X-Git-Url: http://git.droids-corp.org/?a=blobdiff_plain;f=drivers%2Fnet%2Fmlx5%2Fmlx5_txq.c;h=35b3ade8688c60cfe960e261a822f8068ed81697;hb=0bf68c660e932e76087dc8c87f8b1dacba89c2be;hp=5e45748a170a2b33f0d3cd14533a2d2fcbc437c6;hpb=82e75f8323bf8544ee4756787b99bce2ae6a7cb0;p=dpdk.git diff --git a/drivers/net/mlx5/mlx5_txq.c b/drivers/net/mlx5/mlx5_txq.c index 5e45748a17..35b3ade868 100644 --- a/drivers/net/mlx5/mlx5_txq.c +++ b/drivers/net/mlx5/mlx5_txq.c @@ -4,7 +4,6 @@ */ #include -#include #include #include #include @@ -28,12 +27,16 @@ #include #include -#include "mlx5_utils.h" +#include +#include +#include +#include + #include "mlx5_defs.h" +#include "mlx5_utils.h" #include "mlx5.h" #include "mlx5_rxtx.h" #include "mlx5_autoconf.h" -#include "mlx5_glue.h" /** * Allocate TX queue elements. @@ -62,7 +65,7 @@ txq_alloc_elts(struct mlx5_txq_ctrl *txq_ctrl) * @param txq_ctrl * Pointer to TX queue structure. */ -static void +void txq_free_elts(struct mlx5_txq_ctrl *txq_ctrl) { const uint16_t elts_n = 1 << txq_ctrl->txq.elts_n; @@ -80,9 +83,9 @@ txq_free_elts(struct mlx5_txq_ctrl *txq_ctrl) while (elts_tail != elts_head) { struct rte_mbuf *elt = (*elts)[elts_tail & elts_m]; - assert(elt != NULL); + MLX5_ASSERT(elt != NULL); rte_pktmbuf_free_seg(elt); -#ifndef NDEBUG +#ifdef RTE_LIBRTE_MLX5_DEBUG /* Poisoning. */ memset(&(*elts)[elts_tail & elts_m], 0x77, @@ -127,7 +130,8 @@ mlx5_get_tx_port_offloads(struct rte_eth_dev *dev) offloads |= DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM; if (config->tso) offloads |= (DEV_TX_OFFLOAD_VXLAN_TNL_TSO | - DEV_TX_OFFLOAD_GRE_TNL_TSO); + DEV_TX_OFFLOAD_GRE_TNL_TSO | + DEV_TX_OFFLOAD_GENEVE_TNL_TSO); } return offloads; } @@ -146,27 +150,27 @@ mlx5_get_tx_port_offloads(struct rte_eth_dev *dev) * 0 on success, a negative errno value otherwise and rte_errno is set. */ static int -mlx5_tx_queue_pre_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc) +mlx5_tx_queue_pre_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t *desc) { struct mlx5_priv *priv = dev->data->dev_private; - if (desc <= MLX5_TX_COMP_THRESH) { + if (*desc <= MLX5_TX_COMP_THRESH) { DRV_LOG(WARNING, "port %u number of descriptors requested for Tx queue" " %u must be higher than MLX5_TX_COMP_THRESH, using %u" - " instead of %u", - dev->data->port_id, idx, MLX5_TX_COMP_THRESH + 1, desc); - desc = MLX5_TX_COMP_THRESH + 1; + " instead of %u", dev->data->port_id, idx, + MLX5_TX_COMP_THRESH + 1, *desc); + *desc = MLX5_TX_COMP_THRESH + 1; } - if (!rte_is_power_of_2(desc)) { - desc = 1 << log2above(desc); + if (!rte_is_power_of_2(*desc)) { + *desc = 1 << log2above(*desc); DRV_LOG(WARNING, "port %u increased number of descriptors in Tx queue" " %u to the next power of two (%d)", - dev->data->port_id, idx, desc); + dev->data->port_id, idx, *desc); } DRV_LOG(DEBUG, "port %u configuring queue %u for %u descriptors", - dev->data->port_id, idx, desc); + dev->data->port_id, idx, *desc); if (idx >= priv->txqs_n) { DRV_LOG(ERR, "port %u Tx queue index out of range (%u >= %u)", dev->data->port_id, idx, priv->txqs_n); @@ -209,7 +213,7 @@ mlx5_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc, container_of(txq, struct mlx5_txq_ctrl, txq); int res; - res = mlx5_tx_queue_pre_setup(dev, idx, desc); + res = mlx5_tx_queue_pre_setup(dev, idx, &desc); if (res) return res; txq_ctrl = mlx5_txq_new(dev, idx, desc, socket, conf); @@ -250,7 +254,7 @@ mlx5_tx_hairpin_queue_setup(struct rte_eth_dev *dev, uint16_t idx, container_of(txq, struct mlx5_txq_ctrl, txq); int res; - res = mlx5_tx_queue_pre_setup(dev, idx, desc); + res = mlx5_tx_queue_pre_setup(dev, idx, &desc); if (res) return res; if (hairpin_conf->peer_count != 1 || @@ -271,7 +275,6 @@ mlx5_tx_hairpin_queue_setup(struct rte_eth_dev *dev, uint16_t idx, DRV_LOG(DEBUG, "port %u adding Tx queue %u to list", dev->data->port_id, idx); (*priv->txqs)[idx] = &txq_ctrl->txq; - txq_ctrl->type = MLX5_TXQ_TYPE_HAIRPIN; return 0; } @@ -295,9 +298,9 @@ mlx5_tx_queue_release(void *dpdk_txq) priv = txq_ctrl->priv; for (i = 0; (i != priv->txqs_n); ++i) if ((*priv->txqs)[i] == txq) { - mlx5_txq_release(ETH_DEV(priv), i); DRV_LOG(DEBUG, "port %u removing Tx queue %u from list", PORT_ID(priv), txq->idx); + mlx5_txq_release(ETH_DEV(priv), i); break; } } @@ -314,7 +317,7 @@ static void txq_uar_ncattr_init(struct mlx5_txq_ctrl *txq_ctrl, size_t page_size) { struct mlx5_priv *priv = txq_ctrl->priv; - unsigned int cmd; + off_t cmd; txq_ctrl->txq.db_heu = priv->config.dbnc == MLX5_TXDB_HEURISTIC; txq_ctrl->txq.db_nc = 0; @@ -344,8 +347,8 @@ txq_uar_init(struct mlx5_txq_ctrl *txq_ctrl) if (txq_ctrl->type != MLX5_TXQ_TYPE_STANDARD) return; - assert(rte_eal_process_type() == RTE_PROC_PRIMARY); - assert(ppriv); + MLX5_ASSERT(rte_eal_process_type() == RTE_PROC_PRIMARY); + MLX5_ASSERT(ppriv); ppriv->uar_table[txq_ctrl->txq.idx] = txq_ctrl->bf_reg; txq_uar_ncattr_init(txq_ctrl, page_size); #ifndef RTE_ARCH_64 @@ -383,7 +386,7 @@ txq_uar_init_secondary(struct mlx5_txq_ctrl *txq_ctrl, int fd) if (txq_ctrl->type != MLX5_TXQ_TYPE_STANDARD) return 0; - assert(ppriv); + MLX5_ASSERT(ppriv); /* * As rdma-core, UARs are mapped in size of OS page * size. Ref to libmlx5 function: mlx5_init_context() @@ -424,6 +427,30 @@ txq_uar_uninit_secondary(struct mlx5_txq_ctrl *txq_ctrl) munmap(RTE_PTR_ALIGN_FLOOR(addr, page_size), page_size); } +/** + * Deinitialize Tx UAR registers for secondary process. + * + * @param dev + * Pointer to Ethernet device. + */ +void +mlx5_tx_uar_uninit_secondary(struct rte_eth_dev *dev) +{ + struct mlx5_priv *priv = dev->data->dev_private; + struct mlx5_txq_data *txq; + struct mlx5_txq_ctrl *txq_ctrl; + unsigned int i; + + MLX5_ASSERT(rte_eal_process_type() == RTE_PROC_SECONDARY); + for (i = 0; i != priv->txqs_n; ++i) { + if (!(*priv->txqs)[i]) + continue; + txq = (*priv->txqs)[i]; + txq_ctrl = container_of(txq, struct mlx5_txq_ctrl, txq); + txq_uar_uninit_secondary(txq_ctrl); + } +} + /** * Initialize Tx UAR registers for secondary process. * @@ -444,7 +471,7 @@ mlx5_tx_uar_init_secondary(struct rte_eth_dev *dev, int fd) unsigned int i; int ret; - assert(rte_eal_process_type() == RTE_PROC_SECONDARY); + MLX5_ASSERT(rte_eal_process_type() == RTE_PROC_SECONDARY); for (i = 0; i != priv->txqs_n; ++i) { if (!(*priv->txqs)[i]) continue; @@ -452,7 +479,7 @@ mlx5_tx_uar_init_secondary(struct rte_eth_dev *dev, int fd) txq_ctrl = container_of(txq, struct mlx5_txq_ctrl, txq); if (txq_ctrl->type != MLX5_TXQ_TYPE_STANDARD) continue; - assert(txq->idx == (uint16_t)i); + MLX5_ASSERT(txq->idx == (uint16_t)i); ret = txq_uar_init_secondary(txq_ctrl, fd); if (ret) goto error; @@ -490,10 +517,10 @@ mlx5_txq_obj_hairpin_new(struct rte_eth_dev *dev, uint16_t idx) container_of(txq_data, struct mlx5_txq_ctrl, txq); struct mlx5_devx_create_sq_attr attr = { 0 }; struct mlx5_txq_obj *tmpl = NULL; - int ret = 0; + uint32_t max_wq_data; - assert(txq_data); - assert(!txq_ctrl->obj); + MLX5_ASSERT(txq_data); + MLX5_ASSERT(!txq_ctrl->obj); tmpl = rte_calloc_socket(__func__, 1, sizeof(*tmpl), 0, txq_ctrl->socket); if (!tmpl) { @@ -501,39 +528,48 @@ mlx5_txq_obj_hairpin_new(struct rte_eth_dev *dev, uint16_t idx) "port %u Tx queue %u cannot allocate memory resources", dev->data->port_id, txq_data->idx); rte_errno = ENOMEM; - goto error; + return NULL; } tmpl->type = MLX5_TXQ_OBJ_TYPE_DEVX_HAIRPIN; tmpl->txq_ctrl = txq_ctrl; attr.hairpin = 1; attr.tis_lst_sz = 1; - /* Workaround for hairpin startup */ - attr.wq_attr.log_hairpin_num_packets = log2above(32); - /* Workaround for packets larger than 1KB */ - attr.wq_attr.log_hairpin_data_sz = - priv->config.hca_attr.log_max_hairpin_wq_data_sz; + max_wq_data = priv->config.hca_attr.log_max_hairpin_wq_data_sz; + /* Jumbo frames > 9KB should be supported, and more packets. */ + if (priv->config.log_hp_size != (uint32_t)MLX5_ARG_UNSET) { + if (priv->config.log_hp_size > max_wq_data) { + DRV_LOG(ERR, "total data size %u power of 2 is " + "too large for hairpin", + priv->config.log_hp_size); + rte_free(tmpl); + rte_errno = ERANGE; + return NULL; + } + attr.wq_attr.log_hairpin_data_sz = priv->config.log_hp_size; + } else { + attr.wq_attr.log_hairpin_data_sz = + (max_wq_data < MLX5_HAIRPIN_JUMBO_LOG_SIZE) ? + max_wq_data : MLX5_HAIRPIN_JUMBO_LOG_SIZE; + } + /* Set the packets number to the maximum value for performance. */ + attr.wq_attr.log_hairpin_num_packets = + attr.wq_attr.log_hairpin_data_sz - + MLX5_HAIRPIN_QUEUE_STRIDE; attr.tis_num = priv->sh->tis->id; tmpl->sq = mlx5_devx_cmd_create_sq(priv->sh->ctx, &attr); if (!tmpl->sq) { DRV_LOG(ERR, "port %u tx hairpin queue %u can't create sq object", dev->data->port_id, idx); + rte_free(tmpl); rte_errno = errno; - goto error; + return NULL; } DRV_LOG(DEBUG, "port %u sxq %u updated with %p", dev->data->port_id, idx, (void *)&tmpl); rte_atomic32_inc(&tmpl->refcnt); LIST_INSERT_HEAD(&priv->txqsobj, tmpl, next); return tmpl; -error: - ret = rte_errno; /* Save rte_errno before cleanup. */ - if (tmpl->tis) - mlx5_devx_cmd_destroy(tmpl->tis); - if (tmpl->sq) - mlx5_devx_cmd_destroy(tmpl->sq); - rte_errno = ret; /* Restore rte_errno. */ - return NULL; } /** @@ -578,7 +614,7 @@ mlx5_txq_obj_new(struct rte_eth_dev *dev, uint16_t idx, if (priv->config.devx && !priv->sh->tdn) qp.comp_mask |= MLX5DV_QP_MASK_RAW_QP_HANDLES; #endif - assert(txq_data); + MLX5_ASSERT(txq_data); priv->verbs_alloc_ctx.type = MLX5_VERBS_ALLOC_TYPE_TX_QUEUE; priv->verbs_alloc_ctx.obj = txq_ctrl; if (mlx5_getenv_int("MLX5_ENABLE_CQE_COMPRESSION")) { @@ -609,9 +645,9 @@ mlx5_txq_obj_new(struct rte_eth_dev *dev, uint16_t idx, .cap = { /* Max number of outstanding WRs. */ .max_send_wr = - ((priv->sh->device_attr.orig_attr.max_qp_wr < + ((priv->sh->device_attr.max_qp_wr < desc) ? - priv->sh->device_attr.orig_attr.max_qp_wr : + priv->sh->device_attr.max_qp_wr : desc), /* * Max number of scatter/gather elements in a WR, @@ -648,7 +684,7 @@ mlx5_txq_obj_new(struct rte_eth_dev *dev, uint16_t idx, /* Move the QP to this state. */ .qp_state = IBV_QPS_INIT, /* IB device port number. */ - .port_num = (uint8_t)priv->ibv_port, + .port_num = (uint8_t)priv->dev_port, }; ret = mlx5_glue->modify_qp(tmpl.qp, &attr.mod, (IBV_QP_STATE | IBV_QP_PORT)); @@ -717,13 +753,22 @@ mlx5_txq_obj_new(struct rte_eth_dev *dev, uint16_t idx, txq_data->cq_db = cq_info.dbrec; txq_data->cqes = (volatile struct mlx5_cqe *)cq_info.buf; txq_data->cq_ci = 0; -#ifndef NDEBUG txq_data->cq_pi = 0; -#endif txq_data->wqe_ci = 0; txq_data->wqe_pi = 0; txq_data->wqe_comp = 0; txq_data->wqe_thres = txq_data->wqe_s / MLX5_TX_COMP_THRESH_INLINE_DIV; + txq_data->fcqs = rte_calloc_socket(__func__, + txq_data->cqe_s, + sizeof(*txq_data->fcqs), + RTE_CACHE_LINE_SIZE, + txq_ctrl->socket); + if (!txq_data->fcqs) { + DRV_LOG(ERR, "port %u Tx queue %u cannot allocate memory (FCQ)", + dev->data->port_id, idx); + rte_errno = ENOMEM; + goto error; + } #ifdef HAVE_IBV_FLOW_DV_SUPPORT /* * If using DevX need to query and store TIS transport domain value. @@ -772,6 +817,8 @@ error: claim_zero(mlx5_glue->destroy_cq(tmpl.cq)); if (tmpl.qp) claim_zero(mlx5_glue->destroy_qp(tmpl.qp)); + if (txq_data->fcqs) + rte_free(txq_data->fcqs); if (txq_obj) rte_free(txq_obj); priv->verbs_alloc_ctx.type = MLX5_VERBS_ALLOC_TYPE_NONE; @@ -818,7 +865,7 @@ mlx5_txq_obj_get(struct rte_eth_dev *dev, uint16_t idx) int mlx5_txq_obj_release(struct mlx5_txq_obj *txq_obj) { - assert(txq_obj); + MLX5_ASSERT(txq_obj); if (rte_atomic32_dec_and_test(&txq_obj->refcnt)) { if (txq_obj->type == MLX5_TXQ_OBJ_TYPE_DEVX_HAIRPIN) { if (txq_obj->tis) @@ -826,6 +873,8 @@ mlx5_txq_obj_release(struct mlx5_txq_obj *txq_obj) } else { claim_zero(mlx5_glue->destroy_qp(txq_obj->qp)); claim_zero(mlx5_glue->destroy_cq(txq_obj->cq)); + if (txq_obj->txq_ctrl->txq.fcqs) + rte_free(txq_obj->txq_ctrl->txq.fcqs); } LIST_REMOVE(txq_obj, next); rte_free(txq_obj); @@ -899,7 +948,7 @@ txq_calc_inline_max(struct mlx5_txq_ctrl *txq_ctrl) struct mlx5_priv *priv = txq_ctrl->priv; unsigned int wqe_size; - wqe_size = priv->sh->device_attr.orig_attr.max_qp_wr / desc; + wqe_size = priv->sh->device_attr.max_qp_wr / desc; if (!wqe_size) return 0; /* @@ -963,7 +1012,7 @@ txq_set_params(struct mlx5_txq_ctrl *txq_ctrl) * If there is requested minimal amount of data to inline * we MUST enable inlining. This is a case for ConnectX-4 * which usually requires L2 inlined for correct operating - * and ConnectX-4LX which requires L2-L4 inlined to + * and ConnectX-4 Lx which requires L2-L4 inlined to * support E-Switch Flows. */ if (inlen_mode) { @@ -1034,12 +1083,12 @@ txq_set_params(struct mlx5_txq_ctrl *txq_ctrl) * beginning of inlining buffer in Ethernet * Segment. */ - assert(inlen_send >= MLX5_ESEG_MIN_INLINE_SIZE); - assert(inlen_send <= MLX5_WQE_SIZE_MAX + - MLX5_ESEG_MIN_INLINE_SIZE - - MLX5_WQE_CSEG_SIZE - - MLX5_WQE_ESEG_SIZE - - MLX5_WQE_DSEG_SIZE * 2); + MLX5_ASSERT(inlen_send >= MLX5_ESEG_MIN_INLINE_SIZE); + MLX5_ASSERT(inlen_send <= MLX5_WQE_SIZE_MAX + + MLX5_ESEG_MIN_INLINE_SIZE - + MLX5_WQE_CSEG_SIZE - + MLX5_WQE_ESEG_SIZE - + MLX5_WQE_DSEG_SIZE * 2); } else if (inlen_mode) { /* * If minimal inlining is requested we must @@ -1089,12 +1138,12 @@ txq_set_params(struct mlx5_txq_ctrl *txq_ctrl) PORT_ID(priv), inlen_empw, temp); inlen_empw = temp; } - assert(inlen_empw >= MLX5_ESEG_MIN_INLINE_SIZE); - assert(inlen_empw <= MLX5_WQE_SIZE_MAX + - MLX5_DSEG_MIN_INLINE_SIZE - - MLX5_WQE_CSEG_SIZE - - MLX5_WQE_ESEG_SIZE - - MLX5_WQE_DSEG_SIZE); + MLX5_ASSERT(inlen_empw >= MLX5_ESEG_MIN_INLINE_SIZE); + MLX5_ASSERT(inlen_empw <= MLX5_WQE_SIZE_MAX + + MLX5_DSEG_MIN_INLINE_SIZE - + MLX5_WQE_CSEG_SIZE - + MLX5_WQE_ESEG_SIZE - + MLX5_WQE_DSEG_SIZE); txq_ctrl->txq.inlen_empw = inlen_empw; } txq_ctrl->max_inline_data = RTE_MAX(inlen_send, inlen_empw); @@ -1154,7 +1203,7 @@ txq_adjust_params(struct mlx5_txq_ctrl *txq_ctrl) " Tx queue size (%d)", txq_ctrl->txq.inlen_mode, max_inline, priv->dev_data->port_id, - priv->sh->device_attr.orig_attr.max_qp_wr); + priv->sh->device_attr.max_qp_wr); goto error; } if (txq_ctrl->txq.inlen_send > max_inline && @@ -1166,7 +1215,7 @@ txq_adjust_params(struct mlx5_txq_ctrl *txq_ctrl) " Tx queue size (%d)", txq_ctrl->txq.inlen_send, max_inline, priv->dev_data->port_id, - priv->sh->device_attr.orig_attr.max_qp_wr); + priv->sh->device_attr.max_qp_wr); goto error; } if (txq_ctrl->txq.inlen_empw > max_inline && @@ -1178,7 +1227,7 @@ txq_adjust_params(struct mlx5_txq_ctrl *txq_ctrl) " Tx queue size (%d)", txq_ctrl->txq.inlen_empw, max_inline, priv->dev_data->port_id, - priv->sh->device_attr.orig_attr.max_qp_wr); + priv->sh->device_attr.max_qp_wr); goto error; } if (txq_ctrl->txq.tso_en && max_inline < MLX5_MAX_TSO_HEADER) { @@ -1188,7 +1237,7 @@ txq_adjust_params(struct mlx5_txq_ctrl *txq_ctrl) " Tx queue size (%d)", MLX5_MAX_TSO_HEADER, max_inline, priv->dev_data->port_id, - priv->sh->device_attr.orig_attr.max_qp_wr); + priv->sh->device_attr.max_qp_wr); goto error; } if (txq_ctrl->txq.inlen_send > max_inline) { @@ -1209,11 +1258,11 @@ txq_adjust_params(struct mlx5_txq_ctrl *txq_ctrl) } txq_ctrl->max_inline_data = RTE_MAX(txq_ctrl->txq.inlen_send, txq_ctrl->txq.inlen_empw); - assert(txq_ctrl->max_inline_data <= max_inline); - assert(txq_ctrl->txq.inlen_mode <= max_inline); - assert(txq_ctrl->txq.inlen_mode <= txq_ctrl->txq.inlen_send); - assert(txq_ctrl->txq.inlen_mode <= txq_ctrl->txq.inlen_empw || - !txq_ctrl->txq.inlen_empw); + MLX5_ASSERT(txq_ctrl->max_inline_data <= max_inline); + MLX5_ASSERT(txq_ctrl->txq.inlen_mode <= max_inline); + MLX5_ASSERT(txq_ctrl->txq.inlen_mode <= txq_ctrl->txq.inlen_send); + MLX5_ASSERT(txq_ctrl->txq.inlen_mode <= txq_ctrl->txq.inlen_empw || + !txq_ctrl->txq.inlen_empw); return 0; error: rte_errno = ENOMEM; @@ -1258,8 +1307,8 @@ mlx5_txq_new(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc, goto error; } /* Save pointer of global generation number to check memory event. */ - tmpl->txq.mr_ctrl.dev_gen_ptr = &priv->sh->mr.dev_gen; - assert(desc > MLX5_TX_COMP_THRESH); + tmpl->txq.mr_ctrl.dev_gen_ptr = &priv->sh->share_cache.dev_gen; + MLX5_ASSERT(desc > MLX5_TX_COMP_THRESH); tmpl->txq.offloads = conf->offloads | dev->data->dev_conf.txmode.offloads; tmpl->priv = priv; @@ -1273,12 +1322,12 @@ mlx5_txq_new(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc, if (txq_adjust_params(tmpl)) goto error; if (txq_calc_wqebb_cnt(tmpl) > - priv->sh->device_attr.orig_attr.max_qp_wr) { + priv->sh->device_attr.max_qp_wr) { DRV_LOG(ERR, "port %u Tx WQEBB count (%d) exceeds the limit (%d)," " try smaller queue size", dev->data->port_id, txq_calc_wqebb_cnt(tmpl), - priv->sh->device_attr.orig_attr.max_qp_wr); + priv->sh->device_attr.max_qp_wr); rte_errno = ENOMEM; goto error; }