1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
12 #include <sys/types.h>
13 #include <sys/queue.h>
14 #include <netinet/in.h>
21 #include <rte_common.h>
23 #include <rte_memory.h>
24 #include <rte_memcpy.h>
26 #include <rte_launch.h>
27 #include <rte_atomic.h>
28 #include <rte_spinlock.h>
29 #include <rte_cycles.h>
30 #include <rte_prefetch.h>
31 #include <rte_lcore.h>
32 #include <rte_per_lcore.h>
33 #include <rte_branch_prediction.h>
34 #include <rte_interrupts.h>
35 #include <rte_random.h>
36 #include <rte_debug.h>
37 #include <rte_ether.h>
38 #include <rte_ethdev.h>
40 #include <rte_mempool.h>
42 #include <rte_malloc.h>
46 #define RTE_LOGTYPE_L2FWD RTE_LOGTYPE_USER1
47 #define MBUF_NAME "mbuf_pool_%d"
48 #define MBUF_DATA_SIZE RTE_MBUF_DEFAULT_BUF_SIZE
50 #define RING_MASTER_NAME "l2fwd_ring_m2s_"
51 #define RING_SLAVE_NAME "l2fwd_ring_s2m_"
52 #define MAX_NAME_LEN 32
53 /* RECREATE flag indicate needs initialize resource and launch slave_core again */
54 #define SLAVE_RECREATE_FLAG 0x1
55 /* RESTART flag indicate needs restart port and send START command again */
56 #define SLAVE_RESTART_FLAG 0x2
57 #define INVALID_MAPPING_ID ((unsigned)LCORE_ID_ANY)
58 /* Maximum message buffer per slave */
59 #define NB_CORE_MSGBUF 32
65 #define MAX_PKT_BURST 32
66 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
69 * Configurable number of RX/TX ring descriptors
71 #define RTE_TEST_RX_DESC_DEFAULT 128
72 #define RTE_TEST_TX_DESC_DEFAULT 512
73 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
74 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
76 /* ethernet addresses of ports */
77 static struct ether_addr l2fwd_ports_eth_addr[RTE_MAX_ETHPORTS];
79 /* mask of enabled ports */
80 static uint32_t l2fwd_enabled_port_mask = 0;
82 /* list of enabled ports */
83 static uint32_t l2fwd_dst_ports[RTE_MAX_ETHPORTS];
85 static unsigned int l2fwd_rx_queue_per_lcore = 1;
89 struct rte_mbuf *m_table[MAX_PKT_BURST];
92 #define MAX_RX_QUEUE_PER_LCORE 16
93 #define MAX_TX_QUEUE_PER_PORT 16
94 struct lcore_queue_conf {
96 unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
97 } __rte_cache_aligned;
98 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
100 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
102 struct lcore_resource_struct {
103 int enabled; /* Only set in case this lcore involved into packet forwarding */
104 int flags; /* Set only slave need to restart or recreate */
105 unsigned lcore_id; /* lcore ID */
106 unsigned pair_id; /* dependency lcore ID on port */
107 char ring_name[2][MAX_NAME_LEN];
108 /* ring[0] for master send cmd, slave read */
109 /* ring[1] for slave send ack, master read */
110 struct rte_ring *ring[2];
111 int port_num; /* Total port numbers */
112 /* Port id for that lcore to receive packets */
113 uint16_t port[RTE_MAX_ETHPORTS];
114 }__attribute__((packed)) __rte_cache_aligned;
116 static struct lcore_resource_struct lcore_resource[RTE_MAX_LCORE];
117 static struct rte_mempool *message_pool;
118 static rte_spinlock_t res_lock = RTE_SPINLOCK_INITIALIZER;
119 /* use floating processes */
120 static int float_proc = 0;
121 /* Save original cpu affinity */
127 static const struct rte_eth_conf port_conf = {
130 .header_split = 0, /**< Header Split disabled */
131 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
132 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
133 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
134 .hw_strip_crc = 1, /**< CRC stripped by hardware */
137 .mq_mode = ETH_MQ_TX_NONE,
141 static struct rte_mempool * l2fwd_pktmbuf_pool[RTE_MAX_ETHPORTS];
143 /* Per-port statistics struct */
144 struct l2fwd_port_statistics {
148 } __rte_cache_aligned;
149 struct l2fwd_port_statistics *port_statistics;
151 * pointer to lcore ID mapping array, used to return lcore id in case slave
152 * process exited unexpectedly, use only floating process option applied
154 unsigned *mapping_id;
156 /* A tsc-based timer responsible for triggering statistics printout */
157 #define TIMER_MILLISECOND 2000000ULL /* around 1ms at 2 Ghz */
158 #define MAX_TIMER_PERIOD 86400 /* 1 day max */
159 static int64_t timer_period = 10 * TIMER_MILLISECOND * 1000; /* default period is 10 seconds */
161 static int l2fwd_launch_one_lcore(void *dummy);
163 /* Print out statistics on packets dropped */
167 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
170 total_packets_dropped = 0;
171 total_packets_tx = 0;
172 total_packets_rx = 0;
174 const char clr[] = { 27, '[', '2', 'J', '\0' };
175 const char topLeft[] = { 27, '[', '1', ';', '1', 'H','\0' };
177 /* Clear screen and move to top left */
178 printf("%s%s", clr, topLeft);
180 printf("\nPort statistics ====================================");
182 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
183 /* skip disabled ports */
184 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
186 printf("\nStatistics for port %u ------------------------------"
187 "\nPackets sent: %24"PRIu64
188 "\nPackets received: %20"PRIu64
189 "\nPackets dropped: %21"PRIu64,
191 port_statistics[portid].tx,
192 port_statistics[portid].rx,
193 port_statistics[portid].dropped);
195 total_packets_dropped += port_statistics[portid].dropped;
196 total_packets_tx += port_statistics[portid].tx;
197 total_packets_rx += port_statistics[portid].rx;
199 printf("\nAggregate statistics ==============================="
200 "\nTotal packets sent: %18"PRIu64
201 "\nTotal packets received: %14"PRIu64
202 "\nTotal packets dropped: %15"PRIu64,
205 total_packets_dropped);
206 printf("\n====================================================\n");
210 clear_cpu_affinity(void)
214 s = sched_setaffinity(0, cpu_aff.size, &cpu_aff.set);
216 printf("sched_setaffinity failed:%s\n", strerror(errno));
224 get_cpu_affinity(void)
228 cpu_aff.size = sizeof(cpu_set_t);
229 CPU_ZERO(&cpu_aff.set);
231 s = sched_getaffinity(0, cpu_aff.size, &cpu_aff.set);
233 printf("sched_getaffinity failed:%s\n", strerror(errno));
241 * This fnciton demonstrates the approach to create ring in first instance
242 * or re-attach an existed ring in later instance.
244 static struct rte_ring *
245 create_ring(const char *name, unsigned count,
246 int socket_id,unsigned flags)
248 struct rte_ring *ring;
253 /* If already create, just attached it */
254 if (likely((ring = rte_ring_lookup(name)) != NULL))
257 /* First call it, create one */
258 return rte_ring_create(name, count, socket_id, flags);
261 /* Malloc with rte_malloc on structures that shared by master and slave */
263 l2fwd_malloc_shared_struct(void)
265 port_statistics = rte_zmalloc("port_stat",
266 sizeof(struct l2fwd_port_statistics) * RTE_MAX_ETHPORTS,
268 if (port_statistics == NULL)
271 /* allocate mapping_id array */
274 mapping_id = rte_malloc("mapping_id", sizeof(unsigned) * RTE_MAX_LCORE,
277 if (mapping_id == NULL)
280 for (i = 0 ;i < RTE_MAX_LCORE; i++)
281 mapping_id[i] = INVALID_MAPPING_ID;
286 /* Create ring which used for communicate among master and slave */
288 create_ms_ring(unsigned slaveid)
290 unsigned flag = RING_F_SP_ENQ | RING_F_SC_DEQ;
291 struct lcore_resource_struct *res = &lcore_resource[slaveid];
292 unsigned socketid = rte_socket_id();
294 /* Always assume create ring on master socket_id */
295 /* Default only create a ring size 32 */
296 snprintf(res->ring_name[0], MAX_NAME_LEN, "%s%u",
297 RING_MASTER_NAME, slaveid);
298 if ((res->ring[0] = create_ring(res->ring_name[0], NB_CORE_MSGBUF,
299 socketid, flag)) == NULL) {
300 printf("Create m2s ring %s failed\n", res->ring_name[0]);
304 snprintf(res->ring_name[1], MAX_NAME_LEN, "%s%u",
305 RING_SLAVE_NAME, slaveid);
306 if ((res->ring[1] = create_ring(res->ring_name[1], NB_CORE_MSGBUF,
307 socketid, flag)) == NULL) {
308 printf("Create s2m ring %s failed\n", res->ring_name[1]);
315 /* send command to pair in paired master and slave ring */
317 sendcmd(unsigned slaveid, enum l2fwd_cmd cmd, int is_master)
319 struct lcore_resource_struct *res = &lcore_resource[slaveid];
323 /* Only check master, it must be enabled and running if it is slave */
324 if (is_master && !res->enabled)
327 if (res->ring[fd] == NULL)
330 if (rte_mempool_get(message_pool, &msg) < 0) {
331 printf("Error to get message buffer\n");
335 *(enum l2fwd_cmd *)msg = cmd;
337 if (rte_ring_enqueue(res->ring[fd], msg) != 0) {
338 printf("Enqueue error\n");
339 rte_mempool_put(message_pool, msg);
346 /* Get command from pair in paired master and slave ring */
348 getcmd(unsigned slaveid, enum l2fwd_cmd *cmd, int is_master)
350 struct lcore_resource_struct *res = &lcore_resource[slaveid];
352 int fd = !!is_master;
354 /* Only check master, it must be enabled and running if it is slave */
355 if (is_master && (!res->enabled))
358 if (res->ring[fd] == NULL)
361 ret = rte_ring_dequeue(res->ring[fd], &msg);
364 *cmd = *(enum l2fwd_cmd *)msg;
365 rte_mempool_put(message_pool, msg);
370 /* Master send command to slave and wait until ack received or error met */
372 master_sendcmd_with_ack(unsigned slaveid, enum l2fwd_cmd cmd)
374 enum l2fwd_cmd ack_cmd;
377 if (sendcmd(slaveid, cmd, 1) != 0)
378 rte_exit(EXIT_FAILURE, "Failed to send message\n");
382 ret = getcmd(slaveid, &ack_cmd, 1);
383 if (ret == 0 && cmd == ack_cmd)
386 /* If slave not running yet, return an error */
387 if (flib_query_slave_status(slaveid) != ST_RUN) {
396 /* restart all port that assigned to that slave lcore */
398 reset_slave_all_ports(unsigned slaveid)
400 struct lcore_resource_struct *slave = &lcore_resource[slaveid];
403 /* stop/start port */
404 for (i = 0; i < slave->port_num; i++) {
405 char buf_name[RTE_MEMPOOL_NAMESIZE];
406 struct rte_mempool *pool;
407 printf("Stop port :%d\n", slave->port[i]);
408 rte_eth_dev_stop(slave->port[i]);
409 snprintf(buf_name, RTE_MEMPOOL_NAMESIZE, MBUF_NAME, slave->port[i]);
410 pool = rte_mempool_lookup(buf_name);
412 printf("Port %d mempool free object is %u(%u)\n", slave->port[i],
413 rte_mempool_avail_count(pool),
414 (unsigned int)NB_MBUF);
416 printf("Can't find mempool %s\n", buf_name);
418 printf("Start port :%d\n", slave->port[i]);
419 ret = rte_eth_dev_start(slave->port[i]);
427 reset_shared_structures(unsigned slaveid)
430 /* Only port are shared resource here */
431 ret = reset_slave_all_ports(slaveid);
437 * Call this function to re-create resource that needed for slave process that
438 * exited in last instance
441 init_slave_res(unsigned slaveid)
443 struct lcore_resource_struct *slave = &lcore_resource[slaveid];
446 if (!slave->enabled) {
447 printf("Something wrong with lcore=%u enabled=%d\n",slaveid,
452 /* Initialize ring */
453 if (create_ms_ring(slaveid) != 0)
454 rte_exit(EXIT_FAILURE, "failed to create ring for slave %u\n",
457 /* drain un-read buffer if have */
458 while (getcmd(slaveid, &cmd, 1) == 0);
459 while (getcmd(slaveid, &cmd, 0) == 0);
465 recreate_one_slave(unsigned slaveid)
468 /* Re-initialize resource for stalled slave */
469 if ((ret = init_slave_res(slaveid)) != 0) {
470 printf("Init slave=%u failed\n", slaveid);
474 if ((ret = flib_remote_launch(l2fwd_launch_one_lcore, NULL, slaveid))
476 printf("Launch slave %u failed\n", slaveid);
482 * remapping resource belong to slave_id to new lcore that gets from flib_assign_lcore_id(),
483 * used only floating process option applied.
486 * original lcore_id that apply for remapping
489 remapping_slave_resource(unsigned slaveid, unsigned map_id)
492 /* remapping lcore_resource */
493 memcpy(&lcore_resource[map_id], &lcore_resource[slaveid],
494 sizeof(struct lcore_resource_struct));
496 /* remapping lcore_queue_conf */
497 memcpy(&lcore_queue_conf[map_id], &lcore_queue_conf[slaveid],
498 sizeof(struct lcore_queue_conf));
502 reset_pair(unsigned slaveid, unsigned pairid)
505 if ((ret = reset_shared_structures(slaveid)) != 0)
508 if((ret = reset_shared_structures(pairid)) != 0)
512 unsigned map_id = mapping_id[slaveid];
514 if (map_id != INVALID_MAPPING_ID) {
515 printf("%u return mapping id %u\n", slaveid, map_id);
516 flib_free_lcore_id(map_id);
517 mapping_id[slaveid] = INVALID_MAPPING_ID;
520 map_id = mapping_id[pairid];
521 if (map_id != INVALID_MAPPING_ID) {
522 printf("%u return mapping id %u\n", pairid, map_id);
523 flib_free_lcore_id(map_id);
524 mapping_id[pairid] = INVALID_MAPPING_ID;
528 if((ret = recreate_one_slave(slaveid)) != 0)
531 ret = recreate_one_slave(pairid);
538 slave_exit_cb(unsigned slaveid, __attribute__((unused))int stat)
540 struct lcore_resource_struct *slave = &lcore_resource[slaveid];
542 printf("Get slave %u leave info\n", slaveid);
543 if (!slave->enabled) {
544 printf("Lcore=%u not registered for it's exit\n", slaveid);
547 rte_spinlock_lock(&res_lock);
549 /* Change the state and wait master to start them */
550 slave->flags = SLAVE_RECREATE_FLAG;
552 rte_spinlock_unlock(&res_lock);
556 l2fwd_simple_forward(struct rte_mbuf *m, unsigned portid)
558 struct ether_hdr *eth;
562 struct rte_eth_dev_tx_buffer *buffer;
564 dst_port = l2fwd_dst_ports[portid];
565 eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
567 /* 02:00:00:00:00:xx */
568 tmp = ð->d_addr.addr_bytes[0];
569 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
572 ether_addr_copy(&l2fwd_ports_eth_addr[dst_port], ð->s_addr);
574 buffer = tx_buffer[dst_port];
575 sent = rte_eth_tx_buffer(dst_port, 0, buffer, m);
577 port_statistics[dst_port].tx += sent;
580 /* main processing loop */
582 l2fwd_main_loop(void)
584 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
588 uint64_t prev_tsc, diff_tsc, cur_tsc;
589 unsigned i, j, portid, nb_rx;
590 struct lcore_queue_conf *qconf;
591 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S *
593 struct rte_eth_dev_tx_buffer *buffer;
597 lcore_id = rte_lcore_id();
599 qconf = &lcore_queue_conf[lcore_id];
601 if (qconf->n_rx_port == 0) {
602 RTE_LOG(INFO, L2FWD, "lcore %u has nothing to do\n", lcore_id);
606 RTE_LOG(INFO, L2FWD, "entering main loop on lcore %u\n", lcore_id);
608 for (i = 0; i < qconf->n_rx_port; i++) {
609 portid = qconf->rx_port_list[i];
610 RTE_LOG(INFO, L2FWD, " -- lcoreid=%u portid=%u\n", lcore_id,
616 cur_tsc = rte_rdtsc();
618 if (unlikely(getcmd(lcore_id, &cmd, 0) == 0)) {
619 sendcmd(lcore_id, cmd, 0);
621 /* If get stop command, stop forwarding and exit */
622 if (cmd == CMD_STOP) {
628 * TX burst queue drain
630 diff_tsc = cur_tsc - prev_tsc;
631 if (unlikely(diff_tsc > drain_tsc)) {
633 for (i = 0; i < qconf->n_rx_port; i++) {
635 portid = l2fwd_dst_ports[qconf->rx_port_list[i]];
636 buffer = tx_buffer[portid];
638 sent = rte_eth_tx_buffer_flush(portid, 0, buffer);
640 port_statistics[portid].tx += sent;
648 * Read packet from RX queues
650 for (i = 0; i < qconf->n_rx_port; i++) {
652 portid = qconf->rx_port_list[i];
653 nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
654 pkts_burst, MAX_PKT_BURST);
656 port_statistics[portid].rx += nb_rx;
658 for (j = 0; j < nb_rx; j++) {
660 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
661 l2fwd_simple_forward(m, portid);
668 l2fwd_launch_one_lcore(__attribute__((unused)) void *dummy)
670 unsigned lcore_id = rte_lcore_id();
675 /* Change it to floating process, also change it's lcore_id */
676 clear_cpu_affinity();
677 RTE_PER_LCORE(_lcore_id) = 0;
679 if (flib_assign_lcore_id() < 0 ) {
680 printf("flib_assign_lcore_id failed\n");
683 flcore_id = rte_lcore_id();
684 /* Set mapping id, so master can return it after slave exited */
685 mapping_id[lcore_id] = flcore_id;
686 printf("Org lcore_id = %u, cur lcore_id = %u\n",
687 lcore_id, flcore_id);
688 remapping_slave_resource(lcore_id, flcore_id);
693 /* return lcore_id before return */
695 flib_free_lcore_id(rte_lcore_id());
696 mapping_id[lcore_id] = INVALID_MAPPING_ID;
703 l2fwd_usage(const char *prgname)
705 printf("%s [EAL options] -- -p PORTMASK -s COREMASK [-q NQ] -f\n"
706 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
707 " -q NQ: number of queue (=ports) per lcore (default is 1)\n"
708 " -f use floating process which won't bind to any core to run\n"
709 " -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n",
714 l2fwd_parse_portmask(const char *portmask)
719 /* parse hexadecimal string */
720 pm = strtoul(portmask, &end, 16);
721 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
731 l2fwd_parse_nqueue(const char *q_arg)
736 /* parse hexadecimal string */
737 n = strtoul(q_arg, &end, 10);
738 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
742 if (n >= MAX_RX_QUEUE_PER_LCORE)
749 l2fwd_parse_timer_period(const char *q_arg)
754 /* parse number string */
755 n = strtol(q_arg, &end, 10);
756 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
758 if (n >= MAX_TIMER_PERIOD)
764 /* Parse the argument given in the command line of the application */
766 l2fwd_parse_args(int argc, char **argv)
771 char *prgname = argv[0];
772 static struct option lgopts[] = {
779 while ((opt = getopt_long(argc, argvopt, "p:q:T:f",
780 lgopts, &option_index)) != EOF) {
785 l2fwd_enabled_port_mask = l2fwd_parse_portmask(optarg);
786 if (l2fwd_enabled_port_mask == 0) {
787 printf("invalid portmask\n");
788 l2fwd_usage(prgname);
796 l2fwd_rx_queue_per_lcore = l2fwd_parse_nqueue(optarg);
797 if (l2fwd_rx_queue_per_lcore == 0) {
798 printf("invalid queue number\n");
799 l2fwd_usage(prgname);
806 timer_period = l2fwd_parse_timer_period(optarg) * 1000 * TIMER_MILLISECOND;
807 if (timer_period < 0) {
808 printf("invalid timer period\n");
809 l2fwd_usage(prgname);
814 /* use floating process */
821 l2fwd_usage(prgname);
825 l2fwd_usage(prgname);
831 argv[optind-1] = prgname;
834 l2fwd_usage(prgname);
838 optind = 1; /* reset getopt lib */
842 /* Check the link status of all ports in up to 9s, and print them finally */
844 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask)
846 #define CHECK_INTERVAL 100 /* 100ms */
847 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
849 uint8_t count, all_ports_up, print_flag = 0;
850 struct rte_eth_link link;
852 printf("\nChecking link status");
854 for (count = 0; count <= MAX_CHECK_TIME; count++) {
856 for (portid = 0; portid < port_num; portid++) {
857 if ((port_mask & (1 << portid)) == 0)
859 memset(&link, 0, sizeof(link));
860 rte_eth_link_get_nowait(portid, &link);
861 /* print link status if flag set */
862 if (print_flag == 1) {
863 if (link.link_status)
865 "Port%d Link Up- speed %u Mbps- %s\n",
866 portid, link.link_speed,
867 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
868 ("full-duplex") : ("half-duplex\n"));
870 printf("Port %d Link Down\n", portid);
873 /* clear all_ports_up flag if any link down */
874 if (link.link_status == ETH_LINK_DOWN) {
879 /* after finally printing all link status, get out */
883 if (all_ports_up == 0) {
886 rte_delay_ms(CHECK_INTERVAL);
889 /* set the print_flag if all ports up or timeout */
890 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
898 main(int argc, char **argv)
900 struct lcore_queue_conf *qconf;
903 uint16_t nb_ports_available;
904 uint16_t portid, last_port;
905 unsigned rx_lcore_id;
906 unsigned nb_ports_in_mask = 0;
908 uint64_t prev_tsc, diff_tsc, cur_tsc, timer_tsc;
910 /* Save cpu_affinity first, restore it in case it's floating process option */
911 if (get_cpu_affinity() != 0)
912 rte_exit(EXIT_FAILURE, "get_cpu_affinity error\n");
914 /* Also tries to set cpu affinity to detect whether it will fail in child process */
915 if(clear_cpu_affinity() != 0)
916 rte_exit(EXIT_FAILURE, "clear_cpu_affinity error\n");
919 ret = rte_eal_init(argc, argv);
921 rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
925 /* parse application arguments (after the EAL ones) */
926 ret = l2fwd_parse_args(argc, argv);
928 rte_exit(EXIT_FAILURE, "Invalid L2FWD arguments\n");
931 if (flib_init() != 0)
932 rte_exit(EXIT_FAILURE, "flib init error");
935 * Allocated structures that slave lcore would change. For those that slaves are
936 * read only, needn't use malloc to share and global or static variables is ok since
937 * slave inherit all the knowledge that master initialized.
939 if (l2fwd_malloc_shared_struct() != 0)
940 rte_exit(EXIT_FAILURE, "malloc mem failed\n");
942 /* Initialize lcore_resource structures */
943 memset(lcore_resource, 0, sizeof(lcore_resource));
944 for (i = 0; i < RTE_MAX_LCORE; i++)
945 lcore_resource[i].lcore_id = i;
947 nb_ports = rte_eth_dev_count();
949 rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
951 /* create the mbuf pool */
952 for (portid = 0; portid < nb_ports; portid++) {
953 /* skip ports that are not enabled */
954 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
956 char buf_name[RTE_MEMPOOL_NAMESIZE];
957 snprintf(buf_name, RTE_MEMPOOL_NAMESIZE, MBUF_NAME, portid);
958 l2fwd_pktmbuf_pool[portid] =
959 rte_pktmbuf_pool_create(buf_name, NB_MBUF, 32,
960 0, MBUF_DATA_SIZE, rte_socket_id());
961 if (l2fwd_pktmbuf_pool[portid] == NULL)
962 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
964 printf("Create mbuf %s\n", buf_name);
967 /* reset l2fwd_dst_ports */
968 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++)
969 l2fwd_dst_ports[portid] = 0;
973 * Each logical core is assigned a dedicated TX queue on each port.
975 for (portid = 0; portid < nb_ports; portid++) {
976 /* skip ports that are not enabled */
977 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
980 if (nb_ports_in_mask % 2) {
981 l2fwd_dst_ports[portid] = last_port;
982 l2fwd_dst_ports[last_port] = portid;
989 if (nb_ports_in_mask % 2) {
990 printf("Notice: odd number of ports in portmask.\n");
991 l2fwd_dst_ports[last_port] = last_port;
997 /* Initialize the port/queue configuration of each logical core */
998 for (portid = 0; portid < nb_ports; portid++) {
999 struct lcore_resource_struct *res;
1000 /* skip ports that are not enabled */
1001 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
1004 /* get the lcore_id for this port */
1005 /* skip master lcore */
1006 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
1007 rte_get_master_lcore() == rx_lcore_id ||
1008 lcore_queue_conf[rx_lcore_id].n_rx_port ==
1009 l2fwd_rx_queue_per_lcore) {
1012 if (rx_lcore_id >= RTE_MAX_LCORE)
1013 rte_exit(EXIT_FAILURE, "Not enough cores\n");
1016 if (qconf != &lcore_queue_conf[rx_lcore_id])
1017 /* Assigned a new logical core in the loop above. */
1018 qconf = &lcore_queue_conf[rx_lcore_id];
1020 qconf->rx_port_list[qconf->n_rx_port] = portid;
1023 /* Save the port resource info into lcore_resource strucutres */
1024 res = &lcore_resource[rx_lcore_id];
1026 res->port[res->port_num++] = portid;
1028 printf("Lcore %u: RX port %u\n", rx_lcore_id, (unsigned) portid);
1031 nb_ports_available = nb_ports;
1033 /* Initialise each port */
1034 for (portid = 0; portid < nb_ports; portid++) {
1035 /* skip ports that are not enabled */
1036 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0) {
1037 printf("Skipping disabled port %u\n", (unsigned) portid);
1038 nb_ports_available--;
1042 printf("Initializing port %u... ", (unsigned) portid);
1044 ret = rte_eth_dev_configure(portid, 1, 1, &port_conf);
1046 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
1047 ret, (unsigned) portid);
1049 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1052 rte_exit(EXIT_FAILURE,
1053 "rte_eth_dev_adjust_nb_rx_tx_desc: err=%d, port=%u\n",
1054 ret, (unsigned) portid);
1056 rte_eth_macaddr_get(portid,&l2fwd_ports_eth_addr[portid]);
1058 /* init one RX queue */
1060 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
1061 rte_eth_dev_socket_id(portid),
1063 l2fwd_pktmbuf_pool[portid]);
1065 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup:err=%d, port=%u\n",
1066 ret, (unsigned) portid);
1068 /* init one TX queue on each port */
1070 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
1071 rte_eth_dev_socket_id(portid),
1074 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup:err=%d, port=%u\n",
1075 ret, (unsigned) portid);
1077 /* Initialize TX buffers */
1078 tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
1079 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
1080 rte_eth_dev_socket_id(portid));
1081 if (tx_buffer[portid] == NULL)
1082 rte_exit(EXIT_FAILURE, "Cannot allocate buffer for tx on port %u\n",
1085 rte_eth_tx_buffer_init(tx_buffer[portid], MAX_PKT_BURST);
1087 ret = rte_eth_tx_buffer_set_err_callback(tx_buffer[portid],
1088 rte_eth_tx_buffer_count_callback,
1089 &port_statistics[portid].dropped);
1091 rte_exit(EXIT_FAILURE, "Cannot set error callback for "
1092 "tx buffer on port %u\n", (unsigned) portid);
1095 ret = rte_eth_dev_start(portid);
1097 rte_exit(EXIT_FAILURE, "rte_eth_dev_start:err=%d, port=%u\n",
1098 ret, (unsigned) portid);
1102 rte_eth_promiscuous_enable(portid);
1104 printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
1106 l2fwd_ports_eth_addr[portid].addr_bytes[0],
1107 l2fwd_ports_eth_addr[portid].addr_bytes[1],
1108 l2fwd_ports_eth_addr[portid].addr_bytes[2],
1109 l2fwd_ports_eth_addr[portid].addr_bytes[3],
1110 l2fwd_ports_eth_addr[portid].addr_bytes[4],
1111 l2fwd_ports_eth_addr[portid].addr_bytes[5]);
1113 /* initialize port stats */
1114 //memset(&port_statistics, 0, sizeof(port_statistics));
1117 if (!nb_ports_available) {
1118 rte_exit(EXIT_FAILURE,
1119 "All available ports are disabled. Please set portmask.\n");
1122 check_all_ports_link_status(nb_ports, l2fwd_enabled_port_mask);
1124 /* Record pair lcore */
1126 * Since l2fwd example would create pair between different neighbour port, that's
1127 * port 0 receive and forward to port 1, the same to port 1, these 2 ports will have
1128 * dependency. If one port stopped working (killed, for example), the port need to
1129 * be stopped/started again. During the time, another port need to wait until stop/start
1130 * procedure completed. So, record the pair relationship for those lcores working
1133 for (portid = 0; portid < nb_ports; portid++) {
1135 unsigned lcore = 0, pair_lcore = 0;
1136 unsigned j, find_lcore, find_pair_lcore;
1137 /* skip ports that are not enabled */
1138 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
1141 /* Find pair ports' lcores */
1142 find_lcore = find_pair_lcore = 0;
1143 pair_port = l2fwd_dst_ports[portid];
1144 for (i = 0; i < RTE_MAX_LCORE; i++) {
1145 if (!rte_lcore_is_enabled(i))
1147 for (j = 0; j < lcore_queue_conf[i].n_rx_port;j++) {
1148 if (lcore_queue_conf[i].rx_port_list[j] == portid) {
1153 if (lcore_queue_conf[i].rx_port_list[j] == pair_port) {
1155 find_pair_lcore = 1;
1159 if (find_lcore && find_pair_lcore)
1162 if (!find_lcore || !find_pair_lcore)
1163 rte_exit(EXIT_FAILURE, "Not find port=%d pair\n", portid);
1165 printf("lcore %u and %u paired\n", lcore, pair_lcore);
1166 lcore_resource[lcore].pair_id = pair_lcore;
1167 lcore_resource[pair_lcore].pair_id = lcore;
1170 /* Create message buffer for all master and slave */
1171 message_pool = rte_mempool_create("ms_msg_pool",
1172 NB_CORE_MSGBUF * RTE_MAX_LCORE,
1173 sizeof(enum l2fwd_cmd), NB_CORE_MSGBUF / 2,
1174 0, NULL, NULL, NULL, NULL, rte_socket_id(), 0);
1176 if (message_pool == NULL)
1177 rte_exit(EXIT_FAILURE, "Create msg mempool failed\n");
1179 /* Create ring for each master and slave pair, also register cb when slave leaves */
1180 for (i = 0; i < RTE_MAX_LCORE; i++) {
1182 * Only create ring and register slave_exit cb in case that core involved into
1185 if (lcore_resource[i].enabled) {
1186 /* Create ring for master and slave communication */
1187 ret = create_ms_ring(i);
1189 rte_exit(EXIT_FAILURE, "Create ring for lcore=%u failed",
1192 if (flib_register_slave_exit_notify(i,
1193 slave_exit_cb) != 0)
1194 rte_exit(EXIT_FAILURE,
1195 "Register master_trace_slave_exit failed");
1199 /* launch per-lcore init on every lcore except master */
1200 flib_mp_remote_launch(l2fwd_launch_one_lcore, NULL, SKIP_MASTER);
1202 /* print statistics 10 second */
1203 prev_tsc = cur_tsc = rte_rdtsc();
1207 cur_tsc = rte_rdtsc();
1208 diff_tsc = cur_tsc - prev_tsc;
1209 /* if timer is enabled */
1210 if (timer_period > 0) {
1212 /* advance the timer */
1213 timer_tsc += diff_tsc;
1215 /* if timer has reached its timeout */
1216 if (unlikely(timer_tsc >= (uint64_t) timer_period)) {
1219 /* reset the timer */
1226 /* Check any slave need restart or recreate */
1227 rte_spinlock_lock(&res_lock);
1228 for (i = 0; i < RTE_MAX_LCORE; i++) {
1229 struct lcore_resource_struct *res = &lcore_resource[i];
1230 struct lcore_resource_struct *pair = &lcore_resource[res->pair_id];
1232 /* If find slave exited, try to reset pair */
1233 if (res->enabled && res->flags && pair->enabled) {
1235 master_sendcmd_with_ack(pair->lcore_id, CMD_STOP);
1236 rte_spinlock_unlock(&res_lock);
1238 rte_spinlock_lock(&res_lock);
1242 if (reset_pair(res->lcore_id, pair->lcore_id) != 0)
1243 rte_exit(EXIT_FAILURE, "failed to reset slave");
1248 rte_spinlock_unlock(&res_lock);