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
47 #define NB_MBUF_PER_PORT 3000
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 rte_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(__rte_unused struct rte_timer *ptr_timer,
121 __rte_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");
168 l2fwd_simple_forward(struct rte_mbuf *m, unsigned portid)
170 struct rte_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 rte_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 rte_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(__rte_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(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;
457 printf("\nChecking link status");
459 for (count = 0; count <= MAX_CHECK_TIME; count++) {
461 RTE_ETH_FOREACH_DEV(portid) {
462 if ((port_mask & (1 << portid)) == 0)
464 memset(&link, 0, sizeof(link));
465 ret = rte_eth_link_get_nowait(portid, &link);
469 printf("Port %u link get failed: %s\n",
470 portid, rte_strerror(-ret));
473 /* print link status if flag set */
474 if (print_flag == 1) {
475 if (link.link_status)
477 "Port%d Link Up. Speed %u Mbps - %s\n",
478 portid, link.link_speed,
479 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
480 ("full-duplex") : ("half-duplex"));
482 printf("Port %d Link Down\n", portid);
485 /* clear all_ports_up flag if any link down */
486 if (link.link_status == ETH_LINK_DOWN) {
491 /* after finally printing all link status, get out */
495 if (all_ports_up == 0) {
498 rte_delay_ms(CHECK_INTERVAL);
501 /* set the print_flag if all ports up or timeout */
502 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
510 dead_core(__rte_unused void *ptr_data, const int id_core)
512 if (terminate_signal_received)
514 printf("Dead core %i - restarting..\n", id_core);
515 if (rte_eal_get_lcore_state(id_core) == FINISHED) {
516 rte_eal_wait_lcore(id_core);
517 rte_eal_remote_launch(l2fwd_launch_one_lcore, NULL, id_core);
519 printf("..false positive!\n");
524 relay_core_state(void *ptr_data, const int id_core,
525 const enum rte_keepalive_state core_state, uint64_t last_alive)
527 rte_keepalive_relayed_state((struct rte_keepalive_shm *)ptr_data,
528 id_core, core_state, last_alive);
532 main(int argc, char **argv)
534 struct lcore_queue_conf *qconf;
537 uint16_t nb_ports_available = 0;
538 uint16_t portid, last_port;
539 unsigned lcore_id, rx_lcore_id;
540 unsigned nb_ports_in_mask = 0;
541 unsigned int total_nb_mbufs;
542 struct sigaction signal_handler;
543 struct rte_keepalive_shm *ka_shm;
545 memset(&signal_handler, 0, sizeof(signal_handler));
546 terminate_signal_received = 0;
547 signal_handler.sa_handler = &handle_sigterm;
548 if (sigaction(SIGINT, &signal_handler, NULL) == -1 ||
549 sigaction(SIGTERM, &signal_handler, NULL) == -1)
550 rte_exit(EXIT_FAILURE, "SIGNAL\n");
554 ret = rte_eal_init(argc, argv);
556 rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
560 l2fwd_enabled_port_mask = 0;
562 /* parse application arguments (after the EAL ones) */
563 ret = l2fwd_parse_args(argc, argv);
565 rte_exit(EXIT_FAILURE, "Invalid L2FWD arguments\n");
567 nb_ports = rte_eth_dev_count_avail();
569 rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
571 /* create the mbuf pool */
572 total_nb_mbufs = NB_MBUF_PER_PORT * nb_ports;
574 l2fwd_pktmbuf_pool = rte_pktmbuf_pool_create("mbuf_pool",
575 total_nb_mbufs, 32, 0, RTE_MBUF_DEFAULT_BUF_SIZE,
577 if (l2fwd_pktmbuf_pool == NULL)
578 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool\n");
580 /* reset l2fwd_dst_ports */
581 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++)
582 l2fwd_dst_ports[portid] = 0;
586 * Each logical core is assigned a dedicated TX queue on each port.
588 RTE_ETH_FOREACH_DEV(portid) {
589 /* skip ports that are not enabled */
590 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
593 if (nb_ports_in_mask % 2) {
594 l2fwd_dst_ports[portid] = last_port;
595 l2fwd_dst_ports[last_port] = portid;
601 if (nb_ports_in_mask % 2) {
602 printf("Notice: odd number of ports in portmask.\n");
603 l2fwd_dst_ports[last_port] = last_port;
609 /* Initialize the port/queue configuration of each logical core */
610 RTE_ETH_FOREACH_DEV(portid) {
611 /* skip ports that are not enabled */
612 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
615 /* get the lcore_id for this port */
616 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
617 lcore_queue_conf[rx_lcore_id].n_rx_port ==
618 l2fwd_rx_queue_per_lcore) {
620 if (rx_lcore_id >= RTE_MAX_LCORE)
621 rte_exit(EXIT_FAILURE, "Not enough cores\n");
624 if (qconf != &lcore_queue_conf[rx_lcore_id])
625 /* Assigned a new logical core in the loop above. */
626 qconf = &lcore_queue_conf[rx_lcore_id];
628 qconf->rx_port_list[qconf->n_rx_port] = portid;
630 printf("Lcore %u: RX port %u\n",
631 rx_lcore_id, portid);
634 /* Initialise each port */
635 RTE_ETH_FOREACH_DEV(portid) {
636 struct rte_eth_dev_info dev_info;
637 struct rte_eth_rxconf rxq_conf;
638 struct rte_eth_txconf txq_conf;
639 struct rte_eth_conf local_port_conf = port_conf;
641 /* skip ports that are not enabled */
642 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0) {
643 printf("Skipping disabled port %u\n", portid);
646 nb_ports_available++;
649 printf("Initializing port %u... ", portid);
652 ret = rte_eth_dev_info_get(portid, &dev_info);
654 rte_exit(EXIT_FAILURE,
655 "Error during getting device (port %u) info: %s\n",
656 portid, strerror(-ret));
658 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
659 local_port_conf.txmode.offloads |=
660 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
661 ret = rte_eth_dev_configure(portid, 1, 1, &local_port_conf);
663 rte_exit(EXIT_FAILURE,
664 "Cannot configure device: err=%d, port=%u\n",
667 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
670 rte_exit(EXIT_FAILURE,
671 "Cannot adjust number of descriptors: err=%d, port=%u\n",
674 ret = rte_eth_macaddr_get(portid,
675 &l2fwd_ports_eth_addr[portid]);
677 rte_exit(EXIT_FAILURE,
678 "Cannot mac address: err=%d, port=%u\n",
681 /* init one RX queue */
683 rxq_conf = dev_info.default_rxconf;
684 rxq_conf.offloads = local_port_conf.rxmode.offloads;
685 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
686 rte_eth_dev_socket_id(portid),
690 rte_exit(EXIT_FAILURE,
691 "rte_eth_rx_queue_setup:err=%d, port=%u\n",
694 /* init one TX queue on each port */
696 txq_conf = dev_info.default_txconf;
697 txq_conf.offloads = local_port_conf.txmode.offloads;
698 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
699 rte_eth_dev_socket_id(portid),
702 rte_exit(EXIT_FAILURE,
703 "rte_eth_tx_queue_setup:err=%d, port=%u\n",
706 /* Initialize TX buffers */
707 tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
708 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
709 rte_eth_dev_socket_id(portid));
710 if (tx_buffer[portid] == NULL)
711 rte_exit(EXIT_FAILURE, "Cannot allocate buffer for tx on port %u\n",
714 rte_eth_tx_buffer_init(tx_buffer[portid], MAX_PKT_BURST);
716 ret = rte_eth_tx_buffer_set_err_callback(tx_buffer[portid],
717 rte_eth_tx_buffer_count_callback,
718 &port_statistics[portid].dropped);
720 rte_exit(EXIT_FAILURE,
721 "Cannot set error callback for tx buffer on port %u\n",
725 ret = rte_eth_dev_start(portid);
727 rte_exit(EXIT_FAILURE,
728 "rte_eth_dev_start:err=%d, port=%u\n",
731 ret = rte_eth_promiscuous_enable(portid);
733 rte_exit(EXIT_FAILURE,
734 "rte_eth_promiscuous_enable:err=%s, port=%u\n",
735 rte_strerror(-ret), portid);
737 printf("Port %u, MAC address: "
738 "%02X:%02X:%02X:%02X:%02X:%02X\n\n",
740 l2fwd_ports_eth_addr[portid].addr_bytes[0],
741 l2fwd_ports_eth_addr[portid].addr_bytes[1],
742 l2fwd_ports_eth_addr[portid].addr_bytes[2],
743 l2fwd_ports_eth_addr[portid].addr_bytes[3],
744 l2fwd_ports_eth_addr[portid].addr_bytes[4],
745 l2fwd_ports_eth_addr[portid].addr_bytes[5]);
747 /* initialize port stats */
748 memset(&port_statistics, 0, sizeof(port_statistics));
751 if (!nb_ports_available) {
752 rte_exit(EXIT_FAILURE,
753 "All available ports are disabled. Please set portmask.\n");
756 check_all_ports_link_status(l2fwd_enabled_port_mask);
758 struct rte_timer hb_timer, stats_timer;
760 rte_timer_subsystem_init();
761 rte_timer_init(&stats_timer);
764 if (check_period > 0) {
765 ka_shm = rte_keepalive_shm_create();
767 rte_exit(EXIT_FAILURE,
768 "rte_keepalive_shm_create() failed");
769 rte_global_keepalive_info =
770 rte_keepalive_create(&dead_core, ka_shm);
771 if (rte_global_keepalive_info == NULL)
772 rte_exit(EXIT_FAILURE, "init_keep_alive() failed");
773 rte_keepalive_register_relay_callback(rte_global_keepalive_info,
774 relay_core_state, ka_shm);
775 rte_timer_init(&hb_timer);
776 if (rte_timer_reset(&hb_timer,
777 (check_period * rte_get_timer_hz()) / 1000,
780 (void(*)(struct rte_timer*, void*))
781 &rte_keepalive_dispatch_pings,
782 rte_global_keepalive_info
784 rte_exit(EXIT_FAILURE, "Keepalive setup failure.\n");
786 if (timer_period > 0) {
787 if (rte_timer_reset(&stats_timer,
788 (timer_period * rte_get_timer_hz()) / 1000,
793 rte_exit(EXIT_FAILURE, "Stats setup failure.\n");
795 /* launch per-lcore init on every slave lcore */
796 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
797 struct lcore_queue_conf *qconf = &lcore_queue_conf[lcore_id];
799 if (qconf->n_rx_port == 0)
801 "lcore %u has nothing to do\n",
805 rte_eal_remote_launch(
806 l2fwd_launch_one_lcore,
810 rte_keepalive_register_core(rte_global_keepalive_info,
814 while (!terminate_signal_received) {
819 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
820 if (rte_eal_wait_lcore(lcore_id) < 0)
825 rte_keepalive_shm_cleanup(ka_shm);