X-Git-Url: http://git.droids-corp.org/?a=blobdiff_plain;f=drivers%2Fnet%2Fqede%2Fqede_rxtx.c;h=01ea9b47077fd88ae6c3b5484bff6ccab7eada75;hb=f8687bd0b2fc62eb9f6397bbdc75dd03ad4e0048;hp=94f3c78597dad9441c71d029c7866699df777db1;hpb=fbfd99551ca370266f4bfff58ce441cf5cb1203a;p=dpdk.git diff --git a/drivers/net/qede/qede_rxtx.c b/drivers/net/qede/qede_rxtx.c index 94f3c78597..01ea9b4707 100644 --- a/drivers/net/qede/qede_rxtx.c +++ b/drivers/net/qede/qede_rxtx.c @@ -23,8 +23,8 @@ static inline int qede_alloc_rx_buffer(struct qede_rx_queue *rxq) "Failed to allocate rx buffer " "sw_rx_prod %u sw_rx_cons %u mp entries %u free %u", idx, rxq->sw_rx_cons & NUM_RX_BDS(rxq), - rte_mempool_count(rxq->mb_pool), - rte_mempool_free_count(rxq->mb_pool)); + rte_mempool_avail_count(rxq->mb_pool), + rte_mempool_in_use_count(rxq->mb_pool)); return -ENOMEM; } rxq->sw_rx_ring[idx].mbuf = new_mb; @@ -61,7 +61,23 @@ void qede_rx_queue_release(void *rx_queue) rte_free(rxq->sw_rx_ring); rxq->sw_rx_ring = NULL; rte_free(rxq); - rx_queue = NULL; + rxq = NULL; + } +} + +static void qede_tx_queue_release_mbufs(struct qede_tx_queue *txq) +{ + unsigned int i; + + PMD_TX_LOG(DEBUG, txq, "releasing %u mbufs", txq->nb_tx_desc); + + if (txq->sw_tx_ring) { + for (i = 0; i < txq->nb_tx_desc; i++) { + if (txq->sw_tx_ring[i].mbuf) { + rte_pktmbuf_free(txq->sw_tx_ring[i].mbuf); + txq->sw_tx_ring[i].mbuf = NULL; + } + } } } @@ -73,11 +89,11 @@ qede_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx, { struct qede_dev *qdev = dev->data->dev_private; struct ecore_dev *edev = &qdev->edev; - struct rte_eth_dev_data *eth_data = dev->data; + struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode; struct qede_rx_queue *rxq; - uint16_t pkt_len = (uint16_t)dev->data->dev_conf.rxmode.max_rx_pkt_len; + uint16_t max_rx_pkt_len; + uint16_t bufsz; size_t size; - uint16_t data_size; int rc; int i; @@ -111,30 +127,27 @@ qede_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx, rxq->nb_rx_desc = nb_desc; rxq->queue_id = queue_idx; rxq->port_id = dev->data->port_id; - - /* Sanity check */ - data_size = (uint16_t)rte_pktmbuf_data_room_size(mp) - - RTE_PKTMBUF_HEADROOM; - - if (pkt_len > data_size) { - DP_ERR(edev, "MTU %u should not exceed dataroom %u\n", - pkt_len, data_size); - rte_free(rxq); - return -EINVAL; + max_rx_pkt_len = (uint16_t)rxmode->max_rx_pkt_len; + qdev->mtu = max_rx_pkt_len; + + /* Fix up RX buffer size */ + bufsz = (uint16_t)rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM; + if ((rxmode->enable_scatter) || + (max_rx_pkt_len + QEDE_ETH_OVERHEAD) > bufsz) { + if (!dev->data->scattered_rx) { + DP_INFO(edev, "Forcing scatter-gather mode\n"); + dev->data->scattered_rx = 1; + } } + if (dev->data->scattered_rx) + rxq->rx_buf_size = bufsz + QEDE_ETH_OVERHEAD; + else + rxq->rx_buf_size = qdev->mtu + QEDE_ETH_OVERHEAD; + /* Align to cache-line size if needed */ + rxq->rx_buf_size = QEDE_CEIL_TO_CACHE_LINE_SIZE(rxq->rx_buf_size); - qdev->mtu = pkt_len; - rxq->rx_buf_size = pkt_len + QEDE_ETH_OVERHEAD; - - DP_INFO(edev, "MTU = %u ; RX buffer = %u\n", - qdev->mtu, rxq->rx_buf_size); - - if (pkt_len > ETHER_MAX_LEN) { - dev->data->dev_conf.rxmode.jumbo_frame = 1; - DP_NOTICE(edev, false, "jumbo frame enabled\n"); - } else { - dev->data->dev_conf.rxmode.jumbo_frame = 0; - } + DP_INFO(edev, "mtu %u mbufsz %u bd_max_bytes %u scatter_mode %d\n", + qdev->mtu, bufsz, rxq->rx_buf_size, dev->data->scattered_rx); /* Allocate the parallel driver ring for Rx buffers */ size = sizeof(*rxq->sw_rx_ring) * rxq->nb_rx_desc; @@ -156,7 +169,8 @@ qede_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx, ECORE_CHAIN_CNT_TYPE_U16, rxq->nb_rx_desc, sizeof(struct eth_rx_bd), - &rxq->rx_bd_ring); + &rxq->rx_bd_ring, + NULL); if (rc != ECORE_SUCCESS) { DP_NOTICE(edev, false, @@ -166,6 +180,7 @@ qede_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx, rxq->sw_rx_ring = NULL; rte_free(rxq); rxq = NULL; + return -ENOMEM; } /* Allocate FW completion ring */ @@ -175,7 +190,8 @@ qede_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx, ECORE_CHAIN_CNT_TYPE_U16, rxq->nb_rx_desc, sizeof(union eth_rx_cqe), - &rxq->rx_comp_ring); + &rxq->rx_comp_ring, + NULL); if (rc != ECORE_SUCCESS) { DP_NOTICE(edev, false, @@ -185,6 +201,7 @@ qede_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx, rte_free(rxq->sw_rx_ring); rxq->sw_rx_ring = NULL; rte_free(rxq); + return -ENOMEM; } /* Allocate buffers for the Rx ring */ @@ -198,8 +215,6 @@ qede_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx, } dev->data->rx_queues[queue_idx] = rxq; - if (!qdev->rx_queues) - qdev->rx_queues = (struct qede_rx_queue **)dev->data->rx_queues; DP_INFO(edev, "rxq %d num_desc %u rx_buf_size=%u socket %u\n", queue_idx, nb_desc, qdev->mtu, socket_id); @@ -210,22 +225,6 @@ err4: return -ENOMEM; } -static void qede_tx_queue_release_mbufs(struct qede_tx_queue *txq) -{ - unsigned int i; - - PMD_TX_LOG(DEBUG, txq, "releasing %u mbufs\n", txq->nb_tx_desc); - - if (txq->sw_tx_ring != NULL) { - for (i = 0; i < txq->nb_tx_desc; i++) { - if (txq->sw_tx_ring[i].mbuf != NULL) { - rte_pktmbuf_free(txq->sw_tx_ring[i].mbuf); - txq->sw_tx_ring[i].mbuf = NULL; - } - } - } -} - void qede_tx_queue_release(void *tx_queue) { struct qede_tx_queue *txq = tx_queue; @@ -238,7 +237,7 @@ void qede_tx_queue_release(void *tx_queue) } rte_free(txq); } - tx_queue = NULL; + txq = NULL; } int @@ -287,7 +286,8 @@ qede_tx_queue_setup(struct rte_eth_dev *dev, ECORE_CHAIN_CNT_TYPE_U16, txq->nb_tx_desc, sizeof(union eth_tx_bd_types), - &txq->tx_pbl); + &txq->tx_pbl, + NULL); if (rc != ECORE_SUCCESS) { DP_ERR(edev, "Unable to allocate memory for txbd ring on socket %u", @@ -319,10 +319,6 @@ qede_tx_queue_setup(struct rte_eth_dev *dev, (txq->nb_tx_desc - QEDE_DEFAULT_TX_FREE_THRESH); dev->data->tx_queues[queue_idx] = txq; - if (!qdev->tx_queues) - qdev->tx_queues = (struct qede_tx_queue **)dev->data->tx_queues; - - txq->txq_counter = 0; DP_INFO(edev, "txq %u num_desc %u tx_free_thresh %u socket %u\n", @@ -335,32 +331,25 @@ qede_tx_queue_setup(struct rte_eth_dev *dev, static void qede_init_fp(struct qede_dev *qdev) { struct qede_fastpath *fp; - int rss_id, txq_index, tc; + uint8_t i, rss_id, tc; + int fp_rx = qdev->fp_num_rx, rxq = 0, txq = 0; - memset((void *)qdev->fp_array, 0, (QEDE_RSS_CNT(qdev) * + memset((void *)qdev->fp_array, 0, (QEDE_QUEUE_CNT(qdev) * sizeof(*qdev->fp_array))); - memset((void *)qdev->sb_array, 0, (QEDE_RSS_CNT(qdev) * + memset((void *)qdev->sb_array, 0, (QEDE_QUEUE_CNT(qdev) * sizeof(*qdev->sb_array))); - for_each_rss(rss_id) { - fp = &qdev->fp_array[rss_id]; - - fp->qdev = qdev; - fp->rss_id = rss_id; - - /* Point rxq to generic rte queues that was created - * as part of queue creation. - */ - fp->rxq = qdev->rx_queues[rss_id]; - fp->sb_info = &qdev->sb_array[rss_id]; - - for (tc = 0; tc < qdev->num_tc; tc++) { - txq_index = tc * QEDE_RSS_CNT(qdev) + rss_id; - fp->txqs[tc] = qdev->tx_queues[txq_index]; - fp->txqs[tc]->queue_id = txq_index; - /* Updating it to main structure */ - snprintf(fp->name, sizeof(fp->name), "%s-fp-%d", - "qdev", rss_id); + for_each_queue(i) { + fp = &qdev->fp_array[i]; + if (fp_rx) { + fp->type = QEDE_FASTPATH_RX; + fp_rx--; + } else{ + fp->type = QEDE_FASTPATH_TX; } + fp->qdev = qdev; + fp->id = i; + fp->sb_info = &qdev->sb_array[i]; + snprintf(fp->name, sizeof(fp->name), "%s-fp-%d", "qdev", i); } qdev->gro_disable = gro_disable; @@ -386,7 +375,7 @@ int qede_alloc_fp_array(struct qede_dev *qdev) struct ecore_dev *edev = &qdev->edev; int i; - qdev->fp_array = rte_calloc("fp", QEDE_RSS_CNT(qdev), + qdev->fp_array = rte_calloc("fp", QEDE_QUEUE_CNT(qdev), sizeof(*qdev->fp_array), RTE_CACHE_LINE_SIZE); @@ -395,7 +384,7 @@ int qede_alloc_fp_array(struct qede_dev *qdev) return -ENOMEM; } - qdev->sb_array = rte_calloc("sb", QEDE_RSS_CNT(qdev), + qdev->sb_array = rte_calloc("sb", QEDE_QUEUE_CNT(qdev), sizeof(*qdev->sb_array), RTE_CACHE_LINE_SIZE); @@ -437,50 +426,58 @@ qede_alloc_mem_sb(struct qede_dev *qdev, struct ecore_sb_info *sb_info, return 0; } -static int qede_alloc_mem_fp(struct qede_dev *qdev, struct qede_fastpath *fp) -{ - return qede_alloc_mem_sb(qdev, fp->sb_info, fp->rss_id); -} - -static void qede_shrink_txq(struct qede_dev *qdev, uint16_t num_rss) -{ - /* @@@TBD - this should also re-set the qed interrupts */ -} - -/* This function allocates all qede memory at NIC load. */ -static int qede_alloc_mem_load(struct qede_dev *qdev) +int qede_alloc_fp_resc(struct qede_dev *qdev) { - int rc = 0, rss_id; struct ecore_dev *edev = &qdev->edev; + struct qede_fastpath *fp; + uint32_t num_sbs; + uint16_t i; + uint16_t sb_idx; + int rc; - for (rss_id = 0; rss_id < QEDE_RSS_CNT(qdev); rss_id++) { - struct qede_fastpath *fp = &qdev->fp_array[rss_id]; + if (IS_VF(edev)) + ecore_vf_get_num_sbs(ECORE_LEADING_HWFN(edev), &num_sbs); + else + num_sbs = ecore_cxt_get_proto_cid_count + (ECORE_LEADING_HWFN(edev), PROTOCOLID_ETH, NULL); - rc = qede_alloc_mem_fp(qdev, fp); - if (rc) - break; + if (num_sbs == 0) { + DP_ERR(edev, "No status blocks available\n"); + return -EINVAL; } - if (rss_id != QEDE_RSS_CNT(qdev)) { - /* Failed allocating memory for all the queues */ - if (!rss_id) { - DP_ERR(edev, - "Failed to alloc memory for leading queue\n"); - rc = -ENOMEM; - } else { - DP_NOTICE(edev, false, - "Failed to allocate memory for all of " - "RSS queues\n" - "Desired: %d queues, allocated: %d queues\n", - QEDE_RSS_CNT(qdev), rss_id); - qede_shrink_txq(qdev, rss_id); + if (qdev->fp_array) + qede_free_fp_arrays(qdev); + + rc = qede_alloc_fp_array(qdev); + if (rc != 0) + return rc; + + qede_init_fp(qdev); + + for (i = 0; i < QEDE_QUEUE_CNT(qdev); i++) { + fp = &qdev->fp_array[i]; + if (IS_VF(edev)) + sb_idx = i % num_sbs; + else + sb_idx = i; + if (qede_alloc_mem_sb(qdev, fp->sb_info, sb_idx)) { + qede_free_fp_arrays(qdev); + return -ENOMEM; } - qdev->num_rss = rss_id; } return 0; } +void qede_dealloc_fp_resc(struct rte_eth_dev *eth_dev) +{ + struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev); + + qede_free_mem_load(eth_dev); + qede_free_fp_arrays(qdev); +} + static inline void qede_update_rx_prod(struct qede_dev *edev, struct qede_rx_queue *rxq) { @@ -509,149 +506,76 @@ qede_update_rx_prod(struct qede_dev *edev, struct qede_rx_queue *rxq) */ rte_wmb(); - PMD_RX_LOG(DEBUG, rxq, "bd_prod %u cqe_prod %u\n", bd_prod, cqe_prod); -} - -static inline uint32_t -qede_rxfh_indir_default(uint32_t index, uint32_t n_rx_rings) -{ - return index % n_rx_rings; -} - -static void qede_prandom_bytes(uint32_t *buff, size_t bytes) -{ - unsigned int i; - - srand((unsigned int)time(NULL)); - - for (i = 0; i < ECORE_RSS_KEY_SIZE; i++) - buff[i] = rand(); -} - -static int -qede_config_rss(struct rte_eth_dev *eth_dev, - struct qed_update_vport_rss_params *rss_params) -{ - enum rte_eth_rx_mq_mode mode = eth_dev->data->dev_conf.rxmode.mq_mode; - struct rte_eth_rss_conf rss_conf = - eth_dev->data->dev_conf.rx_adv_conf.rss_conf; - struct qede_dev *qdev = eth_dev->data->dev_private; - struct ecore_dev *edev = &qdev->edev; - unsigned int i; - - PMD_INIT_FUNC_TRACE(edev); - - /* Check if RSS conditions are met. - * Note: Even though its meaningless to enable RSS with one queue, it - * could be used to produce RSS Hash, so skipping that check. - */ - - if (!(mode & ETH_MQ_RX_RSS)) { - DP_INFO(edev, "RSS flag is not set\n"); - return -EINVAL; - } - - DP_INFO(edev, "RSS flag is set\n"); - - if (rss_conf.rss_hf == 0) { - DP_NOTICE(edev, false, "No RSS hash function to apply\n"); - return -EINVAL; - } - - if (rss_conf.rss_key != NULL) { - DP_NOTICE(edev, false, - "User provided RSS key is not supported\n"); - return -EINVAL; - } - - memset(rss_params, 0, sizeof(*rss_params)); - - for (i = 0; i < ECORE_RSS_IND_TABLE_SIZE; i++) - rss_params->rss_ind_table[i] = qede_rxfh_indir_default(i, - QEDE_RSS_CNT(qdev)); - - qede_prandom_bytes(rss_params->rss_key, - sizeof(rss_params->rss_key)); - - DP_INFO(edev, "RSS check passes\n"); - - return 0; + PMD_RX_LOG(DEBUG, rxq, "bd_prod %u cqe_prod %u", bd_prod, cqe_prod); } static int qede_start_queues(struct rte_eth_dev *eth_dev, bool clear_stats) { struct qede_dev *qdev = eth_dev->data->dev_private; struct ecore_dev *edev = &qdev->edev; - struct qed_update_vport_rss_params *rss_params = &qdev->rss_params; + struct ecore_queue_start_common_params q_params; struct qed_dev_info *qed_info = &qdev->dev_info.common; struct qed_update_vport_params vport_update_params; - struct qed_start_vport_params start = { 0 }; + struct qede_tx_queue *txq; + struct qede_fastpath *fp; + dma_addr_t p_phys_table; + int txq_index; + uint16_t page_cnt; int vlan_removal_en = 1; int rc, tc, i; - if (!qdev->num_rss) { - DP_ERR(edev, - "Cannot update V-VPORT as active as " - "there are no Rx queues\n"); - return -EINVAL; - } + for_each_queue(i) { + fp = &qdev->fp_array[i]; + if (fp->type & QEDE_FASTPATH_RX) { + p_phys_table = ecore_chain_get_pbl_phys(&fp->rxq-> + rx_comp_ring); + page_cnt = ecore_chain_get_page_cnt(&fp->rxq-> + rx_comp_ring); - start.remove_inner_vlan = vlan_removal_en; - start.gro_enable = !qdev->gro_disable; - start.mtu = qdev->mtu; - start.vport_id = 0; - start.drop_ttl0 = true; - start.clear_stats = clear_stats; + memset(&q_params, 0, sizeof(q_params)); + q_params.queue_id = i; + q_params.vport_id = 0; + q_params.sb = fp->sb_info->igu_sb_id; + q_params.sb_idx = RX_PI; - rc = qdev->ops->vport_start(edev, &start); - if (rc) { - DP_ERR(edev, "Start V-PORT failed %d\n", rc); - return rc; - } - - DP_INFO(edev, - "Start vport ramrod passed, vport_id = %d," - " MTU = %d, vlan_removal_en = %d\n", - start.vport_id, qdev->mtu, vlan_removal_en); - - for_each_rss(i) { - struct qede_fastpath *fp = &qdev->fp_array[i]; - dma_addr_t p_phys_table; - uint16_t page_cnt; - - p_phys_table = ecore_chain_get_pbl_phys(&fp->rxq->rx_comp_ring); - page_cnt = ecore_chain_get_page_cnt(&fp->rxq->rx_comp_ring); + ecore_sb_ack(fp->sb_info, IGU_INT_DISABLE, 0); - ecore_sb_ack(fp->sb_info, IGU_INT_DISABLE, 0); /* @DPDK */ - - rc = qdev->ops->q_rx_start(edev, i, i, 0, - fp->sb_info->igu_sb_id, - RX_PI, + rc = qdev->ops->q_rx_start(edev, i, &q_params, fp->rxq->rx_buf_size, fp->rxq->rx_bd_ring.p_phys_addr, p_phys_table, page_cnt, &fp->rxq->hw_rxq_prod_addr); - if (rc) { - DP_ERR(edev, "Start RXQ #%d failed %d\n", i, rc); - return rc; - } + if (rc) { + DP_ERR(edev, "Start rxq #%d failed %d\n", + fp->rxq->queue_id, rc); + return rc; + } - fp->rxq->hw_cons_ptr = &fp->sb_info->sb_virt->pi_array[RX_PI]; + fp->rxq->hw_cons_ptr = + &fp->sb_info->sb_virt->pi_array[RX_PI]; - qede_update_rx_prod(qdev, fp->rxq); + qede_update_rx_prod(qdev, fp->rxq); + } + if (!(fp->type & QEDE_FASTPATH_TX)) + continue; for (tc = 0; tc < qdev->num_tc; tc++) { - struct qede_tx_queue *txq = fp->txqs[tc]; - int txq_index = tc * QEDE_RSS_CNT(qdev) + i; + txq = fp->txqs[tc]; + txq_index = tc * QEDE_RSS_COUNT(qdev) + i; p_phys_table = ecore_chain_get_pbl_phys(&txq->tx_pbl); page_cnt = ecore_chain_get_page_cnt(&txq->tx_pbl); - rc = qdev->ops->q_tx_start(edev, i, txq_index, - 0, - fp->sb_info->igu_sb_id, - TX_PI(tc), - p_phys_table, page_cnt, + + memset(&q_params, 0, sizeof(q_params)); + q_params.queue_id = txq->queue_id; + q_params.vport_id = 0; + q_params.sb = fp->sb_info->igu_sb_id; + q_params.sb_idx = TX_PI(tc); + + rc = qdev->ops->q_tx_start(edev, i, &q_params, + p_phys_table, + page_cnt, /* **pp_doorbell */ &txq->doorbell_addr); if (rc) { DP_ERR(edev, "Start txq %u failed %d\n", @@ -675,7 +599,9 @@ static int qede_start_queues(struct rte_eth_dev *eth_dev, bool clear_stats) /* Prepare and send the vport enable */ memset(&vport_update_params, 0, sizeof(vport_update_params)); - vport_update_params.vport_id = start.vport_id; + /* Update MTU via vport update */ + vport_update_params.mtu = qdev->mtu; + vport_update_params.vport_id = 0; vport_update_params.update_vport_active_flg = 1; vport_update_params.vport_active_flg = 1; @@ -686,19 +612,6 @@ static int qede_start_queues(struct rte_eth_dev *eth_dev, bool clear_stats) vport_update_params.tx_switching_flg = 1; } - if (!qede_config_rss(eth_dev, rss_params)) { - vport_update_params.update_rss_flg = 1; - - qdev->rss_enabled = 1; - DP_INFO(edev, "Updating RSS flag\n"); - } else { - qdev->rss_enabled = 0; - DP_INFO(edev, "Not Updating RSS flag\n"); - } - - rte_memcpy(&vport_update_params.rss_params, rss_params, - sizeof(*rss_params)); - rc = qdev->ops->vport_update(edev, &vport_update_params); if (rc) { DP_ERR(edev, "Update V-PORT failed %d\n", rc); @@ -708,79 +621,64 @@ static int qede_start_queues(struct rte_eth_dev *eth_dev, bool clear_stats) return 0; } -#ifdef ENC_SUPPORTED static bool qede_tunn_exist(uint16_t flag) { return !!((PARSING_AND_ERR_FLAGS_TUNNELEXIST_MASK << PARSING_AND_ERR_FLAGS_TUNNELEXIST_SHIFT) & flag); } -static inline uint8_t qede_check_tunn_csum(uint16_t flag) +/* + * qede_check_tunn_csum_l4: + * Returns: + * 1 : If L4 csum is enabled AND if the validation has failed. + * 0 : Otherwise + */ +static inline uint8_t qede_check_tunn_csum_l4(uint16_t flag) { - uint8_t tcsum = 0; - uint16_t csum_flag = 0; - if ((PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMWASCALCULATED_MASK << PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMWASCALCULATED_SHIFT) & flag) - csum_flag |= PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMERROR_MASK << - PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMERROR_SHIFT; - - if ((PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_MASK << - PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_SHIFT) & flag) { - csum_flag |= PARSING_AND_ERR_FLAGS_L4CHKSMERROR_MASK << - PARSING_AND_ERR_FLAGS_L4CHKSMERROR_SHIFT; - tcsum = QEDE_TUNN_CSUM_UNNECESSARY; - } - - csum_flag |= PARSING_AND_ERR_FLAGS_TUNNELIPHDRERROR_MASK << - PARSING_AND_ERR_FLAGS_TUNNELIPHDRERROR_SHIFT | - PARSING_AND_ERR_FLAGS_IPHDRERROR_MASK << - PARSING_AND_ERR_FLAGS_IPHDRERROR_SHIFT; + return !!((PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMERROR_MASK << + PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMERROR_SHIFT) & flag); - if (csum_flag & flag) - return QEDE_CSUM_ERROR; - - return QEDE_CSUM_UNNECESSARY | tcsum; -} -#else -static inline uint8_t qede_tunn_exist(uint16_t flag) -{ return 0; } -static inline uint8_t qede_check_tunn_csum(uint16_t flag) +static inline uint8_t qede_check_notunn_csum_l4(uint16_t flag) { - return 0; -} -#endif - -static inline uint8_t qede_check_notunn_csum(uint16_t flag) -{ - uint8_t csum = 0; - uint16_t csum_flag = 0; - if ((PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_MASK << - PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_SHIFT) & flag) { - csum_flag |= PARSING_AND_ERR_FLAGS_L4CHKSMERROR_MASK << - PARSING_AND_ERR_FLAGS_L4CHKSMERROR_SHIFT; - csum = QEDE_CSUM_UNNECESSARY; - } - - csum_flag |= PARSING_AND_ERR_FLAGS_IPHDRERROR_MASK << - PARSING_AND_ERR_FLAGS_IPHDRERROR_SHIFT; + PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_SHIFT) & flag) + return !!((PARSING_AND_ERR_FLAGS_L4CHKSMERROR_MASK << + PARSING_AND_ERR_FLAGS_L4CHKSMERROR_SHIFT) & flag); - if (csum_flag & flag) - return QEDE_CSUM_ERROR; - - return csum; + return 0; } -static inline uint8_t qede_check_csum(uint16_t flag) +static inline uint8_t +qede_check_notunn_csum_l3(struct rte_mbuf *m, uint16_t flag) { - if (likely(!qede_tunn_exist(flag))) - return qede_check_notunn_csum(flag); - else - return qede_check_tunn_csum(flag); + struct ipv4_hdr *ip; + uint16_t pkt_csum; + uint16_t calc_csum; + uint16_t val; + + val = ((PARSING_AND_ERR_FLAGS_IPHDRERROR_MASK << + PARSING_AND_ERR_FLAGS_IPHDRERROR_SHIFT) & flag); + + if (unlikely(val)) { + m->packet_type = qede_rx_cqe_to_pkt_type(flag); + if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) { + ip = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *, + sizeof(struct ether_hdr)); + pkt_csum = ip->hdr_checksum; + ip->hdr_checksum = 0; + calc_csum = rte_ipv4_cksum(ip); + ip->hdr_checksum = pkt_csum; + return (calc_csum != pkt_csum); + } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) { + return 1; + } + } + return 0; } static inline void qede_rx_bd_ring_consume(struct qede_rx_queue *rxq) @@ -825,20 +723,126 @@ qede_recycle_rx_bd_ring(struct qede_rx_queue *rxq, static inline uint32_t qede_rx_cqe_to_pkt_type(uint16_t flags) { - uint32_t p_type; - /* TBD - L4 indications needed ? */ - uint16_t protocol = ((PARSING_AND_ERR_FLAGS_L3TYPE_MASK << - PARSING_AND_ERR_FLAGS_L3TYPE_SHIFT) & flags); - - /* protocol = 3 means LLC/SNAP over Ethernet */ - if (unlikely(protocol == 0 || protocol == 3)) - p_type = RTE_PTYPE_UNKNOWN; - else if (protocol == 1) - p_type = RTE_PTYPE_L3_IPV4; - else if (protocol == 2) - p_type = RTE_PTYPE_L3_IPV6; - - return RTE_PTYPE_L2_ETHER | p_type; + uint16_t val; + + /* Lookup table */ + static const uint32_t + ptype_lkup_tbl[QEDE_PKT_TYPE_MAX] __rte_cache_aligned = { + [QEDE_PKT_TYPE_IPV4] = RTE_PTYPE_L3_IPV4, + [QEDE_PKT_TYPE_IPV6] = RTE_PTYPE_L3_IPV6, + [QEDE_PKT_TYPE_IPV4_TCP] = RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_TCP, + [QEDE_PKT_TYPE_IPV6_TCP] = RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_TCP, + [QEDE_PKT_TYPE_IPV4_UDP] = RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_UDP, + [QEDE_PKT_TYPE_IPV6_UDP] = RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_UDP, + }; + + /* Bits (0..3) provides L3/L4 protocol type */ + val = ((PARSING_AND_ERR_FLAGS_L3TYPE_MASK << + PARSING_AND_ERR_FLAGS_L3TYPE_SHIFT) | + (PARSING_AND_ERR_FLAGS_L4PROTOCOL_MASK << + PARSING_AND_ERR_FLAGS_L4PROTOCOL_SHIFT)) & flags; + + if (val < QEDE_PKT_TYPE_MAX) + return ptype_lkup_tbl[val] | RTE_PTYPE_L2_ETHER; + else + return RTE_PTYPE_UNKNOWN; +} + +static inline uint32_t qede_rx_cqe_to_tunn_pkt_type(uint16_t flags) +{ + uint32_t val; + + /* Lookup table */ + static const uint32_t + ptype_tunn_lkup_tbl[QEDE_PKT_TYPE_TUNN_MAX_TYPE] __rte_cache_aligned = { + [QEDE_PKT_TYPE_UNKNOWN] = RTE_PTYPE_UNKNOWN, + [QEDE_PKT_TYPE_TUNN_GENEVE] = RTE_PTYPE_TUNNEL_GENEVE, + [QEDE_PKT_TYPE_TUNN_GRE] = RTE_PTYPE_TUNNEL_GRE, + [QEDE_PKT_TYPE_TUNN_VXLAN] = RTE_PTYPE_TUNNEL_VXLAN, + [QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_GENEVE] = + RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L2_ETHER, + [QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_GRE] = + RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L2_ETHER, + [QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_VXLAN] = + RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L2_ETHER, + [QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_GENEVE] = + RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L2_ETHER, + [QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_GRE] = + RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L2_ETHER, + [QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_VXLAN] = + RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L2_ETHER, + [QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_GENEVE] = + RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV4, + [QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_GRE] = + RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV4, + [QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_VXLAN] = + RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV4, + [QEDE_PKT_TYPE_TUNN_IPV4_TENID_EXIST_GENEVE] = + RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV4, + [QEDE_PKT_TYPE_TUNN_IPV4_TENID_EXIST_GRE] = + RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV4, + [QEDE_PKT_TYPE_TUNN_IPV4_TENID_EXIST_VXLAN] = + RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV4, + [QEDE_PKT_TYPE_TUNN_IPV6_TENID_NOEXIST_GENEVE] = + RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV6, + [QEDE_PKT_TYPE_TUNN_IPV6_TENID_NOEXIST_GRE] = + RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV6, + [QEDE_PKT_TYPE_TUNN_IPV6_TENID_NOEXIST_VXLAN] = + RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV6, + [QEDE_PKT_TYPE_TUNN_IPV6_TENID_EXIST_GENEVE] = + RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV6, + [QEDE_PKT_TYPE_TUNN_IPV6_TENID_EXIST_GRE] = + RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV6, + [QEDE_PKT_TYPE_TUNN_IPV6_TENID_EXIST_VXLAN] = + RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV6, + }; + + /* Cover bits[4-0] to include tunn_type and next protocol */ + val = ((ETH_TUNNEL_PARSING_FLAGS_TYPE_MASK << + ETH_TUNNEL_PARSING_FLAGS_TYPE_SHIFT) | + (ETH_TUNNEL_PARSING_FLAGS_NEXT_PROTOCOL_MASK << + ETH_TUNNEL_PARSING_FLAGS_NEXT_PROTOCOL_SHIFT)) & flags; + + if (val < QEDE_PKT_TYPE_TUNN_MAX_TYPE) + return ptype_tunn_lkup_tbl[val]; + else + return RTE_PTYPE_UNKNOWN; +} + +static inline int +qede_process_sg_pkts(void *p_rxq, struct rte_mbuf *rx_mb, + uint8_t num_segs, uint16_t pkt_len) +{ + struct qede_rx_queue *rxq = p_rxq; + struct qede_dev *qdev = rxq->qdev; + struct ecore_dev *edev = &qdev->edev; + register struct rte_mbuf *seg1 = NULL; + register struct rte_mbuf *seg2 = NULL; + uint16_t sw_rx_index; + uint16_t cur_size; + + seg1 = rx_mb; + while (num_segs) { + cur_size = pkt_len > rxq->rx_buf_size ? rxq->rx_buf_size : + pkt_len; + if (unlikely(!cur_size)) { + PMD_RX_LOG(ERR, rxq, "Length is 0 while %u BDs" + " left for mapping jumbo", num_segs); + qede_recycle_rx_bd_ring(rxq, qdev, num_segs); + return -EINVAL; + } + sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS(rxq); + seg2 = rxq->sw_rx_ring[sw_rx_index].mbuf; + qede_rx_bd_ring_consume(rxq); + pkt_len -= cur_size; + seg2->data_len = cur_size; + seg1->next = seg2; + seg1 = seg1->next; + num_segs--; + rxq->rx_segs++; + } + + return 0; } uint16_t @@ -853,12 +857,18 @@ qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts) union eth_rx_cqe *cqe; struct eth_fast_path_rx_reg_cqe *fp_cqe; register struct rte_mbuf *rx_mb = NULL; + register struct rte_mbuf *seg1 = NULL; enum eth_rx_cqe_type cqe_type; - uint16_t len, pad; + uint16_t pkt_len; /* Sum of all BD segments */ + uint16_t len; /* Length of first BD */ + uint8_t num_segs = 1; + uint16_t pad; uint16_t preload_idx; uint8_t csum_flag; uint16_t parse_flag; enum rss_hash_type htype; + uint8_t tunn_parse_flag; + uint8_t j; hw_comp_cons = rte_le_to_cpu_16(*rxq->hw_cons_ptr); sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring); @@ -875,9 +885,9 @@ qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts) cqe_type = cqe->fast_path_regular.type; if (unlikely(cqe_type == ETH_RX_CQE_TYPE_SLOW_PATH)) { - PMD_RX_LOG(DEBUG, rxq, "Got a slowath CQE\n"); + PMD_RX_LOG(DEBUG, rxq, "Got a slowath CQE"); - qdev->ops->eth_cqe_completion(edev, fp->rss_id, + qdev->ops->eth_cqe_completion(edev, fp->id, (struct eth_slow_path_rx_cqe *)cqe); goto next_cqe; } @@ -891,12 +901,13 @@ qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts) fp_cqe = &cqe->fast_path_regular; len = rte_le_to_cpu_16(fp_cqe->len_on_first_bd); + pkt_len = rte_le_to_cpu_16(fp_cqe->pkt_len); pad = fp_cqe->placement_offset; assert((len + pad) <= rx_mb->buf_len); PMD_RX_LOG(DEBUG, rxq, "CQE type = 0x%x, flags = 0x%x, vlan = 0x%x" - " len = %u, parsing_flags = %d\n", + " len = %u, parsing_flags = %d", cqe_type, fp_cqe->bitfields, rte_le_to_cpu_16(fp_cqe->vlan_tag), len, rte_le_to_cpu_16(fp_cqe->pars_flags.flags)); @@ -904,53 +915,98 @@ qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts) /* If this is an error packet then drop it */ parse_flag = rte_le_to_cpu_16(cqe->fast_path_regular.pars_flags.flags); - csum_flag = qede_check_csum(parse_flag); - if (unlikely(csum_flag == QEDE_CSUM_ERROR)) { - PMD_RX_LOG(ERR, rxq, - "CQE in CONS = %u has error, flags = 0x%x " - "dropping incoming packet\n", - sw_comp_cons, parse_flag); - rxq->rx_hw_errors++; - qede_recycle_rx_bd_ring(rxq, qdev, fp_cqe->bd_num); - goto next_cqe; + + rx_mb->ol_flags = 0; + + if (qede_tunn_exist(parse_flag)) { + PMD_RX_LOG(DEBUG, rxq, "Rx tunneled packet"); + if (unlikely(qede_check_tunn_csum_l4(parse_flag))) { + PMD_RX_LOG(ERR, rxq, + "L4 csum failed, flags = 0x%x", + parse_flag); + rxq->rx_hw_errors++; + rx_mb->ol_flags |= PKT_RX_L4_CKSUM_BAD; + } else { + tunn_parse_flag = + fp_cqe->tunnel_pars_flags.flags; + rx_mb->packet_type = + qede_rx_cqe_to_tunn_pkt_type( + tunn_parse_flag); + } + } else { + PMD_RX_LOG(DEBUG, rxq, "Rx non-tunneled packet"); + if (unlikely(qede_check_notunn_csum_l4(parse_flag))) { + PMD_RX_LOG(ERR, rxq, + "L4 csum failed, flags = 0x%x", + parse_flag); + rxq->rx_hw_errors++; + rx_mb->ol_flags |= PKT_RX_L4_CKSUM_BAD; + } else if (unlikely(qede_check_notunn_csum_l3(rx_mb, + parse_flag))) { + PMD_RX_LOG(ERR, rxq, + "IP csum failed, flags = 0x%x", + parse_flag); + rxq->rx_hw_errors++; + rx_mb->ol_flags |= PKT_RX_IP_CKSUM_BAD; + } else { + rx_mb->packet_type = + qede_rx_cqe_to_pkt_type(parse_flag); + } } + PMD_RX_LOG(INFO, rxq, "packet_type 0x%x", rx_mb->packet_type); + if (unlikely(qede_alloc_rx_buffer(rxq) != 0)) { PMD_RX_LOG(ERR, rxq, "New buffer allocation failed," - "dropping incoming packet\n"); + "dropping incoming packet"); qede_recycle_rx_bd_ring(rxq, qdev, fp_cqe->bd_num); rte_eth_devices[rxq->port_id]. data->rx_mbuf_alloc_failed++; rxq->rx_alloc_errors++; break; } - qede_rx_bd_ring_consume(rxq); + if (fp_cqe->bd_num > 1) { + PMD_RX_LOG(DEBUG, rxq, "Jumbo-over-BD packet: %02x BDs" + " len on first: %04x Total Len: %04x", + fp_cqe->bd_num, len, pkt_len); + num_segs = fp_cqe->bd_num - 1; + seg1 = rx_mb; + if (qede_process_sg_pkts(p_rxq, seg1, num_segs, + pkt_len - len)) + goto next_cqe; + for (j = 0; j < num_segs; j++) { + if (qede_alloc_rx_buffer(rxq)) { + PMD_RX_LOG(ERR, rxq, + "Buffer allocation failed"); + rte_eth_devices[rxq->port_id]. + data->rx_mbuf_alloc_failed++; + rxq->rx_alloc_errors++; + break; + } + rxq->rx_segs++; + } + } + rxq->rx_segs++; /* for the first segment */ /* Prefetch next mbuf while processing current one. */ preload_idx = rxq->sw_rx_cons & NUM_RX_BDS(rxq); rte_prefetch0(rxq->sw_rx_ring[preload_idx].mbuf); - if (fp_cqe->bd_num != 1) - PMD_RX_LOG(DEBUG, rxq, - "Jumbo-over-BD packet not supported\n"); - - /* Update MBUF fields */ - rx_mb->ol_flags = 0; + /* Update rest of the MBUF fields */ rx_mb->data_off = pad + RTE_PKTMBUF_HEADROOM; - rx_mb->nb_segs = 1; + rx_mb->nb_segs = fp_cqe->bd_num; rx_mb->data_len = len; - rx_mb->pkt_len = len; + rx_mb->pkt_len = pkt_len; rx_mb->port = rxq->port_id; - rx_mb->packet_type = qede_rx_cqe_to_pkt_type(parse_flag); htype = (uint8_t)GET_FIELD(fp_cqe->bitfields, ETH_FAST_PATH_RX_REG_CQE_RSS_HASH_TYPE); - if (qdev->rss_enabled && htype) { + if (qdev->rss_enable && htype) { rx_mb->ol_flags |= PKT_RX_RSS_HASH; rx_mb->hash.rss = rte_le_to_cpu_32(fp_cqe->rss_hash); - PMD_RX_LOG(DEBUG, rxq, "Hash result 0x%x\n", + PMD_RX_LOG(DEBUG, rxq, "Hash result 0x%x", rx_mb->hash.rss); } @@ -968,6 +1024,7 @@ qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts) */ rx_mb->vlan_tci = rte_le_to_cpu_16(fp_cqe->vlan_tag); rx_mb->ol_flags |= PKT_RX_QINQ_PKT; + rx_mb->vlan_tci_outer = 0; } rx_pkts[rx_pkt] = rx_mb; @@ -977,7 +1034,7 @@ next_cqe: sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring); if (rx_pkt == nb_pkts) { PMD_RX_LOG(DEBUG, rxq, - "Budget reached nb_pkts=%u received=%u\n", + "Budget reached nb_pkts=%u received=%u", rx_pkt, nb_pkts); break; } @@ -985,7 +1042,9 @@ next_cqe: qede_update_rx_prod(qdev, rxq); - PMD_RX_LOG(DEBUG, rxq, "rx_pkts=%u core=%d\n", rx_pkt, rte_lcore_id()); + rxq->rcv_pkts += rx_pkt; + + PMD_RX_LOG(DEBUG, rxq, "rx_pkts=%u core=%d", rx_pkt, rte_lcore_id()); return rx_pkt; } @@ -993,24 +1052,28 @@ next_cqe: static inline int qede_free_tx_pkt(struct ecore_dev *edev, struct qede_tx_queue *txq) { - uint16_t idx = TX_CONS(txq); + uint16_t nb_segs, idx = TX_CONS(txq); struct eth_tx_bd *tx_data_bd; struct rte_mbuf *mbuf = txq->sw_tx_ring[idx].mbuf; if (unlikely(!mbuf)) { + PMD_TX_LOG(ERR, txq, "null mbuf"); PMD_TX_LOG(ERR, txq, - "null mbuf nb_tx_desc %u nb_tx_avail %u " - "sw_tx_cons %u sw_tx_prod %u\n", + "tx_desc %u tx_avail %u tx_cons %u tx_prod %u", txq->nb_tx_desc, txq->nb_tx_avail, idx, TX_PROD(txq)); return -1; } - /* Free now */ - rte_pktmbuf_free_seg(mbuf); + nb_segs = mbuf->nb_segs; + while (nb_segs) { + /* It's like consuming rxbuf in recv() */ + ecore_chain_consume(&txq->tx_pbl); + txq->nb_tx_avail++; + nb_segs--; + } + rte_pktmbuf_free(mbuf); txq->sw_tx_ring[idx].mbuf = NULL; - ecore_chain_consume(&txq->tx_pbl); - txq->nb_tx_avail++; return 0; } @@ -1020,52 +1083,103 @@ qede_process_tx_compl(struct ecore_dev *edev, struct qede_tx_queue *txq) { uint16_t tx_compl = 0; uint16_t hw_bd_cons; - int rc; hw_bd_cons = rte_le_to_cpu_16(*txq->hw_cons_ptr); rte_compiler_barrier(); while (hw_bd_cons != ecore_chain_get_cons_idx(&txq->tx_pbl)) { - rc = qede_free_tx_pkt(edev, txq); - if (rc) { - DP_NOTICE(edev, false, - "hw_bd_cons = %d, chain_cons=%d\n", - hw_bd_cons, - ecore_chain_get_cons_idx(&txq->tx_pbl)); + if (qede_free_tx_pkt(edev, txq)) { + PMD_TX_LOG(ERR, txq, + "hw_bd_cons = %u, chain_cons = %u", + hw_bd_cons, + ecore_chain_get_cons_idx(&txq->tx_pbl)); break; } txq->sw_tx_cons++; /* Making TXD available */ tx_compl++; } - PMD_TX_LOG(DEBUG, txq, "Tx compl %u sw_tx_cons %u avail %u\n", + PMD_TX_LOG(DEBUG, txq, "Tx compl %u sw_tx_cons %u avail %u", tx_compl, txq->sw_tx_cons, txq->nb_tx_avail); return tx_compl; } +/* Populate scatter gather buffer descriptor fields */ +static inline uint8_t +qede_encode_sg_bd(struct qede_tx_queue *p_txq, struct rte_mbuf *m_seg, + struct eth_tx_1st_bd *bd1) +{ + struct qede_tx_queue *txq = p_txq; + struct eth_tx_2nd_bd *bd2 = NULL; + struct eth_tx_3rd_bd *bd3 = NULL; + struct eth_tx_bd *tx_bd = NULL; + dma_addr_t mapping; + uint8_t nb_segs = 1; /* min one segment per packet */ + + /* Check for scattered buffers */ + while (m_seg) { + if (nb_segs == 1) { + bd2 = (struct eth_tx_2nd_bd *) + ecore_chain_produce(&txq->tx_pbl); + memset(bd2, 0, sizeof(*bd2)); + mapping = rte_mbuf_data_dma_addr(m_seg); + QEDE_BD_SET_ADDR_LEN(bd2, mapping, m_seg->data_len); + PMD_TX_LOG(DEBUG, txq, "BD2 len %04x", + m_seg->data_len); + } else if (nb_segs == 2) { + bd3 = (struct eth_tx_3rd_bd *) + ecore_chain_produce(&txq->tx_pbl); + memset(bd3, 0, sizeof(*bd3)); + mapping = rte_mbuf_data_dma_addr(m_seg); + QEDE_BD_SET_ADDR_LEN(bd3, mapping, m_seg->data_len); + PMD_TX_LOG(DEBUG, txq, "BD3 len %04x", + m_seg->data_len); + } else { + tx_bd = (struct eth_tx_bd *) + ecore_chain_produce(&txq->tx_pbl); + memset(tx_bd, 0, sizeof(*tx_bd)); + mapping = rte_mbuf_data_dma_addr(m_seg); + QEDE_BD_SET_ADDR_LEN(tx_bd, mapping, m_seg->data_len); + PMD_TX_LOG(DEBUG, txq, "BD len %04x", + m_seg->data_len); + } + nb_segs++; + m_seg = m_seg->next; + } + + /* Return total scattered buffers */ + return nb_segs; +} + uint16_t qede_xmit_pkts(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts) { struct qede_tx_queue *txq = p_txq; struct qede_dev *qdev = txq->qdev; struct ecore_dev *edev = &qdev->edev; - struct qede_fastpath *fp = &qdev->fp_array[txq->queue_id]; - struct eth_tx_1st_bd *first_bd; + struct qede_fastpath *fp; + struct eth_tx_1st_bd *bd1; + struct rte_mbuf *mbuf; + struct rte_mbuf *m_seg = NULL; uint16_t nb_tx_pkts; - uint16_t nb_pkt_sent = 0; uint16_t bd_prod; uint16_t idx; uint16_t tx_count; + uint16_t nb_frags; + uint16_t nb_pkt_sent = 0; + + fp = &qdev->fp_array[QEDE_RSS_COUNT(qdev) + txq->queue_id]; if (unlikely(txq->nb_tx_avail < txq->tx_free_thresh)) { - PMD_TX_LOG(DEBUG, txq, "send=%u avail=%u free_thresh=%u\n", + PMD_TX_LOG(DEBUG, txq, "send=%u avail=%u free_thresh=%u", nb_pkts, txq->nb_tx_avail, txq->tx_free_thresh); (void)qede_process_tx_compl(edev, txq); } - nb_tx_pkts = RTE_MIN(nb_pkts, (txq->nb_tx_avail / MAX_NUM_TX_BDS)); + nb_tx_pkts = RTE_MIN(nb_pkts, (txq->nb_tx_avail / + ETH_TX_MAX_BDS_PER_NON_LSO_PACKET)); if (unlikely(nb_tx_pkts == 0)) { - PMD_TX_LOG(DEBUG, txq, "Out of BDs nb_pkts=%u avail=%u\n", + PMD_TX_LOG(DEBUG, txq, "Out of BDs nb_pkts=%u avail=%u", nb_pkts, txq->nb_tx_avail); return 0; } @@ -1074,111 +1188,171 @@ qede_xmit_pkts(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts) while (nb_tx_pkts--) { /* Fill the entry in the SW ring and the BDs in the FW ring */ idx = TX_PROD(txq); - struct rte_mbuf *mbuf = *tx_pkts++; + mbuf = *tx_pkts++; txq->sw_tx_ring[idx].mbuf = mbuf; - first_bd = (struct eth_tx_1st_bd *) - ecore_chain_produce(&txq->tx_pbl); - first_bd->data.bd_flags.bitfields = - 1 << ETH_TX_1ST_BD_FLAGS_START_BD_SHIFT; + bd1 = (struct eth_tx_1st_bd *)ecore_chain_produce(&txq->tx_pbl); + bd1->data.bd_flags.bitfields = + 1 << ETH_TX_1ST_BD_FLAGS_START_BD_SHIFT; + /* FW 8.10.x specific change */ + bd1->data.bitfields = + (mbuf->pkt_len & ETH_TX_DATA_1ST_BD_PKT_LEN_MASK) + << ETH_TX_DATA_1ST_BD_PKT_LEN_SHIFT; /* Map MBUF linear data for DMA and set in the first BD */ - QEDE_BD_SET_ADDR_LEN(first_bd, rte_mbuf_data_dma_addr(mbuf), + QEDE_BD_SET_ADDR_LEN(bd1, rte_mbuf_data_dma_addr(mbuf), mbuf->data_len); + PMD_TX_LOG(INFO, txq, "BD1 len %04x", mbuf->data_len); + + if (RTE_ETH_IS_TUNNEL_PKT(mbuf->packet_type)) { + PMD_TX_LOG(INFO, txq, "Tx tunnel packet"); + /* First indicate its a tunnel pkt */ + bd1->data.bd_flags.bitfields |= + ETH_TX_DATA_1ST_BD_TUNN_FLAG_MASK << + ETH_TX_DATA_1ST_BD_TUNN_FLAG_SHIFT; + + /* Legacy FW had flipped behavior in regard to this bit + * i.e. it needed to set to prevent FW from touching + * encapsulated packets when it didn't need to. + */ + if (unlikely(txq->is_legacy)) + bd1->data.bitfields ^= + 1 << ETH_TX_DATA_1ST_BD_TUNN_FLAG_SHIFT; + + /* Outer IP checksum offload */ + if (mbuf->ol_flags & PKT_TX_OUTER_IP_CKSUM) { + PMD_TX_LOG(INFO, txq, "OuterIP csum offload"); + bd1->data.bd_flags.bitfields |= + ETH_TX_1ST_BD_FLAGS_TUNN_IP_CSUM_MASK << + ETH_TX_1ST_BD_FLAGS_TUNN_IP_CSUM_SHIFT; + } + + /* Outer UDP checksum offload */ + bd1->data.bd_flags.bitfields |= + ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_MASK << + ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_SHIFT; + } /* Descriptor based VLAN insertion */ if (mbuf->ol_flags & (PKT_TX_VLAN_PKT | PKT_TX_QINQ_PKT)) { - first_bd->data.vlan = rte_cpu_to_le_16(mbuf->vlan_tci); - first_bd->data.bd_flags.bitfields |= + PMD_TX_LOG(INFO, txq, "Insert VLAN 0x%x", + mbuf->vlan_tci); + bd1->data.vlan = rte_cpu_to_le_16(mbuf->vlan_tci); + bd1->data.bd_flags.bitfields |= 1 << ETH_TX_1ST_BD_FLAGS_VLAN_INSERTION_SHIFT; } /* Offload the IP checksum in the hardware */ if (mbuf->ol_flags & PKT_TX_IP_CKSUM) { - first_bd->data.bd_flags.bitfields |= + PMD_TX_LOG(INFO, txq, "IP csum offload"); + bd1->data.bd_flags.bitfields |= 1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT; } /* L4 checksum offload (tcp or udp) */ if (mbuf->ol_flags & (PKT_TX_TCP_CKSUM | PKT_TX_UDP_CKSUM)) { - first_bd->data.bd_flags.bitfields |= + PMD_TX_LOG(INFO, txq, "L4 csum offload"); + bd1->data.bd_flags.bitfields |= 1 << ETH_TX_1ST_BD_FLAGS_L4_CSUM_SHIFT; /* IPv6 + extn. -> later */ } - first_bd->data.nbds = MAX_NUM_TX_BDS; + + /* Handle fragmented MBUF */ + m_seg = mbuf->next; + /* Encode scatter gather buffer descriptors if required */ + nb_frags = qede_encode_sg_bd(txq, m_seg, bd1); + bd1->data.nbds = nb_frags; + txq->nb_tx_avail -= nb_frags; txq->sw_tx_prod++; rte_prefetch0(txq->sw_tx_ring[TX_PROD(txq)].mbuf); - txq->nb_tx_avail--; bd_prod = rte_cpu_to_le_16(ecore_chain_get_prod_idx(&txq->tx_pbl)); nb_pkt_sent++; + txq->xmit_pkts++; + PMD_TX_LOG(INFO, txq, "nbds = %d pkt_len = %04x", + bd1->data.nbds, mbuf->pkt_len); } /* Write value of prod idx into bd_prod */ txq->tx_db.data.bd_prod = bd_prod; rte_wmb(); rte_compiler_barrier(); - DIRECT_REG_WR(edev, txq->doorbell_addr, txq->tx_db.raw); + DIRECT_REG_WR_RELAXED(edev, txq->doorbell_addr, txq->tx_db.raw); rte_wmb(); /* Check again for Tx completions */ (void)qede_process_tx_compl(edev, txq); - PMD_TX_LOG(DEBUG, txq, "to_send=%u can_send=%u sent=%u core=%d\n", + PMD_TX_LOG(DEBUG, txq, "to_send=%u can_send=%u sent=%u core=%d", nb_pkts, tx_count, nb_pkt_sent, rte_lcore_id()); return nb_pkt_sent; } +static void qede_init_fp_queue(struct rte_eth_dev *eth_dev) +{ + struct qede_dev *qdev = eth_dev->data->dev_private; + struct qede_fastpath *fp; + uint8_t i, rss_id, txq_index, tc; + int rxq = 0, txq = 0; + + for_each_queue(i) { + fp = &qdev->fp_array[i]; + if (fp->type & QEDE_FASTPATH_RX) { + fp->rxq = eth_dev->data->rx_queues[i]; + fp->rxq->queue_id = rxq++; + } + + if (fp->type & QEDE_FASTPATH_TX) { + for (tc = 0; tc < qdev->num_tc; tc++) { + txq_index = tc * QEDE_TSS_COUNT(qdev) + txq; + fp->txqs[tc] = + eth_dev->data->tx_queues[txq_index]; + fp->txqs[tc]->queue_id = txq_index; + if (qdev->dev_info.is_legacy) + fp->txqs[tc]->is_legacy = true; + } + txq++; + } + } +} + int qede_dev_start(struct rte_eth_dev *eth_dev) { struct qede_dev *qdev = eth_dev->data->dev_private; struct ecore_dev *edev = &qdev->edev; struct qed_link_output link_output; + struct qede_fastpath *fp; int rc; - DP_INFO(edev, "port %u\n", eth_dev->data->port_id); + DP_INFO(edev, "Device state is %d\n", qdev->state); - if (qdev->state == QEDE_START) { - DP_INFO(edev, "device already started\n"); + if (qdev->state == QEDE_DEV_START) { + DP_INFO(edev, "Port is already started\n"); return 0; } - if (qdev->state == QEDE_CLOSE) { - rc = qede_alloc_fp_array(qdev); - qede_init_fp(qdev); - rc = qede_alloc_mem_load(qdev); - DP_INFO(edev, "Allocated %d RSS queues on %d TC/s\n", - QEDE_RSS_CNT(qdev), qdev->num_tc); - } else if (qdev->state == QEDE_STOP) { - DP_INFO(edev, "restarting port %u\n", eth_dev->data->port_id); - } else { - DP_INFO(edev, "unknown state port %u\n", - eth_dev->data->port_id); - return -EINVAL; - } + if (qdev->state == QEDE_DEV_CONFIG) + qede_init_fp_queue(eth_dev); rc = qede_start_queues(eth_dev, true); - if (rc) { DP_ERR(edev, "Failed to start queues\n"); /* TBD: free */ return rc; } - DP_INFO(edev, "Start VPORT, RXQ and TXQ succeeded\n"); - + /* Bring-up the link */ qede_dev_set_link_state(eth_dev, true); - /* Query whether link is already-up */ - memset(&link_output, 0, sizeof(link_output)); - qdev->ops->common->get_link(edev, &link_output); - DP_NOTICE(edev, false, "link status: %s\n", - link_output.link_up ? "up" : "down"); + /* Reset ring */ + if (qede_reset_fp_rings(qdev)) + return -ENOMEM; - qdev->state = QEDE_START; + /* Start/resume traffic */ + qdev->ops->fastpath_start(edev); - qede_config_rx_mode(eth_dev); + qdev->state = QEDE_DEV_START; - DP_INFO(edev, "dev_state is QEDE_START\n"); + DP_INFO(edev, "dev_state is QEDE_DEV_START\n"); return 0; } @@ -1234,7 +1408,7 @@ static int qede_stop_queues(struct qede_dev *qdev) vport_update_params.vport_active_flg = 0; vport_update_params.update_rss_flg = 0; - DP_INFO(edev, "vport_update\n"); + DP_INFO(edev, "Deactivate vport\n"); rc = qdev->ops->vport_update(edev, &vport_update_params); if (rc) { @@ -1245,129 +1419,157 @@ static int qede_stop_queues(struct qede_dev *qdev) DP_INFO(edev, "Flushing tx queues\n"); /* Flush Tx queues. If needed, request drain from MCP */ - for_each_rss(i) { + for_each_queue(i) { struct qede_fastpath *fp = &qdev->fp_array[i]; - for (tc = 0; tc < qdev->num_tc; tc++) { - struct qede_tx_queue *txq = fp->txqs[tc]; - rc = qede_drain_txq(qdev, txq, true); - if (rc) - return rc; + + if (fp->type & QEDE_FASTPATH_TX) { + for (tc = 0; tc < qdev->num_tc; tc++) { + struct qede_tx_queue *txq = fp->txqs[tc]; + + rc = qede_drain_txq(qdev, txq, true); + if (rc) + return rc; + } } } /* Stop all Queues in reverse order */ - for (i = QEDE_RSS_CNT(qdev) - 1; i >= 0; i--) { + for (i = QEDE_QUEUE_CNT(qdev) - 1; i >= 0; i--) { struct qed_stop_rxq_params rx_params; /* Stop the Tx Queue(s) */ - for (tc = 0; tc < qdev->num_tc; tc++) { - struct qed_stop_txq_params tx_params; - - tx_params.rss_id = i; - tx_params.tx_queue_id = tc * QEDE_RSS_CNT(qdev) + i; - - DP_INFO(edev, "Stopping tx queues\n"); - rc = qdev->ops->q_tx_stop(edev, &tx_params); - if (rc) { - DP_ERR(edev, "Failed to stop TXQ #%d\n", - tx_params.tx_queue_id); - return rc; + if (qdev->fp_array[i].type & QEDE_FASTPATH_TX) { + for (tc = 0; tc < qdev->num_tc; tc++) { + struct qed_stop_txq_params tx_params; + u8 val; + + tx_params.rss_id = i; + val = qdev->fp_array[i].txqs[tc]->queue_id; + tx_params.tx_queue_id = val; + + DP_INFO(edev, "Stopping tx queues\n"); + rc = qdev->ops->q_tx_stop(edev, &tx_params); + if (rc) { + DP_ERR(edev, "Failed to stop TXQ #%d\n", + tx_params.tx_queue_id); + return rc; + } } } /* Stop the Rx Queue */ - memset(&rx_params, 0, sizeof(rx_params)); - rx_params.rss_id = i; - rx_params.rx_queue_id = i; - rx_params.eq_completion_only = 1; + if (qdev->fp_array[i].type & QEDE_FASTPATH_RX) { + memset(&rx_params, 0, sizeof(rx_params)); + rx_params.rss_id = i; + rx_params.rx_queue_id = qdev->fp_array[i].rxq->queue_id; + rx_params.eq_completion_only = 1; - DP_INFO(edev, "Stopping rx queues\n"); + DP_INFO(edev, "Stopping rx queues\n"); - rc = qdev->ops->q_rx_stop(edev, &rx_params); - if (rc) { - DP_ERR(edev, "Failed to stop RXQ #%d\n", i); - return rc; + rc = qdev->ops->q_rx_stop(edev, &rx_params); + if (rc) { + DP_ERR(edev, "Failed to stop RXQ #%d\n", i); + return rc; + } } } - DP_INFO(edev, "Stopping vports\n"); - - /* Stop the vport */ - rc = qdev->ops->vport_stop(edev, 0); - if (rc) - DP_ERR(edev, "Failed to stop VPORT\n"); - - return rc; + return 0; } -void qede_reset_fp_rings(struct qede_dev *qdev) +int qede_reset_fp_rings(struct qede_dev *qdev) { - uint16_t rss_id; + struct qede_fastpath *fp; + struct qede_tx_queue *txq; uint8_t tc; - - for_each_rss(rss_id) { - DP_INFO(&qdev->edev, "reset fp chain for rss %u\n", rss_id); - struct qede_fastpath *fp = &qdev->fp_array[rss_id]; - ecore_chain_reset(&fp->rxq->rx_bd_ring); - ecore_chain_reset(&fp->rxq->rx_comp_ring); - for (tc = 0; tc < qdev->num_tc; tc++) { - struct qede_tx_queue *txq = fp->txqs[tc]; - ecore_chain_reset(&txq->tx_pbl); + uint16_t id, i; + + for_each_queue(id) { + fp = &qdev->fp_array[id]; + + if (fp->type & QEDE_FASTPATH_RX) { + DP_INFO(&qdev->edev, + "Reset FP chain for RSS %u\n", id); + qede_rx_queue_release_mbufs(fp->rxq); + ecore_chain_reset(&fp->rxq->rx_bd_ring); + ecore_chain_reset(&fp->rxq->rx_comp_ring); + fp->rxq->sw_rx_prod = 0; + fp->rxq->sw_rx_cons = 0; + *fp->rxq->hw_cons_ptr = 0; + for (i = 0; i < fp->rxq->nb_rx_desc; i++) { + if (qede_alloc_rx_buffer(fp->rxq)) { + DP_ERR(&qdev->edev, + "RX buffer allocation failed\n"); + return -ENOMEM; + } + } + } + if (fp->type & QEDE_FASTPATH_TX) { + for (tc = 0; tc < qdev->num_tc; tc++) { + txq = fp->txqs[tc]; + qede_tx_queue_release_mbufs(txq); + ecore_chain_reset(&txq->tx_pbl); + txq->sw_tx_cons = 0; + txq->sw_tx_prod = 0; + *txq->hw_cons_ptr = 0; + } } } + + return 0; } /* This function frees all memory of a single fp */ -static void qede_free_mem_fp(struct qede_dev *qdev, struct qede_fastpath *fp) +void qede_free_mem_load(struct rte_eth_dev *eth_dev) { + struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev); + struct qede_fastpath *fp; + uint16_t txq_idx; + uint8_t id; uint8_t tc; - qede_rx_queue_release(fp->rxq); - for (tc = 0; tc < qdev->num_tc; tc++) - qede_tx_queue_release(fp->txqs[tc]); -} - -void qede_free_mem_load(struct qede_dev *qdev) -{ - uint8_t rss_id; - - for_each_rss(rss_id) { - struct qede_fastpath *fp = &qdev->fp_array[rss_id]; - qede_free_mem_fp(qdev, fp); + for_each_queue(id) { + fp = &qdev->fp_array[id]; + if (fp->type & QEDE_FASTPATH_RX) { + if (!fp->rxq) + continue; + qede_rx_queue_release(fp->rxq); + eth_dev->data->rx_queues[id] = NULL; + } else { + for (tc = 0; tc < qdev->num_tc; tc++) { + if (!fp->txqs[tc]) + continue; + txq_idx = fp->txqs[tc]->queue_id; + qede_tx_queue_release(fp->txqs[tc]); + eth_dev->data->tx_queues[txq_idx] = NULL; + } + } } - /* qdev->num_rss = 0; */ } -/* - * Stop an Ethernet device. The device can be restarted with a call to - * rte_eth_dev_start(). - * Do not change link state and do not release sw structures. - */ void qede_dev_stop(struct rte_eth_dev *eth_dev) { struct qede_dev *qdev = eth_dev->data->dev_private; struct ecore_dev *edev = &qdev->edev; - int rc; DP_INFO(edev, "port %u\n", eth_dev->data->port_id); - if (qdev->state != QEDE_START) { - DP_INFO(edev, "device not yet started\n"); + if (qdev->state != QEDE_DEV_START) { + DP_INFO(edev, "Device not yet started\n"); return; } - rc = qede_stop_queues(qdev); - - if (rc) + if (qede_stop_queues(qdev)) DP_ERR(edev, "Didn't succeed to close queues\n"); DP_INFO(edev, "Stopped queues\n"); qdev->ops->fastpath_stop(edev); - qede_reset_fp_rings(qdev); + /* Bring the link down */ + qede_dev_set_link_state(eth_dev, false); - qdev->state = QEDE_STOP; + qdev->state = QEDE_DEV_STOP; - DP_INFO(edev, "dev_state is QEDE_STOP\n"); + DP_INFO(edev, "dev_state is QEDE_DEV_STOP\n"); }