xref: /dpdk/examples/ipv4_multicast/main.c (revision 7e06c0de1952d3109a5b0c4779d7e7d8059c9d78)
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_cycles.h>
24 #include <rte_prefetch.h>
25 #include <rte_lcore.h>
26 #include <rte_per_lcore.h>
27 #include <rte_branch_prediction.h>
28 #include <rte_interrupts.h>
29 #include <rte_random.h>
30 #include <rte_debug.h>
31 #include <rte_ether.h>
32 #include <rte_ethdev.h>
33 #include <rte_mempool.h>
34 #include <rte_mbuf.h>
35 #include <rte_malloc.h>
36 #include <rte_fbk_hash.h>
37 #include <rte_ip.h>
38 
39 #define RTE_LOGTYPE_IPv4_MULTICAST RTE_LOGTYPE_USER1
40 
41 #define MAX_PORTS 16
42 
43 #define	MCAST_CLONE_PORTS	2
44 #define	MCAST_CLONE_SEGS	2
45 
46 #define	PKT_MBUF_DATA_SIZE	RTE_MBUF_DEFAULT_BUF_SIZE
47 #define	NB_PKT_MBUF	8192
48 
49 #define	HDR_MBUF_DATA_SIZE	(2 * RTE_PKTMBUF_HEADROOM)
50 #define	NB_HDR_MBUF	(NB_PKT_MBUF * MAX_PORTS)
51 
52 #define	NB_CLONE_MBUF	(NB_PKT_MBUF * MCAST_CLONE_PORTS * MCAST_CLONE_SEGS * 2)
53 
54 /* allow max jumbo frame 9.5 KB */
55 #define	JUMBO_FRAME_MAX_SIZE	0x2600
56 
57 #define MAX_PKT_BURST 32
58 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
59 
60 /* Configure how many packets ahead to prefetch, when reading packets */
61 #define PREFETCH_OFFSET	3
62 
63 /*
64  * Construct Ethernet multicast address from IPv4 multicast address.
65  * Citing RFC 1112, section 6.4:
66  * "An IP host group address is mapped to an Ethernet multicast address
67  * by placing the low-order 23-bits of the IP address into the low-order
68  * 23 bits of the Ethernet multicast address 01-00-5E-00-00-00 (hex)."
69  */
70 
71 /* Construct Ethernet multicast address from IPv4 multicast Address. 8< */
72 #define	ETHER_ADDR_FOR_IPV4_MCAST(x)	\
73 	(rte_cpu_to_be_64(0x01005e000000ULL | ((x) & 0x7fffff)) >> 16)
74 /* >8 End of Construction of multicast address from IPv4 multicast address. */
75 
76 /*
77  * Configurable number of RX/TX ring descriptors
78  */
79 #define RX_DESC_DEFAULT 1024
80 #define TX_DESC_DEFAULT 1024
81 static uint16_t nb_rxd = RX_DESC_DEFAULT;
82 static uint16_t nb_txd = TX_DESC_DEFAULT;
83 
84 /* ethernet addresses of ports */
85 static struct rte_ether_addr ports_eth_addr[MAX_PORTS];
86 
87 /* mask of enabled ports */
88 static uint32_t enabled_port_mask = 0;
89 
90 static uint16_t nb_ports;
91 
92 static int rx_queue_per_lcore = 1;
93 
94 struct mbuf_table {
95 	uint16_t len;
96 	struct rte_mbuf *m_table[MAX_PKT_BURST];
97 };
98 
99 #define MAX_RX_QUEUE_PER_LCORE 16
100 #define MAX_TX_QUEUE_PER_PORT 16
101 struct __rte_cache_aligned lcore_queue_conf {
102 	uint64_t tx_tsc;
103 	uint16_t n_rx_queue;
104 	uint8_t rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
105 	uint16_t tx_queue_id[MAX_PORTS];
106 	struct mbuf_table tx_mbufs[MAX_PORTS];
107 };
108 static struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
109 
110 static struct rte_eth_conf port_conf = {
111 	.rxmode = {
112 		.mtu = JUMBO_FRAME_MAX_SIZE - RTE_ETHER_HDR_LEN -
113 			RTE_ETHER_CRC_LEN,
114 	},
115 	.txmode = {
116 		.mq_mode = RTE_ETH_MQ_TX_NONE,
117 		.offloads = RTE_ETH_TX_OFFLOAD_MULTI_SEGS,
118 	},
119 };
120 
121 static struct rte_mempool *packet_pool, *header_pool, *clone_pool;
122 
123 
124 /* Multicast */
125 static struct rte_fbk_hash_params mcast_hash_params = {
126 	.name = "MCAST_HASH",
127 	.entries = 1024,
128 	.entries_per_bucket = 4,
129 	.socket_id = 0,
130 	.hash_func = NULL,
131 	.init_val = 0,
132 };
133 
134 struct rte_fbk_hash_table *mcast_hash = NULL;
135 
136 struct mcast_group_params {
137 	uint32_t ip;
138 	uint16_t port_mask;
139 };
140 
141 static struct mcast_group_params mcast_group_table[] = {
142 		{RTE_IPV4(224,0,0,101), 0x1},
143 		{RTE_IPV4(224,0,0,102), 0x2},
144 		{RTE_IPV4(224,0,0,103), 0x3},
145 		{RTE_IPV4(224,0,0,104), 0x4},
146 		{RTE_IPV4(224,0,0,105), 0x5},
147 		{RTE_IPV4(224,0,0,106), 0x6},
148 		{RTE_IPV4(224,0,0,107), 0x7},
149 		{RTE_IPV4(224,0,0,108), 0x8},
150 		{RTE_IPV4(224,0,0,109), 0x9},
151 		{RTE_IPV4(224,0,0,110), 0xA},
152 		{RTE_IPV4(224,0,0,111), 0xB},
153 		{RTE_IPV4(224,0,0,112), 0xC},
154 		{RTE_IPV4(224,0,0,113), 0xD},
155 		{RTE_IPV4(224,0,0,114), 0xE},
156 		{RTE_IPV4(224,0,0,115), 0xF},
157 };
158 
159 /* Send burst of packets on an output interface */
160 static void
send_burst(struct lcore_queue_conf * qconf,uint16_t port)161 send_burst(struct lcore_queue_conf *qconf, uint16_t port)
162 {
163 	struct rte_mbuf **m_table;
164 	uint16_t n, queueid;
165 	int ret;
166 
167 	queueid = qconf->tx_queue_id[port];
168 	m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
169 	n = qconf->tx_mbufs[port].len;
170 
171 	ret = rte_eth_tx_burst(port, queueid, m_table, n);
172 	while (unlikely (ret < n)) {
173 		rte_pktmbuf_free(m_table[ret]);
174 		ret++;
175 	}
176 
177 	qconf->tx_mbufs[port].len = 0;
178 }
179 
180 /* Get number of bits set. 8< */
181 static inline uint32_t
bitcnt(uint32_t v)182 bitcnt(uint32_t v)
183 {
184 	uint32_t n;
185 
186 	for (n = 0; v != 0; v &= v - 1, n++)
187 		;
188 
189 	return n;
190 }
191 /* >8 End of getting number of bits set. */
192 
193 /**
194  * Create the output multicast packet based on the given input packet.
195  * There are two approaches for creating outgoing packet, though both
196  * are based on data zero-copy idea, they differ in few details:
197  * First one creates a clone of the input packet, e.g - walk though all
198  * segments of the input packet, and for each of them create a new packet
199  * mbuf and attach that new mbuf to the segment (refer to rte_pktmbuf_clone()
200  * for more details). Then new mbuf is allocated for the packet header
201  * and is prepended to the 'clone' mbuf.
202  * Second approach doesn't make a clone, it just increment refcnt for all
203  * input packet segments. Then it allocates new mbuf for the packet header
204  * and prepends it to the input packet.
205  * Basically first approach reuses only input packet's data, but creates
206  * it's own copy of packet's metadata. Second approach reuses both input's
207  * packet data and metadata.
208  * The advantage of first approach - is that each outgoing packet has it's
209  * own copy of metadata, so we can safely modify data pointer of the
210  * input packet. That allows us to skip creation if the output packet for
211  * the last destination port, but instead modify input packet's header inplace,
212  * e.g: for N destination ports we need to invoke mcast_out_pkt (N-1) times.
213  * The advantage of second approach - less work for each outgoing packet,
214  * e.g: we skip "clone" operation completely. Though it comes with a price -
215  * input packet's metadata has to be intact. So for N destination ports we
216  * need to invoke mcast_out_pkt N times.
217  * So for small number of outgoing ports (and segments in the input packet)
218  * first approach will be faster.
219  * As number of outgoing ports (and/or input segments) will grow,
220  * second way will become more preferable.
221  *
222  *  @param pkt
223  *  Input packet mbuf.
224  *  @param use_clone
225  *  Control which of the two approaches described above should be used:
226  *  - 0 - use second approach:
227  *    Don't "clone" input packet.
228  *    Prepend new header directly to the input packet
229  *  - 1 - use first approach:
230  *    Make a "clone" of input packet first.
231  *    Prepend new header to the clone of the input packet
232  *  @return
233  *  - The pointer to the new outgoing packet.
234  *  - NULL if operation failed.
235  */
236 
237 /* mcast_out_pkt 8< */
238 static inline struct rte_mbuf *
mcast_out_pkt(struct rte_mbuf * pkt,int use_clone)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 /* >8 End of mcast_out_kt. */
265 
266 /*
267  * Write new Ethernet header to the outgoing packet,
268  * and put it into the outgoing queue for the given port.
269  */
270 
271 /* Write new Ethernet header to outgoing packets. 8< */
272 static inline void
mcast_send_pkt(struct rte_mbuf * pkt,struct rte_ether_addr * dest_addr,struct lcore_queue_conf * qconf,uint16_t port)273 mcast_send_pkt(struct rte_mbuf *pkt, struct rte_ether_addr *dest_addr,
274 		struct lcore_queue_conf *qconf, uint16_t port)
275 {
276 	struct rte_ether_hdr *ethdr;
277 	uint16_t len;
278 
279 	/* Construct Ethernet header. */
280 	ethdr = (struct rte_ether_hdr *)
281 		rte_pktmbuf_prepend(pkt, (uint16_t)sizeof(*ethdr));
282 	RTE_ASSERT(ethdr != NULL);
283 
284 	rte_ether_addr_copy(dest_addr, &ethdr->dst_addr);
285 	rte_ether_addr_copy(&ports_eth_addr[port], &ethdr->src_addr);
286 	ethdr->ether_type = rte_be_to_cpu_16(RTE_ETHER_TYPE_IPV4);
287 
288 	/* Put new packet into the output queue */
289 	len = qconf->tx_mbufs[port].len;
290 	qconf->tx_mbufs[port].m_table[len] = pkt;
291 	qconf->tx_mbufs[port].len = ++len;
292 
293 	/* Transmit packets */
294 	if (unlikely(MAX_PKT_BURST == len))
295 		send_burst(qconf, port);
296 }
297 /* >8 End of writing new Ethernet headers. */
298 
299 /* Multicast forward of the input packet */
300 static inline void
mcast_forward(struct rte_mbuf * m,struct lcore_queue_conf * qconf)301 mcast_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf)
302 {
303 	struct rte_mbuf *mc;
304 	struct rte_ipv4_hdr *iphdr;
305 	uint32_t dest_addr, port_mask, port_num, use_clone;
306 	int32_t hash;
307 	uint16_t port;
308 	union {
309 		uint64_t as_int;
310 		struct rte_ether_addr as_addr;
311 	} dst_eth_addr;
312 
313 	/* Remove the Ethernet header from the input packet. 8< */
314 	iphdr = (struct rte_ipv4_hdr *)
315 		rte_pktmbuf_adj(m, (uint16_t)sizeof(struct rte_ether_hdr));
316 	RTE_ASSERT(iphdr != NULL);
317 
318 	dest_addr = rte_be_to_cpu_32(iphdr->dst_addr);
319 	/* >8 End of removing the Ethernet header from the input packet. */
320 
321 	/*
322 	 * Check that it is a valid multicast address and
323 	 * we have some active ports assigned to it.
324 	 */
325 
326 	/* Check valid multicast address. 8< */
327 	if (!RTE_IS_IPV4_MCAST(dest_addr) ||
328 	    (hash = rte_fbk_hash_lookup(mcast_hash, dest_addr)) <= 0 ||
329 	    (port_mask = hash & enabled_port_mask) == 0) {
330 		rte_pktmbuf_free(m);
331 		return;
332 	}
333 	/* >8 End of valid multicast address check. */
334 
335 	/* Calculate number of destination ports. */
336 	port_num = bitcnt(port_mask);
337 
338 	/* Should we use rte_pktmbuf_clone() or not. 8< */
339 	use_clone = (port_num <= MCAST_CLONE_PORTS &&
340 	    m->nb_segs <= MCAST_CLONE_SEGS);
341 	/* >8 End of using rte_pktmbuf_clone(). */
342 
343 	/* Mark all packet's segments as referenced port_num times */
344 	if (use_clone == 0)
345 		rte_pktmbuf_refcnt_update(m, (uint16_t)port_num);
346 
347 	/* Construct destination ethernet address. 8< */
348 	dst_eth_addr.as_int = ETHER_ADDR_FOR_IPV4_MCAST(dest_addr);
349 	/* >8 End of constructing destination ethernet address. */
350 
351 	/* Packets dispatched to destination ports. 8< */
352 	for (port = 0; use_clone != port_mask; port_mask >>= 1, port++) {
353 
354 		/* Prepare output packet and send it out. */
355 		if ((port_mask & 1) != 0) {
356 			if (likely ((mc = mcast_out_pkt(m, use_clone)) != NULL))
357 				mcast_send_pkt(mc, &dst_eth_addr.as_addr,
358 						qconf, port);
359 			else if (use_clone == 0)
360 				rte_pktmbuf_free(m);
361 		}
362 	}
363 	/* >8 End of packets dispatched to destination ports. */
364 
365 	/*
366 	 * If we making clone packets, then, for the last destination port,
367 	 * we can overwrite input packet's metadata.
368 	 */
369 	if (use_clone != 0)
370 		mcast_send_pkt(m, &dst_eth_addr.as_addr, qconf, port);
371 	else
372 		rte_pktmbuf_free(m);
373 }
374 
375 /* Send burst of outgoing packet, if timeout expires. */
376 static inline void
send_timeout_burst(struct lcore_queue_conf * qconf)377 send_timeout_burst(struct lcore_queue_conf *qconf)
378 {
379 	uint64_t cur_tsc;
380 	uint16_t portid;
381 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
382 
383 	cur_tsc = rte_rdtsc();
384 	if (likely (cur_tsc < qconf->tx_tsc + drain_tsc))
385 		return;
386 
387 	for (portid = 0; portid < MAX_PORTS; portid++) {
388 		if (qconf->tx_mbufs[portid].len != 0)
389 			send_burst(qconf, portid);
390 	}
391 	qconf->tx_tsc = cur_tsc;
392 }
393 
394 /* main processing loop */
395 static int
main_loop(__rte_unused void * dummy)396 main_loop(__rte_unused void *dummy)
397 {
398 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
399 	unsigned lcore_id;
400 	int i, j, nb_rx;
401 	uint16_t portid;
402 	struct lcore_queue_conf *qconf;
403 
404 	lcore_id = rte_lcore_id();
405 	qconf = &lcore_queue_conf[lcore_id];
406 
407 
408 	if (qconf->n_rx_queue == 0) {
409 		RTE_LOG(INFO, IPv4_MULTICAST, "lcore %u has nothing to do\n",
410 		    lcore_id);
411 		return 0;
412 	}
413 
414 	RTE_LOG(INFO, IPv4_MULTICAST, "entering main loop on lcore %u\n",
415 	    lcore_id);
416 
417 	for (i = 0; i < qconf->n_rx_queue; i++) {
418 
419 		portid = qconf->rx_queue_list[i];
420 		RTE_LOG(INFO, IPv4_MULTICAST, " -- lcoreid=%u portid=%d\n",
421 		    lcore_id, portid);
422 	}
423 
424 	while (1) {
425 
426 		/*
427 		 * Read packet from RX queues
428 		 */
429 		for (i = 0; i < qconf->n_rx_queue; i++) {
430 
431 			portid = qconf->rx_queue_list[i];
432 			nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
433 						 MAX_PKT_BURST);
434 
435 			/* Prefetch first packets */
436 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
437 				rte_prefetch0(rte_pktmbuf_mtod(
438 						pkts_burst[j], void *));
439 			}
440 
441 			/* Prefetch and forward already prefetched packets */
442 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
443 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
444 						j + PREFETCH_OFFSET], void *));
445 				mcast_forward(pkts_burst[j], qconf);
446 			}
447 
448 			/* Forward remaining prefetched packets */
449 			for (; j < nb_rx; j++) {
450 				mcast_forward(pkts_burst[j], qconf);
451 			}
452 		}
453 
454 		/* Send out packets from TX queues */
455 		send_timeout_burst(qconf);
456 	}
457 }
458 
459 /* display usage */
460 static void
print_usage(const char * prgname)461 print_usage(const char *prgname)
462 {
463 	printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
464 	    "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
465 	    "  -q NQ: number of queue (=ports) per lcore (default is 1)\n",
466 	    prgname);
467 }
468 
469 static uint32_t
parse_portmask(const char * portmask)470 parse_portmask(const char *portmask)
471 {
472 	char *end = NULL;
473 	unsigned long pm;
474 
475 	/* parse hexadecimal string */
476 	pm = strtoul(portmask, &end, 16);
477 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
478 		return 0;
479 
480 	return (uint32_t)pm;
481 }
482 
483 static int
parse_nqueue(const char * q_arg)484 parse_nqueue(const char *q_arg)
485 {
486 	char *end = NULL;
487 	unsigned long n;
488 
489 	/* parse numerical string */
490 	errno = 0;
491 	n = strtoul(q_arg, &end, 0);
492 	if (errno != 0 || end == NULL || *end != '\0' ||
493 			n == 0 || n >= MAX_RX_QUEUE_PER_LCORE)
494 		return -1;
495 
496 	return n;
497 }
498 
499 /* Parse the argument given in the command line of the application */
500 static int
parse_args(int argc,char ** argv)501 parse_args(int argc, char **argv)
502 {
503 	int opt, ret;
504 	char **argvopt;
505 	int option_index;
506 	char *prgname = argv[0];
507 	static struct option lgopts[] = {
508 		{NULL, 0, 0, 0}
509 	};
510 
511 	argvopt = argv;
512 
513 	while ((opt = getopt_long(argc, argvopt, "p:q:",
514 				  lgopts, &option_index)) != EOF) {
515 
516 		switch (opt) {
517 		/* portmask */
518 		case 'p':
519 			enabled_port_mask = parse_portmask(optarg);
520 			if (enabled_port_mask == 0) {
521 				printf("invalid portmask\n");
522 				print_usage(prgname);
523 				return -1;
524 			}
525 			break;
526 
527 		/* nqueue */
528 		case 'q':
529 			rx_queue_per_lcore = parse_nqueue(optarg);
530 			if (rx_queue_per_lcore < 0) {
531 				printf("invalid queue number\n");
532 				print_usage(prgname);
533 				return -1;
534 			}
535 			break;
536 
537 		default:
538 			print_usage(prgname);
539 			return -1;
540 		}
541 	}
542 
543 	if (optind >= 0)
544 		argv[optind-1] = prgname;
545 
546 	ret = optind-1;
547 	optind = 1; /* reset getopt lib */
548 	return ret;
549 }
550 
551 static void
print_ethaddr(const char * name,struct rte_ether_addr * eth_addr)552 print_ethaddr(const char *name, struct rte_ether_addr *eth_addr)
553 {
554 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
555 	rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
556 	printf("%s%s", name, buf);
557 }
558 
559 /* Hash object is created and loaded. 8< */
560 static int
init_mcast_hash(void)561 init_mcast_hash(void)
562 {
563 	uint32_t i;
564 
565 	mcast_hash_params.socket_id = rte_socket_id();
566 	mcast_hash = rte_fbk_hash_create(&mcast_hash_params);
567 	if (mcast_hash == NULL){
568 		return -1;
569 	}
570 
571 	for (i = 0; i < RTE_DIM(mcast_group_table); i++) {
572 		if (rte_fbk_hash_add_key(mcast_hash,
573 			mcast_group_table[i].ip,
574 			mcast_group_table[i].port_mask) < 0) {
575 			return -1;
576 		}
577 	}
578 
579 	return 0;
580 }
581 /* >8 End of hash object is created and loaded. */
582 
583 /* Check the link status of all ports in up to 9s, and print them finally */
584 static void
check_all_ports_link_status(uint32_t port_mask)585 check_all_ports_link_status(uint32_t port_mask)
586 {
587 #define CHECK_INTERVAL 100 /* 100ms */
588 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
589 	uint16_t portid;
590 	uint8_t count, all_ports_up, print_flag = 0;
591 	struct rte_eth_link link;
592 	int ret;
593 	char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
594 
595 	printf("\nChecking link status");
596 	fflush(stdout);
597 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
598 		all_ports_up = 1;
599 		RTE_ETH_FOREACH_DEV(portid) {
600 			if ((port_mask & (1 << portid)) == 0)
601 				continue;
602 			memset(&link, 0, sizeof(link));
603 			ret = rte_eth_link_get_nowait(portid, &link);
604 			if (ret < 0) {
605 				all_ports_up = 0;
606 				if (print_flag == 1)
607 					printf("Port %u link get failed: %s\n",
608 						portid, rte_strerror(-ret));
609 				continue;
610 			}
611 			/* print link status if flag set */
612 			if (print_flag == 1) {
613 				rte_eth_link_to_str(link_status_text,
614 					sizeof(link_status_text),
615 					&link);
616 				printf("Port %d %s\n", portid,
617 				       link_status_text);
618 				continue;
619 			}
620 			/* clear all_ports_up flag if any link down */
621 			if (link.link_status == RTE_ETH_LINK_DOWN) {
622 				all_ports_up = 0;
623 				break;
624 			}
625 		}
626 		/* after finally printing all link status, get out */
627 		if (print_flag == 1)
628 			break;
629 
630 		if (all_ports_up == 0) {
631 			printf(".");
632 			fflush(stdout);
633 			rte_delay_ms(CHECK_INTERVAL);
634 		}
635 
636 		/* set the print_flag if all ports up or timeout */
637 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
638 			print_flag = 1;
639 			printf("done\n");
640 		}
641 	}
642 }
643 
644 int
main(int argc,char ** argv)645 main(int argc, char **argv)
646 {
647 	struct lcore_queue_conf *qconf;
648 	struct rte_eth_dev_info dev_info;
649 	struct rte_eth_txconf *txconf;
650 	int ret;
651 	uint16_t queueid;
652 	unsigned lcore_id = 0, rx_lcore_id = 0;
653 	uint32_t n_tx_queue, nb_lcores;
654 	uint16_t portid;
655 
656 	/* init EAL */
657 	ret = rte_eal_init(argc, argv);
658 	if (ret < 0)
659 		rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
660 	argc -= ret;
661 	argv += ret;
662 
663 	/* parse application arguments (after the EAL ones) */
664 	ret = parse_args(argc, argv);
665 	if (ret < 0)
666 		rte_exit(EXIT_FAILURE, "Invalid IPV4_MULTICAST parameters\n");
667 
668 	/* Create the mbuf pools. 8< */
669 	packet_pool = rte_pktmbuf_pool_create("packet_pool", NB_PKT_MBUF, 32,
670 		0, PKT_MBUF_DATA_SIZE, rte_socket_id());
671 
672 	if (packet_pool == NULL)
673 		rte_exit(EXIT_FAILURE, "Cannot init packet mbuf pool\n");
674 
675 	header_pool = rte_pktmbuf_pool_create("header_pool", NB_HDR_MBUF, 32,
676 		0, HDR_MBUF_DATA_SIZE, rte_socket_id());
677 
678 	if (header_pool == NULL)
679 		rte_exit(EXIT_FAILURE, "Cannot init header mbuf pool\n");
680 
681 	clone_pool = rte_pktmbuf_pool_create("clone_pool", NB_CLONE_MBUF, 32,
682 		0, 0, rte_socket_id());
683 
684 	if (clone_pool == NULL)
685 		rte_exit(EXIT_FAILURE, "Cannot init clone mbuf pool\n");
686 	/* >8 End of create mbuf pools. */
687 
688 	nb_ports = rte_eth_dev_count_avail();
689 	if (nb_ports == 0)
690 		rte_exit(EXIT_FAILURE, "No physical ports!\n");
691 	if (nb_ports > MAX_PORTS)
692 		nb_ports = MAX_PORTS;
693 
694 	nb_lcores = rte_lcore_count();
695 
696 	/* initialize all ports */
697 	RTE_ETH_FOREACH_DEV(portid) {
698 		struct rte_eth_rxconf rxq_conf;
699 		struct rte_eth_conf local_port_conf = port_conf;
700 
701 		/* skip ports that are not enabled */
702 		if ((enabled_port_mask & (1 << portid)) == 0) {
703 			printf("Skipping disabled port %d\n", portid);
704 			continue;
705 		}
706 
707 		qconf = &lcore_queue_conf[rx_lcore_id];
708 
709 		/* limit the frame size to the maximum supported by NIC */
710 		ret = rte_eth_dev_info_get(portid, &dev_info);
711 		if (ret != 0)
712 			rte_exit(EXIT_FAILURE,
713 				"Error during getting device (port %u) info: %s\n",
714 				portid, strerror(-ret));
715 
716 		local_port_conf.rxmode.mtu = RTE_MIN(
717 		    dev_info.max_mtu,
718 		    local_port_conf.rxmode.mtu);
719 
720 		/* get the lcore_id for this port */
721 		while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
722 		       qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
723 
724 			rx_lcore_id ++;
725 			qconf = &lcore_queue_conf[rx_lcore_id];
726 
727 			if (rx_lcore_id >= RTE_MAX_LCORE)
728 				rte_exit(EXIT_FAILURE, "Not enough cores\n");
729 		}
730 		qconf->rx_queue_list[qconf->n_rx_queue] = portid;
731 		qconf->n_rx_queue++;
732 
733 		/* init port */
734 		printf("Initializing port %d on lcore %u... ", portid,
735 		       rx_lcore_id);
736 		fflush(stdout);
737 
738 		n_tx_queue = nb_lcores;
739 		if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
740 			n_tx_queue = MAX_TX_QUEUE_PER_PORT;
741 
742 		ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
743 					    &local_port_conf);
744 		if (ret < 0)
745 			rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
746 				  ret, portid);
747 
748 		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
749 						       &nb_txd);
750 		if (ret < 0)
751 			rte_exit(EXIT_FAILURE,
752 				 "Cannot adjust number of descriptors: err=%d, port=%d\n",
753 				 ret, portid);
754 
755 		ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
756 		if (ret < 0)
757 			rte_exit(EXIT_FAILURE,
758 				 "Cannot get MAC address: err=%d, port=%d\n",
759 				 ret, portid);
760 
761 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
762 		printf(", ");
763 
764 		/* init one RX queue */
765 		queueid = 0;
766 		printf("rxq=%hu ", queueid);
767 		fflush(stdout);
768 		rxq_conf = dev_info.default_rxconf;
769 		rxq_conf.offloads = local_port_conf.rxmode.offloads;
770 		ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
771 					     rte_eth_dev_socket_id(portid),
772 					     &rxq_conf,
773 					     packet_pool);
774 		if (ret < 0)
775 			rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, port=%d\n",
776 				  ret, portid);
777 
778 		/* init one TX queue per couple (lcore,port) */
779 		queueid = 0;
780 
781 		RTE_LCORE_FOREACH(lcore_id) {
782 			if (rte_lcore_is_enabled(lcore_id) == 0)
783 				continue;
784 			printf("txq=%u,%hu ", lcore_id, queueid);
785 			fflush(stdout);
786 
787 			txconf = &dev_info.default_txconf;
788 			txconf->offloads = local_port_conf.txmode.offloads;
789 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
790 						     rte_lcore_to_socket_id(lcore_id), txconf);
791 			if (ret < 0)
792 				rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
793 					  "port=%d\n", ret, portid);
794 
795 			qconf = &lcore_queue_conf[lcore_id];
796 			qconf->tx_queue_id[portid] = queueid;
797 			queueid++;
798 		}
799 		ret = rte_eth_allmulticast_enable(portid);
800 		if (ret < 0)
801 			rte_exit(EXIT_FAILURE,
802 				"rte_eth_allmulticast_enable: err=%d, port=%d\n",
803 				ret, portid);
804 		/* Start device */
805 		ret = rte_eth_dev_start(portid);
806 		if (ret < 0)
807 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
808 				  ret, portid);
809 
810 		printf("done:\n");
811 	}
812 
813 	check_all_ports_link_status(enabled_port_mask);
814 
815 	/* initialize the multicast hash */
816 	int retval = init_mcast_hash();
817 	if (retval != 0)
818 		rte_exit(EXIT_FAILURE, "Cannot build the multicast hash\n");
819 
820 	/* launch per-lcore init on every lcore */
821 	rte_eal_mp_remote_launch(main_loop, NULL, CALL_MAIN);
822 	RTE_LCORE_FOREACH_WORKER(lcore_id) {
823 		if (rte_eal_wait_lcore(lcore_id) < 0)
824 			return -1;
825 	}
826 
827 	/* clean up the EAL */
828 	rte_eal_cleanup();
829 
830 	return 0;
831 }
832