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.
41 #include <sys/types.h>
42 #include <sys/queue.h>
43 #include <netinet/in.h>
50 #include <rte_common.h>
52 #include <rte_memory.h>
53 #include <rte_memcpy.h>
54 #include <rte_memzone.h>
56 #include <rte_launch.h>
57 #include <rte_atomic.h>
58 #include <rte_spinlock.h>
59 #include <rte_cycles.h>
60 #include <rte_prefetch.h>
61 #include <rte_lcore.h>
62 #include <rte_per_lcore.h>
63 #include <rte_branch_prediction.h>
64 #include <rte_interrupts.h>
66 #include <rte_random.h>
67 #include <rte_debug.h>
68 #include <rte_ether.h>
69 #include <rte_ethdev.h>
71 #include <rte_mempool.h>
73 #include <rte_malloc.h>
77 #define RTE_LOGTYPE_L2FWD RTE_LOGTYPE_USER1
78 #define MBUF_NAME "mbuf_pool_%d"
79 #define MBUF_DATA_SIZE RTE_MBUF_DEFAULT_BUF_SIZE
81 #define RING_MASTER_NAME "l2fwd_ring_m2s_"
82 #define RING_SLAVE_NAME "l2fwd_ring_s2m_"
83 #define MAX_NAME_LEN 32
84 /* RECREATE flag indicate needs initialize resource and launch slave_core again */
85 #define SLAVE_RECREATE_FLAG 0x1
86 /* RESTART flag indicate needs restart port and send START command again */
87 #define SLAVE_RESTART_FLAG 0x2
88 #define INVALID_MAPPING_ID ((unsigned)LCORE_ID_ANY)
89 /* Maximum message buffer per slave */
90 #define NB_CORE_MSGBUF 32
96 #define MAX_PKT_BURST 32
97 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
100 * Configurable number of RX/TX ring descriptors
102 #define RTE_TEST_RX_DESC_DEFAULT 128
103 #define RTE_TEST_TX_DESC_DEFAULT 512
104 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
105 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
107 /* ethernet addresses of ports */
108 static struct ether_addr l2fwd_ports_eth_addr[RTE_MAX_ETHPORTS];
110 /* mask of enabled ports */
111 static uint32_t l2fwd_enabled_port_mask = 0;
113 /* list of enabled ports */
114 static uint32_t l2fwd_dst_ports[RTE_MAX_ETHPORTS];
116 static unsigned int l2fwd_rx_queue_per_lcore = 1;
120 struct rte_mbuf *m_table[MAX_PKT_BURST];
123 #define MAX_RX_QUEUE_PER_LCORE 16
124 #define MAX_TX_QUEUE_PER_PORT 16
125 struct lcore_queue_conf {
127 unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
128 } __rte_cache_aligned;
129 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
131 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
133 struct lcore_resource_struct {
134 int enabled; /* Only set in case this lcore involved into packet forwarding */
135 int flags; /* Set only slave need to restart or recreate */
136 unsigned lcore_id; /* lcore ID */
137 unsigned pair_id; /* dependency lcore ID on port */
138 char ring_name[2][MAX_NAME_LEN];
139 /* ring[0] for master send cmd, slave read */
140 /* ring[1] for slave send ack, master read */
141 struct rte_ring *ring[2];
142 int port_num; /* Total port numbers */
143 uint8_t port[RTE_MAX_ETHPORTS]; /* Port id for that lcore to receive packets */
144 }__attribute__((packed)) __rte_cache_aligned;
146 static struct lcore_resource_struct lcore_resource[RTE_MAX_LCORE];
147 static struct rte_mempool *message_pool;
148 static rte_spinlock_t res_lock = RTE_SPINLOCK_INITIALIZER;
149 /* use floating processes */
150 static int float_proc = 0;
151 /* Save original cpu affinity */
157 static const struct rte_eth_conf port_conf = {
160 .header_split = 0, /**< Header Split disabled */
161 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
162 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
163 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
164 .hw_strip_crc = 1, /**< CRC stripped by hardware */
167 .mq_mode = ETH_MQ_TX_NONE,
171 static struct rte_mempool * l2fwd_pktmbuf_pool[RTE_MAX_ETHPORTS];
173 /* Per-port statistics struct */
174 struct l2fwd_port_statistics {
178 } __rte_cache_aligned;
179 struct l2fwd_port_statistics *port_statistics;
181 * pointer to lcore ID mapping array, used to return lcore id in case slave
182 * process exited unexpectedly, use only floating process option applied
184 unsigned *mapping_id;
186 /* A tsc-based timer responsible for triggering statistics printout */
187 #define TIMER_MILLISECOND 2000000ULL /* around 1ms at 2 Ghz */
188 #define MAX_TIMER_PERIOD 86400 /* 1 day max */
189 static int64_t timer_period = 10 * TIMER_MILLISECOND * 1000; /* default period is 10 seconds */
191 static int l2fwd_launch_one_lcore(void *dummy);
193 /* Print out statistics on packets dropped */
197 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
200 total_packets_dropped = 0;
201 total_packets_tx = 0;
202 total_packets_rx = 0;
204 const char clr[] = { 27, '[', '2', 'J', '\0' };
205 const char topLeft[] = { 27, '[', '1', ';', '1', 'H','\0' };
207 /* Clear screen and move to top left */
208 printf("%s%s", clr, topLeft);
210 printf("\nPort statistics ====================================");
212 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
213 /* skip disabled ports */
214 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
216 printf("\nStatistics for port %u ------------------------------"
217 "\nPackets sent: %24"PRIu64
218 "\nPackets received: %20"PRIu64
219 "\nPackets dropped: %21"PRIu64,
221 port_statistics[portid].tx,
222 port_statistics[portid].rx,
223 port_statistics[portid].dropped);
225 total_packets_dropped += port_statistics[portid].dropped;
226 total_packets_tx += port_statistics[portid].tx;
227 total_packets_rx += port_statistics[portid].rx;
229 printf("\nAggregate statistics ==============================="
230 "\nTotal packets sent: %18"PRIu64
231 "\nTotal packets received: %14"PRIu64
232 "\nTotal packets dropped: %15"PRIu64,
235 total_packets_dropped);
236 printf("\n====================================================\n");
240 clear_cpu_affinity(void)
244 s = sched_setaffinity(0, cpu_aff.size, &cpu_aff.set);
246 printf("sched_setaffinity failed:%s\n", strerror(errno));
254 get_cpu_affinity(void)
258 cpu_aff.size = sizeof(cpu_set_t);
259 CPU_ZERO(&cpu_aff.set);
261 s = sched_getaffinity(0, cpu_aff.size, &cpu_aff.set);
263 printf("sched_getaffinity failed:%s\n", strerror(errno));
271 * This fnciton demonstrates the approach to create ring in first instance
272 * or re-attach an existed ring in later instance.
274 static struct rte_ring *
275 create_ring(const char *name, unsigned count,
276 int socket_id,unsigned flags)
278 struct rte_ring *ring;
283 /* If already create, just attached it */
284 if (likely((ring = rte_ring_lookup(name)) != NULL))
287 /* First call it, create one */
288 return rte_ring_create(name, count, socket_id, flags);
291 /* Malloc with rte_malloc on structures that shared by master and slave */
293 l2fwd_malloc_shared_struct(void)
295 port_statistics = rte_zmalloc("port_stat",
296 sizeof(struct l2fwd_port_statistics) * RTE_MAX_ETHPORTS,
298 if (port_statistics == NULL)
301 /* allocate mapping_id array */
304 mapping_id = rte_malloc("mapping_id", sizeof(unsigned) * RTE_MAX_LCORE,
307 if (mapping_id == NULL)
310 for (i = 0 ;i < RTE_MAX_LCORE; i++)
311 mapping_id[i] = INVALID_MAPPING_ID;
316 /* Create ring which used for communicate among master and slave */
318 create_ms_ring(unsigned slaveid)
320 unsigned flag = RING_F_SP_ENQ | RING_F_SC_DEQ;
321 struct lcore_resource_struct *res = &lcore_resource[slaveid];
322 unsigned socketid = rte_socket_id();
324 /* Always assume create ring on master socket_id */
325 /* Default only create a ring size 32 */
326 snprintf(res->ring_name[0], MAX_NAME_LEN, "%s%u",
327 RING_MASTER_NAME, slaveid);
328 if ((res->ring[0] = create_ring(res->ring_name[0], NB_CORE_MSGBUF,
329 socketid, flag)) == NULL) {
330 printf("Create m2s ring %s failed\n", res->ring_name[0]);
334 snprintf(res->ring_name[1], MAX_NAME_LEN, "%s%u",
335 RING_SLAVE_NAME, slaveid);
336 if ((res->ring[1] = create_ring(res->ring_name[1], NB_CORE_MSGBUF,
337 socketid, flag)) == NULL) {
338 printf("Create s2m ring %s failed\n", res->ring_name[1]);
345 /* send command to pair in paired master and slave ring */
347 sendcmd(unsigned slaveid, enum l2fwd_cmd cmd, int is_master)
349 struct lcore_resource_struct *res = &lcore_resource[slaveid];
353 /* Only check master, it must be enabled and running if it is slave */
354 if (is_master && !res->enabled)
357 if (res->ring[fd] == NULL)
360 if (rte_mempool_get(message_pool, &msg) < 0) {
361 printf("Error to get message buffer\n");
365 *(enum l2fwd_cmd *)msg = cmd;
367 if (rte_ring_enqueue(res->ring[fd], msg) != 0) {
368 printf("Enqueue error\n");
369 rte_mempool_put(message_pool, msg);
376 /* Get command from pair in paired master and slave ring */
378 getcmd(unsigned slaveid, enum l2fwd_cmd *cmd, int is_master)
380 struct lcore_resource_struct *res = &lcore_resource[slaveid];
382 int fd = !!is_master;
384 /* Only check master, it must be enabled and running if it is slave */
385 if (is_master && (!res->enabled))
388 if (res->ring[fd] == NULL)
391 ret = rte_ring_dequeue(res->ring[fd], &msg);
394 *cmd = *(enum l2fwd_cmd *)msg;
395 rte_mempool_put(message_pool, msg);
400 /* Master send command to slave and wait until ack received or error met */
402 master_sendcmd_with_ack(unsigned slaveid, enum l2fwd_cmd cmd)
404 enum l2fwd_cmd ack_cmd;
407 if (sendcmd(slaveid, cmd, 1) != 0)
408 rte_exit(EXIT_FAILURE, "Failed to send message\n");
412 ret = getcmd(slaveid, &ack_cmd, 1);
413 if (ret == 0 && cmd == ack_cmd)
416 /* If slave not running yet, return an error */
417 if (flib_query_slave_status(slaveid) != ST_RUN) {
426 /* restart all port that assigned to that slave lcore */
428 reset_slave_all_ports(unsigned slaveid)
430 struct lcore_resource_struct *slave = &lcore_resource[slaveid];
433 /* stop/start port */
434 for (i = 0; i < slave->port_num; i++) {
435 char buf_name[RTE_MEMPOOL_NAMESIZE];
436 struct rte_mempool *pool;
437 printf("Stop port :%d\n", slave->port[i]);
438 rte_eth_dev_stop(slave->port[i]);
439 snprintf(buf_name, RTE_MEMPOOL_NAMESIZE, MBUF_NAME, slave->port[i]);
440 pool = rte_mempool_lookup(buf_name);
442 printf("Port %d mempool free object is %u(%u)\n", slave->port[i],
443 rte_mempool_avail_count(pool),
444 (unsigned int)NB_MBUF);
446 printf("Can't find mempool %s\n", buf_name);
448 printf("Start port :%d\n", slave->port[i]);
449 ret = rte_eth_dev_start(slave->port[i]);
457 reset_shared_structures(unsigned slaveid)
460 /* Only port are shared resource here */
461 ret = reset_slave_all_ports(slaveid);
467 * Call this function to re-create resource that needed for slave process that
468 * exited in last instance
471 init_slave_res(unsigned slaveid)
473 struct lcore_resource_struct *slave = &lcore_resource[slaveid];
476 if (!slave->enabled) {
477 printf("Something wrong with lcore=%u enabled=%d\n",slaveid,
482 /* Initialize ring */
483 if (create_ms_ring(slaveid) != 0)
484 rte_exit(EXIT_FAILURE, "failed to create ring for slave %u\n",
487 /* drain un-read buffer if have */
488 while (getcmd(slaveid, &cmd, 1) == 0);
489 while (getcmd(slaveid, &cmd, 0) == 0);
495 recreate_one_slave(unsigned slaveid)
498 /* Re-initialize resource for stalled slave */
499 if ((ret = init_slave_res(slaveid)) != 0) {
500 printf("Init slave=%u failed\n", slaveid);
504 if ((ret = flib_remote_launch(l2fwd_launch_one_lcore, NULL, slaveid))
506 printf("Launch slave %u failed\n", slaveid);
512 * remapping resource belong to slave_id to new lcore that gets from flib_assign_lcore_id(),
513 * used only floating process option applied.
516 * original lcore_id that apply for remapping
519 remapping_slave_resource(unsigned slaveid, unsigned map_id)
522 /* remapping lcore_resource */
523 memcpy(&lcore_resource[map_id], &lcore_resource[slaveid],
524 sizeof(struct lcore_resource_struct));
526 /* remapping lcore_queue_conf */
527 memcpy(&lcore_queue_conf[map_id], &lcore_queue_conf[slaveid],
528 sizeof(struct lcore_queue_conf));
532 reset_pair(unsigned slaveid, unsigned pairid)
535 if ((ret = reset_shared_structures(slaveid)) != 0)
538 if((ret = reset_shared_structures(pairid)) != 0)
542 unsigned map_id = mapping_id[slaveid];
544 if (map_id != INVALID_MAPPING_ID) {
545 printf("%u return mapping id %u\n", slaveid, map_id);
546 flib_free_lcore_id(map_id);
547 mapping_id[slaveid] = INVALID_MAPPING_ID;
550 map_id = mapping_id[pairid];
551 if (map_id != INVALID_MAPPING_ID) {
552 printf("%u return mapping id %u\n", pairid, map_id);
553 flib_free_lcore_id(map_id);
554 mapping_id[pairid] = INVALID_MAPPING_ID;
558 if((ret = recreate_one_slave(slaveid)) != 0)
561 ret = recreate_one_slave(pairid);
568 slave_exit_cb(unsigned slaveid, __attribute__((unused))int stat)
570 struct lcore_resource_struct *slave = &lcore_resource[slaveid];
572 printf("Get slave %u leave info\n", slaveid);
573 if (!slave->enabled) {
574 printf("Lcore=%u not registered for it's exit\n", slaveid);
577 rte_spinlock_lock(&res_lock);
579 /* Change the state and wait master to start them */
580 slave->flags = SLAVE_RECREATE_FLAG;
582 rte_spinlock_unlock(&res_lock);
586 l2fwd_simple_forward(struct rte_mbuf *m, unsigned portid)
588 struct ether_hdr *eth;
592 struct rte_eth_dev_tx_buffer *buffer;
594 dst_port = l2fwd_dst_ports[portid];
595 eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
597 /* 02:00:00:00:00:xx */
598 tmp = ð->d_addr.addr_bytes[0];
599 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
602 ether_addr_copy(&l2fwd_ports_eth_addr[dst_port], ð->s_addr);
604 buffer = tx_buffer[dst_port];
605 sent = rte_eth_tx_buffer(dst_port, 0, buffer, m);
607 port_statistics[dst_port].tx += sent;
610 /* main processing loop */
612 l2fwd_main_loop(void)
614 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
618 uint64_t prev_tsc, diff_tsc, cur_tsc;
619 unsigned i, j, portid, nb_rx;
620 struct lcore_queue_conf *qconf;
621 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S *
623 struct rte_eth_dev_tx_buffer *buffer;
627 lcore_id = rte_lcore_id();
629 qconf = &lcore_queue_conf[lcore_id];
631 if (qconf->n_rx_port == 0) {
632 RTE_LOG(INFO, L2FWD, "lcore %u has nothing to do\n", lcore_id);
636 RTE_LOG(INFO, L2FWD, "entering main loop on lcore %u\n", lcore_id);
638 for (i = 0; i < qconf->n_rx_port; i++) {
639 portid = qconf->rx_port_list[i];
640 RTE_LOG(INFO, L2FWD, " -- lcoreid=%u portid=%u\n", lcore_id,
646 cur_tsc = rte_rdtsc();
648 if (unlikely(getcmd(lcore_id, &cmd, 0) == 0)) {
649 sendcmd(lcore_id, cmd, 0);
651 /* If get stop command, stop forwarding and exit */
652 if (cmd == CMD_STOP) {
658 * TX burst queue drain
660 diff_tsc = cur_tsc - prev_tsc;
661 if (unlikely(diff_tsc > drain_tsc)) {
663 for (i = 0; i < qconf->n_rx_port; i++) {
665 portid = l2fwd_dst_ports[qconf->rx_port_list[i]];
666 buffer = tx_buffer[portid];
668 sent = rte_eth_tx_buffer_flush(portid, 0, buffer);
670 port_statistics[portid].tx += sent;
678 * Read packet from RX queues
680 for (i = 0; i < qconf->n_rx_port; i++) {
682 portid = qconf->rx_port_list[i];
683 nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
684 pkts_burst, MAX_PKT_BURST);
686 port_statistics[portid].rx += nb_rx;
688 for (j = 0; j < nb_rx; j++) {
690 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
691 l2fwd_simple_forward(m, portid);
698 l2fwd_launch_one_lcore(__attribute__((unused)) void *dummy)
700 unsigned lcore_id = rte_lcore_id();
705 /* Change it to floating process, also change it's lcore_id */
706 clear_cpu_affinity();
707 RTE_PER_LCORE(_lcore_id) = 0;
709 if (flib_assign_lcore_id() < 0 ) {
710 printf("flib_assign_lcore_id failed\n");
713 flcore_id = rte_lcore_id();
714 /* Set mapping id, so master can return it after slave exited */
715 mapping_id[lcore_id] = flcore_id;
716 printf("Org lcore_id = %u, cur lcore_id = %u\n",
717 lcore_id, flcore_id);
718 remapping_slave_resource(lcore_id, flcore_id);
723 /* return lcore_id before return */
725 flib_free_lcore_id(rte_lcore_id());
726 mapping_id[lcore_id] = INVALID_MAPPING_ID;
733 l2fwd_usage(const char *prgname)
735 printf("%s [EAL options] -- -p PORTMASK -s COREMASK [-q NQ] -f\n"
736 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
737 " -q NQ: number of queue (=ports) per lcore (default is 1)\n"
738 " -f use floating process which won't bind to any core to run\n"
739 " -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n",
744 l2fwd_parse_portmask(const char *portmask)
749 /* parse hexadecimal string */
750 pm = strtoul(portmask, &end, 16);
751 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
761 l2fwd_parse_nqueue(const char *q_arg)
766 /* parse hexadecimal string */
767 n = strtoul(q_arg, &end, 10);
768 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
772 if (n >= MAX_RX_QUEUE_PER_LCORE)
779 l2fwd_parse_timer_period(const char *q_arg)
784 /* parse number string */
785 n = strtol(q_arg, &end, 10);
786 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
788 if (n >= MAX_TIMER_PERIOD)
794 /* Parse the argument given in the command line of the application */
796 l2fwd_parse_args(int argc, char **argv)
801 char *prgname = argv[0];
802 static struct option lgopts[] = {
809 while ((opt = getopt_long(argc, argvopt, "p:q:T:f",
810 lgopts, &option_index)) != EOF) {
815 l2fwd_enabled_port_mask = l2fwd_parse_portmask(optarg);
816 if (l2fwd_enabled_port_mask == 0) {
817 printf("invalid portmask\n");
818 l2fwd_usage(prgname);
826 l2fwd_rx_queue_per_lcore = l2fwd_parse_nqueue(optarg);
827 if (l2fwd_rx_queue_per_lcore == 0) {
828 printf("invalid queue number\n");
829 l2fwd_usage(prgname);
836 timer_period = l2fwd_parse_timer_period(optarg) * 1000 * TIMER_MILLISECOND;
837 if (timer_period < 0) {
838 printf("invalid timer period\n");
839 l2fwd_usage(prgname);
844 /* use floating process */
851 l2fwd_usage(prgname);
855 l2fwd_usage(prgname);
861 argv[optind-1] = prgname;
864 l2fwd_usage(prgname);
868 optind = 1; /* reset getopt lib */
872 /* Check the link status of all ports in up to 9s, and print them finally */
874 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
876 #define CHECK_INTERVAL 100 /* 100ms */
877 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
878 uint8_t portid, count, all_ports_up, print_flag = 0;
879 struct rte_eth_link link;
881 printf("\nChecking link status");
883 for (count = 0; count <= MAX_CHECK_TIME; count++) {
885 for (portid = 0; portid < port_num; portid++) {
886 if ((port_mask & (1 << portid)) == 0)
888 memset(&link, 0, sizeof(link));
889 rte_eth_link_get_nowait(portid, &link);
890 /* print link status if flag set */
891 if (print_flag == 1) {
892 if (link.link_status)
893 printf("Port %d Link Up - speed %u "
894 "Mbps - %s\n", (uint8_t)portid,
895 (unsigned)link.link_speed,
896 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
897 ("full-duplex") : ("half-duplex\n"));
899 printf("Port %d Link Down\n",
903 /* clear all_ports_up flag if any link down */
904 if (link.link_status == ETH_LINK_DOWN) {
909 /* after finally printing all link status, get out */
913 if (all_ports_up == 0) {
916 rte_delay_ms(CHECK_INTERVAL);
919 /* set the print_flag if all ports up or timeout */
920 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
928 main(int argc, char **argv)
930 struct lcore_queue_conf *qconf;
931 struct rte_eth_dev_info dev_info;
934 uint8_t nb_ports_available;
935 uint8_t portid, last_port;
936 unsigned rx_lcore_id;
937 unsigned nb_ports_in_mask = 0;
939 uint64_t prev_tsc, diff_tsc, cur_tsc, timer_tsc;
941 /* Save cpu_affinity first, restore it in case it's floating process option */
942 if (get_cpu_affinity() != 0)
943 rte_exit(EXIT_FAILURE, "get_cpu_affinity error\n");
945 /* Also tries to set cpu affinity to detect whether it will fail in child process */
946 if(clear_cpu_affinity() != 0)
947 rte_exit(EXIT_FAILURE, "clear_cpu_affinity error\n");
950 ret = rte_eal_init(argc, argv);
952 rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
956 /* parse application arguments (after the EAL ones) */
957 ret = l2fwd_parse_args(argc, argv);
959 rte_exit(EXIT_FAILURE, "Invalid L2FWD arguments\n");
962 if (flib_init() != 0)
963 rte_exit(EXIT_FAILURE, "flib init error");
966 * Allocated structures that slave lcore would change. For those that slaves are
967 * read only, needn't use malloc to share and global or static variables is ok since
968 * slave inherit all the knowledge that master initialized.
970 if (l2fwd_malloc_shared_struct() != 0)
971 rte_exit(EXIT_FAILURE, "malloc mem failed\n");
973 /* Initialize lcore_resource structures */
974 memset(lcore_resource, 0, sizeof(lcore_resource));
975 for (i = 0; i < RTE_MAX_LCORE; i++)
976 lcore_resource[i].lcore_id = i;
978 nb_ports = rte_eth_dev_count();
980 rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
982 /* create the mbuf pool */
983 for (portid = 0; portid < nb_ports; portid++) {
984 /* skip ports that are not enabled */
985 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
987 char buf_name[RTE_MEMPOOL_NAMESIZE];
988 snprintf(buf_name, RTE_MEMPOOL_NAMESIZE, MBUF_NAME, portid);
989 l2fwd_pktmbuf_pool[portid] =
990 rte_pktmbuf_pool_create(buf_name, NB_MBUF, 32,
991 0, MBUF_DATA_SIZE, rte_socket_id());
992 if (l2fwd_pktmbuf_pool[portid] == NULL)
993 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
995 printf("Create mbuf %s\n", buf_name);
998 /* reset l2fwd_dst_ports */
999 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++)
1000 l2fwd_dst_ports[portid] = 0;
1004 * Each logical core is assigned a dedicated TX queue on each port.
1006 for (portid = 0; portid < nb_ports; portid++) {
1007 /* skip ports that are not enabled */
1008 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
1011 if (nb_ports_in_mask % 2) {
1012 l2fwd_dst_ports[portid] = last_port;
1013 l2fwd_dst_ports[last_port] = portid;
1020 rte_eth_dev_info_get(portid, &dev_info);
1022 if (nb_ports_in_mask % 2) {
1023 printf("Notice: odd number of ports in portmask.\n");
1024 l2fwd_dst_ports[last_port] = last_port;
1030 /* Initialize the port/queue configuration of each logical core */
1031 for (portid = 0; portid < nb_ports; portid++) {
1032 struct lcore_resource_struct *res;
1033 /* skip ports that are not enabled */
1034 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
1037 /* get the lcore_id for this port */
1038 /* skip master lcore */
1039 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
1040 rte_get_master_lcore() == rx_lcore_id ||
1041 lcore_queue_conf[rx_lcore_id].n_rx_port ==
1042 l2fwd_rx_queue_per_lcore) {
1045 if (rx_lcore_id >= RTE_MAX_LCORE)
1046 rte_exit(EXIT_FAILURE, "Not enough cores\n");
1049 if (qconf != &lcore_queue_conf[rx_lcore_id])
1050 /* Assigned a new logical core in the loop above. */
1051 qconf = &lcore_queue_conf[rx_lcore_id];
1053 qconf->rx_port_list[qconf->n_rx_port] = portid;
1056 /* Save the port resource info into lcore_resource strucutres */
1057 res = &lcore_resource[rx_lcore_id];
1059 res->port[res->port_num++] = portid;
1061 printf("Lcore %u: RX port %u\n", rx_lcore_id, (unsigned) portid);
1064 nb_ports_available = nb_ports;
1066 /* Initialise each port */
1067 for (portid = 0; portid < nb_ports; portid++) {
1068 /* skip ports that are not enabled */
1069 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0) {
1070 printf("Skipping disabled port %u\n", (unsigned) portid);
1071 nb_ports_available--;
1075 printf("Initializing port %u... ", (unsigned) portid);
1077 ret = rte_eth_dev_configure(portid, 1, 1, &port_conf);
1079 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
1080 ret, (unsigned) portid);
1082 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1085 rte_exit(EXIT_FAILURE,
1086 "rte_eth_dev_adjust_nb_rx_tx_desc: err=%d, port=%u\n",
1087 ret, (unsigned) portid);
1089 rte_eth_macaddr_get(portid,&l2fwd_ports_eth_addr[portid]);
1091 /* init one RX queue */
1093 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
1094 rte_eth_dev_socket_id(portid),
1096 l2fwd_pktmbuf_pool[portid]);
1098 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup:err=%d, port=%u\n",
1099 ret, (unsigned) portid);
1101 /* init one TX queue on each port */
1103 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
1104 rte_eth_dev_socket_id(portid),
1107 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup:err=%d, port=%u\n",
1108 ret, (unsigned) portid);
1110 /* Initialize TX buffers */
1111 tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
1112 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
1113 rte_eth_dev_socket_id(portid));
1114 if (tx_buffer[portid] == NULL)
1115 rte_exit(EXIT_FAILURE, "Cannot allocate buffer for tx on port %u\n",
1118 rte_eth_tx_buffer_init(tx_buffer[portid], MAX_PKT_BURST);
1120 ret = rte_eth_tx_buffer_set_err_callback(tx_buffer[portid],
1121 rte_eth_tx_buffer_count_callback,
1122 &port_statistics[portid].dropped);
1124 rte_exit(EXIT_FAILURE, "Cannot set error callback for "
1125 "tx buffer on port %u\n", (unsigned) portid);
1128 ret = rte_eth_dev_start(portid);
1130 rte_exit(EXIT_FAILURE, "rte_eth_dev_start:err=%d, port=%u\n",
1131 ret, (unsigned) portid);
1135 rte_eth_promiscuous_enable(portid);
1137 printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
1139 l2fwd_ports_eth_addr[portid].addr_bytes[0],
1140 l2fwd_ports_eth_addr[portid].addr_bytes[1],
1141 l2fwd_ports_eth_addr[portid].addr_bytes[2],
1142 l2fwd_ports_eth_addr[portid].addr_bytes[3],
1143 l2fwd_ports_eth_addr[portid].addr_bytes[4],
1144 l2fwd_ports_eth_addr[portid].addr_bytes[5]);
1146 /* initialize port stats */
1147 //memset(&port_statistics, 0, sizeof(port_statistics));
1150 if (!nb_ports_available) {
1151 rte_exit(EXIT_FAILURE,
1152 "All available ports are disabled. Please set portmask.\n");
1155 check_all_ports_link_status(nb_ports, l2fwd_enabled_port_mask);
1157 /* Record pair lcore */
1159 * Since l2fwd example would create pair between different neighbour port, that's
1160 * port 0 receive and forward to port 1, the same to port 1, these 2 ports will have
1161 * dependency. If one port stopped working (killed, for example), the port need to
1162 * be stopped/started again. During the time, another port need to wait until stop/start
1163 * procedure completed. So, record the pair relationship for those lcores working
1166 for (portid = 0; portid < nb_ports; portid++) {
1168 unsigned lcore = 0, pair_lcore = 0;
1169 unsigned j, find_lcore, find_pair_lcore;
1170 /* skip ports that are not enabled */
1171 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
1174 /* Find pair ports' lcores */
1175 find_lcore = find_pair_lcore = 0;
1176 pair_port = l2fwd_dst_ports[portid];
1177 for (i = 0; i < RTE_MAX_LCORE; i++) {
1178 if (!rte_lcore_is_enabled(i))
1180 for (j = 0; j < lcore_queue_conf[i].n_rx_port;j++) {
1181 if (lcore_queue_conf[i].rx_port_list[j] == portid) {
1186 if (lcore_queue_conf[i].rx_port_list[j] == pair_port) {
1188 find_pair_lcore = 1;
1192 if (find_lcore && find_pair_lcore)
1195 if (!find_lcore || !find_pair_lcore)
1196 rte_exit(EXIT_FAILURE, "Not find port=%d pair\n", portid);
1198 printf("lcore %u and %u paired\n", lcore, pair_lcore);
1199 lcore_resource[lcore].pair_id = pair_lcore;
1200 lcore_resource[pair_lcore].pair_id = lcore;
1203 /* Create message buffer for all master and slave */
1204 message_pool = rte_mempool_create("ms_msg_pool",
1205 NB_CORE_MSGBUF * RTE_MAX_LCORE,
1206 sizeof(enum l2fwd_cmd), NB_CORE_MSGBUF / 2,
1208 rte_pktmbuf_pool_init, NULL,
1209 rte_pktmbuf_init, NULL,
1210 rte_socket_id(), 0);
1212 if (message_pool == NULL)
1213 rte_exit(EXIT_FAILURE, "Create msg mempool failed\n");
1215 /* Create ring for each master and slave pair, also register cb when slave leaves */
1216 for (i = 0; i < RTE_MAX_LCORE; i++) {
1218 * Only create ring and register slave_exit cb in case that core involved into
1221 if (lcore_resource[i].enabled) {
1222 /* Create ring for master and slave communication */
1223 ret = create_ms_ring(i);
1225 rte_exit(EXIT_FAILURE, "Create ring for lcore=%u failed",
1228 if (flib_register_slave_exit_notify(i,
1229 slave_exit_cb) != 0)
1230 rte_exit(EXIT_FAILURE,
1231 "Register master_trace_slave_exit failed");
1235 /* launch per-lcore init on every lcore except master */
1236 flib_mp_remote_launch(l2fwd_launch_one_lcore, NULL, SKIP_MASTER);
1238 /* print statistics 10 second */
1239 prev_tsc = cur_tsc = rte_rdtsc();
1243 cur_tsc = rte_rdtsc();
1244 diff_tsc = cur_tsc - prev_tsc;
1245 /* if timer is enabled */
1246 if (timer_period > 0) {
1248 /* advance the timer */
1249 timer_tsc += diff_tsc;
1251 /* if timer has reached its timeout */
1252 if (unlikely(timer_tsc >= (uint64_t) timer_period)) {
1255 /* reset the timer */
1262 /* Check any slave need restart or recreate */
1263 rte_spinlock_lock(&res_lock);
1264 for (i = 0; i < RTE_MAX_LCORE; i++) {
1265 struct lcore_resource_struct *res = &lcore_resource[i];
1266 struct lcore_resource_struct *pair = &lcore_resource[res->pair_id];
1268 /* If find slave exited, try to reset pair */
1269 if (res->enabled && res->flags && pair->enabled) {
1271 master_sendcmd_with_ack(pair->lcore_id, CMD_STOP);
1272 rte_spinlock_unlock(&res_lock);
1274 rte_spinlock_lock(&res_lock);
1278 if (reset_pair(res->lcore_id, pair->lcore_id) != 0)
1279 rte_exit(EXIT_FAILURE, "failed to reset slave");
1284 rte_spinlock_unlock(&res_lock);