from eal lib */
+       rte_intr_callback_unregister(&pci_dev->intr_handle,
+                                    nfp_net_dev_interrupt_handler,
+                                    (void *)dev);
+
        /*
         * The ixgbe PMD driver disables the pcie master on the
         * device. The i40e does not...
         */
 }
 
+static void
+nfp_net_promisc_enable(struct rte_eth_dev *dev)
+{
+       uint32_t new_ctrl, update = 0;
+       struct nfp_net_hw *hw;
+
+       PMD_DRV_LOG(DEBUG, "Promiscuous mode enable\n");
+
+       hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       if (!(hw->cap & NFP_NET_CFG_CTRL_PROMISC)) {
+               PMD_INIT_LOG(INFO, "Promiscuous mode not supported\n");
+               return;
+       }
+
+       if (hw->ctrl & NFP_NET_CFG_CTRL_PROMISC) {
+               PMD_DRV_LOG(INFO, "Promiscuous mode already enabled\n");
+               return;
+       }
+
+       new_ctrl = hw->ctrl | NFP_NET_CFG_CTRL_PROMISC;
+       update = NFP_NET_CFG_UPDATE_GEN;
+
+       /*
+        * DPDK sets promiscuous mode on just after this call assuming
+        * it can not fail ...
+        */
+       if (nfp_net_reconfig(hw, new_ctrl, update) < 0)
+               return;
+
+       hw->ctrl = new_ctrl;
+}
+
+static void
+nfp_net_promisc_disable(struct rte_eth_dev *dev)
+{
+       uint32_t new_ctrl, update = 0;
+       struct nfp_net_hw *hw;
+
+       hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       if ((hw->ctrl & NFP_NET_CFG_CTRL_PROMISC) == 0) {
+               PMD_DRV_LOG(INFO, "Promiscuous mode already disabled\n");
+               return;
+       }
+
+       new_ctrl = hw->ctrl & ~NFP_NET_CFG_CTRL_PROMISC;
+       update = NFP_NET_CFG_UPDATE_GEN;
+
+       /*
+        * DPDK sets promiscuous mode off just before this call
+        * assuming it can not fail ...
+        */
+       if (nfp_net_reconfig(hw, new_ctrl, update) < 0)
+               return;
+
+       hw->ctrl = new_ctrl;
+}
+
+/*
+ * return 0 means link status changed, -1 means not changed
+ *
+ * Wait to complete is needed as it can take up to 9 seconds to get the Link
+ * status.
+ */
+static int
+nfp_net_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
+{
+       struct nfp_net_hw *hw;
+       struct rte_eth_link link, old;
+       uint32_t nn_link_status;
+
+       static const uint32_t ls_to_ethtool[] = {
+               [NFP_NET_CFG_STS_LINK_RATE_UNSUPPORTED] = ETH_SPEED_NUM_NONE,
+               [NFP_NET_CFG_STS_LINK_RATE_UNKNOWN]     = ETH_SPEED_NUM_NONE,
+               [NFP_NET_CFG_STS_LINK_RATE_1G]          = ETH_SPEED_NUM_1G,
+               [NFP_NET_CFG_STS_LINK_RATE_10G]         = ETH_SPEED_NUM_10G,
+               [NFP_NET_CFG_STS_LINK_RATE_25G]         = ETH_SPEED_NUM_25G,
+               [NFP_NET_CFG_STS_LINK_RATE_40G]         = ETH_SPEED_NUM_40G,
+               [NFP_NET_CFG_STS_LINK_RATE_50G]         = ETH_SPEED_NUM_50G,
+               [NFP_NET_CFG_STS_LINK_RATE_100G]        = ETH_SPEED_NUM_100G,
+       };
+
+       PMD_DRV_LOG(DEBUG, "Link update\n");
+
+       hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       memset(&old, 0, sizeof(old));
+       nfp_net_dev_atomic_read_link_status(dev, &old);
+
+       nn_link_status = nn_cfg_readl(hw, NFP_NET_CFG_STS);
+
+       memset(&link, 0, sizeof(struct rte_eth_link));
+
+       if (nn_link_status & NFP_NET_CFG_STS_LINK)
+               link.link_status = ETH_LINK_UP;
+
+       link.link_duplex = ETH_LINK_FULL_DUPLEX;
+
+       nn_link_status = (nn_link_status >> NFP_NET_CFG_STS_LINK_RATE_SHIFT) &
+                        NFP_NET_CFG_STS_LINK_RATE_MASK;
+
+       if ((NFD_CFG_MAJOR_VERSION_of(hw->ver) < 4) ||
+           ((NFD_CFG_MINOR_VERSION_of(hw->ver) == 4) &&
+           (NFD_CFG_MINOR_VERSION_of(hw->ver) == 0)))
+               /* We really do not know the speed wil old firmware */
+               link.link_speed = ETH_SPEED_NUM_NONE;
+       else {
+               if (nn_link_status >= RTE_DIM(ls_to_ethtool))
+                       link.link_speed = ETH_SPEED_NUM_NONE;
+               else
+                       link.link_speed = ls_to_ethtool[nn_link_status];
+       }
+
+       if (old.link_status != link.link_status) {
+               nfp_net_dev_atomic_write_link_status(dev, &link);
+               if (link.link_status)
+                       PMD_DRV_LOG(INFO, "NIC Link is Up\n");
+               else
+                       PMD_DRV_LOG(INFO, "NIC Link is Down\n");
+               return 0;
+       }
+
+       return -1;
+}
+
+static void
+nfp_net_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
+{
+       int i;
+       struct nfp_net_hw *hw;
+       struct rte_eth_stats nfp_dev_stats;
+
+       hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       /* RTE_ETHDEV_QUEUE_STAT_CNTRS default value is 16 */
+
+       /* reading per RX ring stats */
+       for (i = 0; i < dev->data->nb_rx_queues; i++) {
+               if (i == RTE_ETHDEV_QUEUE_STAT_CNTRS)
+                       break;
+
+               nfp_dev_stats.q_ipackets[i] =
+                       nn_cfg_readq(hw, NFP_NET_CFG_RXR_STATS(i));
+
+               nfp_dev_stats.q_ipackets[i] -=
+                       hw->eth_stats_base.q_ipackets[i];
+
+               nfp_dev_stats.q_ibytes[i] =
+                       nn_cfg_readq(hw, NFP_NET_CFG_RXR_STATS(i) + 0x8);
+
+               nfp_dev_stats.q_ibytes[i] -=
+                       hw->eth_stats_base.q_ibytes[i];
+       }
+
+       /* reading per TX ring stats */
+       for (i = 0; i < dev->data->nb_tx_queues; i++) {
+               if (i == RTE_ETHDEV_QUEUE_STAT_CNTRS)
+                       break;
+
+               nfp_dev_stats.q_opackets[i] =
+                       nn_cfg_readq(hw, NFP_NET_CFG_TXR_STATS(i));
+
+               nfp_dev_stats.q_opackets[i] -=
+                       hw->eth_stats_base.q_opackets[i];
+
+               nfp_dev_stats.q_obytes[i] =
+                       nn_cfg_readq(hw, NFP_NET_CFG_TXR_STATS(i) + 0x8);
+
+               nfp_dev_stats.q_obytes[i] -=
+                       hw->eth_stats_base.q_obytes[i];
+       }
+
+       nfp_dev_stats.ipackets =
+               nn_cfg_readq(hw, NFP_NET_CFG_STATS_RX_FRAMES);
+
+       nfp_dev_stats.ipackets -= hw->eth_stats_base.ipackets;
+
+       nfp_dev_stats.ibytes =
+               nn_cfg_readq(hw, NFP_NET_CFG_STATS_RX_OCTETS);
+
+       nfp_dev_stats.ibytes -= hw->eth_stats_base.ibytes;
+
+       nfp_dev_stats.opackets =
+               nn_cfg_readq(hw, NFP_NET_CFG_STATS_TX_FRAMES);
+
+       nfp_dev_stats.opackets -= hw->eth_stats_base.opackets;
+
+       nfp_dev_stats.obytes =
+               nn_cfg_readq(hw, NFP_NET_CFG_STATS_TX_OCTETS);
+
+       nfp_dev_stats.obytes -= hw->eth_stats_base.obytes;
+
+       /* reading general device stats */
+       nfp_dev_stats.ierrors =
+               nn_cfg_readq(hw, NFP_NET_CFG_STATS_RX_ERRORS);
+
+       nfp_dev_stats.ierrors -= hw->eth_stats_base.ierrors;
+
+       nfp_dev_stats.oerrors =
+               nn_cfg_readq(hw, NFP_NET_CFG_STATS_TX_ERRORS);
+
+       nfp_dev_stats.oerrors -= hw->eth_stats_base.oerrors;
+
+       /* RX ring mbuf allocation failures */
+       nfp_dev_stats.rx_nombuf = dev->data->rx_mbuf_alloc_failed;
+
+       nfp_dev_stats.imissed =
+               nn_cfg_readq(hw, NFP_NET_CFG_STATS_RX_DISCARDS);
+
+       nfp_dev_stats.imissed -= hw->eth_stats_base.imissed;
+
+       if (stats)
+               memcpy(stats, &nfp_dev_stats, sizeof(*stats));
+}
+
+static void
+nfp_net_stats_reset(struct rte_eth_dev *dev)
+{
+       int i;
+       struct nfp_net_hw *hw;
+
+       hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       /*
+        * hw->eth_stats_base records the per counter starting point.
+        * Lets update it now
+        */
+
+       /* reading per RX ring stats */
+       for (i = 0; i < dev->data->nb_rx_queues; i++) {
+               if (i == RTE_ETHDEV_QUEUE_STAT_CNTRS)
+                       break;
+
+               hw->eth_stats_base.q_ipackets[i] =
+                       nn_cfg_readq(hw, NFP_NET_CFG_RXR_STATS(i));
+
+               hw->eth_stats_base.q_ibytes[i] =
+                       nn_cfg_readq(hw, NFP_NET_CFG_RXR_STATS(i) + 0x8);
+       }
+
+       /* reading per TX ring stats */
+       for (i = 0; i < dev->data->nb_tx_queues; i++) {
+               if (i == RTE_ETHDEV_QUEUE_STAT_CNTRS)
+                       break;
+
+               hw->eth_stats_base.q_opackets[i] =
+                       nn_cfg_readq(hw, NFP_NET_CFG_TXR_STATS(i));
+
+               hw->eth_stats_base.q_obytes[i] =
+                       nn_cfg_readq(hw, NFP_NET_CFG_TXR_STATS(i) + 0x8);
+       }
+
+       hw->eth_stats_base.ipackets =
+               nn_cfg_readq(hw, NFP_NET_CFG_STATS_RX_FRAMES);
+
+       hw->eth_stats_base.ibytes =
+               nn_cfg_readq(hw, NFP_NET_CFG_STATS_RX_OCTETS);
+
+       hw->eth_stats_base.opackets =
+               nn_cfg_readq(hw, NFP_NET_CFG_STATS_TX_FRAMES);
+
+       hw->eth_stats_base.obytes =
+               nn_cfg_readq(hw, NFP_NET_CFG_STATS_TX_OCTETS);
+
+       /* reading general device stats */
+       hw->eth_stats_base.ierrors =
+               nn_cfg_readq(hw, NFP_NET_CFG_STATS_RX_ERRORS);
+
+       hw->eth_stats_base.oerrors =
+               nn_cfg_readq(hw, NFP_NET_CFG_STATS_TX_ERRORS);
+
+       /* RX ring mbuf allocation failures */
+       dev->data->rx_mbuf_alloc_failed = 0;
+
+       hw->eth_stats_base.imissed =
+               nn_cfg_readq(hw, NFP_NET_CFG_STATS_RX_DISCARDS);
+}
+
+static void
+nfp_net_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
+{
+       struct nfp_net_hw *hw;
+
+       hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       dev_info->pci_dev = RTE_DEV_TO_PCI(dev->device);
+       dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
+       dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
+       dev_info->min_rx_bufsize = ETHER_MIN_MTU;
+       dev_info->max_rx_pktlen = hw->mtu;
+       /* Next should change when PF support is implemented */
+       dev_info->max_mac_addrs = 1;
+
+       if (hw->cap & NFP_NET_CFG_CTRL_RXVLAN)
+               dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
+
+       if (hw->cap & NFP_NET_CFG_CTRL_RXCSUM)
+               dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_IPV4_CKSUM |
+                                            DEV_RX_OFFLOAD_UDP_CKSUM |
+                                            DEV_RX_OFFLOAD_TCP_CKSUM;
+
+       if (hw->cap & NFP_NET_CFG_CTRL_TXVLAN)
+               dev_info->tx_offload_capa = DEV_TX_OFFLOAD_VLAN_INSERT;
+
+       if (hw->cap & NFP_NET_CFG_CTRL_TXCSUM)
+               dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_IPV4_CKSUM |
+                                            DEV_TX_OFFLOAD_UDP_CKSUM |
+                                            DEV_TX_OFFLOAD_TCP_CKSUM;
+
+       dev_info->default_rxconf = (struct rte_eth_rxconf) {
+               .rx_thresh = {
+                       .pthresh = DEFAULT_RX_PTHRESH,
+                       .hthresh = DEFAULT_RX_HTHRESH,
+                       .wthresh = DEFAULT_RX_WTHRESH,
+               },
+               .rx_free_thresh = DEFAULT_RX_FREE_THRESH,
+               .rx_drop_en = 0,
+       };
+
+       dev_info->default_txconf = (struct rte_eth_txconf) {
+               .tx_thresh = {
+                       .pthresh = DEFAULT_TX_PTHRESH,
+                       .hthresh = DEFAULT_TX_HTHRESH,
+                       .wthresh = DEFAULT_TX_WTHRESH,
+               },
+               .tx_free_thresh = DEFAULT_TX_FREE_THRESH,
+               .tx_rs_thresh = DEFAULT_TX_RSBIT_THRESH,
+               .txq_flags = ETH_TXQ_FLAGS_NOMULTSEGS |
+                            ETH_TXQ_FLAGS_NOOFFLOADS,
+       };
+
+       dev_info->reta_size = NFP_NET_CFG_RSS_ITBL_SZ;
+       dev_info->hash_key_size = NFP_NET_CFG_RSS_KEY_SZ;
+
+       dev_info->speed_capa = ETH_SPEED_NUM_1G | ETH_LINK_SPEED_10G |
+                              ETH_SPEED_NUM_25G | ETH_SPEED_NUM_40G |
+                              ETH_SPEED_NUM_50G | ETH_LINK_SPEED_100G;
+
+       if (hw->cap & NFP_NET_CFG_CTRL_LSO)
+               dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
+}
+
+static const uint32_t *
+nfp_net_supported_ptypes_get(struct rte_eth_dev *dev)
+{
+       static const uint32_t ptypes[] = {
+               /* refers to nfp_net_set_hash() */
+               RTE_PTYPE_INNER_L3_IPV4,
+               RTE_PTYPE_INNER_L3_IPV6,
+               RTE_PTYPE_INNER_L3_IPV6_EXT,
+               RTE_PTYPE_INNER_L4_MASK,
+               RTE_PTYPE_UNKNOWN
+       };
+
+       if (dev->rx_pkt_burst == nfp_net_recv_pkts)
+               return ptypes;
+       return NULL;
+}
+
 static uint32_t
 nfp_net_rx_queue_count(struct rte_eth_dev *dev, uint16_t queue_idx)
 {
@@ -694,8 +1189,7 @@ nfp_net_rx_queue_count(struct rte_eth_dev *dev, uint16_t queue_idx)
                return 0;
        }
 
-       idx = rxq->rd_p % rxq->rx_count;
-       rxds = &rxq->rxds[idx];
+       idx = rxq->rd_p;
 
        count = 0;
 
@@ -723,6 +1217,184 @@ nfp_net_rx_queue_count(struct rte_eth_dev *dev, uint16_t queue_idx)
        return count;
 }
 
+static int
+nfp_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
+{
+       struct rte_pci_device *pci_dev;
+       struct nfp_net_hw *hw;
+       int base = 0;
+
+       hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       pci_dev = RTE_DEV_TO_PCI(dev->device);
+
+       if (pci_dev->intr_handle.type != RTE_INTR_HANDLE_UIO)
+               base = 1;
+
+       /* Make sure all updates are written before un-masking */
+       rte_wmb();
+       nn_cfg_writeb(hw, NFP_NET_CFG_ICR(base + queue_id),
+                     NFP_NET_CFG_ICR_UNMASKED);
+       return 0;
+}
+
+static int
+nfp_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
+{
+       struct rte_pci_device *pci_dev;
+       struct nfp_net_hw *hw;
+       int base = 0;
+
+       hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       pci_dev = RTE_DEV_TO_PCI(dev->device);
+
+       if (pci_dev->intr_handle.type != RTE_INTR_HANDLE_UIO)
+               base = 1;
+
+       /* Make sure all updates are written before un-masking */
+       rte_wmb();
+       nn_cfg_writeb(hw, NFP_NET_CFG_ICR(base + queue_id), 0x1);
+       return 0;
+}
+
+static void
+nfp_net_dev_link_status_print(struct rte_eth_dev *dev)
+{
+       struct rte_pci_device *pci_dev = RTE_DEV_TO_PCI(dev->device);
+       struct rte_eth_link link;
+
+       memset(&link, 0, sizeof(link));
+       nfp_net_dev_atomic_read_link_status(dev, &link);
+       if (link.link_status)
+               RTE_LOG(INFO, PMD, "Port %d: Link Up - speed %u Mbps - %s\n",
+                       (int)(dev->data->port_id), (unsigned)link.link_speed,
+                       link.link_duplex == ETH_LINK_FULL_DUPLEX
+                       ? "full-duplex" : "half-duplex");
+       else
+               RTE_LOG(INFO, PMD, " Port %d: Link Down\n",
+                       (int)(dev->data->port_id));
+
+       RTE_LOG(INFO, PMD, "PCI Address: %04d:%02d:%02d:%d\n",
+               pci_dev->addr.domain, pci_dev->addr.bus,
+               pci_dev->addr.devid, pci_dev->addr.function);
+}
+
+/* Interrupt configuration and handling */
+
+/*
+ * nfp_net_irq_unmask - Unmask an interrupt
+ *
+ * If MSI-X auto-masking is enabled clear the mask bit, otherwise
+ * clear the ICR for the entry.
+ */
+static void
+nfp_net_irq_unmask(struct rte_eth_dev *dev)
+{
+       struct nfp_net_hw *hw;
+       struct rte_pci_device *pci_dev;
+
+       hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       pci_dev = RTE_DEV_TO_PCI(dev->device);
+
+       if (hw->ctrl & NFP_NET_CFG_CTRL_MSIXAUTO) {
+               /* If MSI-X auto-masking is used, clear the entry */
+               rte_wmb();
+               rte_intr_enable(&pci_dev->intr_handle);
+       } else {
+               /* Make sure all updates are written before un-masking */
+               rte_wmb();
+               nn_cfg_writeb(hw, NFP_NET_CFG_ICR(NFP_NET_IRQ_LSC_IDX),
+                             NFP_NET_CFG_ICR_UNMASKED);
+       }
+}
+
+static void
+nfp_net_dev_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
+                             void *param)
+{
+       int64_t timeout;
+       struct rte_eth_link link;
+       struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
+
+       PMD_DRV_LOG(DEBUG, "We got a LSC interrupt!!!\n");
+
+       /* get the link status */
+       memset(&link, 0, sizeof(link));
+       nfp_net_dev_atomic_read_link_status(dev, &link);
+
+       nfp_net_link_update(dev, 0);
+
+       /* likely to up */
+       if (!link.link_status) {
+               /* handle it 1 sec later, wait it being stable */
+               timeout = NFP_NET_LINK_UP_CHECK_TIMEOUT;
+               /* likely to down */
+       } else {
+               /* handle it 4 sec later, wait it being stable */
+               timeout = NFP_NET_LINK_DOWN_CHECK_TIMEOUT;
+       }
+
+       if (rte_eal_alarm_set(timeout * 1000,
+                             nfp_net_dev_interrupt_delayed_handler,
+                             (void *)dev) < 0) {
+               RTE_LOG(ERR, PMD, "Error setting alarm");
+               /* Unmasking */
+               nfp_net_irq_unmask(dev);
+       }
+}
+
+/*
+ * Interrupt handler which shall be registered for alarm callback for delayed
+ * handling specific interrupt to wait for the stable nic state. As the NIC
+ * interrupt state is not stable for nfp after link is just down, it needs
+ * to wait 4 seconds to get the stable status.
+ *
+ * @param handle   Pointer to interrupt handle.
+ * @param param    The address of parameter (struct rte_eth_dev *)
+ *
+ * @return  void
+ */
+static void
+nfp_net_dev_interrupt_delayed_handler(void *param)
+{
+       struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
+
+       nfp_net_link_update(dev, 0);
+       _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL);
+
+       nfp_net_dev_link_status_print(dev);
+
+       /* Unmasking */
+       nfp_net_irq_unmask(dev);
+}
+
+static int
+nfp_net_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
+{
+       struct nfp_net_hw *hw;
+
+       hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       /* check that mtu is within the allowed range */
+       if ((mtu < ETHER_MIN_MTU) || ((uint32_t)mtu > hw->max_mtu))
+               return -EINVAL;
+
+       /* switch to jumbo mode if needed */
+       if ((uint32_t)mtu > ETHER_MAX_LEN)
+               dev->data->dev_conf.rxmode.jumbo_frame = 1;
+       else
+               dev->data->dev_conf.rxmode.jumbo_frame = 0;
+
+       /* update max frame size */
+       dev->data->dev_conf.rxmode.max_rx_pkt_len = (uint32_t)mtu;
+
+       /* writing to configuration space */
+       nn_cfg_writel(hw, NFP_NET_CFG_MTU, (uint32_t)mtu);
+
+       hw->mtu = mtu;
+
+       return 0;
+}
+
 static int
 nfp_net_rx_queue_setup(struct rte_eth_dev *dev,
                       uint16_t queue_idx, uint16_t nb_desc,
@@ -743,7 +1415,7 @@ nfp_net_rx_queue_setup(struct rte_eth_dev *dev,
            (nb_desc > NFP_NET_MAX_RX_DESC) ||
            (nb_desc < NFP_NET_MIN_RX_DESC)) {
                RTE_LOG(ERR, PMD, "Wrong nb_desc value\n");
-               return (-EINVAL);
+               return -EINVAL;
        }
 
        /*
@@ -759,7 +1431,7 @@ nfp_net_rx_queue_setup(struct rte_eth_dev *dev,
        rxq = rte_zmalloc_socket("ethdev RX queue", sizeof(struct nfp_net_rxq),
                                 RTE_CACHE_LINE_SIZE, socket_id);
        if (rxq == NULL)
-               return (-ENOMEM);
+               return -ENOMEM;
 
        /* Hw queues mapping based on firmware confifguration */
        rxq->qidx = queue_idx;
@@ -796,7 +1468,7 @@ nfp_net_rx_queue_setup(struct rte_eth_dev *dev,
        if (tz == NULL) {
                RTE_LOG(ERR, PMD, "Error allocatig rx dma\n");
                nfp_net_rx_queue_release(rxq);
-               return (-ENOMEM);
+               return -ENOMEM;
        }
 
        /* Saving physical and virtual addresses for the RX ring */
@@ -809,7 +1481,7 @@ nfp_net_rx_queue_setup(struct rte_eth_dev *dev,
                                         RTE_CACHE_LINE_SIZE, socket_id);
        if (rxq->rxbufs == NULL) {
                nfp_net_rx_queue_release(rxq);
-               return (-ENOMEM);
+               return -ENOMEM;
        }
 
        PMD_RX_LOG(DEBUG, "rxbufs=%p hw_ring=%p dma_addr=0x%" PRIx64 "\n",
@@ -847,7 +1519,7 @@ nfp_net_rx_fill_freelist(struct nfp_net_rxq *rxq)
                if (mbuf == NULL) {
                        RTE_LOG(ERR, PMD, "RX mbuf alloc failed queue_id=%u\n",
                                (unsigned)rxq->qidx);
-                       return (-ENOMEM);
+                       return -ENOMEM;
                }
 
                dma_addr = rte_cpu_to_le_64(RTE_MBUF_DMA_ADDR_DEFAULT(mbuf));
@@ -858,8 +1530,6 @@ nfp_net_rx_fill_freelist(struct nfp_net_rxq *rxq)
                rxd->fld.dma_addr_lo = dma_addr & 0xffffffff;
                rxe[i].mbuf = mbuf;
                PMD_RX_LOG(DEBUG, "[%d]: %" PRIx64 "\n", i, dma_addr);
-
-               rxq->wr_p++;
        }
 
        /* Make sure all writes are flushed before telling the hardware */
@@ -925,7 +1595,7 @@ nfp_net_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
                                 RTE_CACHE_LINE_SIZE, socket_id);
        if (txq == NULL) {
                RTE_LOG(ERR, PMD, "Error allocating tx dma\n");
-               return (-ENOMEM);
+               return -ENOMEM;
        }
 
        /*
@@ -939,11 +1609,10 @@ nfp_net_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
        if (tz == NULL) {
                RTE_LOG(ERR, PMD, "Error allocating tx dma\n");
                nfp_net_tx_queue_release(txq);
-               return (-ENOMEM);
+               return -ENOMEM;
        }
 
        txq->tx_count = nb_desc;
-       txq->tail = 0;
        txq->tx_free_thresh = tx_free_thresh;
        txq->tx_pthresh = tx_conf->tx_thresh.pthresh;
        txq->tx_hthresh = tx_conf->tx_thresh.hthresh;
@@ -967,7 +1636,7 @@ nfp_net_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
                                         RTE_CACHE_LINE_SIZE, socket_id);
        if (txq->txbufs == NULL) {
                nfp_net_tx_queue_release(txq);
-               return (-ENOMEM);
+               return -ENOMEM;
        }
        PMD_TX_LOG(DEBUG, "txbufs=%p hw_ring=%p dma_addr=0x%" PRIx64 "\n",
                   txq->txbufs, txq->txds, (unsigned long int)txq->dma);
@@ -987,12 +1656,33 @@ nfp_net_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
        return 0;
 }
 
+/* nfp_net_tx_tso - Set TX descriptor for TSO */
+static inline void
+nfp_net_tx_tso(struct nfp_net_txq *txq, struct nfp_net_tx_desc *txd,
+              struct rte_mbuf *mb)
+{
+       uint64_t ol_flags;
+       struct nfp_net_hw *hw = txq->hw;
+
+       if (!(hw->cap & NFP_NET_CFG_CTRL_LSO))
+               return;
+
+       ol_flags = mb->ol_flags;
+
+       if (!(ol_flags & PKT_TX_TCP_SEG))
+               return;
+
+       txd->l4_offset = mb->l2_len + mb->l3_len + mb->l4_len;
+       txd->lso = rte_cpu_to_le_16(mb->tso_segsz);
+       txd->flags |= PCIE_DESC_TX_LSO;
+}
+
 /* nfp_net_tx_cksum - Set TX CSUM offload flags in TX descriptor */
 static inline void
 nfp_net_tx_cksum(struct nfp_net_txq *txq, struct nfp_net_tx_desc *txd,
                 struct rte_mbuf *mb)
 {
-       uint16_t ol_flags;
+       uint64_t ol_flags;
        struct nfp_net_hw *hw = txq->hw;
 
        if (!(hw->cap & NFP_NET_CFG_CTRL_TXCSUM))
@@ -1013,7 +1703,8 @@ nfp_net_tx_cksum(struct nfp_net_txq *txq, struct nfp_net_tx_desc *txd,
                break;
        }
 
-       txd->flags |= PCIE_DESC_TX_CSUM;
+       if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_L4_MASK))
+               txd->flags |= PCIE_DESC_TX_CSUM;
 }
 
 /* nfp_net_rx_cksum - set mbuf checksum flags based on RX descriptor flags */
@@ -1071,12 +1762,6 @@ nfp_net_set_hash(struct nfp_net_rxq *rxq, struct nfp_net_rx_desc *rxd,
        hash = rte_be_to_cpu_32(*(uint32_t *)NFP_HASH_OFFSET);
        hash_type = rte_be_to_cpu_32(*(uint32_t *)NFP_HASH_TYPE_OFFSET);
 
-       /*
-        * hash type is sharing the same word with input port info
-        * 31-8: input port
-        * 7:0: hash type
-        */
-       hash_type &= 0xff;
        mbuf->hash.rss = hash;
        mbuf->ol_flags |= PKT_RX_RSS_HASH;
 
@@ -1095,29 +1780,6 @@ nfp_net_set_hash(struct nfp_net_rxq *rxq, struct nfp_net_rx_desc *rxd,
        }
 }
 
-/* nfp_net_check_port - Set mbuf in_port field */
-static void
-nfp_net_check_port(struct nfp_net_rx_desc *rxd, struct rte_mbuf *mbuf)
-{
-       uint32_t port;
-
-       if (!(rxd->rxd.flags & PCIE_DESC_RX_INGRESS_PORT)) {
-               mbuf->port = 0;
-               return;
-       }
-
-       port = rte_be_to_cpu_32(*(uint32_t *)((uint8_t *)mbuf->buf_addr +
-                                             mbuf->data_off - 8));
-
-       /*
-        * hash type is sharing the same word with input port info
-        * 31-8: input port
-        * 7:0: hash type
-        */
-       port = (uint8_t)(port >> 8);
-       mbuf->port = port;
-}
-
 static inline void
 nfp_net_mbuf_alloc_failed(struct nfp_net_rxq *rxq)
 {
@@ -1163,7 +1825,6 @@ nfp_net_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
        struct nfp_net_hw *hw;
        struct rte_mbuf *mb;
        struct rte_mbuf *new_mb;
-       int idx;
        uint16_t nb_hold;
        uint64_t dma_addr;
        int avail;
@@ -1174,7 +1835,7 @@ nfp_net_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
                 * DPDK just checks the queue is lower than max queues
                 * enabled. But the queue needs to be configured
                 */
-               RTE_LOG(ERR, PMD, "RX Bad queue\n");
+               RTE_LOG_DP(ERR, PMD, "RX Bad queue\n");
                return -EINVAL;
        }
 
@@ -1183,11 +1844,9 @@ nfp_net_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
        nb_hold = 0;
 
        while (avail < nb_pkts) {
-               idx = rxq->rd_p % rxq->rx_count;
-
-               rxb = &rxq->rxbufs[idx];
+               rxb = &rxq->rxbufs[rxq->rd_p];
                if (unlikely(rxb == NULL)) {
-                       RTE_LOG(ERR, PMD, "rxb does not exist!\n");
+                       RTE_LOG_DP(ERR, PMD, "rxb does not exist!\n");
                        break;
                }
 
@@ -1197,7 +1856,7 @@ nfp_net_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
                 */
                rte_rmb();
 
-               rxds = &rxq->rxds[idx];
+               rxds = &rxq->rxds[rxq->rd_p];
                if ((rxds->rxd.meta_len_dd & PCIE_DESC_RX_DD) == 0)
                        break;
 
@@ -1207,7 +1866,7 @@ nfp_net_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
                 */
                new_mb = rte_pktmbuf_alloc(rxq->mem_pool);
                if (unlikely(new_mb == NULL)) {
-                       RTE_LOG(DEBUG, PMD, "RX mbuf alloc failed port_id=%u "
+                       RTE_LOG_DP(DEBUG, PMD, "RX mbuf alloc failed port_id=%u "
                                "queue_id=%u\n", (unsigned)rxq->port_id,
                                (unsigned)rxq->qidx);
                        nfp_net_mbuf_alloc_failed(rxq);
@@ -1238,7 +1897,7 @@ nfp_net_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
                         * responsibility of avoiding it. But we have
                         * to give some info about the error
                         */
-                       RTE_LOG(ERR, PMD,
+                       RTE_LOG_DP(ERR, PMD,
                                "mbuf overflow likely due to the RX offset.\n"
                                "\t\tYour mbuf size should have extra space for"
                                " RX offset=%u bytes.\n"
@@ -1267,13 +1926,10 @@ nfp_net_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
                /* Checking the checksum flag */
                nfp_net_rx_cksum(rxq, rxds, mb);
 
-               /* Checking the port flag */
-               nfp_net_check_port(rxds, mb);
-
                if ((rxds->rxd.flags & PCIE_DESC_RX_VLAN) &&
                    (hw->ctrl & NFP_NET_CFG_CTRL_RXVLAN)) {
                        mb->vlan_tci = rte_cpu_to_le_32(rxds->rxd.vlan);
-                       mb->ol_flags |= PKT_RX_VLAN_PKT;
+                       mb->ol_flags |= PKT_RX_VLAN_PKT | PKT_RX_VLAN_STRIPPED;
                }
 
                /* Adding the mbuff to the mbuff array passed by the app */
@@ -1288,6 +1944,8 @@ nfp_net_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
                rxds->fld.dma_addr_lo = dma_addr & 0xffffffff;
 
                rxq->rd_p++;
+               if (unlikely(rxq->rd_p == rxq->rx_count)) /* wrapping?*/
+                       rxq->rd_p = 0;
        }
 
        if (nb_hold == 0)
@@ -1333,33 +1991,40 @@ nfp_net_tx_free_bufs(struct nfp_net_txq *txq)
        /* Work out how many packets have been sent */
        qcp_rd_p = nfp_qcp_read(txq->qcp_q, NFP_QCP_READ_PTR);
 
-       if (qcp_rd_p == txq->qcp_rd_p) {
+       if (qcp_rd_p == txq->rd_p) {
                PMD_TX_LOG(DEBUG, "queue %u: It seems harrier is not sending "
                           "packets (%u, %u)\n", txq->qidx,
-                          qcp_rd_p, txq->qcp_rd_p);
+                          qcp_rd_p, txq->rd_p);
                return 0;
        }
 
-       if (qcp_rd_p > txq->qcp_rd_p)
-               todo = qcp_rd_p - txq->qcp_rd_p;
+       if (qcp_rd_p > txq->rd_p)
+               todo = qcp_rd_p - txq->rd_p;
        else
-               todo = qcp_rd_p + txq->tx_count - txq->qcp_rd_p;
+               todo = qcp_rd_p + txq->tx_count - txq->rd_p;
 
-       PMD_TX_LOG(DEBUG, "qcp_rd_p %u, txq->qcp_rd_p: %u, qcp->rd_p: %u\n",
-                  qcp_rd_p, txq->qcp_rd_p, txq->rd_p);
+       PMD_TX_LOG(DEBUG, "qcp_rd_p %u, txq->rd_p: %u, qcp->rd_p: %u\n",
+                  qcp_rd_p, txq->rd_p, txq->rd_p);
 
        if (todo == 0)
                return todo;
 
-       txq->qcp_rd_p += todo;
-       txq->qcp_rd_p %= txq->tx_count;
        txq->rd_p += todo;
+       if (unlikely(txq->rd_p >= txq->tx_count))
+               txq->rd_p -= txq->tx_count;
 
        return todo;
 }
 
 /* Leaving always free descriptors for avoiding wrapping confusion */
-#define NFP_FREE_TX_DESC(t) (t->tx_count - (t->wr_p - t->rd_p) - 8)
+static inline
+uint32_t nfp_free_tx_desc(struct nfp_net_txq *txq)
+{
+       if (txq->wr_p >= txq->rd_p)
+               return txq->tx_count - (txq->wr_p - txq->rd_p) - 8;
+       else
+               return txq->rd_p - txq->wr_p - 8;
+}
 
 /*
  * nfp_net_txq_full - Check if the TX queue free descriptors
@@ -1370,9 +2035,9 @@ nfp_net_tx_free_bufs(struct nfp_net_txq *txq)
  * This function uses the host copy* of read/write pointers
  */
 static inline
-int nfp_net_txq_full(struct nfp_net_txq *txq)
+uint32_t nfp_net_txq_full(struct nfp_net_txq *txq)
 {
-       return NFP_FREE_TX_DESC(txq) < txq->tx_free_thresh;
+       return (nfp_free_tx_desc(txq) < txq->tx_free_thresh);
 }
 
 static uint16_t
@@ -1380,7 +2045,7 @@ nfp_net_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 {
        struct nfp_net_txq *txq;
        struct nfp_net_hw *hw;
-       struct nfp_net_tx_desc *txds;
+       struct nfp_net_tx_desc *txds, txd;
        struct rte_mbuf *pkt;
        uint64_t dma_addr;
        int pkt_size, dma_size;
@@ -1390,15 +2055,15 @@ nfp_net_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 
        txq = tx_queue;
        hw = txq->hw;
-       txds = &txq->txds[txq->tail];
+       txds = &txq->txds[txq->wr_p];
 
        PMD_TX_LOG(DEBUG, "working for queue %u at pos %d and %u packets\n",
-                  txq->qidx, txq->tail, nb_pkts);
+                  txq->qidx, txq->wr_p, nb_pkts);
 
-       if ((NFP_FREE_TX_DESC(txq) < nb_pkts) || (nfp_net_txq_full(txq)))
+       if ((nfp_free_tx_desc(txq) < nb_pkts) || (nfp_net_txq_full(txq)))
                nfp_net_tx_free_bufs(txq);
 
-       free_descs = (uint16_t)NFP_FREE_TX_DESC(txq);
+       free_descs = (uint16_t)nfp_free_tx_desc(txq);
        if (unlikely(free_descs == 0))
                return 0;
 
@@ -1411,7 +2076,7 @@ nfp_net_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
        /* Sending packets */
        while ((i < nb_pkts) && free_descs) {
                /* Grabbing the mbuf linked to the current descriptor */
-               lmbuf = &txq->txbufs[txq->tail].mbuf;
+               lmbuf = &txq->txbufs[txq->wr_p].mbuf;
                /* Warming the cache for releasing the mbuf later on */
                RTE_MBUF_PREFETCH_TO_FREE(*lmbuf);
 
@@ -1429,19 +2094,17 @@ nfp_net_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 
                /*
                 * Checksum and VLAN flags just in the first descriptor for a
-                * multisegment packet
+                * multisegment packet, but TSO info needs to be in all of them.
                 */
-               nfp_net_tx_cksum(txq, txds, pkt);
+               nfp_net_tx_tso(txq, &txd, pkt);
+               nfp_net_tx_cksum(txq, &txd, pkt);
 
                if ((pkt->ol_flags & PKT_TX_VLAN_PKT) &&
                    (hw->cap & NFP_NET_CFG_CTRL_TXVLAN)) {
-                       txds->flags |= PCIE_DESC_TX_VLAN;
-                       txds->vlan = pkt->vlan_tci;
+                       txd.flags |= PCIE_DESC_TX_VLAN;
+                       txd.vlan = pkt->vlan_tci;
                }
 
-               if (pkt->ol_flags & PKT_TX_TCP_SEG)
-                       rte_panic("TSO is not supported\n");
-
                /*
                 * mbuf data_len is the data in one segment and pkt_len data
                 * in the whole packet. When the packet is just one segment,
@@ -1449,13 +2112,20 @@ nfp_net_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
                 */
                pkt_size = pkt->pkt_len;
 
-               while (pkt_size) {
-                       /* Releasing mbuf which was prefetched above */
-                       if (*lmbuf)
-                               rte_pktmbuf_free_seg(*lmbuf);
+               /* Releasing mbuf which was prefetched above */
+               if (*lmbuf)
+                       rte_pktmbuf_free(*lmbuf);
+               /*
+                * Linking mbuf with descriptor for being released
+                * next time descriptor is used
+                */
+               *lmbuf = pkt;
 
+               while (pkt_size) {
+                       /* Copying TSO, VLAN and cksum info */
+                       *txds = txd;
                        dma_size = pkt->data_len;
-                       dma_addr = RTE_MBUF_DATA_DMA_ADDR(pkt);
+                       dma_addr = rte_mbuf_data_dma_addr(pkt);
                        PMD_TX_LOG(DEBUG, "Working with mbuf at dma address:"
                                   "%" PRIx64 "\n", dma_addr);
 
@@ -1467,16 +2137,9 @@ nfp_net_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
                        ASSERT(free_descs > 0);
                        free_descs--;
 
-                       /*
-                        * Linking mbuf with descriptor for being released
-                        * next time descriptor is used
-                        */
-                       *lmbuf = pkt;
-
                        txq->wr_p++;
-                       txq->tail++;
-                       if (unlikely(txq->tail == txq->tx_count)) /* wrapping?*/
-                               txq->tail = 0;
+                       if (unlikely(txq->wr_p == txq->tx_count)) /* wrapping?*/
+                               txq->wr_p = 0;
 
                        pkt_size -= dma_size;
                        if (!pkt_size) {
@@ -1487,7 +2150,7 @@ nfp_net_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
                                pkt = pkt->next;
                        }
                        /* Referencing next free TX descriptor */
-                       txds = &txq->txds[txq->tail];
+                       txds = &txq->txds[txq->wr_p];
                        issued_descs++;
                }
                i++;
@@ -1501,6 +2164,41 @@ xmit_end:
        return i;
 }
 
+static void
+nfp_net_vlan_offload_set(struct rte_eth_dev *dev, int mask)
+{
+       uint32_t new_ctrl, update;
+       struct nfp_net_hw *hw;
+
+       hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       new_ctrl = 0;
+
+       if ((mask & ETH_VLAN_FILTER_OFFLOAD) ||
+           (mask & ETH_VLAN_EXTEND_OFFLOAD))
+               RTE_LOG(INFO, PMD, "No support for ETH_VLAN_FILTER_OFFLOAD or"
+                       " ETH_VLAN_EXTEND_OFFLOAD");
+
+       /* Enable vlan strip if it is not configured yet */
+       if ((mask & ETH_VLAN_STRIP_OFFLOAD) &&
+           !(hw->ctrl & NFP_NET_CFG_CTRL_RXVLAN))
+               new_ctrl = hw->ctrl | NFP_NET_CFG_CTRL_RXVLAN;
+
+       /* Disable vlan strip just if it is configured */
+       if (!(mask & ETH_VLAN_STRIP_OFFLOAD) &&
+           (hw->ctrl & NFP_NET_CFG_CTRL_RXVLAN))
+               new_ctrl = hw->ctrl & ~NFP_NET_CFG_CTRL_RXVLAN;
+
+       if (new_ctrl == 0)
+               return;
+
+       update = NFP_NET_CFG_UPDATE_GEN;
+
+       if (nfp_net_reconfig(hw, new_ctrl, update) < 0)
+               return;
+
+       hw->ctrl = new_ctrl;
+}
+
 /* Update Redirection Table(RETA) of Receive Side Scaling of Ethernet device */
 static int
 nfp_net_reta_update(struct rte_eth_dev *dev,
@@ -1716,11 +2414,20 @@ nfp_net_rss_hash_conf_get(struct rte_eth_dev *dev,
 }
 
 /* Initialise and register driver with DPDK Application */
-static struct eth_dev_ops nfp_net_eth_dev_ops = {
+static const struct eth_dev_ops nfp_net_eth_dev_ops = {
        .dev_configure          = nfp_net_configure,
        .dev_start              = nfp_net_start,
        .dev_stop               = nfp_net_stop,
        .dev_close              = nfp_net_close,
+       .promiscuous_enable     = nfp_net_promisc_enable,
+       .promiscuous_disable    = nfp_net_promisc_disable,
+       .link_update            = nfp_net_link_update,
+       .stats_get              = nfp_net_stats_get,
+       .stats_reset            = nfp_net_stats_reset,
+       .dev_infos_get          = nfp_net_infos_get,
+       .dev_supported_ptypes_get = nfp_net_supported_ptypes_get,
+       .mtu_set                = nfp_net_dev_mtu_set,
+       .vlan_offload_set       = nfp_net_vlan_offload_set,
        .reta_update            = nfp_net_reta_update,
        .reta_query             = nfp_net_reta_query,
        .rss_hash_update        = nfp_net_rss_hash_update,
@@ -1730,6 +2437,8 @@ static struct eth_dev_ops nfp_net_eth_dev_ops = {
        .rx_queue_count         = nfp_net_rx_queue_count,
        .tx_queue_setup         = nfp_net_tx_queue_setup,
        .tx_queue_release       = nfp_net_tx_queue_release,
+       .rx_queue_intr_enable   = nfp_rx_queue_intr_enable,
+       .rx_queue_intr_disable  = nfp_rx_queue_intr_disable,
 };
 
 static int
@@ -1754,7 +2463,10 @@ nfp_net_init(struct rte_eth_dev *eth_dev)
        if (rte_eal_process_type() != RTE_PROC_PRIMARY)
                return 0;
 
-       pci_dev = eth_dev->pci_dev;
+       pci_dev = RTE_DEV_TO_PCI(eth_dev->device);
+       rte_eth_copy_pci_info(eth_dev, pci_dev);
+       eth_dev->data->dev_flags = RTE_ETH_DEV_DETACHABLE;
+
        hw->device_id = pci_dev->id.device_id;
        hw->vendor_id = pci_dev->id.vendor_id;
        hw->subsystem_device_id = pci_dev->id.subsystem_device_id;
@@ -1822,7 +2534,6 @@ nfp_net_init(struct rte_eth_dev *eth_dev)
                     hw->cap & NFP_NET_CFG_CTRL_LSO     ? "TSO "     : "",
                     hw->cap & NFP_NET_CFG_CTRL_RSS     ? "RSS "     : "");
 
-       pci_dev = eth_dev->pci_dev;
        hw->ctrl = 0;
 
        hw->stride_rx = stride;
@@ -1831,6 +2542,9 @@ nfp_net_init(struct rte_eth_dev *eth_dev)
        PMD_INIT_LOG(INFO, "max_rx_queues: %u, max_tx_queues: %u\n",
                     hw->max_rx_queues, hw->max_tx_queues);
 
+       /* Initializing spinlock for reconfigs */
+       rte_spinlock_init(&hw->reconfig_lock);
+
        /* Allocating memory for mac addr */
        eth_dev->data->mac_addrs = rte_zmalloc("mac_addr", ETHER_ADDR_LEN, 0);
        if (eth_dev->data->mac_addrs == NULL) {
@@ -1838,12 +2552,15 @@ nfp_net_init(struct rte_eth_dev *eth_dev)
                return -ENOMEM;
        }
 
-       /* Using random mac addresses for VFs */
-       eth_random_addr(&hw->mac_addr[0]);
+       nfp_net_read_mac(hw);
+
+       if (!is_valid_assigned_ether_addr((struct ether_addr *)&hw->mac_addr))
+               /* Using random mac addresses for VFs */
+               eth_random_addr(&hw->mac_addr[0]);
 
        /* Copying mac address to DPDK eth_dev struct */
-       ether_addr_copy(&eth_dev->data->mac_addrs[0],
-                       (struct ether_addr *)hw->mac_addr);
+       ether_addr_copy((struct ether_addr *)hw->mac_addr,
+                       &eth_dev->data->mac_addrs[0]);
 
        PMD_INIT_LOG(INFO, "port %d VendorID=0x%x DeviceID=0x%x "
                     "mac=%02x:%02x:%02x:%02x:%02x:%02x",
@@ -1852,21 +2569,28 @@ nfp_net_init(struct rte_eth_dev *eth_dev)
                     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
                     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
 
+       /* Registering LSC interrupt handler */
+       rte_intr_callback_register(&pci_dev->intr_handle,
+                                  nfp_net_dev_interrupt_handler,
+                                  (void *)eth_dev);
+
+       /* Telling the firmware about the LSC interrupt entry */
+       nn_cfg_writeb(hw, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX);
+
+       /* Recording current stats counters values */
+       nfp_net_stats_reset(eth_dev);
+
        return 0;
 }
 
 static struct rte_pci_id pci_id_nfp_net_map[] = {
        {
-               .vendor_id = PCI_VENDOR_ID_NETRONOME,
-               .device_id = PCI_DEVICE_ID_NFP6000_PF_NIC,
-               .subsystem_vendor_id = PCI_ANY_ID,
-               .subsystem_device_id = PCI_ANY_ID,
+               RTE_PCI_DEVICE(PCI_VENDOR_ID_NETRONOME,
+                              PCI_DEVICE_ID_NFP6000_PF_NIC)
        },
        {
-               .vendor_id = PCI_VENDOR_ID_NETRONOME,
-               .device_id = PCI_DEVICE_ID_NFP6000_VF_NIC,
-               .subsystem_vendor_id = PCI_ANY_ID,
-               .subsystem_device_id = PCI_ANY_ID,
+               RTE_PCI_DEVICE(PCI_VENDOR_ID_NETRONOME,
+                              PCI_DEVICE_ID_NFP6000_VF_NIC)
        },
        {
                .vendor_id = 0,
@@ -1874,33 +2598,19 @@ static struct rte_pci_id pci_id_nfp_net_map[] = {
 };
 
 static struct eth_driver rte_nfp_net_pmd = {
-       {
-               .name = "rte_nfp_net_pmd",
+       .pci_drv = {
                .id_table = pci_id_nfp_net_map,
                .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
+               .probe = rte_eth_dev_pci_probe,
+               .remove = rte_eth_dev_pci_remove,
        },
        .eth_dev_init = nfp_net_init,
        .dev_private_size = sizeof(struct nfp_net_adapter),
 };
 
-static int
-nfp_net_pmd_init(const char *name __rte_unused,
-                const char *params __rte_unused)
-{
-       PMD_INIT_FUNC_TRACE();
-       PMD_INIT_LOG(INFO, "librte_pmd_nfp_net version %s\n",
-                    NFP_NET_PMD_VERSION);
-
-       rte_eth_driver_register(&rte_nfp_net_pmd);
-       return 0;
-}
-
-static struct rte_driver rte_nfp_net_driver = {
-       .type = PMD_PDEV,
-       .init = nfp_net_pmd_init,
-};
-
-PMD_REGISTER_DRIVER(rte_nfp_net_driver);
+RTE_PMD_REGISTER_PCI(net_nfp, rte_nfp_net_pmd.pci_drv);
+RTE_PMD_REGISTER_PCI_TABLE(net_nfp, pci_id_nfp_net_map);
+RTE_PMD_REGISTER_KMOD_DEP(net_nfp, "* igb_uio | uio_pci_generic | vfio");
 
 /*
  * Local variables: