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>
55 #include <rte_launch.h>
56 #include <rte_atomic.h>
57 #include <rte_spinlock.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>
64 #include <rte_random.h>
65 #include <rte_debug.h>
66 #include <rte_ether.h>
67 #include <rte_ethdev.h>
69 #include <rte_mempool.h>
71 #include <rte_malloc.h>
75 #define RTE_LOGTYPE_L2FWD RTE_LOGTYPE_USER1
76 #define MBUF_NAME "mbuf_pool_%d"
77 #define MBUF_DATA_SIZE RTE_MBUF_DEFAULT_BUF_SIZE
79 #define RING_MASTER_NAME "l2fwd_ring_m2s_"
80 #define RING_SLAVE_NAME "l2fwd_ring_s2m_"
81 #define MAX_NAME_LEN 32
82 /* RECREATE flag indicate needs initialize resource and launch slave_core again */
83 #define SLAVE_RECREATE_FLAG 0x1
84 /* RESTART flag indicate needs restart port and send START command again */
85 #define SLAVE_RESTART_FLAG 0x2
86 #define INVALID_MAPPING_ID ((unsigned)LCORE_ID_ANY)
87 /* Maximum message buffer per slave */
88 #define NB_CORE_MSGBUF 32
94 #define MAX_PKT_BURST 32
95 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
98 * Configurable number of RX/TX ring descriptors
100 #define RTE_TEST_RX_DESC_DEFAULT 128
101 #define RTE_TEST_TX_DESC_DEFAULT 512
102 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
103 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
105 /* ethernet addresses of ports */
106 static struct ether_addr l2fwd_ports_eth_addr[RTE_MAX_ETHPORTS];
108 /* mask of enabled ports */
109 static uint32_t l2fwd_enabled_port_mask = 0;
111 /* list of enabled ports */
112 static uint32_t l2fwd_dst_ports[RTE_MAX_ETHPORTS];
114 static unsigned int l2fwd_rx_queue_per_lcore = 1;
118 struct rte_mbuf *m_table[MAX_PKT_BURST];
121 #define MAX_RX_QUEUE_PER_LCORE 16
122 #define MAX_TX_QUEUE_PER_PORT 16
123 struct lcore_queue_conf {
125 unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
126 } __rte_cache_aligned;
127 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
129 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
131 struct lcore_resource_struct {
132 int enabled; /* Only set in case this lcore involved into packet forwarding */
133 int flags; /* Set only slave need to restart or recreate */
134 unsigned lcore_id; /* lcore ID */
135 unsigned pair_id; /* dependency lcore ID on port */
136 char ring_name[2][MAX_NAME_LEN];
137 /* ring[0] for master send cmd, slave read */
138 /* ring[1] for slave send ack, master read */
139 struct rte_ring *ring[2];
140 int port_num; /* Total port numbers */
141 /* Port id for that lcore to receive packets */
142 uint16_t port[RTE_MAX_ETHPORTS];
143 }__attribute__((packed)) __rte_cache_aligned;
145 static struct lcore_resource_struct lcore_resource[RTE_MAX_LCORE];
146 static struct rte_mempool *message_pool;
147 static rte_spinlock_t res_lock = RTE_SPINLOCK_INITIALIZER;
148 /* use floating processes */
149 static int float_proc = 0;
150 /* Save original cpu affinity */
156 static const struct rte_eth_conf port_conf = {
159 .header_split = 0, /**< Header Split disabled */
160 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
161 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
162 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
163 .hw_strip_crc = 1, /**< CRC stripped by hardware */
166 .mq_mode = ETH_MQ_TX_NONE,
170 static struct rte_mempool * l2fwd_pktmbuf_pool[RTE_MAX_ETHPORTS];
172 /* Per-port statistics struct */
173 struct l2fwd_port_statistics {
177 } __rte_cache_aligned;
178 struct l2fwd_port_statistics *port_statistics;
180 * pointer to lcore ID mapping array, used to return lcore id in case slave
181 * process exited unexpectedly, use only floating process option applied
183 unsigned *mapping_id;
185 /* A tsc-based timer responsible for triggering statistics printout */
186 #define TIMER_MILLISECOND 2000000ULL /* around 1ms at 2 Ghz */
187 #define MAX_TIMER_PERIOD 86400 /* 1 day max */
188 static int64_t timer_period = 10 * TIMER_MILLISECOND * 1000; /* default period is 10 seconds */
190 static int l2fwd_launch_one_lcore(void *dummy);
192 /* Print out statistics on packets dropped */
196 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
199 total_packets_dropped = 0;
200 total_packets_tx = 0;
201 total_packets_rx = 0;
203 const char clr[] = { 27, '[', '2', 'J', '\0' };
204 const char topLeft[] = { 27, '[', '1', ';', '1', 'H','\0' };
206 /* Clear screen and move to top left */
207 printf("%s%s", clr, topLeft);
209 printf("\nPort statistics ====================================");
211 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
212 /* skip disabled ports */
213 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
215 printf("\nStatistics for port %u ------------------------------"
216 "\nPackets sent: %24"PRIu64
217 "\nPackets received: %20"PRIu64
218 "\nPackets dropped: %21"PRIu64,
220 port_statistics[portid].tx,
221 port_statistics[portid].rx,
222 port_statistics[portid].dropped);
224 total_packets_dropped += port_statistics[portid].dropped;
225 total_packets_tx += port_statistics[portid].tx;
226 total_packets_rx += port_statistics[portid].rx;
228 printf("\nAggregate statistics ==============================="
229 "\nTotal packets sent: %18"PRIu64
230 "\nTotal packets received: %14"PRIu64
231 "\nTotal packets dropped: %15"PRIu64,
234 total_packets_dropped);
235 printf("\n====================================================\n");
239 clear_cpu_affinity(void)
243 s = sched_setaffinity(0, cpu_aff.size, &cpu_aff.set);
245 printf("sched_setaffinity failed:%s\n", strerror(errno));
253 get_cpu_affinity(void)
257 cpu_aff.size = sizeof(cpu_set_t);
258 CPU_ZERO(&cpu_aff.set);
260 s = sched_getaffinity(0, cpu_aff.size, &cpu_aff.set);
262 printf("sched_getaffinity failed:%s\n", strerror(errno));
270 * This fnciton demonstrates the approach to create ring in first instance
271 * or re-attach an existed ring in later instance.
273 static struct rte_ring *
274 create_ring(const char *name, unsigned count,
275 int socket_id,unsigned flags)
277 struct rte_ring *ring;
282 /* If already create, just attached it */
283 if (likely((ring = rte_ring_lookup(name)) != NULL))
286 /* First call it, create one */
287 return rte_ring_create(name, count, socket_id, flags);
290 /* Malloc with rte_malloc on structures that shared by master and slave */
292 l2fwd_malloc_shared_struct(void)
294 port_statistics = rte_zmalloc("port_stat",
295 sizeof(struct l2fwd_port_statistics) * RTE_MAX_ETHPORTS,
297 if (port_statistics == NULL)
300 /* allocate mapping_id array */
303 mapping_id = rte_malloc("mapping_id", sizeof(unsigned) * RTE_MAX_LCORE,
306 if (mapping_id == NULL)
309 for (i = 0 ;i < RTE_MAX_LCORE; i++)
310 mapping_id[i] = INVALID_MAPPING_ID;
315 /* Create ring which used for communicate among master and slave */
317 create_ms_ring(unsigned slaveid)
319 unsigned flag = RING_F_SP_ENQ | RING_F_SC_DEQ;
320 struct lcore_resource_struct *res = &lcore_resource[slaveid];
321 unsigned socketid = rte_socket_id();
323 /* Always assume create ring on master socket_id */
324 /* Default only create a ring size 32 */
325 snprintf(res->ring_name[0], MAX_NAME_LEN, "%s%u",
326 RING_MASTER_NAME, slaveid);
327 if ((res->ring[0] = create_ring(res->ring_name[0], NB_CORE_MSGBUF,
328 socketid, flag)) == NULL) {
329 printf("Create m2s ring %s failed\n", res->ring_name[0]);
333 snprintf(res->ring_name[1], MAX_NAME_LEN, "%s%u",
334 RING_SLAVE_NAME, slaveid);
335 if ((res->ring[1] = create_ring(res->ring_name[1], NB_CORE_MSGBUF,
336 socketid, flag)) == NULL) {
337 printf("Create s2m ring %s failed\n", res->ring_name[1]);
344 /* send command to pair in paired master and slave ring */
346 sendcmd(unsigned slaveid, enum l2fwd_cmd cmd, int is_master)
348 struct lcore_resource_struct *res = &lcore_resource[slaveid];
352 /* Only check master, it must be enabled and running if it is slave */
353 if (is_master && !res->enabled)
356 if (res->ring[fd] == NULL)
359 if (rte_mempool_get(message_pool, &msg) < 0) {
360 printf("Error to get message buffer\n");
364 *(enum l2fwd_cmd *)msg = cmd;
366 if (rte_ring_enqueue(res->ring[fd], msg) != 0) {
367 printf("Enqueue error\n");
368 rte_mempool_put(message_pool, msg);
375 /* Get command from pair in paired master and slave ring */
377 getcmd(unsigned slaveid, enum l2fwd_cmd *cmd, int is_master)
379 struct lcore_resource_struct *res = &lcore_resource[slaveid];
381 int fd = !!is_master;
383 /* Only check master, it must be enabled and running if it is slave */
384 if (is_master && (!res->enabled))
387 if (res->ring[fd] == NULL)
390 ret = rte_ring_dequeue(res->ring[fd], &msg);
393 *cmd = *(enum l2fwd_cmd *)msg;
394 rte_mempool_put(message_pool, msg);
399 /* Master send command to slave and wait until ack received or error met */
401 master_sendcmd_with_ack(unsigned slaveid, enum l2fwd_cmd cmd)
403 enum l2fwd_cmd ack_cmd;
406 if (sendcmd(slaveid, cmd, 1) != 0)
407 rte_exit(EXIT_FAILURE, "Failed to send message\n");
411 ret = getcmd(slaveid, &ack_cmd, 1);
412 if (ret == 0 && cmd == ack_cmd)
415 /* If slave not running yet, return an error */
416 if (flib_query_slave_status(slaveid) != ST_RUN) {
425 /* restart all port that assigned to that slave lcore */
427 reset_slave_all_ports(unsigned slaveid)
429 struct lcore_resource_struct *slave = &lcore_resource[slaveid];
432 /* stop/start port */
433 for (i = 0; i < slave->port_num; i++) {
434 char buf_name[RTE_MEMPOOL_NAMESIZE];
435 struct rte_mempool *pool;
436 printf("Stop port :%d\n", slave->port[i]);
437 rte_eth_dev_stop(slave->port[i]);
438 snprintf(buf_name, RTE_MEMPOOL_NAMESIZE, MBUF_NAME, slave->port[i]);
439 pool = rte_mempool_lookup(buf_name);
441 printf("Port %d mempool free object is %u(%u)\n", slave->port[i],
442 rte_mempool_avail_count(pool),
443 (unsigned int)NB_MBUF);
445 printf("Can't find mempool %s\n", buf_name);
447 printf("Start port :%d\n", slave->port[i]);
448 ret = rte_eth_dev_start(slave->port[i]);
456 reset_shared_structures(unsigned slaveid)
459 /* Only port are shared resource here */
460 ret = reset_slave_all_ports(slaveid);
466 * Call this function to re-create resource that needed for slave process that
467 * exited in last instance
470 init_slave_res(unsigned slaveid)
472 struct lcore_resource_struct *slave = &lcore_resource[slaveid];
475 if (!slave->enabled) {
476 printf("Something wrong with lcore=%u enabled=%d\n",slaveid,
481 /* Initialize ring */
482 if (create_ms_ring(slaveid) != 0)
483 rte_exit(EXIT_FAILURE, "failed to create ring for slave %u\n",
486 /* drain un-read buffer if have */
487 while (getcmd(slaveid, &cmd, 1) == 0);
488 while (getcmd(slaveid, &cmd, 0) == 0);
494 recreate_one_slave(unsigned slaveid)
497 /* Re-initialize resource for stalled slave */
498 if ((ret = init_slave_res(slaveid)) != 0) {
499 printf("Init slave=%u failed\n", slaveid);
503 if ((ret = flib_remote_launch(l2fwd_launch_one_lcore, NULL, slaveid))
505 printf("Launch slave %u failed\n", slaveid);
511 * remapping resource belong to slave_id to new lcore that gets from flib_assign_lcore_id(),
512 * used only floating process option applied.
515 * original lcore_id that apply for remapping
518 remapping_slave_resource(unsigned slaveid, unsigned map_id)
521 /* remapping lcore_resource */
522 memcpy(&lcore_resource[map_id], &lcore_resource[slaveid],
523 sizeof(struct lcore_resource_struct));
525 /* remapping lcore_queue_conf */
526 memcpy(&lcore_queue_conf[map_id], &lcore_queue_conf[slaveid],
527 sizeof(struct lcore_queue_conf));
531 reset_pair(unsigned slaveid, unsigned pairid)
534 if ((ret = reset_shared_structures(slaveid)) != 0)
537 if((ret = reset_shared_structures(pairid)) != 0)
541 unsigned map_id = mapping_id[slaveid];
543 if (map_id != INVALID_MAPPING_ID) {
544 printf("%u return mapping id %u\n", slaveid, map_id);
545 flib_free_lcore_id(map_id);
546 mapping_id[slaveid] = INVALID_MAPPING_ID;
549 map_id = mapping_id[pairid];
550 if (map_id != INVALID_MAPPING_ID) {
551 printf("%u return mapping id %u\n", pairid, map_id);
552 flib_free_lcore_id(map_id);
553 mapping_id[pairid] = INVALID_MAPPING_ID;
557 if((ret = recreate_one_slave(slaveid)) != 0)
560 ret = recreate_one_slave(pairid);
567 slave_exit_cb(unsigned slaveid, __attribute__((unused))int stat)
569 struct lcore_resource_struct *slave = &lcore_resource[slaveid];
571 printf("Get slave %u leave info\n", slaveid);
572 if (!slave->enabled) {
573 printf("Lcore=%u not registered for it's exit\n", slaveid);
576 rte_spinlock_lock(&res_lock);
578 /* Change the state and wait master to start them */
579 slave->flags = SLAVE_RECREATE_FLAG;
581 rte_spinlock_unlock(&res_lock);
585 l2fwd_simple_forward(struct rte_mbuf *m, unsigned portid)
587 struct ether_hdr *eth;
591 struct rte_eth_dev_tx_buffer *buffer;
593 dst_port = l2fwd_dst_ports[portid];
594 eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
596 /* 02:00:00:00:00:xx */
597 tmp = ð->d_addr.addr_bytes[0];
598 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
601 ether_addr_copy(&l2fwd_ports_eth_addr[dst_port], ð->s_addr);
603 buffer = tx_buffer[dst_port];
604 sent = rte_eth_tx_buffer(dst_port, 0, buffer, m);
606 port_statistics[dst_port].tx += sent;
609 /* main processing loop */
611 l2fwd_main_loop(void)
613 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
617 uint64_t prev_tsc, diff_tsc, cur_tsc;
618 unsigned i, j, portid, nb_rx;
619 struct lcore_queue_conf *qconf;
620 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S *
622 struct rte_eth_dev_tx_buffer *buffer;
626 lcore_id = rte_lcore_id();
628 qconf = &lcore_queue_conf[lcore_id];
630 if (qconf->n_rx_port == 0) {
631 RTE_LOG(INFO, L2FWD, "lcore %u has nothing to do\n", lcore_id);
635 RTE_LOG(INFO, L2FWD, "entering main loop on lcore %u\n", lcore_id);
637 for (i = 0; i < qconf->n_rx_port; i++) {
638 portid = qconf->rx_port_list[i];
639 RTE_LOG(INFO, L2FWD, " -- lcoreid=%u portid=%u\n", lcore_id,
645 cur_tsc = rte_rdtsc();
647 if (unlikely(getcmd(lcore_id, &cmd, 0) == 0)) {
648 sendcmd(lcore_id, cmd, 0);
650 /* If get stop command, stop forwarding and exit */
651 if (cmd == CMD_STOP) {
657 * TX burst queue drain
659 diff_tsc = cur_tsc - prev_tsc;
660 if (unlikely(diff_tsc > drain_tsc)) {
662 for (i = 0; i < qconf->n_rx_port; i++) {
664 portid = l2fwd_dst_ports[qconf->rx_port_list[i]];
665 buffer = tx_buffer[portid];
667 sent = rte_eth_tx_buffer_flush(portid, 0, buffer);
669 port_statistics[portid].tx += sent;
677 * Read packet from RX queues
679 for (i = 0; i < qconf->n_rx_port; i++) {
681 portid = qconf->rx_port_list[i];
682 nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
683 pkts_burst, MAX_PKT_BURST);
685 port_statistics[portid].rx += nb_rx;
687 for (j = 0; j < nb_rx; j++) {
689 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
690 l2fwd_simple_forward(m, portid);
697 l2fwd_launch_one_lcore(__attribute__((unused)) void *dummy)
699 unsigned lcore_id = rte_lcore_id();
704 /* Change it to floating process, also change it's lcore_id */
705 clear_cpu_affinity();
706 RTE_PER_LCORE(_lcore_id) = 0;
708 if (flib_assign_lcore_id() < 0 ) {
709 printf("flib_assign_lcore_id failed\n");
712 flcore_id = rte_lcore_id();
713 /* Set mapping id, so master can return it after slave exited */
714 mapping_id[lcore_id] = flcore_id;
715 printf("Org lcore_id = %u, cur lcore_id = %u\n",
716 lcore_id, flcore_id);
717 remapping_slave_resource(lcore_id, flcore_id);
722 /* return lcore_id before return */
724 flib_free_lcore_id(rte_lcore_id());
725 mapping_id[lcore_id] = INVALID_MAPPING_ID;
732 l2fwd_usage(const char *prgname)
734 printf("%s [EAL options] -- -p PORTMASK -s COREMASK [-q NQ] -f\n"
735 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
736 " -q NQ: number of queue (=ports) per lcore (default is 1)\n"
737 " -f use floating process which won't bind to any core to run\n"
738 " -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n",
743 l2fwd_parse_portmask(const char *portmask)
748 /* parse hexadecimal string */
749 pm = strtoul(portmask, &end, 16);
750 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
760 l2fwd_parse_nqueue(const char *q_arg)
765 /* parse hexadecimal string */
766 n = strtoul(q_arg, &end, 10);
767 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
771 if (n >= MAX_RX_QUEUE_PER_LCORE)
778 l2fwd_parse_timer_period(const char *q_arg)
783 /* parse number string */
784 n = strtol(q_arg, &end, 10);
785 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
787 if (n >= MAX_TIMER_PERIOD)
793 /* Parse the argument given in the command line of the application */
795 l2fwd_parse_args(int argc, char **argv)
800 char *prgname = argv[0];
801 static struct option lgopts[] = {
808 while ((opt = getopt_long(argc, argvopt, "p:q:T:f",
809 lgopts, &option_index)) != EOF) {
814 l2fwd_enabled_port_mask = l2fwd_parse_portmask(optarg);
815 if (l2fwd_enabled_port_mask == 0) {
816 printf("invalid portmask\n");
817 l2fwd_usage(prgname);
825 l2fwd_rx_queue_per_lcore = l2fwd_parse_nqueue(optarg);
826 if (l2fwd_rx_queue_per_lcore == 0) {
827 printf("invalid queue number\n");
828 l2fwd_usage(prgname);
835 timer_period = l2fwd_parse_timer_period(optarg) * 1000 * TIMER_MILLISECOND;
836 if (timer_period < 0) {
837 printf("invalid timer period\n");
838 l2fwd_usage(prgname);
843 /* use floating process */
850 l2fwd_usage(prgname);
854 l2fwd_usage(prgname);
860 argv[optind-1] = prgname;
863 l2fwd_usage(prgname);
867 optind = 1; /* reset getopt lib */
871 /* Check the link status of all ports in up to 9s, and print them finally */
873 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask)
875 #define CHECK_INTERVAL 100 /* 100ms */
876 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
878 uint8_t 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)
894 "Port%d Link Up- speed %u Mbps- %s\n",
895 portid, link.link_speed,
896 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
897 ("full-duplex") : ("half-duplex\n"));
899 printf("Port %d Link Down\n", portid);
902 /* clear all_ports_up flag if any link down */
903 if (link.link_status == ETH_LINK_DOWN) {
908 /* after finally printing all link status, get out */
912 if (all_ports_up == 0) {
915 rte_delay_ms(CHECK_INTERVAL);
918 /* set the print_flag if all ports up or timeout */
919 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
927 main(int argc, char **argv)
929 struct lcore_queue_conf *qconf;
930 struct rte_eth_dev_info dev_info;
933 uint16_t nb_ports_available;
934 uint16_t portid, last_port;
935 unsigned rx_lcore_id;
936 unsigned nb_ports_in_mask = 0;
938 uint64_t prev_tsc, diff_tsc, cur_tsc, timer_tsc;
940 /* Save cpu_affinity first, restore it in case it's floating process option */
941 if (get_cpu_affinity() != 0)
942 rte_exit(EXIT_FAILURE, "get_cpu_affinity error\n");
944 /* Also tries to set cpu affinity to detect whether it will fail in child process */
945 if(clear_cpu_affinity() != 0)
946 rte_exit(EXIT_FAILURE, "clear_cpu_affinity error\n");
949 ret = rte_eal_init(argc, argv);
951 rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
955 /* parse application arguments (after the EAL ones) */
956 ret = l2fwd_parse_args(argc, argv);
958 rte_exit(EXIT_FAILURE, "Invalid L2FWD arguments\n");
961 if (flib_init() != 0)
962 rte_exit(EXIT_FAILURE, "flib init error");
965 * Allocated structures that slave lcore would change. For those that slaves are
966 * read only, needn't use malloc to share and global or static variables is ok since
967 * slave inherit all the knowledge that master initialized.
969 if (l2fwd_malloc_shared_struct() != 0)
970 rte_exit(EXIT_FAILURE, "malloc mem failed\n");
972 /* Initialize lcore_resource structures */
973 memset(lcore_resource, 0, sizeof(lcore_resource));
974 for (i = 0; i < RTE_MAX_LCORE; i++)
975 lcore_resource[i].lcore_id = i;
977 nb_ports = rte_eth_dev_count();
979 rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
981 /* create the mbuf pool */
982 for (portid = 0; portid < nb_ports; portid++) {
983 /* skip ports that are not enabled */
984 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
986 char buf_name[RTE_MEMPOOL_NAMESIZE];
987 snprintf(buf_name, RTE_MEMPOOL_NAMESIZE, MBUF_NAME, portid);
988 l2fwd_pktmbuf_pool[portid] =
989 rte_pktmbuf_pool_create(buf_name, NB_MBUF, 32,
990 0, MBUF_DATA_SIZE, rte_socket_id());
991 if (l2fwd_pktmbuf_pool[portid] == NULL)
992 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
994 printf("Create mbuf %s\n", buf_name);
997 /* reset l2fwd_dst_ports */
998 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++)
999 l2fwd_dst_ports[portid] = 0;
1003 * Each logical core is assigned a dedicated TX queue on each port.
1005 for (portid = 0; portid < nb_ports; portid++) {
1006 /* skip ports that are not enabled */
1007 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
1010 if (nb_ports_in_mask % 2) {
1011 l2fwd_dst_ports[portid] = last_port;
1012 l2fwd_dst_ports[last_port] = portid;
1019 rte_eth_dev_info_get(portid, &dev_info);
1021 if (nb_ports_in_mask % 2) {
1022 printf("Notice: odd number of ports in portmask.\n");
1023 l2fwd_dst_ports[last_port] = last_port;
1029 /* Initialize the port/queue configuration of each logical core */
1030 for (portid = 0; portid < nb_ports; portid++) {
1031 struct lcore_resource_struct *res;
1032 /* skip ports that are not enabled */
1033 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
1036 /* get the lcore_id for this port */
1037 /* skip master lcore */
1038 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
1039 rte_get_master_lcore() == rx_lcore_id ||
1040 lcore_queue_conf[rx_lcore_id].n_rx_port ==
1041 l2fwd_rx_queue_per_lcore) {
1044 if (rx_lcore_id >= RTE_MAX_LCORE)
1045 rte_exit(EXIT_FAILURE, "Not enough cores\n");
1048 if (qconf != &lcore_queue_conf[rx_lcore_id])
1049 /* Assigned a new logical core in the loop above. */
1050 qconf = &lcore_queue_conf[rx_lcore_id];
1052 qconf->rx_port_list[qconf->n_rx_port] = portid;
1055 /* Save the port resource info into lcore_resource strucutres */
1056 res = &lcore_resource[rx_lcore_id];
1058 res->port[res->port_num++] = portid;
1060 printf("Lcore %u: RX port %u\n", rx_lcore_id, (unsigned) portid);
1063 nb_ports_available = nb_ports;
1065 /* Initialise each port */
1066 for (portid = 0; portid < nb_ports; portid++) {
1067 /* skip ports that are not enabled */
1068 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0) {
1069 printf("Skipping disabled port %u\n", (unsigned) portid);
1070 nb_ports_available--;
1074 printf("Initializing port %u... ", (unsigned) portid);
1076 ret = rte_eth_dev_configure(portid, 1, 1, &port_conf);
1078 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
1079 ret, (unsigned) portid);
1081 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1084 rte_exit(EXIT_FAILURE,
1085 "rte_eth_dev_adjust_nb_rx_tx_desc: err=%d, port=%u\n",
1086 ret, (unsigned) portid);
1088 rte_eth_macaddr_get(portid,&l2fwd_ports_eth_addr[portid]);
1090 /* init one RX queue */
1092 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
1093 rte_eth_dev_socket_id(portid),
1095 l2fwd_pktmbuf_pool[portid]);
1097 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup:err=%d, port=%u\n",
1098 ret, (unsigned) portid);
1100 /* init one TX queue on each port */
1102 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
1103 rte_eth_dev_socket_id(portid),
1106 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup:err=%d, port=%u\n",
1107 ret, (unsigned) portid);
1109 /* Initialize TX buffers */
1110 tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
1111 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
1112 rte_eth_dev_socket_id(portid));
1113 if (tx_buffer[portid] == NULL)
1114 rte_exit(EXIT_FAILURE, "Cannot allocate buffer for tx on port %u\n",
1117 rte_eth_tx_buffer_init(tx_buffer[portid], MAX_PKT_BURST);
1119 ret = rte_eth_tx_buffer_set_err_callback(tx_buffer[portid],
1120 rte_eth_tx_buffer_count_callback,
1121 &port_statistics[portid].dropped);
1123 rte_exit(EXIT_FAILURE, "Cannot set error callback for "
1124 "tx buffer on port %u\n", (unsigned) portid);
1127 ret = rte_eth_dev_start(portid);
1129 rte_exit(EXIT_FAILURE, "rte_eth_dev_start:err=%d, port=%u\n",
1130 ret, (unsigned) portid);
1134 rte_eth_promiscuous_enable(portid);
1136 printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
1138 l2fwd_ports_eth_addr[portid].addr_bytes[0],
1139 l2fwd_ports_eth_addr[portid].addr_bytes[1],
1140 l2fwd_ports_eth_addr[portid].addr_bytes[2],
1141 l2fwd_ports_eth_addr[portid].addr_bytes[3],
1142 l2fwd_ports_eth_addr[portid].addr_bytes[4],
1143 l2fwd_ports_eth_addr[portid].addr_bytes[5]);
1145 /* initialize port stats */
1146 //memset(&port_statistics, 0, sizeof(port_statistics));
1149 if (!nb_ports_available) {
1150 rte_exit(EXIT_FAILURE,
1151 "All available ports are disabled. Please set portmask.\n");
1154 check_all_ports_link_status(nb_ports, l2fwd_enabled_port_mask);
1156 /* Record pair lcore */
1158 * Since l2fwd example would create pair between different neighbour port, that's
1159 * port 0 receive and forward to port 1, the same to port 1, these 2 ports will have
1160 * dependency. If one port stopped working (killed, for example), the port need to
1161 * be stopped/started again. During the time, another port need to wait until stop/start
1162 * procedure completed. So, record the pair relationship for those lcores working
1165 for (portid = 0; portid < nb_ports; portid++) {
1167 unsigned lcore = 0, pair_lcore = 0;
1168 unsigned j, find_lcore, find_pair_lcore;
1169 /* skip ports that are not enabled */
1170 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
1173 /* Find pair ports' lcores */
1174 find_lcore = find_pair_lcore = 0;
1175 pair_port = l2fwd_dst_ports[portid];
1176 for (i = 0; i < RTE_MAX_LCORE; i++) {
1177 if (!rte_lcore_is_enabled(i))
1179 for (j = 0; j < lcore_queue_conf[i].n_rx_port;j++) {
1180 if (lcore_queue_conf[i].rx_port_list[j] == portid) {
1185 if (lcore_queue_conf[i].rx_port_list[j] == pair_port) {
1187 find_pair_lcore = 1;
1191 if (find_lcore && find_pair_lcore)
1194 if (!find_lcore || !find_pair_lcore)
1195 rte_exit(EXIT_FAILURE, "Not find port=%d pair\n", portid);
1197 printf("lcore %u and %u paired\n", lcore, pair_lcore);
1198 lcore_resource[lcore].pair_id = pair_lcore;
1199 lcore_resource[pair_lcore].pair_id = lcore;
1202 /* Create message buffer for all master and slave */
1203 message_pool = rte_mempool_create("ms_msg_pool",
1204 NB_CORE_MSGBUF * RTE_MAX_LCORE,
1205 sizeof(enum l2fwd_cmd), NB_CORE_MSGBUF / 2,
1206 0, NULL, NULL, NULL, NULL, rte_socket_id(), 0);
1208 if (message_pool == NULL)
1209 rte_exit(EXIT_FAILURE, "Create msg mempool failed\n");
1211 /* Create ring for each master and slave pair, also register cb when slave leaves */
1212 for (i = 0; i < RTE_MAX_LCORE; i++) {
1214 * Only create ring and register slave_exit cb in case that core involved into
1217 if (lcore_resource[i].enabled) {
1218 /* Create ring for master and slave communication */
1219 ret = create_ms_ring(i);
1221 rte_exit(EXIT_FAILURE, "Create ring for lcore=%u failed",
1224 if (flib_register_slave_exit_notify(i,
1225 slave_exit_cb) != 0)
1226 rte_exit(EXIT_FAILURE,
1227 "Register master_trace_slave_exit failed");
1231 /* launch per-lcore init on every lcore except master */
1232 flib_mp_remote_launch(l2fwd_launch_one_lcore, NULL, SKIP_MASTER);
1234 /* print statistics 10 second */
1235 prev_tsc = cur_tsc = rte_rdtsc();
1239 cur_tsc = rte_rdtsc();
1240 diff_tsc = cur_tsc - prev_tsc;
1241 /* if timer is enabled */
1242 if (timer_period > 0) {
1244 /* advance the timer */
1245 timer_tsc += diff_tsc;
1247 /* if timer has reached its timeout */
1248 if (unlikely(timer_tsc >= (uint64_t) timer_period)) {
1251 /* reset the timer */
1258 /* Check any slave need restart or recreate */
1259 rte_spinlock_lock(&res_lock);
1260 for (i = 0; i < RTE_MAX_LCORE; i++) {
1261 struct lcore_resource_struct *res = &lcore_resource[i];
1262 struct lcore_resource_struct *pair = &lcore_resource[res->pair_id];
1264 /* If find slave exited, try to reset pair */
1265 if (res->enabled && res->flags && pair->enabled) {
1267 master_sendcmd_with_ack(pair->lcore_id, CMD_STOP);
1268 rte_spinlock_unlock(&res_lock);
1270 rte_spinlock_lock(&res_lock);
1274 if (reset_pair(res->lcore_id, pair->lcore_id) != 0)
1275 rte_exit(EXIT_FAILURE, "failed to reset slave");
1280 rte_spinlock_unlock(&res_lock);