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 128
56 #define RTE_TEST_TX_DESC_DEFAULT 512
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 = {
84 .ignore_offload_bitfield = 1,
85 .offloads = DEV_RX_OFFLOAD_CRC_STRIP,
88 .mq_mode = ETH_MQ_TX_NONE,
92 struct rte_mempool *l2fwd_pktmbuf_pool = NULL;
94 /* Per-port statistics struct */
95 struct l2fwd_port_statistics {
99 } __rte_cache_aligned;
100 struct l2fwd_port_statistics port_statistics[RTE_MAX_ETHPORTS];
102 /* A tsc-based timer responsible for triggering statistics printout */
103 #define TIMER_MILLISECOND 1
104 #define MAX_TIMER_PERIOD 86400 /* 1 day max */
105 static int64_t timer_period = 10 * TIMER_MILLISECOND * 1000; /* 10 seconds */
106 static int64_t check_period = 5; /* default check cycle is 5ms */
108 /* Keepalive structure */
109 struct rte_keepalive *rte_global_keepalive_info;
111 /* Termination signalling */
112 static int terminate_signal_received;
114 /* Termination signal handler */
115 static void handle_sigterm(__rte_unused int value)
117 terminate_signal_received = 1;
120 /* Print out statistics on packets dropped */
122 print_stats(__attribute__((unused)) struct rte_timer *ptr_timer,
123 __attribute__((unused)) void *ptr_data)
125 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
128 total_packets_dropped = 0;
129 total_packets_tx = 0;
130 total_packets_rx = 0;
132 const char clr[] = { 27, '[', '2', 'J', '\0' };
133 const char topLeft[] = { 27, '[', '1', ';', '1', 'H', '\0' };
135 /* Clear screen and move to top left */
136 printf("%s%s", clr, topLeft);
138 printf("\nPort statistics ====================================");
140 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
141 /* skip disabled ports */
142 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
144 printf("\nStatistics for port %u ------------------------------"
145 "\nPackets sent: %24"PRIu64
146 "\nPackets received: %20"PRIu64
147 "\nPackets dropped: %21"PRIu64,
149 port_statistics[portid].tx,
150 port_statistics[portid].rx,
151 port_statistics[portid].dropped);
153 total_packets_dropped += port_statistics[portid].dropped;
154 total_packets_tx += port_statistics[portid].tx;
155 total_packets_rx += port_statistics[portid].rx;
157 printf("\nAggregate statistics ==============================="
158 "\nTotal packets sent: %18"PRIu64
159 "\nTotal packets received: %14"PRIu64
160 "\nTotal packets dropped: %15"PRIu64,
163 total_packets_dropped);
164 printf("\n====================================================\n");
168 l2fwd_simple_forward(struct rte_mbuf *m, unsigned portid)
170 struct ether_hdr *eth;
174 struct rte_eth_dev_tx_buffer *buffer;
176 dst_port = l2fwd_dst_ports[portid];
177 eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
179 /* 02:00:00:00:00:xx */
180 tmp = ð->d_addr.addr_bytes[0];
181 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
184 ether_addr_copy(&l2fwd_ports_eth_addr[dst_port], ð->s_addr);
186 buffer = tx_buffer[dst_port];
187 sent = rte_eth_tx_buffer(dst_port, 0, buffer, m);
189 port_statistics[dst_port].tx += sent;
192 /* main processing loop */
194 l2fwd_main_loop(void)
196 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
200 uint64_t prev_tsc, diff_tsc, cur_tsc;
201 unsigned i, j, portid, nb_rx;
202 struct lcore_queue_conf *qconf;
203 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
204 / US_PER_S * BURST_TX_DRAIN_US;
205 struct rte_eth_dev_tx_buffer *buffer;
209 lcore_id = rte_lcore_id();
210 qconf = &lcore_queue_conf[lcore_id];
212 if (qconf->n_rx_port == 0) {
213 RTE_LOG(INFO, L2FWD, "lcore %u has nothing to do\n", lcore_id);
217 RTE_LOG(INFO, L2FWD, "entering main loop on lcore %u\n", lcore_id);
219 for (i = 0; i < qconf->n_rx_port; i++) {
221 portid = qconf->rx_port_list[i];
222 RTE_LOG(INFO, L2FWD, " -- lcoreid=%u portid=%u\n", lcore_id,
226 uint64_t tsc_initial = rte_rdtsc();
227 uint64_t tsc_lifetime = (rand()&0x07) * rte_get_tsc_hz();
229 while (!terminate_signal_received) {
230 /* Keepalive heartbeat */
231 rte_keepalive_mark_alive(rte_global_keepalive_info);
233 cur_tsc = rte_rdtsc();
236 * Die randomly within 7 secs for demo purposes if
239 if (check_period > 0 && cur_tsc - tsc_initial > tsc_lifetime)
243 * TX burst queue drain
245 diff_tsc = cur_tsc - prev_tsc;
246 if (unlikely(diff_tsc > drain_tsc)) {
248 for (i = 0; i < qconf->n_rx_port; i++) {
250 portid = l2fwd_dst_ports[qconf->rx_port_list[i]];
251 buffer = tx_buffer[portid];
253 sent = rte_eth_tx_buffer_flush(portid, 0, buffer);
255 port_statistics[portid].tx += sent;
263 * Read packet from RX queues
265 for (i = 0; i < qconf->n_rx_port; i++) {
267 portid = qconf->rx_port_list[i];
268 nb_rx = rte_eth_rx_burst(portid, 0,
269 pkts_burst, MAX_PKT_BURST);
271 port_statistics[portid].rx += nb_rx;
273 for (j = 0; j < nb_rx; j++) {
275 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
276 l2fwd_simple_forward(m, portid);
283 l2fwd_launch_one_lcore(__attribute__((unused)) void *dummy)
291 l2fwd_usage(const char *prgname)
293 printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
294 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
295 " -q NQ: number of queue (=ports) per lcore (default is 1)\n"
296 " -K PERIOD: Keepalive check period (5 default; 86400 max)\n"
297 " -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n",
302 l2fwd_parse_portmask(const char *portmask)
307 /* parse hexadecimal string */
308 pm = strtoul(portmask, &end, 16);
309 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
319 l2fwd_parse_nqueue(const char *q_arg)
324 /* parse hexadecimal string */
325 n = strtoul(q_arg, &end, 10);
326 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
330 if (n >= MAX_RX_QUEUE_PER_LCORE)
337 l2fwd_parse_timer_period(const char *q_arg)
342 /* parse number string */
343 n = strtol(q_arg, &end, 10);
344 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
346 if (n >= MAX_TIMER_PERIOD)
353 l2fwd_parse_check_period(const char *q_arg)
358 /* parse number string */
359 n = strtol(q_arg, &end, 10);
360 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
362 if (n >= MAX_TIMER_PERIOD)
368 /* Parse the argument given in the command line of the application */
370 l2fwd_parse_args(int argc, char **argv)
375 char *prgname = argv[0];
376 static struct option lgopts[] = {
382 while ((opt = getopt_long(argc, argvopt, "p:q:T:K:",
383 lgopts, &option_index)) != EOF) {
388 l2fwd_enabled_port_mask = l2fwd_parse_portmask(optarg);
389 if (l2fwd_enabled_port_mask == 0) {
390 printf("invalid portmask\n");
391 l2fwd_usage(prgname);
398 l2fwd_rx_queue_per_lcore = l2fwd_parse_nqueue(optarg);
399 if (l2fwd_rx_queue_per_lcore == 0) {
400 printf("invalid queue number\n");
401 l2fwd_usage(prgname);
408 timer_period = l2fwd_parse_timer_period(optarg)
409 * (int64_t)(1000 * TIMER_MILLISECOND);
410 if (timer_period < 0) {
411 printf("invalid timer period\n");
412 l2fwd_usage(prgname);
419 check_period = l2fwd_parse_check_period(optarg);
420 if (check_period < 0) {
421 printf("invalid check period\n");
422 l2fwd_usage(prgname);
429 l2fwd_usage(prgname);
433 l2fwd_usage(prgname);
439 argv[optind-1] = prgname;
442 optind = 1; /* reset getopt lib */
446 /* Check the link status of all ports in up to 9s, and print them finally */
448 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask)
450 #define CHECK_INTERVAL 100 /* 100ms */
451 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
453 uint8_t count, all_ports_up, print_flag = 0;
454 struct rte_eth_link link;
456 printf("\nChecking link status");
458 for (count = 0; count <= MAX_CHECK_TIME; count++) {
460 for (portid = 0; portid < port_num; portid++) {
461 if ((port_mask & (1 << portid)) == 0)
463 memset(&link, 0, sizeof(link));
464 rte_eth_link_get_nowait(portid, &link);
465 /* print link status if flag set */
466 if (print_flag == 1) {
467 if (link.link_status)
469 "Port%d Link Up. Speed %u Mbps - %s\n",
470 portid, link.link_speed,
471 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
472 ("full-duplex") : ("half-duplex\n"));
474 printf("Port %d Link Down\n", portid);
477 /* clear all_ports_up flag if any link down */
478 if (link.link_status == ETH_LINK_DOWN) {
483 /* after finally printing all link status, get out */
487 if (all_ports_up == 0) {
490 rte_delay_ms(CHECK_INTERVAL);
493 /* set the print_flag if all ports up or timeout */
494 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
502 dead_core(__rte_unused void *ptr_data, const int id_core)
504 if (terminate_signal_received)
506 printf("Dead core %i - restarting..\n", id_core);
507 if (rte_eal_get_lcore_state(id_core) == FINISHED) {
508 rte_eal_wait_lcore(id_core);
509 rte_eal_remote_launch(l2fwd_launch_one_lcore, NULL, id_core);
511 printf("..false positive!\n");
516 relay_core_state(void *ptr_data, const int id_core,
517 const enum rte_keepalive_state core_state, uint64_t last_alive)
519 rte_keepalive_relayed_state((struct rte_keepalive_shm *)ptr_data,
520 id_core, core_state, last_alive);
524 main(int argc, char **argv)
526 struct lcore_queue_conf *qconf;
529 uint16_t nb_ports_available;
530 uint16_t portid, last_port;
531 unsigned lcore_id, rx_lcore_id;
532 unsigned nb_ports_in_mask = 0;
533 struct sigaction signal_handler;
534 struct rte_keepalive_shm *ka_shm;
536 memset(&signal_handler, 0, sizeof(signal_handler));
537 terminate_signal_received = 0;
538 signal_handler.sa_handler = &handle_sigterm;
539 if (sigaction(SIGINT, &signal_handler, NULL) == -1 ||
540 sigaction(SIGTERM, &signal_handler, NULL) == -1)
541 rte_exit(EXIT_FAILURE, "SIGNAL\n");
545 ret = rte_eal_init(argc, argv);
547 rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
551 l2fwd_enabled_port_mask = 0;
553 /* parse application arguments (after the EAL ones) */
554 ret = l2fwd_parse_args(argc, argv);
556 rte_exit(EXIT_FAILURE, "Invalid L2FWD arguments\n");
558 /* create the mbuf pool */
559 l2fwd_pktmbuf_pool = rte_pktmbuf_pool_create("mbuf_pool", NB_MBUF, 32,
560 0, RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
561 if (l2fwd_pktmbuf_pool == NULL)
562 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool\n");
564 nb_ports = rte_eth_dev_count();
566 rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
568 /* reset l2fwd_dst_ports */
569 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++)
570 l2fwd_dst_ports[portid] = 0;
574 * Each logical core is assigned a dedicated TX queue on each port.
576 for (portid = 0; portid < nb_ports; portid++) {
577 /* skip ports that are not enabled */
578 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
581 if (nb_ports_in_mask % 2) {
582 l2fwd_dst_ports[portid] = last_port;
583 l2fwd_dst_ports[last_port] = portid;
589 if (nb_ports_in_mask % 2) {
590 printf("Notice: odd number of ports in portmask.\n");
591 l2fwd_dst_ports[last_port] = last_port;
597 /* Initialize the port/queue configuration of each logical core */
598 for (portid = 0; portid < nb_ports; portid++) {
599 /* skip ports that are not enabled */
600 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
603 /* get the lcore_id for this port */
604 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
605 lcore_queue_conf[rx_lcore_id].n_rx_port ==
606 l2fwd_rx_queue_per_lcore) {
608 if (rx_lcore_id >= RTE_MAX_LCORE)
609 rte_exit(EXIT_FAILURE, "Not enough cores\n");
612 if (qconf != &lcore_queue_conf[rx_lcore_id])
613 /* Assigned a new logical core in the loop above. */
614 qconf = &lcore_queue_conf[rx_lcore_id];
616 qconf->rx_port_list[qconf->n_rx_port] = portid;
618 printf("Lcore %u: RX port %u\n",
619 rx_lcore_id, portid);
622 nb_ports_available = nb_ports;
624 /* Initialise each port */
625 for (portid = 0; portid < nb_ports; portid++) {
626 struct rte_eth_dev_info dev_info;
627 struct rte_eth_rxconf rxq_conf;
628 struct rte_eth_txconf txq_conf;
629 struct rte_eth_conf local_port_conf = port_conf;
631 /* skip ports that are not enabled */
632 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0) {
633 printf("Skipping disabled port %u\n", portid);
634 nb_ports_available--;
638 printf("Initializing port %u... ", portid);
640 rte_eth_dev_info_get(portid, &dev_info);
641 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
642 local_port_conf.txmode.offloads |=
643 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
644 ret = rte_eth_dev_configure(portid, 1, 1, &local_port_conf);
646 rte_exit(EXIT_FAILURE,
647 "Cannot configure device: err=%d, port=%u\n",
650 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
653 rte_exit(EXIT_FAILURE,
654 "Cannot adjust number of descriptors: err=%d, port=%u\n",
657 rte_eth_macaddr_get(portid, &l2fwd_ports_eth_addr[portid]);
659 /* init one RX queue */
661 rxq_conf = dev_info.default_rxconf;
662 rxq_conf.offloads = local_port_conf.rxmode.offloads;
663 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
664 rte_eth_dev_socket_id(portid),
668 rte_exit(EXIT_FAILURE,
669 "rte_eth_rx_queue_setup:err=%d, port=%u\n",
672 /* init one TX queue on each port */
674 txq_conf = dev_info.default_txconf;
675 txq_conf.txq_flags = ETH_TXQ_FLAGS_IGNORE;
676 txq_conf.offloads = local_port_conf.txmode.offloads;
677 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
678 rte_eth_dev_socket_id(portid),
681 rte_exit(EXIT_FAILURE,
682 "rte_eth_tx_queue_setup:err=%d, port=%u\n",
685 /* Initialize TX buffers */
686 tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
687 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
688 rte_eth_dev_socket_id(portid));
689 if (tx_buffer[portid] == NULL)
690 rte_exit(EXIT_FAILURE, "Cannot allocate buffer for tx on port %u\n",
693 rte_eth_tx_buffer_init(tx_buffer[portid], MAX_PKT_BURST);
695 ret = rte_eth_tx_buffer_set_err_callback(tx_buffer[portid],
696 rte_eth_tx_buffer_count_callback,
697 &port_statistics[portid].dropped);
699 rte_exit(EXIT_FAILURE,
700 "Cannot set error callback for tx buffer on port %u\n",
704 ret = rte_eth_dev_start(portid);
706 rte_exit(EXIT_FAILURE,
707 "rte_eth_dev_start:err=%d, port=%u\n",
710 rte_eth_promiscuous_enable(portid);
712 printf("Port %u, MAC address: "
713 "%02X:%02X:%02X:%02X:%02X:%02X\n\n",
715 l2fwd_ports_eth_addr[portid].addr_bytes[0],
716 l2fwd_ports_eth_addr[portid].addr_bytes[1],
717 l2fwd_ports_eth_addr[portid].addr_bytes[2],
718 l2fwd_ports_eth_addr[portid].addr_bytes[3],
719 l2fwd_ports_eth_addr[portid].addr_bytes[4],
720 l2fwd_ports_eth_addr[portid].addr_bytes[5]);
722 /* initialize port stats */
723 memset(&port_statistics, 0, sizeof(port_statistics));
726 if (!nb_ports_available) {
727 rte_exit(EXIT_FAILURE,
728 "All available ports are disabled. Please set portmask.\n");
731 check_all_ports_link_status(nb_ports, l2fwd_enabled_port_mask);
733 struct rte_timer hb_timer, stats_timer;
735 rte_timer_subsystem_init();
736 rte_timer_init(&stats_timer);
739 if (check_period > 0) {
740 ka_shm = rte_keepalive_shm_create();
742 rte_exit(EXIT_FAILURE,
743 "rte_keepalive_shm_create() failed");
744 rte_global_keepalive_info =
745 rte_keepalive_create(&dead_core, ka_shm);
746 if (rte_global_keepalive_info == NULL)
747 rte_exit(EXIT_FAILURE, "init_keep_alive() failed");
748 rte_keepalive_register_relay_callback(rte_global_keepalive_info,
749 relay_core_state, ka_shm);
750 rte_timer_init(&hb_timer);
751 if (rte_timer_reset(&hb_timer,
752 (check_period * rte_get_timer_hz()) / 1000,
755 (void(*)(struct rte_timer*, void*))
756 &rte_keepalive_dispatch_pings,
757 rte_global_keepalive_info
759 rte_exit(EXIT_FAILURE, "Keepalive setup failure.\n");
761 if (timer_period > 0) {
762 if (rte_timer_reset(&stats_timer,
763 (timer_period * rte_get_timer_hz()) / 1000,
768 rte_exit(EXIT_FAILURE, "Stats setup failure.\n");
770 /* launch per-lcore init on every slave lcore */
771 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
772 struct lcore_queue_conf *qconf = &lcore_queue_conf[lcore_id];
774 if (qconf->n_rx_port == 0)
776 "lcore %u has nothing to do\n",
780 rte_eal_remote_launch(
781 l2fwd_launch_one_lcore,
785 rte_keepalive_register_core(rte_global_keepalive_info,
789 while (!terminate_signal_received) {
794 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
795 if (rte_eal_wait_lcore(lcore_id) < 0)
800 rte_keepalive_shm_cleanup(ka_shm);