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>
50 #include <rte_common.h>
52 #include <rte_malloc.h>
53 #include <rte_memory.h>
54 #include <rte_memcpy.h>
55 #include <rte_memzone.h>
57 #include <rte_per_lcore.h>
58 #include <rte_launch.h>
59 #include <rte_atomic.h>
60 #include <rte_cycles.h>
61 #include <rte_prefetch.h>
62 #include <rte_lcore.h>
63 #include <rte_per_lcore.h>
64 #include <rte_branch_prediction.h>
65 #include <rte_interrupts.h>
67 #include <rte_random.h>
68 #include <rte_debug.h>
69 #include <rte_ether.h>
70 #include <rte_ethdev.h>
71 #include <rte_mempool.h>
74 static volatile bool force_quit;
76 /* MAC updating enabled by default */
77 static int mac_updating = 1;
79 #define RTE_LOGTYPE_L2FWD RTE_LOGTYPE_USER1
83 #define MAX_PKT_BURST 32
84 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
85 #define MEMPOOL_CACHE_SIZE 256
88 * Configurable number of RX/TX ring descriptors
90 #define RTE_TEST_RX_DESC_DEFAULT 128
91 #define RTE_TEST_TX_DESC_DEFAULT 512
92 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
93 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
95 /* ethernet addresses of ports */
96 static struct ether_addr l2fwd_ports_eth_addr[RTE_MAX_ETHPORTS];
98 /* mask of enabled ports */
99 static uint32_t l2fwd_enabled_port_mask = 0;
101 /* list of enabled ports */
102 static uint32_t l2fwd_dst_ports[RTE_MAX_ETHPORTS];
104 static unsigned int l2fwd_rx_queue_per_lcore = 1;
106 #define MAX_RX_QUEUE_PER_LCORE 16
107 #define MAX_TX_QUEUE_PER_PORT 16
108 struct lcore_queue_conf {
110 unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
111 } __rte_cache_aligned;
112 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
114 static struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
116 static const struct rte_eth_conf port_conf = {
119 .header_split = 0, /**< Header Split disabled */
120 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
121 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
122 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
123 .hw_strip_crc = 1, /**< CRC stripped by hardware */
126 .mq_mode = ETH_MQ_TX_NONE,
130 struct rte_mempool * l2fwd_pktmbuf_pool = NULL;
132 /* Per-port statistics struct */
133 struct l2fwd_port_statistics {
137 } __rte_cache_aligned;
138 struct l2fwd_port_statistics port_statistics[RTE_MAX_ETHPORTS];
140 #define MAX_TIMER_PERIOD 86400 /* 1 day max */
141 /* A tsc-based timer responsible for triggering statistics printout */
142 static uint64_t timer_period = 10; /* 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");
191 l2fwd_mac_updating(struct rte_mbuf *m, unsigned dest_portid)
193 struct ether_hdr *eth;
196 eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
198 /* 02:00:00:00:00:xx */
199 tmp = ð->d_addr.addr_bytes[0];
200 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dest_portid << 40);
203 ether_addr_copy(&l2fwd_ports_eth_addr[dest_portid], ð->s_addr);
207 l2fwd_simple_forward(struct rte_mbuf *m, unsigned portid)
211 struct rte_eth_dev_tx_buffer *buffer;
213 dst_port = l2fwd_dst_ports[portid];
216 l2fwd_mac_updating(m, dst_port);
218 buffer = tx_buffer[dst_port];
219 sent = rte_eth_tx_buffer(dst_port, 0, buffer, m);
221 port_statistics[dst_port].tx += sent;
224 /* main processing loop */
226 l2fwd_main_loop(void)
228 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
232 uint64_t prev_tsc, diff_tsc, cur_tsc, timer_tsc;
233 unsigned i, j, portid, nb_rx;
234 struct lcore_queue_conf *qconf;
235 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S *
237 struct rte_eth_dev_tx_buffer *buffer;
242 lcore_id = rte_lcore_id();
243 qconf = &lcore_queue_conf[lcore_id];
245 if (qconf->n_rx_port == 0) {
246 RTE_LOG(INFO, L2FWD, "lcore %u has nothing to do\n", lcore_id);
250 RTE_LOG(INFO, L2FWD, "entering main loop on lcore %u\n", lcore_id);
252 for (i = 0; i < qconf->n_rx_port; i++) {
254 portid = qconf->rx_port_list[i];
255 RTE_LOG(INFO, L2FWD, " -- lcoreid=%u portid=%u\n", lcore_id,
260 while (!force_quit) {
262 cur_tsc = rte_rdtsc();
265 * TX burst queue drain
267 diff_tsc = cur_tsc - prev_tsc;
268 if (unlikely(diff_tsc > drain_tsc)) {
270 for (i = 0; i < qconf->n_rx_port; i++) {
272 portid = l2fwd_dst_ports[qconf->rx_port_list[i]];
273 buffer = tx_buffer[portid];
275 sent = rte_eth_tx_buffer_flush(portid, 0, buffer);
277 port_statistics[portid].tx += sent;
281 /* if timer is enabled */
282 if (timer_period > 0) {
284 /* advance the timer */
285 timer_tsc += diff_tsc;
287 /* if timer has reached its timeout */
288 if (unlikely(timer_tsc >= timer_period)) {
290 /* do this only on master core */
291 if (lcore_id == rte_get_master_lcore()) {
293 /* reset the timer */
303 * Read packet from RX queues
305 for (i = 0; i < qconf->n_rx_port; i++) {
307 portid = qconf->rx_port_list[i];
308 nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
309 pkts_burst, MAX_PKT_BURST);
311 port_statistics[portid].rx += nb_rx;
313 for (j = 0; j < nb_rx; j++) {
315 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
316 l2fwd_simple_forward(m, portid);
323 l2fwd_launch_one_lcore(__attribute__((unused)) void *dummy)
331 l2fwd_usage(const char *prgname)
333 printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
334 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
335 " -q NQ: number of queue (=ports) per lcore (default is 1)\n"
336 " -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n"
337 " --[no-]mac-updating: Enable or disable MAC addresses updating (enabled by default)\n"
339 " - The source MAC address is replaced by the TX port MAC address\n"
340 " - The destination MAC address is replaced by 02:00:00:00:00:TX_PORT_ID\n",
345 l2fwd_parse_portmask(const char *portmask)
350 /* parse hexadecimal string */
351 pm = strtoul(portmask, &end, 16);
352 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
362 l2fwd_parse_nqueue(const char *q_arg)
367 /* parse hexadecimal string */
368 n = strtoul(q_arg, &end, 10);
369 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
373 if (n >= MAX_RX_QUEUE_PER_LCORE)
380 l2fwd_parse_timer_period(const char *q_arg)
385 /* parse number string */
386 n = strtol(q_arg, &end, 10);
387 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
389 if (n >= MAX_TIMER_PERIOD)
395 static const char short_options[] =
397 "q:" /* number of queues */
398 "T:" /* timer period */
401 #define CMD_LINE_OPT_MAC_UPDATING "mac-updating"
402 #define CMD_LINE_OPT_NO_MAC_UPDATING "no-mac-updating"
405 /* long options mapped to a short option */
407 /* first long only option value must be >= 256, so that we won't
408 * conflict with short options */
409 CMD_LINE_OPT_MIN_NUM = 256,
412 static const struct option lgopts[] = {
413 { CMD_LINE_OPT_MAC_UPDATING, no_argument, &mac_updating, 1},
414 { CMD_LINE_OPT_NO_MAC_UPDATING, no_argument, &mac_updating, 0},
418 /* Parse the argument given in the command line of the application */
420 l2fwd_parse_args(int argc, char **argv)
422 int opt, ret, timer_secs;
425 char *prgname = argv[0];
429 while ((opt = getopt_long(argc, argvopt, short_options,
430 lgopts, &option_index)) != EOF) {
435 l2fwd_enabled_port_mask = l2fwd_parse_portmask(optarg);
436 if (l2fwd_enabled_port_mask == 0) {
437 printf("invalid portmask\n");
438 l2fwd_usage(prgname);
445 l2fwd_rx_queue_per_lcore = l2fwd_parse_nqueue(optarg);
446 if (l2fwd_rx_queue_per_lcore == 0) {
447 printf("invalid queue number\n");
448 l2fwd_usage(prgname);
455 timer_secs = l2fwd_parse_timer_period(optarg);
456 if (timer_secs < 0) {
457 printf("invalid timer period\n");
458 l2fwd_usage(prgname);
461 timer_period = timer_secs;
469 l2fwd_usage(prgname);
475 argv[optind-1] = prgname;
478 optind = 1; /* reset getopt lib */
482 /* Check the link status of all ports in up to 9s, and print them finally */
484 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
486 #define CHECK_INTERVAL 100 /* 100ms */
487 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
488 uint8_t portid, count, all_ports_up, print_flag = 0;
489 struct rte_eth_link link;
491 printf("\nChecking link status");
493 for (count = 0; count <= MAX_CHECK_TIME; count++) {
497 for (portid = 0; portid < port_num; portid++) {
500 if ((port_mask & (1 << portid)) == 0)
502 memset(&link, 0, sizeof(link));
503 rte_eth_link_get_nowait(portid, &link);
504 /* print link status if flag set */
505 if (print_flag == 1) {
506 if (link.link_status)
507 printf("Port %d Link Up - speed %u "
508 "Mbps - %s\n", (uint8_t)portid,
509 (unsigned)link.link_speed,
510 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
511 ("full-duplex") : ("half-duplex\n"));
513 printf("Port %d Link Down\n",
517 /* clear all_ports_up flag if any link down */
518 if (link.link_status == ETH_LINK_DOWN) {
523 /* after finally printing all link status, get out */
527 if (all_ports_up == 0) {
530 rte_delay_ms(CHECK_INTERVAL);
533 /* set the print_flag if all ports up or timeout */
534 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
542 signal_handler(int signum)
544 if (signum == SIGINT || signum == SIGTERM) {
545 printf("\n\nSignal %d received, preparing to exit...\n",
552 main(int argc, char **argv)
554 struct lcore_queue_conf *qconf;
555 struct rte_eth_dev_info dev_info;
558 uint8_t nb_ports_available;
559 uint8_t portid, last_port;
560 unsigned lcore_id, rx_lcore_id;
561 unsigned nb_ports_in_mask = 0;
564 ret = rte_eal_init(argc, argv);
566 rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
571 signal(SIGINT, signal_handler);
572 signal(SIGTERM, signal_handler);
574 /* parse application arguments (after the EAL ones) */
575 ret = l2fwd_parse_args(argc, argv);
577 rte_exit(EXIT_FAILURE, "Invalid L2FWD arguments\n");
579 printf("MAC updating %s\n", mac_updating ? "enabled" : "disabled");
581 /* convert to number of cycles */
582 timer_period *= rte_get_timer_hz();
584 /* create the mbuf pool */
585 l2fwd_pktmbuf_pool = rte_pktmbuf_pool_create("mbuf_pool", NB_MBUF,
586 MEMPOOL_CACHE_SIZE, 0, RTE_MBUF_DEFAULT_BUF_SIZE,
588 if (l2fwd_pktmbuf_pool == NULL)
589 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool\n");
591 nb_ports = rte_eth_dev_count();
593 rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
595 /* reset l2fwd_dst_ports */
596 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++)
597 l2fwd_dst_ports[portid] = 0;
601 * Each logical core is assigned a dedicated TX queue on each port.
603 for (portid = 0; portid < nb_ports; portid++) {
604 /* skip ports that are not enabled */
605 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
608 if (nb_ports_in_mask % 2) {
609 l2fwd_dst_ports[portid] = last_port;
610 l2fwd_dst_ports[last_port] = portid;
617 rte_eth_dev_info_get(portid, &dev_info);
619 if (nb_ports_in_mask % 2) {
620 printf("Notice: odd number of ports in portmask.\n");
621 l2fwd_dst_ports[last_port] = last_port;
627 /* Initialize the port/queue configuration of each logical core */
628 for (portid = 0; portid < nb_ports; portid++) {
629 /* skip ports that are not enabled */
630 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
633 /* get the lcore_id for this port */
634 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
635 lcore_queue_conf[rx_lcore_id].n_rx_port ==
636 l2fwd_rx_queue_per_lcore) {
638 if (rx_lcore_id >= RTE_MAX_LCORE)
639 rte_exit(EXIT_FAILURE, "Not enough cores\n");
642 if (qconf != &lcore_queue_conf[rx_lcore_id])
643 /* Assigned a new logical core in the loop above. */
644 qconf = &lcore_queue_conf[rx_lcore_id];
646 qconf->rx_port_list[qconf->n_rx_port] = portid;
648 printf("Lcore %u: RX port %u\n", rx_lcore_id, (unsigned) portid);
651 nb_ports_available = nb_ports;
653 /* Initialise each port */
654 for (portid = 0; portid < nb_ports; portid++) {
655 /* skip ports that are not enabled */
656 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0) {
657 printf("Skipping disabled port %u\n", (unsigned) portid);
658 nb_ports_available--;
662 printf("Initializing port %u... ", (unsigned) portid);
664 ret = rte_eth_dev_configure(portid, 1, 1, &port_conf);
666 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
667 ret, (unsigned) portid);
669 rte_eth_macaddr_get(portid,&l2fwd_ports_eth_addr[portid]);
671 /* init one RX queue */
673 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
674 rte_eth_dev_socket_id(portid),
678 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup:err=%d, port=%u\n",
679 ret, (unsigned) portid);
681 /* init one TX queue on each port */
683 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
684 rte_eth_dev_socket_id(portid),
687 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup:err=%d, port=%u\n",
688 ret, (unsigned) portid);
690 /* Initialize TX buffers */
691 tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
692 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
693 rte_eth_dev_socket_id(portid));
694 if (tx_buffer[portid] == NULL)
695 rte_exit(EXIT_FAILURE, "Cannot allocate buffer for tx on port %u\n",
698 rte_eth_tx_buffer_init(tx_buffer[portid], MAX_PKT_BURST);
700 ret = rte_eth_tx_buffer_set_err_callback(tx_buffer[portid],
701 rte_eth_tx_buffer_count_callback,
702 &port_statistics[portid].dropped);
704 rte_exit(EXIT_FAILURE, "Cannot set error callback for "
705 "tx buffer on port %u\n", (unsigned) portid);
708 ret = rte_eth_dev_start(portid);
710 rte_exit(EXIT_FAILURE, "rte_eth_dev_start:err=%d, port=%u\n",
711 ret, (unsigned) portid);
715 rte_eth_promiscuous_enable(portid);
717 printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
719 l2fwd_ports_eth_addr[portid].addr_bytes[0],
720 l2fwd_ports_eth_addr[portid].addr_bytes[1],
721 l2fwd_ports_eth_addr[portid].addr_bytes[2],
722 l2fwd_ports_eth_addr[portid].addr_bytes[3],
723 l2fwd_ports_eth_addr[portid].addr_bytes[4],
724 l2fwd_ports_eth_addr[portid].addr_bytes[5]);
726 /* initialize port stats */
727 memset(&port_statistics, 0, sizeof(port_statistics));
730 if (!nb_ports_available) {
731 rte_exit(EXIT_FAILURE,
732 "All available ports are disabled. Please set portmask.\n");
735 check_all_ports_link_status(nb_ports, l2fwd_enabled_port_mask);
738 /* launch per-lcore init on every lcore */
739 rte_eal_mp_remote_launch(l2fwd_launch_one_lcore, NULL, CALL_MASTER);
740 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
741 if (rte_eal_wait_lcore(lcore_id) < 0) {
747 for (portid = 0; portid < nb_ports; portid++) {
748 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
750 printf("Closing port %d...", portid);
751 rte_eth_dev_stop(portid);
752 rte_eth_dev_close(portid);