xref: /dpdk/examples/ipv4_multicast/main.c (revision 25d11a86c56d50947af33d0b79ede622809bd8b9)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2014 Intel Corporation
3  */
4 
5 #include <stdio.h>
6 #include <stdlib.h>
7 #include <stdint.h>
8 #include <inttypes.h>
9 #include <sys/types.h>
10 #include <string.h>
11 #include <sys/queue.h>
12 #include <stdarg.h>
13 #include <errno.h>
14 #include <getopt.h>
15 
16 #include <rte_common.h>
17 #include <rte_byteorder.h>
18 #include <rte_log.h>
19 #include <rte_memory.h>
20 #include <rte_memcpy.h>
21 #include <rte_eal.h>
22 #include <rte_launch.h>
23 #include <rte_atomic.h>
24 #include <rte_cycles.h>
25 #include <rte_prefetch.h>
26 #include <rte_lcore.h>
27 #include <rte_per_lcore.h>
28 #include <rte_branch_prediction.h>
29 #include <rte_interrupts.h>
30 #include <rte_random.h>
31 #include <rte_debug.h>
32 #include <rte_ether.h>
33 #include <rte_ethdev.h>
34 #include <rte_mempool.h>
35 #include <rte_mbuf.h>
36 #include <rte_malloc.h>
37 #include <rte_fbk_hash.h>
38 #include <rte_ip.h>
39 
40 #define RTE_LOGTYPE_IPv4_MULTICAST RTE_LOGTYPE_USER1
41 
42 #define MAX_PORTS 16
43 
44 #define	MCAST_CLONE_PORTS	2
45 #define	MCAST_CLONE_SEGS	2
46 
47 #define	PKT_MBUF_DATA_SIZE	RTE_MBUF_DEFAULT_BUF_SIZE
48 #define	NB_PKT_MBUF	8192
49 
50 #define	HDR_MBUF_DATA_SIZE	(2 * RTE_PKTMBUF_HEADROOM)
51 #define	NB_HDR_MBUF	(NB_PKT_MBUF * MAX_PORTS)
52 
53 #define	NB_CLONE_MBUF	(NB_PKT_MBUF * MCAST_CLONE_PORTS * MCAST_CLONE_SEGS * 2)
54 
55 /* allow max jumbo frame 9.5 KB */
56 #define	JUMBO_FRAME_MAX_SIZE	0x2600
57 
58 #define MAX_PKT_BURST 32
59 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
60 
61 /* Configure how many packets ahead to prefetch, when reading packets */
62 #define PREFETCH_OFFSET	3
63 
64 /*
65  * Construct Ethernet multicast address from IPv4 multicast address.
66  * Citing RFC 1112, section 6.4:
67  * "An IP host group address is mapped to an Ethernet multicast address
68  * by placing the low-order 23-bits of the IP address into the low-order
69  * 23 bits of the Ethernet multicast address 01-00-5E-00-00-00 (hex)."
70  */
71 #define	ETHER_ADDR_FOR_IPV4_MCAST(x)	\
72 	(rte_cpu_to_be_64(0x01005e000000ULL | ((x) & 0x7fffff)) >> 16)
73 
74 /*
75  * Configurable number of RX/TX ring descriptors
76  */
77 #define RTE_TEST_RX_DESC_DEFAULT 1024
78 #define RTE_TEST_TX_DESC_DEFAULT 1024
79 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
80 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
81 
82 /* ethernet addresses of ports */
83 static struct ether_addr ports_eth_addr[MAX_PORTS];
84 
85 /* mask of enabled ports */
86 static uint32_t enabled_port_mask = 0;
87 
88 static uint16_t nb_ports;
89 
90 static int rx_queue_per_lcore = 1;
91 
92 struct mbuf_table {
93 	uint16_t len;
94 	struct rte_mbuf *m_table[MAX_PKT_BURST];
95 };
96 
97 #define MAX_RX_QUEUE_PER_LCORE 16
98 #define MAX_TX_QUEUE_PER_PORT 16
99 struct lcore_queue_conf {
100 	uint64_t tx_tsc;
101 	uint16_t n_rx_queue;
102 	uint8_t rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
103 	uint16_t tx_queue_id[MAX_PORTS];
104 	struct mbuf_table tx_mbufs[MAX_PORTS];
105 } __rte_cache_aligned;
106 static struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
107 
108 static struct rte_eth_conf port_conf = {
109 	.rxmode = {
110 		.max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
111 		.split_hdr_size = 0,
112 		.offloads = DEV_RX_OFFLOAD_JUMBO_FRAME,
113 	},
114 	.txmode = {
115 		.mq_mode = ETH_MQ_TX_NONE,
116 		.offloads = DEV_TX_OFFLOAD_MULTI_SEGS,
117 	},
118 };
119 
120 static struct rte_mempool *packet_pool, *header_pool, *clone_pool;
121 
122 
123 /* Multicast */
124 static struct rte_fbk_hash_params mcast_hash_params = {
125 	.name = "MCAST_HASH",
126 	.entries = 1024,
127 	.entries_per_bucket = 4,
128 	.socket_id = 0,
129 	.hash_func = NULL,
130 	.init_val = 0,
131 };
132 
133 struct rte_fbk_hash_table *mcast_hash = NULL;
134 
135 struct mcast_group_params {
136 	uint32_t ip;
137 	uint16_t port_mask;
138 };
139 
140 static struct mcast_group_params mcast_group_table[] = {
141 		{IPv4(224,0,0,101), 0x1},
142 		{IPv4(224,0,0,102), 0x2},
143 		{IPv4(224,0,0,103), 0x3},
144 		{IPv4(224,0,0,104), 0x4},
145 		{IPv4(224,0,0,105), 0x5},
146 		{IPv4(224,0,0,106), 0x6},
147 		{IPv4(224,0,0,107), 0x7},
148 		{IPv4(224,0,0,108), 0x8},
149 		{IPv4(224,0,0,109), 0x9},
150 		{IPv4(224,0,0,110), 0xA},
151 		{IPv4(224,0,0,111), 0xB},
152 		{IPv4(224,0,0,112), 0xC},
153 		{IPv4(224,0,0,113), 0xD},
154 		{IPv4(224,0,0,114), 0xE},
155 		{IPv4(224,0,0,115), 0xF},
156 };
157 
158 #define N_MCAST_GROUPS \
159 	(sizeof (mcast_group_table) / sizeof (mcast_group_table[0]))
160 
161 
162 /* Send burst of packets on an output interface */
163 static void
164 send_burst(struct lcore_queue_conf *qconf, uint16_t port)
165 {
166 	struct rte_mbuf **m_table;
167 	uint16_t n, queueid;
168 	int ret;
169 
170 	queueid = qconf->tx_queue_id[port];
171 	m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
172 	n = qconf->tx_mbufs[port].len;
173 
174 	ret = rte_eth_tx_burst(port, queueid, m_table, n);
175 	while (unlikely (ret < n)) {
176 		rte_pktmbuf_free(m_table[ret]);
177 		ret++;
178 	}
179 
180 	qconf->tx_mbufs[port].len = 0;
181 }
182 
183 /* Get number of bits set. */
184 static inline uint32_t
185 bitcnt(uint32_t v)
186 {
187 	uint32_t n;
188 
189 	for (n = 0; v != 0; v &= v - 1, n++)
190 		;
191 
192 	return n;
193 }
194 
195 /**
196  * Create the output multicast packet based on the given input packet.
197  * There are two approaches for creating outgoing packet, though both
198  * are based on data zero-copy idea, they differ in few details:
199  * First one creates a clone of the input packet, e.g - walk though all
200  * segments of the input packet, and for each of them create a new packet
201  * mbuf and attach that new mbuf to the segment (refer to rte_pktmbuf_clone()
202  * for more details). Then new mbuf is allocated for the packet header
203  * and is prepended to the 'clone' mbuf.
204  * Second approach doesn't make a clone, it just increment refcnt for all
205  * input packet segments. Then it allocates new mbuf for the packet header
206  * and prepends it to the input packet.
207  * Basically first approach reuses only input packet's data, but creates
208  * it's own copy of packet's metadata. Second approach reuses both input's
209  * packet data and metadata.
210  * The advantage of first approach - is that each outgoing packet has it's
211  * own copy of metadata, so we can safely modify data pointer of the
212  * input packet. That allows us to skip creation if the output packet for
213  * the last destination port, but instead modify input packet's header inplace,
214  * e.g: for N destination ports we need to invoke mcast_out_pkt (N-1) times.
215  * The advantage of second approach - less work for each outgoing packet,
216  * e.g: we skip "clone" operation completely. Though it comes with a price -
217  * input packet's metadata has to be intact. So for N destination ports we
218  * need to invoke mcast_out_pkt N times.
219  * So for small number of outgoing ports (and segments in the input packet)
220  * first approach will be faster.
221  * As number of outgoing ports (and/or input segments) will grow,
222  * second way will become more preferable.
223  *
224  *  @param pkt
225  *  Input packet mbuf.
226  *  @param use_clone
227  *  Control which of the two approaches described above should be used:
228  *  - 0 - use second approach:
229  *    Don't "clone" input packet.
230  *    Prepend new header directly to the input packet
231  *  - 1 - use first approach:
232  *    Make a "clone" of input packet first.
233  *    Prepend new header to the clone of the input packet
234  *  @return
235  *  - The pointer to the new outgoing packet.
236  *  - NULL if operation failed.
237  */
238 static inline struct rte_mbuf *
239 mcast_out_pkt(struct rte_mbuf *pkt, int use_clone)
240 {
241 	struct rte_mbuf *hdr;
242 
243 	/* Create new mbuf for the header. */
244 	if (unlikely ((hdr = rte_pktmbuf_alloc(header_pool)) == NULL))
245 		return NULL;
246 
247 	/* If requested, then make a new clone packet. */
248 	if (use_clone != 0 &&
249 	    unlikely ((pkt = rte_pktmbuf_clone(pkt, clone_pool)) == NULL)) {
250 		rte_pktmbuf_free(hdr);
251 		return NULL;
252 	}
253 
254 	/* prepend new header */
255 	hdr->next = pkt;
256 
257 	/* update header's fields */
258 	hdr->pkt_len = (uint16_t)(hdr->data_len + pkt->pkt_len);
259 	hdr->nb_segs = pkt->nb_segs + 1;
260 
261 	__rte_mbuf_sanity_check(hdr, 1);
262 	return hdr;
263 }
264 
265 /*
266  * Write new Ethernet header to the outgoing packet,
267  * and put it into the outgoing queue for the given port.
268  */
269 static inline void
270 mcast_send_pkt(struct rte_mbuf *pkt, struct ether_addr *dest_addr,
271 		struct lcore_queue_conf *qconf, uint16_t port)
272 {
273 	struct ether_hdr *ethdr;
274 	uint16_t len;
275 
276 	/* Construct Ethernet header. */
277 	ethdr = (struct ether_hdr *)rte_pktmbuf_prepend(pkt, (uint16_t)sizeof(*ethdr));
278 	RTE_ASSERT(ethdr != NULL);
279 
280 	ether_addr_copy(dest_addr, &ethdr->d_addr);
281 	ether_addr_copy(&ports_eth_addr[port], &ethdr->s_addr);
282 	ethdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
283 
284 	/* Put new packet into the output queue */
285 	len = qconf->tx_mbufs[port].len;
286 	qconf->tx_mbufs[port].m_table[len] = pkt;
287 	qconf->tx_mbufs[port].len = ++len;
288 
289 	/* Transmit packets */
290 	if (unlikely(MAX_PKT_BURST == len))
291 		send_burst(qconf, port);
292 }
293 
294 /* Multicast forward of the input packet */
295 static inline void
296 mcast_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf)
297 {
298 	struct rte_mbuf *mc;
299 	struct ipv4_hdr *iphdr;
300 	uint32_t dest_addr, port_mask, port_num, use_clone;
301 	int32_t hash;
302 	uint16_t port;
303 	union {
304 		uint64_t as_int;
305 		struct ether_addr as_addr;
306 	} dst_eth_addr;
307 
308 	/* Remove the Ethernet header from the input packet */
309 	iphdr = (struct ipv4_hdr *)rte_pktmbuf_adj(m, (uint16_t)sizeof(struct ether_hdr));
310 	RTE_ASSERT(iphdr != NULL);
311 
312 	dest_addr = rte_be_to_cpu_32(iphdr->dst_addr);
313 
314 	/*
315 	 * Check that it is a valid multicast address and
316 	 * we have some active ports assigned to it.
317 	 */
318 	if(!IS_IPV4_MCAST(dest_addr) ||
319 	    (hash = rte_fbk_hash_lookup(mcast_hash, dest_addr)) <= 0 ||
320 	    (port_mask = hash & enabled_port_mask) == 0) {
321 		rte_pktmbuf_free(m);
322 		return;
323 	}
324 
325 	/* Calculate number of destination ports. */
326 	port_num = bitcnt(port_mask);
327 
328 	/* Should we use rte_pktmbuf_clone() or not. */
329 	use_clone = (port_num <= MCAST_CLONE_PORTS &&
330 	    m->nb_segs <= MCAST_CLONE_SEGS);
331 
332 	/* Mark all packet's segments as referenced port_num times */
333 	if (use_clone == 0)
334 		rte_pktmbuf_refcnt_update(m, (uint16_t)port_num);
335 
336 	/* construct destination ethernet address */
337 	dst_eth_addr.as_int = ETHER_ADDR_FOR_IPV4_MCAST(dest_addr);
338 
339 	for (port = 0; use_clone != port_mask; port_mask >>= 1, port++) {
340 
341 		/* Prepare output packet and send it out. */
342 		if ((port_mask & 1) != 0) {
343 			if (likely ((mc = mcast_out_pkt(m, use_clone)) != NULL))
344 				mcast_send_pkt(mc, &dst_eth_addr.as_addr,
345 						qconf, port);
346 			else if (use_clone == 0)
347 				rte_pktmbuf_free(m);
348 		}
349 	}
350 
351 	/*
352 	 * If we making clone packets, then, for the last destination port,
353 	 * we can overwrite input packet's metadata.
354 	 */
355 	if (use_clone != 0)
356 		mcast_send_pkt(m, &dst_eth_addr.as_addr, qconf, port);
357 	else
358 		rte_pktmbuf_free(m);
359 }
360 
361 /* Send burst of outgoing packet, if timeout expires. */
362 static inline void
363 send_timeout_burst(struct lcore_queue_conf *qconf)
364 {
365 	uint64_t cur_tsc;
366 	uint16_t portid;
367 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
368 
369 	cur_tsc = rte_rdtsc();
370 	if (likely (cur_tsc < qconf->tx_tsc + drain_tsc))
371 		return;
372 
373 	for (portid = 0; portid < MAX_PORTS; portid++) {
374 		if (qconf->tx_mbufs[portid].len != 0)
375 			send_burst(qconf, portid);
376 	}
377 	qconf->tx_tsc = cur_tsc;
378 }
379 
380 /* main processing loop */
381 static int
382 main_loop(__rte_unused void *dummy)
383 {
384 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
385 	unsigned lcore_id;
386 	int i, j, nb_rx;
387 	uint16_t portid;
388 	struct lcore_queue_conf *qconf;
389 
390 	lcore_id = rte_lcore_id();
391 	qconf = &lcore_queue_conf[lcore_id];
392 
393 
394 	if (qconf->n_rx_queue == 0) {
395 		RTE_LOG(INFO, IPv4_MULTICAST, "lcore %u has nothing to do\n",
396 		    lcore_id);
397 		return 0;
398 	}
399 
400 	RTE_LOG(INFO, IPv4_MULTICAST, "entering main loop on lcore %u\n",
401 	    lcore_id);
402 
403 	for (i = 0; i < qconf->n_rx_queue; i++) {
404 
405 		portid = qconf->rx_queue_list[i];
406 		RTE_LOG(INFO, IPv4_MULTICAST, " -- lcoreid=%u portid=%d\n",
407 		    lcore_id, portid);
408 	}
409 
410 	while (1) {
411 
412 		/*
413 		 * Read packet from RX queues
414 		 */
415 		for (i = 0; i < qconf->n_rx_queue; i++) {
416 
417 			portid = qconf->rx_queue_list[i];
418 			nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
419 						 MAX_PKT_BURST);
420 
421 			/* Prefetch first packets */
422 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
423 				rte_prefetch0(rte_pktmbuf_mtod(
424 						pkts_burst[j], void *));
425 			}
426 
427 			/* Prefetch and forward already prefetched packets */
428 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
429 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
430 						j + PREFETCH_OFFSET], void *));
431 				mcast_forward(pkts_burst[j], qconf);
432 			}
433 
434 			/* Forward remaining prefetched packets */
435 			for (; j < nb_rx; j++) {
436 				mcast_forward(pkts_burst[j], qconf);
437 			}
438 		}
439 
440 		/* Send out packets from TX queues */
441 		send_timeout_burst(qconf);
442 	}
443 }
444 
445 /* display usage */
446 static void
447 print_usage(const char *prgname)
448 {
449 	printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
450 	    "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
451 	    "  -q NQ: number of queue (=ports) per lcore (default is 1)\n",
452 	    prgname);
453 }
454 
455 static uint32_t
456 parse_portmask(const char *portmask)
457 {
458 	char *end = NULL;
459 	unsigned long pm;
460 
461 	/* parse hexadecimal string */
462 	pm = strtoul(portmask, &end, 16);
463 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
464 		return 0;
465 
466 	return (uint32_t)pm;
467 }
468 
469 static int
470 parse_nqueue(const char *q_arg)
471 {
472 	char *end = NULL;
473 	unsigned long n;
474 
475 	/* parse numerical string */
476 	errno = 0;
477 	n = strtoul(q_arg, &end, 0);
478 	if (errno != 0 || end == NULL || *end != '\0' ||
479 			n == 0 || n >= MAX_RX_QUEUE_PER_LCORE)
480 		return -1;
481 
482 	return n;
483 }
484 
485 /* Parse the argument given in the command line of the application */
486 static int
487 parse_args(int argc, char **argv)
488 {
489 	int opt, ret;
490 	char **argvopt;
491 	int option_index;
492 	char *prgname = argv[0];
493 	static struct option lgopts[] = {
494 		{NULL, 0, 0, 0}
495 	};
496 
497 	argvopt = argv;
498 
499 	while ((opt = getopt_long(argc, argvopt, "p:q:",
500 				  lgopts, &option_index)) != EOF) {
501 
502 		switch (opt) {
503 		/* portmask */
504 		case 'p':
505 			enabled_port_mask = parse_portmask(optarg);
506 			if (enabled_port_mask == 0) {
507 				printf("invalid portmask\n");
508 				print_usage(prgname);
509 				return -1;
510 			}
511 			break;
512 
513 		/* nqueue */
514 		case 'q':
515 			rx_queue_per_lcore = parse_nqueue(optarg);
516 			if (rx_queue_per_lcore < 0) {
517 				printf("invalid queue number\n");
518 				print_usage(prgname);
519 				return -1;
520 			}
521 			break;
522 
523 		default:
524 			print_usage(prgname);
525 			return -1;
526 		}
527 	}
528 
529 	if (optind >= 0)
530 		argv[optind-1] = prgname;
531 
532 	ret = optind-1;
533 	optind = 1; /* reset getopt lib */
534 	return ret;
535 }
536 
537 static void
538 print_ethaddr(const char *name, struct ether_addr *eth_addr)
539 {
540 	char buf[ETHER_ADDR_FMT_SIZE];
541 	ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
542 	printf("%s%s", name, buf);
543 }
544 
545 static int
546 init_mcast_hash(void)
547 {
548 	uint32_t i;
549 
550 	mcast_hash_params.socket_id = rte_socket_id();
551 	mcast_hash = rte_fbk_hash_create(&mcast_hash_params);
552 	if (mcast_hash == NULL){
553 		return -1;
554 	}
555 
556 	for (i = 0; i < N_MCAST_GROUPS; i ++){
557 		if (rte_fbk_hash_add_key(mcast_hash,
558 			mcast_group_table[i].ip,
559 			mcast_group_table[i].port_mask) < 0) {
560 			return -1;
561 		}
562 	}
563 
564 	return 0;
565 }
566 
567 /* Check the link status of all ports in up to 9s, and print them finally */
568 static void
569 check_all_ports_link_status(uint32_t port_mask)
570 {
571 #define CHECK_INTERVAL 100 /* 100ms */
572 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
573 	uint16_t portid;
574 	uint8_t count, all_ports_up, print_flag = 0;
575 	struct rte_eth_link link;
576 
577 	printf("\nChecking link status");
578 	fflush(stdout);
579 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
580 		all_ports_up = 1;
581 		RTE_ETH_FOREACH_DEV(portid) {
582 			if ((port_mask & (1 << portid)) == 0)
583 				continue;
584 			memset(&link, 0, sizeof(link));
585 			rte_eth_link_get_nowait(portid, &link);
586 			/* print link status if flag set */
587 			if (print_flag == 1) {
588 				if (link.link_status)
589 					printf(
590 					"Port%d Link Up. Speed %u Mbps - %s\n",
591 					portid, link.link_speed,
592 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
593 					("full-duplex") : ("half-duplex\n"));
594 				else
595 					printf("Port %d Link Down\n", portid);
596 				continue;
597 			}
598 			/* clear all_ports_up flag if any link down */
599 			if (link.link_status == ETH_LINK_DOWN) {
600 				all_ports_up = 0;
601 				break;
602 			}
603 		}
604 		/* after finally printing all link status, get out */
605 		if (print_flag == 1)
606 			break;
607 
608 		if (all_ports_up == 0) {
609 			printf(".");
610 			fflush(stdout);
611 			rte_delay_ms(CHECK_INTERVAL);
612 		}
613 
614 		/* set the print_flag if all ports up or timeout */
615 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
616 			print_flag = 1;
617 			printf("done\n");
618 		}
619 	}
620 }
621 
622 int
623 main(int argc, char **argv)
624 {
625 	struct lcore_queue_conf *qconf;
626 	struct rte_eth_dev_info dev_info;
627 	struct rte_eth_txconf *txconf;
628 	int ret;
629 	uint16_t queueid;
630 	unsigned lcore_id = 0, rx_lcore_id = 0;
631 	uint32_t n_tx_queue, nb_lcores;
632 	uint16_t portid;
633 
634 	/* init EAL */
635 	ret = rte_eal_init(argc, argv);
636 	if (ret < 0)
637 		rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
638 	argc -= ret;
639 	argv += ret;
640 
641 	/* parse application arguments (after the EAL ones) */
642 	ret = parse_args(argc, argv);
643 	if (ret < 0)
644 		rte_exit(EXIT_FAILURE, "Invalid IPV4_MULTICAST parameters\n");
645 
646 	/* create the mbuf pools */
647 	packet_pool = rte_pktmbuf_pool_create("packet_pool", NB_PKT_MBUF, 32,
648 		0, PKT_MBUF_DATA_SIZE, rte_socket_id());
649 
650 	if (packet_pool == NULL)
651 		rte_exit(EXIT_FAILURE, "Cannot init packet mbuf pool\n");
652 
653 	header_pool = rte_pktmbuf_pool_create("header_pool", NB_HDR_MBUF, 32,
654 		0, HDR_MBUF_DATA_SIZE, rte_socket_id());
655 
656 	if (header_pool == NULL)
657 		rte_exit(EXIT_FAILURE, "Cannot init header mbuf pool\n");
658 
659 	clone_pool = rte_pktmbuf_pool_create("clone_pool", NB_CLONE_MBUF, 32,
660 		0, 0, rte_socket_id());
661 
662 	if (clone_pool == NULL)
663 		rte_exit(EXIT_FAILURE, "Cannot init clone mbuf pool\n");
664 
665 	nb_ports = rte_eth_dev_count_avail();
666 	if (nb_ports == 0)
667 		rte_exit(EXIT_FAILURE, "No physical ports!\n");
668 	if (nb_ports > MAX_PORTS)
669 		nb_ports = MAX_PORTS;
670 
671 	nb_lcores = rte_lcore_count();
672 
673 	/* initialize all ports */
674 	RTE_ETH_FOREACH_DEV(portid) {
675 		struct rte_eth_rxconf rxq_conf;
676 		struct rte_eth_conf local_port_conf = port_conf;
677 
678 		/* skip ports that are not enabled */
679 		if ((enabled_port_mask & (1 << portid)) == 0) {
680 			printf("Skipping disabled port %d\n", portid);
681 			continue;
682 		}
683 
684 		qconf = &lcore_queue_conf[rx_lcore_id];
685 
686 		/* limit the frame size to the maximum supported by NIC */
687 		rte_eth_dev_info_get(portid, &dev_info);
688 		local_port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
689 		    dev_info.max_rx_pktlen,
690 		    local_port_conf.rxmode.max_rx_pkt_len);
691 
692 		/* get the lcore_id for this port */
693 		while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
694 		       qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
695 
696 			rx_lcore_id ++;
697 			qconf = &lcore_queue_conf[rx_lcore_id];
698 
699 			if (rx_lcore_id >= RTE_MAX_LCORE)
700 				rte_exit(EXIT_FAILURE, "Not enough cores\n");
701 		}
702 		qconf->rx_queue_list[qconf->n_rx_queue] = portid;
703 		qconf->n_rx_queue++;
704 
705 		/* init port */
706 		printf("Initializing port %d on lcore %u... ", portid,
707 		       rx_lcore_id);
708 		fflush(stdout);
709 
710 		n_tx_queue = nb_lcores;
711 		if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
712 			n_tx_queue = MAX_TX_QUEUE_PER_PORT;
713 
714 		ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
715 					    &local_port_conf);
716 		if (ret < 0)
717 			rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
718 				  ret, portid);
719 
720 		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
721 						       &nb_txd);
722 		if (ret < 0)
723 			rte_exit(EXIT_FAILURE,
724 				 "Cannot adjust number of descriptors: err=%d, port=%d\n",
725 				 ret, portid);
726 
727 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
728 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
729 		printf(", ");
730 
731 		/* init one RX queue */
732 		queueid = 0;
733 		printf("rxq=%hu ", queueid);
734 		fflush(stdout);
735 		rxq_conf = dev_info.default_rxconf;
736 		rxq_conf.offloads = local_port_conf.rxmode.offloads;
737 		ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
738 					     rte_eth_dev_socket_id(portid),
739 					     &rxq_conf,
740 					     packet_pool);
741 		if (ret < 0)
742 			rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, port=%d\n",
743 				  ret, portid);
744 
745 		/* init one TX queue per couple (lcore,port) */
746 		queueid = 0;
747 
748 		RTE_LCORE_FOREACH(lcore_id) {
749 			if (rte_lcore_is_enabled(lcore_id) == 0)
750 				continue;
751 			printf("txq=%u,%hu ", lcore_id, queueid);
752 			fflush(stdout);
753 
754 			txconf = &dev_info.default_txconf;
755 			txconf->offloads = local_port_conf.txmode.offloads;
756 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
757 						     rte_lcore_to_socket_id(lcore_id), txconf);
758 			if (ret < 0)
759 				rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
760 					  "port=%d\n", ret, portid);
761 
762 			qconf = &lcore_queue_conf[lcore_id];
763 			qconf->tx_queue_id[portid] = queueid;
764 			queueid++;
765 		}
766 		rte_eth_allmulticast_enable(portid);
767 		/* Start device */
768 		ret = rte_eth_dev_start(portid);
769 		if (ret < 0)
770 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
771 				  ret, portid);
772 
773 		printf("done:\n");
774 	}
775 
776 	check_all_ports_link_status(enabled_port_mask);
777 
778 	/* initialize the multicast hash */
779 	int retval = init_mcast_hash();
780 	if (retval != 0)
781 		rte_exit(EXIT_FAILURE, "Cannot build the multicast hash\n");
782 
783 	/* launch per-lcore init on every lcore */
784 	rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
785 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
786 		if (rte_eal_wait_lcore(lcore_id) < 0)
787 			return -1;
788 	}
789 
790 	return 0;
791 }
792