X-Git-Url: http://git.droids-corp.org/?a=blobdiff_plain;f=drivers%2Fnet%2Fmlx5%2Fmlx5.c;h=57a62215a00d02fd4d4c33cd26675fec6fa7dc54;hb=35b2d13fd6fdcbd191f2a30d74648faeb1186c65;hp=9d0b1295191257178b45cbc3e41e8bb0389c711e;hpb=e505508a385826f2731cae825a332af122e9b783;p=dpdk.git diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c index 9d0b129519..57a62215a0 100644 --- a/drivers/net/mlx5/mlx5.c +++ b/drivers/net/mlx5/mlx5.c @@ -101,12 +101,18 @@ /* Allow L3 VXLAN flow creation. */ #define MLX5_L3_VXLAN_EN "l3_vxlan_en" +/* Activate DV E-Switch flow steering. */ +#define MLX5_DV_ESW_EN "dv_esw_en" + /* Activate DV flow steering. */ #define MLX5_DV_FLOW_EN "dv_flow_en" /* Activate Netlink support in VF mode. */ #define MLX5_VF_NL_EN "vf_nl_en" +/* Enable extending memsegs when creating a MR. */ +#define MLX5_MR_EXT_MEMSEG_EN "mr_ext_memseg_en" + /* Select port representors to instantiate. */ #define MLX5_REPRESENTOR "representor" @@ -127,16 +133,342 @@ struct mlx5_shared_data *mlx5_shared_data; /* Spinlock for mlx5_shared_data allocation. */ 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. */ + 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_ibv_shared) mlx5_ibv_list = LIST_HEAD_INITIALIZER(); +static pthread_mutex_t mlx5_ibv_list_mutex = PTHREAD_MUTEX_INITIALIZER; + /** - * Prepare shared data between primary and secondary process. + * Allocate shared IB device context. If there is multiport device the + * master and representors will share this context, if there is single + * port dedicated IB device, the context will be used by only given + * port due to unification. + * + * Routine first searches the context for the specified IB device name, + * if found the shared context assumed and reference counter is incremented. + * If no context found the new one is created and initialized with specified + * IB device context and parameters. + * + * @param[in] spawn + * Pointer to the IB device attributes (name, port, etc). + * + * @return + * Pointer to mlx5_ibv_shared object on success, + * otherwise NULL and rte_errno is set. + */ +static struct mlx5_ibv_shared * +mlx5_alloc_shared_ibctx(const struct mlx5_dev_spawn_data *spawn) +{ + struct mlx5_ibv_shared *sh; + int err = 0; + uint32_t i; + + assert(spawn); + /* Secondary process should not create the shared context. */ + assert(rte_eal_process_type() == RTE_PROC_PRIMARY); + pthread_mutex_lock(&mlx5_ibv_list_mutex); + /* Search for IB context by device name. */ + LIST_FOREACH(sh, &mlx5_ibv_list, next) { + if (!strcmp(sh->ibdev_name, spawn->ibv_dev->name)) { + sh->refcnt++; + goto exit; + } + } + /* No device found, we have to create new shared context. */ + assert(spawn->max_port); + sh = rte_zmalloc("ethdev shared ib context", + sizeof(struct mlx5_ibv_shared) + + spawn->max_port * + sizeof(struct mlx5_ibv_shared_port), + RTE_CACHE_LINE_SIZE); + if (!sh) { + DRV_LOG(ERR, "shared context allocation failure"); + rte_errno = ENOMEM; + goto exit; + } + /* 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"); + } else { + sh->ctx = mlx5_glue->open_device(spawn->ibv_dev); + if (!sh->ctx) { + err = errno ? errno : ENODEV; + goto error; + } + DRV_LOG(DEBUG, "DevX is NOT supported"); + } + err = mlx5_glue->query_device_ex(sh->ctx, NULL, &sh->device_attr); + if (err) { + DRV_LOG(DEBUG, "ibv_query_device_ex() failed"); + goto error; + } + sh->refcnt = 1; + sh->max_port = spawn->max_port; + strncpy(sh->ibdev_name, sh->ctx->device->name, + sizeof(sh->ibdev_name)); + strncpy(sh->ibdev_path, sh->ctx->device->ibdev_path, + sizeof(sh->ibdev_path)); + sh->pci_dev = spawn->pci_dev; + pthread_mutex_init(&sh->intr_mutex, NULL); + /* + * Setting port_id to max unallowed value means + * there is no interrupt subhandler installed for + * the given port index i. + */ + for (i = 0; i < sh->max_port; i++) + sh->port[i].ih_port_id = RTE_MAX_ETHPORTS; + sh->pd = mlx5_glue->alloc_pd(sh->ctx); + if (sh->pd == NULL) { + DRV_LOG(ERR, "PD allocation failure"); + err = ENOMEM; + goto error; + } + /* + * Once the device is added to the list of memory event + * callback, its global MR cache table cannot be expanded + * on the fly because of deadlock. If it overflows, lookup + * should be done by searching MR list linearly, which is slow. + * + * At this point the device is not added to the memory + * event list yet, context is just being created. + */ + err = mlx5_mr_btree_init(&sh->mr.cache, + MLX5_MR_BTREE_CACHE_N * 2, + sh->pci_dev->device.numa_node); + if (err) { + err = rte_errno; + goto error; + } + LIST_INSERT_HEAD(&mlx5_ibv_list, sh, next); +exit: + pthread_mutex_unlock(&mlx5_ibv_list_mutex); + return sh; +error: + pthread_mutex_unlock(&mlx5_ibv_list_mutex); + assert(sh); + if (sh->pd) + claim_zero(mlx5_glue->dealloc_pd(sh->pd)); + if (sh->ctx) + claim_zero(mlx5_glue->close_device(sh->ctx)); + rte_free(sh); + assert(err > 0); + rte_errno = err; + return NULL; +} + +/** + * Free shared IB device context. Decrement counter and if zero free + * all allocated resources and close handles. + * + * @param[in] sh + * Pointer to mlx5_ibv_shared object to free */ static void -mlx5_prepare_shared_data(void) +mlx5_free_shared_ibctx(struct mlx5_ibv_shared *sh) +{ + pthread_mutex_lock(&mlx5_ibv_list_mutex); +#ifndef NDEBUG + /* Check the object presence in the list. */ + struct mlx5_ibv_shared *lctx; + + LIST_FOREACH(lctx, &mlx5_ibv_list, next) + if (lctx == sh) + break; + assert(lctx); + if (lctx != sh) { + DRV_LOG(ERR, "Freeing non-existing shared IB context"); + goto exit; + } +#endif + assert(sh); + assert(sh->refcnt); + /* Secondary process should not free the shared context. */ + assert(rte_eal_process_type() == RTE_PROC_PRIMARY); + if (--sh->refcnt) + goto exit; + /* Release created Memory Regions. */ + mlx5_mr_release(sh); + LIST_REMOVE(sh, next); + /* + * Ensure there is no async event handler installed. + * Only primary process handles async device events. + **/ + assert(!sh->intr_cnt); + if (sh->intr_cnt) + rte_intr_callback_unregister + (&sh->intr_handle, mlx5_dev_interrupt_handler, sh); + pthread_mutex_destroy(&sh->intr_mutex); + if (sh->pd) + claim_zero(mlx5_glue->dealloc_pd(sh->pd)); + if (sh->ctx) + claim_zero(mlx5_glue->close_device(sh->ctx)); + rte_free(sh); +exit: + pthread_mutex_unlock(&mlx5_ibv_list_mutex); +} + +/** + * 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) +{ +#ifdef HAVE_MLX5DV_DR + struct mlx5_ibv_shared *sh = priv->sh; + int err = 0; + void *domain; + + assert(sh); + 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 + 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; + } + return err; +#else + (void)priv; + return 0; +#endif +} + +/** + * 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) +{ +#ifdef HAVE_MLX5DV_DR + struct mlx5_ibv_shared *sh; + + if (!priv->dr_shared) + return; + priv->dr_shared = 0; + sh = priv->sh; + assert(sh); + 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 + pthread_mutex_destroy(&sh->dv_mutex); +#else + (void)priv; +#endif +} + +/** + * Initialize shared data between primary and secondary process. + * + * A memzone is reserved by primary process and secondary processes attach to + * the memzone. + * + * @return + * 0 on success, a negative errno value otherwise and rte_errno is set. + */ +static int +mlx5_init_shared_data(void) { const struct rte_memzone *mz; + int ret = 0; rte_spinlock_lock(&mlx5_shared_data_lock); if (mlx5_shared_data == NULL) { @@ -145,22 +477,31 @@ mlx5_prepare_shared_data(void) mz = rte_memzone_reserve(MZ_MLX5_PMD_SHARED_DATA, sizeof(*mlx5_shared_data), SOCKET_ID_ANY, 0); + if (mz == NULL) { + DRV_LOG(ERR, + "Cannot allocate mlx5 shared data\n"); + ret = -rte_errno; + goto error; + } + mlx5_shared_data = mz->addr; + memset(mlx5_shared_data, 0, sizeof(*mlx5_shared_data)); + rte_spinlock_init(&mlx5_shared_data->lock); } else { /* Lookup allocated shared memory. */ mz = rte_memzone_lookup(MZ_MLX5_PMD_SHARED_DATA); + if (mz == NULL) { + DRV_LOG(ERR, + "Cannot attach mlx5 shared data\n"); + ret = -rte_errno; + goto error; + } + mlx5_shared_data = mz->addr; + memset(&mlx5_local_data, 0, sizeof(mlx5_local_data)); } - if (mz == NULL) - rte_panic("Cannot allocate mlx5 shared data\n"); - mlx5_shared_data = mz->addr; - /* Initialize shared data. */ - if (rte_eal_process_type() == RTE_PROC_PRIMARY) { - LIST_INIT(&mlx5_shared_data->mem_event_cb_list); - rte_rwlock_init(&mlx5_shared_data->mem_event_rwlock); - } - rte_mem_event_callback_register("MLX5_MEM_EVENT_CB", - mlx5_mr_mem_event_cb, NULL); } +error: rte_spinlock_unlock(&mlx5_shared_data_lock); + return ret; } /** @@ -236,6 +577,54 @@ mlx5_free_verbs_buf(void *ptr, void *data __rte_unused) rte_free(ptr); } +/** + * Initialize process private data structure. + * + * @param dev + * Pointer to Ethernet device structure. + * + * @return + * 0 on success, a negative errno value otherwise and rte_errno is set. + */ +int +mlx5_proc_priv_init(struct rte_eth_dev *dev) +{ + struct mlx5_priv *priv = dev->data->dev_private; + struct mlx5_proc_priv *ppriv; + size_t ppriv_size; + + /* + * UAR register table follows the process private structure. BlueFlame + * registers for Tx queues are stored in the table. + */ + ppriv_size = + sizeof(struct mlx5_proc_priv) + priv->txqs_n * sizeof(void *); + ppriv = rte_malloc_socket("mlx5_proc_priv", ppriv_size, + RTE_CACHE_LINE_SIZE, dev->device->numa_node); + if (!ppriv) { + rte_errno = ENOMEM; + return -rte_errno; + } + ppriv->uar_table_sz = ppriv_size; + dev->process_private = ppriv; + return 0; +} + +/** + * Un-initialize process private data structure. + * + * @param dev + * Pointer to Ethernet device structure. + */ +static void +mlx5_proc_priv_uninit(struct rte_eth_dev *dev) +{ + if (!dev->process_private) + return; + rte_free(dev->process_private); + dev->process_private = NULL; +} + /** * DPDK callback to close the device. * @@ -253,7 +642,7 @@ mlx5_dev_close(struct rte_eth_dev *dev) DRV_LOG(DEBUG, "port %u closing device \"%s\"", dev->data->port_id, - ((priv->ctx != NULL) ? priv->ctx->device->name : "")); + ((priv->sh->ctx != NULL) ? priv->sh->ctx->device->name : "")); /* In case mlx5_dev_stop() has not been called. */ mlx5_dev_interrupt_handler_uninstall(dev); mlx5_traffic_disable(dev); @@ -261,6 +650,9 @@ mlx5_dev_close(struct rte_eth_dev *dev) /* Prevent crashes when queues are still in use. */ dev->rx_pkt_burst = removed_rx_burst; dev->tx_pkt_burst = removed_tx_burst; + rte_wmb(); + /* Disable datapath on secondary process. */ + mlx5_mp_req_stop_rxtx(dev); if (priv->rxqs != NULL) { /* XXX race condition if mlx5_rx_burst() is still running. */ usleep(1000); @@ -277,20 +669,18 @@ mlx5_dev_close(struct rte_eth_dev *dev) priv->txqs_n = 0; priv->txqs = NULL; } + mlx5_proc_priv_uninit(dev); mlx5_mprq_free_mp(dev); - mlx5_mr_release(dev); - if (priv->pd != NULL) { - assert(priv->ctx != NULL); - claim_zero(mlx5_glue->dealloc_pd(priv->pd)); - claim_zero(mlx5_glue->close_device(priv->ctx)); - } else - assert(priv->ctx == NULL); + /* Remove from memory callback device list. */ + rte_rwlock_write_lock(&mlx5_shared_data->mem_event_rwlock); + assert(priv->sh); + LIST_REMOVE(priv->sh, mem_event_cb); + rte_rwlock_write_unlock(&mlx5_shared_data->mem_event_rwlock); + mlx5_free_shared_dr(priv); if (priv->rss_conf.rss_key != NULL) rte_free(priv->rss_conf.rss_key); if (priv->reta_idx != NULL) rte_free(priv->reta_idx); - if (priv->primary_socket) - mlx5_socket_uninit(dev); if (priv->config.vf) mlx5_nl_mac_addr_flush(dev); if (priv->nl_socket_route >= 0) @@ -299,6 +689,16 @@ mlx5_dev_close(struct rte_eth_dev *dev) close(priv->nl_socket_rdma); if (priv->tcf_context) mlx5_flow_tcf_context_destroy(priv->tcf_context); + if (priv->sh) { + /* + * Free the shared context in last turn, because the cleanup + * routines above may use some shared fields, like + * mlx5_nl_mac_addr_flush() uses ibdev_path for retrieveing + * ifindex if Netlink fails. + */ + mlx5_free_shared_ibctx(priv->sh); + priv->sh = NULL; + } ret = mlx5_hrxq_ibv_verify(dev); if (ret) DRV_LOG(WARNING, "port %u some hash Rx queue still remain", @@ -329,17 +729,15 @@ mlx5_dev_close(struct rte_eth_dev *dev) dev->data->port_id); if (priv->domain_id != RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID) { unsigned int c = 0; - unsigned int i = mlx5_dev_to_port_id(dev->device, NULL, 0); - uint16_t port_id[i]; + uint16_t port_id; - i = RTE_MIN(mlx5_dev_to_port_id(dev->device, port_id, i), i); - while (i--) { + RTE_ETH_FOREACH_DEV_OF(port_id, dev->device) { struct mlx5_priv *opriv = - rte_eth_devices[port_id[i]].data->dev_private; + rte_eth_devices[port_id].data->dev_private; if (!opriv || opriv->domain_id != priv->domain_id || - &rte_eth_devices[port_id[i]] == dev) + &rte_eth_devices[port_id] == dev) continue; ++c; } @@ -522,8 +920,12 @@ mlx5_args_check(const char *key, const char *val, void *opaque) config->l3_vxlan_en = !!tmp; } else if (strcmp(MLX5_VF_NL_EN, key) == 0) { config->vf_nl_en = !!tmp; + } else if (strcmp(MLX5_DV_ESW_EN, key) == 0) { + config->dv_esw_en = !!tmp; } else if (strcmp(MLX5_DV_FLOW_EN, key) == 0) { config->dv_flow_en = !!tmp; + } else if (strcmp(MLX5_MR_EXT_MEMSEG_EN, key) == 0) { + config->mr_ext_memseg_en = !!tmp; } else { DRV_LOG(WARNING, "%s: unknown parameter", key); rte_errno = EINVAL; @@ -564,7 +966,9 @@ mlx5_args(struct mlx5_dev_config *config, struct rte_devargs *devargs) MLX5_RX_VEC_EN, MLX5_L3_VXLAN_EN, MLX5_VF_NL_EN, + MLX5_DV_ESW_EN, MLX5_DV_FLOW_EN, + MLX5_MR_EXT_MEMSEG_EN, MLX5_REPRESENTOR, NULL, }; @@ -596,118 +1000,49 @@ mlx5_args(struct mlx5_dev_config *config, struct rte_devargs *devargs) static struct rte_pci_driver mlx5_driver; -/* - * Reserved UAR address space for TXQ UAR(hw doorbell) mapping, process - * local resource used by both primary and secondary to avoid duplicate - * reservation. - * The space has to be available on both primary and secondary process, - * TXQ UAR maps to this area using fixed mmap w/o double check. - */ -static void *uar_base; - -static int -find_lower_va_bound(const struct rte_memseg_list *msl, - const struct rte_memseg *ms, void *arg) -{ - void **addr = arg; - - if (msl->external) - return 0; - if (*addr == NULL) - *addr = ms->addr; - else - *addr = RTE_MIN(*addr, ms->addr); - - return 0; -} - -/** - * Reserve UAR address space for primary process. - * - * @param[in] dev - * Pointer to Ethernet device. - * - * @return - * 0 on success, a negative errno value otherwise and rte_errno is set. - */ -static int -mlx5_uar_init_primary(struct rte_eth_dev *dev) -{ - struct mlx5_priv *priv = dev->data->dev_private; - void *addr = (void *)0; - - if (uar_base) { /* UAR address space mapped. */ - priv->uar_base = uar_base; - return 0; - } - /* find out lower bound of hugepage segments */ - rte_memseg_walk(find_lower_va_bound, &addr); - - /* keep distance to hugepages to minimize potential conflicts. */ - addr = RTE_PTR_SUB(addr, (uintptr_t)(MLX5_UAR_OFFSET + MLX5_UAR_SIZE)); - /* anonymous mmap, no real memory consumption. */ - addr = mmap(addr, MLX5_UAR_SIZE, - PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); - if (addr == MAP_FAILED) { - DRV_LOG(ERR, - "port %u failed to reserve UAR address space, please" - " adjust MLX5_UAR_SIZE or try --base-virtaddr", - dev->data->port_id); - rte_errno = ENOMEM; - return -rte_errno; - } - /* Accept either same addr or a new addr returned from mmap if target - * range occupied. - */ - DRV_LOG(INFO, "port %u reserved UAR address space: %p", - dev->data->port_id, addr); - priv->uar_base = addr; /* for primary and secondary UAR re-mmap. */ - uar_base = addr; /* process local, don't reserve again. */ - return 0; -} - /** - * Reserve UAR address space for secondary process, align with - * primary process. + * PMD global initialization. * - * @param[in] dev - * Pointer to Ethernet device. + * Independent from individual device, this function initializes global + * per-PMD data structures distinguishing primary and secondary processes. + * Hence, each initialization is called once per a process. * * @return * 0 on success, a negative errno value otherwise and rte_errno is set. */ static int -mlx5_uar_init_secondary(struct rte_eth_dev *dev) +mlx5_init_once(void) { - struct mlx5_priv *priv = dev->data->dev_private; - void *addr; + struct mlx5_shared_data *sd; + struct mlx5_local_data *ld = &mlx5_local_data; - assert(priv->uar_base); - if (uar_base) { /* already reserved. */ - assert(uar_base == priv->uar_base); - return 0; - } - /* anonymous mmap, no real memory consumption. */ - addr = mmap(priv->uar_base, MLX5_UAR_SIZE, - PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); - if (addr == MAP_FAILED) { - DRV_LOG(ERR, "port %u UAR mmap failed: %p size: %llu", - dev->data->port_id, priv->uar_base, MLX5_UAR_SIZE); - rte_errno = ENXIO; - return -rte_errno; - } - if (priv->uar_base != addr) { - DRV_LOG(ERR, - "port %u UAR address %p size %llu occupied, please" - " adjust MLX5_UAR_OFFSET or try EAL parameter" - " --base-virtaddr", - dev->data->port_id, priv->uar_base, MLX5_UAR_SIZE); - rte_errno = ENXIO; + if (mlx5_init_shared_data()) return -rte_errno; + sd = mlx5_shared_data; + assert(sd); + rte_spinlock_lock(&sd->lock); + switch (rte_eal_process_type()) { + case RTE_PROC_PRIMARY: + if (sd->init_done) + break; + LIST_INIT(&sd->mem_event_cb_list); + rte_rwlock_init(&sd->mem_event_rwlock); + rte_mem_event_callback_register("MLX5_MEM_EVENT_CB", + mlx5_mr_mem_event_cb, NULL); + mlx5_mp_init_primary(); + sd->init_done = true; + break; + case RTE_PROC_SECONDARY: + if (ld->init_done) + break; + mlx5_mp_init_secondary(); + ++sd->secondary_cnt; + ld->init_done = true; + break; + default: + break; } - uar_base = addr; /* process local, don't reserve again */ - DRV_LOG(INFO, "port %u reserved UAR address space: %p", - dev->data->port_id, addr); + rte_spinlock_unlock(&sd->lock); return 0; } @@ -716,12 +1051,10 @@ mlx5_uar_init_secondary(struct rte_eth_dev *dev) * * @param dpdk_dev * Backing DPDK device. - * @param ibv_dev - * Verbs device. + * @param spawn + * Verbs device parameters (name, port, switch_info) to spawn. * @param config * Device configuration parameters. - * @param[in] switch_info - * Switch properties of Ethernet device. * * @return * A valid Ethernet device object on success, NULL otherwise and rte_errno @@ -732,14 +1065,12 @@ mlx5_uar_init_secondary(struct rte_eth_dev *dev) */ static struct rte_eth_dev * mlx5_dev_spawn(struct rte_device *dpdk_dev, - struct ibv_device *ibv_dev, - struct mlx5_dev_config config, - const struct mlx5_switch_info *switch_info) + struct mlx5_dev_spawn_data *spawn, + struct mlx5_dev_config config) { - struct ibv_context *ctx = NULL; - struct ibv_device_attr_ex attr; + const struct mlx5_switch_info *switch_info = &spawn->info; + struct mlx5_ibv_shared *sh = NULL; struct ibv_port_attr port_attr; - struct ibv_pd *pd = NULL; struct mlx5dv_context dv_attr = { .comp_mask = 0 }; struct rte_eth_dev *eth_dev = NULL; struct mlx5_priv *priv = NULL; @@ -756,7 +1087,7 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev, 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 ether_addr mac; + struct rte_ether_addr mac; char name[RTE_ETH_NAME_MAX_LEN]; int own_domain_id = 0; uint16_t port_id; @@ -793,20 +1124,40 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev, rte_errno = EEXIST; return NULL; } - /* Prepare shared data between primary and secondary process. */ - mlx5_prepare_shared_data(); - errno = 0; - ctx = mlx5_glue->dv_open_device(ibv_dev); - if (ctx) { - config.devx = 1; - DRV_LOG(DEBUG, "DEVX is supported"); - } else { - ctx = mlx5_glue->open_device(ibv_dev); - if (!ctx) { - rte_errno = errno ? errno : ENODEV; + DRV_LOG(DEBUG, "naming Ethernet device \"%s\"", name); + if (rte_eal_process_type() == RTE_PROC_SECONDARY) { + 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; + /* Receive command fd from primary process */ + err = mlx5_mp_req_verbs_cmd_fd(eth_dev); + if (err < 0) + return NULL; + /* Remap UAR for Tx queues. */ + err = mlx5_tx_uar_init_secondary(eth_dev, err); + if (err) + return NULL; + /* + * 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; } + sh = mlx5_alloc_shared_ibctx(spawn); + if (!sh) + return NULL; + config.devx = sh->devx; #ifdef HAVE_IBV_MLX5_MOD_SWP dv_attr.comp_mask |= MLX5DV_CONTEXT_MASK_SWP; #endif @@ -820,7 +1171,7 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev, #ifdef HAVE_IBV_DEVICE_STRIDING_RQ_SUPPORT dv_attr.comp_mask |= MLX5DV_CONTEXT_MASK_STRIDING_RQ; #endif - mlx5_glue->dv_query_device(ctx, &dv_attr); + 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"); @@ -905,51 +1256,8 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev, " old OFED/rdma-core version or firmware configuration"); #endif config.mpls_en = mpls_en; - err = mlx5_glue->query_device_ex(ctx, NULL, &attr); - if (err) { - DEBUG("ibv_query_device_ex() failed"); - goto error; - } - DRV_LOG(DEBUG, "naming Ethernet device \"%s\"", name); - if (rte_eal_process_type() == RTE_PROC_SECONDARY) { - eth_dev = rte_eth_dev_attach_secondary(name); - if (eth_dev == NULL) { - DRV_LOG(ERR, "can not attach rte ethdev"); - rte_errno = ENOMEM; - err = rte_errno; - goto error; - } - eth_dev->device = dpdk_dev; - eth_dev->dev_ops = &mlx5_dev_sec_ops; - err = mlx5_uar_init_secondary(eth_dev); - if (err) { - err = rte_errno; - goto error; - } - /* Receive command fd from primary process */ - err = mlx5_socket_connect(eth_dev); - if (err < 0) { - err = rte_errno; - goto error; - } - /* Remap UAR for Tx queues. */ - err = mlx5_tx_uar_remap(eth_dev, err); - if (err) { - err = rte_errno; - goto error; - } - /* - * 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); - claim_zero(mlx5_glue->close_device(ctx)); - return eth_dev; - } /* Check port status. */ - err = mlx5_glue->query_port(ctx, 1, &port_attr); + 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; @@ -963,13 +1271,7 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev, DRV_LOG(DEBUG, "port is not active: \"%s\" (%d)", mlx5_glue->port_state_str(port_attr.state), port_attr.state); - /* Allocate protection domain. */ - pd = mlx5_glue->alloc_pd(ctx); - if (pd == NULL) { - DRV_LOG(ERR, "PD allocation failure"); - err = ENOMEM; - goto error; - } + /* Allocate private eth device data. */ priv = rte_zmalloc("ethdev private structure", sizeof(*priv), RTE_CACHE_LINE_SIZE); @@ -978,14 +1280,9 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev, err = ENOMEM; goto error; } - priv->ctx = ctx; - strncpy(priv->ibdev_name, priv->ctx->device->name, - sizeof(priv->ibdev_name)); - strncpy(priv->ibdev_path, priv->ctx->device->ibdev_path, - sizeof(priv->ibdev_path)); - priv->device_attr = attr; - priv->pd = pd; - priv->mtu = ETHER_MTU; + priv->sh = sh; + priv->ibv_port = spawn->ibv_port; + priv->mtu = RTE_ETHER_MTU; #ifndef RTE_ARCH_64 /* Initialize UAR access locks for 32bit implementations. */ rte_spinlock_init(&priv->uar_lock_cq); @@ -1003,7 +1300,7 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev, * Currently we 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 exapmple, let's have the IB device port 10, it has + * 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 @@ -1019,22 +1316,16 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev, * Look for sibling devices in order to reuse their switch domain * if any, otherwise allocate one. */ - i = mlx5_dev_to_port_id(dpdk_dev, NULL, 0); - if (i > 0) { - uint16_t port_id[i]; - - i = RTE_MIN(mlx5_dev_to_port_id(dpdk_dev, port_id, i), i); - while (i--) { - const struct mlx5_priv *opriv = - rte_eth_devices[port_id[i]].data->dev_private; + RTE_ETH_FOREACH_DEV_OF(port_id, dpdk_dev) { + const struct mlx5_priv *opriv = + rte_eth_devices[port_id].data->dev_private; - if (!opriv || - opriv->domain_id == - RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID) - continue; - priv->domain_id = opriv->domain_id; - break; - } + if (!opriv || + 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); @@ -1053,7 +1344,8 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev, strerror(rte_errno)); goto error; } - config.hw_csum = !!(attr.device_cap_flags_ex & IBV_DEVICE_RAW_IP_CSUM); + 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) && \ @@ -1067,7 +1359,7 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev, } #endif config.ind_table_max_size = - attr.rss_caps.max_rwq_indirection_table_size; + sh->device_attr.rss_caps.max_rwq_indirection_table_size; /* * Remove this check once DPDK supports larger/variable * indirection tables. @@ -1076,18 +1368,18 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev, 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 = !!(attr.raw_packet_caps & + 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 = !!(attr.raw_packet_caps & + 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 = !!attr.rx_pad_end_addr_align; + hw_padding = !!sh->device_attr.rx_pad_end_addr_align; #elif defined(HAVE_IBV_WQ_FLAGS_PCI_WRITE_END_PADDING) - hw_padding = !!(attr.device_cap_flags_ex & + hw_padding = !!(sh->device_attr.device_cap_flags_ex & IBV_DEVICE_PCI_WRITE_END_PADDING); #endif if (config.hw_padding && !hw_padding) { @@ -1096,11 +1388,11 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev, } else if (config.hw_padding) { DRV_LOG(DEBUG, "Rx end alignment padding is enabled"); } - config.tso = (attr.tso_caps.max_tso > 0 && - (attr.tso_caps.supported_qpts & + config.tso = (sh->device_attr.tso_caps.max_tso > 0 && + (sh->device_attr.tso_caps.supported_qpts & (1 << IBV_QPT_RAW_PACKET))); if (config.tso) - config.tso_max_payload_sz = attr.tso_caps.max_tso; + config.tso_max_payload_sz = sh->device_attr.tso_caps.max_tso; /* * MPW is disabled by default, while the Enhanced MPW is enabled * by default. @@ -1157,11 +1449,6 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev, priv->dev_data = eth_dev->data; eth_dev->data->mac_addrs = priv->mac; eth_dev->device = dpdk_dev; - err = mlx5_uar_init_primary(eth_dev); - if (err) { - err = rte_errno; - goto error; - } /* Configure the first MAC address by default. */ if (mlx5_get_mac(eth_dev, &mac.addr_bytes)) { DRV_LOG(ERR, @@ -1241,7 +1528,8 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev, .free = &mlx5_free_verbs_buf, .data = priv, }; - mlx5_glue->dv_set_context_attr(ctx, MLX5DV_CTX_ATTR_BUF_ALLOCATORS, + 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", @@ -1249,12 +1537,33 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev, mlx5_set_link_up(eth_dev); /* * Even though the interrupt handler is not installed yet, - * interrupts will still trigger on the asyn_fd from + * interrupts will still trigger on the async_fd from * Verbs context returned by ibv_open_device(). */ mlx5_link_update(eth_dev, 0); +#ifdef HAVE_IBV_DEVX_OBJ + if (config.devx) { + err = mlx5_devx_cmd_query_hca_attr(sh->ctx, &config.hca_attr); + if (err) { + err = -err; + goto error; + } + } +#endif +#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 /* Store device configuration on private structure. */ priv->config = config; + if (config.dv_flow_en) { + err = mlx5_alloc_shared_dr(priv); + if (err) + goto error; + } /* Supported Verbs flow priority number detection. */ err = mlx5_flow_discover_priorities(eth_dev); if (err < 0) { @@ -1262,27 +1571,16 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev, goto error; } priv->config.flow_prio = err; - /* - * Once the device is added to the list of memory event - * callback, its global MR cache table cannot be expanded - * on the fly because of deadlock. If it overflows, lookup - * should be done by searching MR list linearly, which is slow. - */ - err = mlx5_mr_btree_init(&priv->mr.cache, - MLX5_MR_BTREE_CACHE_N * 2, - eth_dev->device->numa_node); - if (err) { - err = rte_errno; - goto error; - } /* Add device to memory callback list. */ rte_rwlock_write_lock(&mlx5_shared_data->mem_event_rwlock); LIST_INSERT_HEAD(&mlx5_shared_data->mem_event_cb_list, - priv, mem_event_cb); + sh, mem_event_cb); rte_rwlock_write_unlock(&mlx5_shared_data->mem_event_rwlock); return eth_dev; error: if (priv) { + 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) @@ -1295,28 +1593,18 @@ error: if (eth_dev != NULL) eth_dev->data->dev_private = NULL; } - if (pd) - claim_zero(mlx5_glue->dealloc_pd(pd)); 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 (ctx) - claim_zero(mlx5_glue->close_device(ctx)); + if (sh) + mlx5_free_shared_ibctx(sh); assert(err > 0); rte_errno = err; return NULL; } -/** Data associated with devices to spawn. */ -struct mlx5_dev_spawn_data { - unsigned int ifindex; /**< Network interface index. */ - struct mlx5_switch_info info; /**< Switch information. */ - struct ibv_device *ibv_dev; /**< Associated IB device. */ - struct rte_eth_dev *eth_dev; /**< Associated Ethernet device. */ -}; - /** * Comparison callback to sort device data. * @@ -1373,10 +1661,33 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv __rte_unused, struct rte_pci_device *pci_dev) { struct ibv_device **ibv_list; - unsigned int n = 0; + /* + * 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; struct mlx5_dev_config dev_config; int ret; + ret = mlx5_init_once(); + if (ret) { + DRV_LOG(ERR, "unable to init PMD global data: %s", + strerror(rte_errno)); + return -rte_errno; + } assert(pci_drv == &mlx5_driver); errno = 0; ibv_list = mlx5_glue->get_device_list(&ret); @@ -1385,8 +1696,14 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv __rte_unused, 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 = -1; + int nl_rdma = -1; + unsigned int i; while (ret-- > 0) { struct rte_pci_addr pci_addr; @@ -1401,77 +1718,213 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv __rte_unused, continue; DRV_LOG(INFO, "PCI information matches for device \"%s\"", ibv_list[ret]->name); - ibv_match[n++] = ibv_list[ret]; + ibv_match[nd++] = ibv_list[ret]; + } + ibv_match[nd] = NULL; + if (!nd) { + /* No device matches, just complain and bail out. */ + mlx5_glue->free_device_list(ibv_list); + 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; + return ret; + } + nl_route = mlx5_nl_init(NETLINK_ROUTE); + nl_rdma = mlx5_nl_init(NETLINK_RDMA); + 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); } - ibv_match[n] = NULL; - - struct mlx5_dev_spawn_data list[n]; - int nl_route = n ? mlx5_nl_init(NETLINK_ROUTE) : -1; - int nl_rdma = n ? mlx5_nl_init(NETLINK_RDMA) : -1; - unsigned int i; - unsigned int u; - /* - * The existence of several matching entries (n > 1) means port - * representors have been instantiated. No existing Verbs call nor - * /sys 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 either: - * - * 1. No device matches (n == 0), complain and bail out. - * 2. A single IB device matches (n == 1) and is not a representor, - * assume no switch support. - * 3. Otherwise no safe assumptions can be made; complain louder and - * bail out. + * Now we can determine the maximal + * amount of devices to be spawned. */ - for (i = 0; i != n; ++i) { - list[i].ibv_dev = ibv_match[i]; - list[i].eth_dev = NULL; - if (nl_rdma < 0) - list[i].ifindex = 0; - else - list[i].ifindex = mlx5_nl_ifindex - (nl_rdma, list[i].ibv_dev->name, 1); - if (nl_route < 0 || - !list[i].ifindex || - mlx5_nl_switch_info(nl_route, list[i].ifindex, - &list[i].info) || - ((!list[i].info.representor && !list[i].info.master) && - mlx5_sysfs_switch_info(list[i].ifindex, &list[i].info))) { - list[i].ifindex = 0; - memset(&list[i].info, 0, sizeof(list[i].info)); - continue; + struct mlx5_dev_spawn_data list[np ? np : nd]; + + if (np > 1) { + /* + * Single IB device with multiple ports found, + * it may be E-Switch master device and representors. + * We have to perform identification trough the ports. + */ + assert(nl_rdma >= 0); + assert(ns == 0); + assert(nd == 1); + 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].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 && (list[ns].info.representor ^ + list[ns].info.master)) + ns++; } - } - if (nl_rdma >= 0) - close(nl_rdma); - if (nl_route >= 0) - close(nl_route); - /* Count unidentified devices. */ - for (u = 0, i = 0; i != n; ++i) - if (!list[i].info.master && !list[i].info.representor) - ++u; - if (u) { - if (n == 1 && u == 1) { - /* Case #2. */ - DRV_LOG(INFO, "no switch support detected"); - } else { - /* Case #3. */ + 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].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 tell which of the matching devices" - " is the master (lack of kernel support?)"); - n = 0; + "unable to recognize master/representors" + " on the multiple IB devices"); + rte_errno = ENOENT; + ret = -rte_errno; + goto exit; } } + assert(ns); /* * Sort list to probe devices in natural order for users convenience * (i.e. master first, then representors from lowest to highest ID). */ - if (n) - qsort(list, n, sizeof(*list), mlx5_dev_spawn_data_cmp); + qsort(list, ns, sizeof(*list), mlx5_dev_spawn_data_cmp); /* Default configuration. */ dev_config = (struct mlx5_dev_config){ .hw_padding = 0, @@ -1483,14 +1936,16 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv __rte_unused, .txqs_vec = MLX5_ARG_UNSET, .inline_max_packet_sz = MLX5_ARG_UNSET, .vf_nl_en = 1, + .mr_ext_memseg_en = 1, .mprq = { .enabled = 0, /* Disabled by default. */ .stride_num_n = MLX5_MPRQ_STRIDE_NUM_N, .max_memcpy_len = MLX5_MPRQ_MEMCPY_DEFAULT_LEN, .min_rxqs_num = MLX5_MPRQ_MIN_RXQS, }, + .dv_esw_en = 1, }; - /* Device speicific configuration. */ + /* Device specific configuration. */ switch (pci_dev->id.device_id) { case PCI_DEVICE_ID_MELLANOX_CONNECTX5BF: dev_config.txqs_vec = MLX5_VPMD_MAX_TXQS_BLUEFIELD; @@ -1507,12 +1962,12 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv __rte_unused, /* Set architecture-dependent default value if unset. */ if (dev_config.txqs_vec == MLX5_ARG_UNSET) dev_config.txqs_vec = MLX5_VPMD_MAX_TXQS; - for (i = 0; i != n; ++i) { + for (i = 0; i != ns; ++i) { uint32_t restore; list[i].eth_dev = mlx5_dev_spawn(&pci_dev->device, - list[i].ibv_dev, dev_config, - &list[i].info); + &list[i], + dev_config); if (!list[i].eth_dev) { if (rte_errno != EBUSY && rte_errno != EEXIST) break; @@ -1525,16 +1980,7 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv __rte_unused, list[i].eth_dev->data->dev_flags |= restore; rte_eth_dev_probing_finish(list[i].eth_dev); } - mlx5_glue->free_device_list(ibv_list); - if (!n) { - 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; - } else if (i != n) { + if (i != ns) { DRV_LOG(ERR, "probe of PCI device " PCI_PRI_FMT " aborted after" " encountering an error: %s", @@ -1556,6 +2002,18 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv __rte_unused, } else { ret = 0; } +exit: + /* + * Do the routine cleanup: + * - close opened Netlink sockets + * - free the Infiniband device list + */ + if (nl_rdma >= 0) + close(nl_rdma); + if (nl_route >= 0) + close(nl_route); + assert(ibv_list); + mlx5_glue->free_device_list(ibv_list); return ret; } @@ -1574,14 +2032,9 @@ static int mlx5_pci_remove(struct rte_pci_device *pci_dev) { uint16_t port_id; - struct rte_eth_dev *port; - for (port_id = 0; port_id < RTE_MAX_ETHPORTS; port_id++) { - port = &rte_eth_devices[port_id]; - if (port->state != RTE_ETH_DEV_UNUSED && - port->device == &pci_dev->device) - rte_eth_dev_close(port_id); - } + RTE_ETH_FOREACH_DEV_OF(port_id, &pci_dev->device) + rte_eth_dev_close(port_id); return 0; }