.. SPDX-License-Identifier: BSD-3-Clause Copyright(c) 2010-2014 Intel Corporation .. _spp_vf_explain_spp_vf: spp_vf ====== This section describes implementation of key features of ``spp_vf``. ``spp_vf`` consists of master thread and several worker threads, ``forwarder``, ``classifier`` or ``merger``, as slaves. For classifying packets, ``spp_vf`` has a worker thread named ``classifier`` and a table for registering MAC address entries. Initialization -------------- In master thread, data of classifier and VLAN features are initialized after ``rte_eal_init()`` is called. Port capability is a set of data for describing VLAN features. Then, each of worker threads are launched with ``rte_eal_remote_launch()`` on assigned lcores.. .. code-block:: c /* spp_vf.c */ ret = rte_eal_init(argc, argv); /* skipping lines ... */ /* Start worker threads of classifier and forwarder */ unsigned int lcore_id = 0; RTE_LCORE_FOREACH_SLAVE(lcore_id) { rte_eal_remote_launch(slave_main, NULL, lcore_id); } Slave Main ---------- Main function of worker thread is defined as ``slave_main()`` which is called from ``rte_eal_remote_launch()``. Behavior of worker thread is decided in while loop in this function. If lcore status is not ``SPPWK_LCORE_RUNNING``, worker thread does nothing. On the other hand, it does packet forwarding with or without classifying. It classifies incoming packets if component type is ``SPPWK_TYPE_CLS``, or simply forwards packets. .. code-block:: c /* spp_vf.c */ while ((status = spp_get_core_status(lcore_id)) != SPPWK_LCORE_REQ_STOP) { if (status != SPPWK_LCORE_RUNNING) continue; /* skipping lines ... */ /* It is for processing multiple components. */ for (cnt = 0; cnt < core->num; cnt++) { /* Component classification to call a function. */ if (spp_get_component_type(core->id[cnt]) == SPPWK_TYPE_CLS) { /* Component type for classifier. */ ret = classify_packets(core->id[cnt]); if (unlikely(ret != 0)) break; } else { /* Component type for forward or merge. */ ret = forward_packets(core->id[cnt]); if (unlikely(ret != 0)) break; } } Data structure of classifier ---------------------------- Classifier has a set of attributes for classification as struct ``mac_classifier``, which consists of a table of MAC addresses, number of classifying ports, indices of ports and default index of port. Clasifier table is implemented as hash of struct ``rte_hash``. .. code-block:: c /* shared/secondary/spp_worker_th/vf_deps.h */ /* Classifier for MAC addresses. */ struct mac_classifier { struct rte_hash *cls_tbl; /* Hash table for MAC classification. */ int nof_cls_ports; /* Num of ports classified validly. */ int cls_ports[RTE_MAX_ETHPORTS]; /* Ports for classification. */ int default_cls_idx; /* Default index for classification. */ }; Classifier itself is defined as a struct ``cls_comp_info``. There are several attributes in this struct including ``mac_classifier`` or ``cls_port_info`` or so. ``cls_port_info`` is for defining a set of attributes of ports, such as interface type, device ID or packet data. .. code-block:: c /* shared/secondary/spp_worker_th/vf_deps.h */ /* classifier component information */ struct cls_comp_info { char name[STR_LEN_NAME]; /* component name */ int mac_addr_entry; /* mac address entry flag */ struct mac_classifier *mac_clfs[NOF_VLAN]; /* classifiers per VLAN. */ int nof_tx_ports; /* Number of TX ports info entries. */ /* Classifier has one RX port and several TX ports. */ struct cls_port_info rx_port_i; /* RX port info classified. */ struct cls_port_info tx_ports_i[RTE_MAX_ETHPORTS]; /* TX info. */ }; /* Attirbutes of port for classification. */ struct cls_port_info { enum port_type iface_type; int iface_no; /* Index of ports handled by classifier. */ int iface_no_global; /* ID for interface generated by spp_vf */ uint16_t ethdev_port_id; /* Ethdev port ID. */ uint16_t nof_pkts; /* Number of packets in pkts[]. */ struct rte_mbuf *pkts[MAX_PKT_BURST]; /* packets to be classified. */ }; Classifying the packet ---------------------- If component type is ``SPPWK_TYPE_CLS``, worker thread behaves as a classifier, so component calls ``classify_packets()``. In this function, packets from RX port are received with ``sppwk_eth_vlan_rx_burst()`` which is derived from ``rte_eth_rx_burst()`` for adding or deleting VLAN tags. Received packets are classified with ``classify_packet()``. .. code-block:: c /* classifier.c */ n_rx = sppwk_eth_vlan_rx_burst(clsd_data_rx->ethdev_port_id, 0, rx_pkts, MAX_PKT_BURST); /* skipping lines ... */ classify_packet(rx_pkts, n_rx, cmp_info, clsd_data_tx); Packet processing in forwarder and merger ----------------------------------------- Configuration data for forwarder and merger is stored as structured tables ``forward_rxtx``, ``forward_path`` and ``forward_info``. The ``forward_rxtx`` has two member variables for expressing the port to be sent(tx) and to be receive(rx), ``forward_path`` has member variables for expressing the data path. Like as ``mac_classifier``, ``forward_info`` has two tables, one is for updating by commands, the other is for looking up to process packets. .. code-block:: c /* forwarder.c */ /* A set of port info of rx and tx */ struct forward_rxtx { struct spp_port_info rx; /* rx port */ struct spp_port_info tx; /* tx port */ }; /* Information on the path used for forward. */ struct forward_path { char name[STR_LEN_NAME]; /* Component name */ volatile enum sppwk_worker_type wk_type; int nof_rx; /* Number of RX ports */ int nof_tx; /* Number of TX ports */ struct forward_rxtx ports[RTE_MAX_ETHPORTS]; /* Set of RX and TX */ }; /* Information for forward. */ struct forward_info { volatile int ref_index; /* index to reference area */ volatile int upd_index; /* index to update area */ struct forward_path path[SPP_INFO_AREA_MAX]; /* Information of data path */ }; L2 Multicast Support -------------------- ``spp_vf`` supports multicast for resolving ARP requests. It is implemented as ``handle_l2multicast_packet()`` and called from ``classify_packet()`` for incoming multicast packets. .. code-block:: c /* classify_packet() in classifier.c */ /* L2 multicast(include broadcast) ? */ if (unlikely(is_multicast_ether_addr(ð->d_addr))) { RTE_LOG(DEBUG, SPP_CLASSIFIER_MAC, "multicast mac address.\n"); handle_l2multicast_packet(rx_pkts[i], classifier_info, classified_data); continue; } Packets are cloned with ``rte_mbuf_refcnt_update()`` for distributing multicast packets. .. code-block:: c /* classifier.c */ handle_l2multicast_packet(struct rte_mbuf *pkt, struct cls_comp_info *cmp_info, struct cls_port_info *clsd_data) { int i; struct mac_classifier *mac_cls; uint16_t vid = get_vid(pkt); int gen_def_clsd_idx = get_general_default_classified_index(cmp_info); int n_act_clsd; /* skipping lines... */ rte_mbuf_refcnt_update(pkt, (int16_t)(n_act_clsd - 1)); Two phase update for forwarding ------------------------------- Update of netowrk configuration in ``spp_vf`` is done in a short period of time, but not so short considering the time scale of packet forwarding. It might forward packets before the updating is completed possibly. To avoid such kind of situation, ``spp_vf`` has two phase update mechanism. Status info is referred from forwarding process after the update is completed. .. code-block:: c int flush_cmd(void) { int ret; int *p_change_comp; struct sppwk_comp_info *p_comp_info; struct cancel_backup_info *backup_info; sppwk_get_mng_data(NULL, &p_comp_info, NULL, NULL, &p_change_comp, &backup_info); ret = update_port_info(); if (ret < SPPWK_RET_OK) return ret; update_lcore_info(); ret = update_comp_info(p_comp_info, p_change_comp); backup_mng_info(backup_info); return ret; }