X-Git-Url: http://git.droids-corp.org/?a=blobdiff_plain;f=drivers%2Fnet%2Fmlx5%2Fmlx5_rxq.c;h=cbb017bb966cf965f722fd665edb1f57047ddcf0;hb=ce2eabdd43ecfdd3ada379ed884063e82fa266e9;hp=3d7ae7e19470a4c1e64094aa5a911ab3ad26db27;hpb=634efbc2c8c05213696c36e103becc48cdbe534c;p=dpdk.git diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c index 3d7ae7e194..cbb017bb96 100644 --- a/drivers/net/mlx5/mlx5_rxq.c +++ b/drivers/net/mlx5/mlx5_rxq.c @@ -62,6 +62,7 @@ #include "mlx5.h" #include "mlx5_rxtx.h" #include "mlx5_utils.h" +#include "mlx5_autoconf.h" #include "mlx5_defs.h" /* Initialization data for hash RX queues. */ @@ -210,27 +211,27 @@ const size_t rss_hash_default_key_len = sizeof(rss_hash_default_key); * information from hash_rxq_init[]. Nothing is written to flow_attr when * flow_attr_size is not large enough, but the required size is still returned. * - * @param[in] hash_rxq - * Pointer to hash RX queue. + * @param priv + * Pointer to private structure. * @param[out] flow_attr * Pointer to flow attribute structure to fill. Note that the allocated * area must be larger and large enough to hold all flow specifications. * @param flow_attr_size * Entire size of flow_attr and trailing room for flow specifications. + * @param type + * Hash RX queue type to use for flow steering rule. * * @return * Total size of the flow attribute buffer. No errors are defined. */ size_t -hash_rxq_flow_attr(const struct hash_rxq *hash_rxq, - struct ibv_exp_flow_attr *flow_attr, - size_t flow_attr_size) +priv_flow_attr(struct priv *priv, struct ibv_exp_flow_attr *flow_attr, + size_t flow_attr_size, enum hash_rxq_type type) { size_t offset = sizeof(*flow_attr); - enum hash_rxq_type type = hash_rxq->type; const struct hash_rxq_init *init = &hash_rxq_init[type]; - assert(hash_rxq->priv != NULL); + assert(priv != NULL); assert((size_t)type < RTE_DIM(hash_rxq_init)); do { offset += init->flow_spec.hdr.size; @@ -242,9 +243,14 @@ hash_rxq_flow_attr(const struct hash_rxq *hash_rxq, init = &hash_rxq_init[type]; *flow_attr = (struct ibv_exp_flow_attr){ .type = IBV_EXP_FLOW_ATTR_NORMAL, +#ifdef MLX5_FDIR_SUPPORT + /* Priorities < 3 are reserved for flow director. */ + .priority = init->flow_priority + 3, +#else /* MLX5_FDIR_SUPPORT */ .priority = init->flow_priority, +#endif /* MLX5_FDIR_SUPPORT */ .num_of_specs = 0, - .port = hash_rxq->priv->port, + .port = priv->port, .flags = 0, }; do { @@ -259,23 +265,29 @@ hash_rxq_flow_attr(const struct hash_rxq *hash_rxq, } /** - * Return the type corresponding to the n'th bit set. + * Convert hash type position in indirection table initializer to + * hash RX queue type. * * @param table - * The indirection table. - * @param n - * The n'th bit set. + * Indirection table initializer. + * @param pos + * Hash type position. * * @return - * The corresponding hash_rxq_type. + * Hash RX queue type. */ static enum hash_rxq_type -hash_rxq_type_from_n(const struct ind_table_init *table, unsigned int n) +hash_rxq_type_from_pos(const struct ind_table_init *table, unsigned int pos) { - assert(n < table->hash_types_n); - while (((table->hash_types >> n) & 0x1) == 0) - ++n; - return n; + enum hash_rxq_type type = 0; + + assert(pos < table->hash_types_n); + do { + if ((table->hash_types & (1 << type)) && (pos-- == 0)) + break; + ++type; + } while (1); + return type; } /** @@ -300,7 +312,7 @@ priv_make_ind_table_init(struct priv *priv, /* Mandatory to receive frames not handled by normal hash RX queues. */ unsigned int hash_types_sup = 1 << HASH_RXQ_ETH; - rss_hf = priv->dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf; + rss_hf = priv->rss_hf; /* Process other protocols only if more than one queue. */ if (priv->rxqs_n > 1) for (i = 0; (i != hash_rxq_init_n); ++i) @@ -429,7 +441,7 @@ priv_create_hash_rxqs(struct priv *priv) ++i) { struct hash_rxq *hash_rxq = &(*hash_rxqs)[i]; enum hash_rxq_type type = - hash_rxq_type_from_n(&ind_table_init[j], k); + hash_rxq_type_from_pos(&ind_table_init[j], k); struct rte_eth_rss_conf *priv_rss_conf = (*priv->rss_conf)[type]; struct ibv_exp_rx_hash_conf hash_conf = { @@ -453,8 +465,8 @@ priv_create_hash_rxqs(struct priv *priv) .port_num = priv->port, }; - DEBUG("using indirection table %u for hash RX queue %u", - j, i); + DEBUG("using indirection table %u for hash RX queue %u type %d", + j, i, type); *hash_rxq = (struct hash_rxq){ .priv = priv, .qp = ibv_exp_create_qp(priv->ctx, &qp_init_attr), @@ -528,8 +540,11 @@ priv_destroy_hash_rxqs(struct priv *priv) assert(hash_rxq->priv == priv); assert(hash_rxq->qp != NULL); /* Also check that there are no remaining flows. */ - assert(hash_rxq->allmulti_flow == NULL); - assert(hash_rxq->promisc_flow == NULL); + for (j = 0; (j != RTE_DIM(hash_rxq->special_flow)); ++j) + for (k = 0; + (k != RTE_DIM(hash_rxq->special_flow[j])); + ++k) + assert(hash_rxq->special_flow[j][k] == NULL); for (j = 0; (j != RTE_DIM(hash_rxq->mac_flow)); ++j) for (k = 0; (k != RTE_DIM(hash_rxq->mac_flow[j])); ++k) assert(hash_rxq->mac_flow[j][k] == NULL); @@ -567,18 +582,57 @@ priv_allow_flow_type(struct priv *priv, enum hash_rxq_flow_type type) /* Only FLOW_TYPE_PROMISC is allowed when promiscuous mode * has been requested. */ if (priv->promisc_req) - return (type == HASH_RXQ_FLOW_TYPE_PROMISC); + return type == HASH_RXQ_FLOW_TYPE_PROMISC; switch (type) { case HASH_RXQ_FLOW_TYPE_PROMISC: return !!priv->promisc_req; case HASH_RXQ_FLOW_TYPE_ALLMULTI: return !!priv->allmulti_req; + case HASH_RXQ_FLOW_TYPE_BROADCAST: +#ifdef HAVE_FLOW_SPEC_IPV6 + case HASH_RXQ_FLOW_TYPE_IPV6MULTI: +#endif /* HAVE_FLOW_SPEC_IPV6 */ + /* If allmulti is enabled, broadcast and ipv6multi + * are unnecessary. */ + return !priv->allmulti_req; case HASH_RXQ_FLOW_TYPE_MAC: return 1; + default: + /* Unsupported flow type is not allowed. */ + return 0; } return 0; } +/** + * Automatically enable/disable flows according to configuration. + * + * @param priv + * Private structure. + * + * @return + * 0 on success, errno value on failure. + */ +int +priv_rehash_flows(struct priv *priv) +{ + unsigned int i; + + for (i = 0; (i != RTE_DIM((*priv->hash_rxqs)[0].special_flow)); ++i) + if (!priv_allow_flow_type(priv, i)) { + priv_special_flow_disable(priv, i); + } else { + int ret = priv_special_flow_enable(priv, i); + + if (ret) + return ret; + } + if (priv_allow_flow_type(priv, HASH_RXQ_FLOW_TYPE_MAC)) + return priv_mac_addrs_enable(priv); + priv_mac_addrs_disable(priv); + return 0; +} + /** * Allocate RX queue elements with scattered packets support. * @@ -850,6 +904,8 @@ rxq_cleanup(struct rxq *rxq) rxq_free_elts_sp(rxq); else rxq_free_elts(rxq); + rxq->poll = NULL; + rxq->recv = NULL; if (rxq->if_wq != NULL) { assert(rxq->priv != NULL); assert(rxq->priv->ctx != NULL); @@ -1052,6 +1108,10 @@ rxq_rehash(struct rte_eth_dev *dev, struct rxq *rxq) err = EIO; goto error; } + if (tmpl.sp) + tmpl.recv = tmpl.if_wq->recv_sg_list; + else + tmpl.recv = tmpl.if_wq->recv_burst; error: *rxq = tmpl; assert(err >= 0); @@ -1133,11 +1193,7 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc, DEBUG("%p: %s scattered packets support (%u WRs)", (void *)dev, (tmpl.sp ? "enabling" : "disabling"), desc); /* Use the entire RX mempool as the memory region. */ - tmpl.mr = ibv_reg_mr(priv->pd, - (void *)mp->elt_va_start, - (mp->elt_va_end - mp->elt_va_start), - (IBV_ACCESS_LOCAL_WRITE | - IBV_ACCESS_REMOTE_WRITE)); + tmpl.mr = mlx5_mp2mr(priv->pd, mp); if (tmpl.mr == NULL) { ret = EINVAL; ERROR("%p: MR creation failure: %s", @@ -1173,6 +1229,8 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc, priv->device_attr.max_qp_wr); DEBUG("priv->device_attr.max_sge is %d", priv->device_attr.max_sge); + /* Configure VLAN stripping. */ + tmpl.vlan_strip = dev->data->dev_conf.rxmode.hw_vlan_strip; attr.wq = (struct ibv_exp_wq_init_attr){ .wq_context = NULL, /* Could be useful in the future. */ .wq_type = IBV_EXP_WQT_RQ, @@ -1187,9 +1245,58 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc, MLX5_PMD_SGE_WR_N), .pd = priv->pd, .cq = tmpl.cq, - .comp_mask = IBV_EXP_CREATE_WQ_RES_DOMAIN, + .comp_mask = + IBV_EXP_CREATE_WQ_RES_DOMAIN | +#ifdef HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS + IBV_EXP_CREATE_WQ_VLAN_OFFLOADS | +#endif /* HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS */ + 0, .res_domain = tmpl.rd, +#ifdef HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS + .vlan_offloads = (tmpl.vlan_strip ? + IBV_EXP_RECEIVE_WQ_CVLAN_STRIP : + 0), +#endif /* HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS */ }; + +#ifdef HAVE_VERBS_FCS + /* By default, FCS (CRC) is stripped by hardware. */ + if (dev->data->dev_conf.rxmode.hw_strip_crc) { + tmpl.crc_present = 0; + } else if (priv->hw_fcs_strip) { + /* Ask HW/Verbs to leave CRC in place when supported. */ + attr.wq.flags |= IBV_EXP_CREATE_WQ_FLAG_SCATTER_FCS; + attr.wq.comp_mask |= IBV_EXP_CREATE_WQ_FLAGS; + tmpl.crc_present = 1; + } else { + WARN("%p: CRC stripping has been disabled but will still" + " be performed by hardware, make sure MLNX_OFED and" + " firmware are up to date", + (void *)dev); + tmpl.crc_present = 0; + } + DEBUG("%p: CRC stripping is %s, %u bytes will be subtracted from" + " incoming frames to hide it", + (void *)dev, + tmpl.crc_present ? "disabled" : "enabled", + tmpl.crc_present << 2); +#endif /* HAVE_VERBS_FCS */ + +#ifdef HAVE_VERBS_RX_END_PADDING + if (!mlx5_getenv_int("MLX5_PMD_ENABLE_PADDING")) + ; /* Nothing else to do. */ + else if (priv->hw_padding) { + INFO("%p: enabling packet padding on queue %p", + (void *)dev, (void *)rxq); + attr.wq.flags |= IBV_EXP_CREATE_WQ_FLAG_RX_END_PADDING; + attr.wq.comp_mask |= IBV_EXP_CREATE_WQ_FLAGS; + } else + WARN("%p: packet padding has been requested but is not" + " supported, make sure MLNX_OFED and firmware are" + " up to date", + (void *)dev); +#endif /* HAVE_VERBS_RX_END_PADDING */ + tmpl.wq = ibv_exp_create_wq(priv->ctx, &attr.wq); if (tmpl.wq == NULL) { ret = (errno ? errno : EINVAL); @@ -1211,6 +1318,9 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc, DEBUG("%p: RTE port ID: %u", (void *)rxq, tmpl.port_id); attr.params = (struct ibv_exp_query_intf_params){ .intf_scope = IBV_EXP_INTF_GLOBAL, +#ifdef HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS + .intf_version = 1, +#endif /* HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS */ .intf = IBV_EXP_INTF_CQ, .obj = tmpl.cq, }; @@ -1279,6 +1389,16 @@ rxq_setup(struct rte_eth_dev *dev, struct rxq *rxq, uint16_t desc, *rxq = tmpl; DEBUG("%p: rxq updated with %p", (void *)rxq, (void *)&tmpl); assert(ret == 0); + /* Assign function in queue. */ +#ifdef HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS + rxq->poll = rxq->if_cq->poll_length_flags_cvlan; +#else /* HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS */ + rxq->poll = rxq->if_cq->poll_length_flags; +#endif /* HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS */ + if (rxq->sp) + rxq->recv = rxq->if_wq->recv_sg_list; + else + rxq->recv = rxq->if_wq->recv_burst; return 0; error: rxq_cleanup(&tmpl); @@ -1314,6 +1434,9 @@ mlx5_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc, struct rxq *rxq = (*priv->rxqs)[idx]; int ret; + if (mlx5_is_secondary()) + return -E_RTE_SECONDARY; + priv_lock(priv); DEBUG("%p: configuring queue %u for %u descriptors", (void *)dev, idx, desc); @@ -1372,6 +1495,9 @@ mlx5_rx_queue_release(void *dpdk_rxq) struct priv *priv; unsigned int i; + if (mlx5_is_secondary()) + return; + if (rxq == NULL) return; priv = rxq->priv; @@ -1387,3 +1513,43 @@ mlx5_rx_queue_release(void *dpdk_rxq) rte_free(rxq); priv_unlock(priv); } + +/** + * DPDK callback for RX in secondary processes. + * + * This function configures all queues from primary process information + * if necessary before reverting to the normal RX burst callback. + * + * @param dpdk_rxq + * Generic pointer to RX queue structure. + * @param[out] pkts + * Array to store received packets. + * @param pkts_n + * Maximum number of packets in array. + * + * @return + * Number of packets successfully received (<= pkts_n). + */ +uint16_t +mlx5_rx_burst_secondary_setup(void *dpdk_rxq, struct rte_mbuf **pkts, + uint16_t pkts_n) +{ + struct rxq *rxq = dpdk_rxq; + struct priv *priv = mlx5_secondary_data_setup(rxq->priv); + struct priv *primary_priv; + unsigned int index; + + if (priv == NULL) + return 0; + primary_priv = + mlx5_secondary_data[priv->dev->data->port_id].primary_priv; + /* Look for queue index in both private structures. */ + for (index = 0; index != priv->rxqs_n; ++index) + if (((*primary_priv->rxqs)[index] == rxq) || + ((*priv->rxqs)[index] == rxq)) + break; + if (index == priv->rxqs_n) + return 0; + rxq = (*priv->rxqs)[index]; + return priv->dev->rx_pkt_burst(rxq, pkts, pkts_n); +}