4 * Copyright(c) 2010-2014 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_memory.h>
51 #include <rte_memcpy.h>
52 #include <rte_memzone.h>
53 #include <rte_tailq.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>
70 #include <rte_mempool.h>
73 #define RTE_LOGTYPE_L2FWD RTE_LOGTYPE_USER1
75 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
78 #define MAX_PKT_BURST 32
79 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
82 * Configurable number of RX/TX ring descriptors
84 #define RTE_TEST_RX_DESC_DEFAULT 128
85 #define RTE_TEST_TX_DESC_DEFAULT 512
86 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
87 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
89 /* ethernet addresses of ports */
90 static struct ether_addr l2fwd_ports_eth_addr[RTE_MAX_ETHPORTS];
92 /* mask of enabled ports */
93 static uint32_t l2fwd_enabled_port_mask = 0;
95 /* list of enabled ports */
96 static uint32_t l2fwd_dst_ports[RTE_MAX_ETHPORTS];
98 static unsigned int l2fwd_rx_queue_per_lcore = 1;
102 struct rte_mbuf *m_table[MAX_PKT_BURST];
105 #define MAX_RX_QUEUE_PER_LCORE 16
106 #define MAX_TX_QUEUE_PER_PORT 16
107 struct lcore_queue_conf {
109 unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
110 struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
112 } __rte_cache_aligned;
113 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
115 static const struct rte_eth_conf port_conf = {
118 .header_split = 0, /**< Header Split disabled */
119 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
120 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
121 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
122 .hw_strip_crc = 0, /**< CRC stripped by hardware */
125 .mq_mode = ETH_MQ_TX_NONE,
129 struct rte_mempool * l2fwd_pktmbuf_pool = NULL;
131 /* Per-port statistics struct */
132 struct l2fwd_port_statistics {
136 } __rte_cache_aligned;
137 struct l2fwd_port_statistics port_statistics[RTE_MAX_ETHPORTS];
139 /* A tsc-based timer responsible for triggering statistics printout */
140 #define TIMER_MILLISECOND 2000000ULL /* around 1ms at 2 Ghz */
141 #define MAX_TIMER_PERIOD 86400 /* 1 day max */
142 static int64_t timer_period = 10 * TIMER_MILLISECOND * 1000; /* default period is 10 seconds */
144 /* Print out statistics on packets dropped */
148 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
151 total_packets_dropped = 0;
152 total_packets_tx = 0;
153 total_packets_rx = 0;
155 const char clr[] = { 27, '[', '2', 'J', '\0' };
156 const char topLeft[] = { 27, '[', '1', ';', '1', 'H','\0' };
158 /* Clear screen and move to top left */
159 printf("%s%s", clr, topLeft);
161 printf("\nPort statistics ====================================");
163 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
164 /* skip disabled ports */
165 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
167 printf("\nStatistics for port %u ------------------------------"
168 "\nPackets sent: %24"PRIu64
169 "\nPackets received: %20"PRIu64
170 "\nPackets dropped: %21"PRIu64,
172 port_statistics[portid].tx,
173 port_statistics[portid].rx,
174 port_statistics[portid].dropped);
176 total_packets_dropped += port_statistics[portid].dropped;
177 total_packets_tx += port_statistics[portid].tx;
178 total_packets_rx += port_statistics[portid].rx;
180 printf("\nAggregate statistics ==============================="
181 "\nTotal packets sent: %18"PRIu64
182 "\nTotal packets received: %14"PRIu64
183 "\nTotal packets dropped: %15"PRIu64,
186 total_packets_dropped);
187 printf("\n====================================================\n");
190 /* Send the burst of packets on an output interface */
192 l2fwd_send_burst(struct lcore_queue_conf *qconf, unsigned n, uint8_t port)
194 struct rte_mbuf **m_table;
198 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
200 ret = rte_eth_tx_burst(port, (uint16_t) queueid, m_table, (uint16_t) n);
201 port_statistics[port].tx += ret;
202 if (unlikely(ret < n)) {
203 port_statistics[port].dropped += (n - ret);
205 rte_pktmbuf_free(m_table[ret]);
212 /* Enqueue packets for TX and prepare them to be sent */
214 l2fwd_send_packet(struct rte_mbuf *m, uint8_t port)
216 unsigned lcore_id, len;
217 struct lcore_queue_conf *qconf;
219 lcore_id = rte_lcore_id();
221 qconf = &lcore_queue_conf[lcore_id];
222 len = qconf->tx_mbufs[port].len;
223 qconf->tx_mbufs[port].m_table[len] = m;
226 /* enough pkts to be sent */
227 if (unlikely(len == MAX_PKT_BURST)) {
228 l2fwd_send_burst(qconf, MAX_PKT_BURST, port);
232 qconf->tx_mbufs[port].len = len;
237 l2fwd_simple_forward(struct rte_mbuf *m, unsigned portid)
239 struct ether_hdr *eth;
243 dst_port = l2fwd_dst_ports[portid];
244 eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
246 /* 02:00:00:00:00:xx */
247 tmp = ð->d_addr.addr_bytes[0];
248 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
251 ether_addr_copy(&l2fwd_ports_eth_addr[dst_port], ð->s_addr);
253 l2fwd_send_packet(m, (uint8_t) dst_port);
256 /* main processing loop */
258 l2fwd_main_loop(void)
260 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
263 uint64_t prev_tsc, diff_tsc, cur_tsc, timer_tsc;
264 unsigned i, j, portid, nb_rx;
265 struct lcore_queue_conf *qconf;
266 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
271 lcore_id = rte_lcore_id();
272 qconf = &lcore_queue_conf[lcore_id];
274 if (qconf->n_rx_port == 0) {
275 RTE_LOG(INFO, L2FWD, "lcore %u has nothing to do\n", lcore_id);
279 RTE_LOG(INFO, L2FWD, "entering main loop on lcore %u\n", lcore_id);
281 for (i = 0; i < qconf->n_rx_port; i++) {
283 portid = qconf->rx_port_list[i];
284 RTE_LOG(INFO, L2FWD, " -- lcoreid=%u portid=%u\n", lcore_id,
290 cur_tsc = rte_rdtsc();
293 * TX burst queue drain
295 diff_tsc = cur_tsc - prev_tsc;
296 if (unlikely(diff_tsc > drain_tsc)) {
298 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
299 if (qconf->tx_mbufs[portid].len == 0)
301 l2fwd_send_burst(&lcore_queue_conf[lcore_id],
302 qconf->tx_mbufs[portid].len,
304 qconf->tx_mbufs[portid].len = 0;
307 /* if timer is enabled */
308 if (timer_period > 0) {
310 /* advance the timer */
311 timer_tsc += diff_tsc;
313 /* if timer has reached its timeout */
314 if (unlikely(timer_tsc >= (uint64_t) timer_period)) {
316 /* do this only on master core */
317 if (lcore_id == rte_get_master_lcore()) {
319 /* reset the timer */
329 * Read packet from RX queues
331 for (i = 0; i < qconf->n_rx_port; i++) {
333 portid = qconf->rx_port_list[i];
334 nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
335 pkts_burst, MAX_PKT_BURST);
337 port_statistics[portid].rx += nb_rx;
339 for (j = 0; j < nb_rx; j++) {
341 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
342 l2fwd_simple_forward(m, portid);
349 l2fwd_launch_one_lcore(__attribute__((unused)) void *dummy)
357 l2fwd_usage(const char *prgname)
359 printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
360 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
361 " -q NQ: number of queue (=ports) per lcore (default is 1)\n"
362 " -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n",
367 l2fwd_parse_portmask(const char *portmask)
372 /* parse hexadecimal string */
373 pm = strtoul(portmask, &end, 16);
374 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
384 l2fwd_parse_nqueue(const char *q_arg)
389 /* parse hexadecimal string */
390 n = strtoul(q_arg, &end, 10);
391 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
395 if (n >= MAX_RX_QUEUE_PER_LCORE)
402 l2fwd_parse_timer_period(const char *q_arg)
407 /* parse number string */
408 n = strtol(q_arg, &end, 10);
409 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
411 if (n >= MAX_TIMER_PERIOD)
417 /* Parse the argument given in the command line of the application */
419 l2fwd_parse_args(int argc, char **argv)
424 char *prgname = argv[0];
425 static struct option lgopts[] = {
431 while ((opt = getopt_long(argc, argvopt, "p:q:T:",
432 lgopts, &option_index)) != EOF) {
437 l2fwd_enabled_port_mask = l2fwd_parse_portmask(optarg);
438 if (l2fwd_enabled_port_mask == 0) {
439 printf("invalid portmask\n");
440 l2fwd_usage(prgname);
447 l2fwd_rx_queue_per_lcore = l2fwd_parse_nqueue(optarg);
448 if (l2fwd_rx_queue_per_lcore == 0) {
449 printf("invalid queue number\n");
450 l2fwd_usage(prgname);
457 timer_period = l2fwd_parse_timer_period(optarg) * 1000 * TIMER_MILLISECOND;
458 if (timer_period < 0) {
459 printf("invalid timer period\n");
460 l2fwd_usage(prgname);
467 l2fwd_usage(prgname);
471 l2fwd_usage(prgname);
477 argv[optind-1] = prgname;
480 optind = 0; /* reset getopt lib */
484 /* Check the link status of all ports in up to 9s, and print them finally */
486 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
488 #define CHECK_INTERVAL 100 /* 100ms */
489 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
490 uint8_t portid, count, all_ports_up, print_flag = 0;
491 struct rte_eth_link link;
493 printf("\nChecking link status");
495 for (count = 0; count <= MAX_CHECK_TIME; count++) {
497 for (portid = 0; portid < port_num; portid++) {
498 if ((port_mask & (1 << portid)) == 0)
500 memset(&link, 0, sizeof(link));
501 rte_eth_link_get_nowait(portid, &link);
502 /* print link status if flag set */
503 if (print_flag == 1) {
504 if (link.link_status)
505 printf("Port %d Link Up - speed %u "
506 "Mbps - %s\n", (uint8_t)portid,
507 (unsigned)link.link_speed,
508 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
509 ("full-duplex") : ("half-duplex\n"));
511 printf("Port %d Link Down\n",
515 /* clear all_ports_up flag if any link down */
516 if (link.link_status == 0) {
521 /* after finally printing all link status, get out */
525 if (all_ports_up == 0) {
528 rte_delay_ms(CHECK_INTERVAL);
531 /* set the print_flag if all ports up or timeout */
532 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
540 main(int argc, char **argv)
542 struct lcore_queue_conf *qconf;
543 struct rte_eth_dev_info dev_info;
546 uint8_t nb_ports_available;
547 uint8_t portid, last_port;
548 unsigned lcore_id, rx_lcore_id;
549 unsigned nb_ports_in_mask = 0;
552 ret = rte_eal_init(argc, argv);
554 rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
558 /* parse application arguments (after the EAL ones) */
559 ret = l2fwd_parse_args(argc, argv);
561 rte_exit(EXIT_FAILURE, "Invalid L2FWD arguments\n");
563 /* create the mbuf pool */
565 rte_mempool_create("mbuf_pool", NB_MBUF,
567 sizeof(struct rte_pktmbuf_pool_private),
568 rte_pktmbuf_pool_init, NULL,
569 rte_pktmbuf_init, NULL,
571 if (l2fwd_pktmbuf_pool == NULL)
572 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool\n");
574 nb_ports = rte_eth_dev_count();
576 rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
578 if (nb_ports > RTE_MAX_ETHPORTS)
579 nb_ports = RTE_MAX_ETHPORTS;
581 /* reset l2fwd_dst_ports */
582 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++)
583 l2fwd_dst_ports[portid] = 0;
587 * Each logical core is assigned a dedicated TX queue on each port.
589 for (portid = 0; portid < nb_ports; portid++) {
590 /* skip ports that are not enabled */
591 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
594 if (nb_ports_in_mask % 2) {
595 l2fwd_dst_ports[portid] = last_port;
596 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", rx_lcore_id, (unsigned) portid);
637 nb_ports_available = nb_ports;
639 /* Initialise each port */
640 for (portid = 0; portid < nb_ports; portid++) {
641 /* skip ports that are not enabled */
642 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0) {
643 printf("Skipping disabled port %u\n", (unsigned) portid);
644 nb_ports_available--;
648 printf("Initializing port %u... ", (unsigned) portid);
650 ret = rte_eth_dev_configure(portid, 1, 1, &port_conf);
652 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
653 ret, (unsigned) portid);
655 rte_eth_macaddr_get(portid,&l2fwd_ports_eth_addr[portid]);
657 /* init one RX queue */
659 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
660 rte_eth_dev_socket_id(portid),
664 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup:err=%d, port=%u\n",
665 ret, (unsigned) portid);
667 /* init one TX queue on each port */
669 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
670 rte_eth_dev_socket_id(portid),
673 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup:err=%d, port=%u\n",
674 ret, (unsigned) portid);
677 ret = rte_eth_dev_start(portid);
679 rte_exit(EXIT_FAILURE, "rte_eth_dev_start:err=%d, port=%u\n",
680 ret, (unsigned) portid);
684 rte_eth_promiscuous_enable(portid);
686 printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
688 l2fwd_ports_eth_addr[portid].addr_bytes[0],
689 l2fwd_ports_eth_addr[portid].addr_bytes[1],
690 l2fwd_ports_eth_addr[portid].addr_bytes[2],
691 l2fwd_ports_eth_addr[portid].addr_bytes[3],
692 l2fwd_ports_eth_addr[portid].addr_bytes[4],
693 l2fwd_ports_eth_addr[portid].addr_bytes[5]);
695 /* initialize port stats */
696 memset(&port_statistics, 0, sizeof(port_statistics));
699 if (!nb_ports_available) {
700 rte_exit(EXIT_FAILURE,
701 "All available ports are disabled. Please set portmask.\n");
704 check_all_ports_link_status(nb_ports, l2fwd_enabled_port_mask);
706 /* launch per-lcore init on every lcore */
707 rte_eal_mp_remote_launch(l2fwd_launch_one_lcore, NULL, CALL_MASTER);
708 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
709 if (rte_eal_wait_lcore(lcore_id) < 0)