X-Git-Url: http://git.droids-corp.org/?a=blobdiff_plain;f=drivers%2Fnet%2Fmlx4%2Fmlx4.c;h=7f07b8dc09f149a07a56af036584b01f67e0d14b;hb=34f3966c7f81f947e9eebb347dec6a9f68eec4e6;hp=f389e66427db4426e822ed187bcd780f1e89f464;hpb=5feecc57d90b97c579b16d1083ea167f7564530b;p=dpdk.git diff --git a/drivers/net/mlx4/mlx4.c b/drivers/net/mlx4/mlx4.c index f389e66427..7f07b8dc09 100644 --- a/drivers/net/mlx4/mlx4.c +++ b/drivers/net/mlx4/mlx4.c @@ -44,9 +44,15 @@ #include "mlx4.h" #include "mlx4_glue.h" #include "mlx4_flow.h" +#include "mlx4_mr.h" #include "mlx4_rxtx.h" #include "mlx4_utils.h" +struct mlx4_dev_list mlx4_mem_event_cb_list = + LIST_HEAD_INITIALIZER(mlx4_mem_event_cb_list); + +rte_rwlock_t mlx4_mem_event_rwlock = RTE_RWLOCK_INITIALIZER; + /** Configuration structure for device arguments. */ struct mlx4_conf { struct { @@ -61,6 +67,8 @@ const char *pmd_mlx4_init_params[] = { NULL, }; +static void mlx4_dev_stop(struct rte_eth_dev *dev); + /** * DPDK callback for Ethernet device configuration. * @@ -123,6 +131,9 @@ mlx4_dev_start(struct rte_eth_dev *dev) (void *)dev, strerror(-ret)); goto err; } +#ifndef NDEBUG + mlx4_mr_dump_dev(dev); +#endif ret = mlx4_rxq_intr_enable(priv); if (ret) { ERROR("%p: interrupt handler installation failed", @@ -143,8 +154,7 @@ mlx4_dev_start(struct rte_eth_dev *dev) dev->rx_pkt_burst = mlx4_rx_burst; return 0; err: - /* Rollback. */ - priv->started = 0; + mlx4_dev_stop(dev); return ret; } @@ -194,10 +204,12 @@ mlx4_dev_close(struct rte_eth_dev *dev) dev->tx_pkt_burst = mlx4_tx_burst_removed; rte_wmb(); mlx4_flow_clean(priv); + mlx4_rss_deinit(priv); for (i = 0; i != dev->data->nb_rx_queues; ++i) mlx4_rx_queue_release(dev->data->rx_queues[i]); for (i = 0; i != dev->data->nb_tx_queues; ++i) mlx4_tx_queue_release(dev->data->tx_queues[i]); + mlx4_mr_release(dev); if (priv->pd != NULL) { assert(priv->ctx != NULL); claim_zero(mlx4_glue->dealloc_pd(priv->pd)); @@ -385,6 +397,99 @@ free_kvlist: return ret; } +/** + * Interpret RSS capabilities reported by device. + * + * This function returns the set of usable Verbs RSS hash fields, kernel + * quirks taken into account. + * + * @param ctx + * Verbs context. + * @param pd + * Verbs protection domain. + * @param device_attr_ex + * Extended device attributes to interpret. + * + * @return + * Usable RSS hash fields mask in Verbs format. + */ +static uint64_t +mlx4_hw_rss_sup(struct ibv_context *ctx, struct ibv_pd *pd, + struct ibv_device_attr_ex *device_attr_ex) +{ + uint64_t hw_rss_sup = device_attr_ex->rss_caps.rx_hash_fields_mask; + struct ibv_cq *cq = NULL; + struct ibv_wq *wq = NULL; + struct ibv_rwq_ind_table *ind = NULL; + struct ibv_qp *qp = NULL; + + if (!hw_rss_sup) { + WARN("no RSS capabilities reported; disabling support for UDP" + " RSS and inner VXLAN RSS"); + return IBV_RX_HASH_SRC_IPV4 | IBV_RX_HASH_DST_IPV4 | + IBV_RX_HASH_SRC_IPV6 | IBV_RX_HASH_DST_IPV6 | + IBV_RX_HASH_SRC_PORT_TCP | IBV_RX_HASH_DST_PORT_TCP; + } + if (!(hw_rss_sup & IBV_RX_HASH_INNER)) + return hw_rss_sup; + /* + * Although reported as supported, missing code in some Linux + * versions (v4.15, v4.16) prevents the creation of hash QPs with + * inner capability. + * + * There is no choice but to attempt to instantiate a temporary RSS + * context in order to confirm its support. + */ + cq = mlx4_glue->create_cq(ctx, 1, NULL, NULL, 0); + wq = cq ? mlx4_glue->create_wq + (ctx, + &(struct ibv_wq_init_attr){ + .wq_type = IBV_WQT_RQ, + .max_wr = 1, + .max_sge = 1, + .pd = pd, + .cq = cq, + }) : NULL; + ind = wq ? mlx4_glue->create_rwq_ind_table + (ctx, + &(struct ibv_rwq_ind_table_init_attr){ + .log_ind_tbl_size = 0, + .ind_tbl = &wq, + .comp_mask = 0, + }) : NULL; + qp = ind ? mlx4_glue->create_qp_ex + (ctx, + &(struct ibv_qp_init_attr_ex){ + .comp_mask = + (IBV_QP_INIT_ATTR_PD | + IBV_QP_INIT_ATTR_RX_HASH | + IBV_QP_INIT_ATTR_IND_TABLE), + .qp_type = IBV_QPT_RAW_PACKET, + .pd = pd, + .rwq_ind_tbl = ind, + .rx_hash_conf = { + .rx_hash_function = IBV_RX_HASH_FUNC_TOEPLITZ, + .rx_hash_key_len = MLX4_RSS_HASH_KEY_SIZE, + .rx_hash_key = mlx4_rss_hash_key_default, + .rx_hash_fields_mask = hw_rss_sup, + }, + }) : NULL; + if (!qp) { + WARN("disabling unusable inner RSS capability due to kernel" + " quirk"); + hw_rss_sup &= ~IBV_RX_HASH_INNER; + } else { + claim_zero(mlx4_glue->destroy_qp(qp)); + } + if (ind) + claim_zero(mlx4_glue->destroy_rwq_ind_table(ind)); + if (wq) + claim_zero(mlx4_glue->destroy_wq(wq)); + if (cq) + claim_zero(mlx4_glue->destroy_cq(cq)); + return hw_rss_sup; +} + static struct rte_pci_driver mlx4_driver; /** @@ -470,14 +575,14 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) ibv_dev = list[i]; DEBUG("device opened"); if (mlx4_glue->query_device(attr_ctx, &device_attr)) { - rte_errno = ENODEV; + err = ENODEV; goto error; } INFO("%u port(s) detected", device_attr.phys_port_cnt); conf.ports.present |= (UINT64_C(1) << device_attr.phys_port_cnt) - 1; if (mlx4_args(pci_dev->device.devargs, &conf)) { ERROR("failed to process device arguments"); - rte_errno = EINVAL; + err = EINVAL; goto error; } /* Use all ports when none are defined */ @@ -485,7 +590,7 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) conf.ports.enabled = conf.ports.present; /* Retrieve extended device attributes. */ if (mlx4_glue->query_device_ex(attr_ctx, NULL, &device_attr_ex)) { - rte_errno = ENODEV; + err = ENODEV; goto error; } assert(device_attr.max_sge >= MLX4_MAX_SGE); @@ -504,18 +609,18 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) DEBUG("using port %u", port); ctx = mlx4_glue->open_device(ibv_dev); if (ctx == NULL) { - rte_errno = ENODEV; + err = ENODEV; goto port_error; } /* Check port status. */ err = mlx4_glue->query_port(ctx, port, &port_attr); if (err) { - rte_errno = err; - ERROR("port query failed: %s", strerror(rte_errno)); + err = ENODEV; + ERROR("port query failed: %s", strerror(err)); goto port_error; } if (port_attr.link_layer != IBV_LINK_LAYER_ETHERNET) { - rte_errno = ENOTSUP; + err = ENOTSUP; ERROR("port %d is not configured in Ethernet mode", port); goto port_error; @@ -525,15 +630,16 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) port, mlx4_glue->port_state_str(port_attr.state), port_attr.state); /* Make asynchronous FD non-blocking to handle interrupts. */ - if (mlx4_fd_set_non_blocking(ctx->async_fd) < 0) { + err = mlx4_fd_set_non_blocking(ctx->async_fd); + if (err) { ERROR("cannot make asynchronous FD non-blocking: %s", - strerror(rte_errno)); + strerror(err)); goto port_error; } /* Allocate protection domain. */ pd = mlx4_glue->alloc_pd(ctx); if (pd == NULL) { - rte_errno = ENOMEM; + err = ENOMEM; ERROR("PD allocation failure"); goto port_error; } @@ -542,7 +648,7 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) sizeof(*priv), RTE_CACHE_LINE_SIZE); if (priv == NULL) { - rte_errno = ENOMEM; + err = ENOMEM; ERROR("priv allocation failure"); goto port_error; } @@ -562,26 +668,32 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) (device_attr.vendor_part_id == PCI_DEVICE_ID_MELLANOX_CONNECTX3PRO); DEBUG("L2 tunnel checksum offloads are %ssupported", - (priv->hw_csum_l2tun ? "" : "not ")); - priv->hw_rss_sup = device_attr_ex.rss_caps.rx_hash_fields_mask; - if (!priv->hw_rss_sup) { - WARN("no RSS capabilities reported; disabling support" - " for UDP RSS and inner VXLAN RSS"); - /* Fake support for all possible RSS hash fields. */ - priv->hw_rss_sup = ~UINT64_C(0); - priv->hw_rss_sup = mlx4_conv_rss_hf(priv, -1); - /* Filter out known unsupported fields. */ - priv->hw_rss_sup &= - ~(uint64_t)(IBV_RX_HASH_SRC_PORT_UDP | - IBV_RX_HASH_DST_PORT_UDP | - IBV_RX_HASH_INNER); - } + priv->hw_csum_l2tun ? "" : "not "); + priv->hw_rss_sup = mlx4_hw_rss_sup(priv->ctx, priv->pd, + &device_attr_ex); DEBUG("supported RSS hash fields mask: %016" PRIx64, priv->hw_rss_sup); + priv->hw_rss_max_qps = + device_attr_ex.rss_caps.max_rwq_indirection_table_size; + DEBUG("MAX RSS queues %d", priv->hw_rss_max_qps); + priv->hw_fcs_strip = !!(device_attr_ex.raw_packet_caps & + IBV_RAW_PACKET_CAP_SCATTER_FCS); + DEBUG("FCS stripping toggling is %ssupported", + priv->hw_fcs_strip ? "" : "not "); + priv->tso = + ((device_attr_ex.tso_caps.max_tso > 0) && + (device_attr_ex.tso_caps.supported_qpts & + (1 << IBV_QPT_RAW_PACKET))); + if (priv->tso) + priv->tso_max_payload_sz = + device_attr_ex.tso_caps.max_tso; + DEBUG("TSO is %ssupported", + priv->tso ? "" : "not "); /* Configure the first MAC address by default. */ - if (mlx4_get_mac(priv, &mac.addr_bytes)) { + err = mlx4_get_mac(priv, &mac.addr_bytes); + if (err) { ERROR("cannot get MAC address, is mlx4_en loaded?" - " (rte_errno: %s)", strerror(rte_errno)); + " (error: %s)", strerror(err)); goto port_error; } INFO("port %u MAC address is %02x:%02x:%02x:%02x:%02x:%02x", @@ -614,15 +726,14 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) eth_dev = rte_eth_dev_allocate(name); } if (eth_dev == NULL) { + err = ENOMEM; ERROR("can not allocate rte ethdev"); - rte_errno = ENOMEM; goto port_error; } eth_dev->data->dev_private = priv; eth_dev->data->mac_addrs = priv->mac; eth_dev->device = &pci_dev->device; rte_eth_copy_pci_info(eth_dev, pci_dev); - eth_dev->device->driver = &mlx4_driver.driver; /* Initialize local interrupt handle for current port. */ priv->intr_handle = (struct rte_intr_handle){ .fd = -1, @@ -649,19 +760,40 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev) /* Update link status once if waiting for LSC. */ if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) mlx4_link_update(eth_dev, 0); + /* + * 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 = mlx4_mr_btree_init(&priv->mr.cache, + MLX4_MR_BTREE_CACHE_N * 2, + eth_dev->device->numa_node); + if (err) { + /* rte_errno is already set. */ + goto port_error; + } + /* Add device to memory callback list. */ + rte_rwlock_write_lock(&mlx4_mem_event_rwlock); + LIST_INSERT_HEAD(&mlx4_mem_event_cb_list, priv, mem_event_cb); + rte_rwlock_write_unlock(&mlx4_mem_event_rwlock); + rte_eth_dev_probing_finish(eth_dev); continue; port_error: rte_free(priv); + if (eth_dev != NULL) + eth_dev->data->dev_private = NULL; if (pd) claim_zero(mlx4_glue->dealloc_pd(pd)); if (ctx) claim_zero(mlx4_glue->close_device(ctx)); - if (eth_dev) + if (eth_dev != NULL) { + /* mac_addrs must not be freed because part of dev_private */ + eth_dev->data->mac_addrs = NULL; rte_eth_dev_release_port(eth_dev); + } break; } - if (i == device_attr.phys_port_cnt) - return 0; /* * XXX if something went wrong in the loop above, there is a resource * leak (ctx, pd, priv, dpdk ethdev) but we can do nothing about it as @@ -673,8 +805,9 @@ error: claim_zero(mlx4_glue->close_device(attr_ctx)); if (list) mlx4_glue->free_device_list(list); - assert(rte_errno >= 0); - return -rte_errno; + if (err) + rte_errno = err; + return -err; } static const struct rte_pci_id mlx4_pci_id_map[] = { @@ -838,9 +971,7 @@ glue_error: /** * Driver initialization routine. */ -RTE_INIT(rte_mlx4_pmd_init); -static void -rte_mlx4_pmd_init(void) +RTE_INIT(rte_mlx4_pmd_init) { /* * MLX4_DEVICE_FATAL_CLEANUP tells ibv_destroy functions we @@ -876,6 +1007,8 @@ rte_mlx4_pmd_init(void) } mlx4_glue->fork_init(); rte_pci_register(&mlx4_driver); + rte_mem_event_callback_register("MLX4_MEM_EVENT_CB", + mlx4_mr_mem_event_cb, NULL); } RTE_PMD_EXPORT_NAME(net_mlx4, __COUNTER__);