1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
10 #include <sys/types.h>
11 #include <sys/queue.h>
12 #include <netinet/in.h>
20 #include <rte_common.h>
22 #include <rte_malloc.h>
23 #include <rte_memory.h>
24 #include <rte_memcpy.h>
26 #include <rte_launch.h>
27 #include <rte_atomic.h>
28 #include <rte_cycles.h>
29 #include <rte_prefetch.h>
30 #include <rte_lcore.h>
31 #include <rte_per_lcore.h>
32 #include <rte_branch_prediction.h>
33 #include <rte_interrupts.h>
34 #include <rte_random.h>
35 #include <rte_debug.h>
36 #include <rte_ether.h>
37 #include <rte_ethdev.h>
38 #include <rte_mempool.h>
40 #include <rte_timer.h>
41 #include <rte_keepalive.h>
45 #define RTE_LOGTYPE_L2FWD RTE_LOGTYPE_USER1
49 #define MAX_PKT_BURST 32
50 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
53 * Configurable number of RX/TX ring descriptors
55 #define RTE_TEST_RX_DESC_DEFAULT 1024
56 #define RTE_TEST_TX_DESC_DEFAULT 1024
57 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
58 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
60 /* ethernet addresses of ports */
61 static struct ether_addr l2fwd_ports_eth_addr[RTE_MAX_ETHPORTS];
63 /* mask of enabled ports */
64 static uint32_t l2fwd_enabled_port_mask;
66 /* list of enabled ports */
67 static uint32_t l2fwd_dst_ports[RTE_MAX_ETHPORTS];
69 static unsigned int l2fwd_rx_queue_per_lcore = 1;
71 #define MAX_RX_QUEUE_PER_LCORE 16
72 #define MAX_TX_QUEUE_PER_PORT 16
73 struct lcore_queue_conf {
75 unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
76 } __rte_cache_aligned;
77 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
79 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
81 static struct rte_eth_conf port_conf = {
86 .mq_mode = ETH_MQ_TX_NONE,
90 struct rte_mempool *l2fwd_pktmbuf_pool = NULL;
92 /* Per-port statistics struct */
93 struct l2fwd_port_statistics {
97 } __rte_cache_aligned;
98 struct l2fwd_port_statistics port_statistics[RTE_MAX_ETHPORTS];
100 /* A tsc-based timer responsible for triggering statistics printout */
101 #define TIMER_MILLISECOND 1
102 #define MAX_TIMER_PERIOD 86400 /* 1 day max */
103 static int64_t timer_period = 10 * TIMER_MILLISECOND * 1000; /* 10 seconds */
104 static int64_t check_period = 5; /* default check cycle is 5ms */
106 /* Keepalive structure */
107 struct rte_keepalive *rte_global_keepalive_info;
109 /* Termination signalling */
110 static int terminate_signal_received;
112 /* Termination signal handler */
113 static void handle_sigterm(__rte_unused int value)
115 terminate_signal_received = 1;
118 /* Print out statistics on packets dropped */
120 print_stats(__attribute__((unused)) struct rte_timer *ptr_timer,
121 __attribute__((unused)) void *ptr_data)
123 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
126 total_packets_dropped = 0;
127 total_packets_tx = 0;
128 total_packets_rx = 0;
130 const char clr[] = { 27, '[', '2', 'J', '\0' };
131 const char topLeft[] = { 27, '[', '1', ';', '1', 'H', '\0' };
133 /* Clear screen and move to top left */
134 printf("%s%s", clr, topLeft);
136 printf("\nPort statistics ====================================");
138 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
139 /* skip disabled ports */
140 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
142 printf("\nStatistics for port %u ------------------------------"
143 "\nPackets sent: %24"PRIu64
144 "\nPackets received: %20"PRIu64
145 "\nPackets dropped: %21"PRIu64,
147 port_statistics[portid].tx,
148 port_statistics[portid].rx,
149 port_statistics[portid].dropped);
151 total_packets_dropped += port_statistics[portid].dropped;
152 total_packets_tx += port_statistics[portid].tx;
153 total_packets_rx += port_statistics[portid].rx;
155 printf("\nAggregate statistics ==============================="
156 "\nTotal packets sent: %18"PRIu64
157 "\nTotal packets received: %14"PRIu64
158 "\nTotal packets dropped: %15"PRIu64,
161 total_packets_dropped);
162 printf("\n====================================================\n");
166 l2fwd_simple_forward(struct rte_mbuf *m, unsigned portid)
168 struct ether_hdr *eth;
172 struct rte_eth_dev_tx_buffer *buffer;
174 dst_port = l2fwd_dst_ports[portid];
175 eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
177 /* 02:00:00:00:00:xx */
178 tmp = ð->d_addr.addr_bytes[0];
179 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
182 ether_addr_copy(&l2fwd_ports_eth_addr[dst_port], ð->s_addr);
184 buffer = tx_buffer[dst_port];
185 sent = rte_eth_tx_buffer(dst_port, 0, buffer, m);
187 port_statistics[dst_port].tx += sent;
190 /* main processing loop */
192 l2fwd_main_loop(void)
194 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
198 uint64_t prev_tsc, diff_tsc, cur_tsc;
199 unsigned i, j, portid, nb_rx;
200 struct lcore_queue_conf *qconf;
201 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
202 / US_PER_S * BURST_TX_DRAIN_US;
203 struct rte_eth_dev_tx_buffer *buffer;
207 lcore_id = rte_lcore_id();
208 qconf = &lcore_queue_conf[lcore_id];
210 if (qconf->n_rx_port == 0) {
211 RTE_LOG(INFO, L2FWD, "lcore %u has nothing to do\n", lcore_id);
215 RTE_LOG(INFO, L2FWD, "entering main loop on lcore %u\n", lcore_id);
217 for (i = 0; i < qconf->n_rx_port; i++) {
219 portid = qconf->rx_port_list[i];
220 RTE_LOG(INFO, L2FWD, " -- lcoreid=%u portid=%u\n", lcore_id,
224 uint64_t tsc_initial = rte_rdtsc();
225 uint64_t tsc_lifetime = (rand()&0x07) * rte_get_tsc_hz();
227 while (!terminate_signal_received) {
228 /* Keepalive heartbeat */
229 rte_keepalive_mark_alive(rte_global_keepalive_info);
231 cur_tsc = rte_rdtsc();
234 * Die randomly within 7 secs for demo purposes if
237 if (check_period > 0 && cur_tsc - tsc_initial > tsc_lifetime)
241 * TX burst queue drain
243 diff_tsc = cur_tsc - prev_tsc;
244 if (unlikely(diff_tsc > drain_tsc)) {
246 for (i = 0; i < qconf->n_rx_port; i++) {
248 portid = l2fwd_dst_ports[qconf->rx_port_list[i]];
249 buffer = tx_buffer[portid];
251 sent = rte_eth_tx_buffer_flush(portid, 0, buffer);
253 port_statistics[portid].tx += sent;
261 * Read packet from RX queues
263 for (i = 0; i < qconf->n_rx_port; i++) {
265 portid = qconf->rx_port_list[i];
266 nb_rx = rte_eth_rx_burst(portid, 0,
267 pkts_burst, MAX_PKT_BURST);
269 port_statistics[portid].rx += nb_rx;
271 for (j = 0; j < nb_rx; j++) {
273 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
274 l2fwd_simple_forward(m, portid);
281 l2fwd_launch_one_lcore(__attribute__((unused)) void *dummy)
289 l2fwd_usage(const char *prgname)
291 printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
292 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
293 " -q NQ: number of queue (=ports) per lcore (default is 1)\n"
294 " -K PERIOD: Keepalive check period (5 default; 86400 max)\n"
295 " -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n",
300 l2fwd_parse_portmask(const char *portmask)
305 /* parse hexadecimal string */
306 pm = strtoul(portmask, &end, 16);
307 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
317 l2fwd_parse_nqueue(const char *q_arg)
322 /* parse hexadecimal string */
323 n = strtoul(q_arg, &end, 10);
324 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
328 if (n >= MAX_RX_QUEUE_PER_LCORE)
335 l2fwd_parse_timer_period(const char *q_arg)
340 /* parse number string */
341 n = strtol(q_arg, &end, 10);
342 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
344 if (n >= MAX_TIMER_PERIOD)
351 l2fwd_parse_check_period(const char *q_arg)
356 /* parse number string */
357 n = strtol(q_arg, &end, 10);
358 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
360 if (n >= MAX_TIMER_PERIOD)
366 /* Parse the argument given in the command line of the application */
368 l2fwd_parse_args(int argc, char **argv)
373 char *prgname = argv[0];
374 static struct option lgopts[] = {
380 while ((opt = getopt_long(argc, argvopt, "p:q:T:K:",
381 lgopts, &option_index)) != EOF) {
386 l2fwd_enabled_port_mask = l2fwd_parse_portmask(optarg);
387 if (l2fwd_enabled_port_mask == 0) {
388 printf("invalid portmask\n");
389 l2fwd_usage(prgname);
396 l2fwd_rx_queue_per_lcore = l2fwd_parse_nqueue(optarg);
397 if (l2fwd_rx_queue_per_lcore == 0) {
398 printf("invalid queue number\n");
399 l2fwd_usage(prgname);
406 timer_period = l2fwd_parse_timer_period(optarg)
407 * (int64_t)(1000 * TIMER_MILLISECOND);
408 if (timer_period < 0) {
409 printf("invalid timer period\n");
410 l2fwd_usage(prgname);
417 check_period = l2fwd_parse_check_period(optarg);
418 if (check_period < 0) {
419 printf("invalid check period\n");
420 l2fwd_usage(prgname);
427 l2fwd_usage(prgname);
431 l2fwd_usage(prgname);
437 argv[optind-1] = prgname;
440 optind = 1; /* reset getopt lib */
444 /* Check the link status of all ports in up to 9s, and print them finally */
446 check_all_ports_link_status(uint32_t port_mask)
448 #define CHECK_INTERVAL 100 /* 100ms */
449 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
451 uint8_t count, all_ports_up, print_flag = 0;
452 struct rte_eth_link link;
454 printf("\nChecking link status");
456 for (count = 0; count <= MAX_CHECK_TIME; count++) {
458 RTE_ETH_FOREACH_DEV(portid) {
459 if ((port_mask & (1 << portid)) == 0)
461 memset(&link, 0, sizeof(link));
462 rte_eth_link_get_nowait(portid, &link);
463 /* print link status if flag set */
464 if (print_flag == 1) {
465 if (link.link_status)
467 "Port%d Link Up. Speed %u Mbps - %s\n",
468 portid, link.link_speed,
469 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
470 ("full-duplex") : ("half-duplex\n"));
472 printf("Port %d Link Down\n", portid);
475 /* clear all_ports_up flag if any link down */
476 if (link.link_status == ETH_LINK_DOWN) {
481 /* after finally printing all link status, get out */
485 if (all_ports_up == 0) {
488 rte_delay_ms(CHECK_INTERVAL);
491 /* set the print_flag if all ports up or timeout */
492 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
500 dead_core(__rte_unused void *ptr_data, const int id_core)
502 if (terminate_signal_received)
504 printf("Dead core %i - restarting..\n", id_core);
505 if (rte_eal_get_lcore_state(id_core) == FINISHED) {
506 rte_eal_wait_lcore(id_core);
507 rte_eal_remote_launch(l2fwd_launch_one_lcore, NULL, id_core);
509 printf("..false positive!\n");
514 relay_core_state(void *ptr_data, const int id_core,
515 const enum rte_keepalive_state core_state, uint64_t last_alive)
517 rte_keepalive_relayed_state((struct rte_keepalive_shm *)ptr_data,
518 id_core, core_state, last_alive);
522 main(int argc, char **argv)
524 struct lcore_queue_conf *qconf;
527 uint16_t nb_ports_available = 0;
528 uint16_t portid, last_port;
529 unsigned lcore_id, rx_lcore_id;
530 unsigned nb_ports_in_mask = 0;
531 struct sigaction signal_handler;
532 struct rte_keepalive_shm *ka_shm;
534 memset(&signal_handler, 0, sizeof(signal_handler));
535 terminate_signal_received = 0;
536 signal_handler.sa_handler = &handle_sigterm;
537 if (sigaction(SIGINT, &signal_handler, NULL) == -1 ||
538 sigaction(SIGTERM, &signal_handler, NULL) == -1)
539 rte_exit(EXIT_FAILURE, "SIGNAL\n");
543 ret = rte_eal_init(argc, argv);
545 rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
549 l2fwd_enabled_port_mask = 0;
551 /* parse application arguments (after the EAL ones) */
552 ret = l2fwd_parse_args(argc, argv);
554 rte_exit(EXIT_FAILURE, "Invalid L2FWD arguments\n");
556 /* create the mbuf pool */
557 l2fwd_pktmbuf_pool = rte_pktmbuf_pool_create("mbuf_pool", NB_MBUF, 32,
558 0, RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
559 if (l2fwd_pktmbuf_pool == NULL)
560 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool\n");
562 nb_ports = rte_eth_dev_count_avail();
564 rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
566 /* reset l2fwd_dst_ports */
567 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++)
568 l2fwd_dst_ports[portid] = 0;
572 * Each logical core is assigned a dedicated TX queue on each port.
574 RTE_ETH_FOREACH_DEV(portid) {
575 /* skip ports that are not enabled */
576 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
579 if (nb_ports_in_mask % 2) {
580 l2fwd_dst_ports[portid] = last_port;
581 l2fwd_dst_ports[last_port] = portid;
587 if (nb_ports_in_mask % 2) {
588 printf("Notice: odd number of ports in portmask.\n");
589 l2fwd_dst_ports[last_port] = last_port;
595 /* Initialize the port/queue configuration of each logical core */
596 RTE_ETH_FOREACH_DEV(portid) {
597 /* skip ports that are not enabled */
598 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
601 /* get the lcore_id for this port */
602 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
603 lcore_queue_conf[rx_lcore_id].n_rx_port ==
604 l2fwd_rx_queue_per_lcore) {
606 if (rx_lcore_id >= RTE_MAX_LCORE)
607 rte_exit(EXIT_FAILURE, "Not enough cores\n");
610 if (qconf != &lcore_queue_conf[rx_lcore_id])
611 /* Assigned a new logical core in the loop above. */
612 qconf = &lcore_queue_conf[rx_lcore_id];
614 qconf->rx_port_list[qconf->n_rx_port] = portid;
616 printf("Lcore %u: RX port %u\n",
617 rx_lcore_id, portid);
620 /* Initialise each port */
621 RTE_ETH_FOREACH_DEV(portid) {
622 struct rte_eth_dev_info dev_info;
623 struct rte_eth_rxconf rxq_conf;
624 struct rte_eth_txconf txq_conf;
625 struct rte_eth_conf local_port_conf = port_conf;
627 /* skip ports that are not enabled */
628 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0) {
629 printf("Skipping disabled port %u\n", portid);
632 nb_ports_available++;
635 printf("Initializing port %u... ", portid);
637 rte_eth_dev_info_get(portid, &dev_info);
638 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
639 local_port_conf.txmode.offloads |=
640 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
641 ret = rte_eth_dev_configure(portid, 1, 1, &local_port_conf);
643 rte_exit(EXIT_FAILURE,
644 "Cannot configure device: err=%d, port=%u\n",
647 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
650 rte_exit(EXIT_FAILURE,
651 "Cannot adjust number of descriptors: err=%d, port=%u\n",
654 rte_eth_macaddr_get(portid, &l2fwd_ports_eth_addr[portid]);
656 /* init one RX queue */
658 rxq_conf = dev_info.default_rxconf;
659 rxq_conf.offloads = local_port_conf.rxmode.offloads;
660 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
661 rte_eth_dev_socket_id(portid),
665 rte_exit(EXIT_FAILURE,
666 "rte_eth_rx_queue_setup:err=%d, port=%u\n",
669 /* init one TX queue on each port */
671 txq_conf = dev_info.default_txconf;
672 txq_conf.offloads = local_port_conf.txmode.offloads;
673 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
674 rte_eth_dev_socket_id(portid),
677 rte_exit(EXIT_FAILURE,
678 "rte_eth_tx_queue_setup:err=%d, port=%u\n",
681 /* Initialize TX buffers */
682 tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
683 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
684 rte_eth_dev_socket_id(portid));
685 if (tx_buffer[portid] == NULL)
686 rte_exit(EXIT_FAILURE, "Cannot allocate buffer for tx on port %u\n",
689 rte_eth_tx_buffer_init(tx_buffer[portid], MAX_PKT_BURST);
691 ret = rte_eth_tx_buffer_set_err_callback(tx_buffer[portid],
692 rte_eth_tx_buffer_count_callback,
693 &port_statistics[portid].dropped);
695 rte_exit(EXIT_FAILURE,
696 "Cannot set error callback for tx buffer on port %u\n",
700 ret = rte_eth_dev_start(portid);
702 rte_exit(EXIT_FAILURE,
703 "rte_eth_dev_start:err=%d, port=%u\n",
706 rte_eth_promiscuous_enable(portid);
708 printf("Port %u, MAC address: "
709 "%02X:%02X:%02X:%02X:%02X:%02X\n\n",
711 l2fwd_ports_eth_addr[portid].addr_bytes[0],
712 l2fwd_ports_eth_addr[portid].addr_bytes[1],
713 l2fwd_ports_eth_addr[portid].addr_bytes[2],
714 l2fwd_ports_eth_addr[portid].addr_bytes[3],
715 l2fwd_ports_eth_addr[portid].addr_bytes[4],
716 l2fwd_ports_eth_addr[portid].addr_bytes[5]);
718 /* initialize port stats */
719 memset(&port_statistics, 0, sizeof(port_statistics));
722 if (!nb_ports_available) {
723 rte_exit(EXIT_FAILURE,
724 "All available ports are disabled. Please set portmask.\n");
727 check_all_ports_link_status(l2fwd_enabled_port_mask);
729 struct rte_timer hb_timer, stats_timer;
731 rte_timer_subsystem_init();
732 rte_timer_init(&stats_timer);
735 if (check_period > 0) {
736 ka_shm = rte_keepalive_shm_create();
738 rte_exit(EXIT_FAILURE,
739 "rte_keepalive_shm_create() failed");
740 rte_global_keepalive_info =
741 rte_keepalive_create(&dead_core, ka_shm);
742 if (rte_global_keepalive_info == NULL)
743 rte_exit(EXIT_FAILURE, "init_keep_alive() failed");
744 rte_keepalive_register_relay_callback(rte_global_keepalive_info,
745 relay_core_state, ka_shm);
746 rte_timer_init(&hb_timer);
747 if (rte_timer_reset(&hb_timer,
748 (check_period * rte_get_timer_hz()) / 1000,
751 (void(*)(struct rte_timer*, void*))
752 &rte_keepalive_dispatch_pings,
753 rte_global_keepalive_info
755 rte_exit(EXIT_FAILURE, "Keepalive setup failure.\n");
757 if (timer_period > 0) {
758 if (rte_timer_reset(&stats_timer,
759 (timer_period * rte_get_timer_hz()) / 1000,
764 rte_exit(EXIT_FAILURE, "Stats setup failure.\n");
766 /* launch per-lcore init on every slave lcore */
767 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
768 struct lcore_queue_conf *qconf = &lcore_queue_conf[lcore_id];
770 if (qconf->n_rx_port == 0)
772 "lcore %u has nothing to do\n",
776 rte_eal_remote_launch(
777 l2fwd_launch_one_lcore,
781 rte_keepalive_register_core(rte_global_keepalive_info,
785 while (!terminate_signal_received) {
790 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
791 if (rte_eal_wait_lcore(lcore_id) < 0)
796 rte_keepalive_shm_cleanup(ka_shm);