xref: /dpdk/examples/packet_ordering/main.c (revision 6491dbbecebb1e4f07fc970ef90b34119d8be2e3)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2016 Intel Corporation
3  */
4 
5 #include <signal.h>
6 #include <getopt.h>
7 
8 #include <rte_eal.h>
9 #include <rte_common.h>
10 #include <rte_errno.h>
11 #include <rte_ethdev.h>
12 #include <rte_lcore.h>
13 #include <rte_malloc.h>
14 #include <rte_mbuf.h>
15 #include <rte_mempool.h>
16 #include <rte_ring.h>
17 #include <rte_reorder.h>
18 
19 #define RX_DESC_PER_QUEUE 1024
20 #define TX_DESC_PER_QUEUE 1024
21 
22 #define MAX_PKTS_BURST 32
23 #define REORDER_BUFFER_SIZE 8192
24 #define MBUF_PER_POOL 65535
25 #define MBUF_POOL_CACHE_SIZE 250
26 
27 #define RING_SIZE 16384
28 
29 /* Macros for printing using RTE_LOG */
30 #define RTE_LOGTYPE_REORDERAPP          RTE_LOGTYPE_USER1
31 
32 unsigned int portmask;
33 unsigned int disable_reorder;
34 volatile uint8_t quit_signal;
35 
36 static struct rte_mempool *mbuf_pool;
37 
38 static struct rte_eth_conf port_conf_default = {
39 	.rxmode = {
40 		.ignore_offload_bitfield = 1,
41 	},
42 };
43 
44 struct worker_thread_args {
45 	struct rte_ring *ring_in;
46 	struct rte_ring *ring_out;
47 };
48 
49 struct send_thread_args {
50 	struct rte_ring *ring_in;
51 	struct rte_reorder_buffer *buffer;
52 };
53 
54 volatile struct app_stats {
55 	struct {
56 		uint64_t rx_pkts;
57 		uint64_t enqueue_pkts;
58 		uint64_t enqueue_failed_pkts;
59 	} rx __rte_cache_aligned;
60 
61 	struct {
62 		uint64_t dequeue_pkts;
63 		uint64_t enqueue_pkts;
64 		uint64_t enqueue_failed_pkts;
65 	} wkr __rte_cache_aligned;
66 
67 	struct {
68 		uint64_t dequeue_pkts;
69 		/* Too early pkts transmitted directly w/o reordering */
70 		uint64_t early_pkts_txtd_woro;
71 		/* Too early pkts failed from direct transmit */
72 		uint64_t early_pkts_tx_failed_woro;
73 		uint64_t ro_tx_pkts;
74 		uint64_t ro_tx_failed_pkts;
75 	} tx __rte_cache_aligned;
76 } app_stats;
77 
78 /**
79  * Get the last enabled lcore ID
80  *
81  * @return
82  *   The last enabled lcore ID.
83  */
84 static unsigned int
85 get_last_lcore_id(void)
86 {
87 	int i;
88 
89 	for (i = RTE_MAX_LCORE - 1; i >= 0; i--)
90 		if (rte_lcore_is_enabled(i))
91 			return i;
92 	return 0;
93 }
94 
95 /**
96  * Get the previous enabled lcore ID
97  * @param id
98  *  The current lcore ID
99  * @return
100  *   The previous enabled lcore ID or the current lcore
101  *   ID if it is the first available core.
102  */
103 static unsigned int
104 get_previous_lcore_id(unsigned int id)
105 {
106 	int i;
107 
108 	for (i = id - 1; i >= 0; i--)
109 		if (rte_lcore_is_enabled(i))
110 			return i;
111 	return id;
112 }
113 
114 static inline void
115 pktmbuf_free_bulk(struct rte_mbuf *mbuf_table[], unsigned n)
116 {
117 	unsigned int i;
118 
119 	for (i = 0; i < n; i++)
120 		rte_pktmbuf_free(mbuf_table[i]);
121 }
122 
123 /* display usage */
124 static void
125 print_usage(const char *prgname)
126 {
127 	printf("%s [EAL options] -- -p PORTMASK\n"
128 			"  -p PORTMASK: hexadecimal bitmask of ports to configure\n",
129 			prgname);
130 }
131 
132 static int
133 parse_portmask(const char *portmask)
134 {
135 	unsigned long pm;
136 	char *end = NULL;
137 
138 	/* parse hexadecimal string */
139 	pm = strtoul(portmask, &end, 16);
140 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
141 		return -1;
142 
143 	if (pm == 0)
144 		return -1;
145 
146 	return pm;
147 }
148 
149 /* Parse the argument given in the command line of the application */
150 static int
151 parse_args(int argc, char **argv)
152 {
153 	int opt;
154 	int option_index;
155 	char **argvopt;
156 	char *prgname = argv[0];
157 	static struct option lgopts[] = {
158 		{"disable-reorder", 0, 0, 0},
159 		{NULL, 0, 0, 0}
160 	};
161 
162 	argvopt = argv;
163 
164 	while ((opt = getopt_long(argc, argvopt, "p:",
165 					lgopts, &option_index)) != EOF) {
166 		switch (opt) {
167 		/* portmask */
168 		case 'p':
169 			portmask = parse_portmask(optarg);
170 			if (portmask == 0) {
171 				printf("invalid portmask\n");
172 				print_usage(prgname);
173 				return -1;
174 			}
175 			break;
176 		/* long options */
177 		case 0:
178 			if (!strcmp(lgopts[option_index].name, "disable-reorder")) {
179 				printf("reorder disabled\n");
180 				disable_reorder = 1;
181 			}
182 			break;
183 		default:
184 			print_usage(prgname);
185 			return -1;
186 		}
187 	}
188 	if (optind <= 1) {
189 		print_usage(prgname);
190 		return -1;
191 	}
192 
193 	argv[optind-1] = prgname;
194 	optind = 1; /* reset getopt lib */
195 	return 0;
196 }
197 
198 /*
199  * Tx buffer error callback
200  */
201 static void
202 flush_tx_error_callback(struct rte_mbuf **unsent, uint16_t count,
203 		void *userdata __rte_unused) {
204 
205 	/* free the mbufs which failed from transmit */
206 	app_stats.tx.ro_tx_failed_pkts += count;
207 	RTE_LOG_DP(DEBUG, REORDERAPP, "%s:Packet loss with tx_burst\n", __func__);
208 	pktmbuf_free_bulk(unsent, count);
209 
210 }
211 
212 static inline int
213 free_tx_buffers(struct rte_eth_dev_tx_buffer *tx_buffer[]) {
214 	uint16_t port_id;
215 
216 	/* initialize buffers for all ports */
217 	RTE_ETH_FOREACH_DEV(port_id) {
218 		/* skip ports that are not enabled */
219 		if ((portmask & (1 << port_id)) == 0)
220 			continue;
221 
222 		rte_free(tx_buffer[port_id]);
223 	}
224 	return 0;
225 }
226 
227 static inline int
228 configure_tx_buffers(struct rte_eth_dev_tx_buffer *tx_buffer[])
229 {
230 	uint16_t port_id;
231 	int ret;
232 
233 	/* initialize buffers for all ports */
234 	RTE_ETH_FOREACH_DEV(port_id) {
235 		/* skip ports that are not enabled */
236 		if ((portmask & (1 << port_id)) == 0)
237 			continue;
238 
239 		/* Initialize TX buffers */
240 		tx_buffer[port_id] = rte_zmalloc_socket("tx_buffer",
241 				RTE_ETH_TX_BUFFER_SIZE(MAX_PKTS_BURST), 0,
242 				rte_eth_dev_socket_id(port_id));
243 		if (tx_buffer[port_id] == NULL)
244 			rte_exit(EXIT_FAILURE, "Cannot allocate buffer for tx on port %u\n",
245 				 port_id);
246 
247 		rte_eth_tx_buffer_init(tx_buffer[port_id], MAX_PKTS_BURST);
248 
249 		ret = rte_eth_tx_buffer_set_err_callback(tx_buffer[port_id],
250 				flush_tx_error_callback, NULL);
251 		if (ret < 0)
252 			rte_exit(EXIT_FAILURE,
253 			"Cannot set error callback for tx buffer on port %u\n",
254 				 port_id);
255 	}
256 	return 0;
257 }
258 
259 static inline int
260 configure_eth_port(uint16_t port_id)
261 {
262 	struct ether_addr addr;
263 	const uint16_t rxRings = 1, txRings = 1;
264 	int ret;
265 	uint16_t q;
266 	uint16_t nb_rxd = RX_DESC_PER_QUEUE;
267 	uint16_t nb_txd = TX_DESC_PER_QUEUE;
268 	struct rte_eth_dev_info dev_info;
269 	struct rte_eth_txconf txconf;
270 	struct rte_eth_conf port_conf = port_conf_default;
271 
272 	if (!rte_eth_dev_is_valid_port(port_id))
273 		return -1;
274 
275 	rte_eth_dev_info_get(port_id, &dev_info);
276 	if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
277 		port_conf.txmode.offloads |=
278 			DEV_TX_OFFLOAD_MBUF_FAST_FREE;
279 	ret = rte_eth_dev_configure(port_id, rxRings, txRings, &port_conf_default);
280 	if (ret != 0)
281 		return ret;
282 
283 	ret = rte_eth_dev_adjust_nb_rx_tx_desc(port_id, &nb_rxd, &nb_txd);
284 	if (ret != 0)
285 		return ret;
286 
287 	for (q = 0; q < rxRings; q++) {
288 		ret = rte_eth_rx_queue_setup(port_id, q, nb_rxd,
289 				rte_eth_dev_socket_id(port_id), NULL,
290 				mbuf_pool);
291 		if (ret < 0)
292 			return ret;
293 	}
294 
295 	txconf = dev_info.default_txconf;
296 	txconf.txq_flags = ETH_TXQ_FLAGS_IGNORE;
297 	txconf.offloads = port_conf.txmode.offloads;
298 	for (q = 0; q < txRings; q++) {
299 		ret = rte_eth_tx_queue_setup(port_id, q, nb_txd,
300 				rte_eth_dev_socket_id(port_id), &txconf);
301 		if (ret < 0)
302 			return ret;
303 	}
304 
305 	ret = rte_eth_dev_start(port_id);
306 	if (ret < 0)
307 		return ret;
308 
309 	rte_eth_macaddr_get(port_id, &addr);
310 	printf("Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8
311 			" %02"PRIx8" %02"PRIx8" %02"PRIx8"\n",
312 			port_id,
313 			addr.addr_bytes[0], addr.addr_bytes[1],
314 			addr.addr_bytes[2], addr.addr_bytes[3],
315 			addr.addr_bytes[4], addr.addr_bytes[5]);
316 
317 	rte_eth_promiscuous_enable(port_id);
318 
319 	return 0;
320 }
321 
322 static void
323 print_stats(void)
324 {
325 	uint16_t i;
326 	struct rte_eth_stats eth_stats;
327 
328 	printf("\nRX thread stats:\n");
329 	printf(" - Pkts rxd:				%"PRIu64"\n",
330 						app_stats.rx.rx_pkts);
331 	printf(" - Pkts enqd to workers ring:		%"PRIu64"\n",
332 						app_stats.rx.enqueue_pkts);
333 
334 	printf("\nWorker thread stats:\n");
335 	printf(" - Pkts deqd from workers ring:		%"PRIu64"\n",
336 						app_stats.wkr.dequeue_pkts);
337 	printf(" - Pkts enqd to tx ring:		%"PRIu64"\n",
338 						app_stats.wkr.enqueue_pkts);
339 	printf(" - Pkts enq to tx failed:		%"PRIu64"\n",
340 						app_stats.wkr.enqueue_failed_pkts);
341 
342 	printf("\nTX stats:\n");
343 	printf(" - Pkts deqd from tx ring:		%"PRIu64"\n",
344 						app_stats.tx.dequeue_pkts);
345 	printf(" - Ro Pkts transmitted:			%"PRIu64"\n",
346 						app_stats.tx.ro_tx_pkts);
347 	printf(" - Ro Pkts tx failed:			%"PRIu64"\n",
348 						app_stats.tx.ro_tx_failed_pkts);
349 	printf(" - Pkts transmitted w/o reorder:	%"PRIu64"\n",
350 						app_stats.tx.early_pkts_txtd_woro);
351 	printf(" - Pkts tx failed w/o reorder:		%"PRIu64"\n",
352 						app_stats.tx.early_pkts_tx_failed_woro);
353 
354 	RTE_ETH_FOREACH_DEV(i) {
355 		rte_eth_stats_get(i, &eth_stats);
356 		printf("\nPort %u stats:\n", i);
357 		printf(" - Pkts in:   %"PRIu64"\n", eth_stats.ipackets);
358 		printf(" - Pkts out:  %"PRIu64"\n", eth_stats.opackets);
359 		printf(" - In Errs:   %"PRIu64"\n", eth_stats.ierrors);
360 		printf(" - Out Errs:  %"PRIu64"\n", eth_stats.oerrors);
361 		printf(" - Mbuf Errs: %"PRIu64"\n", eth_stats.rx_nombuf);
362 	}
363 }
364 
365 static void
366 int_handler(int sig_num)
367 {
368 	printf("Exiting on signal %d\n", sig_num);
369 	quit_signal = 1;
370 }
371 
372 /**
373  * This thread receives mbufs from the port and affects them an internal
374  * sequence number to keep track of their order of arrival through an
375  * mbuf structure.
376  * The mbufs are then passed to the worker threads via the rx_to_workers
377  * ring.
378  */
379 static int
380 rx_thread(struct rte_ring *ring_out)
381 {
382 	uint32_t seqn = 0;
383 	uint16_t i, ret = 0;
384 	uint16_t nb_rx_pkts;
385 	uint16_t port_id;
386 	struct rte_mbuf *pkts[MAX_PKTS_BURST];
387 
388 	RTE_LOG(INFO, REORDERAPP, "%s() started on lcore %u\n", __func__,
389 							rte_lcore_id());
390 
391 	while (!quit_signal) {
392 
393 		RTE_ETH_FOREACH_DEV(port_id) {
394 			if ((portmask & (1 << port_id)) != 0) {
395 
396 				/* receive packets */
397 				nb_rx_pkts = rte_eth_rx_burst(port_id, 0,
398 								pkts, MAX_PKTS_BURST);
399 				if (nb_rx_pkts == 0) {
400 					RTE_LOG_DP(DEBUG, REORDERAPP,
401 					"%s():Received zero packets\n",	__func__);
402 					continue;
403 				}
404 				app_stats.rx.rx_pkts += nb_rx_pkts;
405 
406 				/* mark sequence number */
407 				for (i = 0; i < nb_rx_pkts; )
408 					pkts[i++]->seqn = seqn++;
409 
410 				/* enqueue to rx_to_workers ring */
411 				ret = rte_ring_enqueue_burst(ring_out,
412 						(void *)pkts, nb_rx_pkts, NULL);
413 				app_stats.rx.enqueue_pkts += ret;
414 				if (unlikely(ret < nb_rx_pkts)) {
415 					app_stats.rx.enqueue_failed_pkts +=
416 									(nb_rx_pkts-ret);
417 					pktmbuf_free_bulk(&pkts[ret], nb_rx_pkts - ret);
418 				}
419 			}
420 		}
421 	}
422 	return 0;
423 }
424 
425 /**
426  * This thread takes bursts of packets from the rx_to_workers ring and
427  * Changes the input port value to output port value. And feds it to
428  * workers_to_tx
429  */
430 static int
431 worker_thread(void *args_ptr)
432 {
433 	const uint16_t nb_ports = rte_eth_dev_count_avail();
434 	uint16_t i, ret = 0;
435 	uint16_t burst_size = 0;
436 	struct worker_thread_args *args;
437 	struct rte_mbuf *burst_buffer[MAX_PKTS_BURST] = { NULL };
438 	struct rte_ring *ring_in, *ring_out;
439 	const unsigned xor_val = (nb_ports > 1);
440 
441 	args = (struct worker_thread_args *) args_ptr;
442 	ring_in  = args->ring_in;
443 	ring_out = args->ring_out;
444 
445 	RTE_LOG(INFO, REORDERAPP, "%s() started on lcore %u\n", __func__,
446 							rte_lcore_id());
447 
448 	while (!quit_signal) {
449 
450 		/* dequeue the mbufs from rx_to_workers ring */
451 		burst_size = rte_ring_dequeue_burst(ring_in,
452 				(void *)burst_buffer, MAX_PKTS_BURST, NULL);
453 		if (unlikely(burst_size == 0))
454 			continue;
455 
456 		__sync_fetch_and_add(&app_stats.wkr.dequeue_pkts, burst_size);
457 
458 		/* just do some operation on mbuf */
459 		for (i = 0; i < burst_size;)
460 			burst_buffer[i++]->port ^= xor_val;
461 
462 		/* enqueue the modified mbufs to workers_to_tx ring */
463 		ret = rte_ring_enqueue_burst(ring_out, (void *)burst_buffer,
464 				burst_size, NULL);
465 		__sync_fetch_and_add(&app_stats.wkr.enqueue_pkts, ret);
466 		if (unlikely(ret < burst_size)) {
467 			/* Return the mbufs to their respective pool, dropping packets */
468 			__sync_fetch_and_add(&app_stats.wkr.enqueue_failed_pkts,
469 					(int)burst_size - ret);
470 			pktmbuf_free_bulk(&burst_buffer[ret], burst_size - ret);
471 		}
472 	}
473 	return 0;
474 }
475 
476 /**
477  * Dequeue mbufs from the workers_to_tx ring and reorder them before
478  * transmitting.
479  */
480 static int
481 send_thread(struct send_thread_args *args)
482 {
483 	int ret;
484 	unsigned int i, dret;
485 	uint16_t nb_dq_mbufs;
486 	uint8_t outp;
487 	unsigned sent;
488 	struct rte_mbuf *mbufs[MAX_PKTS_BURST];
489 	struct rte_mbuf *rombufs[MAX_PKTS_BURST] = {NULL};
490 	static struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
491 
492 	RTE_LOG(INFO, REORDERAPP, "%s() started on lcore %u\n", __func__, rte_lcore_id());
493 
494 	configure_tx_buffers(tx_buffer);
495 
496 	while (!quit_signal) {
497 
498 		/* deque the mbufs from workers_to_tx ring */
499 		nb_dq_mbufs = rte_ring_dequeue_burst(args->ring_in,
500 				(void *)mbufs, MAX_PKTS_BURST, NULL);
501 
502 		if (unlikely(nb_dq_mbufs == 0))
503 			continue;
504 
505 		app_stats.tx.dequeue_pkts += nb_dq_mbufs;
506 
507 		for (i = 0; i < nb_dq_mbufs; i++) {
508 			/* send dequeued mbufs for reordering */
509 			ret = rte_reorder_insert(args->buffer, mbufs[i]);
510 
511 			if (ret == -1 && rte_errno == ERANGE) {
512 				/* Too early pkts should be transmitted out directly */
513 				RTE_LOG_DP(DEBUG, REORDERAPP,
514 						"%s():Cannot reorder early packet "
515 						"direct enqueuing to TX\n", __func__);
516 				outp = mbufs[i]->port;
517 				if ((portmask & (1 << outp)) == 0) {
518 					rte_pktmbuf_free(mbufs[i]);
519 					continue;
520 				}
521 				if (rte_eth_tx_burst(outp, 0, (void *)mbufs[i], 1) != 1) {
522 					rte_pktmbuf_free(mbufs[i]);
523 					app_stats.tx.early_pkts_tx_failed_woro++;
524 				} else
525 					app_stats.tx.early_pkts_txtd_woro++;
526 			} else if (ret == -1 && rte_errno == ENOSPC) {
527 				/**
528 				 * Early pkts just outside of window should be dropped
529 				 */
530 				rte_pktmbuf_free(mbufs[i]);
531 			}
532 		}
533 
534 		/*
535 		 * drain MAX_PKTS_BURST of reordered
536 		 * mbufs for transmit
537 		 */
538 		dret = rte_reorder_drain(args->buffer, rombufs, MAX_PKTS_BURST);
539 		for (i = 0; i < dret; i++) {
540 
541 			struct rte_eth_dev_tx_buffer *outbuf;
542 			uint8_t outp1;
543 
544 			outp1 = rombufs[i]->port;
545 			/* skip ports that are not enabled */
546 			if ((portmask & (1 << outp1)) == 0) {
547 				rte_pktmbuf_free(rombufs[i]);
548 				continue;
549 			}
550 
551 			outbuf = tx_buffer[outp1];
552 			sent = rte_eth_tx_buffer(outp1, 0, outbuf, rombufs[i]);
553 			if (sent)
554 				app_stats.tx.ro_tx_pkts += sent;
555 		}
556 	}
557 
558 	free_tx_buffers(tx_buffer);
559 
560 	return 0;
561 }
562 
563 /**
564  * Dequeue mbufs from the workers_to_tx ring and transmit them
565  */
566 static int
567 tx_thread(struct rte_ring *ring_in)
568 {
569 	uint32_t i, dqnum;
570 	uint8_t outp;
571 	unsigned sent;
572 	struct rte_mbuf *mbufs[MAX_PKTS_BURST];
573 	struct rte_eth_dev_tx_buffer *outbuf;
574 	static struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
575 
576 	RTE_LOG(INFO, REORDERAPP, "%s() started on lcore %u\n", __func__,
577 							rte_lcore_id());
578 
579 	configure_tx_buffers(tx_buffer);
580 
581 	while (!quit_signal) {
582 
583 		/* deque the mbufs from workers_to_tx ring */
584 		dqnum = rte_ring_dequeue_burst(ring_in,
585 				(void *)mbufs, MAX_PKTS_BURST, NULL);
586 
587 		if (unlikely(dqnum == 0))
588 			continue;
589 
590 		app_stats.tx.dequeue_pkts += dqnum;
591 
592 		for (i = 0; i < dqnum; i++) {
593 			outp = mbufs[i]->port;
594 			/* skip ports that are not enabled */
595 			if ((portmask & (1 << outp)) == 0) {
596 				rte_pktmbuf_free(mbufs[i]);
597 				continue;
598 			}
599 
600 			outbuf = tx_buffer[outp];
601 			sent = rte_eth_tx_buffer(outp, 0, outbuf, mbufs[i]);
602 			if (sent)
603 				app_stats.tx.ro_tx_pkts += sent;
604 		}
605 	}
606 
607 	return 0;
608 }
609 
610 int
611 main(int argc, char **argv)
612 {
613 	int ret;
614 	unsigned nb_ports;
615 	unsigned int lcore_id, last_lcore_id, master_lcore_id;
616 	uint16_t port_id;
617 	uint16_t nb_ports_available;
618 	struct worker_thread_args worker_args = {NULL, NULL};
619 	struct send_thread_args send_args = {NULL, NULL};
620 	struct rte_ring *rx_to_workers;
621 	struct rte_ring *workers_to_tx;
622 
623 	/* catch ctrl-c so we can print on exit */
624 	signal(SIGINT, int_handler);
625 
626 	/* Initialize EAL */
627 	ret = rte_eal_init(argc, argv);
628 	if (ret < 0)
629 		return -1;
630 
631 	argc -= ret;
632 	argv += ret;
633 
634 	/* Parse the application specific arguments */
635 	ret = parse_args(argc, argv);
636 	if (ret < 0)
637 		return -1;
638 
639 	/* Check if we have enought cores */
640 	if (rte_lcore_count() < 3)
641 		rte_exit(EXIT_FAILURE, "Error, This application needs at "
642 				"least 3 logical cores to run:\n"
643 				"1 lcore for packet RX\n"
644 				"1 lcore for packet TX\n"
645 				"and at least 1 lcore for worker threads\n");
646 
647 	nb_ports = rte_eth_dev_count_avail();
648 	if (nb_ports == 0)
649 		rte_exit(EXIT_FAILURE, "Error: no ethernet ports detected\n");
650 	if (nb_ports != 1 && (nb_ports & 1))
651 		rte_exit(EXIT_FAILURE, "Error: number of ports must be even, except "
652 				"when using a single port\n");
653 
654 	mbuf_pool = rte_pktmbuf_pool_create("mbuf_pool", MBUF_PER_POOL,
655 			MBUF_POOL_CACHE_SIZE, 0, RTE_MBUF_DEFAULT_BUF_SIZE,
656 			rte_socket_id());
657 	if (mbuf_pool == NULL)
658 		rte_exit(EXIT_FAILURE, "%s\n", rte_strerror(rte_errno));
659 
660 	nb_ports_available = nb_ports;
661 
662 	/* initialize all ports */
663 	RTE_ETH_FOREACH_DEV(port_id) {
664 		/* skip ports that are not enabled */
665 		if ((portmask & (1 << port_id)) == 0) {
666 			printf("\nSkipping disabled port %d\n", port_id);
667 			nb_ports_available--;
668 			continue;
669 		}
670 		/* init port */
671 		printf("Initializing port %u... done\n", port_id);
672 
673 		if (configure_eth_port(port_id) != 0)
674 			rte_exit(EXIT_FAILURE, "Cannot initialize port %"PRIu8"\n",
675 					port_id);
676 	}
677 
678 	if (!nb_ports_available) {
679 		rte_exit(EXIT_FAILURE,
680 			"All available ports are disabled. Please set portmask.\n");
681 	}
682 
683 	/* Create rings for inter core communication */
684 	rx_to_workers = rte_ring_create("rx_to_workers", RING_SIZE, rte_socket_id(),
685 			RING_F_SP_ENQ);
686 	if (rx_to_workers == NULL)
687 		rte_exit(EXIT_FAILURE, "%s\n", rte_strerror(rte_errno));
688 
689 	workers_to_tx = rte_ring_create("workers_to_tx", RING_SIZE, rte_socket_id(),
690 			RING_F_SC_DEQ);
691 	if (workers_to_tx == NULL)
692 		rte_exit(EXIT_FAILURE, "%s\n", rte_strerror(rte_errno));
693 
694 	if (!disable_reorder) {
695 		send_args.buffer = rte_reorder_create("PKT_RO", rte_socket_id(),
696 				REORDER_BUFFER_SIZE);
697 		if (send_args.buffer == NULL)
698 			rte_exit(EXIT_FAILURE, "%s\n", rte_strerror(rte_errno));
699 	}
700 
701 	last_lcore_id   = get_last_lcore_id();
702 	master_lcore_id = rte_get_master_lcore();
703 
704 	worker_args.ring_in  = rx_to_workers;
705 	worker_args.ring_out = workers_to_tx;
706 
707 	/* Start worker_thread() on all the available slave cores but the last 1 */
708 	for (lcore_id = 0; lcore_id <= get_previous_lcore_id(last_lcore_id); lcore_id++)
709 		if (rte_lcore_is_enabled(lcore_id) && lcore_id != master_lcore_id)
710 			rte_eal_remote_launch(worker_thread, (void *)&worker_args,
711 					lcore_id);
712 
713 	if (disable_reorder) {
714 		/* Start tx_thread() on the last slave core */
715 		rte_eal_remote_launch((lcore_function_t *)tx_thread, workers_to_tx,
716 				last_lcore_id);
717 	} else {
718 		send_args.ring_in = workers_to_tx;
719 		/* Start send_thread() on the last slave core */
720 		rte_eal_remote_launch((lcore_function_t *)send_thread,
721 				(void *)&send_args, last_lcore_id);
722 	}
723 
724 	/* Start rx_thread() on the master core */
725 	rx_thread(rx_to_workers);
726 
727 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
728 		if (rte_eal_wait_lcore(lcore_id) < 0)
729 			return -1;
730 	}
731 
732 	print_stats();
733 	return 0;
734 }
735