xref: /dpdk/examples/l2fwd-jobstats/main.c (revision daa02b5cddbb8e11b31d41e2bf7bb1ae64dcae2f)
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_cycles.h>
20 #include <rte_prefetch.h>
21 #include <rte_lcore.h>
22 #include <rte_per_lcore.h>
23 #include <rte_branch_prediction.h>
24 #include <rte_interrupts.h>
25 #include <rte_debug.h>
26 #include <rte_ether.h>
27 #include <rte_ethdev.h>
28 #include <rte_mempool.h>
29 #include <rte_mbuf.h>
30 #include <rte_spinlock.h>
31 
32 #include <rte_errno.h>
33 #include <rte_jobstats.h>
34 #include <rte_timer.h>
35 #include <rte_alarm.h>
36 #include <rte_pause.h>
37 
38 #define RTE_LOGTYPE_L2FWD RTE_LOGTYPE_USER1
39 
40 #define NB_MBUF   8192
41 
42 #define MAX_PKT_BURST 32
43 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
44 
45 /*
46  * Configurable number of RX/TX ring descriptors
47  */
48 #define RTE_TEST_RX_DESC_DEFAULT 1024
49 #define RTE_TEST_TX_DESC_DEFAULT 1024
50 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
51 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
52 
53 /* ethernet addresses of ports */
54 static struct rte_ether_addr l2fwd_ports_eth_addr[RTE_MAX_ETHPORTS];
55 
56 /* mask of enabled ports */
57 static uint32_t l2fwd_enabled_port_mask;
58 
59 /* list of enabled ports */
60 static uint32_t l2fwd_dst_ports[RTE_MAX_ETHPORTS];
61 
62 #define UPDATE_STEP_UP 1
63 #define UPDATE_STEP_DOWN 32
64 
65 static unsigned int l2fwd_rx_queue_per_lcore = 1;
66 
67 #define MAX_RX_QUEUE_PER_LCORE 16
68 #define MAX_TX_QUEUE_PER_PORT 16
69 /* List of queues to be polled for given lcore. 8< */
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 	uint16_t stats_read_pending;
84 	rte_spinlock_t lock;
85 } __rte_cache_aligned;
86 /* >8 End of list of queues to be polled for given lcore. */
87 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
88 
89 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
90 
91 static struct rte_eth_conf port_conf = {
92 	.rxmode = {
93 		.split_hdr_size = 0,
94 	},
95 	.txmode = {
96 		.mq_mode = RTE_ETH_MQ_TX_NONE,
97 	},
98 };
99 
100 struct rte_mempool *l2fwd_pktmbuf_pool = NULL;
101 
102 /* Per-port statistics struct */
103 struct l2fwd_port_statistics {
104 	uint64_t tx;
105 	uint64_t rx;
106 	uint64_t dropped;
107 } __rte_cache_aligned;
108 struct l2fwd_port_statistics port_statistics[RTE_MAX_ETHPORTS];
109 
110 /* 1 day max */
111 #define MAX_TIMER_PERIOD 86400
112 /* default period is 10 seconds */
113 static int64_t timer_period = 10;
114 /* default timer frequency */
115 static double hz;
116 /* BURST_TX_DRAIN_US converted to cycles */
117 uint64_t drain_tsc;
118 /* Convert cycles to ns */
119 static inline double
120 cycles_to_ns(uint64_t cycles)
121 {
122 	double t = cycles;
123 
124 	t *= (double)NS_PER_S;
125 	t /= hz;
126 	return t;
127 }
128 
129 static void
130 show_lcore_stats(unsigned lcore_id)
131 {
132 	struct lcore_queue_conf *qconf = &lcore_queue_conf[lcore_id];
133 	struct rte_jobstats_context *ctx = &qconf->jobs_context;
134 	struct rte_jobstats *job;
135 	uint8_t i;
136 
137 	/* LCore statistics. */
138 	uint64_t stats_period, loop_count;
139 	uint64_t exec, exec_min, exec_max;
140 	uint64_t management, management_min, management_max;
141 	uint64_t busy, busy_min, busy_max;
142 
143 	/* Jobs statistics. */
144 	const uint16_t port_cnt = qconf->n_rx_port;
145 	uint64_t jobs_exec_cnt[port_cnt], jobs_period[port_cnt];
146 	uint64_t jobs_exec[port_cnt], jobs_exec_min[port_cnt],
147 				jobs_exec_max[port_cnt];
148 
149 	uint64_t flush_exec_cnt, flush_period;
150 	uint64_t flush_exec, flush_exec_min, flush_exec_max;
151 
152 	uint64_t idle_exec_cnt;
153 	uint64_t idle_exec, idle_exec_min, idle_exec_max;
154 	uint64_t collection_time = rte_get_timer_cycles();
155 
156 	/* Ask forwarding thread to give us stats. */
157 	__atomic_store_n(&qconf->stats_read_pending, 1, __ATOMIC_RELAXED);
158 	rte_spinlock_lock(&qconf->lock);
159 	__atomic_store_n(&qconf->stats_read_pending, 0, __ATOMIC_RELAXED);
160 
161 	/* Collect context statistics. */
162 	stats_period = ctx->state_time - ctx->start_time;
163 	loop_count = ctx->loop_cnt;
164 
165 	exec = ctx->exec_time;
166 	exec_min = ctx->min_exec_time;
167 	exec_max = ctx->max_exec_time;
168 
169 	management = ctx->management_time;
170 	management_min = ctx->min_management_time;
171 	management_max = ctx->max_management_time;
172 
173 	rte_jobstats_context_reset(ctx);
174 
175 	for (i = 0; i < port_cnt; i++) {
176 		job = &qconf->port_fwd_jobs[i];
177 
178 		jobs_exec_cnt[i] = job->exec_cnt;
179 		jobs_period[i] = job->period;
180 
181 		jobs_exec[i] = job->exec_time;
182 		jobs_exec_min[i] = job->min_exec_time;
183 		jobs_exec_max[i] = job->max_exec_time;
184 
185 		rte_jobstats_reset(job);
186 	}
187 
188 	flush_exec_cnt = qconf->flush_job.exec_cnt;
189 	flush_period = qconf->flush_job.period;
190 	flush_exec = qconf->flush_job.exec_time;
191 	flush_exec_min = qconf->flush_job.min_exec_time;
192 	flush_exec_max = qconf->flush_job.max_exec_time;
193 	rte_jobstats_reset(&qconf->flush_job);
194 
195 	idle_exec_cnt = qconf->idle_job.exec_cnt;
196 	idle_exec = qconf->idle_job.exec_time;
197 	idle_exec_min = qconf->idle_job.min_exec_time;
198 	idle_exec_max = qconf->idle_job.max_exec_time;
199 	rte_jobstats_reset(&qconf->idle_job);
200 
201 	rte_spinlock_unlock(&qconf->lock);
202 
203 	exec -= idle_exec;
204 	busy = exec + management;
205 	busy_min = exec_min + management_min;
206 	busy_max = exec_max + management_max;
207 
208 
209 	collection_time = rte_get_timer_cycles() - collection_time;
210 
211 #define STAT_FMT "\n%-18s %'14.0f %6.1f%% %'10.0f %'10.0f %'10.0f"
212 
213 	printf("\n----------------"
214 			"\nLCore %3u: statistics (time in ns, collected in %'9.0f)"
215 			"\n%-18s %14s %7s %10s %10s %10s "
216 			"\n%-18s %'14.0f"
217 			"\n%-18s %'14" PRIu64
218 			STAT_FMT /* Exec */
219 			STAT_FMT /* Management */
220 			STAT_FMT /* Busy */
221 			STAT_FMT, /* Idle  */
222 			lcore_id, cycles_to_ns(collection_time),
223 			"Stat type", "total", "%total", "avg", "min", "max",
224 			"Stats duration:", cycles_to_ns(stats_period),
225 			"Loop count:", loop_count,
226 			"Exec time",
227 			cycles_to_ns(exec), exec * 100.0 / stats_period,
228 			cycles_to_ns(loop_count  ? exec / loop_count : 0),
229 			cycles_to_ns(exec_min),
230 			cycles_to_ns(exec_max),
231 			"Management time",
232 			cycles_to_ns(management), management * 100.0 / stats_period,
233 			cycles_to_ns(loop_count  ? management / loop_count : 0),
234 			cycles_to_ns(management_min),
235 			cycles_to_ns(management_max),
236 			"Exec + management",
237 			cycles_to_ns(busy),  busy * 100.0 / stats_period,
238 			cycles_to_ns(loop_count ? busy / loop_count : 0),
239 			cycles_to_ns(busy_min),
240 			cycles_to_ns(busy_max),
241 			"Idle (job)",
242 			cycles_to_ns(idle_exec), idle_exec * 100.0 / stats_period,
243 			cycles_to_ns(idle_exec_cnt ? idle_exec / idle_exec_cnt : 0),
244 			cycles_to_ns(idle_exec_min),
245 			cycles_to_ns(idle_exec_max));
246 
247 	for (i = 0; i < qconf->n_rx_port; i++) {
248 		job = &qconf->port_fwd_jobs[i];
249 		printf("\n\nJob %" PRIu32 ": %-20s "
250 				"\n%-18s %'14" PRIu64
251 				"\n%-18s %'14.0f"
252 				STAT_FMT,
253 				i, job->name,
254 				"Exec count:", jobs_exec_cnt[i],
255 				"Exec period: ", cycles_to_ns(jobs_period[i]),
256 				"Exec time",
257 				cycles_to_ns(jobs_exec[i]), jobs_exec[i] * 100.0 / stats_period,
258 				cycles_to_ns(jobs_exec_cnt[i] ? jobs_exec[i] / jobs_exec_cnt[i]
259 						: 0),
260 				cycles_to_ns(jobs_exec_min[i]),
261 				cycles_to_ns(jobs_exec_max[i]));
262 	}
263 
264 	if (qconf->n_rx_port > 0) {
265 		job = &qconf->flush_job;
266 		printf("\n\nJob %" PRIu32 ": %-20s "
267 				"\n%-18s %'14" PRIu64
268 				"\n%-18s %'14.0f"
269 				STAT_FMT,
270 				i, job->name,
271 				"Exec count:", flush_exec_cnt,
272 				"Exec period: ", cycles_to_ns(flush_period),
273 				"Exec time",
274 				cycles_to_ns(flush_exec), flush_exec * 100.0 / stats_period,
275 				cycles_to_ns(flush_exec_cnt ? flush_exec / flush_exec_cnt : 0),
276 				cycles_to_ns(flush_exec_min),
277 				cycles_to_ns(flush_exec_max));
278 	}
279 }
280 
281 /* Print out statistics on packets dropped */
282 static void
283 show_stats_cb(__rte_unused void *param)
284 {
285 	uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
286 	unsigned portid, lcore_id;
287 
288 	total_packets_dropped = 0;
289 	total_packets_tx = 0;
290 	total_packets_rx = 0;
291 
292 	const char clr[] = { 27, '[', '2', 'J', '\0' };
293 	const char topLeft[] = { 27, '[', '1', ';', '1', 'H', '\0' };
294 
295 	/* Clear screen and move to top left */
296 	printf("%s%s"
297 			"\nPort statistics ===================================",
298 			clr, topLeft);
299 
300 	for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
301 		/* skip disabled ports */
302 		if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
303 			continue;
304 		printf("\nStatistics for port %u ------------------------------"
305 				"\nPackets sent: %24"PRIu64
306 				"\nPackets received: %20"PRIu64
307 				"\nPackets dropped: %21"PRIu64,
308 				portid,
309 				port_statistics[portid].tx,
310 				port_statistics[portid].rx,
311 				port_statistics[portid].dropped);
312 
313 		total_packets_dropped += port_statistics[portid].dropped;
314 		total_packets_tx += port_statistics[portid].tx;
315 		total_packets_rx += port_statistics[portid].rx;
316 	}
317 
318 	printf("\nAggregate statistics ==============================="
319 			"\nTotal packets sent: %18"PRIu64
320 			"\nTotal packets received: %14"PRIu64
321 			"\nTotal packets dropped: %15"PRIu64
322 			"\n====================================================",
323 			total_packets_tx,
324 			total_packets_rx,
325 			total_packets_dropped);
326 
327 	RTE_LCORE_FOREACH(lcore_id) {
328 		if (lcore_queue_conf[lcore_id].n_rx_port > 0)
329 			show_lcore_stats(lcore_id);
330 	}
331 
332 	printf("\n====================================================\n");
333 
334 	fflush(stdout);
335 
336 	rte_eal_alarm_set(timer_period * US_PER_S, show_stats_cb, NULL);
337 }
338 
339 /* Start of l2fwd_simple_forward. 8< */
340 static void
341 l2fwd_simple_forward(struct rte_mbuf *m, unsigned portid)
342 {
343 	struct rte_ether_hdr *eth;
344 	void *tmp;
345 	int sent;
346 	unsigned dst_port;
347 	struct rte_eth_dev_tx_buffer *buffer;
348 
349 	dst_port = l2fwd_dst_ports[portid];
350 	eth = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
351 
352 	/* 02:00:00:00:00:xx */
353 	tmp = &eth->dst_addr.addr_bytes[0];
354 	*((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
355 
356 	/* src addr */
357 	rte_ether_addr_copy(&l2fwd_ports_eth_addr[dst_port], &eth->src_addr);
358 
359 	buffer = tx_buffer[dst_port];
360 	sent = rte_eth_tx_buffer(dst_port, 0, buffer, m);
361 	if (sent)
362 		port_statistics[dst_port].tx += sent;
363 }
364 /* >8 End of l2fwd_simple_forward. */
365 
366 static void
367 l2fwd_job_update_cb(struct rte_jobstats *job, int64_t result)
368 {
369 	int64_t err = job->target - result;
370 	int64_t histeresis = job->target / 8;
371 
372 	if (err < -histeresis) {
373 		if (job->min_period + UPDATE_STEP_DOWN < job->period)
374 			job->period -= UPDATE_STEP_DOWN;
375 	} else if (err > histeresis) {
376 		if (job->period + UPDATE_STEP_UP < job->max_period)
377 			job->period += UPDATE_STEP_UP;
378 	}
379 }
380 
381 static void
382 l2fwd_fwd_job(__rte_unused struct rte_timer *timer, void *arg)
383 {
384 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
385 	struct rte_mbuf *m;
386 
387 	const uint16_t port_idx = (uintptr_t) arg;
388 	const unsigned lcore_id = rte_lcore_id();
389 	struct lcore_queue_conf *qconf = &lcore_queue_conf[lcore_id];
390 	struct rte_jobstats *job = &qconf->port_fwd_jobs[port_idx];
391 	const uint16_t portid = qconf->rx_port_list[port_idx];
392 
393 	uint8_t j;
394 	uint16_t total_nb_rx;
395 
396 	rte_jobstats_start(&qconf->jobs_context, job);
397 
398 	/* Call rx burst 2 times. This allow rte_jobstats logic to see if this
399 	 * function must be called more frequently. */
400 
401 	/* Call rx burst 2 times. 8< */
402 	total_nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
403 			MAX_PKT_BURST);
404 
405 	for (j = 0; j < total_nb_rx; j++) {
406 		m = pkts_burst[j];
407 		rte_prefetch0(rte_pktmbuf_mtod(m, void *));
408 		l2fwd_simple_forward(m, portid);
409 	}
410 	/* >8 End of call rx burst 2 times. */
411 
412 	/* Read second try. 8< */
413 	if (total_nb_rx == MAX_PKT_BURST) {
414 		const uint16_t nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
415 				MAX_PKT_BURST);
416 
417 		total_nb_rx += nb_rx;
418 		for (j = 0; j < nb_rx; j++) {
419 			m = pkts_burst[j];
420 			rte_prefetch0(rte_pktmbuf_mtod(m, void *));
421 			l2fwd_simple_forward(m, portid);
422 		}
423 	}
424 	/* >8 End of read second try. */
425 
426 	port_statistics[portid].rx += total_nb_rx;
427 
428 	/* Adjust period time in which we are running here. 8< */
429 	if (rte_jobstats_finish(job, total_nb_rx) != 0) {
430 		rte_timer_reset(&qconf->rx_timers[port_idx], job->period, PERIODICAL,
431 				lcore_id, l2fwd_fwd_job, arg);
432 	}
433 	/* >8 End of adjust period time in which we are running. */
434 }
435 
436 /* Draining TX queue of each port. 8< */
437 static void
438 l2fwd_flush_job(__rte_unused struct rte_timer *timer, __rte_unused void *arg)
439 {
440 	uint64_t now;
441 	unsigned lcore_id;
442 	struct lcore_queue_conf *qconf;
443 	uint16_t portid;
444 	unsigned i;
445 	uint32_t sent;
446 	struct rte_eth_dev_tx_buffer *buffer;
447 
448 	lcore_id = rte_lcore_id();
449 	qconf = &lcore_queue_conf[lcore_id];
450 
451 	rte_jobstats_start(&qconf->jobs_context, &qconf->flush_job);
452 
453 	now = rte_get_timer_cycles();
454 	lcore_id = rte_lcore_id();
455 	qconf = &lcore_queue_conf[lcore_id];
456 
457 	for (i = 0; i < qconf->n_rx_port; i++) {
458 		portid = l2fwd_dst_ports[qconf->rx_port_list[i]];
459 
460 		if (qconf->next_flush_time[portid] <= now)
461 			continue;
462 
463 		buffer = tx_buffer[portid];
464 		sent = rte_eth_tx_buffer_flush(portid, 0, buffer);
465 		if (sent)
466 			port_statistics[portid].tx += sent;
467 
468 		qconf->next_flush_time[portid] = rte_get_timer_cycles() + drain_tsc;
469 	}
470 
471 	/* Pass target to indicate that this job is happy of time interwal
472 	 * in which it was called. */
473 	rte_jobstats_finish(&qconf->flush_job, qconf->flush_job.target);
474 }
475 /* >8 End of draining TX queue of each port. */
476 
477 /* main processing loop */
478 static void
479 l2fwd_main_loop(void)
480 {
481 	unsigned lcore_id;
482 	unsigned i, portid;
483 	struct lcore_queue_conf *qconf;
484 	uint8_t stats_read_pending = 0;
485 	uint8_t need_manage;
486 
487 	lcore_id = rte_lcore_id();
488 	qconf = &lcore_queue_conf[lcore_id];
489 
490 	if (qconf->n_rx_port == 0) {
491 		RTE_LOG(INFO, L2FWD, "lcore %u has nothing to do\n", lcore_id);
492 		return;
493 	}
494 
495 	RTE_LOG(INFO, L2FWD, "entering main loop on lcore %u\n", lcore_id);
496 
497 	for (i = 0; i < qconf->n_rx_port; i++) {
498 
499 		portid = qconf->rx_port_list[i];
500 		RTE_LOG(INFO, L2FWD, " -- lcoreid=%u portid=%u\n", lcore_id,
501 			portid);
502 	}
503 
504 	rte_jobstats_init(&qconf->idle_job, "idle", 0, 0, 0, 0);
505 
506 	/* Minimize impact of stats reading. 8< */
507 	for (;;) {
508 		rte_spinlock_lock(&qconf->lock);
509 
510 		do {
511 			rte_jobstats_context_start(&qconf->jobs_context);
512 
513 			/* Do the Idle job:
514 			 * - Read stats_read_pending flag
515 			 * - check if some real job need to be executed
516 			 */
517 			rte_jobstats_start(&qconf->jobs_context, &qconf->idle_job);
518 
519 			uint64_t repeats = 0;
520 
521 			do {
522 				uint8_t i;
523 				uint64_t now = rte_get_timer_cycles();
524 
525 				repeats++;
526 				need_manage = qconf->flush_timer.expire < now;
527 				/* Check if we was esked to give a stats. */
528 				stats_read_pending = __atomic_load_n(&qconf->stats_read_pending,
529 						__ATOMIC_RELAXED);
530 				need_manage |= stats_read_pending;
531 
532 				for (i = 0; i < qconf->n_rx_port && !need_manage; i++)
533 					need_manage = qconf->rx_timers[i].expire < now;
534 
535 			} while (!need_manage);
536 
537 			if (likely(repeats != 1))
538 				rte_jobstats_finish(&qconf->idle_job, qconf->idle_job.target);
539 			else
540 				rte_jobstats_abort(&qconf->idle_job);
541 
542 			rte_timer_manage();
543 			rte_jobstats_context_finish(&qconf->jobs_context);
544 		} while (likely(stats_read_pending == 0));
545 
546 		rte_spinlock_unlock(&qconf->lock);
547 		rte_pause();
548 	}
549 	/* >8 End of minimize impact of stats reading. */
550 }
551 
552 static int
553 l2fwd_launch_one_lcore(__rte_unused void *dummy)
554 {
555 	l2fwd_main_loop();
556 	return 0;
557 }
558 
559 /* display usage */
560 static void
561 l2fwd_usage(const char *prgname)
562 {
563 	printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
564 	       "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
565 	       "  -q NQ: number of queue (=ports) per lcore (default is 1)\n"
566 		   "  -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n"
567 		   "  -l set system default locale instead of default (\"C\" locale) for thousands separator in stats.",
568 	       prgname);
569 }
570 
571 static int
572 l2fwd_parse_portmask(const char *portmask)
573 {
574 	char *end = NULL;
575 	unsigned long pm;
576 
577 	/* parse hexadecimal string */
578 	pm = strtoul(portmask, &end, 16);
579 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
580 		return 0;
581 
582 	return pm;
583 }
584 
585 static unsigned int
586 l2fwd_parse_nqueue(const char *q_arg)
587 {
588 	char *end = NULL;
589 	unsigned long n;
590 
591 	/* parse hexadecimal string */
592 	n = strtoul(q_arg, &end, 10);
593 	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
594 		return 0;
595 	if (n == 0)
596 		return 0;
597 	if (n >= MAX_RX_QUEUE_PER_LCORE)
598 		return 0;
599 
600 	return n;
601 }
602 
603 static int
604 l2fwd_parse_timer_period(const char *q_arg)
605 {
606 	char *end = NULL;
607 	int n;
608 
609 	/* parse number string */
610 	n = strtol(q_arg, &end, 10);
611 	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
612 		return -1;
613 	if (n >= MAX_TIMER_PERIOD)
614 		return -1;
615 
616 	return n;
617 }
618 
619 /* Parse the argument given in the command line of the application */
620 static int
621 l2fwd_parse_args(int argc, char **argv)
622 {
623 	int opt, ret;
624 	char **argvopt;
625 	int option_index;
626 	char *prgname = argv[0];
627 	static struct option lgopts[] = {
628 		{NULL, 0, 0, 0}
629 	};
630 
631 	argvopt = argv;
632 
633 	while ((opt = getopt_long(argc, argvopt, "p:q:T:l",
634 				  lgopts, &option_index)) != EOF) {
635 
636 		switch (opt) {
637 		/* portmask */
638 		case 'p':
639 			l2fwd_enabled_port_mask = l2fwd_parse_portmask(optarg);
640 			if (l2fwd_enabled_port_mask == 0) {
641 				printf("invalid portmask\n");
642 				l2fwd_usage(prgname);
643 				return -1;
644 			}
645 			break;
646 
647 		/* nqueue */
648 		case 'q':
649 			l2fwd_rx_queue_per_lcore = l2fwd_parse_nqueue(optarg);
650 			if (l2fwd_rx_queue_per_lcore == 0) {
651 				printf("invalid queue number\n");
652 				l2fwd_usage(prgname);
653 				return -1;
654 			}
655 			break;
656 
657 		/* timer period */
658 		case 'T':
659 			timer_period = l2fwd_parse_timer_period(optarg);
660 			if (timer_period < 0) {
661 				printf("invalid timer period\n");
662 				l2fwd_usage(prgname);
663 				return -1;
664 			}
665 			break;
666 
667 		/* For thousands separator in printf. */
668 		case 'l':
669 			setlocale(LC_ALL, "");
670 			break;
671 
672 		/* long options */
673 		case 0:
674 			l2fwd_usage(prgname);
675 			return -1;
676 
677 		default:
678 			l2fwd_usage(prgname);
679 			return -1;
680 		}
681 	}
682 
683 	if (optind >= 0)
684 		argv[optind-1] = prgname;
685 
686 	ret = optind-1;
687 	optind = 1; /* reset getopt lib */
688 	return ret;
689 }
690 
691 /* Check the link status of all ports in up to 9s, and print them finally */
692 static void
693 check_all_ports_link_status(uint32_t port_mask)
694 {
695 #define CHECK_INTERVAL 100 /* 100ms */
696 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
697 	uint16_t portid;
698 	uint8_t count, all_ports_up, print_flag = 0;
699 	struct rte_eth_link link;
700 	int ret;
701 	char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
702 
703 	printf("\nChecking link status");
704 	fflush(stdout);
705 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
706 		all_ports_up = 1;
707 		RTE_ETH_FOREACH_DEV(portid) {
708 			if ((port_mask & (1 << portid)) == 0)
709 				continue;
710 			memset(&link, 0, sizeof(link));
711 			ret = rte_eth_link_get_nowait(portid, &link);
712 			if (ret < 0) {
713 				all_ports_up = 0;
714 				if (print_flag == 1)
715 					printf("Port %u link get failed: %s\n",
716 						portid, rte_strerror(-ret));
717 				continue;
718 			}
719 			/* print link status if flag set */
720 			if (print_flag == 1) {
721 				rte_eth_link_to_str(link_status_text,
722 					sizeof(link_status_text), &link);
723 				printf("Port %d %s\n", portid,
724 				       link_status_text);
725 				continue;
726 			}
727 			/* clear all_ports_up flag if any link down */
728 			if (link.link_status == RTE_ETH_LINK_DOWN) {
729 				all_ports_up = 0;
730 				break;
731 			}
732 		}
733 		/* after finally printing all link status, get out */
734 		if (print_flag == 1)
735 			break;
736 
737 		if (all_ports_up == 0) {
738 			printf(".");
739 			fflush(stdout);
740 			rte_delay_ms(CHECK_INTERVAL);
741 		}
742 
743 		/* set the print_flag if all ports up or timeout */
744 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
745 			print_flag = 1;
746 			printf("done\n");
747 		}
748 	}
749 }
750 
751 int
752 main(int argc, char **argv)
753 {
754 	struct lcore_queue_conf *qconf;
755 	unsigned lcore_id, rx_lcore_id;
756 	unsigned nb_ports_in_mask = 0;
757 	int ret;
758 	char name[RTE_JOBSTATS_NAMESIZE];
759 	uint16_t nb_ports;
760 	uint16_t nb_ports_available = 0;
761 	uint16_t portid, last_port;
762 	uint8_t i;
763 
764 	/* Init EAL. 8< */
765 	ret = rte_eal_init(argc, argv);
766 	if (ret < 0)
767 		rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
768 	argc -= ret;
769 	argv += ret;
770 
771 	/* parse application arguments (after the EAL ones) */
772 	ret = l2fwd_parse_args(argc, argv);
773 	if (ret < 0)
774 		rte_exit(EXIT_FAILURE, "Invalid L2FWD arguments\n");
775 		/* >8 End of init EAL. */
776 
777 	rte_timer_subsystem_init();
778 
779 	/* fetch default timer frequency. */
780 	hz = rte_get_timer_hz();
781 
782 	/* Create the mbuf pool. 8< */
783 	l2fwd_pktmbuf_pool =
784 		rte_pktmbuf_pool_create("mbuf_pool", NB_MBUF, 32,
785 			0, RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
786 	if (l2fwd_pktmbuf_pool == NULL)
787 		rte_exit(EXIT_FAILURE, "Cannot init mbuf pool\n");
788 	/* >8 End of creation of mbuf pool. */
789 	nb_ports = rte_eth_dev_count_avail();
790 	if (nb_ports == 0)
791 		rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
792 
793 	/* Reset l2fwd_dst_ports. 8< */
794 	for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++)
795 		l2fwd_dst_ports[portid] = 0;
796 	last_port = 0;
797 
798 	/*
799 	 * Each logical core is assigned a dedicated TX queue on each port.
800 	 */
801 	RTE_ETH_FOREACH_DEV(portid) {
802 		/* skip ports that are not enabled */
803 		if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
804 			continue;
805 
806 		if (nb_ports_in_mask % 2) {
807 			l2fwd_dst_ports[portid] = last_port;
808 			l2fwd_dst_ports[last_port] = portid;
809 		} else
810 			last_port = portid;
811 
812 		nb_ports_in_mask++;
813 	}
814 	/* >8 End of reset l2fwd_dst_ports. */
815 	if (nb_ports_in_mask % 2) {
816 		printf("Notice: odd number of ports in portmask.\n");
817 		l2fwd_dst_ports[last_port] = last_port;
818 	}
819 
820 	rx_lcore_id = 0;
821 	qconf = NULL;
822 
823 	/* Initialize the port/queue configuration of each logical core */
824 	RTE_ETH_FOREACH_DEV(portid) {
825 		/* skip ports that are not enabled */
826 		if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
827 			continue;
828 
829 		/* get the lcore_id for this port */
830 		while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
831 		       lcore_queue_conf[rx_lcore_id].n_rx_port ==
832 		       l2fwd_rx_queue_per_lcore) {
833 			rx_lcore_id++;
834 			if (rx_lcore_id >= RTE_MAX_LCORE)
835 				rte_exit(EXIT_FAILURE, "Not enough cores\n");
836 		}
837 
838 		if (qconf != &lcore_queue_conf[rx_lcore_id])
839 			/* Assigned a new logical core in the loop above. */
840 			qconf = &lcore_queue_conf[rx_lcore_id];
841 
842 		qconf->rx_port_list[qconf->n_rx_port] = portid;
843 		qconf->n_rx_port++;
844 		printf("Lcore %u: RX port %u\n", rx_lcore_id, portid);
845 	}
846 
847 	/* Initialise each port */
848 	RTE_ETH_FOREACH_DEV(portid) {
849 		struct rte_eth_dev_info dev_info;
850 		struct rte_eth_rxconf rxq_conf;
851 		struct rte_eth_txconf txq_conf;
852 		struct rte_eth_conf local_port_conf = port_conf;
853 
854 		/* skip ports that are not enabled */
855 		if ((l2fwd_enabled_port_mask & (1 << portid)) == 0) {
856 			printf("Skipping disabled port %u\n", portid);
857 			continue;
858 		}
859 		nb_ports_available++;
860 
861 		/* init port */
862 		printf("Initializing port %u... ", portid);
863 		fflush(stdout);
864 
865 		ret = rte_eth_dev_info_get(portid, &dev_info);
866 		if (ret != 0)
867 			rte_exit(EXIT_FAILURE,
868 				"Error during getting device (port %u) info: %s\n",
869 				portid, strerror(-ret));
870 
871 		if (dev_info.tx_offload_capa & RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE)
872 			local_port_conf.txmode.offloads |=
873 				RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE;
874 		/* Configure the RX and TX queues. 8< */
875 		ret = rte_eth_dev_configure(portid, 1, 1, &local_port_conf);
876 		if (ret < 0)
877 			rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
878 				  ret, portid);
879 		/* >8 End of configuring the RX and TX queues. */
880 
881 		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
882 						       &nb_txd);
883 		if (ret < 0)
884 			rte_exit(EXIT_FAILURE,
885 				 "Cannot adjust number of descriptors: err=%d, port=%u\n",
886 				 ret, portid);
887 
888 		ret = rte_eth_macaddr_get(portid,
889 					  &l2fwd_ports_eth_addr[portid]);
890 		if (ret < 0)
891 			rte_exit(EXIT_FAILURE,
892 				 "Cannot get MAC address: err=%d, port=%u\n",
893 				 ret, portid);
894 
895 		/* init one RX queue */
896 		fflush(stdout);
897 		rxq_conf = dev_info.default_rxconf;
898 		rxq_conf.offloads = local_port_conf.rxmode.offloads;
899 		/* RX queue initialization. 8< */
900 		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
901 					     rte_eth_dev_socket_id(portid),
902 					     &rxq_conf,
903 					     l2fwd_pktmbuf_pool);
904 		if (ret < 0)
905 			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup:err=%d, port=%u\n",
906 				  ret, portid);
907 		/* >8 End of RX queue initialization. */
908 
909 		/* Init one TX queue on each port. 8< */
910 		txq_conf = dev_info.default_txconf;
911 		txq_conf.offloads = local_port_conf.txmode.offloads;
912 		fflush(stdout);
913 		ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
914 				rte_eth_dev_socket_id(portid),
915 				&txq_conf);
916 		if (ret < 0)
917 			rte_exit(EXIT_FAILURE,
918 			"rte_eth_tx_queue_setup:err=%d, port=%u\n",
919 				ret, portid);
920 		/* >8 End of init one TX queue on each port. */
921 
922 		/* Initialize TX buffers */
923 		tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
924 				RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
925 				rte_eth_dev_socket_id(portid));
926 		if (tx_buffer[portid] == NULL)
927 			rte_exit(EXIT_FAILURE, "Cannot allocate buffer for tx on port %u\n",
928 					portid);
929 
930 		rte_eth_tx_buffer_init(tx_buffer[portid], MAX_PKT_BURST);
931 
932 		ret = rte_eth_tx_buffer_set_err_callback(tx_buffer[portid],
933 				rte_eth_tx_buffer_count_callback,
934 				&port_statistics[portid].dropped);
935 		if (ret < 0)
936 			rte_exit(EXIT_FAILURE,
937 			"Cannot set error callback for tx buffer on port %u\n",
938 				 portid);
939 
940 		/* Start device */
941 		ret = rte_eth_dev_start(portid);
942 		if (ret < 0)
943 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start:err=%d, port=%u\n",
944 				  ret, portid);
945 
946 		printf("done:\n");
947 
948 		ret = rte_eth_promiscuous_enable(portid);
949 		if (ret != 0) {
950 			rte_exit(EXIT_FAILURE,
951 				 "rte_eth_promiscuous_enable:err=%s, port=%u\n",
952 				 rte_strerror(-ret), portid);
953 			return ret;
954 
955 		}
956 
957 		printf("Port %u, MAC address: " RTE_ETHER_ADDR_PRT_FMT "\n\n",
958 			portid,
959 			RTE_ETHER_ADDR_BYTES(&l2fwd_ports_eth_addr[portid]));
960 
961 		/* initialize port stats */
962 		memset(&port_statistics, 0, sizeof(port_statistics));
963 	}
964 
965 	if (!nb_ports_available) {
966 		rte_exit(EXIT_FAILURE,
967 			"All available ports are disabled. Please set portmask.\n");
968 	}
969 
970 	check_all_ports_link_status(l2fwd_enabled_port_mask);
971 
972 	drain_tsc = (hz + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
973 
974 	RTE_LCORE_FOREACH(lcore_id) {
975 		qconf = &lcore_queue_conf[lcore_id];
976 
977 		rte_spinlock_init(&qconf->lock);
978 
979 		if (rte_jobstats_context_init(&qconf->jobs_context) != 0)
980 			rte_panic("Jobs stats context for core %u init failed\n", lcore_id);
981 
982 		if (qconf->n_rx_port == 0) {
983 			RTE_LOG(INFO, L2FWD,
984 				"lcore %u: no ports so no jobs stats context initialization\n",
985 				lcore_id);
986 			continue;
987 		}
988 		/* Add flush job. 8< */
989 
990 		/* Set fixed period by setting min = max = initial period. Set target to
991 		 * zero as it is irrelevant for this job.
992 		 */
993 		rte_jobstats_init(&qconf->flush_job, "flush", drain_tsc, drain_tsc,
994 				drain_tsc, 0);
995 
996 		rte_timer_init(&qconf->flush_timer);
997 		ret = rte_timer_reset(&qconf->flush_timer, drain_tsc, PERIODICAL,
998 				lcore_id, &l2fwd_flush_job, NULL);
999 
1000 		if (ret < 0) {
1001 			rte_exit(1, "Failed to reset flush job timer for lcore %u: %s",
1002 					lcore_id, rte_strerror(-ret));
1003 		}
1004 		/* >8 End of add flush job. */
1005 
1006 		for (i = 0; i < qconf->n_rx_port; i++) {
1007 			struct rte_jobstats *job = &qconf->port_fwd_jobs[i];
1008 
1009 			portid = qconf->rx_port_list[i];
1010 			printf("Setting forward job for port %u\n", portid);
1011 
1012 			snprintf(name, RTE_DIM(name), "port %u fwd", portid);
1013 			/* Setup forward job. 8< */
1014 
1015 			/* Set min, max and initial period. Set target to MAX_PKT_BURST as
1016 			 * this is desired optimal RX/TX burst size.
1017 			 */
1018 			rte_jobstats_init(job, name, 0, drain_tsc, 0, MAX_PKT_BURST);
1019 			rte_jobstats_set_update_period_function(job, l2fwd_job_update_cb);
1020 
1021 			rte_timer_init(&qconf->rx_timers[i]);
1022 			ret = rte_timer_reset(&qconf->rx_timers[i], 0, PERIODICAL, lcore_id,
1023 					&l2fwd_fwd_job, (void *)(uintptr_t)i);
1024 
1025 			if (ret < 0) {
1026 				rte_exit(1, "Failed to reset lcore %u port %u job timer: %s",
1027 						lcore_id, qconf->rx_port_list[i], rte_strerror(-ret));
1028 			}
1029 			/* >8 End of forward job. */
1030 		}
1031 	}
1032 
1033 	if (timer_period)
1034 		rte_eal_alarm_set(timer_period * MS_PER_S, show_stats_cb, NULL);
1035 	else
1036 		RTE_LOG(INFO, L2FWD, "Stats display disabled\n");
1037 
1038 	/* launch per-lcore init on every lcore */
1039 	rte_eal_mp_remote_launch(l2fwd_launch_one_lcore, NULL, CALL_MAIN);
1040 	RTE_LCORE_FOREACH_WORKER(lcore_id) {
1041 		if (rte_eal_wait_lcore(lcore_id) < 0)
1042 			return -1;
1043 	}
1044 
1045 	/* clean up the EAL */
1046 	rte_eal_cleanup();
1047 
1048 	return 0;
1049 }
1050