net/mlx5: refactor PCI probing on Linux
authorOphir Munk <ophirmu@mellanox.com>
Wed, 3 Jun 2020 15:06:00 +0000 (15:06 +0000)
committerFerruh Yigit <ferruh.yigit@intel.com>
Tue, 16 Jun 2020 17:21:07 +0000 (19:21 +0200)
Refactor PCI probing related code. Move Linux specific functions (as
well as verbs and dv related code) from mlx5.c file to linux/mlx5_os.c
file.

Signed-off-by: Ophir Munk <ophirmu@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
drivers/net/mlx5/linux/mlx5_os.c
drivers/net/mlx5/mlx5.c
drivers/net/mlx5/mlx5.h

index 7d60683..2ecb1a5 100644 (file)
@@ -39,6 +39,7 @@
 #include <mlx5_glue.h>
 #include <mlx5_devx_cmds.h>
 #include <mlx5_common.h>
+#include <mlx5_common_mp.h>
 
 #include "mlx5_defs.h"
 #include "mlx5.h"
 #include "mlx5_flow.h"
 #include "rte_pmd_mlx5.h"
 
+#define MLX5_TAGS_HLIST_ARRAY_SIZE 8192
+
+#ifndef HAVE_IBV_MLX5_MOD_MPW
+#define MLX5DV_CONTEXT_FLAGS_MPW_ALLOWED (1 << 2)
+#define MLX5DV_CONTEXT_FLAGS_ENHANCED_MPW (1 << 3)
+#endif
+
+#ifndef HAVE_IBV_MLX5_MOD_CQE_128B_COMP
+#define MLX5DV_CONTEXT_FLAGS_CQE_128B_COMP (1 << 4)
+#endif
+
 /**
  * Get ibv device name. Given an ibv_context pointer - return a
  * pointer to the corresponding device name.
@@ -166,3 +178,1785 @@ mlx5_os_get_dev_attr(void *ctx, struct mlx5_dev_attr *device_attr)
 
        return err;
 }
+
+/**
+ * Verbs callback to allocate a memory. This function should allocate the space
+ * according to the size provided residing inside a huge page.
+ * Please note that all allocation must respect the alignment from libmlx5
+ * (i.e. currently sysconf(_SC_PAGESIZE)).
+ *
+ * @param[in] size
+ *   The size in bytes of the memory to allocate.
+ * @param[in] data
+ *   A pointer to the callback data.
+ *
+ * @return
+ *   Allocated buffer, NULL otherwise and rte_errno is set.
+ */
+static void *
+mlx5_alloc_verbs_buf(size_t size, void *data)
+{
+       struct mlx5_priv *priv = data;
+       void *ret;
+       size_t alignment = sysconf(_SC_PAGESIZE);
+       unsigned int socket = SOCKET_ID_ANY;
+
+       if (priv->verbs_alloc_ctx.type == MLX5_VERBS_ALLOC_TYPE_TX_QUEUE) {
+               const struct mlx5_txq_ctrl *ctrl = priv->verbs_alloc_ctx.obj;
+
+               socket = ctrl->socket;
+       } else if (priv->verbs_alloc_ctx.type ==
+                  MLX5_VERBS_ALLOC_TYPE_RX_QUEUE) {
+               const struct mlx5_rxq_ctrl *ctrl = priv->verbs_alloc_ctx.obj;
+
+               socket = ctrl->socket;
+       }
+       MLX5_ASSERT(data != NULL);
+       ret = rte_malloc_socket(__func__, size, alignment, socket);
+       if (!ret && size)
+               rte_errno = ENOMEM;
+       return ret;
+}
+
+/**
+ * Verbs callback to free a memory.
+ *
+ * @param[in] ptr
+ *   A pointer to the memory to free.
+ * @param[in] data
+ *   A pointer to the callback data.
+ */
+static void
+mlx5_free_verbs_buf(void *ptr, void *data __rte_unused)
+{
+       MLX5_ASSERT(data != NULL);
+       rte_free(ptr);
+}
+
+/**
+ * Initialize DR related data within private structure.
+ * Routine checks the reference counter and does actual
+ * resources creation/initialization only if counter is zero.
+ *
+ * @param[in] priv
+ *   Pointer to the private device data structure.
+ *
+ * @return
+ *   Zero on success, positive error code otherwise.
+ */
+static int
+mlx5_alloc_shared_dr(struct mlx5_priv *priv)
+{
+       struct mlx5_dev_ctx_shared *sh = priv->sh;
+       char s[MLX5_HLIST_NAMESIZE];
+       int err = 0;
+
+       if (!sh->flow_tbls)
+               err = mlx5_alloc_table_hash_list(priv);
+       else
+               DRV_LOG(DEBUG, "sh->flow_tbls[%p] already created, reuse\n",
+                       (void *)sh->flow_tbls);
+       if (err)
+               return err;
+       /* Create tags hash list table. */
+       snprintf(s, sizeof(s), "%s_tags", sh->ibdev_name);
+       sh->tag_table = mlx5_hlist_create(s, MLX5_TAGS_HLIST_ARRAY_SIZE);
+       if (!sh->tag_table) {
+               DRV_LOG(ERR, "tags with hash creation failed.\n");
+               err = ENOMEM;
+               goto error;
+       }
+#ifdef HAVE_MLX5DV_DR
+       void *domain;
+
+       if (sh->dv_refcnt) {
+               /* Shared DV/DR structures is already initialized. */
+               sh->dv_refcnt++;
+               priv->dr_shared = 1;
+               return 0;
+       }
+       /* Reference counter is zero, we should initialize structures. */
+       domain = mlx5_glue->dr_create_domain(sh->ctx,
+                                            MLX5DV_DR_DOMAIN_TYPE_NIC_RX);
+       if (!domain) {
+               DRV_LOG(ERR, "ingress mlx5dv_dr_create_domain failed");
+               err = errno;
+               goto error;
+       }
+       sh->rx_domain = domain;
+       domain = mlx5_glue->dr_create_domain(sh->ctx,
+                                            MLX5DV_DR_DOMAIN_TYPE_NIC_TX);
+       if (!domain) {
+               DRV_LOG(ERR, "egress mlx5dv_dr_create_domain failed");
+               err = errno;
+               goto error;
+       }
+       pthread_mutex_init(&sh->dv_mutex, NULL);
+       sh->tx_domain = domain;
+#ifdef HAVE_MLX5DV_DR_ESWITCH
+       if (priv->config.dv_esw_en) {
+               domain  = mlx5_glue->dr_create_domain
+                       (sh->ctx, MLX5DV_DR_DOMAIN_TYPE_FDB);
+               if (!domain) {
+                       DRV_LOG(ERR, "FDB mlx5dv_dr_create_domain failed");
+                       err = errno;
+                       goto error;
+               }
+               sh->fdb_domain = domain;
+               sh->esw_drop_action = mlx5_glue->dr_create_flow_action_drop();
+       }
+#endif
+       if (priv->config.reclaim_mode == MLX5_RCM_AGGR) {
+               mlx5_glue->dr_reclaim_domain_memory(sh->rx_domain, 1);
+               mlx5_glue->dr_reclaim_domain_memory(sh->tx_domain, 1);
+               if (sh->fdb_domain)
+                       mlx5_glue->dr_reclaim_domain_memory(sh->fdb_domain, 1);
+       }
+       sh->pop_vlan_action = mlx5_glue->dr_create_flow_action_pop_vlan();
+#endif /* HAVE_MLX5DV_DR */
+       sh->dv_refcnt++;
+       priv->dr_shared = 1;
+       return 0;
+error:
+       /* Rollback the created objects. */
+       if (sh->rx_domain) {
+               mlx5_glue->dr_destroy_domain(sh->rx_domain);
+               sh->rx_domain = NULL;
+       }
+       if (sh->tx_domain) {
+               mlx5_glue->dr_destroy_domain(sh->tx_domain);
+               sh->tx_domain = NULL;
+       }
+       if (sh->fdb_domain) {
+               mlx5_glue->dr_destroy_domain(sh->fdb_domain);
+               sh->fdb_domain = NULL;
+       }
+       if (sh->esw_drop_action) {
+               mlx5_glue->destroy_flow_action(sh->esw_drop_action);
+               sh->esw_drop_action = NULL;
+       }
+       if (sh->pop_vlan_action) {
+               mlx5_glue->destroy_flow_action(sh->pop_vlan_action);
+               sh->pop_vlan_action = NULL;
+       }
+       if (sh->tag_table) {
+               /* tags should be destroyed with flow before. */
+               mlx5_hlist_destroy(sh->tag_table, NULL, NULL);
+               sh->tag_table = NULL;
+       }
+       mlx5_free_table_hash_list(priv);
+       return err;
+}
+
+/**
+ * Destroy DR related data within private structure.
+ *
+ * @param[in] priv
+ *   Pointer to the private device data structure.
+ */
+void
+mlx5_os_free_shared_dr(struct mlx5_priv *priv)
+{
+       struct mlx5_dev_ctx_shared *sh;
+
+       if (!priv->dr_shared)
+               return;
+       priv->dr_shared = 0;
+       sh = priv->sh;
+       MLX5_ASSERT(sh);
+#ifdef HAVE_MLX5DV_DR
+       MLX5_ASSERT(sh->dv_refcnt);
+       if (sh->dv_refcnt && --sh->dv_refcnt)
+               return;
+       if (sh->rx_domain) {
+               mlx5_glue->dr_destroy_domain(sh->rx_domain);
+               sh->rx_domain = NULL;
+       }
+       if (sh->tx_domain) {
+               mlx5_glue->dr_destroy_domain(sh->tx_domain);
+               sh->tx_domain = NULL;
+       }
+#ifdef HAVE_MLX5DV_DR_ESWITCH
+       if (sh->fdb_domain) {
+               mlx5_glue->dr_destroy_domain(sh->fdb_domain);
+               sh->fdb_domain = NULL;
+       }
+       if (sh->esw_drop_action) {
+               mlx5_glue->destroy_flow_action(sh->esw_drop_action);
+               sh->esw_drop_action = NULL;
+       }
+#endif
+       if (sh->pop_vlan_action) {
+               mlx5_glue->destroy_flow_action(sh->pop_vlan_action);
+               sh->pop_vlan_action = NULL;
+       }
+       pthread_mutex_destroy(&sh->dv_mutex);
+#endif /* HAVE_MLX5DV_DR */
+       if (sh->tag_table) {
+               /* tags should be destroyed with flow before. */
+               mlx5_hlist_destroy(sh->tag_table, NULL, NULL);
+               sh->tag_table = NULL;
+       }
+       mlx5_free_table_hash_list(priv);
+}
+
+/**
+ * Spawn an Ethernet device from Verbs information.
+ *
+ * @param dpdk_dev
+ *   Backing DPDK device.
+ * @param spawn
+ *   Verbs device parameters (name, port, switch_info) to spawn.
+ * @param config
+ *   Device configuration parameters.
+ *
+ * @return
+ *   A valid Ethernet device object on success, NULL otherwise and rte_errno
+ *   is set. The following errors are defined:
+ *
+ *   EBUSY: device is not supposed to be spawned.
+ *   EEXIST: device is already spawned
+ */
+static struct rte_eth_dev *
+mlx5_dev_spawn(struct rte_device *dpdk_dev,
+              struct mlx5_dev_spawn_data *spawn,
+              struct mlx5_dev_config config)
+{
+       const struct mlx5_switch_info *switch_info = &spawn->info;
+       struct mlx5_dev_ctx_shared *sh = NULL;
+       struct ibv_port_attr port_attr;
+       struct mlx5dv_context dv_attr = { .comp_mask = 0 };
+       struct rte_eth_dev *eth_dev = NULL;
+       struct mlx5_priv *priv = NULL;
+       int err = 0;
+       unsigned int hw_padding = 0;
+       unsigned int mps;
+       unsigned int cqe_comp;
+       unsigned int cqe_pad = 0;
+       unsigned int tunnel_en = 0;
+       unsigned int mpls_en = 0;
+       unsigned int swp = 0;
+       unsigned int mprq = 0;
+       unsigned int mprq_min_stride_size_n = 0;
+       unsigned int mprq_max_stride_size_n = 0;
+       unsigned int mprq_min_stride_num_n = 0;
+       unsigned int mprq_max_stride_num_n = 0;
+       struct rte_ether_addr mac;
+       char name[RTE_ETH_NAME_MAX_LEN];
+       int own_domain_id = 0;
+       uint16_t port_id;
+       unsigned int i;
+#ifdef HAVE_MLX5DV_DR_DEVX_PORT
+       struct mlx5dv_devx_port devx_port = { .comp_mask = 0 };
+#endif
+
+       /* Determine if this port representor is supposed to be spawned. */
+       if (switch_info->representor && dpdk_dev->devargs) {
+               struct rte_eth_devargs eth_da;
+
+               err = rte_eth_devargs_parse(dpdk_dev->devargs->args, &eth_da);
+               if (err) {
+                       rte_errno = -err;
+                       DRV_LOG(ERR, "failed to process device arguments: %s",
+                               strerror(rte_errno));
+                       return NULL;
+               }
+               for (i = 0; i < eth_da.nb_representor_ports; ++i)
+                       if (eth_da.representor_ports[i] ==
+                           (uint16_t)switch_info->port_name)
+                               break;
+               if (i == eth_da.nb_representor_ports) {
+                       rte_errno = EBUSY;
+                       return NULL;
+               }
+       }
+       /* Build device name. */
+       if (spawn->pf_bond <  0) {
+               /* Single device. */
+               if (!switch_info->representor)
+                       strlcpy(name, dpdk_dev->name, sizeof(name));
+               else
+                       snprintf(name, sizeof(name), "%s_representor_%u",
+                                dpdk_dev->name, switch_info->port_name);
+       } else {
+               /* Bonding device. */
+               if (!switch_info->representor)
+                       snprintf(name, sizeof(name), "%s_%s",
+                                dpdk_dev->name, spawn->ibv_dev->name);
+               else
+                       snprintf(name, sizeof(name), "%s_%s_representor_%u",
+                                dpdk_dev->name, spawn->ibv_dev->name,
+                                switch_info->port_name);
+       }
+       /* check if the device is already spawned */
+       if (rte_eth_dev_get_port_by_name(name, &port_id) == 0) {
+               rte_errno = EEXIST;
+               return NULL;
+       }
+       DRV_LOG(DEBUG, "naming Ethernet device \"%s\"", name);
+       if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
+               struct mlx5_mp_id mp_id;
+
+               eth_dev = rte_eth_dev_attach_secondary(name);
+               if (eth_dev == NULL) {
+                       DRV_LOG(ERR, "can not attach rte ethdev");
+                       rte_errno = ENOMEM;
+                       return NULL;
+               }
+               eth_dev->device = dpdk_dev;
+               eth_dev->dev_ops = &mlx5_dev_sec_ops;
+               err = mlx5_proc_priv_init(eth_dev);
+               if (err)
+                       return NULL;
+               mp_id.port_id = eth_dev->data->port_id;
+               strlcpy(mp_id.name, MLX5_MP_NAME, RTE_MP_MAX_NAME_LEN);
+               /* Receive command fd from primary process */
+               err = mlx5_mp_req_verbs_cmd_fd(&mp_id);
+               if (err < 0)
+                       goto err_secondary;
+               /* Remap UAR for Tx queues. */
+               err = mlx5_tx_uar_init_secondary(eth_dev, err);
+               if (err)
+                       goto err_secondary;
+               /*
+                * Ethdev pointer is still required as input since
+                * the primary device is not accessible from the
+                * secondary process.
+                */
+               eth_dev->rx_pkt_burst = mlx5_select_rx_function(eth_dev);
+               eth_dev->tx_pkt_burst = mlx5_select_tx_function(eth_dev);
+               return eth_dev;
+err_secondary:
+               mlx5_dev_close(eth_dev);
+               return NULL;
+       }
+       /*
+        * Some parameters ("tx_db_nc" in particularly) are needed in
+        * advance to create dv/verbs device context. We proceed the
+        * devargs here to get ones, and later proceed devargs again
+        * to override some hardware settings.
+        */
+       err = mlx5_args(&config, dpdk_dev->devargs);
+       if (err) {
+               err = rte_errno;
+               DRV_LOG(ERR, "failed to process device arguments: %s",
+                       strerror(rte_errno));
+               goto error;
+       }
+       sh = mlx5_alloc_shared_ibctx(spawn, &config);
+       if (!sh)
+               return NULL;
+       config.devx = sh->devx;
+#ifdef HAVE_MLX5DV_DR_ACTION_DEST_DEVX_TIR
+       config.dest_tir = 1;
+#endif
+#ifdef HAVE_IBV_MLX5_MOD_SWP
+       dv_attr.comp_mask |= MLX5DV_CONTEXT_MASK_SWP;
+#endif
+       /*
+        * Multi-packet send is supported by ConnectX-4 Lx PF as well
+        * as all ConnectX-5 devices.
+        */
+#ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
+       dv_attr.comp_mask |= MLX5DV_CONTEXT_MASK_TUNNEL_OFFLOADS;
+#endif
+#ifdef HAVE_IBV_DEVICE_STRIDING_RQ_SUPPORT
+       dv_attr.comp_mask |= MLX5DV_CONTEXT_MASK_STRIDING_RQ;
+#endif
+       mlx5_glue->dv_query_device(sh->ctx, &dv_attr);
+       if (dv_attr.flags & MLX5DV_CONTEXT_FLAGS_MPW_ALLOWED) {
+               if (dv_attr.flags & MLX5DV_CONTEXT_FLAGS_ENHANCED_MPW) {
+                       DRV_LOG(DEBUG, "enhanced MPW is supported");
+                       mps = MLX5_MPW_ENHANCED;
+               } else {
+                       DRV_LOG(DEBUG, "MPW is supported");
+                       mps = MLX5_MPW;
+               }
+       } else {
+               DRV_LOG(DEBUG, "MPW isn't supported");
+               mps = MLX5_MPW_DISABLED;
+       }
+#ifdef HAVE_IBV_MLX5_MOD_SWP
+       if (dv_attr.comp_mask & MLX5DV_CONTEXT_MASK_SWP)
+               swp = dv_attr.sw_parsing_caps.sw_parsing_offloads;
+       DRV_LOG(DEBUG, "SWP support: %u", swp);
+#endif
+       config.swp = !!swp;
+#ifdef HAVE_IBV_DEVICE_STRIDING_RQ_SUPPORT
+       if (dv_attr.comp_mask & MLX5DV_CONTEXT_MASK_STRIDING_RQ) {
+               struct mlx5dv_striding_rq_caps mprq_caps =
+                       dv_attr.striding_rq_caps;
+
+               DRV_LOG(DEBUG, "\tmin_single_stride_log_num_of_bytes: %d",
+                       mprq_caps.min_single_stride_log_num_of_bytes);
+               DRV_LOG(DEBUG, "\tmax_single_stride_log_num_of_bytes: %d",
+                       mprq_caps.max_single_stride_log_num_of_bytes);
+               DRV_LOG(DEBUG, "\tmin_single_wqe_log_num_of_strides: %d",
+                       mprq_caps.min_single_wqe_log_num_of_strides);
+               DRV_LOG(DEBUG, "\tmax_single_wqe_log_num_of_strides: %d",
+                       mprq_caps.max_single_wqe_log_num_of_strides);
+               DRV_LOG(DEBUG, "\tsupported_qpts: %d",
+                       mprq_caps.supported_qpts);
+               DRV_LOG(DEBUG, "device supports Multi-Packet RQ");
+               mprq = 1;
+               mprq_min_stride_size_n =
+                       mprq_caps.min_single_stride_log_num_of_bytes;
+               mprq_max_stride_size_n =
+                       mprq_caps.max_single_stride_log_num_of_bytes;
+               mprq_min_stride_num_n =
+                       mprq_caps.min_single_wqe_log_num_of_strides;
+               mprq_max_stride_num_n =
+                       mprq_caps.max_single_wqe_log_num_of_strides;
+       }
+#endif
+       if (RTE_CACHE_LINE_SIZE == 128 &&
+           !(dv_attr.flags & MLX5DV_CONTEXT_FLAGS_CQE_128B_COMP))
+               cqe_comp = 0;
+       else
+               cqe_comp = 1;
+       config.cqe_comp = cqe_comp;
+#ifdef HAVE_IBV_MLX5_MOD_CQE_128B_PAD
+       /* Whether device supports 128B Rx CQE padding. */
+       cqe_pad = RTE_CACHE_LINE_SIZE == 128 &&
+                 (dv_attr.flags & MLX5DV_CONTEXT_FLAGS_CQE_128B_PAD);
+#endif
+#ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
+       if (dv_attr.comp_mask & MLX5DV_CONTEXT_MASK_TUNNEL_OFFLOADS) {
+               tunnel_en = ((dv_attr.tunnel_offloads_caps &
+                             MLX5DV_RAW_PACKET_CAP_TUNNELED_OFFLOAD_VXLAN) &&
+                            (dv_attr.tunnel_offloads_caps &
+                             MLX5DV_RAW_PACKET_CAP_TUNNELED_OFFLOAD_GRE) &&
+                            (dv_attr.tunnel_offloads_caps &
+                             MLX5DV_RAW_PACKET_CAP_TUNNELED_OFFLOAD_GENEVE));
+       }
+       DRV_LOG(DEBUG, "tunnel offloading is %ssupported",
+               tunnel_en ? "" : "not ");
+#else
+       DRV_LOG(WARNING,
+               "tunnel offloading disabled due to old OFED/rdma-core version");
+#endif
+       config.tunnel_en = tunnel_en;
+#ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
+       mpls_en = ((dv_attr.tunnel_offloads_caps &
+                   MLX5DV_RAW_PACKET_CAP_TUNNELED_OFFLOAD_CW_MPLS_OVER_GRE) &&
+                  (dv_attr.tunnel_offloads_caps &
+                   MLX5DV_RAW_PACKET_CAP_TUNNELED_OFFLOAD_CW_MPLS_OVER_UDP));
+       DRV_LOG(DEBUG, "MPLS over GRE/UDP tunnel offloading is %ssupported",
+               mpls_en ? "" : "not ");
+#else
+       DRV_LOG(WARNING, "MPLS over GRE/UDP tunnel offloading disabled due to"
+               " old OFED/rdma-core version or firmware configuration");
+#endif
+       config.mpls_en = mpls_en;
+       /* Check port status. */
+       err = mlx5_glue->query_port(sh->ctx, spawn->ibv_port, &port_attr);
+       if (err) {
+               DRV_LOG(ERR, "port query failed: %s", strerror(err));
+               goto error;
+       }
+       if (port_attr.link_layer != IBV_LINK_LAYER_ETHERNET) {
+               DRV_LOG(ERR, "port is not configured in Ethernet mode");
+               err = EINVAL;
+               goto error;
+       }
+       if (port_attr.state != IBV_PORT_ACTIVE)
+               DRV_LOG(DEBUG, "port is not active: \"%s\" (%d)",
+                       mlx5_glue->port_state_str(port_attr.state),
+                       port_attr.state);
+       /* Allocate private eth device data. */
+       priv = rte_zmalloc("ethdev private structure",
+                          sizeof(*priv),
+                          RTE_CACHE_LINE_SIZE);
+       if (priv == NULL) {
+               DRV_LOG(ERR, "priv allocation failure");
+               err = ENOMEM;
+               goto error;
+       }
+       priv->sh = sh;
+       priv->ibv_port = spawn->ibv_port;
+       priv->pci_dev = spawn->pci_dev;
+       priv->mtu = RTE_ETHER_MTU;
+       priv->mp_id.port_id = port_id;
+       strlcpy(priv->mp_id.name, MLX5_MP_NAME, RTE_MP_MAX_NAME_LEN);
+#ifndef RTE_ARCH_64
+       /* Initialize UAR access locks for 32bit implementations. */
+       rte_spinlock_init(&priv->uar_lock_cq);
+       for (i = 0; i < MLX5_UAR_PAGE_NUM_MAX; i++)
+               rte_spinlock_init(&priv->uar_lock[i]);
+#endif
+       /* Some internal functions rely on Netlink sockets, open them now. */
+       priv->nl_socket_rdma = mlx5_nl_init(NETLINK_RDMA);
+       priv->nl_socket_route = mlx5_nl_init(NETLINK_ROUTE);
+       priv->representor = !!switch_info->representor;
+       priv->master = !!switch_info->master;
+       priv->domain_id = RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID;
+       priv->vport_meta_tag = 0;
+       priv->vport_meta_mask = 0;
+       priv->pf_bond = spawn->pf_bond;
+#ifdef HAVE_MLX5DV_DR_DEVX_PORT
+       /*
+        * The DevX port query API is implemented. E-Switch may use
+        * either vport or reg_c[0] metadata register to match on
+        * vport index. The engaged part of metadata register is
+        * defined by mask.
+        */
+       if (switch_info->representor || switch_info->master) {
+               devx_port.comp_mask = MLX5DV_DEVX_PORT_VPORT |
+                                     MLX5DV_DEVX_PORT_MATCH_REG_C_0;
+               err = mlx5_glue->devx_port_query(sh->ctx, spawn->ibv_port,
+                                                &devx_port);
+               if (err) {
+                       DRV_LOG(WARNING,
+                               "can't query devx port %d on device %s",
+                               spawn->ibv_port, spawn->ibv_dev->name);
+                       devx_port.comp_mask = 0;
+               }
+       }
+       if (devx_port.comp_mask & MLX5DV_DEVX_PORT_MATCH_REG_C_0) {
+               priv->vport_meta_tag = devx_port.reg_c_0.value;
+               priv->vport_meta_mask = devx_port.reg_c_0.mask;
+               if (!priv->vport_meta_mask) {
+                       DRV_LOG(ERR, "vport zero mask for port %d"
+                                    " on bonding device %s",
+                                    spawn->ibv_port, spawn->ibv_dev->name);
+                       err = ENOTSUP;
+                       goto error;
+               }
+               if (priv->vport_meta_tag & ~priv->vport_meta_mask) {
+                       DRV_LOG(ERR, "invalid vport tag for port %d"
+                                    " on bonding device %s",
+                                    spawn->ibv_port, spawn->ibv_dev->name);
+                       err = ENOTSUP;
+                       goto error;
+               }
+       }
+       if (devx_port.comp_mask & MLX5DV_DEVX_PORT_VPORT) {
+               priv->vport_id = devx_port.vport_num;
+       } else if (spawn->pf_bond >= 0) {
+               DRV_LOG(ERR, "can't deduce vport index for port %d"
+                            " on bonding device %s",
+                            spawn->ibv_port, spawn->ibv_dev->name);
+               err = ENOTSUP;
+               goto error;
+       } else {
+               /* Suppose vport index in compatible way. */
+               priv->vport_id = switch_info->representor ?
+                                switch_info->port_name + 1 : -1;
+       }
+#else
+       /*
+        * Kernel/rdma_core support single E-Switch per PF configurations
+        * only and vport_id field contains the vport index for
+        * associated VF, which is deduced from representor port name.
+        * For example, let's have the IB device port 10, it has
+        * attached network device eth0, which has port name attribute
+        * pf0vf2, we can deduce the VF number as 2, and set vport index
+        * as 3 (2+1). This assigning schema should be changed if the
+        * multiple E-Switch instances per PF configurations or/and PCI
+        * subfunctions are added.
+        */
+       priv->vport_id = switch_info->representor ?
+                        switch_info->port_name + 1 : -1;
+#endif
+       /* representor_id field keeps the unmodified VF index. */
+       priv->representor_id = switch_info->representor ?
+                              switch_info->port_name : -1;
+       /*
+        * Look for sibling devices in order to reuse their switch domain
+        * if any, otherwise allocate one.
+        */
+       MLX5_ETH_FOREACH_DEV(port_id, priv->pci_dev) {
+               const struct mlx5_priv *opriv =
+                       rte_eth_devices[port_id].data->dev_private;
+
+               if (!opriv ||
+                   opriv->sh != priv->sh ||
+                       opriv->domain_id ==
+                       RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID)
+                       continue;
+               priv->domain_id = opriv->domain_id;
+               break;
+       }
+       if (priv->domain_id == RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID) {
+               err = rte_eth_switch_domain_alloc(&priv->domain_id);
+               if (err) {
+                       err = rte_errno;
+                       DRV_LOG(ERR, "unable to allocate switch domain: %s",
+                               strerror(rte_errno));
+                       goto error;
+               }
+               own_domain_id = 1;
+       }
+       /* Override some values set by hardware configuration. */
+       mlx5_args(&config, dpdk_dev->devargs);
+       err = mlx5_dev_check_sibling_config(priv, &config);
+       if (err)
+               goto error;
+       config.hw_csum = !!(sh->device_attr.device_cap_flags_ex &
+                           IBV_DEVICE_RAW_IP_CSUM);
+       DRV_LOG(DEBUG, "checksum offloading is %ssupported",
+               (config.hw_csum ? "" : "not "));
+#if !defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42) && \
+       !defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
+       DRV_LOG(DEBUG, "counters are not supported");
+#endif
+#if !defined(HAVE_IBV_FLOW_DV_SUPPORT) || !defined(HAVE_MLX5DV_DR)
+       if (config.dv_flow_en) {
+               DRV_LOG(WARNING, "DV flow is not supported");
+               config.dv_flow_en = 0;
+       }
+#endif
+       config.ind_table_max_size =
+               sh->device_attr.max_rwq_indirection_table_size;
+       /*
+        * Remove this check once DPDK supports larger/variable
+        * indirection tables.
+        */
+       if (config.ind_table_max_size > (unsigned int)ETH_RSS_RETA_SIZE_512)
+               config.ind_table_max_size = ETH_RSS_RETA_SIZE_512;
+       DRV_LOG(DEBUG, "maximum Rx indirection table size is %u",
+               config.ind_table_max_size);
+       config.hw_vlan_strip = !!(sh->device_attr.raw_packet_caps &
+                                 IBV_RAW_PACKET_CAP_CVLAN_STRIPPING);
+       DRV_LOG(DEBUG, "VLAN stripping is %ssupported",
+               (config.hw_vlan_strip ? "" : "not "));
+       config.hw_fcs_strip = !!(sh->device_attr.raw_packet_caps &
+                                IBV_RAW_PACKET_CAP_SCATTER_FCS);
+       DRV_LOG(DEBUG, "FCS stripping configuration is %ssupported",
+               (config.hw_fcs_strip ? "" : "not "));
+#if defined(HAVE_IBV_WQ_FLAG_RX_END_PADDING)
+       hw_padding = !!sh->device_attr.rx_pad_end_addr_align;
+#elif defined(HAVE_IBV_WQ_FLAGS_PCI_WRITE_END_PADDING)
+       hw_padding = !!(sh->device_attr.device_cap_flags_ex &
+                       IBV_DEVICE_PCI_WRITE_END_PADDING);
+#endif
+       if (config.hw_padding && !hw_padding) {
+               DRV_LOG(DEBUG, "Rx end alignment padding isn't supported");
+               config.hw_padding = 0;
+       } else if (config.hw_padding) {
+               DRV_LOG(DEBUG, "Rx end alignment padding is enabled");
+       }
+       config.tso = (sh->device_attr.max_tso > 0 &&
+                     (sh->device_attr.tso_supported_qpts &
+                      (1 << IBV_QPT_RAW_PACKET)));
+       if (config.tso)
+               config.tso_max_payload_sz = sh->device_attr.max_tso;
+       /*
+        * MPW is disabled by default, while the Enhanced MPW is enabled
+        * by default.
+        */
+       if (config.mps == MLX5_ARG_UNSET)
+               config.mps = (mps == MLX5_MPW_ENHANCED) ? MLX5_MPW_ENHANCED :
+                                                         MLX5_MPW_DISABLED;
+       else
+               config.mps = config.mps ? mps : MLX5_MPW_DISABLED;
+       DRV_LOG(INFO, "%sMPS is %s",
+               config.mps == MLX5_MPW_ENHANCED ? "enhanced " :
+               config.mps == MLX5_MPW ? "legacy " : "",
+               config.mps != MLX5_MPW_DISABLED ? "enabled" : "disabled");
+       if (config.cqe_comp && !cqe_comp) {
+               DRV_LOG(WARNING, "Rx CQE compression isn't supported");
+               config.cqe_comp = 0;
+       }
+       if (config.cqe_pad && !cqe_pad) {
+               DRV_LOG(WARNING, "Rx CQE padding isn't supported");
+               config.cqe_pad = 0;
+       } else if (config.cqe_pad) {
+               DRV_LOG(INFO, "Rx CQE padding is enabled");
+       }
+       if (config.devx) {
+               priv->counter_fallback = 0;
+               err = mlx5_devx_cmd_query_hca_attr(sh->ctx, &config.hca_attr);
+               if (err) {
+                       err = -err;
+                       goto error;
+               }
+               if (!config.hca_attr.flow_counters_dump)
+                       priv->counter_fallback = 1;
+#ifndef HAVE_IBV_DEVX_ASYNC
+               priv->counter_fallback = 1;
+#endif
+               if (priv->counter_fallback)
+                       DRV_LOG(INFO, "Use fall-back DV counter management");
+               /* Check for LRO support. */
+               if (config.dest_tir && config.hca_attr.lro_cap &&
+                   config.dv_flow_en) {
+                       /* TBD check tunnel lro caps. */
+                       config.lro.supported = config.hca_attr.lro_cap;
+                       DRV_LOG(DEBUG, "Device supports LRO");
+                       /*
+                        * If LRO timeout is not configured by application,
+                        * use the minimal supported value.
+                        */
+                       if (!config.lro.timeout)
+                               config.lro.timeout =
+                               config.hca_attr.lro_timer_supported_periods[0];
+                       DRV_LOG(DEBUG, "LRO session timeout set to %d usec",
+                               config.lro.timeout);
+               }
+#if defined(HAVE_MLX5DV_DR) && defined(HAVE_MLX5_DR_CREATE_ACTION_FLOW_METER)
+               if (config.hca_attr.qos.sup && config.hca_attr.qos.srtcm_sup &&
+                   config.dv_flow_en) {
+                       uint8_t reg_c_mask =
+                               config.hca_attr.qos.flow_meter_reg_c_ids;
+                       /*
+                        * Meter needs two REG_C's for color match and pre-sfx
+                        * flow match. Here get the REG_C for color match.
+                        * REG_C_0 and REG_C_1 is reserved for metadata feature.
+                        */
+                       reg_c_mask &= 0xfc;
+                       if (__builtin_popcount(reg_c_mask) < 1) {
+                               priv->mtr_en = 0;
+                               DRV_LOG(WARNING, "No available register for"
+                                       " meter.");
+                       } else {
+                               priv->mtr_color_reg = ffs(reg_c_mask) - 1 +
+                                                     REG_C_0;
+                               priv->mtr_en = 1;
+                               priv->mtr_reg_share =
+                                     config.hca_attr.qos.flow_meter_reg_share;
+                               DRV_LOG(DEBUG, "The REG_C meter uses is %d",
+                                       priv->mtr_color_reg);
+                       }
+               }
+#endif
+       }
+       if (config.mprq.enabled && mprq) {
+               if (config.mprq.stride_num_n &&
+                   (config.mprq.stride_num_n > mprq_max_stride_num_n ||
+                    config.mprq.stride_num_n < mprq_min_stride_num_n)) {
+                       config.mprq.stride_num_n =
+                               RTE_MIN(RTE_MAX(MLX5_MPRQ_STRIDE_NUM_N,
+                                               mprq_min_stride_num_n),
+                                       mprq_max_stride_num_n);
+                       DRV_LOG(WARNING,
+                               "the number of strides"
+                               " for Multi-Packet RQ is out of range,"
+                               " setting default value (%u)",
+                               1 << config.mprq.stride_num_n);
+               }
+               if (config.mprq.stride_size_n &&
+                   (config.mprq.stride_size_n > mprq_max_stride_size_n ||
+                    config.mprq.stride_size_n < mprq_min_stride_size_n)) {
+                       config.mprq.stride_size_n =
+                               RTE_MIN(RTE_MAX(MLX5_MPRQ_STRIDE_SIZE_N,
+                                               mprq_min_stride_size_n),
+                                       mprq_max_stride_size_n);
+                       DRV_LOG(WARNING,
+                               "the size of a stride"
+                               " for Multi-Packet RQ is out of range,"
+                               " setting default value (%u)",
+                               1 << config.mprq.stride_size_n);
+               }
+               config.mprq.min_stride_size_n = mprq_min_stride_size_n;
+               config.mprq.max_stride_size_n = mprq_max_stride_size_n;
+       } else if (config.mprq.enabled && !mprq) {
+               DRV_LOG(WARNING, "Multi-Packet RQ isn't supported");
+               config.mprq.enabled = 0;
+       }
+       if (config.max_dump_files_num == 0)
+               config.max_dump_files_num = 128;
+       eth_dev = rte_eth_dev_allocate(name);
+       if (eth_dev == NULL) {
+               DRV_LOG(ERR, "can not allocate rte ethdev");
+               err = ENOMEM;
+               goto error;
+       }
+       /* Flag to call rte_eth_dev_release_port() in rte_eth_dev_close(). */
+       eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
+       if (priv->representor) {
+               eth_dev->data->dev_flags |= RTE_ETH_DEV_REPRESENTOR;
+               eth_dev->data->representor_id = priv->representor_id;
+       }
+       /*
+        * Store associated network device interface index. This index
+        * is permanent throughout the lifetime of device. So, we may store
+        * the ifindex here and use the cached value further.
+        */
+       MLX5_ASSERT(spawn->ifindex);
+       priv->if_index = spawn->ifindex;
+       eth_dev->data->dev_private = priv;
+       priv->dev_data = eth_dev->data;
+       eth_dev->data->mac_addrs = priv->mac;
+       eth_dev->device = dpdk_dev;
+       /* Configure the first MAC address by default. */
+       if (mlx5_get_mac(eth_dev, &mac.addr_bytes)) {
+               DRV_LOG(ERR,
+                       "port %u cannot get MAC address, is mlx5_en"
+                       " loaded? (errno: %s)",
+                       eth_dev->data->port_id, strerror(rte_errno));
+               err = ENODEV;
+               goto error;
+       }
+       DRV_LOG(INFO,
+               "port %u MAC address is %02x:%02x:%02x:%02x:%02x:%02x",
+               eth_dev->data->port_id,
+               mac.addr_bytes[0], mac.addr_bytes[1],
+               mac.addr_bytes[2], mac.addr_bytes[3],
+               mac.addr_bytes[4], mac.addr_bytes[5]);
+#ifdef RTE_LIBRTE_MLX5_DEBUG
+       {
+               char ifname[IF_NAMESIZE];
+
+               if (mlx5_get_ifname(eth_dev, &ifname) == 0)
+                       DRV_LOG(DEBUG, "port %u ifname is \"%s\"",
+                               eth_dev->data->port_id, ifname);
+               else
+                       DRV_LOG(DEBUG, "port %u ifname is unknown",
+                               eth_dev->data->port_id);
+       }
+#endif
+       /* Get actual MTU if possible. */
+       err = mlx5_get_mtu(eth_dev, &priv->mtu);
+       if (err) {
+               err = rte_errno;
+               goto error;
+       }
+       DRV_LOG(DEBUG, "port %u MTU is %u", eth_dev->data->port_id,
+               priv->mtu);
+       /* Initialize burst functions to prevent crashes before link-up. */
+       eth_dev->rx_pkt_burst = removed_rx_burst;
+       eth_dev->tx_pkt_burst = removed_tx_burst;
+       eth_dev->dev_ops = &mlx5_dev_ops;
+       /* Register MAC address. */
+       claim_zero(mlx5_mac_addr_add(eth_dev, &mac, 0, 0));
+       if (config.vf && config.vf_nl_en)
+               mlx5_nl_mac_addr_sync(priv->nl_socket_route,
+                                     mlx5_ifindex(eth_dev),
+                                     eth_dev->data->mac_addrs,
+                                     MLX5_MAX_MAC_ADDRESSES);
+       priv->flows = 0;
+       priv->ctrl_flows = 0;
+       TAILQ_INIT(&priv->flow_meters);
+       TAILQ_INIT(&priv->flow_meter_profiles);
+       /* Hint libmlx5 to use PMD allocator for data plane resources */
+       struct mlx5dv_ctx_allocators alctr = {
+               .alloc = &mlx5_alloc_verbs_buf,
+               .free = &mlx5_free_verbs_buf,
+               .data = priv,
+       };
+       mlx5_glue->dv_set_context_attr(sh->ctx,
+                                      MLX5DV_CTX_ATTR_BUF_ALLOCATORS,
+                                      (void *)((uintptr_t)&alctr));
+       /* Bring Ethernet device up. */
+       DRV_LOG(DEBUG, "port %u forcing Ethernet interface up",
+               eth_dev->data->port_id);
+       mlx5_set_link_up(eth_dev);
+       /*
+        * Even though the interrupt handler is not installed yet,
+        * interrupts will still trigger on the async_fd from
+        * Verbs context returned by ibv_open_device().
+        */
+       mlx5_link_update(eth_dev, 0);
+#ifdef HAVE_MLX5DV_DR_ESWITCH
+       if (!(config.hca_attr.eswitch_manager && config.dv_flow_en &&
+             (switch_info->representor || switch_info->master)))
+               config.dv_esw_en = 0;
+#else
+       config.dv_esw_en = 0;
+#endif
+       /* Detect minimal data bytes to inline. */
+       mlx5_set_min_inline(spawn, &config);
+       /* Store device configuration on private structure. */
+       priv->config = config;
+       /* Create context for virtual machine VLAN workaround. */
+       priv->vmwa_context = mlx5_vlan_vmwa_init(eth_dev, spawn->ifindex);
+       if (config.dv_flow_en) {
+               err = mlx5_alloc_shared_dr(priv);
+               if (err)
+                       goto error;
+               /*
+                * RSS id is shared with meter flow id. Meter flow id can only
+                * use the 24 MSB of the register.
+                */
+               priv->qrss_id_pool = mlx5_flow_id_pool_alloc(UINT32_MAX >>
+                                    MLX5_MTR_COLOR_BITS);
+               if (!priv->qrss_id_pool) {
+                       DRV_LOG(ERR, "can't create flow id pool");
+                       err = ENOMEM;
+                       goto error;
+               }
+       }
+       /* Supported Verbs flow priority number detection. */
+       err = mlx5_flow_discover_priorities(eth_dev);
+       if (err < 0) {
+               err = -err;
+               goto error;
+       }
+       priv->config.flow_prio = err;
+       if (!priv->config.dv_esw_en &&
+           priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
+               DRV_LOG(WARNING, "metadata mode %u is not supported "
+                                "(no E-Switch)", priv->config.dv_xmeta_en);
+               priv->config.dv_xmeta_en = MLX5_XMETA_MODE_LEGACY;
+       }
+       mlx5_set_metadata_mask(eth_dev);
+       if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
+           !priv->sh->dv_regc0_mask) {
+               DRV_LOG(ERR, "metadata mode %u is not supported "
+                            "(no metadata reg_c[0] is available)",
+                            priv->config.dv_xmeta_en);
+                       err = ENOTSUP;
+                       goto error;
+       }
+       /*
+        * Allocate the buffer for flow creating, just once.
+        * The allocation must be done before any flow creating.
+        */
+       mlx5_flow_alloc_intermediate(eth_dev);
+       /* Query availability of metadata reg_c's. */
+       err = mlx5_flow_discover_mreg_c(eth_dev);
+       if (err < 0) {
+               err = -err;
+               goto error;
+       }
+       if (!mlx5_flow_ext_mreg_supported(eth_dev)) {
+               DRV_LOG(DEBUG,
+                       "port %u extensive metadata register is not supported",
+                       eth_dev->data->port_id);
+               if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
+                       DRV_LOG(ERR, "metadata mode %u is not supported "
+                                    "(no metadata registers available)",
+                                    priv->config.dv_xmeta_en);
+                       err = ENOTSUP;
+                       goto error;
+               }
+       }
+       if (priv->config.dv_flow_en &&
+           priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
+           mlx5_flow_ext_mreg_supported(eth_dev) &&
+           priv->sh->dv_regc0_mask) {
+               priv->mreg_cp_tbl = mlx5_hlist_create(MLX5_FLOW_MREG_HNAME,
+                                                     MLX5_FLOW_MREG_HTABLE_SZ);
+               if (!priv->mreg_cp_tbl) {
+                       err = ENOMEM;
+                       goto error;
+               }
+       }
+       return eth_dev;
+error:
+       if (priv) {
+               if (priv->mreg_cp_tbl)
+                       mlx5_hlist_destroy(priv->mreg_cp_tbl, NULL, NULL);
+               if (priv->sh)
+                       mlx5_os_free_shared_dr(priv);
+               if (priv->nl_socket_route >= 0)
+                       close(priv->nl_socket_route);
+               if (priv->nl_socket_rdma >= 0)
+                       close(priv->nl_socket_rdma);
+               if (priv->vmwa_context)
+                       mlx5_vlan_vmwa_exit(priv->vmwa_context);
+               if (priv->qrss_id_pool)
+                       mlx5_flow_id_pool_release(priv->qrss_id_pool);
+               if (own_domain_id)
+                       claim_zero(rte_eth_switch_domain_free(priv->domain_id));
+               rte_free(priv);
+               if (eth_dev != NULL)
+                       eth_dev->data->dev_private = NULL;
+       }
+       if (eth_dev != NULL) {
+               /* mac_addrs must not be freed alone because part of
+                * dev_private
+                **/
+               eth_dev->data->mac_addrs = NULL;
+               rte_eth_dev_release_port(eth_dev);
+       }
+       if (sh)
+               mlx5_free_shared_ibctx(sh);
+       MLX5_ASSERT(err > 0);
+       rte_errno = err;
+       return NULL;
+}
+
+/**
+ * Comparison callback to sort device data.
+ *
+ * This is meant to be used with qsort().
+ *
+ * @param a[in]
+ *   Pointer to pointer to first data object.
+ * @param b[in]
+ *   Pointer to pointer to second data object.
+ *
+ * @return
+ *   0 if both objects are equal, less than 0 if the first argument is less
+ *   than the second, greater than 0 otherwise.
+ */
+static int
+mlx5_dev_spawn_data_cmp(const void *a, const void *b)
+{
+       const struct mlx5_switch_info *si_a =
+               &((const struct mlx5_dev_spawn_data *)a)->info;
+       const struct mlx5_switch_info *si_b =
+               &((const struct mlx5_dev_spawn_data *)b)->info;
+       int ret;
+
+       /* Master device first. */
+       ret = si_b->master - si_a->master;
+       if (ret)
+               return ret;
+       /* Then representor devices. */
+       ret = si_b->representor - si_a->representor;
+       if (ret)
+               return ret;
+       /* Unidentified devices come last in no specific order. */
+       if (!si_a->representor)
+               return 0;
+       /* Order representors by name. */
+       return si_a->port_name - si_b->port_name;
+}
+
+/**
+ * Match PCI information for possible slaves of bonding device.
+ *
+ * @param[in] ibv_dev
+ *   Pointer to Infiniband device structure.
+ * @param[in] pci_dev
+ *   Pointer to PCI device structure to match PCI address.
+ * @param[in] nl_rdma
+ *   Netlink RDMA group socket handle.
+ *
+ * @return
+ *   negative value if no bonding device found, otherwise
+ *   positive index of slave PF in bonding.
+ */
+static int
+mlx5_device_bond_pci_match(const struct ibv_device *ibv_dev,
+                          const struct rte_pci_device *pci_dev,
+                          int nl_rdma)
+{
+       char ifname[IF_NAMESIZE + 1];
+       unsigned int ifindex;
+       unsigned int np, i;
+       FILE *file = NULL;
+       int pf = -1;
+
+       /*
+        * Try to get master device name. If something goes
+        * wrong suppose the lack of kernel support and no
+        * bonding devices.
+        */
+       if (nl_rdma < 0)
+               return -1;
+       if (!strstr(ibv_dev->name, "bond"))
+               return -1;
+       np = mlx5_nl_portnum(nl_rdma, ibv_dev->name);
+       if (!np)
+               return -1;
+       /*
+        * The Master device might not be on the predefined
+        * port (not on port index 1, it is not garanted),
+        * we have to scan all Infiniband device port and
+        * find master.
+        */
+       for (i = 1; i <= np; ++i) {
+               /* Check whether Infiniband port is populated. */
+               ifindex = mlx5_nl_ifindex(nl_rdma, ibv_dev->name, i);
+               if (!ifindex)
+                       continue;
+               if (!if_indextoname(ifindex, ifname))
+                       continue;
+               /* Try to read bonding slave names from sysfs. */
+               MKSTR(slaves,
+                     "/sys/class/net/%s/master/bonding/slaves", ifname);
+               file = fopen(slaves, "r");
+               if (file)
+                       break;
+       }
+       if (!file)
+               return -1;
+       /* Use safe format to check maximal buffer length. */
+       MLX5_ASSERT(atol(RTE_STR(IF_NAMESIZE)) == IF_NAMESIZE);
+       while (fscanf(file, "%" RTE_STR(IF_NAMESIZE) "s", ifname) == 1) {
+               char tmp_str[IF_NAMESIZE + 32];
+               struct rte_pci_addr pci_addr;
+               struct mlx5_switch_info info;
+
+               /* Process slave interface names in the loop. */
+               snprintf(tmp_str, sizeof(tmp_str),
+                        "/sys/class/net/%s", ifname);
+               if (mlx5_dev_to_pci_addr(tmp_str, &pci_addr)) {
+                       DRV_LOG(WARNING, "can not get PCI address"
+                                        " for netdev \"%s\"", ifname);
+                       continue;
+               }
+               if (pci_dev->addr.domain != pci_addr.domain ||
+                   pci_dev->addr.bus != pci_addr.bus ||
+                   pci_dev->addr.devid != pci_addr.devid ||
+                   pci_dev->addr.function != pci_addr.function)
+                       continue;
+               /* Slave interface PCI address match found. */
+               fclose(file);
+               snprintf(tmp_str, sizeof(tmp_str),
+                        "/sys/class/net/%s/phys_port_name", ifname);
+               file = fopen(tmp_str, "rb");
+               if (!file)
+                       break;
+               info.name_type = MLX5_PHYS_PORT_NAME_TYPE_NOTSET;
+               if (fscanf(file, "%32s", tmp_str) == 1)
+                       mlx5_translate_port_name(tmp_str, &info);
+               if (info.name_type == MLX5_PHYS_PORT_NAME_TYPE_LEGACY ||
+                   info.name_type == MLX5_PHYS_PORT_NAME_TYPE_UPLINK)
+                       pf = info.port_name;
+               break;
+       }
+       if (file)
+               fclose(file);
+       return pf;
+}
+
+/**
+ * DPDK callback to register a PCI device.
+ *
+ * This function spawns Ethernet devices out of a given PCI device.
+ *
+ * @param[in] pci_drv
+ *   PCI driver structure (mlx5_driver).
+ * @param[in] pci_dev
+ *   PCI device information.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_os_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
+                 struct rte_pci_device *pci_dev)
+{
+       struct ibv_device **ibv_list;
+       /*
+        * Number of found IB Devices matching with requested PCI BDF.
+        * nd != 1 means there are multiple IB devices over the same
+        * PCI device and we have representors and master.
+        */
+       unsigned int nd = 0;
+       /*
+        * Number of found IB device Ports. nd = 1 and np = 1..n means
+        * we have the single multiport IB device, and there may be
+        * representors attached to some of found ports.
+        */
+       unsigned int np = 0;
+       /*
+        * Number of DPDK ethernet devices to Spawn - either over
+        * multiple IB devices or multiple ports of single IB device.
+        * Actually this is the number of iterations to spawn.
+        */
+       unsigned int ns = 0;
+       /*
+        * Bonding device
+        *   < 0 - no bonding device (single one)
+        *  >= 0 - bonding device (value is slave PF index)
+        */
+       int bd = -1;
+       struct mlx5_dev_spawn_data *list = NULL;
+       struct mlx5_dev_config dev_config;
+       int ret;
+
+       if (mlx5_class_get(pci_dev->device.devargs) != MLX5_CLASS_NET) {
+               DRV_LOG(DEBUG, "Skip probing - should be probed by other mlx5"
+                       " driver.");
+               return 1;
+       }
+       if (rte_eal_process_type() == RTE_PROC_PRIMARY)
+               mlx5_pmd_socket_init();
+       ret = mlx5_init_once();
+       if (ret) {
+               DRV_LOG(ERR, "unable to init PMD global data: %s",
+                       strerror(rte_errno));
+               return -rte_errno;
+       }
+       MLX5_ASSERT(pci_drv == &mlx5_driver);
+       errno = 0;
+       ibv_list = mlx5_glue->get_device_list(&ret);
+       if (!ibv_list) {
+               rte_errno = errno ? errno : ENOSYS;
+               DRV_LOG(ERR, "cannot list devices, is ib_uverbs loaded?");
+               return -rte_errno;
+       }
+       /*
+        * First scan the list of all Infiniband devices to find
+        * matching ones, gathering into the list.
+        */
+       struct ibv_device *ibv_match[ret + 1];
+       int nl_route = mlx5_nl_init(NETLINK_ROUTE);
+       int nl_rdma = mlx5_nl_init(NETLINK_RDMA);
+       unsigned int i;
+
+       while (ret-- > 0) {
+               struct rte_pci_addr pci_addr;
+
+               DRV_LOG(DEBUG, "checking device \"%s\"", ibv_list[ret]->name);
+               bd = mlx5_device_bond_pci_match
+                               (ibv_list[ret], pci_dev, nl_rdma);
+               if (bd >= 0) {
+                       /*
+                        * Bonding device detected. Only one match is allowed,
+                        * the bonding is supported over multi-port IB device,
+                        * there should be no matches on representor PCI
+                        * functions or non VF LAG bonding devices with
+                        * specified address.
+                        */
+                       if (nd) {
+                               DRV_LOG(ERR,
+                                       "multiple PCI match on bonding device"
+                                       "\"%s\" found", ibv_list[ret]->name);
+                               rte_errno = ENOENT;
+                               ret = -rte_errno;
+                               goto exit;
+                       }
+                       DRV_LOG(INFO, "PCI information matches for"
+                                     " slave %d bonding device \"%s\"",
+                                     bd, ibv_list[ret]->name);
+                       ibv_match[nd++] = ibv_list[ret];
+                       break;
+               }
+               if (mlx5_dev_to_pci_addr
+                       (ibv_list[ret]->ibdev_path, &pci_addr))
+                       continue;
+               if (pci_dev->addr.domain != pci_addr.domain ||
+                   pci_dev->addr.bus != pci_addr.bus ||
+                   pci_dev->addr.devid != pci_addr.devid ||
+                   pci_dev->addr.function != pci_addr.function)
+                       continue;
+               DRV_LOG(INFO, "PCI information matches for device \"%s\"",
+                       ibv_list[ret]->name);
+               ibv_match[nd++] = ibv_list[ret];
+       }
+       ibv_match[nd] = NULL;
+       if (!nd) {
+               /* No device matches, just complain and bail out. */
+               DRV_LOG(WARNING,
+                       "no Verbs device matches PCI device " PCI_PRI_FMT ","
+                       " are kernel drivers loaded?",
+                       pci_dev->addr.domain, pci_dev->addr.bus,
+                       pci_dev->addr.devid, pci_dev->addr.function);
+               rte_errno = ENOENT;
+               ret = -rte_errno;
+               goto exit;
+       }
+       if (nd == 1) {
+               /*
+                * Found single matching device may have multiple ports.
+                * Each port may be representor, we have to check the port
+                * number and check the representors existence.
+                */
+               if (nl_rdma >= 0)
+                       np = mlx5_nl_portnum(nl_rdma, ibv_match[0]->name);
+               if (!np)
+                       DRV_LOG(WARNING, "can not get IB device \"%s\""
+                                        " ports number", ibv_match[0]->name);
+               if (bd >= 0 && !np) {
+                       DRV_LOG(ERR, "can not get ports"
+                                    " for bonding device");
+                       rte_errno = ENOENT;
+                       ret = -rte_errno;
+                       goto exit;
+               }
+       }
+#ifndef HAVE_MLX5DV_DR_DEVX_PORT
+       if (bd >= 0) {
+               /*
+                * This may happen if there is VF LAG kernel support and
+                * application is compiled with older rdma_core library.
+                */
+               DRV_LOG(ERR,
+                       "No kernel/verbs support for VF LAG bonding found.");
+               rte_errno = ENOTSUP;
+               ret = -rte_errno;
+               goto exit;
+       }
+#endif
+       /*
+        * Now we can determine the maximal
+        * amount of devices to be spawned.
+        */
+       list = rte_zmalloc("device spawn data",
+                        sizeof(struct mlx5_dev_spawn_data) *
+                        (np ? np : nd),
+                        RTE_CACHE_LINE_SIZE);
+       if (!list) {
+               DRV_LOG(ERR, "spawn data array allocation failure");
+               rte_errno = ENOMEM;
+               ret = -rte_errno;
+               goto exit;
+       }
+       if (bd >= 0 || np > 1) {
+               /*
+                * Single IB device with multiple ports found,
+                * it may be E-Switch master device and representors.
+                * We have to perform identification through the ports.
+                */
+               MLX5_ASSERT(nl_rdma >= 0);
+               MLX5_ASSERT(ns == 0);
+               MLX5_ASSERT(nd == 1);
+               MLX5_ASSERT(np);
+               for (i = 1; i <= np; ++i) {
+                       list[ns].max_port = np;
+                       list[ns].ibv_port = i;
+                       list[ns].ibv_dev = ibv_match[0];
+                       list[ns].eth_dev = NULL;
+                       list[ns].pci_dev = pci_dev;
+                       list[ns].pf_bond = bd;
+                       list[ns].ifindex = mlx5_nl_ifindex
+                                       (nl_rdma, list[ns].ibv_dev->name, i);
+                       if (!list[ns].ifindex) {
+                               /*
+                                * No network interface index found for the
+                                * specified port, it means there is no
+                                * representor on this port. It's OK,
+                                * there can be disabled ports, for example
+                                * if sriov_numvfs < sriov_totalvfs.
+                                */
+                               continue;
+                       }
+                       ret = -1;
+                       if (nl_route >= 0)
+                               ret = mlx5_nl_switch_info
+                                              (nl_route,
+                                               list[ns].ifindex,
+                                               &list[ns].info);
+                       if (ret || (!list[ns].info.representor &&
+                                   !list[ns].info.master)) {
+                               /*
+                                * We failed to recognize representors with
+                                * Netlink, let's try to perform the task
+                                * with sysfs.
+                                */
+                               ret =  mlx5_sysfs_switch_info
+                                               (list[ns].ifindex,
+                                                &list[ns].info);
+                       }
+                       if (!ret && bd >= 0) {
+                               switch (list[ns].info.name_type) {
+                               case MLX5_PHYS_PORT_NAME_TYPE_UPLINK:
+                                       if (list[ns].info.port_name == bd)
+                                               ns++;
+                                       break;
+                               case MLX5_PHYS_PORT_NAME_TYPE_PFVF:
+                                       if (list[ns].info.pf_num == bd)
+                                               ns++;
+                                       break;
+                               default:
+                                       break;
+                               }
+                               continue;
+                       }
+                       if (!ret && (list[ns].info.representor ^
+                                    list[ns].info.master))
+                               ns++;
+               }
+               if (!ns) {
+                       DRV_LOG(ERR,
+                               "unable to recognize master/representors"
+                               " on the IB device with multiple ports");
+                       rte_errno = ENOENT;
+                       ret = -rte_errno;
+                       goto exit;
+               }
+       } else {
+               /*
+                * The existence of several matching entries (nd > 1) means
+                * port representors have been instantiated. No existing Verbs
+                * call nor sysfs entries can tell them apart, this can only
+                * be done through Netlink calls assuming kernel drivers are
+                * recent enough to support them.
+                *
+                * In the event of identification failure through Netlink,
+                * try again through sysfs, then:
+                *
+                * 1. A single IB device matches (nd == 1) with single
+                *    port (np=0/1) and is not a representor, assume
+                *    no switch support.
+                *
+                * 2. Otherwise no safe assumptions can be made;
+                *    complain louder and bail out.
+                */
+               for (i = 0; i != nd; ++i) {
+                       memset(&list[ns].info, 0, sizeof(list[ns].info));
+                       list[ns].max_port = 1;
+                       list[ns].ibv_port = 1;
+                       list[ns].ibv_dev = ibv_match[i];
+                       list[ns].eth_dev = NULL;
+                       list[ns].pci_dev = pci_dev;
+                       list[ns].pf_bond = -1;
+                       list[ns].ifindex = 0;
+                       if (nl_rdma >= 0)
+                               list[ns].ifindex = mlx5_nl_ifindex
+                                       (nl_rdma, list[ns].ibv_dev->name, 1);
+                       if (!list[ns].ifindex) {
+                               char ifname[IF_NAMESIZE];
+
+                               /*
+                                * Netlink failed, it may happen with old
+                                * ib_core kernel driver (before 4.16).
+                                * We can assume there is old driver because
+                                * here we are processing single ports IB
+                                * devices. Let's try sysfs to retrieve
+                                * the ifindex. The method works for
+                                * master device only.
+                                */
+                               if (nd > 1) {
+                                       /*
+                                        * Multiple devices found, assume
+                                        * representors, can not distinguish
+                                        * master/representor and retrieve
+                                        * ifindex via sysfs.
+                                        */
+                                       continue;
+                               }
+                               ret = mlx5_get_master_ifname
+                                       (ibv_match[i]->ibdev_path, &ifname);
+                               if (!ret)
+                                       list[ns].ifindex =
+                                               if_nametoindex(ifname);
+                               if (!list[ns].ifindex) {
+                                       /*
+                                        * No network interface index found
+                                        * for the specified device, it means
+                                        * there it is neither representor
+                                        * nor master.
+                                        */
+                                       continue;
+                               }
+                       }
+                       ret = -1;
+                       if (nl_route >= 0)
+                               ret = mlx5_nl_switch_info
+                                              (nl_route,
+                                               list[ns].ifindex,
+                                               &list[ns].info);
+                       if (ret || (!list[ns].info.representor &&
+                                   !list[ns].info.master)) {
+                               /*
+                                * We failed to recognize representors with
+                                * Netlink, let's try to perform the task
+                                * with sysfs.
+                                */
+                               ret =  mlx5_sysfs_switch_info
+                                               (list[ns].ifindex,
+                                                &list[ns].info);
+                       }
+                       if (!ret && (list[ns].info.representor ^
+                                    list[ns].info.master)) {
+                               ns++;
+                       } else if ((nd == 1) &&
+                                  !list[ns].info.representor &&
+                                  !list[ns].info.master) {
+                               /*
+                                * Single IB device with
+                                * one physical port and
+                                * attached network device.
+                                * May be SRIOV is not enabled
+                                * or there is no representors.
+                                */
+                               DRV_LOG(INFO, "no E-Switch support detected");
+                               ns++;
+                               break;
+                       }
+               }
+               if (!ns) {
+                       DRV_LOG(ERR,
+                               "unable to recognize master/representors"
+                               " on the multiple IB devices");
+                       rte_errno = ENOENT;
+                       ret = -rte_errno;
+                       goto exit;
+               }
+       }
+       MLX5_ASSERT(ns);
+       /*
+        * Sort list to probe devices in natural order for users convenience
+        * (i.e. master first, then representors from lowest to highest ID).
+        */
+       qsort(list, ns, sizeof(*list), mlx5_dev_spawn_data_cmp);
+       /* Default configuration. */
+       dev_config = (struct mlx5_dev_config){
+               .hw_padding = 0,
+               .mps = MLX5_ARG_UNSET,
+               .dbnc = MLX5_ARG_UNSET,
+               .rx_vec_en = 1,
+               .txq_inline_max = MLX5_ARG_UNSET,
+               .txq_inline_min = MLX5_ARG_UNSET,
+               .txq_inline_mpw = MLX5_ARG_UNSET,
+               .txqs_inline = MLX5_ARG_UNSET,
+               .vf_nl_en = 1,
+               .mr_ext_memseg_en = 1,
+               .mprq = {
+                       .enabled = 0, /* Disabled by default. */
+                       .stride_num_n = 0,
+                       .stride_size_n = 0,
+                       .max_memcpy_len = MLX5_MPRQ_MEMCPY_DEFAULT_LEN,
+                       .min_rxqs_num = MLX5_MPRQ_MIN_RXQS,
+               },
+               .dv_esw_en = 1,
+               .dv_flow_en = 1,
+               .log_hp_size = MLX5_ARG_UNSET,
+       };
+       /* Device specific configuration. */
+       switch (pci_dev->id.device_id) {
+       case PCI_DEVICE_ID_MELLANOX_CONNECTX4VF:
+       case PCI_DEVICE_ID_MELLANOX_CONNECTX4LXVF:
+       case PCI_DEVICE_ID_MELLANOX_CONNECTX5VF:
+       case PCI_DEVICE_ID_MELLANOX_CONNECTX5EXVF:
+       case PCI_DEVICE_ID_MELLANOX_CONNECTX5BFVF:
+       case PCI_DEVICE_ID_MELLANOX_CONNECTX6VF:
+       case PCI_DEVICE_ID_MELLANOX_CONNECTX6DXVF:
+               dev_config.vf = 1;
+               break;
+       default:
+               break;
+       }
+       for (i = 0; i != ns; ++i) {
+               uint32_t restore;
+
+               list[i].eth_dev = mlx5_dev_spawn(&pci_dev->device,
+                                                &list[i],
+                                                dev_config);
+               if (!list[i].eth_dev) {
+                       if (rte_errno != EBUSY && rte_errno != EEXIST)
+                               break;
+                       /* Device is disabled or already spawned. Ignore it. */
+                       continue;
+               }
+               restore = list[i].eth_dev->data->dev_flags;
+               rte_eth_copy_pci_info(list[i].eth_dev, pci_dev);
+               /* Restore non-PCI flags cleared by the above call. */
+               list[i].eth_dev->data->dev_flags |= restore;
+               rte_eth_dev_probing_finish(list[i].eth_dev);
+       }
+       if (i != ns) {
+               DRV_LOG(ERR,
+                       "probe of PCI device " PCI_PRI_FMT " aborted after"
+                       " encountering an error: %s",
+                       pci_dev->addr.domain, pci_dev->addr.bus,
+                       pci_dev->addr.devid, pci_dev->addr.function,
+                       strerror(rte_errno));
+               ret = -rte_errno;
+               /* Roll back. */
+               while (i--) {
+                       if (!list[i].eth_dev)
+                               continue;
+                       mlx5_dev_close(list[i].eth_dev);
+                       /* mac_addrs must not be freed because in dev_private */
+                       list[i].eth_dev->data->mac_addrs = NULL;
+                       claim_zero(rte_eth_dev_release_port(list[i].eth_dev));
+               }
+               /* Restore original error. */
+               rte_errno = -ret;
+       } else {
+               ret = 0;
+       }
+exit:
+       /*
+        * Do the routine cleanup:
+        * - close opened Netlink sockets
+        * - free allocated spawn data array
+        * - free the Infiniband device list
+        */
+       if (nl_rdma >= 0)
+               close(nl_rdma);
+       if (nl_route >= 0)
+               close(nl_route);
+       if (list)
+               rte_free(list);
+       MLX5_ASSERT(ibv_list);
+       mlx5_glue->free_device_list(ibv_list);
+       return ret;
+}
+
+static int
+mlx5_config_doorbell_mapping_env(const struct mlx5_dev_config *config)
+{
+       char *env;
+       int value;
+
+       MLX5_ASSERT(rte_eal_process_type() == RTE_PROC_PRIMARY);
+       /* Get environment variable to store. */
+       env = getenv(MLX5_SHUT_UP_BF);
+       value = env ? !!strcmp(env, "0") : MLX5_ARG_UNSET;
+       if (config->dbnc == MLX5_ARG_UNSET)
+               setenv(MLX5_SHUT_UP_BF, MLX5_SHUT_UP_BF_DEFAULT, 1);
+       else
+               setenv(MLX5_SHUT_UP_BF,
+                      config->dbnc == MLX5_TXDB_NCACHED ? "1" : "0", 1);
+       return value;
+}
+
+static void
+mlx5_restore_doorbell_mapping_env(int value)
+{
+       MLX5_ASSERT(rte_eal_process_type() == RTE_PROC_PRIMARY);
+       /* Restore the original environment variable state. */
+       if (value == MLX5_ARG_UNSET)
+               unsetenv(MLX5_SHUT_UP_BF);
+       else
+               setenv(MLX5_SHUT_UP_BF, value ? "1" : "0", 1);
+}
+
+/**
+ * Extract pdn of PD object using DV API.
+ *
+ * @param[in] pd
+ *   Pointer to the verbs PD object.
+ * @param[out] pdn
+ *   Pointer to the PD object number variable.
+ *
+ * @return
+ *   0 on success, error value otherwise.
+ */
+int
+mlx5_os_get_pdn(void *pd, uint32_t *pdn)
+{
+#ifdef HAVE_IBV_FLOW_DV_SUPPORT
+       struct mlx5dv_obj obj;
+       struct mlx5dv_pd pd_info;
+       int ret = 0;
+
+       obj.pd.in = pd;
+       obj.pd.out = &pd_info;
+       ret = mlx5_glue->dv_init_obj(&obj, MLX5DV_OBJ_PD);
+       if (ret) {
+               DRV_LOG(DEBUG, "Fail to get PD object info");
+               return ret;
+       }
+       *pdn = pd_info.pdn;
+       return 0;
+#else
+       (void)pd;
+       (void)pdn;
+       return -ENOTSUP;
+#endif /* HAVE_IBV_FLOW_DV_SUPPORT */
+}
+
+/**
+ * Function API to open IB device.
+ *
+ * This function calls the Linux glue APIs to open a device.
+ *
+ * @param[in] spawn
+ *   Pointer to the IB device attributes (name, port, etc).
+ * @param[out] config
+ *   Pointer to device configuration structure.
+ * @param[out] sh
+ *   Pointer to shared context structure.
+ *
+ * @return
+ *   0 on success, a positive error value otherwise.
+ */
+int
+mlx5_os_open_device(const struct mlx5_dev_spawn_data *spawn,
+                    const struct mlx5_dev_config *config,
+                    struct mlx5_dev_ctx_shared *sh)
+{
+       int dbmap_env;
+       int err = 0;
+       /*
+        * Configure environment variable "MLX5_BF_SHUT_UP"
+        * before the device creation. The rdma_core library
+        * checks the variable at device creation and
+        * stores the result internally.
+        */
+       dbmap_env = mlx5_config_doorbell_mapping_env(config);
+       /* Try to open IB device with DV first, then usual Verbs. */
+       errno = 0;
+       sh->ctx = mlx5_glue->dv_open_device(spawn->ibv_dev);
+       if (sh->ctx) {
+               sh->devx = 1;
+               DRV_LOG(DEBUG, "DevX is supported");
+               /* The device is created, no need for environment. */
+               mlx5_restore_doorbell_mapping_env(dbmap_env);
+       } else {
+               /* The environment variable is still configured. */
+               sh->ctx = mlx5_glue->open_device(spawn->ibv_dev);
+               err = errno ? errno : ENODEV;
+               /*
+                * The environment variable is not needed anymore,
+                * all device creation attempts are completed.
+                */
+               mlx5_restore_doorbell_mapping_env(dbmap_env);
+               if (!sh->ctx)
+                       return err;
+               DRV_LOG(DEBUG, "DevX is NOT supported");
+               err = 0;
+       }
+       return err;
+}
+
+/**
+ * Install shared asynchronous device events handler.
+ * This function is implemented to support event sharing
+ * between multiple ports of single IB device.
+ *
+ * @param sh
+ *   Pointer to mlx5_dev_ctx_shared object.
+ */
+void
+mlx5_os_dev_shared_handler_install(struct mlx5_dev_ctx_shared *sh)
+{
+       int ret;
+       int flags;
+
+       sh->intr_handle.fd = -1;
+       flags = fcntl(((struct ibv_context *)sh->ctx)->async_fd, F_GETFL);
+       ret = fcntl(((struct ibv_context *)sh->ctx)->async_fd,
+                   F_SETFL, flags | O_NONBLOCK);
+       if (ret) {
+               DRV_LOG(INFO, "failed to change file descriptor async event"
+                       " queue");
+       } else {
+               sh->intr_handle.fd = ((struct ibv_context *)sh->ctx)->async_fd;
+               sh->intr_handle.type = RTE_INTR_HANDLE_EXT;
+               if (rte_intr_callback_register(&sh->intr_handle,
+                                       mlx5_dev_interrupt_handler, sh)) {
+                       DRV_LOG(INFO, "Fail to install the shared interrupt.");
+                       sh->intr_handle.fd = -1;
+               }
+       }
+       if (sh->devx) {
+#ifdef HAVE_IBV_DEVX_ASYNC
+               sh->intr_handle_devx.fd = -1;
+               sh->devx_comp = mlx5_glue->devx_create_cmd_comp(sh->ctx);
+               if (!sh->devx_comp) {
+                       DRV_LOG(INFO, "failed to allocate devx_comp.");
+                       return;
+               }
+               flags = fcntl(sh->devx_comp->fd, F_GETFL);
+               ret = fcntl(sh->devx_comp->fd, F_SETFL, flags | O_NONBLOCK);
+               if (ret) {
+                       DRV_LOG(INFO, "failed to change file descriptor"
+                               " devx comp");
+                       return;
+               }
+               sh->intr_handle_devx.fd = sh->devx_comp->fd;
+               sh->intr_handle_devx.type = RTE_INTR_HANDLE_EXT;
+               if (rte_intr_callback_register(&sh->intr_handle_devx,
+                                       mlx5_dev_interrupt_handler_devx, sh)) {
+                       DRV_LOG(INFO, "Fail to install the devx shared"
+                               " interrupt.");
+                       sh->intr_handle_devx.fd = -1;
+               }
+#endif /* HAVE_IBV_DEVX_ASYNC */
+       }
+}
+
+/**
+ * Uninstall shared asynchronous device events handler.
+ * This function is implemented to support event sharing
+ * between multiple ports of single IB device.
+ *
+ * @param dev
+ *   Pointer to mlx5_dev_ctx_shared object.
+ */
+void
+mlx5_os_dev_shared_handler_uninstall(struct mlx5_dev_ctx_shared *sh)
+{
+       if (sh->intr_handle.fd >= 0)
+               mlx5_intr_callback_unregister(&sh->intr_handle,
+                                             mlx5_dev_interrupt_handler, sh);
+#ifdef HAVE_IBV_DEVX_ASYNC
+       if (sh->intr_handle_devx.fd >= 0)
+               rte_intr_callback_unregister(&sh->intr_handle_devx,
+                                 mlx5_dev_interrupt_handler_devx, sh);
+       if (sh->devx_comp)
+               mlx5_glue->devx_destroy_cmd_comp(sh->devx_comp);
+#endif
+}
index 4f7b4d3..f62ad12 100644 (file)
@@ -10,7 +10,6 @@
 #include <stdlib.h>
 #include <errno.h>
 #include <net/if.h>
-#include <fcntl.h>
 #include <sys/mman.h>
 #include <linux/rtnetlink.h>
 
 /* Flow memory reclaim mode. */
 #define MLX5_RECLAIM_MEM "reclaim_mem_mode"
 
-#ifndef HAVE_IBV_MLX5_MOD_MPW
-#define MLX5DV_CONTEXT_FLAGS_MPW_ALLOWED (1 << 2)
-#define MLX5DV_CONTEXT_FLAGS_ENHANCED_MPW (1 << 3)
-#endif
-
-#ifndef HAVE_IBV_MLX5_MOD_CQE_128B_COMP
-#define MLX5DV_CONTEXT_FLAGS_CQE_128B_COMP (1 << 4)
-#endif
-
 static const char *MZ_MLX5_PMD_SHARED_DATA = "mlx5_pmd_shared_data";
 
 /* Shared memory between primary and secondary processes. */
@@ -183,22 +173,9 @@ static rte_spinlock_t mlx5_shared_data_lock = RTE_SPINLOCK_INITIALIZER;
 
 /* Process local data for secondary processes. */
 static struct mlx5_local_data mlx5_local_data;
-
 /** Driver-specific log messages type. */
 int mlx5_logtype;
 
-/** Data associated with devices to spawn. */
-struct mlx5_dev_spawn_data {
-       uint32_t ifindex; /**< Network interface index. */
-       uint32_t max_port; /**< IB device maximal port index. */
-       uint32_t ibv_port; /**< IB device physical port index. */
-       int pf_bond; /**< bonding device PF index. < 0 - no bonding */
-       struct mlx5_switch_info info; /**< Switch information. */
-       struct ibv_device *ibv_dev; /**< Associated IB device. */
-       struct rte_eth_dev *eth_dev; /**< Associated Ethernet device. */
-       struct rte_pci_device *pci_dev; /**< Backend PCI device. */
-};
-
 static LIST_HEAD(, mlx5_dev_ctx_shared) mlx5_ibv_list = LIST_HEAD_INITIALIZER();
 static pthread_mutex_t mlx5_ibv_list_mutex = PTHREAD_MUTEX_INITIALIZER;
 
@@ -320,7 +297,6 @@ static struct mlx5_indexed_pool_config mlx5_ipool_cfg[] = {
 #define MLX5_ID_GENERATION_ARRAY_FACTOR 16
 
 #define MLX5_FLOW_TABLE_HLIST_ARRAY_SIZE 4096
-#define MLX5_TAGS_HLIST_ARRAY_SIZE 8192
 
 /**
  * Allocate ID pool structure.
@@ -602,146 +578,6 @@ mlx5_flow_ipool_destroy(struct mlx5_dev_ctx_shared *sh)
                mlx5_ipool_destroy(sh->ipool[i]);
 }
 
-/**
- * Extract pdn of PD object using DV API.
- *
- * @param[in] pd
- *   Pointer to the verbs PD object.
- * @param[out] pdn
- *   Pointer to the PD object number variable.
- *
- * @return
- *   0 on success, error value otherwise.
- */
-#ifdef HAVE_IBV_FLOW_DV_SUPPORT
-static int
-mlx5_get_pdn(struct ibv_pd *pd __rte_unused, uint32_t *pdn __rte_unused)
-{
-       struct mlx5dv_obj obj;
-       struct mlx5dv_pd pd_info;
-       int ret = 0;
-
-       obj.pd.in = pd;
-       obj.pd.out = &pd_info;
-       ret = mlx5_glue->dv_init_obj(&obj, MLX5DV_OBJ_PD);
-       if (ret) {
-               DRV_LOG(DEBUG, "Fail to get PD object info");
-               return ret;
-       }
-       *pdn = pd_info.pdn;
-       return 0;
-}
-#endif /* HAVE_IBV_FLOW_DV_SUPPORT */
-
-static int
-mlx5_config_doorbell_mapping_env(const struct mlx5_dev_config *config)
-{
-       char *env;
-       int value;
-
-       MLX5_ASSERT(rte_eal_process_type() == RTE_PROC_PRIMARY);
-       /* Get environment variable to store. */
-       env = getenv(MLX5_SHUT_UP_BF);
-       value = env ? !!strcmp(env, "0") : MLX5_ARG_UNSET;
-       if (config->dbnc == MLX5_ARG_UNSET)
-               setenv(MLX5_SHUT_UP_BF, MLX5_SHUT_UP_BF_DEFAULT, 1);
-       else
-               setenv(MLX5_SHUT_UP_BF,
-                      config->dbnc == MLX5_TXDB_NCACHED ? "1" : "0", 1);
-       return value;
-}
-
-static void
-mlx5_restore_doorbell_mapping_env(int value)
-{
-       MLX5_ASSERT(rte_eal_process_type() == RTE_PROC_PRIMARY);
-       /* Restore the original environment variable state. */
-       if (value == MLX5_ARG_UNSET)
-               unsetenv(MLX5_SHUT_UP_BF);
-       else
-               setenv(MLX5_SHUT_UP_BF, value ? "1" : "0", 1);
-}
-
-/**
- * Install shared asynchronous device events handler.
- * This function is implemented to support event sharing
- * between multiple ports of single IB device.
- *
- * @param sh
- *   Pointer to mlx5_dev_ctx_shared object.
- */
-static void
-mlx5_dev_shared_handler_install(struct mlx5_dev_ctx_shared *sh)
-{
-       int ret;
-       int flags;
-
-       sh->intr_handle.fd = -1;
-       flags = fcntl(((struct ibv_context *)sh->ctx)->async_fd, F_GETFL);
-       ret = fcntl(((struct ibv_context *)sh->ctx)->async_fd,
-                   F_SETFL, flags | O_NONBLOCK);
-       if (ret) {
-               DRV_LOG(INFO, "failed to change file descriptor async event"
-                       " queue");
-       } else {
-               sh->intr_handle.fd = ((struct ibv_context *)sh->ctx)->async_fd;
-               sh->intr_handle.type = RTE_INTR_HANDLE_EXT;
-               if (rte_intr_callback_register(&sh->intr_handle,
-                                       mlx5_dev_interrupt_handler, sh)) {
-                       DRV_LOG(INFO, "Fail to install the shared interrupt.");
-                       sh->intr_handle.fd = -1;
-               }
-       }
-       if (sh->devx) {
-#ifdef HAVE_IBV_DEVX_ASYNC
-               sh->intr_handle_devx.fd = -1;
-               sh->devx_comp = mlx5_glue->devx_create_cmd_comp(sh->ctx);
-               if (!sh->devx_comp) {
-                       DRV_LOG(INFO, "failed to allocate devx_comp.");
-                       return;
-               }
-               flags = fcntl(sh->devx_comp->fd, F_GETFL);
-               ret = fcntl(sh->devx_comp->fd, F_SETFL, flags | O_NONBLOCK);
-               if (ret) {
-                       DRV_LOG(INFO, "failed to change file descriptor"
-                               " devx comp");
-                       return;
-               }
-               sh->intr_handle_devx.fd = sh->devx_comp->fd;
-               sh->intr_handle_devx.type = RTE_INTR_HANDLE_EXT;
-               if (rte_intr_callback_register(&sh->intr_handle_devx,
-                                       mlx5_dev_interrupt_handler_devx, sh)) {
-                       DRV_LOG(INFO, "Fail to install the devx shared"
-                               " interrupt.");
-                       sh->intr_handle_devx.fd = -1;
-               }
-#endif /* HAVE_IBV_DEVX_ASYNC */
-       }
-}
-
-/**
- * Uninstall shared asynchronous device events handler.
- * This function is implemented to support event sharing
- * between multiple ports of single IB device.
- *
- * @param dev
- *   Pointer to mlx5_dev_ctx_shared object.
- */
-static void
-mlx5_dev_shared_handler_uninstall(struct mlx5_dev_ctx_shared *sh)
-{
-       if (sh->intr_handle.fd >= 0)
-               mlx5_intr_callback_unregister(&sh->intr_handle,
-                                             mlx5_dev_interrupt_handler, sh);
-#ifdef HAVE_IBV_DEVX_ASYNC
-       if (sh->intr_handle_devx.fd >= 0)
-               rte_intr_callback_unregister(&sh->intr_handle_devx,
-                                 mlx5_dev_interrupt_handler_devx, sh);
-       if (sh->devx_comp)
-               mlx5_glue->devx_destroy_cmd_comp(sh->devx_comp);
-#endif
-}
-
 /**
  * Allocate shared IB device context. If there is multiport device the
  * master and representors will share this context, if there is single
@@ -762,17 +598,14 @@ mlx5_dev_shared_handler_uninstall(struct mlx5_dev_ctx_shared *sh)
  *   Pointer to mlx5_dev_ctx_shared object on success,
  *   otherwise NULL and rte_errno is set.
  */
-static struct mlx5_dev_ctx_shared *
+struct mlx5_dev_ctx_shared *
 mlx5_alloc_shared_ibctx(const struct mlx5_dev_spawn_data *spawn,
                        const struct mlx5_dev_config *config)
 {
        struct mlx5_dev_ctx_shared *sh;
-       int dbmap_env;
        int err = 0;
        uint32_t i;
-#ifdef HAVE_IBV_FLOW_DV_SUPPORT
        struct mlx5_devx_tis_attr tis_attr = { 0 };
-#endif
 
        MLX5_ASSERT(spawn);
        /* Secondary process should not create the shared context. */
@@ -797,34 +630,9 @@ mlx5_alloc_shared_ibctx(const struct mlx5_dev_spawn_data *spawn,
                rte_errno  = ENOMEM;
                goto exit;
        }
-       /*
-        * Configure environment variable "MLX5_BF_SHUT_UP"
-        * before the device creation. The rdma_core library
-        * checks the variable at device creation and
-        * stores the result internally.
-        */
-       dbmap_env = mlx5_config_doorbell_mapping_env(config);
-       /* Try to open IB device with DV first, then usual Verbs. */
-       errno = 0;
-       sh->ctx = mlx5_glue->dv_open_device(spawn->ibv_dev);
-       if (sh->ctx) {
-               sh->devx = 1;
-               DRV_LOG(DEBUG, "DevX is supported");
-               /* The device is created, no need for environment. */
-               mlx5_restore_doorbell_mapping_env(dbmap_env);
-       } else {
-               /* The environment variable is still configured. */
-               sh->ctx = mlx5_glue->open_device(spawn->ibv_dev);
-               err = errno ? errno : ENODEV;
-               /*
-                * The environment variable is not needed anymore,
-                * all device creation attempts are completed.
-                */
-               mlx5_restore_doorbell_mapping_env(dbmap_env);
-               if (!sh->ctx)
-                       goto error;
-               DRV_LOG(DEBUG, "DevX is NOT supported");
-       }
+       err = mlx5_os_open_device(spawn, config, sh);
+       if (!sh->ctx)
+               goto error;
        err = mlx5_os_get_dev_attr(sh->ctx, &sh->device_attr);
        if (err) {
                DRV_LOG(DEBUG, "mlx5_os_get_dev_attr() failed");
@@ -851,9 +659,8 @@ mlx5_alloc_shared_ibctx(const struct mlx5_dev_spawn_data *spawn,
                err = ENOMEM;
                goto error;
        }
-#ifdef HAVE_IBV_FLOW_DV_SUPPORT
        if (sh->devx) {
-               err = mlx5_get_pdn(sh->pd, &sh->pdn);
+               err = mlx5_os_get_pdn(sh->pd, &sh->pdn);
                if (err) {
                        DRV_LOG(ERR, "Fail to extract pdn from PD");
                        goto error;
@@ -879,7 +686,6 @@ mlx5_alloc_shared_ibctx(const struct mlx5_dev_spawn_data *spawn,
                err = ENOMEM;
                goto error;
        }
-#endif /* HAVE_IBV_FLOW_DV_SUPPORT */
        /*
         * Once the device is added to the list of memory event
         * callback, its global MR cache table cannot be expanded
@@ -896,7 +702,7 @@ mlx5_alloc_shared_ibctx(const struct mlx5_dev_spawn_data *spawn,
                err = rte_errno;
                goto error;
        }
-       mlx5_dev_shared_handler_install(sh);
+       mlx5_os_dev_shared_handler_install(sh);
        mlx5_flow_aging_init(sh);
        mlx5_flow_counters_mng_init(sh);
        mlx5_flow_ipool_create(sh, config);
@@ -936,7 +742,7 @@ error:
  * @param[in] sh
  *   Pointer to mlx5_dev_ctx_shared object to free
  */
-static void
+void
 mlx5_free_shared_ibctx(struct mlx5_dev_ctx_shared *sh)
 {
        pthread_mutex_lock(&mlx5_ibv_list_mutex);
@@ -973,7 +779,7 @@ mlx5_free_shared_ibctx(struct mlx5_dev_ctx_shared *sh)
         **/
        mlx5_flow_counters_mng_close(sh);
        mlx5_flow_ipool_destroy(sh);
-       mlx5_dev_shared_handler_uninstall(sh);
+       mlx5_os_dev_shared_handler_uninstall(sh);
        if (sh->pd)
                claim_zero(mlx5_glue->dealloc_pd(sh->pd));
        if (sh->tis)
@@ -995,7 +801,7 @@ exit:
  * @param[in] priv
  *   Pointer to the private device data structure.
  */
-static void
+void
 mlx5_free_table_hash_list(struct mlx5_priv *priv)
 {
        struct mlx5_dev_ctx_shared *sh = priv->sh;
@@ -1052,7 +858,7 @@ mlx5_free_table_hash_list(struct mlx5_priv *priv)
  * @return
  *   Zero on success, positive error code otherwise.
  */
-static int
+int
 mlx5_alloc_table_hash_list(struct mlx5_priv *priv)
 {
        struct mlx5_dev_ctx_shared *sh = priv->sh;
@@ -1126,173 +932,6 @@ error:
        return err;
 }
 
-/**
- * Initialize DR related data within private structure.
- * Routine checks the reference counter and does actual
- * resources creation/initialization only if counter is zero.
- *
- * @param[in] priv
- *   Pointer to the private device data structure.
- *
- * @return
- *   Zero on success, positive error code otherwise.
- */
-static int
-mlx5_alloc_shared_dr(struct mlx5_priv *priv)
-{
-       struct mlx5_dev_ctx_shared *sh = priv->sh;
-       char s[MLX5_HLIST_NAMESIZE];
-       int err = 0;
-
-       if (!sh->flow_tbls)
-               err = mlx5_alloc_table_hash_list(priv);
-       else
-               DRV_LOG(DEBUG, "sh->flow_tbls[%p] already created, reuse\n",
-                       (void *)sh->flow_tbls);
-       if (err)
-               return err;
-       /* Create tags hash list table. */
-       snprintf(s, sizeof(s), "%s_tags", sh->ibdev_name);
-       sh->tag_table = mlx5_hlist_create(s, MLX5_TAGS_HLIST_ARRAY_SIZE);
-       if (!sh->tag_table) {
-               DRV_LOG(ERR, "tags with hash creation failed.\n");
-               err = ENOMEM;
-               goto error;
-       }
-#ifdef HAVE_MLX5DV_DR
-       void *domain;
-
-       if (sh->dv_refcnt) {
-               /* Shared DV/DR structures is already initialized. */
-               sh->dv_refcnt++;
-               priv->dr_shared = 1;
-               return 0;
-       }
-       /* Reference counter is zero, we should initialize structures. */
-       domain = mlx5_glue->dr_create_domain(sh->ctx,
-                                            MLX5DV_DR_DOMAIN_TYPE_NIC_RX);
-       if (!domain) {
-               DRV_LOG(ERR, "ingress mlx5dv_dr_create_domain failed");
-               err = errno;
-               goto error;
-       }
-       sh->rx_domain = domain;
-       domain = mlx5_glue->dr_create_domain(sh->ctx,
-                                            MLX5DV_DR_DOMAIN_TYPE_NIC_TX);
-       if (!domain) {
-               DRV_LOG(ERR, "egress mlx5dv_dr_create_domain failed");
-               err = errno;
-               goto error;
-       }
-       pthread_mutex_init(&sh->dv_mutex, NULL);
-       sh->tx_domain = domain;
-#ifdef HAVE_MLX5DV_DR_ESWITCH
-       if (priv->config.dv_esw_en) {
-               domain  = mlx5_glue->dr_create_domain
-                       (sh->ctx, MLX5DV_DR_DOMAIN_TYPE_FDB);
-               if (!domain) {
-                       DRV_LOG(ERR, "FDB mlx5dv_dr_create_domain failed");
-                       err = errno;
-                       goto error;
-               }
-               sh->fdb_domain = domain;
-               sh->esw_drop_action = mlx5_glue->dr_create_flow_action_drop();
-       }
-#endif
-       if (priv->config.reclaim_mode == MLX5_RCM_AGGR) {
-               mlx5_glue->dr_reclaim_domain_memory(sh->rx_domain, 1);
-               mlx5_glue->dr_reclaim_domain_memory(sh->tx_domain, 1);
-               if (sh->fdb_domain)
-                       mlx5_glue->dr_reclaim_domain_memory(sh->fdb_domain, 1);
-       }
-       sh->pop_vlan_action = mlx5_glue->dr_create_flow_action_pop_vlan();
-#endif /* HAVE_MLX5DV_DR */
-       sh->dv_refcnt++;
-       priv->dr_shared = 1;
-       return 0;
-error:
-       /* Rollback the created objects. */
-       if (sh->rx_domain) {
-               mlx5_glue->dr_destroy_domain(sh->rx_domain);
-               sh->rx_domain = NULL;
-       }
-       if (sh->tx_domain) {
-               mlx5_glue->dr_destroy_domain(sh->tx_domain);
-               sh->tx_domain = NULL;
-       }
-       if (sh->fdb_domain) {
-               mlx5_glue->dr_destroy_domain(sh->fdb_domain);
-               sh->fdb_domain = NULL;
-       }
-       if (sh->esw_drop_action) {
-               mlx5_glue->destroy_flow_action(sh->esw_drop_action);
-               sh->esw_drop_action = NULL;
-       }
-       if (sh->pop_vlan_action) {
-               mlx5_glue->destroy_flow_action(sh->pop_vlan_action);
-               sh->pop_vlan_action = NULL;
-       }
-       if (sh->tag_table) {
-               /* tags should be destroyed with flow before. */
-               mlx5_hlist_destroy(sh->tag_table, NULL, NULL);
-               sh->tag_table = NULL;
-       }
-       mlx5_free_table_hash_list(priv);
-       return err;
-}
-
-/**
- * Destroy DR related data within private structure.
- *
- * @param[in] priv
- *   Pointer to the private device data structure.
- */
-static void
-mlx5_free_shared_dr(struct mlx5_priv *priv)
-{
-       struct mlx5_dev_ctx_shared *sh;
-
-       if (!priv->dr_shared)
-               return;
-       priv->dr_shared = 0;
-       sh = priv->sh;
-       MLX5_ASSERT(sh);
-#ifdef HAVE_MLX5DV_DR
-       MLX5_ASSERT(sh->dv_refcnt);
-       if (sh->dv_refcnt && --sh->dv_refcnt)
-               return;
-       if (sh->rx_domain) {
-               mlx5_glue->dr_destroy_domain(sh->rx_domain);
-               sh->rx_domain = NULL;
-       }
-       if (sh->tx_domain) {
-               mlx5_glue->dr_destroy_domain(sh->tx_domain);
-               sh->tx_domain = NULL;
-       }
-#ifdef HAVE_MLX5DV_DR_ESWITCH
-       if (sh->fdb_domain) {
-               mlx5_glue->dr_destroy_domain(sh->fdb_domain);
-               sh->fdb_domain = NULL;
-       }
-       if (sh->esw_drop_action) {
-               mlx5_glue->destroy_flow_action(sh->esw_drop_action);
-               sh->esw_drop_action = NULL;
-       }
-#endif
-       if (sh->pop_vlan_action) {
-               mlx5_glue->destroy_flow_action(sh->pop_vlan_action);
-               sh->pop_vlan_action = NULL;
-       }
-       pthread_mutex_destroy(&sh->dv_mutex);
-#endif /* HAVE_MLX5DV_DR */
-       if (sh->tag_table) {
-               /* tags should be destroyed with flow before. */
-               mlx5_hlist_destroy(sh->tag_table, NULL, NULL);
-               sh->tag_table = NULL;
-       }
-       mlx5_free_table_hash_list(priv);
-}
-
 /**
  * Initialize shared data between primary and secondary process.
  *
@@ -1361,60 +1000,6 @@ mlx5_getenv_int(const char *name)
        return atoi(val);
 }
 
-/**
- * Verbs callback to allocate a memory. This function should allocate the space
- * according to the size provided residing inside a huge page.
- * Please note that all allocation must respect the alignment from libmlx5
- * (i.e. currently sysconf(_SC_PAGESIZE)).
- *
- * @param[in] size
- *   The size in bytes of the memory to allocate.
- * @param[in] data
- *   A pointer to the callback data.
- *
- * @return
- *   Allocated buffer, NULL otherwise and rte_errno is set.
- */
-static void *
-mlx5_alloc_verbs_buf(size_t size, void *data)
-{
-       struct mlx5_priv *priv = data;
-       void *ret;
-       size_t alignment = sysconf(_SC_PAGESIZE);
-       unsigned int socket = SOCKET_ID_ANY;
-
-       if (priv->verbs_alloc_ctx.type == MLX5_VERBS_ALLOC_TYPE_TX_QUEUE) {
-               const struct mlx5_txq_ctrl *ctrl = priv->verbs_alloc_ctx.obj;
-
-               socket = ctrl->socket;
-       } else if (priv->verbs_alloc_ctx.type ==
-                  MLX5_VERBS_ALLOC_TYPE_RX_QUEUE) {
-               const struct mlx5_rxq_ctrl *ctrl = priv->verbs_alloc_ctx.obj;
-
-               socket = ctrl->socket;
-       }
-       MLX5_ASSERT(data != NULL);
-       ret = rte_malloc_socket(__func__, size, alignment, socket);
-       if (!ret && size)
-               rte_errno = ENOMEM;
-       return ret;
-}
-
-/**
- * Verbs callback to free a memory.
- *
- * @param[in] ptr
- *   A pointer to the memory to free.
- * @param[in] data
- *   A pointer to the callback data.
- */
-static void
-mlx5_free_verbs_buf(void *ptr, void *data __rte_unused)
-{
-       MLX5_ASSERT(data != NULL);
-       rte_free(ptr);
-}
-
 /**
  * DPDK callback to add udp tunnel port
  *
@@ -1496,7 +1081,7 @@ mlx5_proc_priv_uninit(struct rte_eth_dev *dev)
  * @param dev
  *   Pointer to Ethernet device structure.
  */
-static void
+void
 mlx5_dev_close(struct rte_eth_dev *dev)
 {
        struct mlx5_priv *priv = dev->data->dev_private;
@@ -1558,7 +1143,7 @@ mlx5_dev_close(struct rte_eth_dev *dev)
        if (priv->mreg_cp_tbl)
                mlx5_hlist_destroy(priv->mreg_cp_tbl, NULL, NULL);
        mlx5_mprq_free_mp(dev);
-       mlx5_free_shared_dr(priv);
+       mlx5_os_free_shared_dr(priv);
        if (priv->rss_conf.rss_key != NULL)
                rte_free(priv->rss_conf.rss_key);
        if (priv->reta_idx != NULL)
@@ -1696,7 +1281,7 @@ const struct eth_dev_ops mlx5_dev_ops = {
 };
 
 /* Available operations from secondary process. */
-static const struct eth_dev_ops mlx5_dev_sec_ops = {
+const struct eth_dev_ops mlx5_dev_sec_ops = {
        .stats_get = mlx5_stats_get,
        .stats_reset = mlx5_stats_reset,
        .xstats_get = mlx5_xstats_get,
@@ -1904,7 +1489,7 @@ mlx5_args_check(const char *key, const char *val, void *opaque)
  * @return
  *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
-static int
+int
 mlx5_args(struct mlx5_dev_config *config, struct rte_devargs *devargs)
 {
        const char **params = (const char *[]){
@@ -1970,8 +1555,6 @@ mlx5_args(struct mlx5_dev_config *config, struct rte_devargs *devargs)
        return 0;
 }
 
-static struct rte_pci_driver mlx5_driver;
-
 /**
  * PMD global initialization.
  *
@@ -1982,7 +1565,7 @@ static struct rte_pci_driver mlx5_driver;
  * @return
  *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
-static int
+int
 mlx5_init_once(void)
 {
        struct mlx5_shared_data *sd;
@@ -2042,7 +1625,7 @@ out:
  * @param config
  *   Device configuration parameters.
  */
-static void
+void
 mlx5_set_min_inline(struct mlx5_dev_spawn_data *spawn,
                    struct mlx5_dev_config *config)
 {
@@ -2152,7 +1735,7 @@ exit:
  * @param [in] dev
  *   Pointer to Ethernet device.
  */
-static void
+void
 mlx5_set_metadata_mask(struct rte_eth_dev *dev)
 {
        struct mlx5_priv *priv = dev->data->dev_private;
@@ -2340,21 +1923,20 @@ rte_pmd_mlx5_get_dyn_flag_names(char *names[], unsigned int n)
 }
 
 /**
- * Check sibling device configurations.
+ * Comparison callback to sort device data.
  *
- * Sibling devices sharing the Infiniband device context
- * should have compatible configurations. This regards
- * representors and bonding slaves.
+ * This is meant to be used with qsort().
  *
- * @param priv
- *   Private device descriptor.
- * @param config
- *   Configuration of the device is going to be created.
+ * @param a[in]
+ *   Pointer to pointer to first data object.
+ * @param b[in]
+ *   Pointer to pointer to second data object.
  *
  * @return
- *   0 on success, EINVAL otherwise
+ *   0 if both objects are equal, less than 0 if the first argument is less
+ *   than the second, greater than 0 otherwise.
  */
-static int
+int
 mlx5_dev_check_sibling_config(struct mlx5_priv *priv,
                              struct mlx5_dev_config *config)
 {
@@ -2392,1367 +1974,6 @@ mlx5_dev_check_sibling_config(struct mlx5_priv *priv,
        }
        return 0;
 }
-/**
- * Spawn an Ethernet device from Verbs information.
- *
- * @param dpdk_dev
- *   Backing DPDK device.
- * @param spawn
- *   Verbs device parameters (name, port, switch_info) to spawn.
- * @param config
- *   Device configuration parameters.
- *
- * @return
- *   A valid Ethernet device object on success, NULL otherwise and rte_errno
- *   is set. The following errors are defined:
- *
- *   EBUSY: device is not supposed to be spawned.
- *   EEXIST: device is already spawned
- */
-static struct rte_eth_dev *
-mlx5_dev_spawn(struct rte_device *dpdk_dev,
-              struct mlx5_dev_spawn_data *spawn,
-              struct mlx5_dev_config config)
-{
-       const struct mlx5_switch_info *switch_info = &spawn->info;
-       struct mlx5_dev_ctx_shared *sh = NULL;
-       struct ibv_port_attr port_attr;
-       struct mlx5dv_context dv_attr = { .comp_mask = 0 };
-       struct rte_eth_dev *eth_dev = NULL;
-       struct mlx5_priv *priv = NULL;
-       int err = 0;
-       unsigned int hw_padding = 0;
-       unsigned int mps;
-       unsigned int cqe_comp;
-       unsigned int cqe_pad = 0;
-       unsigned int tunnel_en = 0;
-       unsigned int mpls_en = 0;
-       unsigned int swp = 0;
-       unsigned int mprq = 0;
-       unsigned int mprq_min_stride_size_n = 0;
-       unsigned int mprq_max_stride_size_n = 0;
-       unsigned int mprq_min_stride_num_n = 0;
-       unsigned int mprq_max_stride_num_n = 0;
-       struct rte_ether_addr mac;
-       char name[RTE_ETH_NAME_MAX_LEN];
-       int own_domain_id = 0;
-       uint16_t port_id;
-       unsigned int i;
-#ifdef HAVE_MLX5DV_DR_DEVX_PORT
-       struct mlx5dv_devx_port devx_port = { .comp_mask = 0 };
-#endif
-
-       /* Determine if this port representor is supposed to be spawned. */
-       if (switch_info->representor && dpdk_dev->devargs) {
-               struct rte_eth_devargs eth_da;
-
-               err = rte_eth_devargs_parse(dpdk_dev->devargs->args, &eth_da);
-               if (err) {
-                       rte_errno = -err;
-                       DRV_LOG(ERR, "failed to process device arguments: %s",
-                               strerror(rte_errno));
-                       return NULL;
-               }
-               for (i = 0; i < eth_da.nb_representor_ports; ++i)
-                       if (eth_da.representor_ports[i] ==
-                           (uint16_t)switch_info->port_name)
-                               break;
-               if (i == eth_da.nb_representor_ports) {
-                       rte_errno = EBUSY;
-                       return NULL;
-               }
-       }
-       /* Build device name. */
-       if (spawn->pf_bond <  0) {
-               /* Single device. */
-               if (!switch_info->representor)
-                       strlcpy(name, dpdk_dev->name, sizeof(name));
-               else
-                       snprintf(name, sizeof(name), "%s_representor_%u",
-                                dpdk_dev->name, switch_info->port_name);
-       } else {
-               /* Bonding device. */
-               if (!switch_info->representor)
-                       snprintf(name, sizeof(name), "%s_%s",
-                                dpdk_dev->name, spawn->ibv_dev->name);
-               else
-                       snprintf(name, sizeof(name), "%s_%s_representor_%u",
-                                dpdk_dev->name, spawn->ibv_dev->name,
-                                switch_info->port_name);
-       }
-       /* check if the device is already spawned */
-       if (rte_eth_dev_get_port_by_name(name, &port_id) == 0) {
-               rte_errno = EEXIST;
-               return NULL;
-       }
-       DRV_LOG(DEBUG, "naming Ethernet device \"%s\"", name);
-       if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
-               struct mlx5_mp_id mp_id;
-
-               eth_dev = rte_eth_dev_attach_secondary(name);
-               if (eth_dev == NULL) {
-                       DRV_LOG(ERR, "can not attach rte ethdev");
-                       rte_errno = ENOMEM;
-                       return NULL;
-               }
-               eth_dev->device = dpdk_dev;
-               eth_dev->dev_ops = &mlx5_dev_sec_ops;
-               err = mlx5_proc_priv_init(eth_dev);
-               if (err)
-                       return NULL;
-               mp_id.port_id = eth_dev->data->port_id;
-               strlcpy(mp_id.name, MLX5_MP_NAME, RTE_MP_MAX_NAME_LEN);
-               /* Receive command fd from primary process */
-               err = mlx5_mp_req_verbs_cmd_fd(&mp_id);
-               if (err < 0)
-                       goto err_secondary;
-               /* Remap UAR for Tx queues. */
-               err = mlx5_tx_uar_init_secondary(eth_dev, err);
-               if (err)
-                       goto err_secondary;
-               /*
-                * Ethdev pointer is still required as input since
-                * the primary device is not accessible from the
-                * secondary process.
-                */
-               eth_dev->rx_pkt_burst = mlx5_select_rx_function(eth_dev);
-               eth_dev->tx_pkt_burst = mlx5_select_tx_function(eth_dev);
-               return eth_dev;
-err_secondary:
-               mlx5_dev_close(eth_dev);
-               return NULL;
-       }
-       /*
-        * Some parameters ("tx_db_nc" in particularly) are needed in
-        * advance to create dv/verbs device context. We proceed the
-        * devargs here to get ones, and later proceed devargs again
-        * to override some hardware settings.
-        */
-       err = mlx5_args(&config, dpdk_dev->devargs);
-       if (err) {
-               err = rte_errno;
-               DRV_LOG(ERR, "failed to process device arguments: %s",
-                       strerror(rte_errno));
-               goto error;
-       }
-       sh = mlx5_alloc_shared_ibctx(spawn, &config);
-       if (!sh)
-               return NULL;
-       config.devx = sh->devx;
-#ifdef HAVE_MLX5DV_DR_ACTION_DEST_DEVX_TIR
-       config.dest_tir = 1;
-#endif
-#ifdef HAVE_IBV_MLX5_MOD_SWP
-       dv_attr.comp_mask |= MLX5DV_CONTEXT_MASK_SWP;
-#endif
-       /*
-        * Multi-packet send is supported by ConnectX-4 Lx PF as well
-        * as all ConnectX-5 devices.
-        */
-#ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
-       dv_attr.comp_mask |= MLX5DV_CONTEXT_MASK_TUNNEL_OFFLOADS;
-#endif
-#ifdef HAVE_IBV_DEVICE_STRIDING_RQ_SUPPORT
-       dv_attr.comp_mask |= MLX5DV_CONTEXT_MASK_STRIDING_RQ;
-#endif
-       mlx5_glue->dv_query_device(sh->ctx, &dv_attr);
-       if (dv_attr.flags & MLX5DV_CONTEXT_FLAGS_MPW_ALLOWED) {
-               if (dv_attr.flags & MLX5DV_CONTEXT_FLAGS_ENHANCED_MPW) {
-                       DRV_LOG(DEBUG, "enhanced MPW is supported");
-                       mps = MLX5_MPW_ENHANCED;
-               } else {
-                       DRV_LOG(DEBUG, "MPW is supported");
-                       mps = MLX5_MPW;
-               }
-       } else {
-               DRV_LOG(DEBUG, "MPW isn't supported");
-               mps = MLX5_MPW_DISABLED;
-       }
-#ifdef HAVE_IBV_MLX5_MOD_SWP
-       if (dv_attr.comp_mask & MLX5DV_CONTEXT_MASK_SWP)
-               swp = dv_attr.sw_parsing_caps.sw_parsing_offloads;
-       DRV_LOG(DEBUG, "SWP support: %u", swp);
-#endif
-       config.swp = !!swp;
-#ifdef HAVE_IBV_DEVICE_STRIDING_RQ_SUPPORT
-       if (dv_attr.comp_mask & MLX5DV_CONTEXT_MASK_STRIDING_RQ) {
-               struct mlx5dv_striding_rq_caps mprq_caps =
-                       dv_attr.striding_rq_caps;
-
-               DRV_LOG(DEBUG, "\tmin_single_stride_log_num_of_bytes: %d",
-                       mprq_caps.min_single_stride_log_num_of_bytes);
-               DRV_LOG(DEBUG, "\tmax_single_stride_log_num_of_bytes: %d",
-                       mprq_caps.max_single_stride_log_num_of_bytes);
-               DRV_LOG(DEBUG, "\tmin_single_wqe_log_num_of_strides: %d",
-                       mprq_caps.min_single_wqe_log_num_of_strides);
-               DRV_LOG(DEBUG, "\tmax_single_wqe_log_num_of_strides: %d",
-                       mprq_caps.max_single_wqe_log_num_of_strides);
-               DRV_LOG(DEBUG, "\tsupported_qpts: %d",
-                       mprq_caps.supported_qpts);
-               DRV_LOG(DEBUG, "device supports Multi-Packet RQ");
-               mprq = 1;
-               mprq_min_stride_size_n =
-                       mprq_caps.min_single_stride_log_num_of_bytes;
-               mprq_max_stride_size_n =
-                       mprq_caps.max_single_stride_log_num_of_bytes;
-               mprq_min_stride_num_n =
-                       mprq_caps.min_single_wqe_log_num_of_strides;
-               mprq_max_stride_num_n =
-                       mprq_caps.max_single_wqe_log_num_of_strides;
-       }
-#endif
-       if (RTE_CACHE_LINE_SIZE == 128 &&
-           !(dv_attr.flags & MLX5DV_CONTEXT_FLAGS_CQE_128B_COMP))
-               cqe_comp = 0;
-       else
-               cqe_comp = 1;
-       config.cqe_comp = cqe_comp;
-#ifdef HAVE_IBV_MLX5_MOD_CQE_128B_PAD
-       /* Whether device supports 128B Rx CQE padding. */
-       cqe_pad = RTE_CACHE_LINE_SIZE == 128 &&
-                 (dv_attr.flags & MLX5DV_CONTEXT_FLAGS_CQE_128B_PAD);
-#endif
-#ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
-       if (dv_attr.comp_mask & MLX5DV_CONTEXT_MASK_TUNNEL_OFFLOADS) {
-               tunnel_en = ((dv_attr.tunnel_offloads_caps &
-                             MLX5DV_RAW_PACKET_CAP_TUNNELED_OFFLOAD_VXLAN) &&
-                            (dv_attr.tunnel_offloads_caps &
-                             MLX5DV_RAW_PACKET_CAP_TUNNELED_OFFLOAD_GRE) &&
-                            (dv_attr.tunnel_offloads_caps &
-                             MLX5DV_RAW_PACKET_CAP_TUNNELED_OFFLOAD_GENEVE));
-       }
-       DRV_LOG(DEBUG, "tunnel offloading is %ssupported",
-               tunnel_en ? "" : "not ");
-#else
-       DRV_LOG(WARNING,
-               "tunnel offloading disabled due to old OFED/rdma-core version");
-#endif
-       config.tunnel_en = tunnel_en;
-#ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
-       mpls_en = ((dv_attr.tunnel_offloads_caps &
-                   MLX5DV_RAW_PACKET_CAP_TUNNELED_OFFLOAD_CW_MPLS_OVER_GRE) &&
-                  (dv_attr.tunnel_offloads_caps &
-                   MLX5DV_RAW_PACKET_CAP_TUNNELED_OFFLOAD_CW_MPLS_OVER_UDP));
-       DRV_LOG(DEBUG, "MPLS over GRE/UDP tunnel offloading is %ssupported",
-               mpls_en ? "" : "not ");
-#else
-       DRV_LOG(WARNING, "MPLS over GRE/UDP tunnel offloading disabled due to"
-               " old OFED/rdma-core version or firmware configuration");
-#endif
-       config.mpls_en = mpls_en;
-       /* Check port status. */
-       err = mlx5_glue->query_port(sh->ctx, spawn->ibv_port, &port_attr);
-       if (err) {
-               DRV_LOG(ERR, "port query failed: %s", strerror(err));
-               goto error;
-       }
-       if (port_attr.link_layer != IBV_LINK_LAYER_ETHERNET) {
-               DRV_LOG(ERR, "port is not configured in Ethernet mode");
-               err = EINVAL;
-               goto error;
-       }
-       if (port_attr.state != IBV_PORT_ACTIVE)
-               DRV_LOG(DEBUG, "port is not active: \"%s\" (%d)",
-                       mlx5_glue->port_state_str(port_attr.state),
-                       port_attr.state);
-       /* Allocate private eth device data. */
-       priv = rte_zmalloc("ethdev private structure",
-                          sizeof(*priv),
-                          RTE_CACHE_LINE_SIZE);
-       if (priv == NULL) {
-               DRV_LOG(ERR, "priv allocation failure");
-               err = ENOMEM;
-               goto error;
-       }
-       priv->sh = sh;
-       priv->ibv_port = spawn->ibv_port;
-       priv->pci_dev = spawn->pci_dev;
-       priv->mtu = RTE_ETHER_MTU;
-       priv->mp_id.port_id = port_id;
-       strlcpy(priv->mp_id.name, MLX5_MP_NAME, RTE_MP_MAX_NAME_LEN);
-#ifndef RTE_ARCH_64
-       /* Initialize UAR access locks for 32bit implementations. */
-       rte_spinlock_init(&priv->uar_lock_cq);
-       for (i = 0; i < MLX5_UAR_PAGE_NUM_MAX; i++)
-               rte_spinlock_init(&priv->uar_lock[i]);
-#endif
-       /* Some internal functions rely on Netlink sockets, open them now. */
-       priv->nl_socket_rdma = mlx5_nl_init(NETLINK_RDMA);
-       priv->nl_socket_route = mlx5_nl_init(NETLINK_ROUTE);
-       priv->representor = !!switch_info->representor;
-       priv->master = !!switch_info->master;
-       priv->domain_id = RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID;
-       priv->vport_meta_tag = 0;
-       priv->vport_meta_mask = 0;
-       priv->pf_bond = spawn->pf_bond;
-#ifdef HAVE_MLX5DV_DR_DEVX_PORT
-       /*
-        * The DevX port query API is implemented. E-Switch may use
-        * either vport or reg_c[0] metadata register to match on
-        * vport index. The engaged part of metadata register is
-        * defined by mask.
-        */
-       if (switch_info->representor || switch_info->master) {
-               devx_port.comp_mask = MLX5DV_DEVX_PORT_VPORT |
-                                     MLX5DV_DEVX_PORT_MATCH_REG_C_0;
-               err = mlx5_glue->devx_port_query(sh->ctx, spawn->ibv_port,
-                                                &devx_port);
-               if (err) {
-                       DRV_LOG(WARNING,
-                               "can't query devx port %d on device %s",
-                               spawn->ibv_port, spawn->ibv_dev->name);
-                       devx_port.comp_mask = 0;
-               }
-       }
-       if (devx_port.comp_mask & MLX5DV_DEVX_PORT_MATCH_REG_C_0) {
-               priv->vport_meta_tag = devx_port.reg_c_0.value;
-               priv->vport_meta_mask = devx_port.reg_c_0.mask;
-               if (!priv->vport_meta_mask) {
-                       DRV_LOG(ERR, "vport zero mask for port %d"
-                                    " on bonding device %s",
-                                    spawn->ibv_port, spawn->ibv_dev->name);
-                       err = ENOTSUP;
-                       goto error;
-               }
-               if (priv->vport_meta_tag & ~priv->vport_meta_mask) {
-                       DRV_LOG(ERR, "invalid vport tag for port %d"
-                                    " on bonding device %s",
-                                    spawn->ibv_port, spawn->ibv_dev->name);
-                       err = ENOTSUP;
-                       goto error;
-               }
-       }
-       if (devx_port.comp_mask & MLX5DV_DEVX_PORT_VPORT) {
-               priv->vport_id = devx_port.vport_num;
-       } else if (spawn->pf_bond >= 0) {
-               DRV_LOG(ERR, "can't deduce vport index for port %d"
-                            " on bonding device %s",
-                            spawn->ibv_port, spawn->ibv_dev->name);
-               err = ENOTSUP;
-               goto error;
-       } else {
-               /* Suppose vport index in compatible way. */
-               priv->vport_id = switch_info->representor ?
-                                switch_info->port_name + 1 : -1;
-       }
-#else
-       /*
-        * Kernel/rdma_core support single E-Switch per PF configurations
-        * only and vport_id field contains the vport index for
-        * associated VF, which is deduced from representor port name.
-        * For example, let's have the IB device port 10, it has
-        * attached network device eth0, which has port name attribute
-        * pf0vf2, we can deduce the VF number as 2, and set vport index
-        * as 3 (2+1). This assigning schema should be changed if the
-        * multiple E-Switch instances per PF configurations or/and PCI
-        * subfunctions are added.
-        */
-       priv->vport_id = switch_info->representor ?
-                        switch_info->port_name + 1 : -1;
-#endif
-       /* representor_id field keeps the unmodified VF index. */
-       priv->representor_id = switch_info->representor ?
-                              switch_info->port_name : -1;
-       /*
-        * Look for sibling devices in order to reuse their switch domain
-        * if any, otherwise allocate one.
-        */
-       MLX5_ETH_FOREACH_DEV(port_id, priv->pci_dev) {
-               const struct mlx5_priv *opriv =
-                       rte_eth_devices[port_id].data->dev_private;
-
-               if (!opriv ||
-                   opriv->sh != priv->sh ||
-                       opriv->domain_id ==
-                       RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID)
-                       continue;
-               priv->domain_id = opriv->domain_id;
-               break;
-       }
-       if (priv->domain_id == RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID) {
-               err = rte_eth_switch_domain_alloc(&priv->domain_id);
-               if (err) {
-                       err = rte_errno;
-                       DRV_LOG(ERR, "unable to allocate switch domain: %s",
-                               strerror(rte_errno));
-                       goto error;
-               }
-               own_domain_id = 1;
-       }
-       /* Override some values set by hardware configuration. */
-       mlx5_args(&config, dpdk_dev->devargs);
-       err = mlx5_dev_check_sibling_config(priv, &config);
-       if (err)
-               goto error;
-       config.hw_csum = !!(sh->device_attr.device_cap_flags_ex &
-                           IBV_DEVICE_RAW_IP_CSUM);
-       DRV_LOG(DEBUG, "checksum offloading is %ssupported",
-               (config.hw_csum ? "" : "not "));
-#if !defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42) && \
-       !defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
-       DRV_LOG(DEBUG, "counters are not supported");
-#endif
-#if !defined(HAVE_IBV_FLOW_DV_SUPPORT) || !defined(HAVE_MLX5DV_DR)
-       if (config.dv_flow_en) {
-               DRV_LOG(WARNING, "DV flow is not supported");
-               config.dv_flow_en = 0;
-       }
-#endif
-       config.ind_table_max_size =
-               sh->device_attr.max_rwq_indirection_table_size;
-       /*
-        * Remove this check once DPDK supports larger/variable
-        * indirection tables.
-        */
-       if (config.ind_table_max_size > (unsigned int)ETH_RSS_RETA_SIZE_512)
-               config.ind_table_max_size = ETH_RSS_RETA_SIZE_512;
-       DRV_LOG(DEBUG, "maximum Rx indirection table size is %u",
-               config.ind_table_max_size);
-       config.hw_vlan_strip = !!(sh->device_attr.raw_packet_caps &
-                                 IBV_RAW_PACKET_CAP_CVLAN_STRIPPING);
-       DRV_LOG(DEBUG, "VLAN stripping is %ssupported",
-               (config.hw_vlan_strip ? "" : "not "));
-       config.hw_fcs_strip = !!(sh->device_attr.raw_packet_caps &
-                                IBV_RAW_PACKET_CAP_SCATTER_FCS);
-       DRV_LOG(DEBUG, "FCS stripping configuration is %ssupported",
-               (config.hw_fcs_strip ? "" : "not "));
-#if defined(HAVE_IBV_WQ_FLAG_RX_END_PADDING)
-       hw_padding = !!sh->device_attr.rx_pad_end_addr_align;
-#elif defined(HAVE_IBV_WQ_FLAGS_PCI_WRITE_END_PADDING)
-       hw_padding = !!(sh->device_attr.device_cap_flags_ex &
-                       IBV_DEVICE_PCI_WRITE_END_PADDING);
-#endif
-       if (config.hw_padding && !hw_padding) {
-               DRV_LOG(DEBUG, "Rx end alignment padding isn't supported");
-               config.hw_padding = 0;
-       } else if (config.hw_padding) {
-               DRV_LOG(DEBUG, "Rx end alignment padding is enabled");
-       }
-       config.tso = (sh->device_attr.max_tso > 0 &&
-                     (sh->device_attr.tso_supported_qpts &
-                      (1 << IBV_QPT_RAW_PACKET)));
-       if (config.tso)
-               config.tso_max_payload_sz = sh->device_attr.max_tso;
-       /*
-        * MPW is disabled by default, while the Enhanced MPW is enabled
-        * by default.
-        */
-       if (config.mps == MLX5_ARG_UNSET)
-               config.mps = (mps == MLX5_MPW_ENHANCED) ? MLX5_MPW_ENHANCED :
-                                                         MLX5_MPW_DISABLED;
-       else
-               config.mps = config.mps ? mps : MLX5_MPW_DISABLED;
-       DRV_LOG(INFO, "%sMPS is %s",
-               config.mps == MLX5_MPW_ENHANCED ? "enhanced " :
-               config.mps == MLX5_MPW ? "legacy " : "",
-               config.mps != MLX5_MPW_DISABLED ? "enabled" : "disabled");
-       if (config.cqe_comp && !cqe_comp) {
-               DRV_LOG(WARNING, "Rx CQE compression isn't supported");
-               config.cqe_comp = 0;
-       }
-       if (config.cqe_pad && !cqe_pad) {
-               DRV_LOG(WARNING, "Rx CQE padding isn't supported");
-               config.cqe_pad = 0;
-       } else if (config.cqe_pad) {
-               DRV_LOG(INFO, "Rx CQE padding is enabled");
-       }
-       if (config.devx) {
-               priv->counter_fallback = 0;
-               err = mlx5_devx_cmd_query_hca_attr(sh->ctx, &config.hca_attr);
-               if (err) {
-                       err = -err;
-                       goto error;
-               }
-               if (!config.hca_attr.flow_counters_dump)
-                       priv->counter_fallback = 1;
-#ifndef HAVE_IBV_DEVX_ASYNC
-               priv->counter_fallback = 1;
-#endif
-               if (priv->counter_fallback)
-                       DRV_LOG(INFO, "Use fall-back DV counter management");
-               /* Check for LRO support. */
-               if (config.dest_tir && config.hca_attr.lro_cap &&
-                   config.dv_flow_en) {
-                       /* TBD check tunnel lro caps. */
-                       config.lro.supported = config.hca_attr.lro_cap;
-                       DRV_LOG(DEBUG, "Device supports LRO");
-                       /*
-                        * If LRO timeout is not configured by application,
-                        * use the minimal supported value.
-                        */
-                       if (!config.lro.timeout)
-                               config.lro.timeout =
-                               config.hca_attr.lro_timer_supported_periods[0];
-                       DRV_LOG(DEBUG, "LRO session timeout set to %d usec",
-                               config.lro.timeout);
-               }
-#if defined(HAVE_MLX5DV_DR) && defined(HAVE_MLX5_DR_CREATE_ACTION_FLOW_METER)
-               if (config.hca_attr.qos.sup && config.hca_attr.qos.srtcm_sup &&
-                   config.dv_flow_en) {
-                       uint8_t reg_c_mask =
-                               config.hca_attr.qos.flow_meter_reg_c_ids;
-                       /*
-                        * Meter needs two REG_C's for color match and pre-sfx
-                        * flow match. Here get the REG_C for color match.
-                        * REG_C_0 and REG_C_1 is reserved for metadata feature.
-                        */
-                       reg_c_mask &= 0xfc;
-                       if (__builtin_popcount(reg_c_mask) < 1) {
-                               priv->mtr_en = 0;
-                               DRV_LOG(WARNING, "No available register for"
-                                       " meter.");
-                       } else {
-                               priv->mtr_color_reg = ffs(reg_c_mask) - 1 +
-                                                     REG_C_0;
-                               priv->mtr_en = 1;
-                               priv->mtr_reg_share =
-                                     config.hca_attr.qos.flow_meter_reg_share;
-                               DRV_LOG(DEBUG, "The REG_C meter uses is %d",
-                                       priv->mtr_color_reg);
-                       }
-               }
-#endif
-       }
-       if (config.mprq.enabled && mprq) {
-               if (config.mprq.stride_num_n &&
-                   (config.mprq.stride_num_n > mprq_max_stride_num_n ||
-                    config.mprq.stride_num_n < mprq_min_stride_num_n)) {
-                       config.mprq.stride_num_n =
-                               RTE_MIN(RTE_MAX(MLX5_MPRQ_STRIDE_NUM_N,
-                                               mprq_min_stride_num_n),
-                                       mprq_max_stride_num_n);
-                       DRV_LOG(WARNING,
-                               "the number of strides"
-                               " for Multi-Packet RQ is out of range,"
-                               " setting default value (%u)",
-                               1 << config.mprq.stride_num_n);
-               }
-               if (config.mprq.stride_size_n &&
-                   (config.mprq.stride_size_n > mprq_max_stride_size_n ||
-                    config.mprq.stride_size_n < mprq_min_stride_size_n)) {
-                       config.mprq.stride_size_n =
-                               RTE_MIN(RTE_MAX(MLX5_MPRQ_STRIDE_SIZE_N,
-                                               mprq_min_stride_size_n),
-                                       mprq_max_stride_size_n);
-                       DRV_LOG(WARNING,
-                               "the size of a stride"
-                               " for Multi-Packet RQ is out of range,"
-                               " setting default value (%u)",
-                               1 << config.mprq.stride_size_n);
-               }
-               config.mprq.min_stride_size_n = mprq_min_stride_size_n;
-               config.mprq.max_stride_size_n = mprq_max_stride_size_n;
-       } else if (config.mprq.enabled && !mprq) {
-               DRV_LOG(WARNING, "Multi-Packet RQ isn't supported");
-               config.mprq.enabled = 0;
-       }
-       if (config.max_dump_files_num == 0)
-               config.max_dump_files_num = 128;
-       eth_dev = rte_eth_dev_allocate(name);
-       if (eth_dev == NULL) {
-               DRV_LOG(ERR, "can not allocate rte ethdev");
-               err = ENOMEM;
-               goto error;
-       }
-       /* Flag to call rte_eth_dev_release_port() in rte_eth_dev_close(). */
-       eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
-       if (priv->representor) {
-               eth_dev->data->dev_flags |= RTE_ETH_DEV_REPRESENTOR;
-               eth_dev->data->representor_id = priv->representor_id;
-       }
-       /*
-        * Store associated network device interface index. This index
-        * is permanent throughout the lifetime of device. So, we may store
-        * the ifindex here and use the cached value further.
-        */
-       MLX5_ASSERT(spawn->ifindex);
-       priv->if_index = spawn->ifindex;
-       eth_dev->data->dev_private = priv;
-       priv->dev_data = eth_dev->data;
-       eth_dev->data->mac_addrs = priv->mac;
-       eth_dev->device = dpdk_dev;
-       /* Configure the first MAC address by default. */
-       if (mlx5_get_mac(eth_dev, &mac.addr_bytes)) {
-               DRV_LOG(ERR,
-                       "port %u cannot get MAC address, is mlx5_en"
-                       " loaded? (errno: %s)",
-                       eth_dev->data->port_id, strerror(rte_errno));
-               err = ENODEV;
-               goto error;
-       }
-       DRV_LOG(INFO,
-               "port %u MAC address is %02x:%02x:%02x:%02x:%02x:%02x",
-               eth_dev->data->port_id,
-               mac.addr_bytes[0], mac.addr_bytes[1],
-               mac.addr_bytes[2], mac.addr_bytes[3],
-               mac.addr_bytes[4], mac.addr_bytes[5]);
-#ifdef RTE_LIBRTE_MLX5_DEBUG
-       {
-               char ifname[IF_NAMESIZE];
-
-               if (mlx5_get_ifname(eth_dev, &ifname) == 0)
-                       DRV_LOG(DEBUG, "port %u ifname is \"%s\"",
-                               eth_dev->data->port_id, ifname);
-               else
-                       DRV_LOG(DEBUG, "port %u ifname is unknown",
-                               eth_dev->data->port_id);
-       }
-#endif
-       /* Get actual MTU if possible. */
-       err = mlx5_get_mtu(eth_dev, &priv->mtu);
-       if (err) {
-               err = rte_errno;
-               goto error;
-       }
-       DRV_LOG(DEBUG, "port %u MTU is %u", eth_dev->data->port_id,
-               priv->mtu);
-       /* Initialize burst functions to prevent crashes before link-up. */
-       eth_dev->rx_pkt_burst = removed_rx_burst;
-       eth_dev->tx_pkt_burst = removed_tx_burst;
-       eth_dev->dev_ops = &mlx5_dev_ops;
-       /* Register MAC address. */
-       claim_zero(mlx5_mac_addr_add(eth_dev, &mac, 0, 0));
-       if (config.vf && config.vf_nl_en)
-               mlx5_nl_mac_addr_sync(priv->nl_socket_route,
-                                     mlx5_ifindex(eth_dev),
-                                     eth_dev->data->mac_addrs,
-                                     MLX5_MAX_MAC_ADDRESSES);
-       priv->flows = 0;
-       priv->ctrl_flows = 0;
-       TAILQ_INIT(&priv->flow_meters);
-       TAILQ_INIT(&priv->flow_meter_profiles);
-       /* Hint libmlx5 to use PMD allocator for data plane resources */
-       struct mlx5dv_ctx_allocators alctr = {
-               .alloc = &mlx5_alloc_verbs_buf,
-               .free = &mlx5_free_verbs_buf,
-               .data = priv,
-       };
-       mlx5_glue->dv_set_context_attr(sh->ctx,
-                                      MLX5DV_CTX_ATTR_BUF_ALLOCATORS,
-                                      (void *)((uintptr_t)&alctr));
-       /* Bring Ethernet device up. */
-       DRV_LOG(DEBUG, "port %u forcing Ethernet interface up",
-               eth_dev->data->port_id);
-       mlx5_set_link_up(eth_dev);
-       /*
-        * Even though the interrupt handler is not installed yet,
-        * interrupts will still trigger on the async_fd from
-        * Verbs context returned by ibv_open_device().
-        */
-       mlx5_link_update(eth_dev, 0);
-#ifdef HAVE_MLX5DV_DR_ESWITCH
-       if (!(config.hca_attr.eswitch_manager && config.dv_flow_en &&
-             (switch_info->representor || switch_info->master)))
-               config.dv_esw_en = 0;
-#else
-       config.dv_esw_en = 0;
-#endif
-       /* Detect minimal data bytes to inline. */
-       mlx5_set_min_inline(spawn, &config);
-       /* Store device configuration on private structure. */
-       priv->config = config;
-       /* Create context for virtual machine VLAN workaround. */
-       priv->vmwa_context = mlx5_vlan_vmwa_init(eth_dev, spawn->ifindex);
-       if (config.dv_flow_en) {
-               err = mlx5_alloc_shared_dr(priv);
-               if (err)
-                       goto error;
-               /*
-                * RSS id is shared with meter flow id. Meter flow id can only
-                * use the 24 MSB of the register.
-                */
-               priv->qrss_id_pool = mlx5_flow_id_pool_alloc(UINT32_MAX >>
-                                    MLX5_MTR_COLOR_BITS);
-               if (!priv->qrss_id_pool) {
-                       DRV_LOG(ERR, "can't create flow id pool");
-                       err = ENOMEM;
-                       goto error;
-               }
-       }
-       /* Supported Verbs flow priority number detection. */
-       err = mlx5_flow_discover_priorities(eth_dev);
-       if (err < 0) {
-               err = -err;
-               goto error;
-       }
-       priv->config.flow_prio = err;
-       if (!priv->config.dv_esw_en &&
-           priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
-               DRV_LOG(WARNING, "metadata mode %u is not supported "
-                                "(no E-Switch)", priv->config.dv_xmeta_en);
-               priv->config.dv_xmeta_en = MLX5_XMETA_MODE_LEGACY;
-       }
-       mlx5_set_metadata_mask(eth_dev);
-       if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
-           !priv->sh->dv_regc0_mask) {
-               DRV_LOG(ERR, "metadata mode %u is not supported "
-                            "(no metadata reg_c[0] is available)",
-                            priv->config.dv_xmeta_en);
-                       err = ENOTSUP;
-                       goto error;
-       }
-       /*
-        * Allocate the buffer for flow creating, just once.
-        * The allocation must be done before any flow creating.
-        */
-       mlx5_flow_alloc_intermediate(eth_dev);
-       /* Query availibility of metadata reg_c's. */
-       err = mlx5_flow_discover_mreg_c(eth_dev);
-       if (err < 0) {
-               err = -err;
-               goto error;
-       }
-       if (!mlx5_flow_ext_mreg_supported(eth_dev)) {
-               DRV_LOG(DEBUG,
-                       "port %u extensive metadata register is not supported",
-                       eth_dev->data->port_id);
-               if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
-                       DRV_LOG(ERR, "metadata mode %u is not supported "
-                                    "(no metadata registers available)",
-                                    priv->config.dv_xmeta_en);
-                       err = ENOTSUP;
-                       goto error;
-               }
-       }
-       if (priv->config.dv_flow_en &&
-           priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
-           mlx5_flow_ext_mreg_supported(eth_dev) &&
-           priv->sh->dv_regc0_mask) {
-               priv->mreg_cp_tbl = mlx5_hlist_create(MLX5_FLOW_MREG_HNAME,
-                                                     MLX5_FLOW_MREG_HTABLE_SZ);
-               if (!priv->mreg_cp_tbl) {
-                       err = ENOMEM;
-                       goto error;
-               }
-       }
-       return eth_dev;
-error:
-       if (priv) {
-               if (priv->mreg_cp_tbl)
-                       mlx5_hlist_destroy(priv->mreg_cp_tbl, NULL, NULL);
-               if (priv->sh)
-                       mlx5_free_shared_dr(priv);
-               if (priv->nl_socket_route >= 0)
-                       close(priv->nl_socket_route);
-               if (priv->nl_socket_rdma >= 0)
-                       close(priv->nl_socket_rdma);
-               if (priv->vmwa_context)
-                       mlx5_vlan_vmwa_exit(priv->vmwa_context);
-               if (priv->qrss_id_pool)
-                       mlx5_flow_id_pool_release(priv->qrss_id_pool);
-               if (own_domain_id)
-                       claim_zero(rte_eth_switch_domain_free(priv->domain_id));
-               rte_free(priv);
-               if (eth_dev != NULL)
-                       eth_dev->data->dev_private = NULL;
-       }
-       if (eth_dev != NULL) {
-               /* mac_addrs must not be freed alone because part of dev_private */
-               eth_dev->data->mac_addrs = NULL;
-               rte_eth_dev_release_port(eth_dev);
-       }
-       if (sh)
-               mlx5_free_shared_ibctx(sh);
-       MLX5_ASSERT(err > 0);
-       rte_errno = err;
-       return NULL;
-}
-
-/**
- * Comparison callback to sort device data.
- *
- * This is meant to be used with qsort().
- *
- * @param a[in]
- *   Pointer to pointer to first data object.
- * @param b[in]
- *   Pointer to pointer to second data object.
- *
- * @return
- *   0 if both objects are equal, less than 0 if the first argument is less
- *   than the second, greater than 0 otherwise.
- */
-static int
-mlx5_dev_spawn_data_cmp(const void *a, const void *b)
-{
-       const struct mlx5_switch_info *si_a =
-               &((const struct mlx5_dev_spawn_data *)a)->info;
-       const struct mlx5_switch_info *si_b =
-               &((const struct mlx5_dev_spawn_data *)b)->info;
-       int ret;
-
-       /* Master device first. */
-       ret = si_b->master - si_a->master;
-       if (ret)
-               return ret;
-       /* Then representor devices. */
-       ret = si_b->representor - si_a->representor;
-       if (ret)
-               return ret;
-       /* Unidentified devices come last in no specific order. */
-       if (!si_a->representor)
-               return 0;
-       /* Order representors by name. */
-       return si_a->port_name - si_b->port_name;
-}
-
-/**
- * Match PCI information for possible slaves of bonding device.
- *
- * @param[in] ibv_dev
- *   Pointer to Infiniband device structure.
- * @param[in] pci_dev
- *   Pointer to PCI device structure to match PCI address.
- * @param[in] nl_rdma
- *   Netlink RDMA group socket handle.
- *
- * @return
- *   negative value if no bonding device found, otherwise
- *   positive index of slave PF in bonding.
- */
-static int
-mlx5_device_bond_pci_match(const struct ibv_device *ibv_dev,
-                          const struct rte_pci_device *pci_dev,
-                          int nl_rdma)
-{
-       char ifname[IF_NAMESIZE + 1];
-       unsigned int ifindex;
-       unsigned int np, i;
-       FILE *file = NULL;
-       int pf = -1;
-
-       /*
-        * Try to get master device name. If something goes
-        * wrong suppose the lack of kernel support and no
-        * bonding devices.
-        */
-       if (nl_rdma < 0)
-               return -1;
-       if (!strstr(ibv_dev->name, "bond"))
-               return -1;
-       np = mlx5_nl_portnum(nl_rdma, ibv_dev->name);
-       if (!np)
-               return -1;
-       /*
-        * The Master device might not be on the predefined
-        * port (not on port index 1, it is not garanted),
-        * we have to scan all Infiniband device port and
-        * find master.
-        */
-       for (i = 1; i <= np; ++i) {
-               /* Check whether Infiniband port is populated. */
-               ifindex = mlx5_nl_ifindex(nl_rdma, ibv_dev->name, i);
-               if (!ifindex)
-                       continue;
-               if (!if_indextoname(ifindex, ifname))
-                       continue;
-               /* Try to read bonding slave names from sysfs. */
-               MKSTR(slaves,
-                     "/sys/class/net/%s/master/bonding/slaves", ifname);
-               file = fopen(slaves, "r");
-               if (file)
-                       break;
-       }
-       if (!file)
-               return -1;
-       /* Use safe format to check maximal buffer length. */
-       MLX5_ASSERT(atol(RTE_STR(IF_NAMESIZE)) == IF_NAMESIZE);
-       while (fscanf(file, "%" RTE_STR(IF_NAMESIZE) "s", ifname) == 1) {
-               char tmp_str[IF_NAMESIZE + 32];
-               struct rte_pci_addr pci_addr;
-               struct mlx5_switch_info info;
-
-               /* Process slave interface names in the loop. */
-               snprintf(tmp_str, sizeof(tmp_str),
-                        "/sys/class/net/%s", ifname);
-               if (mlx5_dev_to_pci_addr(tmp_str, &pci_addr)) {
-                       DRV_LOG(WARNING, "can not get PCI address"
-                                        " for netdev \"%s\"", ifname);
-                       continue;
-               }
-               if (pci_dev->addr.domain != pci_addr.domain ||
-                   pci_dev->addr.bus != pci_addr.bus ||
-                   pci_dev->addr.devid != pci_addr.devid ||
-                   pci_dev->addr.function != pci_addr.function)
-                       continue;
-               /* Slave interface PCI address match found. */
-               fclose(file);
-               snprintf(tmp_str, sizeof(tmp_str),
-                        "/sys/class/net/%s/phys_port_name", ifname);
-               file = fopen(tmp_str, "rb");
-               if (!file)
-                       break;
-               info.name_type = MLX5_PHYS_PORT_NAME_TYPE_NOTSET;
-               if (fscanf(file, "%32s", tmp_str) == 1)
-                       mlx5_translate_port_name(tmp_str, &info);
-               if (info.name_type == MLX5_PHYS_PORT_NAME_TYPE_LEGACY ||
-                   info.name_type == MLX5_PHYS_PORT_NAME_TYPE_UPLINK)
-                       pf = info.port_name;
-               break;
-       }
-       if (file)
-               fclose(file);
-       return pf;
-}
-
-/**
- * DPDK callback to register a PCI device.
- *
- * This function spawns Ethernet devices out of a given PCI device.
- *
- * @param[in] pci_drv
- *   PCI driver structure (mlx5_driver).
- * @param[in] pci_dev
- *   PCI device information.
- *
- * @return
- *   0 on success, a negative errno value otherwise and rte_errno is set.
- */
-static int
-mlx5_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
-              struct rte_pci_device *pci_dev)
-{
-       struct ibv_device **ibv_list;
-       /*
-        * Number of found IB Devices matching with requested PCI BDF.
-        * nd != 1 means there are multiple IB devices over the same
-        * PCI device and we have representors and master.
-        */
-       unsigned int nd = 0;
-       /*
-        * Number of found IB device Ports. nd = 1 and np = 1..n means
-        * we have the single multiport IB device, and there may be
-        * representors attached to some of found ports.
-        */
-       unsigned int np = 0;
-       /*
-        * Number of DPDK ethernet devices to Spawn - either over
-        * multiple IB devices or multiple ports of single IB device.
-        * Actually this is the number of iterations to spawn.
-        */
-       unsigned int ns = 0;
-       /*
-        * Bonding device
-        *   < 0 - no bonding device (single one)
-        *  >= 0 - bonding device (value is slave PF index)
-        */
-       int bd = -1;
-       struct mlx5_dev_spawn_data *list = NULL;
-       struct mlx5_dev_config dev_config;
-       int ret;
-
-       if (mlx5_class_get(pci_dev->device.devargs) != MLX5_CLASS_NET) {
-               DRV_LOG(DEBUG, "Skip probing - should be probed by other mlx5"
-                       " driver.");
-               return 1;
-       }
-       if (rte_eal_process_type() == RTE_PROC_PRIMARY)
-               mlx5_pmd_socket_init();
-       ret = mlx5_init_once();
-       if (ret) {
-               DRV_LOG(ERR, "unable to init PMD global data: %s",
-                       strerror(rte_errno));
-               return -rte_errno;
-       }
-       MLX5_ASSERT(pci_drv == &mlx5_driver);
-       errno = 0;
-       ibv_list = mlx5_glue->get_device_list(&ret);
-       if (!ibv_list) {
-               rte_errno = errno ? errno : ENOSYS;
-               DRV_LOG(ERR, "cannot list devices, is ib_uverbs loaded?");
-               return -rte_errno;
-       }
-       /*
-        * First scan the list of all Infiniband devices to find
-        * matching ones, gathering into the list.
-        */
-       struct ibv_device *ibv_match[ret + 1];
-       int nl_route = mlx5_nl_init(NETLINK_ROUTE);
-       int nl_rdma = mlx5_nl_init(NETLINK_RDMA);
-       unsigned int i;
-
-       while (ret-- > 0) {
-               struct rte_pci_addr pci_addr;
-
-               DRV_LOG(DEBUG, "checking device \"%s\"", ibv_list[ret]->name);
-               bd = mlx5_device_bond_pci_match
-                               (ibv_list[ret], pci_dev, nl_rdma);
-               if (bd >= 0) {
-                       /*
-                        * Bonding device detected. Only one match is allowed,
-                        * the bonding is supported over multi-port IB device,
-                        * there should be no matches on representor PCI
-                        * functions or non VF LAG bonding devices with
-                        * specified address.
-                        */
-                       if (nd) {
-                               DRV_LOG(ERR,
-                                       "multiple PCI match on bonding device"
-                                       "\"%s\" found", ibv_list[ret]->name);
-                               rte_errno = ENOENT;
-                               ret = -rte_errno;
-                               goto exit;
-                       }
-                       DRV_LOG(INFO, "PCI information matches for"
-                                     " slave %d bonding device \"%s\"",
-                                     bd, ibv_list[ret]->name);
-                       ibv_match[nd++] = ibv_list[ret];
-                       break;
-               }
-               if (mlx5_dev_to_pci_addr
-                       (ibv_list[ret]->ibdev_path, &pci_addr))
-                       continue;
-               if (pci_dev->addr.domain != pci_addr.domain ||
-                   pci_dev->addr.bus != pci_addr.bus ||
-                   pci_dev->addr.devid != pci_addr.devid ||
-                   pci_dev->addr.function != pci_addr.function)
-                       continue;
-               DRV_LOG(INFO, "PCI information matches for device \"%s\"",
-                       ibv_list[ret]->name);
-               ibv_match[nd++] = ibv_list[ret];
-       }
-       ibv_match[nd] = NULL;
-       if (!nd) {
-               /* No device matches, just complain and bail out. */
-               DRV_LOG(WARNING,
-                       "no Verbs device matches PCI device " PCI_PRI_FMT ","
-                       " are kernel drivers loaded?",
-                       pci_dev->addr.domain, pci_dev->addr.bus,
-                       pci_dev->addr.devid, pci_dev->addr.function);
-               rte_errno = ENOENT;
-               ret = -rte_errno;
-               goto exit;
-       }
-       if (nd == 1) {
-               /*
-                * Found single matching device may have multiple ports.
-                * Each port may be representor, we have to check the port
-                * number and check the representors existence.
-                */
-               if (nl_rdma >= 0)
-                       np = mlx5_nl_portnum(nl_rdma, ibv_match[0]->name);
-               if (!np)
-                       DRV_LOG(WARNING, "can not get IB device \"%s\""
-                                        " ports number", ibv_match[0]->name);
-               if (bd >= 0 && !np) {
-                       DRV_LOG(ERR, "can not get ports"
-                                    " for bonding device");
-                       rte_errno = ENOENT;
-                       ret = -rte_errno;
-                       goto exit;
-               }
-       }
-#ifndef HAVE_MLX5DV_DR_DEVX_PORT
-       if (bd >= 0) {
-               /*
-                * This may happen if there is VF LAG kernel support and
-                * application is compiled with older rdma_core library.
-                */
-               DRV_LOG(ERR,
-                       "No kernel/verbs support for VF LAG bonding found.");
-               rte_errno = ENOTSUP;
-               ret = -rte_errno;
-               goto exit;
-       }
-#endif
-       /*
-        * Now we can determine the maximal
-        * amount of devices to be spawned.
-        */
-       list = rte_zmalloc("device spawn data",
-                        sizeof(struct mlx5_dev_spawn_data) *
-                        (np ? np : nd),
-                        RTE_CACHE_LINE_SIZE);
-       if (!list) {
-               DRV_LOG(ERR, "spawn data array allocation failure");
-               rte_errno = ENOMEM;
-               ret = -rte_errno;
-               goto exit;
-       }
-       if (bd >= 0 || np > 1) {
-               /*
-                * Single IB device with multiple ports found,
-                * it may be E-Switch master device and representors.
-                * We have to perform identification through the ports.
-                */
-               MLX5_ASSERT(nl_rdma >= 0);
-               MLX5_ASSERT(ns == 0);
-               MLX5_ASSERT(nd == 1);
-               MLX5_ASSERT(np);
-               for (i = 1; i <= np; ++i) {
-                       list[ns].max_port = np;
-                       list[ns].ibv_port = i;
-                       list[ns].ibv_dev = ibv_match[0];
-                       list[ns].eth_dev = NULL;
-                       list[ns].pci_dev = pci_dev;
-                       list[ns].pf_bond = bd;
-                       list[ns].ifindex = mlx5_nl_ifindex
-                                       (nl_rdma, list[ns].ibv_dev->name, i);
-                       if (!list[ns].ifindex) {
-                               /*
-                                * No network interface index found for the
-                                * specified port, it means there is no
-                                * representor on this port. It's OK,
-                                * there can be disabled ports, for example
-                                * if sriov_numvfs < sriov_totalvfs.
-                                */
-                               continue;
-                       }
-                       ret = -1;
-                       if (nl_route >= 0)
-                               ret = mlx5_nl_switch_info
-                                              (nl_route,
-                                               list[ns].ifindex,
-                                               &list[ns].info);
-                       if (ret || (!list[ns].info.representor &&
-                                   !list[ns].info.master)) {
-                               /*
-                                * We failed to recognize representors with
-                                * Netlink, let's try to perform the task
-                                * with sysfs.
-                                */
-                               ret =  mlx5_sysfs_switch_info
-                                               (list[ns].ifindex,
-                                                &list[ns].info);
-                       }
-                       if (!ret && bd >= 0) {
-                               switch (list[ns].info.name_type) {
-                               case MLX5_PHYS_PORT_NAME_TYPE_UPLINK:
-                                       if (list[ns].info.port_name == bd)
-                                               ns++;
-                                       break;
-                               case MLX5_PHYS_PORT_NAME_TYPE_PFVF:
-                                       if (list[ns].info.pf_num == bd)
-                                               ns++;
-                                       break;
-                               default:
-                                       break;
-                               }
-                               continue;
-                       }
-                       if (!ret && (list[ns].info.representor ^
-                                    list[ns].info.master))
-                               ns++;
-               }
-               if (!ns) {
-                       DRV_LOG(ERR,
-                               "unable to recognize master/representors"
-                               " on the IB device with multiple ports");
-                       rte_errno = ENOENT;
-                       ret = -rte_errno;
-                       goto exit;
-               }
-       } else {
-               /*
-                * The existence of several matching entries (nd > 1) means
-                * port representors have been instantiated. No existing Verbs
-                * call nor sysfs entries can tell them apart, this can only
-                * be done through Netlink calls assuming kernel drivers are
-                * recent enough to support them.
-                *
-                * In the event of identification failure through Netlink,
-                * try again through sysfs, then:
-                *
-                * 1. A single IB device matches (nd == 1) with single
-                *    port (np=0/1) and is not a representor, assume
-                *    no switch support.
-                *
-                * 2. Otherwise no safe assumptions can be made;
-                *    complain louder and bail out.
-                */
-               np = 1;
-               for (i = 0; i != nd; ++i) {
-                       memset(&list[ns].info, 0, sizeof(list[ns].info));
-                       list[ns].max_port = 1;
-                       list[ns].ibv_port = 1;
-                       list[ns].ibv_dev = ibv_match[i];
-                       list[ns].eth_dev = NULL;
-                       list[ns].pci_dev = pci_dev;
-                       list[ns].pf_bond = -1;
-                       list[ns].ifindex = 0;
-                       if (nl_rdma >= 0)
-                               list[ns].ifindex = mlx5_nl_ifindex
-                                       (nl_rdma, list[ns].ibv_dev->name, 1);
-                       if (!list[ns].ifindex) {
-                               char ifname[IF_NAMESIZE];
-
-                               /*
-                                * Netlink failed, it may happen with old
-                                * ib_core kernel driver (before 4.16).
-                                * We can assume there is old driver because
-                                * here we are processing single ports IB
-                                * devices. Let's try sysfs to retrieve
-                                * the ifindex. The method works for
-                                * master device only.
-                                */
-                               if (nd > 1) {
-                                       /*
-                                        * Multiple devices found, assume
-                                        * representors, can not distinguish
-                                        * master/representor and retrieve
-                                        * ifindex via sysfs.
-                                        */
-                                       continue;
-                               }
-                               ret = mlx5_get_master_ifname
-                                       (ibv_match[i]->ibdev_path, &ifname);
-                               if (!ret)
-                                       list[ns].ifindex =
-                                               if_nametoindex(ifname);
-                               if (!list[ns].ifindex) {
-                                       /*
-                                        * No network interface index found
-                                        * for the specified device, it means
-                                        * there it is neither representor
-                                        * nor master.
-                                        */
-                                       continue;
-                               }
-                       }
-                       ret = -1;
-                       if (nl_route >= 0)
-                               ret = mlx5_nl_switch_info
-                                              (nl_route,
-                                               list[ns].ifindex,
-                                               &list[ns].info);
-                       if (ret || (!list[ns].info.representor &&
-                                   !list[ns].info.master)) {
-                               /*
-                                * We failed to recognize representors with
-                                * Netlink, let's try to perform the task
-                                * with sysfs.
-                                */
-                               ret =  mlx5_sysfs_switch_info
-                                               (list[ns].ifindex,
-                                                &list[ns].info);
-                       }
-                       if (!ret && (list[ns].info.representor ^
-                                    list[ns].info.master)) {
-                               ns++;
-                       } else if ((nd == 1) &&
-                                  !list[ns].info.representor &&
-                                  !list[ns].info.master) {
-                               /*
-                                * Single IB device with
-                                * one physical port and
-                                * attached network device.
-                                * May be SRIOV is not enabled
-                                * or there is no representors.
-                                */
-                               DRV_LOG(INFO, "no E-Switch support detected");
-                               ns++;
-                               break;
-                       }
-               }
-               if (!ns) {
-                       DRV_LOG(ERR,
-                               "unable to recognize master/representors"
-                               " on the multiple IB devices");
-                       rte_errno = ENOENT;
-                       ret = -rte_errno;
-                       goto exit;
-               }
-       }
-       MLX5_ASSERT(ns);
-       /*
-        * Sort list to probe devices in natural order for users convenience
-        * (i.e. master first, then representors from lowest to highest ID).
-        */
-       qsort(list, ns, sizeof(*list), mlx5_dev_spawn_data_cmp);
-       /* Default configuration. */
-       dev_config = (struct mlx5_dev_config){
-               .hw_padding = 0,
-               .mps = MLX5_ARG_UNSET,
-               .dbnc = MLX5_ARG_UNSET,
-               .rx_vec_en = 1,
-               .txq_inline_max = MLX5_ARG_UNSET,
-               .txq_inline_min = MLX5_ARG_UNSET,
-               .txq_inline_mpw = MLX5_ARG_UNSET,
-               .txqs_inline = MLX5_ARG_UNSET,
-               .vf_nl_en = 1,
-               .mr_ext_memseg_en = 1,
-               .mprq = {
-                       .enabled = 0, /* Disabled by default. */
-                       .stride_num_n = 0,
-                       .stride_size_n = 0,
-                       .max_memcpy_len = MLX5_MPRQ_MEMCPY_DEFAULT_LEN,
-                       .min_rxqs_num = MLX5_MPRQ_MIN_RXQS,
-               },
-               .dv_esw_en = 1,
-               .dv_flow_en = 1,
-               .log_hp_size = MLX5_ARG_UNSET,
-       };
-       /* Device specific configuration. */
-       switch (pci_dev->id.device_id) {
-       case PCI_DEVICE_ID_MELLANOX_CONNECTX4VF:
-       case PCI_DEVICE_ID_MELLANOX_CONNECTX4LXVF:
-       case PCI_DEVICE_ID_MELLANOX_CONNECTX5VF:
-       case PCI_DEVICE_ID_MELLANOX_CONNECTX5EXVF:
-       case PCI_DEVICE_ID_MELLANOX_CONNECTX5BFVF:
-       case PCI_DEVICE_ID_MELLANOX_CONNECTX6VF:
-       case PCI_DEVICE_ID_MELLANOX_CONNECTX6DXVF:
-               dev_config.vf = 1;
-               break;
-       default:
-               break;
-       }
-       for (i = 0; i != ns; ++i) {
-               uint32_t restore;
-
-               list[i].eth_dev = mlx5_dev_spawn(&pci_dev->device,
-                                                &list[i],
-                                                dev_config);
-               if (!list[i].eth_dev) {
-                       if (rte_errno != EBUSY && rte_errno != EEXIST)
-                               break;
-                       /* Device is disabled or already spawned. Ignore it. */
-                       continue;
-               }
-               restore = list[i].eth_dev->data->dev_flags;
-               rte_eth_copy_pci_info(list[i].eth_dev, pci_dev);
-               /* Restore non-PCI flags cleared by the above call. */
-               list[i].eth_dev->data->dev_flags |= restore;
-               rte_eth_dev_probing_finish(list[i].eth_dev);
-       }
-       if (i != ns) {
-               DRV_LOG(ERR,
-                       "probe of PCI device " PCI_PRI_FMT " aborted after"
-                       " encountering an error: %s",
-                       pci_dev->addr.domain, pci_dev->addr.bus,
-                       pci_dev->addr.devid, pci_dev->addr.function,
-                       strerror(rte_errno));
-               ret = -rte_errno;
-               /* Roll back. */
-               while (i--) {
-                       if (!list[i].eth_dev)
-                               continue;
-                       mlx5_dev_close(list[i].eth_dev);
-                       /* mac_addrs must not be freed because in dev_private */
-                       list[i].eth_dev->data->mac_addrs = NULL;
-                       claim_zero(rte_eth_dev_release_port(list[i].eth_dev));
-               }
-               /* Restore original error. */
-               rte_errno = -ret;
-       } else {
-               ret = 0;
-       }
-exit:
-       /*
-        * Do the routine cleanup:
-        * - close opened Netlink sockets
-        * - free allocated spawn data array
-        * - free the Infiniband device list
-        */
-       if (nl_rdma >= 0)
-               close(nl_rdma);
-       if (nl_route >= 0)
-               close(nl_route);
-       if (list)
-               rte_free(list);
-       MLX5_ASSERT(ibv_list);
-       mlx5_glue->free_device_list(ibv_list);
-       return ret;
-}
 
 /**
  * Look for the ethernet device belonging to mlx5 driver.
@@ -3762,7 +1983,7 @@ exit:
  * @param[in] pci_dev
  *   Pointer to the hint PCI device. When device is being probed
  *   the its siblings (master and preceding representors might
- *   not have assigned driver yet (because the mlx5_pci_probe()
+ *   not have assigned driver yet (because the mlx5_os_pci_probe()
  *   is not completed yet, for this case match on hint PCI
  *   device may be used to detect sibling device.
  *
@@ -3884,12 +2105,12 @@ static const struct rte_pci_id mlx5_pci_id_map[] = {
        }
 };
 
-static struct rte_pci_driver mlx5_driver = {
+struct rte_pci_driver mlx5_driver = {
        .driver = {
                .name = MLX5_DRIVER_NAME
        },
        .id_table = mlx5_pci_id_map,
-       .probe = mlx5_pci_probe,
+       .probe = mlx5_os_pci_probe,
        .remove = mlx5_pci_remove,
        .dma_map = mlx5_dma_map,
        .dma_unmap = mlx5_dma_unmap,
index ec4ba87..f5d9aad 100644 (file)
@@ -94,6 +94,18 @@ struct mlx5_dev_attr {
        char            fw_ver[64];
 };
 
+/** Data associated with devices to spawn. */
+struct mlx5_dev_spawn_data {
+       uint32_t ifindex; /**< Network interface index. */
+       uint32_t max_port; /**< IB device maximal port index. */
+       uint32_t ibv_port; /**< IB device physical port index. */
+       int pf_bond; /**< bonding device PF index. < 0 - no bonding */
+       struct mlx5_switch_info info; /**< Switch information. */
+       struct ibv_device *ibv_dev; /**< Associated IB device. */
+       struct rte_eth_dev *eth_dev; /**< Associated Ethernet device. */
+       struct rte_pci_device *pci_dev; /**< Backend PCI device. */
+};
+
 /** Key string for IPC. */
 #define MLX5_MP_NAME "net_mlx5_mp"
 
@@ -116,6 +128,11 @@ struct mlx5_local_data {
 };
 
 extern struct mlx5_shared_data *mlx5_shared_data;
+extern struct rte_pci_driver mlx5_driver;
+
+/* Dev ops structs */
+extern const struct eth_dev_ops mlx5_dev_sec_ops;
+extern const struct eth_dev_ops mlx5_dev_ops;
 
 struct mlx5_counter_ctrl {
        /* Name of the counter. */
@@ -670,12 +687,26 @@ int32_t mlx5_release_dbr(struct rte_eth_dev *dev, uint32_t umem_id,
 int mlx5_udp_tunnel_port_add(struct rte_eth_dev *dev,
                              struct rte_eth_udp_tunnel *udp_tunnel);
 uint16_t mlx5_eth_find_next(uint16_t port_id, struct rte_pci_device *pci_dev);
+void mlx5_dev_close(struct rte_eth_dev *dev);
 
 /* Macro to iterate over all valid ports for mlx5 driver. */
 #define MLX5_ETH_FOREACH_DEV(port_id, pci_dev) \
        for (port_id = mlx5_eth_find_next(0, pci_dev); \
             port_id < RTE_MAX_ETHPORTS; \
             port_id = mlx5_eth_find_next(port_id + 1, pci_dev))
+int mlx5_args(struct mlx5_dev_config *config, struct rte_devargs *devargs);
+struct mlx5_dev_ctx_shared *
+mlx5_alloc_shared_ibctx(const struct mlx5_dev_spawn_data *spawn,
+                       const struct mlx5_dev_config *config);
+void mlx5_free_shared_ibctx(struct mlx5_dev_ctx_shared *sh);
+void mlx5_free_table_hash_list(struct mlx5_priv *priv);
+int mlx5_alloc_table_hash_list(struct mlx5_priv *priv);
+void mlx5_set_min_inline(struct mlx5_dev_spawn_data *spawn,
+                        struct mlx5_dev_config *config);
+void mlx5_set_metadata_mask(struct rte_eth_dev *dev);
+int mlx5_dev_check_sibling_config(struct mlx5_priv *priv,
+                                 struct mlx5_dev_config *config);
+int mlx5_init_once(void);
 
 /* mlx5_ethdev.c */
 
@@ -876,9 +907,18 @@ struct mlx5_flow_meter *mlx5_flow_meter_attach
 void mlx5_flow_meter_detach(struct mlx5_flow_meter *fm);
 
 /* mlx5_os.c */
+struct rte_pci_driver;
 const char *mlx5_os_get_ctx_device_name(void *ctx);
 const char *mlx5_os_get_ctx_device_path(void *ctx);
 uint32_t mlx5_os_get_umem_id(void *umem);
 int mlx5_os_get_dev_attr(void *ctx, struct mlx5_dev_attr *dev_attr);
-
+void mlx5_os_free_shared_dr(struct mlx5_priv *priv);
+int mlx5_os_open_device(const struct mlx5_dev_spawn_data *spawn,
+                        const struct mlx5_dev_config *config,
+                        struct mlx5_dev_ctx_shared *sh);
+int mlx5_os_get_pdn(void *pd, uint32_t *pdn);
+int mlx5_os_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
+                      struct rte_pci_device *pci_dev);
+void mlx5_os_dev_shared_handler_install(struct mlx5_dev_ctx_shared *sh);
+void mlx5_os_dev_shared_handler_uninstall(struct mlx5_dev_ctx_shared *sh);
 #endif /* RTE_PMD_MLX5_H_ */