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 <rte_common.h>
43 #include <rte_malloc.h>
44 #include <rte_memory.h>
45 #include <rte_memcpy.h>
46 #include <rte_memzone.h>
48 #include <rte_launch.h>
49 #include <rte_atomic.h>
50 #include <rte_cycles.h>
51 #include <rte_prefetch.h>
52 #include <rte_lcore.h>
53 #include <rte_per_lcore.h>
54 #include <rte_branch_prediction.h>
55 #include <rte_interrupts.h>
57 #include <rte_debug.h>
58 #include <rte_ether.h>
59 #include <rte_ethdev.h>
60 #include <rte_mempool.h>
62 #include <rte_spinlock.h>
64 #include <rte_errno.h>
65 #include <rte_jobstats.h>
66 #include <rte_timer.h>
67 #include <rte_alarm.h>
68 #include <rte_pause.h>
70 #define RTE_LOGTYPE_L2FWD RTE_LOGTYPE_USER1
74 #define MAX_PKT_BURST 32
75 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
78 * Configurable number of RX/TX ring descriptors
80 #define RTE_TEST_RX_DESC_DEFAULT 128
81 #define RTE_TEST_TX_DESC_DEFAULT 512
82 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
83 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
85 /* ethernet addresses of ports */
86 static struct ether_addr l2fwd_ports_eth_addr[RTE_MAX_ETHPORTS];
88 /* mask of enabled ports */
89 static uint32_t l2fwd_enabled_port_mask;
91 /* list of enabled ports */
92 static uint32_t l2fwd_dst_ports[RTE_MAX_ETHPORTS];
94 #define UPDATE_STEP_UP 1
95 #define UPDATE_STEP_DOWN 32
97 static unsigned int l2fwd_rx_queue_per_lcore = 1;
99 #define MAX_RX_QUEUE_PER_LCORE 16
100 #define MAX_TX_QUEUE_PER_PORT 16
101 struct lcore_queue_conf {
103 unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
104 uint64_t next_flush_time[RTE_MAX_ETHPORTS];
106 struct rte_timer rx_timers[MAX_RX_QUEUE_PER_LCORE];
107 struct rte_jobstats port_fwd_jobs[MAX_RX_QUEUE_PER_LCORE];
109 struct rte_timer flush_timer;
110 struct rte_jobstats flush_job;
111 struct rte_jobstats idle_job;
112 struct rte_jobstats_context jobs_context;
114 rte_atomic16_t stats_read_pending;
116 } __rte_cache_aligned;
117 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
119 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
121 static const struct rte_eth_conf port_conf = {
124 .header_split = 0, /**< Header Split disabled */
125 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
126 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
127 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
128 .hw_strip_crc = 1, /**< CRC stripped by hardware */
131 .mq_mode = ETH_MQ_TX_NONE,
135 struct rte_mempool *l2fwd_pktmbuf_pool = NULL;
137 /* Per-port statistics struct */
138 struct l2fwd_port_statistics {
142 } __rte_cache_aligned;
143 struct l2fwd_port_statistics port_statistics[RTE_MAX_ETHPORTS];
146 #define MAX_TIMER_PERIOD 86400
147 /* default period is 10 seconds */
148 static int64_t timer_period = 10;
149 /* default timer frequency */
151 /* BURST_TX_DRAIN_US converted to cycles */
153 /* Convert cycles to ns */
155 cycles_to_ns(uint64_t cycles)
159 t *= (double)NS_PER_S;
165 show_lcore_stats(unsigned lcore_id)
167 struct lcore_queue_conf *qconf = &lcore_queue_conf[lcore_id];
168 struct rte_jobstats_context *ctx = &qconf->jobs_context;
169 struct rte_jobstats *job;
172 /* LCore statistics. */
173 uint64_t stats_period, loop_count;
174 uint64_t exec, exec_min, exec_max;
175 uint64_t management, management_min, management_max;
176 uint64_t busy, busy_min, busy_max;
178 /* Jobs statistics. */
179 const uint8_t port_cnt = qconf->n_rx_port;
180 uint64_t jobs_exec_cnt[port_cnt], jobs_period[port_cnt];
181 uint64_t jobs_exec[port_cnt], jobs_exec_min[port_cnt],
182 jobs_exec_max[port_cnt];
184 uint64_t flush_exec_cnt, flush_period;
185 uint64_t flush_exec, flush_exec_min, flush_exec_max;
187 uint64_t idle_exec_cnt;
188 uint64_t idle_exec, idle_exec_min, idle_exec_max;
189 uint64_t collection_time = rte_get_timer_cycles();
191 /* Ask forwarding thread to give us stats. */
192 rte_atomic16_set(&qconf->stats_read_pending, 1);
193 rte_spinlock_lock(&qconf->lock);
194 rte_atomic16_set(&qconf->stats_read_pending, 0);
196 /* Collect context statistics. */
197 stats_period = ctx->state_time - ctx->start_time;
198 loop_count = ctx->loop_cnt;
200 exec = ctx->exec_time;
201 exec_min = ctx->min_exec_time;
202 exec_max = ctx->max_exec_time;
204 management = ctx->management_time;
205 management_min = ctx->min_management_time;
206 management_max = ctx->max_management_time;
208 rte_jobstats_context_reset(ctx);
210 for (i = 0; i < port_cnt; i++) {
211 job = &qconf->port_fwd_jobs[i];
213 jobs_exec_cnt[i] = job->exec_cnt;
214 jobs_period[i] = job->period;
216 jobs_exec[i] = job->exec_time;
217 jobs_exec_min[i] = job->min_exec_time;
218 jobs_exec_max[i] = job->max_exec_time;
220 rte_jobstats_reset(job);
223 flush_exec_cnt = qconf->flush_job.exec_cnt;
224 flush_period = qconf->flush_job.period;
225 flush_exec = qconf->flush_job.exec_time;
226 flush_exec_min = qconf->flush_job.min_exec_time;
227 flush_exec_max = qconf->flush_job.max_exec_time;
228 rte_jobstats_reset(&qconf->flush_job);
230 idle_exec_cnt = qconf->idle_job.exec_cnt;
231 idle_exec = qconf->idle_job.exec_time;
232 idle_exec_min = qconf->idle_job.min_exec_time;
233 idle_exec_max = qconf->idle_job.max_exec_time;
234 rte_jobstats_reset(&qconf->idle_job);
236 rte_spinlock_unlock(&qconf->lock);
239 busy = exec + management;
240 busy_min = exec_min + management_min;
241 busy_max = exec_max + management_max;
244 collection_time = rte_get_timer_cycles() - collection_time;
246 #define STAT_FMT "\n%-18s %'14.0f %6.1f%% %'10.0f %'10.0f %'10.0f"
248 printf("\n----------------"
249 "\nLCore %3u: statistics (time in ns, collected in %'9.0f)"
250 "\n%-18s %14s %7s %10s %10s %10s "
252 "\n%-18s %'14" PRIu64
254 STAT_FMT /* Management */
257 lcore_id, cycles_to_ns(collection_time),
258 "Stat type", "total", "%total", "avg", "min", "max",
259 "Stats duration:", cycles_to_ns(stats_period),
260 "Loop count:", loop_count,
262 cycles_to_ns(exec), exec * 100.0 / stats_period,
263 cycles_to_ns(loop_count ? exec / loop_count : 0),
264 cycles_to_ns(exec_min),
265 cycles_to_ns(exec_max),
267 cycles_to_ns(management), management * 100.0 / stats_period,
268 cycles_to_ns(loop_count ? management / loop_count : 0),
269 cycles_to_ns(management_min),
270 cycles_to_ns(management_max),
272 cycles_to_ns(busy), busy * 100.0 / stats_period,
273 cycles_to_ns(loop_count ? busy / loop_count : 0),
274 cycles_to_ns(busy_min),
275 cycles_to_ns(busy_max),
277 cycles_to_ns(idle_exec), idle_exec * 100.0 / stats_period,
278 cycles_to_ns(idle_exec_cnt ? idle_exec / idle_exec_cnt : 0),
279 cycles_to_ns(idle_exec_min),
280 cycles_to_ns(idle_exec_max));
282 for (i = 0; i < qconf->n_rx_port; i++) {
283 job = &qconf->port_fwd_jobs[i];
284 printf("\n\nJob %" PRIu32 ": %-20s "
285 "\n%-18s %'14" PRIu64
289 "Exec count:", jobs_exec_cnt[i],
290 "Exec period: ", cycles_to_ns(jobs_period[i]),
292 cycles_to_ns(jobs_exec[i]), jobs_exec[i] * 100.0 / stats_period,
293 cycles_to_ns(jobs_exec_cnt[i] ? jobs_exec[i] / jobs_exec_cnt[i]
295 cycles_to_ns(jobs_exec_min[i]),
296 cycles_to_ns(jobs_exec_max[i]));
299 if (qconf->n_rx_port > 0) {
300 job = &qconf->flush_job;
301 printf("\n\nJob %" PRIu32 ": %-20s "
302 "\n%-18s %'14" PRIu64
306 "Exec count:", flush_exec_cnt,
307 "Exec period: ", cycles_to_ns(flush_period),
309 cycles_to_ns(flush_exec), flush_exec * 100.0 / stats_period,
310 cycles_to_ns(flush_exec_cnt ? flush_exec / flush_exec_cnt : 0),
311 cycles_to_ns(flush_exec_min),
312 cycles_to_ns(flush_exec_max));
316 /* Print out statistics on packets dropped */
318 show_stats_cb(__rte_unused void *param)
320 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
321 unsigned portid, lcore_id;
323 total_packets_dropped = 0;
324 total_packets_tx = 0;
325 total_packets_rx = 0;
327 const char clr[] = { 27, '[', '2', 'J', '\0' };
328 const char topLeft[] = { 27, '[', '1', ';', '1', 'H', '\0' };
330 /* Clear screen and move to top left */
332 "\nPort statistics ===================================",
335 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
336 /* skip disabled ports */
337 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
339 printf("\nStatistics for port %u ------------------------------"
340 "\nPackets sent: %24"PRIu64
341 "\nPackets received: %20"PRIu64
342 "\nPackets dropped: %21"PRIu64,
344 port_statistics[portid].tx,
345 port_statistics[portid].rx,
346 port_statistics[portid].dropped);
348 total_packets_dropped += port_statistics[portid].dropped;
349 total_packets_tx += port_statistics[portid].tx;
350 total_packets_rx += port_statistics[portid].rx;
353 printf("\nAggregate statistics ==============================="
354 "\nTotal packets sent: %18"PRIu64
355 "\nTotal packets received: %14"PRIu64
356 "\nTotal packets dropped: %15"PRIu64
357 "\n====================================================",
360 total_packets_dropped);
362 RTE_LCORE_FOREACH(lcore_id) {
363 if (lcore_queue_conf[lcore_id].n_rx_port > 0)
364 show_lcore_stats(lcore_id);
367 printf("\n====================================================\n");
368 rte_eal_alarm_set(timer_period * US_PER_S, show_stats_cb, NULL);
372 l2fwd_simple_forward(struct rte_mbuf *m, unsigned portid)
374 struct ether_hdr *eth;
378 struct rte_eth_dev_tx_buffer *buffer;
380 dst_port = l2fwd_dst_ports[portid];
381 eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
383 /* 02:00:00:00:00:xx */
384 tmp = ð->d_addr.addr_bytes[0];
385 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
388 ether_addr_copy(&l2fwd_ports_eth_addr[dst_port], ð->s_addr);
390 buffer = tx_buffer[dst_port];
391 sent = rte_eth_tx_buffer(dst_port, 0, buffer, m);
393 port_statistics[dst_port].tx += sent;
397 l2fwd_job_update_cb(struct rte_jobstats *job, int64_t result)
399 int64_t err = job->target - result;
400 int64_t histeresis = job->target / 8;
402 if (err < -histeresis) {
403 if (job->min_period + UPDATE_STEP_DOWN < job->period)
404 job->period -= UPDATE_STEP_DOWN;
405 } else if (err > histeresis) {
406 if (job->period + UPDATE_STEP_UP < job->max_period)
407 job->period += UPDATE_STEP_UP;
412 l2fwd_fwd_job(__rte_unused struct rte_timer *timer, void *arg)
414 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
417 const uint8_t port_idx = (uintptr_t) arg;
418 const unsigned lcore_id = rte_lcore_id();
419 struct lcore_queue_conf *qconf = &lcore_queue_conf[lcore_id];
420 struct rte_jobstats *job = &qconf->port_fwd_jobs[port_idx];
421 const uint8_t portid = qconf->rx_port_list[port_idx];
424 uint16_t total_nb_rx;
426 rte_jobstats_start(&qconf->jobs_context, job);
428 /* Call rx burst 2 times. This allow rte_jobstats logic to see if this
429 * function must be called more frequently. */
431 total_nb_rx = rte_eth_rx_burst((uint8_t) portid, 0, pkts_burst,
434 for (j = 0; j < total_nb_rx; j++) {
436 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
437 l2fwd_simple_forward(m, portid);
440 if (total_nb_rx == MAX_PKT_BURST) {
441 const uint16_t nb_rx = rte_eth_rx_burst((uint8_t) portid, 0, pkts_burst,
444 total_nb_rx += nb_rx;
445 for (j = 0; j < nb_rx; j++) {
447 rte_prefetch0(rte_pktmbuf_mtod(m, void *));
448 l2fwd_simple_forward(m, portid);
452 port_statistics[portid].rx += total_nb_rx;
454 /* Adjust period time in which we are running here. */
455 if (rte_jobstats_finish(job, total_nb_rx) != 0) {
456 rte_timer_reset(&qconf->rx_timers[port_idx], job->period, PERIODICAL,
457 lcore_id, l2fwd_fwd_job, arg);
462 l2fwd_flush_job(__rte_unused struct rte_timer *timer, __rte_unused void *arg)
466 struct lcore_queue_conf *qconf;
470 struct rte_eth_dev_tx_buffer *buffer;
472 lcore_id = rte_lcore_id();
473 qconf = &lcore_queue_conf[lcore_id];
475 rte_jobstats_start(&qconf->jobs_context, &qconf->flush_job);
477 now = rte_get_timer_cycles();
478 lcore_id = rte_lcore_id();
479 qconf = &lcore_queue_conf[lcore_id];
481 for (i = 0; i < qconf->n_rx_port; i++) {
482 portid = l2fwd_dst_ports[qconf->rx_port_list[i]];
484 if (qconf->next_flush_time[portid] <= now)
487 buffer = tx_buffer[portid];
488 sent = rte_eth_tx_buffer_flush(portid, 0, buffer);
490 port_statistics[portid].tx += sent;
492 qconf->next_flush_time[portid] = rte_get_timer_cycles() + drain_tsc;
495 /* Pass target to indicate that this job is happy of time interwal
496 * in which it was called. */
497 rte_jobstats_finish(&qconf->flush_job, qconf->flush_job.target);
500 /* main processing loop */
502 l2fwd_main_loop(void)
506 struct lcore_queue_conf *qconf;
507 uint8_t stats_read_pending = 0;
510 lcore_id = rte_lcore_id();
511 qconf = &lcore_queue_conf[lcore_id];
513 if (qconf->n_rx_port == 0) {
514 RTE_LOG(INFO, L2FWD, "lcore %u has nothing to do\n", lcore_id);
518 RTE_LOG(INFO, L2FWD, "entering main loop on lcore %u\n", lcore_id);
520 for (i = 0; i < qconf->n_rx_port; i++) {
522 portid = qconf->rx_port_list[i];
523 RTE_LOG(INFO, L2FWD, " -- lcoreid=%u portid=%u\n", lcore_id,
527 rte_jobstats_init(&qconf->idle_job, "idle", 0, 0, 0, 0);
530 rte_spinlock_lock(&qconf->lock);
533 rte_jobstats_context_start(&qconf->jobs_context);
536 * - Read stats_read_pending flag
537 * - check if some real job need to be executed
539 rte_jobstats_start(&qconf->jobs_context, &qconf->idle_job);
541 uint64_t repeats = 0;
545 uint64_t now = rte_get_timer_cycles();
548 need_manage = qconf->flush_timer.expire < now;
549 /* Check if we was esked to give a stats. */
551 rte_atomic16_read(&qconf->stats_read_pending);
552 need_manage |= stats_read_pending;
554 for (i = 0; i < qconf->n_rx_port && !need_manage; i++)
555 need_manage = qconf->rx_timers[i].expire < now;
557 } while (!need_manage);
559 if (likely(repeats != 1))
560 rte_jobstats_finish(&qconf->idle_job, qconf->idle_job.target);
562 rte_jobstats_abort(&qconf->idle_job);
565 rte_jobstats_context_finish(&qconf->jobs_context);
566 } while (likely(stats_read_pending == 0));
568 rte_spinlock_unlock(&qconf->lock);
574 l2fwd_launch_one_lcore(__attribute__((unused)) void *dummy)
582 l2fwd_usage(const char *prgname)
584 printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
585 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
586 " -q NQ: number of queue (=ports) per lcore (default is 1)\n"
587 " -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n"
588 " -l set system default locale instead of default (\"C\" locale) for thousands separator in stats.",
593 l2fwd_parse_portmask(const char *portmask)
598 /* parse hexadecimal string */
599 pm = strtoul(portmask, &end, 16);
600 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
610 l2fwd_parse_nqueue(const char *q_arg)
615 /* parse hexadecimal string */
616 n = strtoul(q_arg, &end, 10);
617 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
621 if (n >= MAX_RX_QUEUE_PER_LCORE)
628 l2fwd_parse_timer_period(const char *q_arg)
633 /* parse number string */
634 n = strtol(q_arg, &end, 10);
635 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
637 if (n >= MAX_TIMER_PERIOD)
643 /* Parse the argument given in the command line of the application */
645 l2fwd_parse_args(int argc, char **argv)
650 char *prgname = argv[0];
651 static struct option lgopts[] = {
657 while ((opt = getopt_long(argc, argvopt, "p:q:T:l",
658 lgopts, &option_index)) != EOF) {
663 l2fwd_enabled_port_mask = l2fwd_parse_portmask(optarg);
664 if (l2fwd_enabled_port_mask == 0) {
665 printf("invalid portmask\n");
666 l2fwd_usage(prgname);
673 l2fwd_rx_queue_per_lcore = l2fwd_parse_nqueue(optarg);
674 if (l2fwd_rx_queue_per_lcore == 0) {
675 printf("invalid queue number\n");
676 l2fwd_usage(prgname);
683 timer_period = l2fwd_parse_timer_period(optarg);
684 if (timer_period < 0) {
685 printf("invalid timer period\n");
686 l2fwd_usage(prgname);
691 /* For thousands separator in printf. */
693 setlocale(LC_ALL, "");
698 l2fwd_usage(prgname);
702 l2fwd_usage(prgname);
708 argv[optind-1] = prgname;
711 optind = 1; /* reset getopt lib */
715 /* Check the link status of all ports in up to 9s, and print them finally */
717 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
719 #define CHECK_INTERVAL 100 /* 100ms */
720 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
721 uint8_t portid, count, all_ports_up, print_flag = 0;
722 struct rte_eth_link link;
724 printf("\nChecking link status");
726 for (count = 0; count <= MAX_CHECK_TIME; count++) {
728 for (portid = 0; portid < port_num; portid++) {
729 if ((port_mask & (1 << portid)) == 0)
731 memset(&link, 0, sizeof(link));
732 rte_eth_link_get_nowait(portid, &link);
733 /* print link status if flag set */
734 if (print_flag == 1) {
735 if (link.link_status)
736 printf("Port %d Link Up - speed %u "
737 "Mbps - %s\n", (uint8_t)portid,
738 (unsigned)link.link_speed,
739 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
740 ("full-duplex") : ("half-duplex\n"));
742 printf("Port %d Link Down\n",
746 /* clear all_ports_up flag if any link down */
747 if (link.link_status == ETH_LINK_DOWN) {
752 /* after finally printing all link status, get out */
756 if (all_ports_up == 0) {
759 rte_delay_ms(CHECK_INTERVAL);
762 /* set the print_flag if all ports up or timeout */
763 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
771 main(int argc, char **argv)
773 struct lcore_queue_conf *qconf;
774 struct rte_eth_dev_info dev_info;
775 unsigned lcore_id, rx_lcore_id;
776 unsigned nb_ports_in_mask = 0;
778 char name[RTE_JOBSTATS_NAMESIZE];
780 uint8_t nb_ports_available;
781 uint8_t portid, last_port;
785 ret = rte_eal_init(argc, argv);
787 rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
791 /* parse application arguments (after the EAL ones) */
792 ret = l2fwd_parse_args(argc, argv);
794 rte_exit(EXIT_FAILURE, "Invalid L2FWD arguments\n");
796 rte_timer_subsystem_init();
798 /* fetch default timer frequency. */
799 hz = rte_get_timer_hz();
801 /* create the mbuf pool */
803 rte_pktmbuf_pool_create("mbuf_pool", NB_MBUF, 32,
804 0, RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
805 if (l2fwd_pktmbuf_pool == NULL)
806 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool\n");
808 nb_ports = rte_eth_dev_count();
810 rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
812 /* reset l2fwd_dst_ports */
813 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++)
814 l2fwd_dst_ports[portid] = 0;
818 * Each logical core is assigned a dedicated TX queue on each port.
820 for (portid = 0; portid < nb_ports; portid++) {
821 /* skip ports that are not enabled */
822 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
825 if (nb_ports_in_mask % 2) {
826 l2fwd_dst_ports[portid] = last_port;
827 l2fwd_dst_ports[last_port] = portid;
833 rte_eth_dev_info_get(portid, &dev_info);
835 if (nb_ports_in_mask % 2) {
836 printf("Notice: odd number of ports in portmask.\n");
837 l2fwd_dst_ports[last_port] = last_port;
843 /* Initialize the port/queue configuration of each logical core */
844 for (portid = 0; portid < nb_ports; portid++) {
845 /* skip ports that are not enabled */
846 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
849 /* get the lcore_id for this port */
850 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
851 lcore_queue_conf[rx_lcore_id].n_rx_port ==
852 l2fwd_rx_queue_per_lcore) {
854 if (rx_lcore_id >= RTE_MAX_LCORE)
855 rte_exit(EXIT_FAILURE, "Not enough cores\n");
858 if (qconf != &lcore_queue_conf[rx_lcore_id])
859 /* Assigned a new logical core in the loop above. */
860 qconf = &lcore_queue_conf[rx_lcore_id];
862 qconf->rx_port_list[qconf->n_rx_port] = portid;
864 printf("Lcore %u: RX port %u\n", rx_lcore_id, (unsigned) portid);
867 nb_ports_available = nb_ports;
869 /* Initialise each port */
870 for (portid = 0; portid < nb_ports; portid++) {
871 /* skip ports that are not enabled */
872 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0) {
873 printf("Skipping disabled port %u\n", (unsigned) portid);
874 nb_ports_available--;
878 printf("Initializing port %u... ", (unsigned) portid);
880 ret = rte_eth_dev_configure(portid, 1, 1, &port_conf);
882 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
883 ret, (unsigned) portid);
885 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
888 rte_exit(EXIT_FAILURE,
889 "Cannot adjust number of descriptors: err=%d, port=%u\n",
890 ret, (unsigned) portid);
892 rte_eth_macaddr_get(portid, &l2fwd_ports_eth_addr[portid]);
894 /* init one RX queue */
896 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
897 rte_eth_dev_socket_id(portid),
901 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup:err=%d, port=%u\n",
902 ret, (unsigned) portid);
904 /* init one TX queue on each port */
906 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
907 rte_eth_dev_socket_id(portid),
910 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup:err=%d, port=%u\n",
911 ret, (unsigned) portid);
913 /* Initialize TX buffers */
914 tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
915 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
916 rte_eth_dev_socket_id(portid));
917 if (tx_buffer[portid] == NULL)
918 rte_exit(EXIT_FAILURE, "Cannot allocate buffer for tx on port %u\n",
921 rte_eth_tx_buffer_init(tx_buffer[portid], MAX_PKT_BURST);
923 ret = rte_eth_tx_buffer_set_err_callback(tx_buffer[portid],
924 rte_eth_tx_buffer_count_callback,
925 &port_statistics[portid].dropped);
927 rte_exit(EXIT_FAILURE, "Cannot set error callback for "
928 "tx buffer on port %u\n", (unsigned) portid);
931 ret = rte_eth_dev_start(portid);
933 rte_exit(EXIT_FAILURE, "rte_eth_dev_start:err=%d, port=%u\n",
934 ret, (unsigned) portid);
938 rte_eth_promiscuous_enable(portid);
940 printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
942 l2fwd_ports_eth_addr[portid].addr_bytes[0],
943 l2fwd_ports_eth_addr[portid].addr_bytes[1],
944 l2fwd_ports_eth_addr[portid].addr_bytes[2],
945 l2fwd_ports_eth_addr[portid].addr_bytes[3],
946 l2fwd_ports_eth_addr[portid].addr_bytes[4],
947 l2fwd_ports_eth_addr[portid].addr_bytes[5]);
949 /* initialize port stats */
950 memset(&port_statistics, 0, sizeof(port_statistics));
953 if (!nb_ports_available) {
954 rte_exit(EXIT_FAILURE,
955 "All available ports are disabled. Please set portmask.\n");
958 check_all_ports_link_status(nb_ports, l2fwd_enabled_port_mask);
960 drain_tsc = (hz + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
962 RTE_LCORE_FOREACH(lcore_id) {
963 qconf = &lcore_queue_conf[lcore_id];
965 rte_spinlock_init(&qconf->lock);
967 if (rte_jobstats_context_init(&qconf->jobs_context) != 0)
968 rte_panic("Jobs stats context for core %u init failed\n", lcore_id);
970 if (qconf->n_rx_port == 0) {
972 "lcore %u: no ports so no jobs stats context initialization\n",
977 * Set fixed period by setting min = max = initial period. Set target to
978 * zero as it is irrelevant for this job. */
979 rte_jobstats_init(&qconf->flush_job, "flush", drain_tsc, drain_tsc,
982 rte_timer_init(&qconf->flush_timer);
983 ret = rte_timer_reset(&qconf->flush_timer, drain_tsc, PERIODICAL,
984 lcore_id, &l2fwd_flush_job, NULL);
987 rte_exit(1, "Failed to reset flush job timer for lcore %u: %s",
988 lcore_id, rte_strerror(-ret));
991 for (i = 0; i < qconf->n_rx_port; i++) {
992 struct rte_jobstats *job = &qconf->port_fwd_jobs[i];
994 portid = qconf->rx_port_list[i];
995 printf("Setting forward job for port %u\n", portid);
997 snprintf(name, RTE_DIM(name), "port %u fwd", portid);
998 /* Setup forward job.
999 * Set min, max and initial period. Set target to MAX_PKT_BURST as
1000 * this is desired optimal RX/TX burst size. */
1001 rte_jobstats_init(job, name, 0, drain_tsc, 0, MAX_PKT_BURST);
1002 rte_jobstats_set_update_period_function(job, l2fwd_job_update_cb);
1004 rte_timer_init(&qconf->rx_timers[i]);
1005 ret = rte_timer_reset(&qconf->rx_timers[i], 0, PERIODICAL, lcore_id,
1006 &l2fwd_fwd_job, (void *)(uintptr_t)i);
1009 rte_exit(1, "Failed to reset lcore %u port %u job timer: %s",
1010 lcore_id, qconf->rx_port_list[i], rte_strerror(-ret));
1016 rte_eal_alarm_set(timer_period * MS_PER_S, show_stats_cb, NULL);
1018 RTE_LOG(INFO, L2FWD, "Stats display disabled\n");
1020 /* launch per-lcore init on every lcore */
1021 rte_eal_mp_remote_launch(l2fwd_launch_one_lcore, NULL, CALL_MASTER);
1022 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1023 if (rte_eal_wait_lcore(lcore_id) < 0)