net/mlx5: fix representor ID check for sampling
[dpdk.git] / drivers / net / nfp / nfp_net.c
index bce4d96..a30e78d 100644 (file)
@@ -17,8 +17,8 @@
 #include <rte_common.h>
 #include <rte_log.h>
 #include <rte_debug.h>
-#include <rte_ethdev_driver.h>
-#include <rte_ethdev_pci.h>
+#include <ethdev_driver.h>
+#include <ethdev_pci.h>
 #include <rte_dev.h>
 #include <rte_ether.h>
 #include <rte_malloc.h>
@@ -30,6 +30,8 @@
 #include <rte_spinlock.h>
 #include <rte_service_component.h>
 
+#include "eal_firmware.h"
+
 #include "nfpcore/nfp_cpp.h"
 #include "nfpcore/nfp_nffw.h"
 #include "nfpcore/nfp_hwinfo.h"
@@ -50,7 +52,7 @@
 #include <errno.h>
 
 /* Prototypes */
-static void nfp_net_close(struct rte_eth_dev *dev);
+static int 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(void *param);
 static void nfp_net_dev_interrupt_delayed_handler(void *param);
@@ -58,6 +60,10 @@ static int nfp_net_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
 static int 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_pf_init(struct rte_pci_device *pci_dev);
+static int nfp_pf_secondary_init(struct rte_pci_device *pci_dev);
+static int nfp_pci_uninit(struct rte_eth_dev *eth_dev);
+static int nfp_init_phyports(struct nfp_pf_dev *pf_dev);
 static int nfp_net_link_update(struct rte_eth_dev *dev, int wait_to_complete);
 static int nfp_net_promisc_enable(struct rte_eth_dev *dev);
 static int nfp_net_promisc_disable(struct rte_eth_dev *dev);
@@ -80,7 +86,7 @@ static int nfp_net_start(struct rte_eth_dev *dev);
 static int nfp_net_stats_get(struct rte_eth_dev *dev,
                              struct rte_eth_stats *stats);
 static int nfp_net_stats_reset(struct rte_eth_dev *dev);
-static void nfp_net_stop(struct rte_eth_dev *dev);
+static int 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);
 
@@ -94,6 +100,13 @@ static int nfp_net_rss_hash_write(struct rte_eth_dev *dev,
                        struct rte_eth_rss_conf *rss_conf);
 static int nfp_set_mac_addr(struct rte_eth_dev *dev,
                             struct rte_ether_addr *mac_addr);
+static int32_t nfp_cpp_bridge_service_func(void *args);
+static void nfp_register_cpp_service(struct nfp_cpp *cpp);
+static int nfp_fw_setup(struct rte_pci_device *dev,
+                       struct nfp_cpp *cpp,
+                       struct nfp_eth_table *nfp_eth_table,
+                       struct nfp_hwinfo *hwinfo);
+
 
 /* The offset of the queue controller queues in the PCIe Target */
 #define NFP_PCIE_QUEUE(_q) (0x80000 + (NFP_QCP_QUEUE_ADDR_SZ * ((_q) & 0xff)))
@@ -486,16 +499,16 @@ nfp_eth_copy_mac(uint8_t *dst, const uint8_t *src)
 }
 
 static int
-nfp_net_pf_read_mac(struct nfp_net_hw *hw, int port)
+nfp_net_pf_read_mac(struct nfp_pf_dev *pf_dev, int port)
 {
        struct nfp_eth_table *nfp_eth_table;
+       struct nfp_net_hw *hw = NULL;
+
+       /* Grab a pointer to the correct physical port */
+       hw = pf_dev->ports[port];
+
+       nfp_eth_table = nfp_eth_read_ports(pf_dev->cpp);
 
-       nfp_eth_table = nfp_eth_read_ports(hw->cpp);
-       /*
-        * hw points to port0 private data. We need hw now pointing to
-        * right port.
-        */
-       hw += port;
        nfp_eth_copy_mac((uint8_t *)&hw->mac_addr,
                         (uint8_t *)&nfp_eth_table->ports[port].mac_addr);
 
@@ -674,12 +687,14 @@ nfp_net_start(struct rte_eth_dev *dev)
        struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
        uint32_t new_ctrl, update = 0;
        struct nfp_net_hw *hw;
+       struct nfp_pf_dev *pf_dev;
        struct rte_eth_conf *dev_conf;
        struct rte_eth_rxmode *rxmode;
        uint32_t intr_vector;
        int ret;
 
        hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       pf_dev = NFP_NET_DEV_PRIVATE_TO_PF(dev->data->dev_private);
 
        PMD_INIT_LOG(DEBUG, "Start");
 
@@ -691,7 +706,7 @@ nfp_net_start(struct rte_eth_dev *dev)
 
        /* check and configure queue intr-vector mapping */
        if (dev->data->dev_conf.intr_conf.rxq != 0) {
-               if (hw->pf_multiport_enabled) {
+               if (pf_dev->multiport) {
                        PMD_INIT_LOG(ERR, "PMD rx interrupt is not supported "
                                          "with NFP multiport PF");
                                return -EINVAL;
@@ -755,13 +770,13 @@ nfp_net_start(struct rte_eth_dev *dev)
                goto error;
        }
 
-       if (hw->is_pf) {
+       if (hw->is_phyport) {
                if (rte_eal_process_type() == RTE_PROC_PRIMARY)
                        /* Configure the physical port up */
-                       nfp_eth_set_configured(hw->cpp, hw->pf_port_idx, 1);
+                       nfp_eth_set_configured(hw->cpp, hw->nfp_idx, 1);
                else
                        nfp_eth_set_configured(dev->process_private,
-                                              hw->pf_port_idx, 1);
+                                              hw->nfp_idx, 1);
        }
 
        hw->ctrl = new_ctrl;
@@ -788,7 +803,7 @@ error:
 }
 
 /* Stop device: disable rx and tx functions to allow for reconfiguring. */
-static void
+static int
 nfp_net_stop(struct rte_eth_dev *dev)
 {
        int i;
@@ -811,14 +826,16 @@ nfp_net_stop(struct rte_eth_dev *dev)
                        (struct nfp_net_rxq *)dev->data->rx_queues[i]);
        }
 
-       if (hw->is_pf) {
+       if (hw->is_phyport) {
                if (rte_eal_process_type() == RTE_PROC_PRIMARY)
                        /* Configure the physical port down */
-                       nfp_eth_set_configured(hw->cpp, hw->pf_port_idx, 0);
+                       nfp_eth_set_configured(hw->cpp, hw->nfp_idx, 0);
                else
                        nfp_eth_set_configured(dev->process_private,
-                                              hw->pf_port_idx, 0);
+                                              hw->nfp_idx, 0);
        }
+
+       return 0;
 }
 
 /* Set the link up. */
@@ -831,15 +848,15 @@ nfp_net_set_link_up(struct rte_eth_dev *dev)
 
        hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 
-       if (!hw->is_pf)
+       if (!hw->is_phyport)
                return -ENOTSUP;
 
        if (rte_eal_process_type() == RTE_PROC_PRIMARY)
                /* Configure the physical port down */
-               return nfp_eth_set_configured(hw->cpp, hw->pf_port_idx, 1);
+               return nfp_eth_set_configured(hw->cpp, hw->nfp_idx, 1);
        else
                return nfp_eth_set_configured(dev->process_private,
-                                             hw->pf_port_idx, 1);
+                                             hw->nfp_idx, 1);
 }
 
 /* Set the link down. */
@@ -852,25 +869,28 @@ nfp_net_set_link_down(struct rte_eth_dev *dev)
 
        hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 
-       if (!hw->is_pf)
+       if (!hw->is_phyport)
                return -ENOTSUP;
 
        if (rte_eal_process_type() == RTE_PROC_PRIMARY)
                /* Configure the physical port down */
-               return nfp_eth_set_configured(hw->cpp, hw->pf_port_idx, 0);
+               return nfp_eth_set_configured(hw->cpp, hw->nfp_idx, 0);
        else
                return nfp_eth_set_configured(dev->process_private,
-                                             hw->pf_port_idx, 0);
+                                             hw->nfp_idx, 0);
 }
 
 /* Reset and stop device. The device can not be restarted. */
-static void
+static int
 nfp_net_close(struct rte_eth_dev *dev)
 {
        struct nfp_net_hw *hw;
        struct rte_pci_device *pci_dev;
        int i;
 
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+               return 0;
+
        PMD_INIT_LOG(DEBUG, "Close");
 
        hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
@@ -894,8 +914,34 @@ nfp_net_close(struct rte_eth_dev *dev)
                        (struct nfp_net_rxq *)dev->data->rx_queues[i]);
        }
 
+       /* Only free PF resources after all physical ports have been closed */
+       if (pci_dev->id.device_id == PCI_DEVICE_ID_NFP4000_PF_NIC ||
+           pci_dev->id.device_id == PCI_DEVICE_ID_NFP6000_PF_NIC) {
+               struct nfp_pf_dev *pf_dev;
+               pf_dev = NFP_NET_DEV_PRIVATE_TO_PF(dev->data->dev_private);
+
+               /* Mark this port as unused and free device priv resources*/
+               nn_cfg_writeb(hw, NFP_NET_CFG_LSC, 0xff);
+               pf_dev->ports[hw->idx] = NULL;
+               rte_eth_dev_release_port(dev);
+
+               for (i = 0; i < pf_dev->total_phyports; i++) {
+                       /* Check to see if ports are still in use */
+                       if (pf_dev->ports[i])
+                               return 0;
+               }
+
+               /* Now it is safe to free all PF resources */
+               PMD_INIT_LOG(INFO, "Freeing PF resources");
+               nfp_cpp_area_free(pf_dev->ctrl_area);
+               nfp_cpp_area_free(pf_dev->hwqueues_area);
+               free(pf_dev->hwinfo);
+               free(pf_dev->sym_tbl);
+               nfp_cpp_free(pf_dev->cpp);
+               rte_free(pf_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,
@@ -906,6 +952,8 @@ nfp_net_close(struct rte_eth_dev *dev)
         * The ixgbe PMD driver disables the pcie master on the
         * device. The i40e does not...
         */
+
+       return 0;
 }
 
 static int
@@ -1213,9 +1261,6 @@ nfp_net_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
                                             DEV_RX_OFFLOAD_UDP_CKSUM |
                                             DEV_RX_OFFLOAD_TCP_CKSUM;
 
-       dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME |
-                                    DEV_RX_OFFLOAD_RSS_HASH;
-
        if (hw->cap & NFP_NET_CFG_CTRL_TXVLAN)
                dev_info->tx_offload_capa = DEV_TX_OFFLOAD_VLAN_INSERT;
 
@@ -1250,15 +1295,36 @@ nfp_net_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
                .tx_rs_thresh = DEFAULT_TX_RSBIT_THRESH,
        };
 
-       dev_info->flow_type_rss_offloads = ETH_RSS_IPV4 |
-                                          ETH_RSS_NONFRAG_IPV4_TCP |
-                                          ETH_RSS_NONFRAG_IPV4_UDP |
-                                          ETH_RSS_IPV6 |
-                                          ETH_RSS_NONFRAG_IPV6_TCP |
-                                          ETH_RSS_NONFRAG_IPV6_UDP;
+       dev_info->rx_desc_lim = (struct rte_eth_desc_lim) {
+               .nb_max = NFP_NET_MAX_RX_DESC,
+               .nb_min = NFP_NET_MIN_RX_DESC,
+               .nb_align = NFP_ALIGN_RING_DESC,
+       };
+
+       dev_info->tx_desc_lim = (struct rte_eth_desc_lim) {
+               .nb_max = NFP_NET_MAX_TX_DESC,
+               .nb_min = NFP_NET_MIN_TX_DESC,
+               .nb_align = NFP_ALIGN_RING_DESC,
+               .nb_seg_max = NFP_TX_MAX_SEG,
+               .nb_mtu_seg_max = NFP_TX_MAX_MTU_SEG,
+       };
+
+       /* All NFP devices support jumbo frames */
+       dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
+
+       if (hw->cap & NFP_NET_CFG_CTRL_RSS) {
+               dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_RSS_HASH;
 
-       dev_info->reta_size = NFP_NET_CFG_RSS_ITBL_SZ;
-       dev_info->hash_key_size = NFP_NET_CFG_RSS_KEY_SZ;
+               dev_info->flow_type_rss_offloads = ETH_RSS_IPV4 |
+                                                  ETH_RSS_NONFRAG_IPV4_TCP |
+                                                  ETH_RSS_NONFRAG_IPV4_UDP |
+                                                  ETH_RSS_IPV6 |
+                                                  ETH_RSS_NONFRAG_IPV6_TCP |
+                                                  ETH_RSS_NONFRAG_IPV6_UDP;
+
+               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_LINK_SPEED_1G | ETH_LINK_SPEED_10G |
                               ETH_LINK_SPEED_25G | ETH_LINK_SPEED_40G |
@@ -1460,7 +1526,7 @@ 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);
+       rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 
        nfp_net_dev_link_status_print(dev);
 
@@ -1487,7 +1553,7 @@ nfp_net_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
        }
 
        /* switch to jumbo mode if needed */
-       if ((uint32_t)mtu > RTE_ETHER_MAX_LEN)
+       if ((uint32_t)mtu > RTE_ETHER_MTU)
                dev->data->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
        else
                dev->data->dev_conf.rxmode.offloads &= ~DEV_RX_OFFLOAD_JUMBO_FRAME;
@@ -1513,15 +1579,17 @@ nfp_net_rx_queue_setup(struct rte_eth_dev *dev,
        const struct rte_memzone *tz;
        struct nfp_net_rxq *rxq;
        struct nfp_net_hw *hw;
+       uint32_t rx_desc_sz;
 
        hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 
        PMD_INIT_FUNC_TRACE();
 
        /* Validating number of descriptors */
-       if (((nb_desc * sizeof(struct nfp_net_rx_desc)) % 128) != 0 ||
-           (nb_desc > NFP_NET_MAX_RX_DESC) ||
-           (nb_desc < NFP_NET_MIN_RX_DESC)) {
+       rx_desc_sz = nb_desc * sizeof(struct nfp_net_rx_desc);
+       if (rx_desc_sz % NFP_ALIGN_RING_DESC != 0 ||
+           nb_desc > NFP_NET_MAX_RX_DESC ||
+           nb_desc < NFP_NET_MIN_RX_DESC) {
                PMD_DRV_LOG(ERR, "Wrong nb_desc value");
                return -EINVAL;
        }
@@ -1660,15 +1728,17 @@ nfp_net_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
        struct nfp_net_txq *txq;
        uint16_t tx_free_thresh;
        struct nfp_net_hw *hw;
+       uint32_t tx_desc_sz;
 
        hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 
        PMD_INIT_FUNC_TRACE();
 
        /* Validating number of descriptors */
-       if (((nb_desc * sizeof(struct nfp_net_tx_desc)) % 128) != 0 ||
-           (nb_desc > NFP_NET_MAX_TX_DESC) ||
-           (nb_desc < NFP_NET_MIN_TX_DESC)) {
+       tx_desc_sz = nb_desc * sizeof(struct nfp_net_tx_desc);
+       if (tx_desc_sz % NFP_ALIGN_RING_DESC != 0 ||
+           nb_desc > NFP_NET_MAX_TX_DESC ||
+           nb_desc < NFP_NET_MIN_TX_DESC) {
                PMD_DRV_LOG(ERR, "Wrong nb_desc value");
                return -EINVAL;
        }
@@ -2353,11 +2423,6 @@ nfp_net_vlan_offload_set(struct rte_eth_dev *dev, int mask)
        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))
-               PMD_DRV_LOG(INFO, "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))
@@ -2626,6 +2691,9 @@ nfp_net_rss_hash_conf_get(struct rte_eth_dev *dev,
        if (cfg_rss_ctrl & NFP_NET_CFG_RSS_IPV6)
                rss_hf |= ETH_RSS_NONFRAG_IPV4_UDP | ETH_RSS_NONFRAG_IPV6_UDP;
 
+       /* Propagate current RSS hash functions to caller */
+       rss_conf->rss_hf = rss_hf;
+
        /* Reading the key size */
        rss_conf->rss_key_len = nn_cfg_readl(hw, NFP_NET_CFG_RSS_KEY_SZ);
 
@@ -2703,50 +2771,19 @@ static const struct eth_dev_ops nfp_net_eth_dev_ops = {
        .rss_hash_conf_get      = nfp_net_rss_hash_conf_get,
        .rx_queue_setup         = nfp_net_rx_queue_setup,
        .rx_queue_release       = nfp_net_rx_queue_release,
-       .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,
 };
 
-/*
- * All eth_dev created got its private data, but before nfp_net_init, that
- * private data is referencing private data for all the PF ports. This is due
- * to how the vNIC bars are mapped based on first port, so all ports need info
- * about port 0 private data. Inside nfp_net_init the private data pointer is
- * changed to the right address for each port once the bars have been mapped.
- *
- * This functions helps to find out which port and therefore which offset
- * inside the private data array to use.
- */
-static int
-get_pf_port_number(char *name)
-{
-       char *pf_str = name;
-       int size = 0;
-
-       while ((*pf_str != '_') && (*pf_str != '\0') && (size++ < 30))
-               pf_str++;
-
-       if (size == 30)
-               /*
-                * This should not happen at all and it would mean major
-                * implementation fault.
-                */
-               rte_panic("nfp_net: problem with pf device name\n");
-
-       /* Expecting _portX with X within [0,7] */
-       pf_str += 5;
-
-       return (int)strtol(pf_str, NULL, 10);
-}
 
 static int
 nfp_net_init(struct rte_eth_dev *eth_dev)
 {
        struct rte_pci_device *pci_dev;
-       struct nfp_net_hw *hw, *hwport0;
+       struct nfp_pf_dev *pf_dev;
+       struct nfp_net_hw *hw;
 
        uint64_t tx_bar_off = 0, rx_bar_off = 0;
        uint32_t start_q;
@@ -2758,6 +2795,9 @@ nfp_net_init(struct rte_eth_dev *eth_dev)
 
        pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
 
+       /* Use backpointer here to the PF of this eth_dev */
+       pf_dev = NFP_NET_DEV_PRIVATE_TO_PF(eth_dev->data->dev_private);
+
        /* NFP can not handle DMA addresses requiring more than 40 bits */
        if (rte_mem_check_dma_mask(40)) {
                RTE_LOG(ERR, PMD, "device %s can not be used:",
@@ -2768,25 +2808,27 @@ nfp_net_init(struct rte_eth_dev *eth_dev)
 
        if ((pci_dev->id.device_id == PCI_DEVICE_ID_NFP4000_PF_NIC) ||
            (pci_dev->id.device_id == PCI_DEVICE_ID_NFP6000_PF_NIC)) {
-               port = get_pf_port_number(eth_dev->data->name);
+               port = ((struct nfp_net_hw *)eth_dev->data->dev_private)->idx;
                if (port < 0 || port > 7) {
                        PMD_DRV_LOG(ERR, "Port value is wrong");
                        return -ENODEV;
                }
 
-               PMD_INIT_LOG(DEBUG, "Working with PF port value %d", port);
+               /* Use PF array of physical ports to get pointer to
+                * this specific port
+                */
+               hw = pf_dev->ports[port];
 
-               /* This points to port 0 private data */
-               hwport0 = NFP_NET_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
+               PMD_INIT_LOG(DEBUG, "Working with physical port number: %d, "
+                                   "NFP internal port number: %d",
+                                   port, hw->nfp_idx);
 
-               /* This points to the specific port private data */
-               hw = &hwport0[port];
        } else {
                hw = NFP_NET_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
-               hwport0 = 0;
        }
 
        eth_dev->dev_ops = &nfp_net_eth_dev_ops;
+       eth_dev->rx_queue_count = nfp_net_rx_queue_count;
        eth_dev->rx_pkt_burst = &nfp_net_recv_pkts;
        eth_dev->tx_pkt_burst = &nfp_net_xmit_pkts;
 
@@ -2813,25 +2855,18 @@ nfp_net_init(struct rte_eth_dev *eth_dev)
                return -ENODEV;
        }
 
-       if (hw->is_pf && port == 0) {
-               hw->ctrl_bar = nfp_rtsym_map(hw->sym_tbl, "_pf0_net_bar0",
-                                            hw->total_ports * 32768,
-                                            &hw->ctrl_area);
-               if (!hw->ctrl_bar) {
-                       printf("nfp_rtsym_map fails for _pf0_net_ctrl_bar");
-                       return -EIO;
+       if (hw->is_phyport) {
+               if (port == 0) {
+                       hw->ctrl_bar = pf_dev->ctrl_bar;
+               } else {
+                       if (!pf_dev->ctrl_bar)
+                               return -ENODEV;
+                       /* Use port offset in pf ctrl_bar for this
+                        * ports control bar
+                        */
+                       hw->ctrl_bar = pf_dev->ctrl_bar +
+                                      (port * NFP_PF_CSR_SLICE_SIZE);
                }
-
-               PMD_INIT_LOG(DEBUG, "ctrl bar: %p", hw->ctrl_bar);
-       }
-
-       if (port > 0) {
-               if (!hwport0->ctrl_bar)
-                       return -ENODEV;
-
-               /* address based on port0 offset */
-               hw->ctrl_bar = hwport0->ctrl_bar +
-                              (port * NFP_PF_CSR_SLICE_SIZE);
        }
 
        PMD_INIT_LOG(DEBUG, "ctrl bar: %p", hw->ctrl_bar);
@@ -2858,26 +2893,9 @@ nfp_net_init(struct rte_eth_dev *eth_dev)
        PMD_INIT_LOG(DEBUG, "tx_bar_off: 0x%" PRIx64 "", tx_bar_off);
        PMD_INIT_LOG(DEBUG, "rx_bar_off: 0x%" PRIx64 "", rx_bar_off);
 
-       if (hw->is_pf && port == 0) {
-               /* configure access to tx/rx vNIC BARs */
-               hwport0->hw_queues = nfp_cpp_map_area(hw->cpp, 0, 0,
-                                                     NFP_PCIE_QUEUE(0),
-                                                     NFP_QCP_QUEUE_AREA_SZ,
-                                                     &hw->hwqueues_area);
-
-               if (!hwport0->hw_queues) {
-                       printf("nfp_rtsym_map fails for net.qc");
-                       err = -EIO;
-                       goto dev_err_ctrl_map;
-               }
-
-               PMD_INIT_LOG(DEBUG, "tx/rx bar address: 0x%p",
-                                   hwport0->hw_queues);
-       }
-
-       if (hw->is_pf) {
-               hw->tx_bar = hwport0->hw_queues + tx_bar_off;
-               hw->rx_bar = hwport0->hw_queues + rx_bar_off;
+       if (hw->is_phyport) {
+               hw->tx_bar = pf_dev->hw_queues + tx_bar_off;
+               hw->rx_bar = pf_dev->hw_queues + rx_bar_off;
                eth_dev->data->dev_private = hw;
        } else {
                hw->tx_bar = (uint8_t *)pci_dev->mem_resource[2].addr +
@@ -2946,8 +2964,8 @@ nfp_net_init(struct rte_eth_dev *eth_dev)
                goto dev_err_queues_map;
        }
 
-       if (hw->is_pf) {
-               nfp_net_pf_read_mac(hwport0, port);
+       if (hw->is_phyport) {
+               nfp_net_pf_read_mac(pf_dev, port);
                nfp_net_write_mac(hw, (uint8_t *)&hw->mac_addr);
        } else {
                nfp_net_vf_read_mac(hw);
@@ -2969,6 +2987,8 @@ nfp_net_init(struct rte_eth_dev *eth_dev)
        if (!(hw->cap & NFP_NET_CFG_CTRL_LIVE_ADDR))
                eth_dev->data->dev_flags |= RTE_ETH_DEV_NOLIVE_MAC_ADDR;
 
+       eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
+
        PMD_INIT_LOG(INFO, "port %d VendorID=0x%x DeviceID=0x%x "
                     "mac=%02x:%02x:%02x:%02x:%02x:%02x",
                     eth_dev->data->port_id, pci_dev->id.vendor_id,
@@ -3344,133 +3364,16 @@ nfp_cpp_bridge_service_func(void *args)
        return 0;
 }
 
-static int
-nfp_pf_create_dev(struct rte_pci_device *dev, int port, int ports,
-                 struct nfp_cpp *cpp, struct nfp_hwinfo *hwinfo,
-                 int phys_port, struct nfp_rtsym_table *sym_tbl, void **priv)
-{
-       struct rte_eth_dev *eth_dev;
-       struct nfp_net_hw *hw = NULL;
-       char *port_name;
-       struct rte_service_spec service;
-       int retval;
-
-       port_name = rte_zmalloc("nfp_pf_port_name", 100, 0);
-       if (!port_name)
-               return -ENOMEM;
-
-       if (ports > 1)
-               snprintf(port_name, 100, "%s_port%d", dev->device.name, port);
-       else
-               strlcat(port_name, dev->device.name, 100);
-
-
-       if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
-               eth_dev = rte_eth_dev_allocate(port_name);
-               if (!eth_dev) {
-                       rte_free(port_name);
-                       return -ENODEV;
-               }
-               if (port == 0) {
-                       *priv = rte_zmalloc(port_name,
-                                           sizeof(struct nfp_net_adapter) *
-                                           ports, RTE_CACHE_LINE_SIZE);
-                       if (!*priv) {
-                               rte_free(port_name);
-                               rte_eth_dev_release_port(eth_dev);
-                               return -ENOMEM;
-                       }
-               }
-               eth_dev->data->dev_private = *priv;
-
-               /*
-                * dev_private pointing to port0 dev_private because we need
-                * to configure vNIC bars based on port0 at nfp_net_init.
-                * Then dev_private is adjusted per port.
-                */
-               hw = (struct nfp_net_hw *)(eth_dev->data->dev_private) + port;
-               hw->cpp = cpp;
-               hw->hwinfo = hwinfo;
-               hw->sym_tbl = sym_tbl;
-               hw->pf_port_idx = phys_port;
-               hw->is_pf = 1;
-               if (ports > 1)
-                       hw->pf_multiport_enabled = 1;
-
-               hw->total_ports = ports;
-       } else {
-               eth_dev = rte_eth_dev_attach_secondary(port_name);
-               if (!eth_dev) {
-                       RTE_LOG(ERR, EAL, "secondary process attach failed, "
-                               "ethdev doesn't exist");
-                       rte_free(port_name);
-                       return -ENODEV;
-               }
-               eth_dev->process_private = cpp;
-       }
-
-       eth_dev->device = &dev->device;
-       rte_eth_copy_pci_info(eth_dev, dev);
-
-       retval = nfp_net_init(eth_dev);
-
-       if (retval) {
-               retval = -ENODEV;
-               goto probe_failed;
-       } else {
-               rte_eth_dev_probing_finish(eth_dev);
-       }
-
-       rte_free(port_name);
-
-       if (port == 0) {
-               /*
-                * The rte_service needs to be created just once per PMD.
-                * And the cpp handler needs to be linked to the service.
-                * Secondary processes will be used for debugging DPDK apps
-                * when requiring to use the CPP interface for accessing NFP
-                * components. And the cpp handler for secondary processes is
-                * available at this point.
-                */
-               memset(&service, 0, sizeof(struct rte_service_spec));
-               snprintf(service.name, sizeof(service.name), "nfp_cpp_service");
-               service.callback = nfp_cpp_bridge_service_func;
-               service.callback_userdata = (void *)cpp;
-
-               hw = (struct nfp_net_hw *)(eth_dev->data->dev_private);
-
-               if (rte_service_component_register(&service,
-                                                  &hw->nfp_cpp_service_id))
-                       RTE_LOG(ERR, PMD, "NFP CPP bridge service register() failed");
-               else
-                       RTE_LOG(DEBUG, PMD, "NFP CPP bridge service registered");
-       }
-
-       return retval;
-
-probe_failed:
-       rte_free(port_name);
-       /* free ports private data if primary process */
-       if (rte_eal_process_type() == RTE_PROC_PRIMARY)
-               rte_free(eth_dev->data->dev_private);
-
-       rte_eth_dev_release_port(eth_dev);
-
-       return retval;
-}
-
 #define DEFAULT_FW_PATH       "/lib/firmware/netronome"
 
 static int
 nfp_fw_upload(struct rte_pci_device *dev, struct nfp_nsp *nsp, char *card)
 {
        struct nfp_cpp *cpp = nsp->cpp;
-       int fw_f;
-       char *fw_buf;
+       void *fw_buf;
        char fw_name[125];
        char serial[40];
-       struct stat file_stat;
-       off_t fsize, bytes;
+       size_t fsize;
 
        /* Looking for firmware file in order of priority */
 
@@ -3483,66 +3386,34 @@ nfp_fw_upload(struct rte_pci_device *dev, struct nfp_nsp *nsp, char *card)
 
        snprintf(fw_name, sizeof(fw_name), "%s/%s.nffw", DEFAULT_FW_PATH,
                        serial);
-
        PMD_DRV_LOG(DEBUG, "Trying with fw file: %s", fw_name);
-       fw_f = open(fw_name, O_RDONLY);
-       if (fw_f >= 0)
-               goto read_fw;
+       if (rte_firmware_read(fw_name, &fw_buf, &fsize) == 0)
+               goto load_fw;
 
        /* Then try the PCI name */
        snprintf(fw_name, sizeof(fw_name), "%s/pci-%s.nffw", DEFAULT_FW_PATH,
                        dev->device.name);
-
        PMD_DRV_LOG(DEBUG, "Trying with fw file: %s", fw_name);
-       fw_f = open(fw_name, O_RDONLY);
-       if (fw_f >= 0)
-               goto read_fw;
+       if (rte_firmware_read(fw_name, &fw_buf, &fsize) == 0)
+               goto load_fw;
 
        /* Finally try the card type and media */
        snprintf(fw_name, sizeof(fw_name), "%s/%s", DEFAULT_FW_PATH, card);
        PMD_DRV_LOG(DEBUG, "Trying with fw file: %s", fw_name);
-       fw_f = open(fw_name, O_RDONLY);
-       if (fw_f < 0) {
+       if (rte_firmware_read(fw_name, &fw_buf, &fsize) < 0) {
                PMD_DRV_LOG(INFO, "Firmware file %s not found.", fw_name);
                return -ENOENT;
        }
 
-read_fw:
-       if (fstat(fw_f, &file_stat) < 0) {
-               PMD_DRV_LOG(INFO, "Firmware file %s size is unknown", fw_name);
-               close(fw_f);
-               return -ENOENT;
-       }
-
-       fsize = file_stat.st_size;
-       PMD_DRV_LOG(INFO, "Firmware file found at %s with size: %" PRIu64 "",
-                           fw_name, (uint64_t)fsize);
-
-       fw_buf = malloc((size_t)fsize);
-       if (!fw_buf) {
-               PMD_DRV_LOG(INFO, "malloc failed for fw buffer");
-               close(fw_f);
-               return -ENOMEM;
-       }
-       memset(fw_buf, 0, fsize);
-
-       bytes = read(fw_f, fw_buf, fsize);
-       if (bytes != fsize) {
-               PMD_DRV_LOG(INFO, "Reading fw to buffer failed."
-                                  "Just %" PRIu64 " of %" PRIu64 " bytes read",
-                                  (uint64_t)bytes, (uint64_t)fsize);
-               free(fw_buf);
-               close(fw_f);
-               return -EIO;
-       }
+load_fw:
+       PMD_DRV_LOG(INFO, "Firmware file found at %s with size: %zu",
+               fw_name, fsize);
 
        PMD_DRV_LOG(INFO, "Uploading the firmware ...");
-       nfp_nsp_load_fw(nsp, fw_buf, bytes);
+       nfp_nsp_load_fw(nsp, fw_buf, fsize);
        PMD_DRV_LOG(INFO, "Done");
 
        free(fw_buf);
-       close(fw_f);
-
        return 0;
 }
 
@@ -3592,20 +3463,121 @@ nfp_fw_setup(struct rte_pci_device *dev, struct nfp_cpp *cpp,
        return err;
 }
 
-static int nfp_pf_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
-                           struct rte_pci_device *dev)
+static int nfp_init_phyports(struct nfp_pf_dev *pf_dev)
 {
+       struct nfp_net_hw *hw;
+       struct rte_eth_dev *eth_dev;
+       struct nfp_eth_table *nfp_eth_table = NULL;
+       int ret = 0;
+       int i;
+
+       nfp_eth_table = nfp_eth_read_ports(pf_dev->cpp);
+       if (!nfp_eth_table) {
+               PMD_INIT_LOG(ERR, "Error reading NFP ethernet table");
+               ret = -EIO;
+               goto error;
+       }
+
+       /* Loop through all physical ports on PF */
+       for (i = 0; i < pf_dev->total_phyports; i++) {
+               const unsigned int numa_node = rte_socket_id();
+               char port_name[RTE_ETH_NAME_MAX_LEN];
+
+               snprintf(port_name, sizeof(port_name), "%s_port%d",
+                        pf_dev->pci_dev->device.name, i);
+
+               /* Allocate a eth_dev for this phyport */
+               eth_dev = rte_eth_dev_allocate(port_name);
+               if (!eth_dev) {
+                       ret = -ENODEV;
+                       goto port_cleanup;
+               }
+
+               /* Allocate memory for this phyport */
+               eth_dev->data->dev_private =
+                       rte_zmalloc_socket(port_name, sizeof(struct nfp_net_hw),
+                                          RTE_CACHE_LINE_SIZE, numa_node);
+               if (!eth_dev->data->dev_private) {
+                       ret = -ENOMEM;
+                       rte_eth_dev_release_port(eth_dev);
+                       goto port_cleanup;
+               }
+
+               hw = NFP_NET_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
+
+               /* Add this device to the PF's array of physical ports */
+               pf_dev->ports[i] = hw;
+
+               hw->pf_dev = pf_dev;
+               hw->cpp = pf_dev->cpp;
+               hw->eth_dev = eth_dev;
+               hw->idx = i;
+               hw->nfp_idx = nfp_eth_table->ports[i].index;
+               hw->is_phyport = true;
+
+               eth_dev->device = &pf_dev->pci_dev->device;
+
+               /* ctrl/tx/rx BAR mappings and remaining init happens in
+                * nfp_net_init
+                */
+               ret = nfp_net_init(eth_dev);
+
+               if (ret) {
+                       ret = -ENODEV;
+                       goto port_cleanup;
+               }
+
+               rte_eth_dev_probing_finish(eth_dev);
+
+       } /* End loop, all ports on this PF */
+       ret = 0;
+       goto eth_table_cleanup;
+
+port_cleanup:
+       for (i = 0; i < pf_dev->total_phyports; i++) {
+               if (pf_dev->ports[i] && pf_dev->ports[i]->eth_dev) {
+                       struct rte_eth_dev *tmp_dev;
+                       tmp_dev = pf_dev->ports[i]->eth_dev;
+                       rte_eth_dev_release_port(tmp_dev);
+                       pf_dev->ports[i] = NULL;
+               }
+       }
+eth_table_cleanup:
+       free(nfp_eth_table);
+error:
+       return ret;
+}
+
+static void nfp_register_cpp_service(struct nfp_cpp *cpp)
+{
+       uint32_t *cpp_service_id = NULL;
+       struct rte_service_spec service;
+
+       memset(&service, 0, sizeof(struct rte_service_spec));
+       snprintf(service.name, sizeof(service.name), "nfp_cpp_service");
+       service.callback = nfp_cpp_bridge_service_func;
+       service.callback_userdata = (void *)cpp;
+
+       if (rte_service_component_register(&service,
+                                          cpp_service_id))
+               RTE_LOG(WARNING, PMD, "NFP CPP bridge service register() failed");
+       else
+               RTE_LOG(DEBUG, PMD, "NFP CPP bridge service registered");
+}
+
+static int nfp_pf_init(struct rte_pci_device *pci_dev)
+{
+       struct nfp_pf_dev *pf_dev = NULL;
        struct nfp_cpp *cpp;
        struct nfp_hwinfo *hwinfo;
        struct nfp_rtsym_table *sym_tbl;
        struct nfp_eth_table *nfp_eth_table = NULL;
+       char name[RTE_ETH_NAME_MAX_LEN];
        int total_ports;
-       void *priv = 0;
        int ret = -ENODEV;
        int err;
-       int i;
 
-       if (!dev)
+       if (!pci_dev)
                return ret;
 
        /*
@@ -3615,75 +3587,218 @@ static int nfp_pf_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
         * interface. Here we avoid this telling to the CPP init code to
         * use a lock file if UIO is being used.
         */
-       if (dev->kdrv == RTE_KDRV_VFIO)
-               cpp = nfp_cpp_from_device_name(dev, 0);
+       if (pci_dev->kdrv == RTE_PCI_KDRV_VFIO)
+               cpp = nfp_cpp_from_device_name(pci_dev, 0);
        else
-               cpp = nfp_cpp_from_device_name(dev, 1);
+               cpp = nfp_cpp_from_device_name(pci_dev, 1);
 
        if (!cpp) {
-               PMD_DRV_LOG(ERR, "A CPP handle can not be obtained");
+               PMD_INIT_LOG(ERR, "A CPP handle can not be obtained");
                ret = -EIO;
                goto error;
        }
 
        hwinfo = nfp_hwinfo_read(cpp);
        if (!hwinfo) {
-               PMD_DRV_LOG(ERR, "Error reading hwinfo table");
-               return -EIO;
+               PMD_INIT_LOG(ERR, "Error reading hwinfo table");
+               ret = -EIO;
+               goto error;
        }
 
        nfp_eth_table = nfp_eth_read_ports(cpp);
        if (!nfp_eth_table) {
-               PMD_DRV_LOG(ERR, "Error reading NFP ethernet table");
-               return -EIO;
+               PMD_INIT_LOG(ERR, "Error reading NFP ethernet table");
+               ret = -EIO;
+               goto hwinfo_cleanup;
        }
 
-       if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
-               if (nfp_fw_setup(dev, cpp, nfp_eth_table, hwinfo)) {
-                       PMD_DRV_LOG(INFO, "Error when uploading firmware");
-                       ret = -EIO;
-                       goto error;
-               }
+       if (nfp_fw_setup(pci_dev, cpp, nfp_eth_table, hwinfo)) {
+               PMD_INIT_LOG(ERR, "Error when uploading firmware");
+               ret = -EIO;
+               goto eth_table_cleanup;
        }
 
        /* Now the symbol table should be there */
        sym_tbl = nfp_rtsym_table_read(cpp);
        if (!sym_tbl) {
-               PMD_DRV_LOG(ERR, "Something is wrong with the firmware"
+               PMD_INIT_LOG(ERR, "Something is wrong with the firmware"
                                " symbol table");
                ret = -EIO;
-               goto error;
+               goto eth_table_cleanup;
        }
 
        total_ports = nfp_rtsym_read_le(sym_tbl, "nfd_cfg_pf0_num_ports", &err);
        if (total_ports != (int)nfp_eth_table->count) {
                PMD_DRV_LOG(ERR, "Inconsistent number of ports");
                ret = -EIO;
-               goto error;
+               goto sym_tbl_cleanup;
        }
-       PMD_INIT_LOG(INFO, "Total pf ports: %d", total_ports);
+
+       PMD_INIT_LOG(INFO, "Total physical ports: %d", total_ports);
 
        if (total_ports <= 0 || total_ports > 8) {
-               PMD_DRV_LOG(ERR, "nfd_cfg_pf0_num_ports symbol with wrong value");
+               PMD_INIT_LOG(ERR, "nfd_cfg_pf0_num_ports symbol with wrong value");
                ret = -ENODEV;
-               goto error;
+               goto sym_tbl_cleanup;
+       }
+       /* Allocate memory for the PF "device" */
+       snprintf(name, sizeof(name), "nfp_pf%d", 0);
+       pf_dev = rte_zmalloc(name, sizeof(*pf_dev), 0);
+       if (!pf_dev) {
+               ret = -ENOMEM;
+               goto sym_tbl_cleanup;
        }
 
-       for (i = 0; i < total_ports; i++) {
-               ret = nfp_pf_create_dev(dev, i, total_ports, cpp, hwinfo,
-                                       nfp_eth_table->ports[i].index,
-                                       sym_tbl, &priv);
-               if (ret)
-                       break;
+       /* Populate the newly created PF device */
+       pf_dev->cpp = cpp;
+       pf_dev->hwinfo = hwinfo;
+       pf_dev->sym_tbl = sym_tbl;
+       pf_dev->total_phyports = total_ports;
+
+       if (total_ports > 1)
+               pf_dev->multiport = true;
+
+       pf_dev->pci_dev = pci_dev;
+
+       /* Map the symbol table */
+       pf_dev->ctrl_bar = nfp_rtsym_map(pf_dev->sym_tbl, "_pf0_net_bar0",
+                                    pf_dev->total_phyports * 32768,
+                                    &pf_dev->ctrl_area);
+       if (!pf_dev->ctrl_bar) {
+               PMD_INIT_LOG(ERR, "nfp_rtsym_map fails for _pf0_net_ctrl_bar");
+               ret = -EIO;
+               goto pf_cleanup;
        }
 
-error:
+       PMD_INIT_LOG(DEBUG, "ctrl bar: %p", pf_dev->ctrl_bar);
+
+       /* configure access to tx/rx vNIC BARs */
+       pf_dev->hw_queues = nfp_cpp_map_area(pf_dev->cpp, 0, 0,
+                                             NFP_PCIE_QUEUE(0),
+                                             NFP_QCP_QUEUE_AREA_SZ,
+                                             &pf_dev->hwqueues_area);
+       if (!pf_dev->hw_queues) {
+               PMD_INIT_LOG(ERR, "nfp_rtsym_map fails for net.qc");
+               ret = -EIO;
+               goto ctrl_area_cleanup;
+       }
+
+       PMD_INIT_LOG(DEBUG, "tx/rx bar address: 0x%p", pf_dev->hw_queues);
+
+       /* Initialize and prep physical ports now
+        * This will loop through all physical ports
+        */
+       ret = nfp_init_phyports(pf_dev);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Could not create physical ports");
+               goto hwqueues_cleanup;
+       }
+
+       /* register the CPP bridge service here for primary use */
+       nfp_register_cpp_service(pf_dev->cpp);
+
+       return 0;
+
+hwqueues_cleanup:
+       nfp_cpp_area_free(pf_dev->hwqueues_area);
+ctrl_area_cleanup:
+       nfp_cpp_area_free(pf_dev->ctrl_area);
+pf_cleanup:
+       rte_free(pf_dev);
+sym_tbl_cleanup:
+       free(sym_tbl);
+eth_table_cleanup:
        free(nfp_eth_table);
+hwinfo_cleanup:
+       free(hwinfo);
+error:
        return ret;
 }
 
-int nfp_logtype_init;
-int nfp_logtype_driver;
+/*
+ * When attaching to the NFP4000/6000 PF on a secondary process there
+ * is no need to initialize the PF again. Only minimal work is required
+ * here
+ */
+static int nfp_pf_secondary_init(struct rte_pci_device *pci_dev)
+{
+       struct nfp_cpp *cpp;
+       struct nfp_rtsym_table *sym_tbl;
+       int total_ports;
+       int i;
+       int err;
+
+       if (!pci_dev)
+               return -ENODEV;
+
+       /*
+        * When device bound to UIO, the device could be used, by mistake,
+        * by two DPDK apps, and the UIO driver does not avoid it. This
+        * could lead to a serious problem when configuring the NFP CPP
+        * interface. Here we avoid this telling to the CPP init code to
+        * use a lock file if UIO is being used.
+        */
+       if (pci_dev->kdrv == RTE_PCI_KDRV_VFIO)
+               cpp = nfp_cpp_from_device_name(pci_dev, 0);
+       else
+               cpp = nfp_cpp_from_device_name(pci_dev, 1);
+
+       if (!cpp) {
+               PMD_INIT_LOG(ERR, "A CPP handle can not be obtained");
+               return -EIO;
+       }
+
+       /*
+        * We don't have access to the PF created in the primary process
+        * here so we have to read the number of ports from firmware
+        */
+       sym_tbl = nfp_rtsym_table_read(cpp);
+       if (!sym_tbl) {
+               PMD_INIT_LOG(ERR, "Something is wrong with the firmware"
+                               " symbol table");
+               return -EIO;
+       }
+
+       total_ports = nfp_rtsym_read_le(sym_tbl, "nfd_cfg_pf0_num_ports", &err);
+
+       for (i = 0; i < total_ports; i++) {
+               struct rte_eth_dev *eth_dev;
+               char port_name[RTE_ETH_NAME_MAX_LEN];
+
+               snprintf(port_name, sizeof(port_name), "%s_port%d",
+                        pci_dev->device.name, i);
+
+               PMD_DRV_LOG(DEBUG, "Secondary attaching to port %s",
+                   port_name);
+               eth_dev = rte_eth_dev_attach_secondary(port_name);
+               if (!eth_dev) {
+                       RTE_LOG(ERR, EAL,
+                       "secondary process attach failed, "
+                       "ethdev doesn't exist");
+                       return -ENODEV;
+               }
+               eth_dev->process_private = cpp;
+               eth_dev->dev_ops = &nfp_net_eth_dev_ops;
+               eth_dev->rx_queue_count = nfp_net_rx_queue_count;
+               eth_dev->rx_pkt_burst = &nfp_net_recv_pkts;
+               eth_dev->tx_pkt_burst = &nfp_net_xmit_pkts;
+               rte_eth_dev_probing_finish(eth_dev);
+       }
+
+       /* Register the CPP bridge service for the secondary too */
+       nfp_register_cpp_service(cpp);
+
+       return 0;
+}
+
+static int nfp_pf_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
+                           struct rte_pci_device *dev)
+{
+       if (rte_eal_process_type() == RTE_PROC_PRIMARY)
+               return nfp_pf_init(dev);
+       else
+               return nfp_pf_secondary_init(dev);
+}
 
 static const struct rte_pci_id pci_id_nfp_pf_net_map[] = {
        {
@@ -3709,6 +3824,29 @@ static const struct rte_pci_id pci_id_nfp_vf_net_map[] = {
        },
 };
 
+static int nfp_pci_uninit(struct rte_eth_dev *eth_dev)
+{
+       struct rte_pci_device *pci_dev;
+       uint16_t port_id;
+
+       pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
+
+       if (pci_dev->id.device_id == PCI_DEVICE_ID_NFP4000_PF_NIC ||
+           pci_dev->id.device_id == PCI_DEVICE_ID_NFP6000_PF_NIC) {
+               /* Free up all physical ports under PF */
+               RTE_ETH_FOREACH_DEV_OF(port_id, &pci_dev->device)
+                       rte_eth_dev_close(port_id);
+               /*
+                * Ports can be closed and freed but hotplugging is not
+                * currently supported
+                */
+               return -ENOTSUP;
+       }
+
+       /* VF cleanup, just free private port data */
+       return nfp_net_close(eth_dev);
+}
+
 static int eth_nfp_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
        struct rte_pci_device *pci_dev)
 {
@@ -3718,34 +3856,7 @@ static int eth_nfp_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
 
 static int eth_nfp_pci_remove(struct rte_pci_device *pci_dev)
 {
-       struct rte_eth_dev *eth_dev;
-       struct nfp_net_hw *hw, *hwport0;
-       int port = 0;
-
-       eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
-       if ((pci_dev->id.device_id == PCI_DEVICE_ID_NFP4000_PF_NIC) ||
-           (pci_dev->id.device_id == PCI_DEVICE_ID_NFP6000_PF_NIC)) {
-               port = get_pf_port_number(eth_dev->data->name);
-               /*
-                * hotplug is not possible with multiport PF although freeing
-                * data structures can be done for first port.
-                */
-               if (port != 0)
-                       return -ENOTSUP;
-               hwport0 = NFP_NET_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
-               hw = &hwport0[port];
-               nfp_cpp_area_free(hw->ctrl_area);
-               nfp_cpp_area_free(hw->hwqueues_area);
-               free(hw->hwinfo);
-               free(hw->sym_tbl);
-               nfp_cpp_free(hw->cpp);
-       } else {
-               hw = NFP_NET_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
-       }
-       /* hotplug is not possible with multiport PF */
-       if (hw->pf_multiport_enabled)
-               return -ENOTSUP;
-       return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
+       return rte_eth_dev_pci_generic_remove(pci_dev, nfp_pci_uninit);
 }
 
 static struct rte_pci_driver rte_nfp_net_pf_pmd = {
@@ -3768,16 +3879,8 @@ RTE_PMD_REGISTER_PCI_TABLE(net_nfp_pf, pci_id_nfp_pf_net_map);
 RTE_PMD_REGISTER_PCI_TABLE(net_nfp_vf, pci_id_nfp_vf_net_map);
 RTE_PMD_REGISTER_KMOD_DEP(net_nfp_pf, "* igb_uio | uio_pci_generic | vfio");
 RTE_PMD_REGISTER_KMOD_DEP(net_nfp_vf, "* igb_uio | uio_pci_generic | vfio");
-
-RTE_INIT(nfp_init_log)
-{
-       nfp_logtype_init = rte_log_register("pmd.net.nfp.init");
-       if (nfp_logtype_init >= 0)
-               rte_log_set_level(nfp_logtype_init, RTE_LOG_NOTICE);
-       nfp_logtype_driver = rte_log_register("pmd.net.nfp.driver");
-       if (nfp_logtype_driver >= 0)
-               rte_log_set_level(nfp_logtype_driver, RTE_LOG_NOTICE);
-}
+RTE_LOG_REGISTER_SUFFIX(nfp_logtype_init, init, NOTICE);
+RTE_LOG_REGISTER_SUFFIX(nfp_logtype_driver, driver, NOTICE);
 /*
  * Local variables:
  * c-file-style: "Linux"