net/qede: enhance Rx CPU utilization
[dpdk.git] / drivers / net / qede / qede_rxtx.c
index a28dd0a..c92c442 100644 (file)
@@ -1518,6 +1518,228 @@ print_rx_bd_info(struct rte_mbuf *m, struct qede_rx_queue *rxq,
 }
 #endif
 
+uint16_t
+qede_recv_pkts_regular(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
+{
+       struct eth_fast_path_rx_reg_cqe *fp_cqe = NULL;
+       register struct rte_mbuf *rx_mb = NULL;
+       struct qede_rx_queue *rxq = p_rxq;
+       struct qede_dev *qdev = rxq->qdev;
+       struct ecore_dev *edev = &qdev->edev;
+       union eth_rx_cqe *cqe;
+       uint64_t ol_flags;
+       enum eth_rx_cqe_type cqe_type;
+       int rss_enable = qdev->rss_enable;
+       int rx_alloc_count = 0;
+       uint32_t packet_type;
+       uint32_t rss_hash;
+       uint16_t vlan_tci, port_id;
+       uint16_t hw_comp_cons, sw_comp_cons, sw_rx_index, num_rx_bds;
+       uint16_t rx_pkt = 0;
+       uint16_t pkt_len = 0;
+       uint16_t len; /* Length of first BD */
+       uint16_t preload_idx;
+       uint16_t parse_flag;
+#ifdef RTE_LIBRTE_QEDE_DEBUG_RX
+       uint8_t bitfield_val;
+#endif
+       uint8_t offset, flags, bd_num;
+
+
+       /* Allocate buffers that we used in previous loop */
+       if (rxq->rx_alloc_count) {
+               if (unlikely(qede_alloc_rx_bulk_mbufs(rxq,
+                            rxq->rx_alloc_count))) {
+                       struct rte_eth_dev *dev;
+
+                       PMD_RX_LOG(ERR, rxq,
+                                  "New buffer allocation failed,"
+                                  "dropping incoming packetn");
+                       dev = &rte_eth_devices[rxq->port_id];
+                       dev->data->rx_mbuf_alloc_failed +=
+                                                       rxq->rx_alloc_count;
+                       rxq->rx_alloc_errors += rxq->rx_alloc_count;
+                       return 0;
+               }
+               qede_update_rx_prod(qdev, rxq);
+               rxq->rx_alloc_count = 0;
+       }
+
+       hw_comp_cons = rte_le_to_cpu_16(*rxq->hw_cons_ptr);
+       sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
+
+       rte_rmb();
+
+       if (hw_comp_cons == sw_comp_cons)
+               return 0;
+
+       num_rx_bds =  NUM_RX_BDS(rxq);
+       port_id = rxq->port_id;
+
+       while (sw_comp_cons != hw_comp_cons) {
+               ol_flags = 0;
+               packet_type = RTE_PTYPE_UNKNOWN;
+               vlan_tci = 0;
+               rss_hash = 0;
+
+               /* Get the CQE from the completion ring */
+               cqe =
+                   (union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring);
+               cqe_type = cqe->fast_path_regular.type;
+               PMD_RX_LOG(INFO, rxq, "Rx CQE type %d\n", cqe_type);
+
+               if (likely(cqe_type == ETH_RX_CQE_TYPE_REGULAR)) {
+                       fp_cqe = &cqe->fast_path_regular;
+               } else {
+                       if (cqe_type == ETH_RX_CQE_TYPE_SLOW_PATH) {
+                               PMD_RX_LOG(INFO, rxq, "Got unexpected slowpath CQE\n");
+                               ecore_eth_cqe_completion
+                                       (&edev->hwfns[rxq->queue_id %
+                                                     edev->num_hwfns],
+                                        (struct eth_slow_path_rx_cqe *)cqe);
+                       }
+                       goto next_cqe;
+               }
+
+               /* Get the data from the SW ring */
+               sw_rx_index = rxq->sw_rx_cons & num_rx_bds;
+               rx_mb = rxq->sw_rx_ring[sw_rx_index].mbuf;
+               assert(rx_mb != NULL);
+
+               parse_flag = rte_le_to_cpu_16(fp_cqe->pars_flags.flags);
+               offset = fp_cqe->placement_offset;
+               len = rte_le_to_cpu_16(fp_cqe->len_on_first_bd);
+               pkt_len = rte_le_to_cpu_16(fp_cqe->pkt_len);
+               vlan_tci = rte_le_to_cpu_16(fp_cqe->vlan_tag);
+               rss_hash = rte_le_to_cpu_32(fp_cqe->rss_hash);
+               bd_num = fp_cqe->bd_num;
+#ifdef RTE_LIBRTE_QEDE_DEBUG_RX
+               bitfield_val = fp_cqe->bitfields;
+#endif
+
+               if (unlikely(qede_tunn_exist(parse_flag))) {
+                       PMD_RX_LOG(INFO, rxq, "Rx tunneled packet\n");
+                       if (unlikely(qede_check_tunn_csum_l4(parse_flag))) {
+                               PMD_RX_LOG(ERR, rxq,
+                                           "L4 csum failed, flags = 0x%x\n",
+                                           parse_flag);
+                               rxq->rx_hw_errors++;
+                               ol_flags |= PKT_RX_L4_CKSUM_BAD;
+                       } else {
+                               ol_flags |= PKT_RX_L4_CKSUM_GOOD;
+                       }
+
+                       if (unlikely(qede_check_tunn_csum_l3(parse_flag))) {
+                               PMD_RX_LOG(ERR, rxq,
+                                       "Outer L3 csum failed, flags = 0x%x\n",
+                                       parse_flag);
+                               rxq->rx_hw_errors++;
+                               ol_flags |= PKT_RX_EIP_CKSUM_BAD;
+                       } else {
+                               ol_flags |= PKT_RX_IP_CKSUM_GOOD;
+                       }
+
+                       flags = fp_cqe->tunnel_pars_flags.flags;
+
+                       /* Tunnel_type */
+                       packet_type =
+                               qede_rx_cqe_to_tunn_pkt_type(flags);
+
+                       /* Inner header */
+                       packet_type |=
+                             qede_rx_cqe_to_pkt_type_inner(parse_flag);
+
+                       /* Outer L3/L4 types is not available in CQE */
+                       packet_type |= qede_rx_cqe_to_pkt_type_outer(rx_mb);
+
+                       /* Outer L3/L4 types is not available in CQE.
+                        * Need to add offset to parse correctly,
+                        */
+                       rx_mb->data_off = offset + RTE_PKTMBUF_HEADROOM;
+                       packet_type |= qede_rx_cqe_to_pkt_type_outer(rx_mb);
+               } else {
+                       packet_type |= qede_rx_cqe_to_pkt_type(parse_flag);
+               }
+
+               /* Common handling for non-tunnel packets and for inner
+                * headers in the case of tunnel.
+                */
+               if (unlikely(qede_check_notunn_csum_l4(parse_flag))) {
+                       PMD_RX_LOG(ERR, rxq,
+                                   "L4 csum failed, flags = 0x%x\n",
+                                   parse_flag);
+                       rxq->rx_hw_errors++;
+                       ol_flags |= PKT_RX_L4_CKSUM_BAD;
+               } else {
+                       ol_flags |= PKT_RX_L4_CKSUM_GOOD;
+               }
+               if (unlikely(qede_check_notunn_csum_l3(rx_mb, parse_flag))) {
+                       PMD_RX_LOG(ERR, rxq, "IP csum failed, flags = 0x%x\n",
+                                  parse_flag);
+                       rxq->rx_hw_errors++;
+                       ol_flags |= PKT_RX_IP_CKSUM_BAD;
+               } else {
+                       ol_flags |= PKT_RX_IP_CKSUM_GOOD;
+               }
+
+               if (unlikely(CQE_HAS_VLAN(parse_flag) ||
+                            CQE_HAS_OUTER_VLAN(parse_flag))) {
+                       /* Note: FW doesn't indicate Q-in-Q packet */
+                       ol_flags |= PKT_RX_VLAN;
+                       if (qdev->vlan_strip_flg) {
+                               ol_flags |= PKT_RX_VLAN_STRIPPED;
+                               rx_mb->vlan_tci = vlan_tci;
+                       }
+               }
+
+               if (rss_enable) {
+                       ol_flags |= PKT_RX_RSS_HASH;
+                       rx_mb->hash.rss = rss_hash;
+               }
+
+               rx_alloc_count++;
+               qede_rx_bd_ring_consume(rxq);
+
+               /* Prefetch next mbuf while processing current one. */
+               preload_idx = rxq->sw_rx_cons & num_rx_bds;
+               rte_prefetch0(rxq->sw_rx_ring[preload_idx].mbuf);
+
+               /* Update rest of the MBUF fields */
+               rx_mb->data_off = offset + RTE_PKTMBUF_HEADROOM;
+               rx_mb->port = port_id;
+               rx_mb->ol_flags = ol_flags;
+               rx_mb->data_len = len;
+               rx_mb->packet_type = packet_type;
+#ifdef RTE_LIBRTE_QEDE_DEBUG_RX
+               print_rx_bd_info(rx_mb, rxq, bitfield_val);
+#endif
+               rx_mb->nb_segs = bd_num;
+               rx_mb->pkt_len = pkt_len;
+
+               rx_pkts[rx_pkt] = rx_mb;
+               rx_pkt++;
+
+next_cqe:
+               ecore_chain_recycle_consumed(&rxq->rx_comp_ring);
+               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",
+                                  rx_pkt, nb_pkts);
+                       break;
+               }
+       }
+
+       /* Request number of bufferes to be allocated in next loop */
+       rxq->rx_alloc_count = rx_alloc_count;
+
+       rxq->rcv_pkts += rx_pkt;
+       rxq->rx_segs += rx_pkt;
+       PMD_RX_LOG(DEBUG, rxq, "rx_pkts=%u core=%d", rx_pkt, rte_lcore_id());
+
+       return rx_pkt;
+}
+
 uint16_t
 qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
 {