xref: /dpdk/examples/l2fwd-jobstats/main.c (revision 92e68d9c97730ede6539d8f4c5748065aae96134)
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 RX_DESC_DEFAULT 1024
49 #define TX_DESC_DEFAULT 1024
50 static uint16_t nb_rxd = RX_DESC_DEFAULT;
51 static uint16_t nb_txd = 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 __rte_cache_aligned 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_ATOMIC(uint16_t) stats_read_pending;
84 	rte_spinlock_t lock;
85 };
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 	.txmode = {
93 		.mq_mode = RTE_ETH_MQ_TX_NONE,
94 	},
95 };
96 
97 struct rte_mempool *l2fwd_pktmbuf_pool = NULL;
98 
99 /* Per-port statistics struct */
100 struct __rte_cache_aligned l2fwd_port_statistics {
101 	uint64_t tx;
102 	uint64_t rx;
103 	uint64_t dropped;
104 };
105 struct l2fwd_port_statistics port_statistics[RTE_MAX_ETHPORTS];
106 
107 /* 1 day max */
108 #define MAX_TIMER_PERIOD 86400
109 /* default period is 10 seconds */
110 static int64_t timer_period = 10;
111 /* default timer frequency */
112 static double hz;
113 /* BURST_TX_DRAIN_US converted to cycles */
114 uint64_t drain_tsc;
115 /* Convert cycles to ns */
116 static inline double
cycles_to_ns(uint64_t cycles)117 cycles_to_ns(uint64_t cycles)
118 {
119 	double t = cycles;
120 
121 	t *= (double)NS_PER_S;
122 	t /= hz;
123 	return t;
124 }
125 
126 static void
show_lcore_stats(unsigned lcore_id)127 show_lcore_stats(unsigned lcore_id)
128 {
129 	struct lcore_queue_conf *qconf = &lcore_queue_conf[lcore_id];
130 	struct rte_jobstats_context *ctx = &qconf->jobs_context;
131 	struct rte_jobstats *job;
132 	uint8_t i;
133 
134 	/* LCore statistics. */
135 	uint64_t stats_period, loop_count;
136 	uint64_t exec, exec_min, exec_max;
137 	uint64_t management, management_min, management_max;
138 	uint64_t busy, busy_min, busy_max;
139 
140 	/* Jobs statistics. */
141 	const uint16_t port_cnt = qconf->n_rx_port;
142 	uint64_t jobs_exec_cnt[port_cnt], jobs_period[port_cnt];
143 	uint64_t jobs_exec[port_cnt], jobs_exec_min[port_cnt],
144 				jobs_exec_max[port_cnt];
145 
146 	uint64_t flush_exec_cnt, flush_period;
147 	uint64_t flush_exec, flush_exec_min, flush_exec_max;
148 
149 	uint64_t idle_exec_cnt;
150 	uint64_t idle_exec, idle_exec_min, idle_exec_max;
151 	uint64_t collection_time = rte_get_timer_cycles();
152 
153 	/* Ask forwarding thread to give us stats. */
154 	rte_atomic_store_explicit(&qconf->stats_read_pending, 1, rte_memory_order_relaxed);
155 	rte_spinlock_lock(&qconf->lock);
156 	rte_atomic_store_explicit(&qconf->stats_read_pending, 0, rte_memory_order_relaxed);
157 
158 	/* Collect context statistics. */
159 	stats_period = ctx->state_time - ctx->start_time;
160 	loop_count = ctx->loop_cnt;
161 
162 	exec = ctx->exec_time;
163 	exec_min = ctx->min_exec_time;
164 	exec_max = ctx->max_exec_time;
165 
166 	management = ctx->management_time;
167 	management_min = ctx->min_management_time;
168 	management_max = ctx->max_management_time;
169 
170 	rte_jobstats_context_reset(ctx);
171 
172 	for (i = 0; i < port_cnt; i++) {
173 		job = &qconf->port_fwd_jobs[i];
174 
175 		jobs_exec_cnt[i] = job->exec_cnt;
176 		jobs_period[i] = job->period;
177 
178 		jobs_exec[i] = job->exec_time;
179 		jobs_exec_min[i] = job->min_exec_time;
180 		jobs_exec_max[i] = job->max_exec_time;
181 
182 		rte_jobstats_reset(job);
183 	}
184 
185 	flush_exec_cnt = qconf->flush_job.exec_cnt;
186 	flush_period = qconf->flush_job.period;
187 	flush_exec = qconf->flush_job.exec_time;
188 	flush_exec_min = qconf->flush_job.min_exec_time;
189 	flush_exec_max = qconf->flush_job.max_exec_time;
190 	rte_jobstats_reset(&qconf->flush_job);
191 
192 	idle_exec_cnt = qconf->idle_job.exec_cnt;
193 	idle_exec = qconf->idle_job.exec_time;
194 	idle_exec_min = qconf->idle_job.min_exec_time;
195 	idle_exec_max = qconf->idle_job.max_exec_time;
196 	rte_jobstats_reset(&qconf->idle_job);
197 
198 	rte_spinlock_unlock(&qconf->lock);
199 
200 	exec -= idle_exec;
201 	busy = exec + management;
202 	busy_min = exec_min + management_min;
203 	busy_max = exec_max + management_max;
204 
205 
206 	collection_time = rte_get_timer_cycles() - collection_time;
207 
208 #define STAT_FMT "\n%-18s %'14.0f %6.1f%% %'10.0f %'10.0f %'10.0f"
209 
210 	printf("\n----------------"
211 			"\nLCore %3u: statistics (time in ns, collected in %'9.0f)"
212 			"\n%-18s %14s %7s %10s %10s %10s "
213 			"\n%-18s %'14.0f"
214 			"\n%-18s %'14" PRIu64
215 			STAT_FMT /* Exec */
216 			STAT_FMT /* Management */
217 			STAT_FMT /* Busy */
218 			STAT_FMT, /* Idle  */
219 			lcore_id, cycles_to_ns(collection_time),
220 			"Stat type", "total", "%total", "avg", "min", "max",
221 			"Stats duration:", cycles_to_ns(stats_period),
222 			"Loop count:", loop_count,
223 			"Exec time",
224 			cycles_to_ns(exec), exec * 100.0 / stats_period,
225 			cycles_to_ns(loop_count  ? exec / loop_count : 0),
226 			cycles_to_ns(exec_min),
227 			cycles_to_ns(exec_max),
228 			"Management time",
229 			cycles_to_ns(management), management * 100.0 / stats_period,
230 			cycles_to_ns(loop_count  ? management / loop_count : 0),
231 			cycles_to_ns(management_min),
232 			cycles_to_ns(management_max),
233 			"Exec + management",
234 			cycles_to_ns(busy),  busy * 100.0 / stats_period,
235 			cycles_to_ns(loop_count ? busy / loop_count : 0),
236 			cycles_to_ns(busy_min),
237 			cycles_to_ns(busy_max),
238 			"Idle (job)",
239 			cycles_to_ns(idle_exec), idle_exec * 100.0 / stats_period,
240 			cycles_to_ns(idle_exec_cnt ? idle_exec / idle_exec_cnt : 0),
241 			cycles_to_ns(idle_exec_min),
242 			cycles_to_ns(idle_exec_max));
243 
244 	for (i = 0; i < qconf->n_rx_port; i++) {
245 		job = &qconf->port_fwd_jobs[i];
246 		printf("\n\nJob %" PRIu32 ": %-20s "
247 				"\n%-18s %'14" PRIu64
248 				"\n%-18s %'14.0f"
249 				STAT_FMT,
250 				i, job->name,
251 				"Exec count:", jobs_exec_cnt[i],
252 				"Exec period: ", cycles_to_ns(jobs_period[i]),
253 				"Exec time",
254 				cycles_to_ns(jobs_exec[i]), jobs_exec[i] * 100.0 / stats_period,
255 				cycles_to_ns(jobs_exec_cnt[i] ? jobs_exec[i] / jobs_exec_cnt[i]
256 						: 0),
257 				cycles_to_ns(jobs_exec_min[i]),
258 				cycles_to_ns(jobs_exec_max[i]));
259 	}
260 
261 	if (qconf->n_rx_port > 0) {
262 		job = &qconf->flush_job;
263 		printf("\n\nJob %" PRIu32 ": %-20s "
264 				"\n%-18s %'14" PRIu64
265 				"\n%-18s %'14.0f"
266 				STAT_FMT,
267 				i, job->name,
268 				"Exec count:", flush_exec_cnt,
269 				"Exec period: ", cycles_to_ns(flush_period),
270 				"Exec time",
271 				cycles_to_ns(flush_exec), flush_exec * 100.0 / stats_period,
272 				cycles_to_ns(flush_exec_cnt ? flush_exec / flush_exec_cnt : 0),
273 				cycles_to_ns(flush_exec_min),
274 				cycles_to_ns(flush_exec_max));
275 	}
276 }
277 
278 /* Print out statistics on packets dropped */
279 static void
show_stats_cb(__rte_unused void * param)280 show_stats_cb(__rte_unused void *param)
281 {
282 	uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
283 	unsigned portid, lcore_id;
284 
285 	total_packets_dropped = 0;
286 	total_packets_tx = 0;
287 	total_packets_rx = 0;
288 
289 	const char clr[] = { 27, '[', '2', 'J', '\0' };
290 	const char topLeft[] = { 27, '[', '1', ';', '1', 'H', '\0' };
291 
292 	/* Clear screen and move to top left */
293 	printf("%s%s"
294 			"\nPort statistics ===================================",
295 			clr, topLeft);
296 
297 	for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
298 		/* skip disabled ports */
299 		if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
300 			continue;
301 		printf("\nStatistics for port %u ------------------------------"
302 				"\nPackets sent: %24"PRIu64
303 				"\nPackets received: %20"PRIu64
304 				"\nPackets dropped: %21"PRIu64,
305 				portid,
306 				port_statistics[portid].tx,
307 				port_statistics[portid].rx,
308 				port_statistics[portid].dropped);
309 
310 		total_packets_dropped += port_statistics[portid].dropped;
311 		total_packets_tx += port_statistics[portid].tx;
312 		total_packets_rx += port_statistics[portid].rx;
313 	}
314 
315 	printf("\nAggregate statistics ==============================="
316 			"\nTotal packets sent: %18"PRIu64
317 			"\nTotal packets received: %14"PRIu64
318 			"\nTotal packets dropped: %15"PRIu64
319 			"\n====================================================",
320 			total_packets_tx,
321 			total_packets_rx,
322 			total_packets_dropped);
323 
324 	RTE_LCORE_FOREACH(lcore_id) {
325 		if (lcore_queue_conf[lcore_id].n_rx_port > 0)
326 			show_lcore_stats(lcore_id);
327 	}
328 
329 	printf("\n====================================================\n");
330 
331 	fflush(stdout);
332 
333 	rte_eal_alarm_set(timer_period * US_PER_S, show_stats_cb, NULL);
334 }
335 
336 /* Start of l2fwd_simple_forward. 8< */
337 static void
l2fwd_simple_forward(struct rte_mbuf * m,unsigned portid)338 l2fwd_simple_forward(struct rte_mbuf *m, unsigned portid)
339 {
340 	struct rte_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 rte_ether_hdr *);
348 
349 	/* 02:00:00:00:00:xx */
350 	tmp = &eth->dst_addr.addr_bytes[0];
351 	*((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
352 
353 	/* src addr */
354 	rte_ether_addr_copy(&l2fwd_ports_eth_addr[dst_port], &eth->src_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 /* >8 End of l2fwd_simple_forward. */
362 
363 static void
l2fwd_job_update_cb(struct rte_jobstats * job,int64_t result)364 l2fwd_job_update_cb(struct rte_jobstats *job, int64_t result)
365 {
366 	int64_t err = job->target - result;
367 	int64_t histeresis = job->target / 8;
368 
369 	if (err < -histeresis) {
370 		if (job->min_period + UPDATE_STEP_DOWN < job->period)
371 			job->period -= UPDATE_STEP_DOWN;
372 	} else if (err > histeresis) {
373 		if (job->period + UPDATE_STEP_UP < job->max_period)
374 			job->period += UPDATE_STEP_UP;
375 	}
376 }
377 
378 static void
l2fwd_fwd_job(__rte_unused struct rte_timer * timer,void * arg)379 l2fwd_fwd_job(__rte_unused struct rte_timer *timer, void *arg)
380 {
381 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
382 	struct rte_mbuf *m;
383 
384 	const uint16_t port_idx = (uintptr_t) arg;
385 	const unsigned lcore_id = rte_lcore_id();
386 	struct lcore_queue_conf *qconf = &lcore_queue_conf[lcore_id];
387 	struct rte_jobstats *job = &qconf->port_fwd_jobs[port_idx];
388 	const uint16_t portid = qconf->rx_port_list[port_idx];
389 
390 	uint8_t j;
391 	uint16_t total_nb_rx;
392 
393 	rte_jobstats_start(&qconf->jobs_context, job);
394 
395 	/* Call rx burst 2 times. This allow rte_jobstats logic to see if this
396 	 * function must be called more frequently. */
397 
398 	/* Call rx burst 2 times. 8< */
399 	total_nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
400 			MAX_PKT_BURST);
401 
402 	for (j = 0; j < total_nb_rx; j++) {
403 		m = pkts_burst[j];
404 		rte_prefetch0(rte_pktmbuf_mtod(m, void *));
405 		l2fwd_simple_forward(m, portid);
406 	}
407 	/* >8 End of call rx burst 2 times. */
408 
409 	/* Read second try. 8< */
410 	if (total_nb_rx == MAX_PKT_BURST) {
411 		const uint16_t nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
412 				MAX_PKT_BURST);
413 
414 		total_nb_rx += nb_rx;
415 		for (j = 0; j < nb_rx; j++) {
416 			m = pkts_burst[j];
417 			rte_prefetch0(rte_pktmbuf_mtod(m, void *));
418 			l2fwd_simple_forward(m, portid);
419 		}
420 	}
421 	/* >8 End of read second try. */
422 
423 	port_statistics[portid].rx += total_nb_rx;
424 
425 	/* Adjust period time in which we are running here. 8< */
426 	if (rte_jobstats_finish(job, total_nb_rx) != 0) {
427 		rte_timer_reset(&qconf->rx_timers[port_idx], job->period, PERIODICAL,
428 				lcore_id, l2fwd_fwd_job, arg);
429 	}
430 	/* >8 End of adjust period time in which we are running. */
431 }
432 
433 /* Draining TX queue of each port. 8< */
434 static void
l2fwd_flush_job(__rte_unused struct rte_timer * timer,__rte_unused void * arg)435 l2fwd_flush_job(__rte_unused struct rte_timer *timer, __rte_unused void *arg)
436 {
437 	uint64_t now;
438 	unsigned lcore_id;
439 	struct lcore_queue_conf *qconf;
440 	uint16_t portid;
441 	unsigned i;
442 	uint32_t sent;
443 	struct rte_eth_dev_tx_buffer *buffer;
444 
445 	lcore_id = rte_lcore_id();
446 	qconf = &lcore_queue_conf[lcore_id];
447 
448 	rte_jobstats_start(&qconf->jobs_context, &qconf->flush_job);
449 
450 	now = rte_get_timer_cycles();
451 	lcore_id = rte_lcore_id();
452 	qconf = &lcore_queue_conf[lcore_id];
453 
454 	for (i = 0; i < qconf->n_rx_port; i++) {
455 		portid = l2fwd_dst_ports[qconf->rx_port_list[i]];
456 
457 		if (qconf->next_flush_time[portid] <= now)
458 			continue;
459 
460 		buffer = tx_buffer[portid];
461 		sent = rte_eth_tx_buffer_flush(portid, 0, buffer);
462 		if (sent)
463 			port_statistics[portid].tx += sent;
464 
465 		qconf->next_flush_time[portid] = rte_get_timer_cycles() + drain_tsc;
466 	}
467 
468 	/* Pass target to indicate that this job is happy of time interval
469 	 * in which it was called. */
470 	rte_jobstats_finish(&qconf->flush_job, qconf->flush_job.target);
471 }
472 /* >8 End of draining TX queue of each port. */
473 
474 /* main processing loop */
475 static void
l2fwd_main_loop(void)476 l2fwd_main_loop(void)
477 {
478 	unsigned lcore_id;
479 	unsigned i, portid;
480 	struct lcore_queue_conf *qconf;
481 	uint8_t stats_read_pending = 0;
482 	uint8_t need_manage;
483 
484 	lcore_id = rte_lcore_id();
485 	qconf = &lcore_queue_conf[lcore_id];
486 
487 	if (qconf->n_rx_port == 0) {
488 		RTE_LOG(INFO, L2FWD, "lcore %u has nothing to do\n", lcore_id);
489 		return;
490 	}
491 
492 	RTE_LOG(INFO, L2FWD, "entering main loop on lcore %u\n", lcore_id);
493 
494 	for (i = 0; i < qconf->n_rx_port; i++) {
495 
496 		portid = qconf->rx_port_list[i];
497 		RTE_LOG(INFO, L2FWD, " -- lcoreid=%u portid=%u\n", lcore_id,
498 			portid);
499 	}
500 
501 	rte_jobstats_init(&qconf->idle_job, "idle", 0, 0, 0, 0);
502 
503 	/* Minimize impact of stats reading. 8< */
504 	for (;;) {
505 		rte_spinlock_lock(&qconf->lock);
506 
507 		do {
508 			rte_jobstats_context_start(&qconf->jobs_context);
509 
510 			/* Do the Idle job:
511 			 * - Read stats_read_pending flag
512 			 * - check if some real job need to be executed
513 			 */
514 			rte_jobstats_start(&qconf->jobs_context, &qconf->idle_job);
515 
516 			uint64_t repeats = 0;
517 
518 			do {
519 				uint8_t i;
520 				uint64_t now = rte_get_timer_cycles();
521 
522 				repeats++;
523 				need_manage = qconf->flush_timer.expire < now;
524 				/* Check if we was esked to give a stats. */
525 				stats_read_pending = rte_atomic_load_explicit(
526 					&qconf->stats_read_pending,
527 					rte_memory_order_relaxed);
528 				need_manage |= stats_read_pending;
529 
530 				for (i = 0; i < qconf->n_rx_port && !need_manage; i++)
531 					need_manage = qconf->rx_timers[i].expire < now;
532 
533 			} while (!need_manage);
534 
535 			if (likely(repeats != 1))
536 				rte_jobstats_finish(&qconf->idle_job, qconf->idle_job.target);
537 			else
538 				rte_jobstats_abort(&qconf->idle_job);
539 
540 			rte_timer_manage();
541 			rte_jobstats_context_finish(&qconf->jobs_context);
542 		} while (likely(stats_read_pending == 0));
543 
544 		rte_spinlock_unlock(&qconf->lock);
545 		rte_pause();
546 	}
547 	/* >8 End of minimize impact of stats reading. */
548 }
549 
550 static int
l2fwd_launch_one_lcore(__rte_unused void * dummy)551 l2fwd_launch_one_lcore(__rte_unused void *dummy)
552 {
553 	l2fwd_main_loop();
554 	return 0;
555 }
556 
557 /* display usage */
558 static void
l2fwd_usage(const char * prgname)559 l2fwd_usage(const char *prgname)
560 {
561 	printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
562 	       "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
563 	       "  -q NQ: number of queue (=ports) per lcore (default is 1)\n"
564 		   "  -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n"
565 		   "  -l set system default locale instead of default (\"C\" locale) for thousands separator in stats.",
566 	       prgname);
567 }
568 
569 static int
l2fwd_parse_portmask(const char * portmask)570 l2fwd_parse_portmask(const char *portmask)
571 {
572 	char *end = NULL;
573 	unsigned long pm;
574 
575 	/* parse hexadecimal string */
576 	pm = strtoul(portmask, &end, 16);
577 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
578 		return 0;
579 
580 	return pm;
581 }
582 
583 static unsigned int
l2fwd_parse_nqueue(const char * q_arg)584 l2fwd_parse_nqueue(const char *q_arg)
585 {
586 	char *end = NULL;
587 	unsigned long n;
588 
589 	/* parse hexadecimal string */
590 	n = strtoul(q_arg, &end, 10);
591 	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
592 		return 0;
593 	if (n == 0)
594 		return 0;
595 	if (n >= MAX_RX_QUEUE_PER_LCORE)
596 		return 0;
597 
598 	return n;
599 }
600 
601 static int
l2fwd_parse_timer_period(const char * q_arg)602 l2fwd_parse_timer_period(const char *q_arg)
603 {
604 	char *end = NULL;
605 	int n;
606 
607 	/* parse number string */
608 	n = strtol(q_arg, &end, 10);
609 	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
610 		return -1;
611 	if (n >= MAX_TIMER_PERIOD)
612 		return -1;
613 
614 	return n;
615 }
616 
617 /* Parse the argument given in the command line of the application */
618 static int
l2fwd_parse_args(int argc,char ** argv)619 l2fwd_parse_args(int argc, char **argv)
620 {
621 	int opt, ret;
622 	char **argvopt;
623 	int option_index;
624 	char *prgname = argv[0];
625 	static struct option lgopts[] = {
626 		{NULL, 0, 0, 0}
627 	};
628 
629 	argvopt = argv;
630 
631 	while ((opt = getopt_long(argc, argvopt, "p:q:T:l",
632 				  lgopts, &option_index)) != EOF) {
633 
634 		switch (opt) {
635 		/* portmask */
636 		case 'p':
637 			l2fwd_enabled_port_mask = l2fwd_parse_portmask(optarg);
638 			if (l2fwd_enabled_port_mask == 0) {
639 				printf("invalid portmask\n");
640 				l2fwd_usage(prgname);
641 				return -1;
642 			}
643 			break;
644 
645 		/* nqueue */
646 		case 'q':
647 			l2fwd_rx_queue_per_lcore = l2fwd_parse_nqueue(optarg);
648 			if (l2fwd_rx_queue_per_lcore == 0) {
649 				printf("invalid queue number\n");
650 				l2fwd_usage(prgname);
651 				return -1;
652 			}
653 			break;
654 
655 		/* timer period */
656 		case 'T':
657 			timer_period = l2fwd_parse_timer_period(optarg);
658 			if (timer_period < 0) {
659 				printf("invalid timer period\n");
660 				l2fwd_usage(prgname);
661 				return -1;
662 			}
663 			break;
664 
665 		/* For thousands separator in printf. */
666 		case 'l':
667 			setlocale(LC_ALL, "");
668 			break;
669 
670 		/* long options */
671 		case 0:
672 			l2fwd_usage(prgname);
673 			return -1;
674 
675 		default:
676 			l2fwd_usage(prgname);
677 			return -1;
678 		}
679 	}
680 
681 	if (optind >= 0)
682 		argv[optind-1] = prgname;
683 
684 	ret = optind-1;
685 	optind = 1; /* reset getopt lib */
686 	return ret;
687 }
688 
689 /* Check the link status of all ports in up to 9s, and print them finally */
690 static void
check_all_ports_link_status(uint32_t port_mask)691 check_all_ports_link_status(uint32_t port_mask)
692 {
693 #define CHECK_INTERVAL 100 /* 100ms */
694 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
695 	uint16_t portid;
696 	uint8_t count, all_ports_up, print_flag = 0;
697 	struct rte_eth_link link;
698 	int ret;
699 	char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
700 
701 	printf("\nChecking link status");
702 	fflush(stdout);
703 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
704 		all_ports_up = 1;
705 		RTE_ETH_FOREACH_DEV(portid) {
706 			if ((port_mask & (1 << portid)) == 0)
707 				continue;
708 			memset(&link, 0, sizeof(link));
709 			ret = rte_eth_link_get_nowait(portid, &link);
710 			if (ret < 0) {
711 				all_ports_up = 0;
712 				if (print_flag == 1)
713 					printf("Port %u link get failed: %s\n",
714 						portid, rte_strerror(-ret));
715 				continue;
716 			}
717 			/* print link status if flag set */
718 			if (print_flag == 1) {
719 				rte_eth_link_to_str(link_status_text,
720 					sizeof(link_status_text), &link);
721 				printf("Port %d %s\n", portid,
722 				       link_status_text);
723 				continue;
724 			}
725 			/* clear all_ports_up flag if any link down */
726 			if (link.link_status == RTE_ETH_LINK_DOWN) {
727 				all_ports_up = 0;
728 				break;
729 			}
730 		}
731 		/* after finally printing all link status, get out */
732 		if (print_flag == 1)
733 			break;
734 
735 		if (all_ports_up == 0) {
736 			printf(".");
737 			fflush(stdout);
738 			rte_delay_ms(CHECK_INTERVAL);
739 		}
740 
741 		/* set the print_flag if all ports up or timeout */
742 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
743 			print_flag = 1;
744 			printf("done\n");
745 		}
746 	}
747 }
748 
749 int
main(int argc,char ** argv)750 main(int argc, char **argv)
751 {
752 	struct lcore_queue_conf *qconf;
753 	unsigned lcore_id, rx_lcore_id;
754 	unsigned nb_ports_in_mask = 0;
755 	int ret;
756 	char name[RTE_JOBSTATS_NAMESIZE];
757 	uint16_t nb_ports;
758 	uint16_t nb_ports_available = 0;
759 	uint16_t portid, last_port;
760 	uint8_t i;
761 
762 	/* Init EAL. 8< */
763 	ret = rte_eal_init(argc, argv);
764 	if (ret < 0)
765 		rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
766 	argc -= ret;
767 	argv += ret;
768 
769 	/* parse application arguments (after the EAL ones) */
770 	ret = l2fwd_parse_args(argc, argv);
771 	if (ret < 0)
772 		rte_exit(EXIT_FAILURE, "Invalid L2FWD arguments\n");
773 		/* >8 End of init EAL. */
774 
775 	rte_timer_subsystem_init();
776 
777 	/* fetch default timer frequency. */
778 	hz = rte_get_timer_hz();
779 
780 	/* Create the mbuf pool. 8< */
781 	l2fwd_pktmbuf_pool =
782 		rte_pktmbuf_pool_create("mbuf_pool", NB_MBUF, 32,
783 			0, RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
784 	if (l2fwd_pktmbuf_pool == NULL)
785 		rte_exit(EXIT_FAILURE, "Cannot init mbuf pool\n");
786 	/* >8 End of creation of mbuf pool. */
787 	nb_ports = rte_eth_dev_count_avail();
788 	if (nb_ports == 0)
789 		rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
790 
791 	/* Reset l2fwd_dst_ports. 8< */
792 	for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++)
793 		l2fwd_dst_ports[portid] = 0;
794 	last_port = 0;
795 
796 	/*
797 	 * Each logical core is assigned a dedicated TX queue on each port.
798 	 */
799 	RTE_ETH_FOREACH_DEV(portid) {
800 		/* skip ports that are not enabled */
801 		if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
802 			continue;
803 
804 		if (nb_ports_in_mask % 2) {
805 			l2fwd_dst_ports[portid] = last_port;
806 			l2fwd_dst_ports[last_port] = portid;
807 		} else
808 			last_port = portid;
809 
810 		nb_ports_in_mask++;
811 	}
812 	/* >8 End of reset l2fwd_dst_ports. */
813 	if (nb_ports_in_mask % 2) {
814 		printf("Notice: odd number of ports in portmask.\n");
815 		l2fwd_dst_ports[last_port] = last_port;
816 	}
817 
818 	rx_lcore_id = 0;
819 	qconf = NULL;
820 
821 	/* Initialize the port/queue configuration of each logical core */
822 	RTE_ETH_FOREACH_DEV(portid) {
823 		/* skip ports that are not enabled */
824 		if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
825 			continue;
826 
827 		/* get the lcore_id for this port */
828 		while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
829 		       lcore_queue_conf[rx_lcore_id].n_rx_port ==
830 		       l2fwd_rx_queue_per_lcore) {
831 			rx_lcore_id++;
832 			if (rx_lcore_id >= RTE_MAX_LCORE)
833 				rte_exit(EXIT_FAILURE, "Not enough cores\n");
834 		}
835 
836 		if (qconf != &lcore_queue_conf[rx_lcore_id])
837 			/* Assigned a new logical core in the loop above. */
838 			qconf = &lcore_queue_conf[rx_lcore_id];
839 
840 		qconf->rx_port_list[qconf->n_rx_port] = portid;
841 		qconf->n_rx_port++;
842 		printf("Lcore %u: RX port %u\n", rx_lcore_id, portid);
843 	}
844 
845 	/* Initialise each port */
846 	RTE_ETH_FOREACH_DEV(portid) {
847 		struct rte_eth_dev_info dev_info;
848 		struct rte_eth_rxconf rxq_conf;
849 		struct rte_eth_txconf txq_conf;
850 		struct rte_eth_conf local_port_conf = port_conf;
851 
852 		/* skip ports that are not enabled */
853 		if ((l2fwd_enabled_port_mask & (1 << portid)) == 0) {
854 			printf("Skipping disabled port %u\n", portid);
855 			continue;
856 		}
857 		nb_ports_available++;
858 
859 		/* init port */
860 		printf("Initializing port %u... ", portid);
861 		fflush(stdout);
862 
863 		ret = rte_eth_dev_info_get(portid, &dev_info);
864 		if (ret != 0)
865 			rte_exit(EXIT_FAILURE,
866 				"Error during getting device (port %u) info: %s\n",
867 				portid, strerror(-ret));
868 
869 		if (dev_info.tx_offload_capa & RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE)
870 			local_port_conf.txmode.offloads |=
871 				RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE;
872 		/* Configure the RX and TX queues. 8< */
873 		ret = rte_eth_dev_configure(portid, 1, 1, &local_port_conf);
874 		if (ret < 0)
875 			rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
876 				  ret, portid);
877 		/* >8 End of configuring the RX and TX queues. */
878 
879 		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
880 						       &nb_txd);
881 		if (ret < 0)
882 			rte_exit(EXIT_FAILURE,
883 				 "Cannot adjust number of descriptors: err=%d, port=%u\n",
884 				 ret, portid);
885 
886 		ret = rte_eth_macaddr_get(portid,
887 					  &l2fwd_ports_eth_addr[portid]);
888 		if (ret < 0)
889 			rte_exit(EXIT_FAILURE,
890 				 "Cannot get MAC address: err=%d, port=%u\n",
891 				 ret, portid);
892 
893 		/* init one RX queue */
894 		fflush(stdout);
895 		rxq_conf = dev_info.default_rxconf;
896 		rxq_conf.offloads = local_port_conf.rxmode.offloads;
897 		/* RX queue initialization. 8< */
898 		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
899 					     rte_eth_dev_socket_id(portid),
900 					     &rxq_conf,
901 					     l2fwd_pktmbuf_pool);
902 		if (ret < 0)
903 			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup:err=%d, port=%u\n",
904 				  ret, portid);
905 		/* >8 End of RX queue initialization. */
906 
907 		/* Init one TX queue on each port. 8< */
908 		txq_conf = dev_info.default_txconf;
909 		txq_conf.offloads = local_port_conf.txmode.offloads;
910 		fflush(stdout);
911 		ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
912 				rte_eth_dev_socket_id(portid),
913 				&txq_conf);
914 		if (ret < 0)
915 			rte_exit(EXIT_FAILURE,
916 			"rte_eth_tx_queue_setup:err=%d, port=%u\n",
917 				ret, portid);
918 		/* >8 End of init one TX queue on each port. */
919 
920 		/* Initialize TX buffers */
921 		tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
922 				RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
923 				rte_eth_dev_socket_id(portid));
924 		if (tx_buffer[portid] == NULL)
925 			rte_exit(EXIT_FAILURE, "Cannot allocate buffer for tx on port %u\n",
926 					portid);
927 
928 		rte_eth_tx_buffer_init(tx_buffer[portid], MAX_PKT_BURST);
929 
930 		ret = rte_eth_tx_buffer_set_err_callback(tx_buffer[portid],
931 				rte_eth_tx_buffer_count_callback,
932 				&port_statistics[portid].dropped);
933 		if (ret < 0)
934 			rte_exit(EXIT_FAILURE,
935 			"Cannot set error callback for tx buffer on port %u\n",
936 				 portid);
937 
938 		/* Start device */
939 		ret = rte_eth_dev_start(portid);
940 		if (ret < 0)
941 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start:err=%d, port=%u\n",
942 				  ret, portid);
943 
944 		printf("done:\n");
945 
946 		ret = rte_eth_promiscuous_enable(portid);
947 		if (ret != 0) {
948 			rte_exit(EXIT_FAILURE,
949 				 "rte_eth_promiscuous_enable:err=%s, port=%u\n",
950 				 rte_strerror(-ret), portid);
951 			return ret;
952 
953 		}
954 
955 		printf("Port %u, MAC address: " RTE_ETHER_ADDR_PRT_FMT "\n\n",
956 			portid,
957 			RTE_ETHER_ADDR_BYTES(&l2fwd_ports_eth_addr[portid]));
958 
959 		/* initialize port stats */
960 		memset(&port_statistics, 0, sizeof(port_statistics));
961 	}
962 
963 	if (!nb_ports_available) {
964 		rte_exit(EXIT_FAILURE,
965 			"All available ports are disabled. Please set portmask.\n");
966 	}
967 
968 	check_all_ports_link_status(l2fwd_enabled_port_mask);
969 
970 	drain_tsc = (hz + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
971 
972 	RTE_LCORE_FOREACH(lcore_id) {
973 		qconf = &lcore_queue_conf[lcore_id];
974 
975 		rte_spinlock_init(&qconf->lock);
976 
977 		if (rte_jobstats_context_init(&qconf->jobs_context) != 0)
978 			rte_panic("Jobs stats context for core %u init failed\n", lcore_id);
979 
980 		if (qconf->n_rx_port == 0) {
981 			RTE_LOG(INFO, L2FWD,
982 				"lcore %u: no ports so no jobs stats context initialization\n",
983 				lcore_id);
984 			continue;
985 		}
986 		/* Add flush job. 8< */
987 
988 		/* Set fixed period by setting min = max = initial period. Set target to
989 		 * zero as it is irrelevant for this job.
990 		 */
991 		rte_jobstats_init(&qconf->flush_job, "flush", drain_tsc, drain_tsc,
992 				drain_tsc, 0);
993 
994 		rte_timer_init(&qconf->flush_timer);
995 		ret = rte_timer_reset(&qconf->flush_timer, drain_tsc, PERIODICAL,
996 				lcore_id, &l2fwd_flush_job, NULL);
997 
998 		if (ret < 0) {
999 			rte_exit(1, "Failed to reset flush job timer for lcore %u: %s",
1000 					lcore_id, rte_strerror(-ret));
1001 		}
1002 		/* >8 End of add flush job. */
1003 
1004 		for (i = 0; i < qconf->n_rx_port; i++) {
1005 			struct rte_jobstats *job = &qconf->port_fwd_jobs[i];
1006 
1007 			portid = qconf->rx_port_list[i];
1008 			printf("Setting forward job for port %u\n", portid);
1009 
1010 			snprintf(name, RTE_DIM(name), "port %u fwd", portid);
1011 			/* Setup forward job. 8< */
1012 
1013 			/* Set min, max and initial period. Set target to MAX_PKT_BURST as
1014 			 * this is desired optimal RX/TX burst size.
1015 			 */
1016 			rte_jobstats_init(job, name, 0, drain_tsc, 0, MAX_PKT_BURST);
1017 			rte_jobstats_set_update_period_function(job, l2fwd_job_update_cb);
1018 
1019 			rte_timer_init(&qconf->rx_timers[i]);
1020 			ret = rte_timer_reset(&qconf->rx_timers[i], 0, PERIODICAL, lcore_id,
1021 					&l2fwd_fwd_job, (void *)(uintptr_t)i);
1022 
1023 			if (ret < 0) {
1024 				rte_exit(1, "Failed to reset lcore %u port %u job timer: %s",
1025 						lcore_id, qconf->rx_port_list[i], rte_strerror(-ret));
1026 			}
1027 			/* >8 End of forward job. */
1028 		}
1029 	}
1030 
1031 	if (timer_period)
1032 		rte_eal_alarm_set(timer_period * MS_PER_S, show_stats_cb, NULL);
1033 	else
1034 		RTE_LOG(INFO, L2FWD, "Stats display disabled\n");
1035 
1036 	/* launch per-lcore init on every lcore */
1037 	rte_eal_mp_remote_launch(l2fwd_launch_one_lcore, NULL, CALL_MAIN);
1038 	RTE_LCORE_FOREACH_WORKER(lcore_id) {
1039 		if (rte_eal_wait_lcore(lcore_id) < 0)
1040 			return -1;
1041 	}
1042 
1043 	/* clean up the EAL */
1044 	rte_eal_cleanup();
1045 
1046 	return 0;
1047 }
1048