4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of Intel Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
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_per_lcore.h>
57 #include <rte_launch.h>
58 #include <rte_atomic.h>
59 #include <rte_spinlock.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>
72 #include <rte_mempool.h>
74 #include <rte_malloc.h>
78 #define RTE_LOGTYPE_L2FWD RTE_LOGTYPE_USER1
79 #define MBUF_NAME "mbuf_pool_%d"
81 (RTE_MBUF_DEFAULT_DATAROOM + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
83 #define RING_MASTER_NAME "l2fwd_ring_m2s_"
84 #define RING_SLAVE_NAME "l2fwd_ring_s2m_"
85 #define MAX_NAME_LEN 32
86 /* RECREATE flag indicate needs initialize resource and launch slave_core again */
87 #define SLAVE_RECREATE_FLAG 0x1
88 /* RESTART flag indicate needs restart port and send START command again */
89 #define SLAVE_RESTART_FLAG 0x2
90 #define INVALID_MAPPING_ID ((unsigned)LCORE_ID_ANY)
91 /* Maximum message buffer per slave */
92 #define NB_CORE_MSGBUF 32
98 #define MAX_PKT_BURST 32
99 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
102 * Configurable number of RX/TX ring descriptors
104 #define RTE_TEST_RX_DESC_DEFAULT 128
105 #define RTE_TEST_TX_DESC_DEFAULT 512
106 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
107 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
109 /* ethernet addresses of ports */
110 static struct ether_addr l2fwd_ports_eth_addr[RTE_MAX_ETHPORTS];
112 /* mask of enabled ports */
113 static uint32_t l2fwd_enabled_port_mask = 0;
115 /* list of enabled ports */
116 static uint32_t l2fwd_dst_ports[RTE_MAX_ETHPORTS];
118 static unsigned int l2fwd_rx_queue_per_lcore = 1;
122 struct rte_mbuf *m_table[MAX_PKT_BURST];
125 #define MAX_RX_QUEUE_PER_LCORE 16
126 #define MAX_TX_QUEUE_PER_PORT 16
127 struct lcore_queue_conf {
129 unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
130 struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
132 } __rte_cache_aligned;
133 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
135 struct lcore_resource_struct {
136 int enabled; /* Only set in case this lcore involved into packet forwarding */
137 int flags; /* Set only slave need to restart or recreate */
138 unsigned lcore_id; /* lcore ID */
139 unsigned pair_id; /* dependency lcore ID on port */
140 char ring_name[2][MAX_NAME_LEN];
141 /* ring[0] for master send cmd, slave read */
142 /* ring[1] for slave send ack, master read */
143 struct rte_ring *ring[2];
144 int port_num; /* Total port numbers */
145 uint8_t port[RTE_MAX_ETHPORTS]; /* Port id for that lcore to receive packets */
146 }__attribute__((packed)) __rte_cache_aligned;
148 static struct lcore_resource_struct lcore_resource[RTE_MAX_LCORE];
149 static struct rte_mempool *message_pool;
150 static rte_spinlock_t res_lock = RTE_SPINLOCK_INITIALIZER;
151 /* use floating processes */
152 static int float_proc = 0;
153 /* Save original cpu affinity */
159 static const struct rte_eth_conf port_conf = {
162 .header_split = 0, /**< Header Split disabled */
163 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
164 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
165 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
166 .hw_strip_crc = 0, /**< CRC stripped by hardware */
169 .mq_mode = ETH_MQ_TX_NONE,
173 static struct rte_mempool * l2fwd_pktmbuf_pool[RTE_MAX_ETHPORTS];
175 /* Per-port statistics struct */
176 struct l2fwd_port_statistics {
180 } __rte_cache_aligned;
181 struct l2fwd_port_statistics *port_statistics;
183 * pointer to lcore ID mapping array, used to return lcore id in case slave
184 * process exited unexpectedly, use only floating process option applied
186 unsigned *mapping_id;
188 /* A tsc-based timer responsible for triggering statistics printout */
189 #define TIMER_MILLISECOND 2000000ULL /* around 1ms at 2 Ghz */
190 #define MAX_TIMER_PERIOD 86400 /* 1 day max */
191 static int64_t timer_period = 10 * TIMER_MILLISECOND * 1000; /* default period is 10 seconds */
193 static int l2fwd_launch_one_lcore(void *dummy);
195 /* Print out statistics on packets dropped */
199 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
202 total_packets_dropped = 0;
203 total_packets_tx = 0;
204 total_packets_rx = 0;
206 const char clr[] = { 27, '[', '2', 'J', '\0' };
207 const char topLeft[] = { 27, '[', '1', ';', '1', 'H','\0' };
209 /* Clear screen and move to top left */
210 printf("%s%s", clr, topLeft);
212 printf("\nPort statistics ====================================");
214 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
215 /* skip disabled ports */
216 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
218 printf("\nStatistics for port %u ------------------------------"
219 "\nPackets sent: %24"PRIu64
220 "\nPackets received: %20"PRIu64
221 "\nPackets dropped: %21"PRIu64,
223 port_statistics[portid].tx,
224 port_statistics[portid].rx,
225 port_statistics[portid].dropped);
227 total_packets_dropped += port_statistics[portid].dropped;
228 total_packets_tx += port_statistics[portid].tx;
229 total_packets_rx += port_statistics[portid].rx;
231 printf("\nAggregate statistics ==============================="
232 "\nTotal packets sent: %18"PRIu64
233 "\nTotal packets received: %14"PRIu64
234 "\nTotal packets dropped: %15"PRIu64,
237 total_packets_dropped);
238 printf("\n====================================================\n");
242 clear_cpu_affinity(void)
246 s = sched_setaffinity(0, cpu_aff.size, &cpu_aff.set);
248 printf("sched_setaffinity failed:%s\n", strerror(errno));
256 get_cpu_affinity(void)
260 cpu_aff.size = sizeof(cpu_set_t);
261 CPU_ZERO(&cpu_aff.set);
263 s = sched_getaffinity(0, cpu_aff.size, &cpu_aff.set);
265 printf("sched_getaffinity failed:%s\n", strerror(errno));
273 * This fnciton demonstrates the approach to create ring in first instance
274 * or re-attach an existed ring in later instance.
276 static struct rte_ring *
277 create_ring(const char *name, unsigned count,
278 int socket_id,unsigned flags)
280 struct rte_ring *ring;
285 /* If already create, just attached it */
286 if (likely((ring = rte_ring_lookup(name)) != NULL))
289 /* First call it, create one */
290 return rte_ring_create(name, count, socket_id, flags);
293 /* Malloc with rte_malloc on structures that shared by master and slave */
295 l2fwd_malloc_shared_struct(void)
297 port_statistics = rte_zmalloc("port_stat",
298 sizeof(struct l2fwd_port_statistics) * RTE_MAX_ETHPORTS,
300 if (port_statistics == NULL)
303 /* allocate mapping_id array */
306 mapping_id = rte_malloc("mapping_id", sizeof(unsigned) * RTE_MAX_LCORE,
309 if (mapping_id == NULL)
312 for (i = 0 ;i < RTE_MAX_LCORE; i++)
313 mapping_id[i] = INVALID_MAPPING_ID;
318 /* Create ring which used for communicate among master and slave */
320 create_ms_ring(unsigned slaveid)
322 unsigned flag = RING_F_SP_ENQ | RING_F_SC_DEQ;
323 struct lcore_resource_struct *res = &lcore_resource[slaveid];
324 unsigned socketid = rte_socket_id();
326 /* Always assume create ring on master socket_id */
327 /* Default only create a ring size 32 */
328 snprintf(res->ring_name[0], MAX_NAME_LEN, "%s%u",
329 RING_MASTER_NAME, slaveid);
330 if ((res->ring[0] = create_ring(res->ring_name[0], NB_CORE_MSGBUF,
331 socketid, flag)) == NULL) {
332 printf("Create m2s ring %s failed\n", res->ring_name[0]);
336 snprintf(res->ring_name[1], MAX_NAME_LEN, "%s%u",
337 RING_SLAVE_NAME, slaveid);
338 if ((res->ring[1] = create_ring(res->ring_name[1], NB_CORE_MSGBUF,
339 socketid, flag)) == NULL) {
340 printf("Create s2m ring %s failed\n", res->ring_name[1]);
347 /* send command to pair in paired master and slave ring */
349 sendcmd(unsigned slaveid, enum l2fwd_cmd cmd, int is_master)
351 struct lcore_resource_struct *res = &lcore_resource[slaveid];
355 /* Only check master, it must be enabled and running if it is slave */
356 if (is_master && !res->enabled)
359 if (res->ring[fd] == NULL)
362 if (rte_mempool_get(message_pool, &msg) < 0) {
363 printf("Error to get message buffer\n");
367 *(enum l2fwd_cmd *)msg = cmd;
369 if (rte_ring_enqueue(res->ring[fd], msg) != 0) {
370 printf("Enqueue error\n");
371 rte_mempool_put(message_pool, msg);
378 /* Get command from pair in paired master and slave ring */
380 getcmd(unsigned slaveid, enum l2fwd_cmd *cmd, int is_master)
382 struct lcore_resource_struct *res = &lcore_resource[slaveid];
384 int fd = !!is_master;
386 /* Only check master, it must be enabled and running if it is slave */
387 if (is_master && (!res->enabled))
390 if (res->ring[fd] == NULL)
393 ret = rte_ring_dequeue(res->ring[fd], &msg);
396 *cmd = *(enum l2fwd_cmd *)msg;
397 rte_mempool_put(message_pool, msg);
402 /* Master send command to slave and wait until ack received or error met */
404 master_sendcmd_with_ack(unsigned slaveid, enum l2fwd_cmd cmd)
406 enum l2fwd_cmd ack_cmd;
409 if (sendcmd(slaveid, cmd, 1) != 0)
410 rte_exit(EXIT_FAILURE, "Failed to send message\n");
414 ret = getcmd(slaveid, &ack_cmd, 1);
415 if (ret == 0 && cmd == ack_cmd)
418 /* If slave not running yet, return an error */
419 if (flib_query_slave_status(slaveid) != ST_RUN) {
428 /* restart all port that assigned to that slave lcore */
430 reset_slave_all_ports(unsigned slaveid)
432 struct lcore_resource_struct *slave = &lcore_resource[slaveid];
435 /* stop/start port */
436 for (i = 0; i < slave->port_num; i++) {
437 char buf_name[RTE_MEMPOOL_NAMESIZE];
438 struct rte_mempool *pool;
439 printf("Stop port :%d\n", slave->port[i]);
440 rte_eth_dev_stop(slave->port[i]);
441 snprintf(buf_name, RTE_MEMPOOL_NAMESIZE, MBUF_NAME, slave->port[i]);
442 pool = rte_mempool_lookup(buf_name);
444 printf("Port %d mempool free object is %u(%u)\n", slave->port[i],
445 rte_mempool_count(pool), (unsigned)NB_MBUF);
447 printf("Can't find mempool %s\n", buf_name);
449 printf("Start port :%d\n", slave->port[i]);
450 ret = rte_eth_dev_start(slave->port[i]);
458 reset_shared_structures(unsigned slaveid)
461 /* Only port are shared resource here */
462 ret = reset_slave_all_ports(slaveid);
468 * Call this function to re-create resource that needed for slave process that
469 * exited in last instance
472 init_slave_res(unsigned slaveid)
474 struct lcore_resource_struct *slave = &lcore_resource[slaveid];
477 if (!slave->enabled) {
478 printf("Something wrong with lcore=%u enabled=%d\n",slaveid,
483 /* Initialize ring */
484 if (create_ms_ring(slaveid) != 0)
485 rte_exit(EXIT_FAILURE, "failed to create ring for slave %u\n",
488 /* drain un-read buffer if have */
489 while (getcmd(slaveid, &cmd, 1) == 0);
490 while (getcmd(slaveid, &cmd, 0) == 0);
496 recreate_one_slave(unsigned slaveid)
499 /* Re-initialize resource for stalled slave */
500 if ((ret = init_slave_res(slaveid)) != 0) {
501 printf("Init slave=%u failed\n", slaveid);
505 if ((ret = flib_remote_launch(l2fwd_launch_one_lcore, NULL, slaveid))
507 printf("Launch slave %u failed\n", slaveid);
513 * remapping resource belong to slave_id to new lcore that gets from flib_assign_lcore_id(),
514 * used only floating process option applied.
517 * original lcore_id that apply for remapping
520 remapping_slave_resource(unsigned slaveid, unsigned map_id)
523 /* remapping lcore_resource */
524 memcpy(&lcore_resource[map_id], &lcore_resource[slaveid],
525 sizeof(struct lcore_resource_struct));
527 /* remapping lcore_queue_conf */
528 memcpy(&lcore_queue_conf[map_id], &lcore_queue_conf[slaveid],
529 sizeof(struct lcore_queue_conf));
533 reset_pair(unsigned slaveid, unsigned pairid)
536 if ((ret = reset_shared_structures(slaveid)) != 0)
539 if((ret = reset_shared_structures(pairid)) != 0)
543 unsigned map_id = mapping_id[slaveid];
545 if (map_id != INVALID_MAPPING_ID) {
546 printf("%u return mapping id %u\n", slaveid, map_id);
547 flib_free_lcore_id(map_id);
548 mapping_id[slaveid] = INVALID_MAPPING_ID;
551 map_id = mapping_id[pairid];
552 if (map_id != INVALID_MAPPING_ID) {
553 printf("%u return mapping id %u\n", pairid, map_id);
554 flib_free_lcore_id(map_id);
555 mapping_id[pairid] = INVALID_MAPPING_ID;
559 if((ret = recreate_one_slave(slaveid)) != 0)
562 ret = recreate_one_slave(pairid);
569 slave_exit_cb(unsigned slaveid, __attribute__((unused))int stat)
571 struct lcore_resource_struct *slave = &lcore_resource[slaveid];
573 printf("Get slave %u leave info\n", slaveid);
574 if (!slave->enabled) {
575 printf("Lcore=%u not registered for it's exit\n", slaveid);
578 rte_spinlock_lock(&res_lock);
580 /* Change the state and wait master to start them */
581 slave->flags = SLAVE_RECREATE_FLAG;
583 rte_spinlock_unlock(&res_lock);
586 /* Send the packet on an output interface */
588 l2fwd_send_burst(struct lcore_queue_conf *qconf, unsigned n, uint8_t port)
590 struct rte_mbuf **m_table;
594 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
596 ret = rte_eth_tx_burst(port, (uint16_t) queueid, m_table, (uint16_t) n);
597 port_statistics[port].tx += ret;
598 if (unlikely(ret < n)) {
599 port_statistics[port].dropped += (n - ret);
601 rte_pktmbuf_free(m_table[ret]);
608 /* Send the packet on an output interface */
610 l2fwd_send_packet(struct rte_mbuf *m, uint8_t port)
612 unsigned lcore_id, len;
613 struct lcore_queue_conf *qconf;
615 lcore_id = rte_lcore_id();
617 qconf = &lcore_queue_conf[lcore_id];
618 len = qconf->tx_mbufs[port].len;
619 qconf->tx_mbufs[port].m_table[len] = m;
622 /* enough pkts to be sent */
623 if (unlikely(len == MAX_PKT_BURST)) {
624 l2fwd_send_burst(qconf, MAX_PKT_BURST, port);
628 qconf->tx_mbufs[port].len = len;
633 l2fwd_simple_forward(struct rte_mbuf *m, unsigned portid)
635 struct ether_hdr *eth;
639 dst_port = l2fwd_dst_ports[portid];
640 eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
642 /* 02:00:00:00:00:xx */
643 tmp = ð->d_addr.addr_bytes[0];
644 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
647 ether_addr_copy(&l2fwd_ports_eth_addr[dst_port], ð->s_addr);
649 l2fwd_send_packet(m, (uint8_t) dst_port);
652 /* main processing loop */
654 l2fwd_main_loop(void)
656 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
659 uint64_t prev_tsc, diff_tsc, cur_tsc;
660 unsigned i, j, portid, nb_rx;
661 struct lcore_queue_conf *qconf;
662 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
666 lcore_id = rte_lcore_id();
668 qconf = &lcore_queue_conf[lcore_id];
670 if (qconf->n_rx_port == 0) {
671 RTE_LOG(INFO, L2FWD, "lcore %u has nothing to do\n", lcore_id);
675 RTE_LOG(INFO, L2FWD, "entering main loop on lcore %u\n", lcore_id);
677 for (i = 0; i < qconf->n_rx_port; i++) {
678 portid = qconf->rx_port_list[i];
679 RTE_LOG(INFO, L2FWD, " -- lcoreid=%u portid=%u\n", lcore_id,
685 cur_tsc = rte_rdtsc();
687 if (unlikely(getcmd(lcore_id, &cmd, 0) == 0)) {
688 sendcmd(lcore_id, cmd, 0);
690 /* If get stop command, stop forwarding and exit */
691 if (cmd == CMD_STOP) {
697 * TX burst queue drain
699 diff_tsc = cur_tsc - prev_tsc;
700 if (unlikely(diff_tsc > drain_tsc)) {
702 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
703 if (qconf->tx_mbufs[portid].len == 0)
705 l2fwd_send_burst(&lcore_queue_conf[lcore_id],
706 qconf->tx_mbufs[portid].len,
708 qconf->tx_mbufs[portid].len = 0;
713 * Read packet from RX queues
715 for (i = 0; i < qconf->n_rx_port; i++) {
717 portid = qconf->rx_port_list[i];
718 nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
719 pkts_burst, MAX_PKT_BURST);
721 port_statistics[portid].rx += nb_rx;
723 for (j = 0; j < nb_rx; j++) {
725 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
726 l2fwd_simple_forward(m, portid);
733 l2fwd_launch_one_lcore(__attribute__((unused)) void *dummy)
735 unsigned lcore_id = rte_lcore_id();
740 /* Change it to floating process, also change it's lcore_id */
741 clear_cpu_affinity();
742 RTE_PER_LCORE(_lcore_id) = 0;
744 if (flib_assign_lcore_id() < 0 ) {
745 printf("flib_assign_lcore_id failed\n");
748 flcore_id = rte_lcore_id();
749 /* Set mapping id, so master can return it after slave exited */
750 mapping_id[lcore_id] = flcore_id;
751 printf("Org lcore_id = %u, cur lcore_id = %u\n",
752 lcore_id, flcore_id);
753 remapping_slave_resource(lcore_id, flcore_id);
758 /* return lcore_id before return */
760 flib_free_lcore_id(rte_lcore_id());
761 mapping_id[lcore_id] = INVALID_MAPPING_ID;
768 l2fwd_usage(const char *prgname)
770 printf("%s [EAL options] -- -p PORTMASK -s COREMASK [-q NQ] -f\n"
771 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
772 " -q NQ: number of queue (=ports) per lcore (default is 1)\n"
773 " -f use floating process which won't bind to any core to run\n"
774 " -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n",
779 l2fwd_parse_portmask(const char *portmask)
784 /* parse hexadecimal string */
785 pm = strtoul(portmask, &end, 16);
786 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
796 l2fwd_parse_nqueue(const char *q_arg)
801 /* parse hexadecimal string */
802 n = strtoul(q_arg, &end, 10);
803 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
807 if (n >= MAX_RX_QUEUE_PER_LCORE)
814 l2fwd_parse_timer_period(const char *q_arg)
819 /* parse number string */
820 n = strtol(q_arg, &end, 10);
821 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
823 if (n >= MAX_TIMER_PERIOD)
829 /* Parse the argument given in the command line of the application */
831 l2fwd_parse_args(int argc, char **argv)
836 char *prgname = argv[0];
837 static struct option lgopts[] = {
844 while ((opt = getopt_long(argc, argvopt, "p:q:T:f",
845 lgopts, &option_index)) != EOF) {
850 l2fwd_enabled_port_mask = l2fwd_parse_portmask(optarg);
851 if (l2fwd_enabled_port_mask == 0) {
852 printf("invalid portmask\n");
853 l2fwd_usage(prgname);
861 l2fwd_rx_queue_per_lcore = l2fwd_parse_nqueue(optarg);
862 if (l2fwd_rx_queue_per_lcore == 0) {
863 printf("invalid queue number\n");
864 l2fwd_usage(prgname);
871 timer_period = l2fwd_parse_timer_period(optarg) * 1000 * TIMER_MILLISECOND;
872 if (timer_period < 0) {
873 printf("invalid timer period\n");
874 l2fwd_usage(prgname);
879 /* use floating process */
886 l2fwd_usage(prgname);
890 l2fwd_usage(prgname);
896 argv[optind-1] = prgname;
899 l2fwd_usage(prgname);
903 optind = 0; /* reset getopt lib */
907 /* Check the link status of all ports in up to 9s, and print them finally */
909 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
911 #define CHECK_INTERVAL 100 /* 100ms */
912 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
913 uint8_t portid, count, all_ports_up, print_flag = 0;
914 struct rte_eth_link link;
916 printf("\nChecking link status");
918 for (count = 0; count <= MAX_CHECK_TIME; count++) {
920 for (portid = 0; portid < port_num; portid++) {
921 if ((port_mask & (1 << portid)) == 0)
923 memset(&link, 0, sizeof(link));
924 rte_eth_link_get_nowait(portid, &link);
925 /* print link status if flag set */
926 if (print_flag == 1) {
927 if (link.link_status)
928 printf("Port %d Link Up - speed %u "
929 "Mbps - %s\n", (uint8_t)portid,
930 (unsigned)link.link_speed,
931 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
932 ("full-duplex") : ("half-duplex\n"));
934 printf("Port %d Link Down\n",
938 /* clear all_ports_up flag if any link down */
939 if (link.link_status == 0) {
944 /* after finally printing all link status, get out */
948 if (all_ports_up == 0) {
951 rte_delay_ms(CHECK_INTERVAL);
954 /* set the print_flag if all ports up or timeout */
955 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
963 main(int argc, char **argv)
965 struct lcore_queue_conf *qconf;
966 struct rte_eth_dev_info dev_info;
969 uint8_t nb_ports_available;
970 uint8_t portid, last_port;
971 unsigned rx_lcore_id;
972 unsigned nb_ports_in_mask = 0;
975 uint64_t prev_tsc, diff_tsc, cur_tsc, timer_tsc;
977 /* Save cpu_affinity first, restore it in case it's floating process option */
978 if (get_cpu_affinity() != 0)
979 rte_exit(EXIT_FAILURE, "get_cpu_affinity error\n");
981 /* Also tries to set cpu affinity to detect whether it will fail in child process */
982 if(clear_cpu_affinity() != 0)
983 rte_exit(EXIT_FAILURE, "clear_cpu_affinity error\n");
986 ret = rte_eal_init(argc, argv);
988 rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
992 /* parse application arguments (after the EAL ones) */
993 ret = l2fwd_parse_args(argc, argv);
995 rte_exit(EXIT_FAILURE, "Invalid L2FWD arguments\n");
998 if (flib_init() != 0)
999 rte_exit(EXIT_FAILURE, "flib init error");
1002 * Allocated structures that slave lcore would change. For those that slaves are
1003 * read only, needn't use malloc to share and global or static variables is ok since
1004 * slave inherit all the knowledge that master initialized.
1006 if (l2fwd_malloc_shared_struct() != 0)
1007 rte_exit(EXIT_FAILURE, "malloc mem failed\n");
1009 /* Initialize lcore_resource structures */
1010 memset(lcore_resource, 0, sizeof(lcore_resource));
1011 for (i = 0; i < RTE_MAX_LCORE; i++)
1012 lcore_resource[i].lcore_id = i;
1014 nb_ports = rte_eth_dev_count();
1016 rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
1018 if (nb_ports > RTE_MAX_ETHPORTS)
1019 nb_ports = RTE_MAX_ETHPORTS;
1021 /* create the mbuf pool */
1022 for (portid = 0; portid < nb_ports; portid++) {
1023 /* skip ports that are not enabled */
1024 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
1026 char buf_name[RTE_MEMPOOL_NAMESIZE];
1027 flags = MEMPOOL_F_SP_PUT | MEMPOOL_F_SC_GET;
1028 snprintf(buf_name, RTE_MEMPOOL_NAMESIZE, MBUF_NAME, portid);
1029 l2fwd_pktmbuf_pool[portid] =
1030 rte_mempool_create(buf_name, NB_MBUF,
1032 sizeof(struct rte_pktmbuf_pool_private),
1033 rte_pktmbuf_pool_init, NULL,
1034 rte_pktmbuf_init, NULL,
1035 rte_socket_id(), flags);
1036 if (l2fwd_pktmbuf_pool[portid] == NULL)
1037 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool\n");
1039 printf("Create mbuf %s\n", buf_name);
1042 /* reset l2fwd_dst_ports */
1043 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++)
1044 l2fwd_dst_ports[portid] = 0;
1048 * Each logical core is assigned a dedicated TX queue on each port.
1050 for (portid = 0; portid < nb_ports; portid++) {
1051 /* skip ports that are not enabled */
1052 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
1055 if (nb_ports_in_mask % 2) {
1056 l2fwd_dst_ports[portid] = last_port;
1057 l2fwd_dst_ports[last_port] = portid;
1064 rte_eth_dev_info_get(portid, &dev_info);
1066 if (nb_ports_in_mask % 2) {
1067 printf("Notice: odd number of ports in portmask.\n");
1068 l2fwd_dst_ports[last_port] = last_port;
1074 /* Initialize the port/queue configuration of each logical core */
1075 for (portid = 0; portid < nb_ports; portid++) {
1076 struct lcore_resource_struct *res;
1077 /* skip ports that are not enabled */
1078 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
1081 /* get the lcore_id for this port */
1082 /* skip master lcore */
1083 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
1084 rte_get_master_lcore() == rx_lcore_id ||
1085 lcore_queue_conf[rx_lcore_id].n_rx_port ==
1086 l2fwd_rx_queue_per_lcore) {
1089 if (rx_lcore_id >= RTE_MAX_LCORE)
1090 rte_exit(EXIT_FAILURE, "Not enough cores\n");
1093 if (qconf != &lcore_queue_conf[rx_lcore_id])
1094 /* Assigned a new logical core in the loop above. */
1095 qconf = &lcore_queue_conf[rx_lcore_id];
1097 qconf->rx_port_list[qconf->n_rx_port] = portid;
1100 /* Save the port resource info into lcore_resource strucutres */
1101 res = &lcore_resource[rx_lcore_id];
1103 res->port[res->port_num++] = portid;
1105 printf("Lcore %u: RX port %u\n", rx_lcore_id, (unsigned) portid);
1108 nb_ports_available = nb_ports;
1110 /* Initialise each port */
1111 for (portid = 0; portid < nb_ports; portid++) {
1112 /* skip ports that are not enabled */
1113 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0) {
1114 printf("Skipping disabled port %u\n", (unsigned) portid);
1115 nb_ports_available--;
1119 printf("Initializing port %u... ", (unsigned) portid);
1121 ret = rte_eth_dev_configure(portid, 1, 1, &port_conf);
1123 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
1124 ret, (unsigned) portid);
1126 rte_eth_macaddr_get(portid,&l2fwd_ports_eth_addr[portid]);
1128 /* init one RX queue */
1130 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
1131 rte_eth_dev_socket_id(portid),
1133 l2fwd_pktmbuf_pool[portid]);
1135 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup:err=%d, port=%u\n",
1136 ret, (unsigned) portid);
1138 /* init one TX queue on each port */
1140 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
1141 rte_eth_dev_socket_id(portid),
1144 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup:err=%d, port=%u\n",
1145 ret, (unsigned) portid);
1148 ret = rte_eth_dev_start(portid);
1150 rte_exit(EXIT_FAILURE, "rte_eth_dev_start:err=%d, port=%u\n",
1151 ret, (unsigned) portid);
1155 rte_eth_promiscuous_enable(portid);
1157 printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
1159 l2fwd_ports_eth_addr[portid].addr_bytes[0],
1160 l2fwd_ports_eth_addr[portid].addr_bytes[1],
1161 l2fwd_ports_eth_addr[portid].addr_bytes[2],
1162 l2fwd_ports_eth_addr[portid].addr_bytes[3],
1163 l2fwd_ports_eth_addr[portid].addr_bytes[4],
1164 l2fwd_ports_eth_addr[portid].addr_bytes[5]);
1166 /* initialize port stats */
1167 //memset(&port_statistics, 0, sizeof(port_statistics));
1170 if (!nb_ports_available) {
1171 rte_exit(EXIT_FAILURE,
1172 "All available ports are disabled. Please set portmask.\n");
1175 check_all_ports_link_status(nb_ports, l2fwd_enabled_port_mask);
1177 /* Record pair lcore */
1179 * Since l2fwd example would create pair between different neighbour port, that's
1180 * port 0 receive and forward to port 1, the same to port 1, these 2 ports will have
1181 * dependency. If one port stopped working (killed, for example), the port need to
1182 * be stopped/started again. During the time, another port need to wait until stop/start
1183 * procedure completed. So, record the pair relationship for those lcores working
1186 for (portid = 0; portid < nb_ports; portid++) {
1188 unsigned lcore = 0, pair_lcore = 0;
1189 unsigned j, find_lcore, find_pair_lcore;
1190 /* skip ports that are not enabled */
1191 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
1194 /* Find pair ports' lcores */
1195 find_lcore = find_pair_lcore = 0;
1196 pair_port = l2fwd_dst_ports[portid];
1197 for (i = 0; i < RTE_MAX_LCORE; i++) {
1198 if (!rte_lcore_is_enabled(i))
1200 for (j = 0; j < lcore_queue_conf[i].n_rx_port;j++) {
1201 if (lcore_queue_conf[i].rx_port_list[j] == portid) {
1206 if (lcore_queue_conf[i].rx_port_list[j] == pair_port) {
1208 find_pair_lcore = 1;
1212 if (find_lcore && find_pair_lcore)
1215 if (!find_lcore || !find_pair_lcore)
1216 rte_exit(EXIT_FAILURE, "Not find port=%d pair\n", portid);
1218 printf("lcore %u and %u paired\n", lcore, pair_lcore);
1219 lcore_resource[lcore].pair_id = pair_lcore;
1220 lcore_resource[pair_lcore].pair_id = lcore;
1223 /* Create message buffer for all master and slave */
1224 message_pool = rte_mempool_create("ms_msg_pool",
1225 NB_CORE_MSGBUF * RTE_MAX_LCORE,
1226 sizeof(enum l2fwd_cmd), NB_CORE_MSGBUF / 2,
1228 rte_pktmbuf_pool_init, NULL,
1229 rte_pktmbuf_init, NULL,
1230 rte_socket_id(), 0);
1232 if (message_pool == NULL)
1233 rte_exit(EXIT_FAILURE, "Create msg mempool failed\n");
1235 /* Create ring for each master and slave pair, also register cb when slave leaves */
1236 for (i = 0; i < RTE_MAX_LCORE; i++) {
1238 * Only create ring and register slave_exit cb in case that core involved into
1241 if (lcore_resource[i].enabled) {
1242 /* Create ring for master and slave communication */
1243 ret = create_ms_ring(i);
1245 rte_exit(EXIT_FAILURE, "Create ring for lcore=%u failed",
1248 if (flib_register_slave_exit_notify(i,
1249 slave_exit_cb) != 0)
1250 rte_exit(EXIT_FAILURE,
1251 "Register master_trace_slave_exit failed");
1255 /* launch per-lcore init on every lcore except master */
1256 flib_mp_remote_launch(l2fwd_launch_one_lcore, NULL, SKIP_MASTER);
1258 /* print statistics 10 second */
1259 prev_tsc = cur_tsc = rte_rdtsc();
1263 cur_tsc = rte_rdtsc();
1264 diff_tsc = cur_tsc - prev_tsc;
1265 /* if timer is enabled */
1266 if (timer_period > 0) {
1268 /* advance the timer */
1269 timer_tsc += diff_tsc;
1271 /* if timer has reached its timeout */
1272 if (unlikely(timer_tsc >= (uint64_t) timer_period)) {
1275 /* reset the timer */
1282 /* Check any slave need restart or recreate */
1283 rte_spinlock_lock(&res_lock);
1284 for (i = 0; i < RTE_MAX_LCORE; i++) {
1285 struct lcore_resource_struct *res = &lcore_resource[i];
1286 struct lcore_resource_struct *pair = &lcore_resource[res->pair_id];
1288 /* If find slave exited, try to reset pair */
1289 if (res->enabled && res->flags && pair->enabled) {
1291 master_sendcmd_with_ack(pair->lcore_id, CMD_STOP);
1292 rte_spinlock_unlock(&res_lock);
1294 rte_spinlock_lock(&res_lock);
1298 if (reset_pair(res->lcore_id, pair->lcore_id) != 0)
1299 rte_exit(EXIT_FAILURE, "failed to reset slave");
1304 rte_spinlock_unlock(&res_lock);