net/sfc: support group flows in tunnel offload
[dpdk.git] / drivers / net / bnxt / bnxt_ethdev.c
index e7fc254..a98f93a 100644 (file)
@@ -580,13 +580,14 @@ static int bnxt_register_fc_ctx_mem(struct bnxt *bp)
        return rc;
 }
 
-static int bnxt_alloc_ctx_mem_buf(char *type, size_t size,
+static int bnxt_alloc_ctx_mem_buf(struct bnxt *bp, char *type, size_t size,
                                  struct bnxt_ctx_mem_buf_info *ctx)
 {
        if (!ctx)
                return -EINVAL;
 
-       ctx->va = rte_zmalloc(type, size, 0);
+       ctx->va = rte_zmalloc_socket(type, size, 0,
+                                    bp->eth_dev->device->numa_node);
        if (ctx->va == NULL)
                return -ENOMEM;
        rte_mem_lock_page(ctx->va);
@@ -610,7 +611,7 @@ static int bnxt_init_fc_ctx_mem(struct bnxt *bp)
        sprintf(type, "bnxt_rx_fc_in_" PCI_PRI_FMT, pdev->addr.domain,
                pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
        /* 4 bytes for each counter-id */
-       rc = bnxt_alloc_ctx_mem_buf(type,
+       rc = bnxt_alloc_ctx_mem_buf(bp, type,
                                    max_fc * 4,
                                    &bp->flow_stat->rx_fc_in_tbl);
        if (rc)
@@ -619,7 +620,7 @@ static int bnxt_init_fc_ctx_mem(struct bnxt *bp)
        sprintf(type, "bnxt_rx_fc_out_" PCI_PRI_FMT, pdev->addr.domain,
                pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
        /* 16 bytes for each counter - 8 bytes pkt_count, 8 bytes byte_count */
-       rc = bnxt_alloc_ctx_mem_buf(type,
+       rc = bnxt_alloc_ctx_mem_buf(bp, type,
                                    max_fc * 16,
                                    &bp->flow_stat->rx_fc_out_tbl);
        if (rc)
@@ -628,7 +629,7 @@ static int bnxt_init_fc_ctx_mem(struct bnxt *bp)
        sprintf(type, "bnxt_tx_fc_in_" PCI_PRI_FMT, pdev->addr.domain,
                pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
        /* 4 bytes for each counter-id */
-       rc = bnxt_alloc_ctx_mem_buf(type,
+       rc = bnxt_alloc_ctx_mem_buf(bp, type,
                                    max_fc * 4,
                                    &bp->flow_stat->tx_fc_in_tbl);
        if (rc)
@@ -637,7 +638,7 @@ static int bnxt_init_fc_ctx_mem(struct bnxt *bp)
        sprintf(type, "bnxt_tx_fc_out_" PCI_PRI_FMT, pdev->addr.domain,
                pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
        /* 16 bytes for each counter - 8 bytes pkt_count, 8 bytes byte_count */
-       rc = bnxt_alloc_ctx_mem_buf(type,
+       rc = bnxt_alloc_ctx_mem_buf(bp, type,
                                    max_fc * 16,
                                    &bp->flow_stat->tx_fc_out_tbl);
        if (rc)
@@ -807,6 +808,26 @@ skip_cosq_cfg:
                }
        }
 
+       for (j = 0; j < bp->tx_nr_rings; j++) {
+               struct bnxt_tx_queue *txq = bp->tx_queues[j];
+
+               if (!txq->tx_deferred_start) {
+                       bp->eth_dev->data->tx_queue_state[j] =
+                               RTE_ETH_QUEUE_STATE_STARTED;
+                       txq->tx_started = true;
+               }
+       }
+
+       for (j = 0; j < bp->rx_nr_rings; j++) {
+               struct bnxt_rx_queue *rxq = bp->rx_queues[j];
+
+               if (!rxq->rx_deferred_start) {
+                       bp->eth_dev->data->rx_queue_state[j] =
+                               RTE_ETH_QUEUE_STATE_STARTED;
+                       rxq->rx_started = true;
+               }
+       }
+
        rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, &bp->vnic_info[0], 0, NULL);
        if (rc) {
                PMD_DRV_LOG(ERR,
@@ -980,12 +1001,18 @@ static int bnxt_dev_info_get_op(struct rte_eth_dev *eth_dev,
        dev_info->rx_offload_capa = BNXT_DEV_RX_OFFLOAD_SUPPORT;
        if (bp->flags & BNXT_FLAG_PTP_SUPPORTED)
                dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TIMESTAMP;
+       if (bp->vnic_cap_flags & BNXT_VNIC_CAP_VLAN_RX_STRIP)
+               dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_STRIP;
        dev_info->tx_queue_offload_capa = DEV_TX_OFFLOAD_MBUF_FAST_FREE;
        dev_info->tx_offload_capa = BNXT_DEV_TX_OFFLOAD_SUPPORT |
                                    dev_info->tx_queue_offload_capa;
+       if (bp->fw_cap & BNXT_FW_CAP_VLAN_TX_INSERT)
+               dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_VLAN_INSERT;
        dev_info->flow_type_rss_offloads = BNXT_ETH_RSS_SUPPORT;
 
        dev_info->speed_capa = bnxt_get_speed_capabilities(bp);
+       dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
+                            RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
 
        dev_info->default_rxconf = (struct rte_eth_rxconf) {
                .rx_thresh = {
@@ -1008,7 +1035,6 @@ static int bnxt_dev_info_get_op(struct rte_eth_dev *eth_dev,
        };
        eth_dev->data->dev_conf.intr_conf.lsc = 1;
 
-       eth_dev->data->dev_conf.intr_conf.rxq = 1;
        dev_info->rx_desc_lim.nb_min = BNXT_MIN_RING_DESC;
        dev_info->rx_desc_lim.nb_max = BNXT_MAX_RX_RING_DESC;
        dev_info->tx_desc_lim.nb_min = BNXT_MIN_RING_DESC;
@@ -1841,11 +1867,6 @@ out:
        if (new.link_status != eth_dev->data->dev_link.link_status ||
            new.link_speed != eth_dev->data->dev_link.link_speed) {
                rte_eth_linkstatus_set(eth_dev, &new);
-
-               rte_eth_dev_callback_process(eth_dev,
-                                            RTE_ETH_EVENT_INTR_LSC,
-                                            NULL);
-
                bnxt_print_link_info(eth_dev);
        }
 
@@ -2152,7 +2173,8 @@ static int bnxt_rss_hash_update_op(struct rte_eth_dev *eth_dev,
 
        if (rss_conf->rss_key_len != HW_HASH_KEY_SIZE) {
                PMD_DRV_LOG(ERR,
-                           "Invalid hashkey length, should be 16 bytes\n");
+                           "Invalid hashkey length, should be %d bytes\n",
+                           HW_HASH_KEY_SIZE);
                return -EINVAL;
        }
        memcpy(vnic->rss_hash_key, rss_conf->rss_key, rss_conf->rss_key_len);
@@ -2351,7 +2373,6 @@ bnxt_udp_tunnel_port_add_op(struct rte_eth_dev *eth_dev,
                }
                tunnel_type =
                        HWRM_TUNNEL_DST_PORT_ALLOC_INPUT_TUNNEL_TYPE_VXLAN;
-               bp->vxlan_port_cnt++;
                break;
        case RTE_TUNNEL_TYPE_GENEVE:
                if (bp->geneve_port_cnt) {
@@ -2366,7 +2387,6 @@ bnxt_udp_tunnel_port_add_op(struct rte_eth_dev *eth_dev,
                }
                tunnel_type =
                        HWRM_TUNNEL_DST_PORT_ALLOC_INPUT_TUNNEL_TYPE_GENEVE;
-               bp->geneve_port_cnt++;
                break;
        default:
                PMD_DRV_LOG(ERR, "Tunnel type is not supported\n");
@@ -2374,6 +2394,18 @@ bnxt_udp_tunnel_port_add_op(struct rte_eth_dev *eth_dev,
        }
        rc = bnxt_hwrm_tunnel_dst_port_alloc(bp, udp_tunnel->udp_port,
                                             tunnel_type);
+
+       if (rc != 0)
+               return rc;
+
+       if (tunnel_type ==
+           HWRM_TUNNEL_DST_PORT_ALLOC_INPUT_TUNNEL_TYPE_VXLAN)
+               bp->vxlan_port_cnt++;
+
+       if (tunnel_type ==
+           HWRM_TUNNEL_DST_PORT_ALLOC_INPUT_TUNNEL_TYPE_GENEVE)
+               bp->geneve_port_cnt++;
+
        return rc;
 }
 
@@ -3064,6 +3096,9 @@ int bnxt_mtu_set_op(struct rte_eth_dev *eth_dev, uint16_t new_mtu)
        if (!rc)
                eth_dev->data->dev_conf.rxmode.max_rx_pkt_len = new_pkt_size;
 
+       if (bnxt_hwrm_config_host_mtu(bp))
+               PMD_DRV_LOG(WARNING, "Failed to configure host MTU\n");
+
        PMD_DRV_LOG(INFO, "New MTU is %d\n", new_mtu);
 
        return rc;
@@ -3123,7 +3158,7 @@ bnxt_rx_queue_count_op(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 {
        struct bnxt *bp = (struct bnxt *)dev->data->dev_private;
        struct bnxt_cp_ring_info *cpr;
-       uint32_t desc = 0, raw_cons;
+       uint32_t desc = 0, raw_cons, cp_ring_size;
        struct bnxt_rx_queue *rxq;
        struct rx_pkt_cmpl *rxcmp;
        int rc;
@@ -3135,6 +3170,7 @@ bnxt_rx_queue_count_op(struct rte_eth_dev *dev, uint16_t rx_queue_id)
        rxq = dev->data->rx_queues[rx_queue_id];
        cpr = rxq->cp_ring;
        raw_cons = cpr->cp_raw_cons;
+       cp_ring_size = cpr->cp_ring_struct->ring_size;
 
        while (1) {
                uint32_t agg_cnt, cons, cmpl_type;
@@ -3142,7 +3178,7 @@ bnxt_rx_queue_count_op(struct rte_eth_dev *dev, uint16_t rx_queue_id)
                cons = RING_CMP(cpr->cp_ring_struct, raw_cons);
                rxcmp = (struct rx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
 
-               if (!CMP_VALID(rxcmp, raw_cons, cpr->cp_ring_struct))
+               if (!bnxt_cpr_cmp_valid(rxcmp, raw_cons, cp_ring_size))
                        break;
 
                cmpl_type = CMP_TYPE(rxcmp);
@@ -3186,7 +3222,7 @@ bnxt_rx_descriptor_status_op(void *rx_queue, uint16_t offset)
        struct bnxt_rx_queue *rxq = rx_queue;
        struct bnxt_cp_ring_info *cpr;
        struct bnxt_rx_ring_info *rxr;
-       uint32_t desc, raw_cons;
+       uint32_t desc, raw_cons, cp_ring_size;
        struct bnxt *bp = rxq->bp;
        struct rx_pkt_cmpl *rxcmp;
        int rc;
@@ -3200,6 +3236,7 @@ bnxt_rx_descriptor_status_op(void *rx_queue, uint16_t offset)
 
        rxr = rxq->rx_ring;
        cpr = rxq->cp_ring;
+       cp_ring_size = cpr->cp_ring_struct->ring_size;
 
        /*
         * For the vector receive case, the completion at the requested
@@ -3216,7 +3253,7 @@ bnxt_rx_descriptor_status_op(void *rx_queue, uint16_t offset)
                cons = RING_CMP(cpr->cp_ring_struct, raw_cons);
                rxcmp = (struct rx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
 
-               if (CMP_VALID(rxcmp, raw_cons, cpr->cp_ring_struct))
+               if (bnxt_cpr_cmp_valid(rxcmp, raw_cons, cp_ring_size))
                        return RTE_ETH_RX_DESC_DONE;
 
                /* Check whether rx desc has an mbuf attached. */
@@ -3242,7 +3279,7 @@ bnxt_rx_descriptor_status_op(void *rx_queue, uint16_t offset)
                cons = RING_CMP(cpr->cp_ring_struct, raw_cons);
                rxcmp = (struct rx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
 
-               if (!CMP_VALID(rxcmp, raw_cons, cpr->cp_ring_struct))
+               if (!bnxt_cpr_cmp_valid(rxcmp, raw_cons, cp_ring_size))
                        break;
 
                cmpl_type = CMP_TYPE(rxcmp);
@@ -3294,41 +3331,47 @@ static int
 bnxt_tx_descriptor_status_op(void *tx_queue, uint16_t offset)
 {
        struct bnxt_tx_queue *txq = (struct bnxt_tx_queue *)tx_queue;
-       struct bnxt_tx_ring_info *txr;
-       struct bnxt_cp_ring_info *cpr;
-       struct rte_mbuf **tx_buf;
-       struct tx_pkt_cmpl *txcmp;
-       uint32_t cons, cp_cons;
+       struct bnxt_cp_ring_info *cpr = txq->cp_ring;
+       uint32_t ring_mask, raw_cons, nb_tx_pkts = 0;
+       struct cmpl_base *cp_desc_ring;
        int rc;
 
-       if (!txq)
-               return -EINVAL;
-
        rc = is_bnxt_in_error(txq->bp);
        if (rc)
                return rc;
 
-       cpr = txq->cp_ring;
-       txr = txq->tx_ring;
-
        if (offset >= txq->nb_tx_desc)
                return -EINVAL;
 
-       cons = RING_CMP(cpr->cp_ring_struct, offset);
-       txcmp = (struct tx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
-       cp_cons = cpr->cp_raw_cons;
+       /* Return "desc done" if descriptor is available for use. */
+       if (bnxt_tx_bds_in_hw(txq) <= offset)
+               return RTE_ETH_TX_DESC_DONE;
 
-       if (cons > cp_cons) {
-               if (CMPL_VALID(txcmp, cpr->valid))
-                       return RTE_ETH_TX_DESC_UNAVAIL;
-       } else {
-               if (CMPL_VALID(txcmp, !cpr->valid))
-                       return RTE_ETH_TX_DESC_UNAVAIL;
+       raw_cons = cpr->cp_raw_cons;
+       cp_desc_ring = cpr->cp_desc_ring;
+       ring_mask = cpr->cp_ring_struct->ring_mask;
+
+       /* Check to see if hw has posted a completion for the descriptor. */
+       while (1) {
+               struct tx_cmpl *txcmp;
+               uint32_t cons;
+
+               cons = RING_CMPL(ring_mask, raw_cons);
+               txcmp = (struct tx_cmpl *)&cp_desc_ring[cons];
+
+               if (!bnxt_cpr_cmp_valid(txcmp, raw_cons, ring_mask + 1))
+                       break;
+
+               if (CMP_TYPE(txcmp) == TX_CMPL_TYPE_TX_L2)
+                       nb_tx_pkts += rte_le_to_cpu_32(txcmp->opaque);
+
+               if (nb_tx_pkts > offset)
+                       return RTE_ETH_TX_DESC_DONE;
+
+               raw_cons = NEXT_RAW_CMP(raw_cons);
        }
-       tx_buf = &txr->tx_buf_ring[cons];
-       if (*tx_buf == NULL)
-               return RTE_ETH_TX_DESC_DONE;
 
+       /* Descriptor is pending transmit, not yet completed by hardware. */
        return RTE_ETH_TX_DESC_FULL;
 }
 
@@ -4346,6 +4389,10 @@ err_start:
 err:
        bp->flags |= BNXT_FLAG_FATAL_ERROR;
        bnxt_uninit_resources(bp, false);
+       if (bp->eth_dev->data->dev_conf.intr_conf.rmv)
+               rte_eth_dev_callback_process(bp->eth_dev,
+                                            RTE_ETH_EVENT_INTR_RMV,
+                                            NULL);
        pthread_mutex_unlock(&bp->err_recovery_lock);
        PMD_DRV_LOG(ERR, "Failed to recover from FW reset\n");
 }
@@ -4420,11 +4467,11 @@ static int bnxt_fw_reset_all(struct bnxt *bp)
        int rc = 0;
 
        if (info->flags & BNXT_FLAG_ERROR_RECOVERY_HOST) {
-               /* Reset through master function driver */
+               /* Reset through primary function driver */
                for (i = 0; i < info->reg_array_cnt; i++)
                        bnxt_write_fw_reset_reg(bp, i);
                /* Wait for time specified by FW after triggering reset */
-               rte_delay_ms(info->master_func_wait_period_after_reset);
+               rte_delay_ms(info->primary_func_wait_period_after_reset);
        } else if (info->flags & BNXT_FLAG_ERROR_RECOVERY_CO_CPU) {
                /* Reset with the help of Kong processor */
                rc = bnxt_hwrm_fw_reset(bp);
@@ -4441,8 +4488,8 @@ static void bnxt_fw_reset_cb(void *arg)
        struct bnxt_error_recovery_info *info = bp->recovery_info;
        int rc = 0;
 
-       /* Only Master function can do FW reset */
-       if (bnxt_is_master_func(bp) &&
+       /* Only Primary function can do FW reset */
+       if (bnxt_is_primary_func(bp) &&
            bnxt_is_recovery_enabled(bp)) {
                rc = bnxt_fw_reset_all(bp);
                if (rc) {
@@ -4470,8 +4517,8 @@ static void bnxt_fw_reset_cb(void *arg)
  * advertised by FW in HWRM_ERROR_RECOVERY_QCFG.
  * When the driver detects heartbeat stop or change in reset_counter,
  * it has to trigger a reset to recover from the error condition.
- * A “master PF” is the function who will have the privilege to
- * initiate the chimp reset. The master PF will be elected by the
+ * A “primary function” is the function who will have the privilege to
+ * initiate the chimp reset. The primary function will be elected by the
  * firmware and will be notified through async message.
  */
 static void bnxt_check_fw_health(void *arg)
@@ -4509,8 +4556,8 @@ reset:
 
        PMD_DRV_LOG(ERR, "Detected FW dead condition\n");
 
-       if (bnxt_is_master_func(bp))
-               wait_msec = info->master_func_wait_period;
+       if (bnxt_is_primary_func(bp))
+               wait_msec = info->primary_func_wait_period;
        else
                wait_msec = info->normal_func_wait_period;
 
@@ -4662,7 +4709,7 @@ static int bnxt_alloc_ctx_mem_blk(struct bnxt *bp,
                if (!mz) {
                        mz = rte_memzone_reserve_aligned(mz_name,
                                                rmem->nr_pages * 8,
-                                               SOCKET_ID_ANY,
+                                               bp->eth_dev->device->numa_node,
                                                RTE_MEMZONE_2MB |
                                                RTE_MEMZONE_SIZE_HINT_ONLY |
                                                RTE_MEMZONE_IOVA_CONTIG,
@@ -4685,7 +4732,7 @@ static int bnxt_alloc_ctx_mem_blk(struct bnxt *bp,
        if (!mz) {
                mz = rte_memzone_reserve_aligned(mz_name,
                                                 mem_size,
-                                                SOCKET_ID_ANY,
+                                                bp->eth_dev->device->numa_node,
                                                 RTE_MEMZONE_1GB |
                                                 RTE_MEMZONE_SIZE_HINT_ONLY |
                                                 RTE_MEMZONE_IOVA_CONTIG,
@@ -4979,7 +5026,7 @@ static int bnxt_setup_mac_addr(struct rte_eth_dev *eth_dev)
                PMD_DRV_LOG(INFO, "VF MAC address not assigned by Host PF\n");
                bnxt_eth_hw_addr_random(bp->mac_addr);
                PMD_DRV_LOG(INFO,
-                           "Assign random MAC:%02X:%02X:%02X:%02X:%02X:%02X\n",
+                           "Assign random MAC:" RTE_ETHER_ADDR_PRT_FMT "\n",
                            bp->mac_addr[0], bp->mac_addr[1], bp->mac_addr[2],
                            bp->mac_addr[3], bp->mac_addr[4], bp->mac_addr[5]);
 
@@ -6133,7 +6180,7 @@ bnxt_uninit_resources(struct bnxt *bp, bool reconfig_dev)
                bp->pf->vf_req_buf = NULL;
        }
 
-       rc = bnxt_hwrm_func_driver_unregister(bp, 0);
+       rc = bnxt_hwrm_func_driver_unregister(bp);
        bp->flags &= ~BNXT_FLAG_REGISTERED;
        bnxt_free_ctx_mem(bp);
        if (!reconfig_dev) {
@@ -6500,6 +6547,7 @@ static int bnxt_pci_remove(struct rte_pci_device *pci_dev)
 static struct rte_pci_driver bnxt_rte_pmd = {
        .id_table = bnxt_pci_id_map,
        .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC |
+                       RTE_PCI_DRV_INTR_RMV |
                        RTE_PCI_DRV_PROBE_AGAIN, /* Needed in case of VF-REPs
                                                  * and OVS-DPDK
                                                  */