xref: /dpdk/examples/l2fwd-jobstats/main.c (revision 89f0711f9ddfb5822da9d34f384b92f72a61c4dc)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2016 Intel Corporation
3  */
4 
5 #include <locale.h>
6 #include <stdlib.h>
7 #include <string.h>
8 #include <stdint.h>
9 #include <ctype.h>
10 #include <getopt.h>
11 
12 #include <rte_common.h>
13 #include <rte_log.h>
14 #include <rte_malloc.h>
15 #include <rte_memory.h>
16 #include <rte_memcpy.h>
17 #include <rte_eal.h>
18 #include <rte_launch.h>
19 #include <rte_atomic.h>
20 #include <rte_cycles.h>
21 #include <rte_prefetch.h>
22 #include <rte_lcore.h>
23 #include <rte_per_lcore.h>
24 #include <rte_branch_prediction.h>
25 #include <rte_interrupts.h>
26 #include <rte_debug.h>
27 #include <rte_ether.h>
28 #include <rte_ethdev.h>
29 #include <rte_mempool.h>
30 #include <rte_mbuf.h>
31 #include <rte_spinlock.h>
32 
33 #include <rte_errno.h>
34 #include <rte_jobstats.h>
35 #include <rte_timer.h>
36 #include <rte_alarm.h>
37 #include <rte_pause.h>
38 
39 #define RTE_LOGTYPE_L2FWD RTE_LOGTYPE_USER1
40 
41 #define NB_MBUF   8192
42 
43 #define MAX_PKT_BURST 32
44 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
45 
46 /*
47  * Configurable number of RX/TX ring descriptors
48  */
49 #define RTE_TEST_RX_DESC_DEFAULT 128
50 #define RTE_TEST_TX_DESC_DEFAULT 512
51 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
52 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
53 
54 /* ethernet addresses of ports */
55 static struct ether_addr l2fwd_ports_eth_addr[RTE_MAX_ETHPORTS];
56 
57 /* mask of enabled ports */
58 static uint32_t l2fwd_enabled_port_mask;
59 
60 /* list of enabled ports */
61 static uint32_t l2fwd_dst_ports[RTE_MAX_ETHPORTS];
62 
63 #define UPDATE_STEP_UP 1
64 #define UPDATE_STEP_DOWN 32
65 
66 static unsigned int l2fwd_rx_queue_per_lcore = 1;
67 
68 #define MAX_RX_QUEUE_PER_LCORE 16
69 #define MAX_TX_QUEUE_PER_PORT 16
70 struct lcore_queue_conf {
71 	unsigned n_rx_port;
72 	unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
73 	uint64_t next_flush_time[RTE_MAX_ETHPORTS];
74 
75 	struct rte_timer rx_timers[MAX_RX_QUEUE_PER_LCORE];
76 	struct rte_jobstats port_fwd_jobs[MAX_RX_QUEUE_PER_LCORE];
77 
78 	struct rte_timer flush_timer;
79 	struct rte_jobstats flush_job;
80 	struct rte_jobstats idle_job;
81 	struct rte_jobstats_context jobs_context;
82 
83 	rte_atomic16_t stats_read_pending;
84 	rte_spinlock_t lock;
85 } __rte_cache_aligned;
86 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
87 
88 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
89 
90 static struct rte_eth_conf port_conf = {
91 	.rxmode = {
92 		.split_hdr_size = 0,
93 		.ignore_offload_bitfield = 1,
94 		.offloads = DEV_RX_OFFLOAD_CRC_STRIP,
95 	},
96 	.txmode = {
97 		.mq_mode = ETH_MQ_TX_NONE,
98 	},
99 };
100 
101 struct rte_mempool *l2fwd_pktmbuf_pool = NULL;
102 
103 /* Per-port statistics struct */
104 struct l2fwd_port_statistics {
105 	uint64_t tx;
106 	uint64_t rx;
107 	uint64_t dropped;
108 } __rte_cache_aligned;
109 struct l2fwd_port_statistics port_statistics[RTE_MAX_ETHPORTS];
110 
111 /* 1 day max */
112 #define MAX_TIMER_PERIOD 86400
113 /* default period is 10 seconds */
114 static int64_t timer_period = 10;
115 /* default timer frequency */
116 static double hz;
117 /* BURST_TX_DRAIN_US converted to cycles */
118 uint64_t drain_tsc;
119 /* Convert cycles to ns */
120 static inline double
121 cycles_to_ns(uint64_t cycles)
122 {
123 	double t = cycles;
124 
125 	t *= (double)NS_PER_S;
126 	t /= hz;
127 	return t;
128 }
129 
130 static void
131 show_lcore_stats(unsigned lcore_id)
132 {
133 	struct lcore_queue_conf *qconf = &lcore_queue_conf[lcore_id];
134 	struct rte_jobstats_context *ctx = &qconf->jobs_context;
135 	struct rte_jobstats *job;
136 	uint8_t i;
137 
138 	/* LCore statistics. */
139 	uint64_t stats_period, loop_count;
140 	uint64_t exec, exec_min, exec_max;
141 	uint64_t management, management_min, management_max;
142 	uint64_t busy, busy_min, busy_max;
143 
144 	/* Jobs statistics. */
145 	const uint16_t port_cnt = qconf->n_rx_port;
146 	uint64_t jobs_exec_cnt[port_cnt], jobs_period[port_cnt];
147 	uint64_t jobs_exec[port_cnt], jobs_exec_min[port_cnt],
148 				jobs_exec_max[port_cnt];
149 
150 	uint64_t flush_exec_cnt, flush_period;
151 	uint64_t flush_exec, flush_exec_min, flush_exec_max;
152 
153 	uint64_t idle_exec_cnt;
154 	uint64_t idle_exec, idle_exec_min, idle_exec_max;
155 	uint64_t collection_time = rte_get_timer_cycles();
156 
157 	/* Ask forwarding thread to give us stats. */
158 	rte_atomic16_set(&qconf->stats_read_pending, 1);
159 	rte_spinlock_lock(&qconf->lock);
160 	rte_atomic16_set(&qconf->stats_read_pending, 0);
161 
162 	/* Collect context statistics. */
163 	stats_period = ctx->state_time - ctx->start_time;
164 	loop_count = ctx->loop_cnt;
165 
166 	exec = ctx->exec_time;
167 	exec_min = ctx->min_exec_time;
168 	exec_max = ctx->max_exec_time;
169 
170 	management = ctx->management_time;
171 	management_min = ctx->min_management_time;
172 	management_max = ctx->max_management_time;
173 
174 	rte_jobstats_context_reset(ctx);
175 
176 	for (i = 0; i < port_cnt; i++) {
177 		job = &qconf->port_fwd_jobs[i];
178 
179 		jobs_exec_cnt[i] = job->exec_cnt;
180 		jobs_period[i] = job->period;
181 
182 		jobs_exec[i] = job->exec_time;
183 		jobs_exec_min[i] = job->min_exec_time;
184 		jobs_exec_max[i] = job->max_exec_time;
185 
186 		rte_jobstats_reset(job);
187 	}
188 
189 	flush_exec_cnt = qconf->flush_job.exec_cnt;
190 	flush_period = qconf->flush_job.period;
191 	flush_exec = qconf->flush_job.exec_time;
192 	flush_exec_min = qconf->flush_job.min_exec_time;
193 	flush_exec_max = qconf->flush_job.max_exec_time;
194 	rte_jobstats_reset(&qconf->flush_job);
195 
196 	idle_exec_cnt = qconf->idle_job.exec_cnt;
197 	idle_exec = qconf->idle_job.exec_time;
198 	idle_exec_min = qconf->idle_job.min_exec_time;
199 	idle_exec_max = qconf->idle_job.max_exec_time;
200 	rte_jobstats_reset(&qconf->idle_job);
201 
202 	rte_spinlock_unlock(&qconf->lock);
203 
204 	exec -= idle_exec;
205 	busy = exec + management;
206 	busy_min = exec_min + management_min;
207 	busy_max = exec_max + management_max;
208 
209 
210 	collection_time = rte_get_timer_cycles() - collection_time;
211 
212 #define STAT_FMT "\n%-18s %'14.0f %6.1f%% %'10.0f %'10.0f %'10.0f"
213 
214 	printf("\n----------------"
215 			"\nLCore %3u: statistics (time in ns, collected in %'9.0f)"
216 			"\n%-18s %14s %7s %10s %10s %10s "
217 			"\n%-18s %'14.0f"
218 			"\n%-18s %'14" PRIu64
219 			STAT_FMT /* Exec */
220 			STAT_FMT /* Management */
221 			STAT_FMT /* Busy */
222 			STAT_FMT, /* Idle  */
223 			lcore_id, cycles_to_ns(collection_time),
224 			"Stat type", "total", "%total", "avg", "min", "max",
225 			"Stats duration:", cycles_to_ns(stats_period),
226 			"Loop count:", loop_count,
227 			"Exec time",
228 			cycles_to_ns(exec), exec * 100.0 / stats_period,
229 			cycles_to_ns(loop_count  ? exec / loop_count : 0),
230 			cycles_to_ns(exec_min),
231 			cycles_to_ns(exec_max),
232 			"Management time",
233 			cycles_to_ns(management), management * 100.0 / stats_period,
234 			cycles_to_ns(loop_count  ? management / loop_count : 0),
235 			cycles_to_ns(management_min),
236 			cycles_to_ns(management_max),
237 			"Exec + management",
238 			cycles_to_ns(busy),  busy * 100.0 / stats_period,
239 			cycles_to_ns(loop_count ? busy / loop_count : 0),
240 			cycles_to_ns(busy_min),
241 			cycles_to_ns(busy_max),
242 			"Idle (job)",
243 			cycles_to_ns(idle_exec), idle_exec * 100.0 / stats_period,
244 			cycles_to_ns(idle_exec_cnt ? idle_exec / idle_exec_cnt : 0),
245 			cycles_to_ns(idle_exec_min),
246 			cycles_to_ns(idle_exec_max));
247 
248 	for (i = 0; i < qconf->n_rx_port; i++) {
249 		job = &qconf->port_fwd_jobs[i];
250 		printf("\n\nJob %" PRIu32 ": %-20s "
251 				"\n%-18s %'14" PRIu64
252 				"\n%-18s %'14.0f"
253 				STAT_FMT,
254 				i, job->name,
255 				"Exec count:", jobs_exec_cnt[i],
256 				"Exec period: ", cycles_to_ns(jobs_period[i]),
257 				"Exec time",
258 				cycles_to_ns(jobs_exec[i]), jobs_exec[i] * 100.0 / stats_period,
259 				cycles_to_ns(jobs_exec_cnt[i] ? jobs_exec[i] / jobs_exec_cnt[i]
260 						: 0),
261 				cycles_to_ns(jobs_exec_min[i]),
262 				cycles_to_ns(jobs_exec_max[i]));
263 	}
264 
265 	if (qconf->n_rx_port > 0) {
266 		job = &qconf->flush_job;
267 		printf("\n\nJob %" PRIu32 ": %-20s "
268 				"\n%-18s %'14" PRIu64
269 				"\n%-18s %'14.0f"
270 				STAT_FMT,
271 				i, job->name,
272 				"Exec count:", flush_exec_cnt,
273 				"Exec period: ", cycles_to_ns(flush_period),
274 				"Exec time",
275 				cycles_to_ns(flush_exec), flush_exec * 100.0 / stats_period,
276 				cycles_to_ns(flush_exec_cnt ? flush_exec / flush_exec_cnt : 0),
277 				cycles_to_ns(flush_exec_min),
278 				cycles_to_ns(flush_exec_max));
279 	}
280 }
281 
282 /* Print out statistics on packets dropped */
283 static void
284 show_stats_cb(__rte_unused void *param)
285 {
286 	uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
287 	unsigned portid, lcore_id;
288 
289 	total_packets_dropped = 0;
290 	total_packets_tx = 0;
291 	total_packets_rx = 0;
292 
293 	const char clr[] = { 27, '[', '2', 'J', '\0' };
294 	const char topLeft[] = { 27, '[', '1', ';', '1', 'H', '\0' };
295 
296 	/* Clear screen and move to top left */
297 	printf("%s%s"
298 			"\nPort statistics ===================================",
299 			clr, topLeft);
300 
301 	for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
302 		/* skip disabled ports */
303 		if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
304 			continue;
305 		printf("\nStatistics for port %u ------------------------------"
306 				"\nPackets sent: %24"PRIu64
307 				"\nPackets received: %20"PRIu64
308 				"\nPackets dropped: %21"PRIu64,
309 				portid,
310 				port_statistics[portid].tx,
311 				port_statistics[portid].rx,
312 				port_statistics[portid].dropped);
313 
314 		total_packets_dropped += port_statistics[portid].dropped;
315 		total_packets_tx += port_statistics[portid].tx;
316 		total_packets_rx += port_statistics[portid].rx;
317 	}
318 
319 	printf("\nAggregate statistics ==============================="
320 			"\nTotal packets sent: %18"PRIu64
321 			"\nTotal packets received: %14"PRIu64
322 			"\nTotal packets dropped: %15"PRIu64
323 			"\n====================================================",
324 			total_packets_tx,
325 			total_packets_rx,
326 			total_packets_dropped);
327 
328 	RTE_LCORE_FOREACH(lcore_id) {
329 		if (lcore_queue_conf[lcore_id].n_rx_port > 0)
330 			show_lcore_stats(lcore_id);
331 	}
332 
333 	printf("\n====================================================\n");
334 	rte_eal_alarm_set(timer_period * US_PER_S, show_stats_cb, NULL);
335 }
336 
337 static void
338 l2fwd_simple_forward(struct rte_mbuf *m, unsigned portid)
339 {
340 	struct ether_hdr *eth;
341 	void *tmp;
342 	int sent;
343 	unsigned dst_port;
344 	struct rte_eth_dev_tx_buffer *buffer;
345 
346 	dst_port = l2fwd_dst_ports[portid];
347 	eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
348 
349 	/* 02:00:00:00:00:xx */
350 	tmp = &eth->d_addr.addr_bytes[0];
351 	*((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
352 
353 	/* src addr */
354 	ether_addr_copy(&l2fwd_ports_eth_addr[dst_port], &eth->s_addr);
355 
356 	buffer = tx_buffer[dst_port];
357 	sent = rte_eth_tx_buffer(dst_port, 0, buffer, m);
358 	if (sent)
359 		port_statistics[dst_port].tx += sent;
360 }
361 
362 static void
363 l2fwd_job_update_cb(struct rte_jobstats *job, int64_t result)
364 {
365 	int64_t err = job->target - result;
366 	int64_t histeresis = job->target / 8;
367 
368 	if (err < -histeresis) {
369 		if (job->min_period + UPDATE_STEP_DOWN < job->period)
370 			job->period -= UPDATE_STEP_DOWN;
371 	} else if (err > histeresis) {
372 		if (job->period + UPDATE_STEP_UP < job->max_period)
373 			job->period += UPDATE_STEP_UP;
374 	}
375 }
376 
377 static void
378 l2fwd_fwd_job(__rte_unused struct rte_timer *timer, void *arg)
379 {
380 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
381 	struct rte_mbuf *m;
382 
383 	const uint16_t port_idx = (uintptr_t) arg;
384 	const unsigned lcore_id = rte_lcore_id();
385 	struct lcore_queue_conf *qconf = &lcore_queue_conf[lcore_id];
386 	struct rte_jobstats *job = &qconf->port_fwd_jobs[port_idx];
387 	const uint16_t portid = qconf->rx_port_list[port_idx];
388 
389 	uint8_t j;
390 	uint16_t total_nb_rx;
391 
392 	rte_jobstats_start(&qconf->jobs_context, job);
393 
394 	/* Call rx burst 2 times. This allow rte_jobstats logic to see if this
395 	 * function must be called more frequently. */
396 
397 	total_nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
398 			MAX_PKT_BURST);
399 
400 	for (j = 0; j < total_nb_rx; j++) {
401 		m = pkts_burst[j];
402 		rte_prefetch0(rte_pktmbuf_mtod(m, void *));
403 		l2fwd_simple_forward(m, portid);
404 	}
405 
406 	if (total_nb_rx == MAX_PKT_BURST) {
407 		const uint16_t nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
408 				MAX_PKT_BURST);
409 
410 		total_nb_rx += nb_rx;
411 		for (j = 0; j < nb_rx; j++) {
412 			m = pkts_burst[j];
413 			rte_prefetch0(rte_pktmbuf_mtod(m, void *));
414 			l2fwd_simple_forward(m, portid);
415 		}
416 	}
417 
418 	port_statistics[portid].rx += total_nb_rx;
419 
420 	/* Adjust period time in which we are running here. */
421 	if (rte_jobstats_finish(job, total_nb_rx) != 0) {
422 		rte_timer_reset(&qconf->rx_timers[port_idx], job->period, PERIODICAL,
423 				lcore_id, l2fwd_fwd_job, arg);
424 	}
425 }
426 
427 static void
428 l2fwd_flush_job(__rte_unused struct rte_timer *timer, __rte_unused void *arg)
429 {
430 	uint64_t now;
431 	unsigned lcore_id;
432 	struct lcore_queue_conf *qconf;
433 	uint16_t portid;
434 	unsigned i;
435 	uint32_t sent;
436 	struct rte_eth_dev_tx_buffer *buffer;
437 
438 	lcore_id = rte_lcore_id();
439 	qconf = &lcore_queue_conf[lcore_id];
440 
441 	rte_jobstats_start(&qconf->jobs_context, &qconf->flush_job);
442 
443 	now = rte_get_timer_cycles();
444 	lcore_id = rte_lcore_id();
445 	qconf = &lcore_queue_conf[lcore_id];
446 
447 	for (i = 0; i < qconf->n_rx_port; i++) {
448 		portid = l2fwd_dst_ports[qconf->rx_port_list[i]];
449 
450 		if (qconf->next_flush_time[portid] <= now)
451 			continue;
452 
453 		buffer = tx_buffer[portid];
454 		sent = rte_eth_tx_buffer_flush(portid, 0, buffer);
455 		if (sent)
456 			port_statistics[portid].tx += sent;
457 
458 		qconf->next_flush_time[portid] = rte_get_timer_cycles() + drain_tsc;
459 	}
460 
461 	/* Pass target to indicate that this job is happy of time interwal
462 	 * in which it was called. */
463 	rte_jobstats_finish(&qconf->flush_job, qconf->flush_job.target);
464 }
465 
466 /* main processing loop */
467 static void
468 l2fwd_main_loop(void)
469 {
470 	unsigned lcore_id;
471 	unsigned i, portid;
472 	struct lcore_queue_conf *qconf;
473 	uint8_t stats_read_pending = 0;
474 	uint8_t need_manage;
475 
476 	lcore_id = rte_lcore_id();
477 	qconf = &lcore_queue_conf[lcore_id];
478 
479 	if (qconf->n_rx_port == 0) {
480 		RTE_LOG(INFO, L2FWD, "lcore %u has nothing to do\n", lcore_id);
481 		return;
482 	}
483 
484 	RTE_LOG(INFO, L2FWD, "entering main loop on lcore %u\n", lcore_id);
485 
486 	for (i = 0; i < qconf->n_rx_port; i++) {
487 
488 		portid = qconf->rx_port_list[i];
489 		RTE_LOG(INFO, L2FWD, " -- lcoreid=%u portid=%u\n", lcore_id,
490 			portid);
491 	}
492 
493 	rte_jobstats_init(&qconf->idle_job, "idle", 0, 0, 0, 0);
494 
495 	for (;;) {
496 		rte_spinlock_lock(&qconf->lock);
497 
498 		do {
499 			rte_jobstats_context_start(&qconf->jobs_context);
500 
501 			/* Do the Idle job:
502 			 * - Read stats_read_pending flag
503 			 * - check if some real job need to be executed
504 			 */
505 			rte_jobstats_start(&qconf->jobs_context, &qconf->idle_job);
506 
507 			uint64_t repeats = 0;
508 
509 			do {
510 				uint8_t i;
511 				uint64_t now = rte_get_timer_cycles();
512 
513 				repeats++;
514 				need_manage = qconf->flush_timer.expire < now;
515 				/* Check if we was esked to give a stats. */
516 				stats_read_pending =
517 						rte_atomic16_read(&qconf->stats_read_pending);
518 				need_manage |= stats_read_pending;
519 
520 				for (i = 0; i < qconf->n_rx_port && !need_manage; i++)
521 					need_manage = qconf->rx_timers[i].expire < now;
522 
523 			} while (!need_manage);
524 
525 			if (likely(repeats != 1))
526 				rte_jobstats_finish(&qconf->idle_job, qconf->idle_job.target);
527 			else
528 				rte_jobstats_abort(&qconf->idle_job);
529 
530 			rte_timer_manage();
531 			rte_jobstats_context_finish(&qconf->jobs_context);
532 		} while (likely(stats_read_pending == 0));
533 
534 		rte_spinlock_unlock(&qconf->lock);
535 		rte_pause();
536 	}
537 }
538 
539 static int
540 l2fwd_launch_one_lcore(__attribute__((unused)) void *dummy)
541 {
542 	l2fwd_main_loop();
543 	return 0;
544 }
545 
546 /* display usage */
547 static void
548 l2fwd_usage(const char *prgname)
549 {
550 	printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
551 	       "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
552 	       "  -q NQ: number of queue (=ports) per lcore (default is 1)\n"
553 		   "  -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n"
554 		   "  -l set system default locale instead of default (\"C\" locale) for thousands separator in stats.",
555 	       prgname);
556 }
557 
558 static int
559 l2fwd_parse_portmask(const char *portmask)
560 {
561 	char *end = NULL;
562 	unsigned long pm;
563 
564 	/* parse hexadecimal string */
565 	pm = strtoul(portmask, &end, 16);
566 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
567 		return -1;
568 
569 	if (pm == 0)
570 		return -1;
571 
572 	return pm;
573 }
574 
575 static unsigned int
576 l2fwd_parse_nqueue(const char *q_arg)
577 {
578 	char *end = NULL;
579 	unsigned long n;
580 
581 	/* parse hexadecimal string */
582 	n = strtoul(q_arg, &end, 10);
583 	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
584 		return 0;
585 	if (n == 0)
586 		return 0;
587 	if (n >= MAX_RX_QUEUE_PER_LCORE)
588 		return 0;
589 
590 	return n;
591 }
592 
593 static int
594 l2fwd_parse_timer_period(const char *q_arg)
595 {
596 	char *end = NULL;
597 	int n;
598 
599 	/* parse number string */
600 	n = strtol(q_arg, &end, 10);
601 	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
602 		return -1;
603 	if (n >= MAX_TIMER_PERIOD)
604 		return -1;
605 
606 	return n;
607 }
608 
609 /* Parse the argument given in the command line of the application */
610 static int
611 l2fwd_parse_args(int argc, char **argv)
612 {
613 	int opt, ret;
614 	char **argvopt;
615 	int option_index;
616 	char *prgname = argv[0];
617 	static struct option lgopts[] = {
618 		{NULL, 0, 0, 0}
619 	};
620 
621 	argvopt = argv;
622 
623 	while ((opt = getopt_long(argc, argvopt, "p:q:T:l",
624 				  lgopts, &option_index)) != EOF) {
625 
626 		switch (opt) {
627 		/* portmask */
628 		case 'p':
629 			l2fwd_enabled_port_mask = l2fwd_parse_portmask(optarg);
630 			if (l2fwd_enabled_port_mask == 0) {
631 				printf("invalid portmask\n");
632 				l2fwd_usage(prgname);
633 				return -1;
634 			}
635 			break;
636 
637 		/* nqueue */
638 		case 'q':
639 			l2fwd_rx_queue_per_lcore = l2fwd_parse_nqueue(optarg);
640 			if (l2fwd_rx_queue_per_lcore == 0) {
641 				printf("invalid queue number\n");
642 				l2fwd_usage(prgname);
643 				return -1;
644 			}
645 			break;
646 
647 		/* timer period */
648 		case 'T':
649 			timer_period = l2fwd_parse_timer_period(optarg);
650 			if (timer_period < 0) {
651 				printf("invalid timer period\n");
652 				l2fwd_usage(prgname);
653 				return -1;
654 			}
655 			break;
656 
657 		/* For thousands separator in printf. */
658 		case 'l':
659 			setlocale(LC_ALL, "");
660 			break;
661 
662 		/* long options */
663 		case 0:
664 			l2fwd_usage(prgname);
665 			return -1;
666 
667 		default:
668 			l2fwd_usage(prgname);
669 			return -1;
670 		}
671 	}
672 
673 	if (optind >= 0)
674 		argv[optind-1] = prgname;
675 
676 	ret = optind-1;
677 	optind = 1; /* reset getopt lib */
678 	return ret;
679 }
680 
681 /* Check the link status of all ports in up to 9s, and print them finally */
682 static void
683 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask)
684 {
685 #define CHECK_INTERVAL 100 /* 100ms */
686 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
687 	uint16_t portid;
688 	uint8_t count, all_ports_up, print_flag = 0;
689 	struct rte_eth_link link;
690 
691 	printf("\nChecking link status");
692 	fflush(stdout);
693 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
694 		all_ports_up = 1;
695 		for (portid = 0; portid < port_num; portid++) {
696 			if ((port_mask & (1 << portid)) == 0)
697 				continue;
698 			memset(&link, 0, sizeof(link));
699 			rte_eth_link_get_nowait(portid, &link);
700 			/* print link status if flag set */
701 			if (print_flag == 1) {
702 				if (link.link_status)
703 					printf(
704 					"Port%d Link Up. Speed %u Mbps - %s\n",
705 						portid, link.link_speed,
706 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
707 					("full-duplex") : ("half-duplex\n"));
708 				else
709 					printf("Port %d Link Down\n", portid);
710 				continue;
711 			}
712 			/* clear all_ports_up flag if any link down */
713 			if (link.link_status == ETH_LINK_DOWN) {
714 				all_ports_up = 0;
715 				break;
716 			}
717 		}
718 		/* after finally printing all link status, get out */
719 		if (print_flag == 1)
720 			break;
721 
722 		if (all_ports_up == 0) {
723 			printf(".");
724 			fflush(stdout);
725 			rte_delay_ms(CHECK_INTERVAL);
726 		}
727 
728 		/* set the print_flag if all ports up or timeout */
729 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
730 			print_flag = 1;
731 			printf("done\n");
732 		}
733 	}
734 }
735 
736 int
737 main(int argc, char **argv)
738 {
739 	struct lcore_queue_conf *qconf;
740 	unsigned lcore_id, rx_lcore_id;
741 	unsigned nb_ports_in_mask = 0;
742 	int ret;
743 	char name[RTE_JOBSTATS_NAMESIZE];
744 	uint16_t nb_ports;
745 	uint16_t nb_ports_available;
746 	uint16_t portid, last_port;
747 	uint8_t i;
748 
749 	/* init EAL */
750 	ret = rte_eal_init(argc, argv);
751 	if (ret < 0)
752 		rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
753 	argc -= ret;
754 	argv += ret;
755 
756 	/* parse application arguments (after the EAL ones) */
757 	ret = l2fwd_parse_args(argc, argv);
758 	if (ret < 0)
759 		rte_exit(EXIT_FAILURE, "Invalid L2FWD arguments\n");
760 
761 	rte_timer_subsystem_init();
762 
763 	/* fetch default timer frequency. */
764 	hz = rte_get_timer_hz();
765 
766 	/* create the mbuf pool */
767 	l2fwd_pktmbuf_pool =
768 		rte_pktmbuf_pool_create("mbuf_pool", NB_MBUF, 32,
769 			0, RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
770 	if (l2fwd_pktmbuf_pool == NULL)
771 		rte_exit(EXIT_FAILURE, "Cannot init mbuf pool\n");
772 
773 	nb_ports = rte_eth_dev_count();
774 	if (nb_ports == 0)
775 		rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
776 
777 	/* reset l2fwd_dst_ports */
778 	for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++)
779 		l2fwd_dst_ports[portid] = 0;
780 	last_port = 0;
781 
782 	/*
783 	 * Each logical core is assigned a dedicated TX queue on each port.
784 	 */
785 	for (portid = 0; portid < nb_ports; portid++) {
786 		/* skip ports that are not enabled */
787 		if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
788 			continue;
789 
790 		if (nb_ports_in_mask % 2) {
791 			l2fwd_dst_ports[portid] = last_port;
792 			l2fwd_dst_ports[last_port] = portid;
793 		} else
794 			last_port = portid;
795 
796 		nb_ports_in_mask++;
797 	}
798 	if (nb_ports_in_mask % 2) {
799 		printf("Notice: odd number of ports in portmask.\n");
800 		l2fwd_dst_ports[last_port] = last_port;
801 	}
802 
803 	rx_lcore_id = 0;
804 	qconf = NULL;
805 
806 	/* Initialize the port/queue configuration of each logical core */
807 	for (portid = 0; portid < nb_ports; portid++) {
808 		/* skip ports that are not enabled */
809 		if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
810 			continue;
811 
812 		/* get the lcore_id for this port */
813 		while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
814 		       lcore_queue_conf[rx_lcore_id].n_rx_port ==
815 		       l2fwd_rx_queue_per_lcore) {
816 			rx_lcore_id++;
817 			if (rx_lcore_id >= RTE_MAX_LCORE)
818 				rte_exit(EXIT_FAILURE, "Not enough cores\n");
819 		}
820 
821 		if (qconf != &lcore_queue_conf[rx_lcore_id])
822 			/* Assigned a new logical core in the loop above. */
823 			qconf = &lcore_queue_conf[rx_lcore_id];
824 
825 		qconf->rx_port_list[qconf->n_rx_port] = portid;
826 		qconf->n_rx_port++;
827 		printf("Lcore %u: RX port %u\n", rx_lcore_id, portid);
828 	}
829 
830 	nb_ports_available = nb_ports;
831 
832 	/* Initialise each port */
833 	for (portid = 0; portid < nb_ports; portid++) {
834 		struct rte_eth_dev_info dev_info;
835 		struct rte_eth_rxconf rxq_conf;
836 		struct rte_eth_txconf txq_conf;
837 		struct rte_eth_conf local_port_conf = port_conf;
838 
839 		/* skip ports that are not enabled */
840 		if ((l2fwd_enabled_port_mask & (1 << portid)) == 0) {
841 			printf("Skipping disabled port %u\n", portid);
842 			nb_ports_available--;
843 			continue;
844 		}
845 		/* init port */
846 		printf("Initializing port %u... ", portid);
847 		fflush(stdout);
848 		rte_eth_dev_info_get(portid, &dev_info);
849 		if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
850 			local_port_conf.txmode.offloads |=
851 				DEV_TX_OFFLOAD_MBUF_FAST_FREE;
852 		ret = rte_eth_dev_configure(portid, 1, 1, &local_port_conf);
853 		if (ret < 0)
854 			rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
855 				  ret, portid);
856 
857 		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
858 						       &nb_txd);
859 		if (ret < 0)
860 			rte_exit(EXIT_FAILURE,
861 				 "Cannot adjust number of descriptors: err=%d, port=%u\n",
862 				 ret, portid);
863 
864 		rte_eth_macaddr_get(portid, &l2fwd_ports_eth_addr[portid]);
865 
866 		/* init one RX queue */
867 		fflush(stdout);
868 		rxq_conf = dev_info.default_rxconf;
869 		rxq_conf.offloads = local_port_conf.rxmode.offloads;
870 		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
871 					     rte_eth_dev_socket_id(portid),
872 					     &rxq_conf,
873 					     l2fwd_pktmbuf_pool);
874 		if (ret < 0)
875 			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup:err=%d, port=%u\n",
876 				  ret, portid);
877 
878 		/* init one TX queue on each port */
879 		txq_conf = dev_info.default_txconf;
880 		txq_conf.txq_flags = ETH_TXQ_FLAGS_IGNORE;
881 		txq_conf.offloads = local_port_conf.txmode.offloads;
882 		fflush(stdout);
883 		ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
884 				rte_eth_dev_socket_id(portid),
885 				&txq_conf);
886 		if (ret < 0)
887 			rte_exit(EXIT_FAILURE,
888 			"rte_eth_tx_queue_setup:err=%d, port=%u\n",
889 				ret, portid);
890 
891 		/* Initialize TX buffers */
892 		tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
893 				RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
894 				rte_eth_dev_socket_id(portid));
895 		if (tx_buffer[portid] == NULL)
896 			rte_exit(EXIT_FAILURE, "Cannot allocate buffer for tx on port %u\n",
897 					portid);
898 
899 		rte_eth_tx_buffer_init(tx_buffer[portid], MAX_PKT_BURST);
900 
901 		ret = rte_eth_tx_buffer_set_err_callback(tx_buffer[portid],
902 				rte_eth_tx_buffer_count_callback,
903 				&port_statistics[portid].dropped);
904 		if (ret < 0)
905 			rte_exit(EXIT_FAILURE,
906 			"Cannot set error callback for tx buffer on port %u\n",
907 				 portid);
908 
909 		/* Start device */
910 		ret = rte_eth_dev_start(portid);
911 		if (ret < 0)
912 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start:err=%d, port=%u\n",
913 				  ret, portid);
914 
915 		printf("done:\n");
916 
917 		rte_eth_promiscuous_enable(portid);
918 
919 		printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
920 				portid,
921 				l2fwd_ports_eth_addr[portid].addr_bytes[0],
922 				l2fwd_ports_eth_addr[portid].addr_bytes[1],
923 				l2fwd_ports_eth_addr[portid].addr_bytes[2],
924 				l2fwd_ports_eth_addr[portid].addr_bytes[3],
925 				l2fwd_ports_eth_addr[portid].addr_bytes[4],
926 				l2fwd_ports_eth_addr[portid].addr_bytes[5]);
927 
928 		/* initialize port stats */
929 		memset(&port_statistics, 0, sizeof(port_statistics));
930 	}
931 
932 	if (!nb_ports_available) {
933 		rte_exit(EXIT_FAILURE,
934 			"All available ports are disabled. Please set portmask.\n");
935 	}
936 
937 	check_all_ports_link_status(nb_ports, l2fwd_enabled_port_mask);
938 
939 	drain_tsc = (hz + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
940 
941 	RTE_LCORE_FOREACH(lcore_id) {
942 		qconf = &lcore_queue_conf[lcore_id];
943 
944 		rte_spinlock_init(&qconf->lock);
945 
946 		if (rte_jobstats_context_init(&qconf->jobs_context) != 0)
947 			rte_panic("Jobs stats context for core %u init failed\n", lcore_id);
948 
949 		if (qconf->n_rx_port == 0) {
950 			RTE_LOG(INFO, L2FWD,
951 				"lcore %u: no ports so no jobs stats context initialization\n",
952 				lcore_id);
953 			continue;
954 		}
955 		/* Add flush job.
956 		 * Set fixed period by setting min = max = initial period. Set target to
957 		 * zero as it is irrelevant for this job. */
958 		rte_jobstats_init(&qconf->flush_job, "flush", drain_tsc, drain_tsc,
959 				drain_tsc, 0);
960 
961 		rte_timer_init(&qconf->flush_timer);
962 		ret = rte_timer_reset(&qconf->flush_timer, drain_tsc, PERIODICAL,
963 				lcore_id, &l2fwd_flush_job, NULL);
964 
965 		if (ret < 0) {
966 			rte_exit(1, "Failed to reset flush job timer for lcore %u: %s",
967 					lcore_id, rte_strerror(-ret));
968 		}
969 
970 		for (i = 0; i < qconf->n_rx_port; i++) {
971 			struct rte_jobstats *job = &qconf->port_fwd_jobs[i];
972 
973 			portid = qconf->rx_port_list[i];
974 			printf("Setting forward job for port %u\n", portid);
975 
976 			snprintf(name, RTE_DIM(name), "port %u fwd", portid);
977 			/* Setup forward job.
978 			 * Set min, max and initial period. Set target to MAX_PKT_BURST as
979 			 * this is desired optimal RX/TX burst size. */
980 			rte_jobstats_init(job, name, 0, drain_tsc, 0, MAX_PKT_BURST);
981 			rte_jobstats_set_update_period_function(job, l2fwd_job_update_cb);
982 
983 			rte_timer_init(&qconf->rx_timers[i]);
984 			ret = rte_timer_reset(&qconf->rx_timers[i], 0, PERIODICAL, lcore_id,
985 					&l2fwd_fwd_job, (void *)(uintptr_t)i);
986 
987 			if (ret < 0) {
988 				rte_exit(1, "Failed to reset lcore %u port %u job timer: %s",
989 						lcore_id, qconf->rx_port_list[i], rte_strerror(-ret));
990 			}
991 		}
992 	}
993 
994 	if (timer_period)
995 		rte_eal_alarm_set(timer_period * MS_PER_S, show_stats_cb, NULL);
996 	else
997 		RTE_LOG(INFO, L2FWD, "Stats display disabled\n");
998 
999 	/* launch per-lcore init on every lcore */
1000 	rte_eal_mp_remote_launch(l2fwd_launch_one_lcore, NULL, CALL_MASTER);
1001 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1002 		if (rte_eal_wait_lcore(lcore_id) < 0)
1003 			return -1;
1004 	}
1005 
1006 	return 0;
1007 }
1008