4 * Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of Intel Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
39 #include <sys/types.h>
40 #include <sys/queue.h>
41 #include <netinet/in.h>
48 #include <rte_common.h>
50 #include <rte_malloc.h>
51 #include <rte_memory.h>
52 #include <rte_memcpy.h>
53 #include <rte_memzone.h>
55 #include <rte_per_lcore.h>
56 #include <rte_launch.h>
57 #include <rte_atomic.h>
58 #include <rte_cycles.h>
59 #include <rte_prefetch.h>
60 #include <rte_lcore.h>
61 #include <rte_per_lcore.h>
62 #include <rte_branch_prediction.h>
63 #include <rte_interrupts.h>
65 #include <rte_random.h>
66 #include <rte_debug.h>
67 #include <rte_ether.h>
68 #include <rte_ethdev.h>
69 #include <rte_mempool.h>
71 #include <rte_timer.h>
72 #include <rte_keepalive.h>
76 #define RTE_LOGTYPE_L2FWD RTE_LOGTYPE_USER1
80 #define MAX_PKT_BURST 32
81 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
84 * Configurable number of RX/TX ring descriptors
86 #define RTE_TEST_RX_DESC_DEFAULT 128
87 #define RTE_TEST_TX_DESC_DEFAULT 512
88 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
89 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
91 /* ethernet addresses of ports */
92 static struct ether_addr l2fwd_ports_eth_addr[RTE_MAX_ETHPORTS];
94 /* mask of enabled ports */
95 static uint32_t l2fwd_enabled_port_mask;
97 /* list of enabled ports */
98 static uint32_t l2fwd_dst_ports[RTE_MAX_ETHPORTS];
100 static unsigned int l2fwd_rx_queue_per_lcore = 1;
102 #define MAX_RX_QUEUE_PER_LCORE 16
103 #define MAX_TX_QUEUE_PER_PORT 16
104 struct lcore_queue_conf {
106 unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
107 } __rte_cache_aligned;
108 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
110 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
112 static const struct rte_eth_conf port_conf = {
115 .header_split = 0, /**< Header Split disabled */
116 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
117 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
118 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
119 .hw_strip_crc = 0, /**< CRC stripped by hardware */
122 .mq_mode = ETH_MQ_TX_NONE,
126 struct rte_mempool *l2fwd_pktmbuf_pool = NULL;
128 /* Per-port statistics struct */
129 struct l2fwd_port_statistics {
133 } __rte_cache_aligned;
134 struct l2fwd_port_statistics port_statistics[RTE_MAX_ETHPORTS];
136 /* A tsc-based timer responsible for triggering statistics printout */
137 #define TIMER_MILLISECOND 1
138 #define MAX_TIMER_PERIOD 86400 /* 1 day max */
139 static int64_t timer_period = 10 * TIMER_MILLISECOND * 1000; /* 10 seconds */
140 static int64_t check_period = 5; /* default check cycle is 5ms */
142 /* Keepalive structure */
143 struct rte_keepalive *rte_global_keepalive_info;
145 /* Print out statistics on packets dropped */
147 print_stats(__attribute__((unused)) struct rte_timer *ptr_timer,
148 __attribute__((unused)) void *ptr_data)
150 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
153 total_packets_dropped = 0;
154 total_packets_tx = 0;
155 total_packets_rx = 0;
157 const char clr[] = { 27, '[', '2', 'J', '\0' };
158 const char topLeft[] = { 27, '[', '1', ';', '1', 'H', '\0' };
160 /* Clear screen and move to top left */
161 printf("%s%s", clr, topLeft);
163 printf("\nPort statistics ====================================");
165 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
166 /* skip disabled ports */
167 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
169 printf("\nStatistics for port %u ------------------------------"
170 "\nPackets sent: %24"PRIu64
171 "\nPackets received: %20"PRIu64
172 "\nPackets dropped: %21"PRIu64,
174 port_statistics[portid].tx,
175 port_statistics[portid].rx,
176 port_statistics[portid].dropped);
178 total_packets_dropped += port_statistics[portid].dropped;
179 total_packets_tx += port_statistics[portid].tx;
180 total_packets_rx += port_statistics[portid].rx;
182 printf("\nAggregate statistics ==============================="
183 "\nTotal packets sent: %18"PRIu64
184 "\nTotal packets received: %14"PRIu64
185 "\nTotal packets dropped: %15"PRIu64,
188 total_packets_dropped);
189 printf("\n====================================================\n");
193 l2fwd_simple_forward(struct rte_mbuf *m, unsigned portid)
195 struct ether_hdr *eth;
199 struct rte_eth_dev_tx_buffer *buffer;
201 dst_port = l2fwd_dst_ports[portid];
202 eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
204 /* 02:00:00:00:00:xx */
205 tmp = ð->d_addr.addr_bytes[0];
206 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
209 ether_addr_copy(&l2fwd_ports_eth_addr[dst_port], ð->s_addr);
211 buffer = tx_buffer[dst_port];
212 sent = rte_eth_tx_buffer(dst_port, 0, buffer, m);
214 port_statistics[dst_port].tx += sent;
217 /* main processing loop */
219 l2fwd_main_loop(void)
221 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
225 uint64_t prev_tsc, diff_tsc, cur_tsc;
226 unsigned i, j, portid, nb_rx;
227 struct lcore_queue_conf *qconf;
228 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
229 / US_PER_S * BURST_TX_DRAIN_US;
230 struct rte_eth_dev_tx_buffer *buffer;
234 lcore_id = rte_lcore_id();
235 qconf = &lcore_queue_conf[lcore_id];
237 if (qconf->n_rx_port == 0) {
238 RTE_LOG(INFO, L2FWD, "lcore %u has nothing to do\n", lcore_id);
242 RTE_LOG(INFO, L2FWD, "entering main loop on lcore %u\n", lcore_id);
244 for (i = 0; i < qconf->n_rx_port; i++) {
246 portid = qconf->rx_port_list[i];
247 RTE_LOG(INFO, L2FWD, " -- lcoreid=%u portid=%u\n", lcore_id,
251 uint64_t tsc_initial = rte_rdtsc();
252 uint64_t tsc_lifetime = (rand()&0x07) * rte_get_tsc_hz();
255 /* Keepalive heartbeat */
256 rte_keepalive_mark_alive(rte_global_keepalive_info);
258 cur_tsc = rte_rdtsc();
261 * Die randomly within 7 secs for demo purposes if
264 if (check_period > 0 && cur_tsc - tsc_initial > tsc_lifetime)
268 * TX burst queue drain
270 diff_tsc = cur_tsc - prev_tsc;
271 if (unlikely(diff_tsc > drain_tsc)) {
273 for (i = 0; i < qconf->n_rx_port; i++) {
275 portid = l2fwd_dst_ports[qconf->rx_port_list[i]];
276 buffer = tx_buffer[portid];
278 sent = rte_eth_tx_buffer_flush(portid, 0, buffer);
280 port_statistics[portid].tx += sent;
288 * Read packet from RX queues
290 for (i = 0; i < qconf->n_rx_port; i++) {
292 portid = qconf->rx_port_list[i];
293 nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
294 pkts_burst, MAX_PKT_BURST);
296 port_statistics[portid].rx += nb_rx;
298 for (j = 0; j < nb_rx; j++) {
300 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
301 l2fwd_simple_forward(m, portid);
308 l2fwd_launch_one_lcore(__attribute__((unused)) void *dummy)
316 l2fwd_usage(const char *prgname)
318 printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
319 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
320 " -q NQ: number of queue (=ports) per lcore (default is 1)\n"
321 " -K PERIOD: Keepalive check period (5 default; 86400 max)\n"
322 " -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n",
327 l2fwd_parse_portmask(const char *portmask)
332 /* parse hexadecimal string */
333 pm = strtoul(portmask, &end, 16);
334 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
344 l2fwd_parse_nqueue(const char *q_arg)
349 /* parse hexadecimal string */
350 n = strtoul(q_arg, &end, 10);
351 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
355 if (n >= MAX_RX_QUEUE_PER_LCORE)
362 l2fwd_parse_timer_period(const char *q_arg)
367 /* parse number string */
368 n = strtol(q_arg, &end, 10);
369 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
371 if (n >= MAX_TIMER_PERIOD)
378 l2fwd_parse_check_period(const char *q_arg)
383 /* parse number string */
384 n = strtol(q_arg, &end, 10);
385 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
387 if (n >= MAX_TIMER_PERIOD)
393 /* Parse the argument given in the command line of the application */
395 l2fwd_parse_args(int argc, char **argv)
400 char *prgname = argv[0];
401 static struct option lgopts[] = {
407 while ((opt = getopt_long(argc, argvopt, "p:q:T:K:",
408 lgopts, &option_index)) != EOF) {
413 l2fwd_enabled_port_mask = l2fwd_parse_portmask(optarg);
414 if (l2fwd_enabled_port_mask == 0) {
415 printf("invalid portmask\n");
416 l2fwd_usage(prgname);
423 l2fwd_rx_queue_per_lcore = l2fwd_parse_nqueue(optarg);
424 if (l2fwd_rx_queue_per_lcore == 0) {
425 printf("invalid queue number\n");
426 l2fwd_usage(prgname);
433 timer_period = l2fwd_parse_timer_period(optarg)
434 * (int64_t)(1000 * TIMER_MILLISECOND);
435 if (timer_period < 0) {
436 printf("invalid timer period\n");
437 l2fwd_usage(prgname);
444 check_period = l2fwd_parse_check_period(optarg);
445 if (check_period < 0) {
446 printf("invalid check period\n");
447 l2fwd_usage(prgname);
454 l2fwd_usage(prgname);
458 l2fwd_usage(prgname);
464 argv[optind-1] = prgname;
467 optind = 1; /* reset getopt lib */
471 /* Check the link status of all ports in up to 9s, and print them finally */
473 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
475 #define CHECK_INTERVAL 100 /* 100ms */
476 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
477 uint8_t portid, count, all_ports_up, print_flag = 0;
478 struct rte_eth_link link;
480 printf("\nChecking link status");
482 for (count = 0; count <= MAX_CHECK_TIME; count++) {
484 for (portid = 0; portid < port_num; portid++) {
485 if ((port_mask & (1 << portid)) == 0)
487 memset(&link, 0, sizeof(link));
488 rte_eth_link_get_nowait(portid, &link);
489 /* print link status if flag set */
490 if (print_flag == 1) {
491 if (link.link_status)
492 printf("Port %d Link Up - speed %u "
493 "Mbps - %s\n", (uint8_t)portid,
494 (unsigned)link.link_speed,
495 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
496 ("full-duplex") : ("half-duplex\n"));
498 printf("Port %d Link Down\n",
502 /* clear all_ports_up flag if any link down */
503 if (link.link_status == ETH_LINK_DOWN) {
508 /* after finally printing all link status, get out */
512 if (all_ports_up == 0) {
515 rte_delay_ms(CHECK_INTERVAL);
518 /* set the print_flag if all ports up or timeout */
519 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
527 dead_core(__rte_unused void *ptr_data, const int id_core)
529 printf("Dead core %i - restarting..\n", id_core);
530 if (rte_eal_get_lcore_state(id_core) == FINISHED) {
531 rte_eal_wait_lcore(id_core);
532 rte_eal_remote_launch(l2fwd_launch_one_lcore, NULL, id_core);
534 printf("..false positive!\n");
539 relay_core_state(void *ptr_data, const int id_core,
540 const enum rte_keepalive_state core_state, uint64_t last_alive)
542 rte_keepalive_relayed_state((struct rte_keepalive_shm *)ptr_data,
543 id_core, core_state, last_alive);
547 main(int argc, char **argv)
549 struct lcore_queue_conf *qconf;
550 struct rte_eth_dev_info dev_info;
553 uint8_t nb_ports_available;
554 uint8_t portid, last_port;
555 unsigned lcore_id, rx_lcore_id;
556 unsigned nb_ports_in_mask = 0;
559 ret = rte_eal_init(argc, argv);
561 rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
565 l2fwd_enabled_port_mask = 0;
567 /* parse application arguments (after the EAL ones) */
568 ret = l2fwd_parse_args(argc, argv);
570 rte_exit(EXIT_FAILURE, "Invalid L2FWD arguments\n");
572 /* create the mbuf pool */
573 l2fwd_pktmbuf_pool = rte_pktmbuf_pool_create("mbuf_pool", NB_MBUF, 32,
574 0, RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
575 if (l2fwd_pktmbuf_pool == NULL)
576 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool\n");
578 nb_ports = rte_eth_dev_count();
580 rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
582 /* reset l2fwd_dst_ports */
583 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++)
584 l2fwd_dst_ports[portid] = 0;
588 * Each logical core is assigned a dedicated TX queue on each port.
590 for (portid = 0; portid < nb_ports; portid++) {
591 /* skip ports that are not enabled */
592 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
595 if (nb_ports_in_mask % 2) {
596 l2fwd_dst_ports[portid] = last_port;
597 l2fwd_dst_ports[last_port] = portid;
603 rte_eth_dev_info_get(portid, &dev_info);
605 if (nb_ports_in_mask % 2) {
606 printf("Notice: odd number of ports in portmask.\n");
607 l2fwd_dst_ports[last_port] = last_port;
613 /* Initialize the port/queue configuration of each logical core */
614 for (portid = 0; portid < nb_ports; portid++) {
615 /* skip ports that are not enabled */
616 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
619 /* get the lcore_id for this port */
620 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
621 lcore_queue_conf[rx_lcore_id].n_rx_port ==
622 l2fwd_rx_queue_per_lcore) {
624 if (rx_lcore_id >= RTE_MAX_LCORE)
625 rte_exit(EXIT_FAILURE, "Not enough cores\n");
628 if (qconf != &lcore_queue_conf[rx_lcore_id])
629 /* Assigned a new logical core in the loop above. */
630 qconf = &lcore_queue_conf[rx_lcore_id];
632 qconf->rx_port_list[qconf->n_rx_port] = portid;
634 printf("Lcore %u: RX port %u\n",
635 rx_lcore_id, (unsigned) portid);
638 nb_ports_available = nb_ports;
640 /* Initialise each port */
641 for (portid = 0; portid < nb_ports; portid++) {
642 /* skip ports that are not enabled */
643 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0) {
644 printf("Skipping disabled port %u\n",
646 nb_ports_available--;
650 printf("Initializing port %u... ", (unsigned) portid);
652 ret = rte_eth_dev_configure(portid, 1, 1, &port_conf);
654 rte_exit(EXIT_FAILURE,
655 "Cannot configure device: err=%d, port=%u\n",
656 ret, (unsigned) portid);
658 rte_eth_macaddr_get(portid, &l2fwd_ports_eth_addr[portid]);
660 /* init one RX queue */
662 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
663 rte_eth_dev_socket_id(portid),
667 rte_exit(EXIT_FAILURE,
668 "rte_eth_rx_queue_setup:err=%d, port=%u\n",
669 ret, (unsigned) portid);
671 /* init one TX queue on each port */
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",
679 ret, (unsigned) portid);
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, "Cannot set error callback for "
696 "tx buffer on port %u\n", (unsigned) portid);
699 ret = rte_eth_dev_start(portid);
701 rte_exit(EXIT_FAILURE,
702 "rte_eth_dev_start:err=%d, port=%u\n",
703 ret, (unsigned) portid);
705 rte_eth_promiscuous_enable(portid);
707 printf("Port %u, MAC address: "
708 "%02X:%02X:%02X:%02X:%02X:%02X\n\n",
710 l2fwd_ports_eth_addr[portid].addr_bytes[0],
711 l2fwd_ports_eth_addr[portid].addr_bytes[1],
712 l2fwd_ports_eth_addr[portid].addr_bytes[2],
713 l2fwd_ports_eth_addr[portid].addr_bytes[3],
714 l2fwd_ports_eth_addr[portid].addr_bytes[4],
715 l2fwd_ports_eth_addr[portid].addr_bytes[5]);
717 /* initialize port stats */
718 memset(&port_statistics, 0, sizeof(port_statistics));
721 if (!nb_ports_available) {
722 rte_exit(EXIT_FAILURE,
723 "All available ports are disabled. Please set portmask.\n");
726 check_all_ports_link_status(nb_ports, l2fwd_enabled_port_mask);
728 struct rte_timer hb_timer, stats_timer;
730 rte_timer_subsystem_init();
731 rte_timer_init(&stats_timer);
733 if (check_period > 0) {
734 struct rte_keepalive_shm *ka_shm;
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,
790 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
791 if (rte_eal_wait_lcore(lcore_id) < 0)