doc: add GRE option flow item to feature list
[dpdk.git] / examples / l3fwd-graph / main.c
index 7ea7885..7f00c65 100644 (file)
 #include <rte_cycles.h>
 #include <rte_eal.h>
 #include <rte_ethdev.h>
+#include <rte_graph_worker.h>
+#include <rte_launch.h>
 #include <rte_lcore.h>
 #include <rte_log.h>
 #include <rte_mempool.h>
+#include <rte_node_eth_api.h>
+#include <rte_node_ip4_api.h>
 #include <rte_per_lcore.h>
 #include <rte_string_fns.h>
 #include <rte_vect.h>
@@ -75,12 +79,17 @@ static uint32_t enabled_port_mask;
 struct lcore_rx_queue {
        uint16_t port_id;
        uint8_t queue_id;
+       char node_name[RTE_NODE_NAMESIZE];
 };
 
 /* Lcore conf */
 struct lcore_conf {
        uint16_t n_rx_queue;
        struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
+
+       struct rte_graph *graph;
+       char name[RTE_GRAPH_NAMESIZE];
+       rte_graph_t graph_id;
 } __rte_cache_aligned;
 
 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
@@ -102,23 +111,43 @@ static uint16_t nb_lcore_params = RTE_DIM(lcore_params_array_default);
 
 static struct rte_eth_conf port_conf = {
        .rxmode = {
-               .mq_mode = ETH_MQ_RX_RSS,
-               .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
+               .mq_mode = RTE_ETH_MQ_RX_RSS,
                .split_hdr_size = 0,
        },
        .rx_adv_conf = {
                .rss_conf = {
                                .rss_key = NULL,
-                               .rss_hf = ETH_RSS_IP,
+                               .rss_hf = RTE_ETH_RSS_IP,
                },
        },
        .txmode = {
-               .mq_mode = ETH_MQ_TX_NONE,
+               .mq_mode = RTE_ETH_MQ_TX_NONE,
        },
 };
 
+static uint32_t max_pkt_len;
+
 static struct rte_mempool *pktmbuf_pool[RTE_MAX_ETHPORTS][NB_SOCKETS];
 
+static struct rte_node_ethdev_config ethdev_conf[RTE_MAX_ETHPORTS];
+
+struct ipv4_l3fwd_lpm_route {
+       uint32_t ip;
+       uint8_t depth;
+       uint8_t if_out;
+};
+
+#define IPV4_L3FWD_LPM_NUM_ROUTES                                              \
+       (sizeof(ipv4_l3fwd_lpm_route_array) /                                  \
+        sizeof(ipv4_l3fwd_lpm_route_array[0]))
+/* 198.18.0.0/16 are set aside for RFC2544 benchmarking. */
+static struct ipv4_l3fwd_lpm_route ipv4_l3fwd_lpm_route_array[] = {
+       {RTE_IPV4(198, 18, 0, 0), 24, 0}, {RTE_IPV4(198, 18, 1, 0), 24, 1},
+       {RTE_IPV4(198, 18, 2, 0), 24, 2}, {RTE_IPV4(198, 18, 3, 0), 24, 3},
+       {RTE_IPV4(198, 18, 4, 0), 24, 4}, {RTE_IPV4(198, 18, 5, 0), 24, 5},
+       {RTE_IPV4(198, 18, 6, 0), 24, 6}, {RTE_IPV4(198, 18, 7, 0), 24, 7},
+};
+
 static int
 check_lcore_params(void)
 {
@@ -139,8 +168,8 @@ check_lcore_params(void)
                        return -1;
                }
 
-               if (lcore == rte_get_master_lcore()) {
-                       printf("Error: lcore %u is master lcore\n", lcore);
+               if (lcore == rte_get_main_lcore()) {
+                       printf("Error: lcore %u is main lcore\n", lcore);
                        return -1;
                }
                socketid = rte_lcore_to_socket_id(lcore);
@@ -231,7 +260,7 @@ print_usage(const char *prgname)
                " [-P]"
                " --config (port,queue,lcore)[,(port,queue,lcore)]"
                " [--eth-dest=X,MM:MM:MM:MM:MM:MM]"
-               " [--enable-jumbo [--max-pkt-len PKTLEN]]"
+               " [--max-pkt-len PKTLEN]"
                " [--no-numa]"
                " [--per-port-pool]\n\n"
 
@@ -240,9 +269,7 @@ print_usage(const char *prgname)
                "  --config (port,queue,lcore): Rx queue configuration\n"
                "  --eth-dest=X,MM:MM:MM:MM:MM:MM: Ethernet destination for "
                "port X\n"
-               "  --enable-jumbo: Enable jumbo frames\n"
-               "  --max-pkt-len: Under the premise of enabling jumbo,\n"
-               "                 maximum packet length in decimal (64-9600)\n"
+               "  --max-pkt-len PKTLEN: maximum packet length in decimal (64-9600)\n"
                "  --no-numa: Disable numa awareness\n"
                "  --per-port-pool: Use separate buffer pool per port\n\n",
                prgname);
@@ -274,10 +301,7 @@ parse_portmask(const char *portmask)
        /* Parse hexadecimal string */
        pm = strtoul(portmask, &end, 16);
        if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
-               return -1;
-
-       if (pm == 0)
-               return -1;
+               return 0;
 
        return pm;
 }
@@ -379,7 +403,7 @@ static const char short_options[] = "p:" /* portmask */
 #define CMD_LINE_OPT_CONFIG       "config"
 #define CMD_LINE_OPT_ETH_DEST     "eth-dest"
 #define CMD_LINE_OPT_NO_NUMA      "no-numa"
-#define CMD_LINE_OPT_ENABLE_JUMBO  "enable-jumbo"
+#define CMD_LINE_OPT_MAX_PKT_LEN   "max-pkt-len"
 #define CMD_LINE_OPT_PER_PORT_POOL "per-port-pool"
 enum {
        /* Long options mapped to a short option */
@@ -391,7 +415,7 @@ enum {
        CMD_LINE_OPT_CONFIG_NUM,
        CMD_LINE_OPT_ETH_DEST_NUM,
        CMD_LINE_OPT_NO_NUMA_NUM,
-       CMD_LINE_OPT_ENABLE_JUMBO_NUM,
+       CMD_LINE_OPT_MAX_PKT_LEN_NUM,
        CMD_LINE_OPT_PARSE_PER_PORT_POOL,
 };
 
@@ -399,7 +423,7 @@ static const struct option lgopts[] = {
        {CMD_LINE_OPT_CONFIG, 1, 0, CMD_LINE_OPT_CONFIG_NUM},
        {CMD_LINE_OPT_ETH_DEST, 1, 0, CMD_LINE_OPT_ETH_DEST_NUM},
        {CMD_LINE_OPT_NO_NUMA, 0, 0, CMD_LINE_OPT_NO_NUMA_NUM},
-       {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, CMD_LINE_OPT_ENABLE_JUMBO_NUM},
+       {CMD_LINE_OPT_MAX_PKT_LEN, 1, 0, CMD_LINE_OPT_MAX_PKT_LEN_NUM},
        {CMD_LINE_OPT_PER_PORT_POOL, 0, 0, CMD_LINE_OPT_PARSE_PER_PORT_POOL},
        {NULL, 0, 0, 0},
 };
@@ -465,28 +489,8 @@ parse_args(int argc, char **argv)
                        numa_on = 0;
                        break;
 
-               case CMD_LINE_OPT_ENABLE_JUMBO_NUM: {
-                       const struct option lenopts = {"max-pkt-len",
-                                                      required_argument, 0, 0};
-
-                       port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
-                       port_conf.txmode.offloads |= DEV_TX_OFFLOAD_MULTI_SEGS;
-
-                       /*
-                        * if no max-pkt-len set, use the default
-                        * value RTE_ETHER_MAX_LEN.
-                        */
-                       if (getopt_long(argc, argvopt, "", &lenopts,
-                                       &option_index) == 0) {
-                               ret = parse_max_pkt_len(optarg);
-                               if (ret < 64 || ret > MAX_JUMBO_PKT_LEN) {
-                                       fprintf(stderr, "Invalid maximum "
-                                                       "packet length\n");
-                                       print_usage(prgname);
-                                       return -1;
-                               }
-                               port_conf.rxmode.max_rx_pkt_len = ret;
-                       }
+               case CMD_LINE_OPT_MAX_PKT_LEN_NUM: {
+                       max_pkt_len = parse_max_pkt_len(optarg);
                        break;
                }
 
@@ -570,6 +574,8 @@ check_all_ports_link_status(uint32_t port_mask)
        uint8_t count, all_ports_up, print_flag = 0;
        struct rte_eth_link link;
        uint16_t portid;
+       int ret;
+       char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
 
        printf("\nChecking link status");
        fflush(stdout);
@@ -584,23 +590,24 @@ check_all_ports_link_status(uint32_t port_mask)
                        if ((port_mask & (1 << portid)) == 0)
                                continue;
                        memset(&link, 0, sizeof(link));
-                       rte_eth_link_get_nowait(portid, &link);
+                       ret = rte_eth_link_get_nowait(portid, &link);
+                       if (ret < 0) {
+                               all_ports_up = 0;
+                               if (print_flag == 1)
+                                       printf("Port %u link get failed: %s\n",
+                                               portid, rte_strerror(-ret));
+                               continue;
+                       }
                        /* Print link status if flag set */
                        if (print_flag == 1) {
-                               if (link.link_status)
-                                       printf("Port%d Link Up. Speed %u Mbps "
-                                              "-%s\n",
-                                              portid, link.link_speed,
-                                              (link.link_duplex ==
-                                               ETH_LINK_FULL_DUPLEX)
-                                                      ? ("full-duplex")
-                                                      : ("half-duplex\n"));
-                               else
-                                       printf("Port %d Link Down\n", portid);
+                               rte_eth_link_to_str(link_status_text,
+                                       sizeof(link_status_text), &link);
+                               printf("Port %d %s\n", portid,
+                                      link_status_text);
                                continue;
                        }
                        /* Clear all_ports_up flag if any link down */
-                       if (link.link_status == ETH_LINK_DOWN) {
+                       if (link.link_status == RTE_ETH_LINK_DOWN) {
                                all_ports_up = 0;
                                break;
                        }
@@ -633,17 +640,126 @@ signal_handler(int signum)
        }
 }
 
+static void
+print_stats(void)
+{
+       const char topLeft[] = {27, '[', '1', ';', '1', 'H', '\0'};
+       const char clr[] = {27, '[', '2', 'J', '\0'};
+       struct rte_graph_cluster_stats_param s_param;
+       struct rte_graph_cluster_stats *stats;
+       const char *pattern = "worker_*";
+
+       /* Prepare stats object */
+       memset(&s_param, 0, sizeof(s_param));
+       s_param.f = stdout;
+       s_param.socket_id = SOCKET_ID_ANY;
+       s_param.graph_patterns = &pattern;
+       s_param.nb_graph_patterns = 1;
+
+       stats = rte_graph_cluster_stats_create(&s_param);
+       if (stats == NULL)
+               rte_exit(EXIT_FAILURE, "Unable to create stats object\n");
+
+       while (!force_quit) {
+               /* Clear screen and move to top left */
+               printf("%s%s", clr, topLeft);
+               rte_graph_cluster_stats_get(stats, 0);
+               rte_delay_ms(1E3);
+       }
+
+       rte_graph_cluster_stats_destroy(stats);
+}
+
+/* Main processing loop. 8< */
+static int
+graph_main_loop(void *conf)
+{
+       struct lcore_conf *qconf;
+       struct rte_graph *graph;
+       uint32_t lcore_id;
+
+       RTE_SET_USED(conf);
+
+       lcore_id = rte_lcore_id();
+       qconf = &lcore_conf[lcore_id];
+       graph = qconf->graph;
+
+       if (!graph) {
+               RTE_LOG(INFO, L3FWD_GRAPH, "Lcore %u has nothing to do\n",
+                       lcore_id);
+               return 0;
+       }
+
+       RTE_LOG(INFO, L3FWD_GRAPH,
+               "Entering main loop on lcore %u, graph %s(%p)\n", lcore_id,
+               qconf->name, graph);
+
+       while (likely(!force_quit))
+               rte_graph_walk(graph);
+
+       return 0;
+}
+/* >8 End of main processing loop. */
+
+static uint32_t
+eth_dev_get_overhead_len(uint32_t max_rx_pktlen, uint16_t max_mtu)
+{
+       uint32_t overhead_len;
+
+       if (max_mtu != UINT16_MAX && max_rx_pktlen > max_mtu)
+               overhead_len = max_rx_pktlen - max_mtu;
+       else
+               overhead_len = RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN;
+
+       return overhead_len;
+}
+
+static int
+config_port_max_pkt_len(struct rte_eth_conf *conf,
+               struct rte_eth_dev_info *dev_info)
+{
+       uint32_t overhead_len;
+
+       if (max_pkt_len == 0)
+               return 0;
+
+       if (max_pkt_len < RTE_ETHER_MIN_LEN || max_pkt_len > MAX_JUMBO_PKT_LEN)
+               return -1;
+
+       overhead_len = eth_dev_get_overhead_len(dev_info->max_rx_pktlen,
+                       dev_info->max_mtu);
+       conf->rxmode.mtu = max_pkt_len - overhead_len;
+
+       if (conf->rxmode.mtu > RTE_ETHER_MTU)
+               conf->txmode.offloads |= RTE_ETH_TX_OFFLOAD_MULTI_SEGS;
+
+       return 0;
+}
+
 int
 main(int argc, char **argv)
 {
+       /* Rewrite data of src and dst ether addr */
+       uint8_t rewrite_data[2 * sizeof(struct rte_ether_addr)];
+       /* Graph initialization. 8< */
+       static const char * const default_patterns[] = {
+               "ip4*",
+               "ethdev_tx-*",
+               "pkt_drop",
+       };
        uint8_t nb_rx_queue, queue, socketid;
+       struct rte_graph_param graph_conf;
        struct rte_eth_dev_info dev_info;
+       uint32_t nb_ports, nb_conf = 0;
        uint32_t n_tx_queue, nb_lcores;
        struct rte_eth_txconf *txconf;
-       uint16_t queueid, portid;
+       uint16_t queueid, portid, i;
+       const char **node_patterns;
        struct lcore_conf *qconf;
+       uint16_t nb_graphs = 0;
+       uint16_t nb_patterns;
+       uint8_t rewrite_len;
        uint32_t lcore_id;
-       uint32_t nb_ports;
        int ret;
 
        /* Init EAL */
@@ -682,7 +798,7 @@ main(int argc, char **argv)
        nb_ports = rte_eth_dev_count_avail();
        nb_lcores = rte_lcore_count();
 
-       /* Initialize all ports */
+       /* Initialize all ports. 8< */
        RTE_ETH_FOREACH_DEV(portid)
        {
                struct rte_eth_conf local_port_conf = port_conf;
@@ -705,9 +821,16 @@ main(int argc, char **argv)
                       nb_rx_queue, n_tx_queue);
 
                rte_eth_dev_info_get(portid, &dev_info);
-               if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
+
+               ret = config_port_max_pkt_len(&local_port_conf, &dev_info);
+               if (ret != 0)
+                       rte_exit(EXIT_FAILURE,
+                               "Invalid max packet length: %u (port %u)\n",
+                               max_pkt_len, portid);
+
+               if (dev_info.tx_offload_capa & RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE)
                        local_port_conf.txmode.offloads |=
-                               DEV_TX_OFFLOAD_MBUF_FAST_FREE;
+                               RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE;
 
                local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
                        dev_info.flow_type_rss_offloads;
@@ -792,6 +915,18 @@ main(int argc, char **argv)
                        queueid++;
                }
 
+               /* Setup ethdev node config */
+               ethdev_conf[nb_conf].port_id = portid;
+               ethdev_conf[nb_conf].num_rx_queues = nb_rx_queue;
+               ethdev_conf[nb_conf].num_tx_queues = n_tx_queue;
+               if (!per_port_pool)
+                       ethdev_conf[nb_conf].mp = pktmbuf_pool[0];
+
+               else
+                       ethdev_conf[nb_conf].mp = pktmbuf_pool[portid];
+               ethdev_conf[nb_conf].mp_count = NB_SOCKETS;
+
+               nb_conf++;
                printf("\n");
        }
 
@@ -835,11 +970,25 @@ main(int argc, char **argv)
                                         "port=%d\n",
                                         ret, portid);
 
+                       /* Add this queue node to its graph */
+                       snprintf(qconf->rx_queue_list[queue].node_name,
+                                RTE_NODE_NAMESIZE, "ethdev_rx-%u-%u", portid,
+                                queueid);
                }
+
+               /* Alloc a graph to this lcore only if source exists  */
+               if (qconf->n_rx_queue)
+                       nb_graphs++;
        }
 
        printf("\n");
 
+       /* Ethdev node config, skip rx queue mapping */
+       ret = rte_node_eth_config(ethdev_conf, nb_conf, nb_graphs);
+       /* >8 End of graph creation. */
+       if (ret)
+               rte_exit(EXIT_FAILURE, "rte_node_eth_config: err=%d\n", ret);
+
        /* Start ports */
        RTE_ETH_FOREACH_DEV(portid)
        {
@@ -867,15 +1016,142 @@ main(int argc, char **argv)
 
        check_all_ports_link_status(enabled_port_mask);
 
+       /* Graph Initialization */
+       nb_patterns = RTE_DIM(default_patterns);
+       node_patterns = malloc((MAX_RX_QUEUE_PER_LCORE + nb_patterns) *
+                              sizeof(*node_patterns));
+       if (!node_patterns)
+               return -ENOMEM;
+       memcpy(node_patterns, default_patterns,
+              nb_patterns * sizeof(*node_patterns));
+
+       memset(&graph_conf, 0, sizeof(graph_conf));
+       graph_conf.node_patterns = node_patterns;
+
+       for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
+               rte_graph_t graph_id;
+               rte_edge_t i;
+
+               if (rte_lcore_is_enabled(lcore_id) == 0)
+                       continue;
+
+               qconf = &lcore_conf[lcore_id];
+
+               /* Skip graph creation if no source exists */
+               if (!qconf->n_rx_queue)
+                       continue;
+
+               /* Add rx node patterns of this lcore */
+               for (i = 0; i < qconf->n_rx_queue; i++) {
+                       graph_conf.node_patterns[nb_patterns + i] =
+                               qconf->rx_queue_list[i].node_name;
+               }
+
+               graph_conf.nb_node_patterns = nb_patterns + i;
+               graph_conf.socket_id = rte_lcore_to_socket_id(lcore_id);
+
+               snprintf(qconf->name, sizeof(qconf->name), "worker_%u",
+                        lcore_id);
+
+               graph_id = rte_graph_create(qconf->name, &graph_conf);
+               if (graph_id == RTE_GRAPH_ID_INVALID)
+                       rte_exit(EXIT_FAILURE,
+                                "rte_graph_create(): graph_id invalid"
+                                " for lcore %u\n", lcore_id);
+
+               qconf->graph_id = graph_id;
+               qconf->graph = rte_graph_lookup(qconf->name);
+               /* >8 End of graph initialization. */
+               if (!qconf->graph)
+                       rte_exit(EXIT_FAILURE,
+                                "rte_graph_lookup(): graph %s not found\n",
+                                qconf->name);
+       }
+
+       memset(&rewrite_data, 0, sizeof(rewrite_data));
+       rewrite_len = sizeof(rewrite_data);
+
+       /* Add route to ip4 graph infra. 8< */
+       for (i = 0; i < IPV4_L3FWD_LPM_NUM_ROUTES; i++) {
+               char route_str[INET6_ADDRSTRLEN * 4];
+               char abuf[INET6_ADDRSTRLEN];
+               struct in_addr in;
+               uint32_t dst_port;
+
+               /* Skip unused ports */
+               if ((1 << ipv4_l3fwd_lpm_route_array[i].if_out &
+                    enabled_port_mask) == 0)
+                       continue;
+
+               dst_port = ipv4_l3fwd_lpm_route_array[i].if_out;
+
+               in.s_addr = htonl(ipv4_l3fwd_lpm_route_array[i].ip);
+               snprintf(route_str, sizeof(route_str), "%s / %d (%d)",
+                        inet_ntop(AF_INET, &in, abuf, sizeof(abuf)),
+                        ipv4_l3fwd_lpm_route_array[i].depth,
+                        ipv4_l3fwd_lpm_route_array[i].if_out);
+
+               /* Use route index 'i' as next hop id */
+               ret = rte_node_ip4_route_add(
+                       ipv4_l3fwd_lpm_route_array[i].ip,
+                       ipv4_l3fwd_lpm_route_array[i].depth, i,
+                       RTE_NODE_IP4_LOOKUP_NEXT_REWRITE);
+
+               if (ret < 0)
+                       rte_exit(EXIT_FAILURE,
+                                "Unable to add ip4 route %s to graph\n",
+                                route_str);
+
+               memcpy(rewrite_data, val_eth + dst_port, rewrite_len);
+
+               /* Add next hop rewrite data for id 'i' */
+               ret = rte_node_ip4_rewrite_add(i, rewrite_data,
+                                              rewrite_len, dst_port);
+               if (ret < 0)
+                       rte_exit(EXIT_FAILURE,
+                                "Unable to add next hop %u for "
+                                "route %s\n", i, route_str);
+
+               RTE_LOG(INFO, L3FWD_GRAPH, "Added route %s, next_hop %u\n",
+                       route_str, i);
+       }
+       /* >8 End of adding route to ip4 graph infa. */
+
+       /* Launch per-lcore init on every worker lcore */
+       rte_eal_mp_remote_launch(graph_main_loop, NULL, SKIP_MAIN);
+
+       /* Accumulate and print stats on main until exit */
+       if (rte_graph_has_stats_feature())
+               print_stats();
+
+       /* Wait for worker cores to exit */
+       ret = 0;
+       RTE_LCORE_FOREACH_WORKER(lcore_id) {
+               ret = rte_eal_wait_lcore(lcore_id);
+               /* Destroy graph */
+               if (ret < 0 || rte_graph_destroy(
+                       rte_graph_from_name(lcore_conf[lcore_id].name))) {
+                       ret = -1;
+                       break;
+               }
+       }
+       free(node_patterns);
+
        /* Stop ports */
        RTE_ETH_FOREACH_DEV(portid) {
                if ((enabled_port_mask & (1 << portid)) == 0)
                        continue;
                printf("Closing port %d...", portid);
-               rte_eth_dev_stop(portid);
+               ret = rte_eth_dev_stop(portid);
+               if (ret != 0)
+                       printf("Failed to stop port %u: %s\n",
+                              portid, rte_strerror(-ret));
                rte_eth_dev_close(portid);
                printf(" Done\n");
        }
+
+       /* clean up the EAL */
+       rte_eal_cleanup();
        printf("Bye...\n");
 
        return ret;