xref: /dpdk/examples/ipv4_multicast/main.c (revision dada9ef6edc59015b6674b5a95258787c71401b0)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2012 Intel Corporation. All rights reserved.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  *
33  */
34 
35 #include <stdio.h>
36 #include <stdlib.h>
37 #include <stdint.h>
38 #include <inttypes.h>
39 #include <sys/types.h>
40 #include <string.h>
41 #include <sys/queue.h>
42 #include <stdarg.h>
43 #include <errno.h>
44 #include <getopt.h>
45 
46 #include <rte_common.h>
47 #include <rte_byteorder.h>
48 #include <rte_log.h>
49 #include <rte_tailq.h>
50 #include <rte_memory.h>
51 #include <rte_memcpy.h>
52 #include <rte_memzone.h>
53 #include <rte_eal.h>
54 #include <rte_per_lcore.h>
55 #include <rte_launch.h>
56 #include <rte_atomic.h>
57 #include <rte_cycles.h>
58 #include <rte_prefetch.h>
59 #include <rte_lcore.h>
60 #include <rte_per_lcore.h>
61 #include <rte_branch_prediction.h>
62 #include <rte_interrupts.h>
63 #include <rte_pci.h>
64 #include <rte_random.h>
65 #include <rte_debug.h>
66 #include <rte_ether.h>
67 #include <rte_ethdev.h>
68 #include <rte_ring.h>
69 #include <rte_mempool.h>
70 #include <rte_mbuf.h>
71 #include <rte_malloc.h>
72 #include <rte_hash_crc.h>
73 #include <rte_fbk_hash.h>
74 #include <rte_ip.h>
75 
76 #include "main.h"
77 
78 #define RTE_LOGTYPE_IPv4_MULTICAST RTE_LOGTYPE_USER1
79 
80 #define MAX_PORTS 16
81 
82 #define	MCAST_CLONE_PORTS	2
83 #define	MCAST_CLONE_SEGS	2
84 
85 #define	PKT_MBUF_SIZE	(2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
86 #define	NB_PKT_MBUF	8192
87 
88 #define	HDR_MBUF_SIZE	(sizeof(struct rte_mbuf) + 2 * RTE_PKTMBUF_HEADROOM)
89 #define	NB_HDR_MBUF	(NB_PKT_MBUF * MAX_PORTS)
90 
91 #define	CLONE_MBUF_SIZE	(sizeof(struct rte_mbuf))
92 #define	NB_CLONE_MBUF	(NB_PKT_MBUF * MCAST_CLONE_PORTS * MCAST_CLONE_SEGS * 2)
93 
94 /* allow max jumbo frame 9.5 KB */
95 #define	JUMBO_FRAME_MAX_SIZE	0x2600
96 
97 /*
98  * RX and TX Prefetch, Host, and Write-back threshold values should be
99  * carefully set for optimal performance. Consult the network
100  * controller's datasheet and supporting DPDK documentation for guidance
101  * on how these parameters should be set.
102  */
103 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
104 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
105 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
106 
107 /*
108  * These default values are optimized for use with the Intel(R) 82599 10 GbE
109  * Controller and the DPDK ixgbe PMD. Consider using other values for other
110  * network controllers and/or network drivers.
111  */
112 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
113 #define TX_HTHRESH 0  /**< Default values of TX host threshold reg. */
114 #define TX_WTHRESH 0  /**< Default values of TX write-back threshold reg. */
115 
116 #define MAX_PKT_BURST 32
117 #define BURST_TX_DRAIN 200000ULL /* around 100us at 2 Ghz */
118 
119 #define SOCKET0 0
120 
121 /* Configure how many packets ahead to prefetch, when reading packets */
122 #define PREFETCH_OFFSET	3
123 
124 /*
125  * Construct Ethernet multicast address from IPv4 multicast address.
126  * Citing RFC 1112, section 6.4:
127  * "An IP host group address is mapped to an Ethernet multicast address
128  * by placing the low-order 23-bits of the IP address into the low-order
129  * 23 bits of the Ethernet multicast address 01-00-5E-00-00-00 (hex)."
130  */
131 #define	ETHER_ADDR_FOR_IPV4_MCAST(x)	\
132 	(rte_cpu_to_be_64(0x01005e000000ULL | ((x) & 0x7fffff)) >> 16)
133 
134 /*
135  * Configurable number of RX/TX ring descriptors
136  */
137 #define RTE_TEST_RX_DESC_DEFAULT 128
138 #define RTE_TEST_TX_DESC_DEFAULT 512
139 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
140 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
141 
142 /* ethernet addresses of ports */
143 static struct ether_addr ports_eth_addr[MAX_PORTS];
144 
145 /* mask of enabled ports */
146 static uint32_t enabled_port_mask = 0;
147 
148 static uint8_t nb_ports = 0;
149 
150 static int rx_queue_per_lcore = 1;
151 
152 struct mbuf_table {
153 	uint16_t len;
154 	struct rte_mbuf *m_table[MAX_PKT_BURST];
155 };
156 
157 #define MAX_RX_QUEUE_PER_LCORE 16
158 #define MAX_TX_QUEUE_PER_PORT 16
159 struct lcore_queue_conf {
160 	uint64_t tx_tsc;
161 	uint16_t n_rx_queue;
162 	uint8_t rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
163 	uint16_t tx_queue_id[MAX_PORTS];
164 	struct mbuf_table tx_mbufs[MAX_PORTS];
165 } __rte_cache_aligned;
166 static struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
167 
168 static const struct rte_eth_conf port_conf = {
169 	.rxmode = {
170 		.max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
171 		.split_hdr_size = 0,
172 		.header_split   = 0, /**< Header Split disabled */
173 		.hw_ip_checksum = 0, /**< IP checksum offload disabled */
174 		.hw_vlan_filter = 0, /**< VLAN filtering disabled */
175 		.jumbo_frame    = 1, /**< Jumbo Frame Support enabled */
176 		.hw_strip_crc   = 0, /**< CRC stripped by hardware */
177 	},
178 	.txmode = {
179 	},
180 };
181 
182 static const struct rte_eth_rxconf rx_conf = {
183 	.rx_thresh = {
184 		.pthresh = RX_PTHRESH,
185 		.hthresh = RX_HTHRESH,
186 		.wthresh = RX_WTHRESH,
187 	},
188 };
189 
190 static const struct rte_eth_txconf tx_conf = {
191 	.tx_thresh = {
192 		.pthresh = TX_PTHRESH,
193 		.hthresh = TX_HTHRESH,
194 		.wthresh = TX_WTHRESH,
195 	},
196 	.tx_free_thresh = 0, /* Use PMD default values */
197 	.tx_rs_thresh = 0, /* Use PMD default values */
198 };
199 
200 static struct rte_mempool *packet_pool, *header_pool, *clone_pool;
201 
202 
203 /* Multicast */
204 static struct rte_fbk_hash_params mcast_hash_params = {
205 	.name = "MCAST_HASH",
206 	.entries = 1024,
207 	.entries_per_bucket = 4,
208 	.socket_id = SOCKET0,
209 	.hash_func = NULL,
210 	.init_val = 0,
211 };
212 
213 struct rte_fbk_hash_table *mcast_hash = NULL;
214 
215 struct mcast_group_params {
216 	uint32_t ip;
217 	uint16_t port_mask;
218 };
219 
220 static struct mcast_group_params mcast_group_table[] = {
221 		{IPv4(224,0,0,101), 0x1},
222 		{IPv4(224,0,0,102), 0x2},
223 		{IPv4(224,0,0,103), 0x3},
224 		{IPv4(224,0,0,104), 0x4},
225 		{IPv4(224,0,0,105), 0x5},
226 		{IPv4(224,0,0,106), 0x6},
227 		{IPv4(224,0,0,107), 0x7},
228 		{IPv4(224,0,0,108), 0x8},
229 		{IPv4(224,0,0,109), 0x9},
230 		{IPv4(224,0,0,110), 0xA},
231 		{IPv4(224,0,0,111), 0xB},
232 		{IPv4(224,0,0,112), 0xC},
233 		{IPv4(224,0,0,113), 0xD},
234 		{IPv4(224,0,0,114), 0xE},
235 		{IPv4(224,0,0,115), 0xF},
236 };
237 
238 #define N_MCAST_GROUPS \
239 	(sizeof (mcast_group_table) / sizeof (mcast_group_table[0]))
240 
241 
242 /* Send burst of packets on an output interface */
243 static void
244 send_burst(struct lcore_queue_conf *qconf, uint8_t port)
245 {
246 	struct rte_mbuf **m_table;
247 	uint16_t n, queueid;
248 	int ret;
249 
250 	queueid = qconf->tx_queue_id[port];
251 	m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
252 	n = qconf->tx_mbufs[port].len;
253 
254 	ret = rte_eth_tx_burst(port, queueid, m_table, n);
255 	while (unlikely (ret < n)) {
256 		rte_pktmbuf_free(m_table[ret]);
257 		ret++;
258 	}
259 
260 	qconf->tx_mbufs[port].len = 0;
261 }
262 
263 /* Get number of bits set. */
264 static inline uint32_t
265 bitcnt(uint32_t v)
266 {
267 	uint32_t n;
268 
269 	for (n = 0; v != 0; v &= v - 1, n++)
270 		;
271 
272 	return (n);
273 }
274 
275 /**
276  * Create the output multicast packet based on the given input packet.
277  * There are two approaches for creating outgoing packet, though both
278  * are based on data zero-copy idea, they differ in few details:
279  * First one creates a clone of the input packet, e.g - walk though all
280  * segments of the input packet, and for each of them create a new packet
281  * mbuf and attach that new mbuf to the segment (refer to rte_pktmbuf_clone()
282  * for more details). Then new mbuf is allocated for the packet header
283  * and is prepended to the 'clone' mbuf.
284  * Second approach doesn't make a clone, it just increment refcnt for all
285  * input packet segments. Then it allocates new mbuf for the packet header
286  * and prepends it to the input packet.
287  * Basically first approach reuses only input packet's data, but creates
288  * it's own copy of packet's metadata. Second approach reuses both input's
289  * packet data and metadata.
290  * The advantage of first approach - is that each outgoing packet has it's
291  * own copy of metadata, so we can safely modify data pointer of the
292  * input packet. That allows us to skip creation if the output packet for
293  * the last destination port, but instead modify input packet's header inplace,
294  * e.g: for N destination ports we need to invoke mcast_out_pkt (N-1) times.
295  * The advantage of second approach - less work for each outgoing packet,
296  * e.g: we skip "clone" operation completely. Though it comes with a price -
297  * input packet's metadata has to be intact. So for N destination ports we
298  * need to invoke mcast_out_pkt N times.
299  * So for small number of outgoing ports (and segments in the input packet)
300  * first approach will be faster.
301  * As number of outgoing ports (and/or input segments) will grow,
302  * second way will become more preferable.
303  *
304  *  @param pkt
305  *  Input packet mbuf.
306  *  @param use_clone
307  *  Control which of the two approaches described above should be used:
308  *  - 0 - use second approach:
309  *    Don't "clone" input packet.
310  *    Prepend new header directly to the input packet
311  *  - 1 - use first approach:
312  *    Make a "clone" of input packet first.
313  *    Prepend new header to the clone of the input packet
314  *  @return
315  *  - The pointer to the new outgoing packet.
316  *  - NULL if operation failed.
317  */
318 static inline struct rte_mbuf *
319 mcast_out_pkt(struct rte_mbuf *pkt, int use_clone)
320 {
321 	struct rte_mbuf *hdr;
322 
323 	/* Create new mbuf for the header. */
324 	if (unlikely ((hdr = rte_pktmbuf_alloc(header_pool)) == NULL))
325 		return (NULL);
326 
327 	/* If requested, then make a new clone packet. */
328 	if (use_clone != 0 &&
329 	    unlikely ((pkt = rte_pktmbuf_clone(pkt, clone_pool)) == NULL)) {
330 		rte_pktmbuf_free(hdr);
331 		return (NULL);
332 	}
333 
334 	/* prepend new header */
335 	hdr->pkt.next = pkt;
336 
337 
338 	/* update header's fields */
339 	hdr->pkt.pkt_len = (uint16_t)(hdr->pkt.data_len + pkt->pkt.pkt_len);
340 	hdr->pkt.nb_segs = (uint8_t)(pkt->pkt.nb_segs + 1);
341 
342 	/* copy metadata from source packet*/
343 	hdr->pkt.in_port = pkt->pkt.in_port;
344 	hdr->pkt.vlan_tci = pkt->pkt.vlan_tci;
345 	hdr->pkt.l2_len = pkt->pkt.l2_len;
346 	hdr->pkt.l3_len = pkt->pkt.l3_len;
347 	hdr->pkt.hash = pkt->pkt.hash;
348 
349 	hdr->ol_flags = pkt->ol_flags;
350 
351 	__rte_mbuf_sanity_check(hdr, RTE_MBUF_PKT, 1);
352 	return (hdr);
353 }
354 
355 /*
356  * Write new Ethernet header to the outgoing packet,
357  * and put it into the outgoing queue for the given port.
358  */
359 static inline void
360 mcast_send_pkt(struct rte_mbuf *pkt, struct ether_addr *dest_addr,
361 		struct lcore_queue_conf *qconf, uint8_t port)
362 {
363 	struct ether_hdr *ethdr;
364 	uint16_t len;
365 
366 	/* Construct Ethernet header. */
367 	ethdr = (struct ether_hdr *)rte_pktmbuf_prepend(pkt, (uint16_t)sizeof(*ethdr));
368 	RTE_MBUF_ASSERT(ethdr != NULL);
369 
370 	ether_addr_copy(dest_addr, &ethdr->d_addr);
371 	ether_addr_copy(&ports_eth_addr[port], &ethdr->s_addr);
372 	ethdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
373 
374 	/* Put new packet into the output queue */
375 	len = qconf->tx_mbufs[port].len;
376 	qconf->tx_mbufs[port].m_table[len] = pkt;
377 	qconf->tx_mbufs[port].len = ++len;
378 
379 	/* Transmit packets */
380 	if (unlikely(MAX_PKT_BURST == len))
381 		send_burst(qconf, port);
382 }
383 
384 /* Multicast forward of the input packet */
385 static inline void
386 mcast_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf)
387 {
388 	struct rte_mbuf *mc;
389 	struct ipv4_hdr *iphdr;
390 	uint32_t dest_addr, port_mask, port_num, use_clone;
391 	int32_t hash;
392 	uint8_t port;
393 	union {
394 		uint64_t as_int;
395 		struct ether_addr as_addr;
396 	} dst_eth_addr;
397 
398 	/* Remove the Ethernet header from the input packet */
399 	iphdr = (struct ipv4_hdr *)rte_pktmbuf_adj(m, (uint16_t)sizeof(struct ether_hdr));
400 	RTE_MBUF_ASSERT(iphdr != NULL);
401 
402 	dest_addr = rte_be_to_cpu_32(iphdr->dst_addr);
403 
404 	/*
405 	 * Check that it is a valid multicast address and
406 	 * we have some active ports assigned to it.
407 	 */
408 	if(!IS_IPV4_MCAST(dest_addr) ||
409 	    (hash = rte_fbk_hash_lookup(mcast_hash, dest_addr)) <= 0 ||
410 	    (port_mask = hash & enabled_port_mask) == 0) {
411 		rte_pktmbuf_free(m);
412 		return;
413 	}
414 
415 	/* Calculate number of destination ports. */
416 	port_num = bitcnt(port_mask);
417 
418 	/* Should we use rte_pktmbuf_clone() or not. */
419 	use_clone = (port_num <= MCAST_CLONE_PORTS &&
420 	    m->pkt.nb_segs <= MCAST_CLONE_SEGS);
421 
422 	/* Mark all packet's segments as referenced port_num times */
423 	if (use_clone == 0)
424 		rte_pktmbuf_refcnt_update(m, (uint16_t)port_num);
425 
426 	/* construct destination ethernet address */
427 	dst_eth_addr.as_int = ETHER_ADDR_FOR_IPV4_MCAST(dest_addr);
428 
429 	for (port = 0; use_clone != port_mask; port_mask >>= 1, port++) {
430 
431 		/* Prepare output packet and send it out. */
432 		if ((port_mask & 1) != 0) {
433 			if (likely ((mc = mcast_out_pkt(m, use_clone)) != NULL))
434 				mcast_send_pkt(mc, &dst_eth_addr.as_addr,
435 						qconf, port);
436 			else if (use_clone == 0)
437 				rte_pktmbuf_free(m);
438 		}
439 	}
440 
441 	/*
442 	 * If we making clone packets, then, for the last destination port,
443 	 * we can overwrite input packet's metadata.
444 	 */
445 	if (use_clone != 0)
446 		mcast_send_pkt(m, &dst_eth_addr.as_addr, qconf, port);
447 	else
448 		rte_pktmbuf_free(m);
449 }
450 
451 /* Send burst of outgoing packet, if timeout expires. */
452 static inline void
453 send_timeout_burst(struct lcore_queue_conf *qconf)
454 {
455 	uint64_t cur_tsc;
456 	uint8_t portid;
457 
458 	cur_tsc = rte_rdtsc();
459 	if (likely (cur_tsc < qconf->tx_tsc + BURST_TX_DRAIN))
460 		return;
461 
462 	for (portid = 0; portid < MAX_PORTS; portid++) {
463 		if (qconf->tx_mbufs[portid].len != 0)
464 			send_burst(qconf, portid);
465 	}
466 	qconf->tx_tsc = cur_tsc;
467 }
468 
469 /* main processing loop */
470 static __attribute__((noreturn)) int
471 main_loop(__rte_unused void *dummy)
472 {
473 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
474 	uint32_t lcore_id;
475 	int i, j, nb_rx;
476 	uint8_t portid;
477 	struct lcore_queue_conf *qconf;
478 
479 	lcore_id = rte_lcore_id();
480 	qconf = &lcore_queue_conf[lcore_id];
481 
482 
483 	if (qconf->n_rx_queue == 0) {
484 		RTE_LOG(INFO, IPv4_MULTICAST, "lcore %u has nothing to do\n",
485 		    lcore_id);
486 		while(1);
487 	}
488 
489 	RTE_LOG(INFO, IPv4_MULTICAST, "entering main loop on lcore %u\n",
490 	    lcore_id);
491 
492 	for (i = 0; i < qconf->n_rx_queue; i++) {
493 
494 		portid = qconf->rx_queue_list[i];
495 		RTE_LOG(INFO, IPv4_MULTICAST, " -- lcoreid=%u portid=%d\n",
496 		    lcore_id, (int) portid);
497 	}
498 
499 	while (1) {
500 
501 		/*
502 		 * Read packet from RX queues
503 		 */
504 		for (i = 0; i < qconf->n_rx_queue; i++) {
505 
506 			portid = qconf->rx_queue_list[i];
507 			nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
508 						 MAX_PKT_BURST);
509 
510 			/* Prefetch first packets */
511 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
512 				rte_prefetch0(rte_pktmbuf_mtod(
513 						pkts_burst[j], void *));
514 			}
515 
516 			/* Prefetch and forward already prefetched packets */
517 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
518 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
519 						j + PREFETCH_OFFSET], void *));
520 				mcast_forward(pkts_burst[j], qconf);
521 			}
522 
523 			/* Forward remaining prefetched packets */
524 			for (; j < nb_rx; j++) {
525 				mcast_forward(pkts_burst[j], qconf);
526 			}
527 		}
528 
529 		/* Send out packets from TX queues */
530 		send_timeout_burst(qconf);
531 	}
532 }
533 
534 /* display usage */
535 static void
536 print_usage(const char *prgname)
537 {
538 	printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
539 	    "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
540 	    "  -q NQ: number of queue (=ports) per lcore (default is 1)\n",
541 	    prgname);
542 }
543 
544 static uint32_t
545 parse_portmask(const char *portmask)
546 {
547 	char *end = NULL;
548 	unsigned long pm;
549 
550 	/* parse hexadecimal string */
551 	pm = strtoul(portmask, &end, 16);
552 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
553 		return 0;
554 
555 	return ((uint32_t)pm);
556 }
557 
558 static int
559 parse_nqueue(const char *q_arg)
560 {
561 	char *end = NULL;
562 	unsigned long n;
563 
564 	/* parse numerical string */
565 	errno = 0;
566 	n = strtoul(q_arg, &end, 0);
567 	if (errno != 0 || end == NULL || *end != '\0' ||
568 			n == 0 || n >= MAX_RX_QUEUE_PER_LCORE)
569 		return (-1);
570 
571 	return (n);
572 }
573 
574 /* Parse the argument given in the command line of the application */
575 static int
576 parse_args(int argc, char **argv)
577 {
578 	int opt, ret;
579 	char **argvopt;
580 	int option_index;
581 	char *prgname = argv[0];
582 	static struct option lgopts[] = {
583 		{NULL, 0, 0, 0}
584 	};
585 
586 	argvopt = argv;
587 
588 	while ((opt = getopt_long(argc, argvopt, "p:q:",
589 				  lgopts, &option_index)) != EOF) {
590 
591 		switch (opt) {
592 		/* portmask */
593 		case 'p':
594 			enabled_port_mask = parse_portmask(optarg);
595 			if (enabled_port_mask == 0) {
596 				printf("invalid portmask\n");
597 				print_usage(prgname);
598 				return -1;
599 			}
600 			break;
601 
602 		/* nqueue */
603 		case 'q':
604 			rx_queue_per_lcore = parse_nqueue(optarg);
605 			if (rx_queue_per_lcore < 0) {
606 				printf("invalid queue number\n");
607 				print_usage(prgname);
608 				return -1;
609 			}
610 			break;
611 
612 		default:
613 			print_usage(prgname);
614 			return -1;
615 		}
616 	}
617 
618 	if (optind >= 0)
619 		argv[optind-1] = prgname;
620 
621 	ret = optind-1;
622 	optind = 0; /* reset getopt lib */
623 	return ret;
624 }
625 
626 static void
627 print_ethaddr(const char *name, struct ether_addr *eth_addr)
628 {
629 	printf("%s%02X:%02X:%02X:%02X:%02X:%02X", name,
630 	       eth_addr->addr_bytes[0],
631 	       eth_addr->addr_bytes[1],
632 	       eth_addr->addr_bytes[2],
633 	       eth_addr->addr_bytes[3],
634 	       eth_addr->addr_bytes[4],
635 	       eth_addr->addr_bytes[5]);
636 }
637 
638 static int
639 init_mcast_hash(void)
640 {
641 	uint32_t i;
642 
643 	mcast_hash = rte_fbk_hash_create(&mcast_hash_params);
644 	if (mcast_hash == NULL){
645 		return -1;
646 	}
647 
648 	for (i = 0; i < N_MCAST_GROUPS; i ++){
649 		if (rte_fbk_hash_add_key(mcast_hash,
650 			mcast_group_table[i].ip,
651 			mcast_group_table[i].port_mask) < 0) {
652 			return -1;
653 		}
654 	}
655 
656 	return 0;
657 }
658 
659 int
660 MAIN(int argc, char **argv)
661 {
662 	struct lcore_queue_conf *qconf;
663 	struct rte_eth_link link;
664 	int ret;
665 	uint16_t queueid;
666 	unsigned lcore_id = 0, rx_lcore_id = 0;;
667 	uint32_t n_tx_queue, nb_lcores;
668 	uint8_t portid;
669 
670 	/* init EAL */
671 	ret = rte_eal_init(argc, argv);
672 	if (ret < 0)
673 		rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
674 	argc -= ret;
675 	argv += ret;
676 
677 	/* parse application arguments (after the EAL ones) */
678 	ret = parse_args(argc, argv);
679 	if (ret < 0)
680 		rte_exit(EXIT_FAILURE, "Invalid IPV4_MULTICAST parameters\n");
681 
682 	/* create the mbuf pools */
683 	packet_pool = rte_mempool_create("packet_pool", NB_PKT_MBUF,
684 	    PKT_MBUF_SIZE, 32, sizeof(struct rte_pktmbuf_pool_private),
685 	    rte_pktmbuf_pool_init, NULL, rte_pktmbuf_init, NULL,
686 	    SOCKET0, 0);
687 
688 	if (packet_pool == NULL)
689 		rte_exit(EXIT_FAILURE, "Cannot init packet mbuf pool\n");
690 
691 	header_pool = rte_mempool_create("header_pool", NB_HDR_MBUF,
692 	    HDR_MBUF_SIZE, 32, 0, NULL, NULL, rte_pktmbuf_init, NULL,
693 	    SOCKET0, 0);
694 
695 	if (header_pool == NULL)
696 		rte_exit(EXIT_FAILURE, "Cannot init header mbuf pool\n");
697 
698 	clone_pool = rte_mempool_create("clone_pool", NB_CLONE_MBUF,
699 	    CLONE_MBUF_SIZE, 32, 0, NULL, NULL, rte_pktmbuf_init, NULL,
700 	    SOCKET0, 0);
701 
702 	if (clone_pool == NULL)
703 		rte_exit(EXIT_FAILURE, "Cannot init clone mbuf pool\n");
704 
705 	/* init driver */
706 #ifdef RTE_LIBRTE_IGB_PMD
707 	if (rte_igb_pmd_init() < 0)
708 		rte_exit(EXIT_FAILURE, "Cannot init igb pmd\n");
709 #endif
710 #ifdef RTE_LIBRTE_IXGBE_PMD
711 	if (rte_ixgbe_pmd_init() < 0)
712 		rte_exit(EXIT_FAILURE, "Cannot init ixgbe pmd\n");
713 #endif
714 
715 	if (rte_eal_pci_probe() < 0)
716 		rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");
717 
718 	nb_ports = rte_eth_dev_count();
719 	if (nb_ports == 0)
720 		rte_exit(EXIT_FAILURE, "No physical ports!\n");
721 	if (nb_ports > MAX_PORTS)
722 		nb_ports = MAX_PORTS;
723 
724 	nb_lcores = rte_lcore_count();
725 
726 	/* initialize all ports */
727 	for (portid = 0; portid < nb_ports; portid++) {
728 		/* skip ports that are not enabled */
729 		if ((enabled_port_mask & (1 << portid)) == 0) {
730 			printf("Skipping disabled port %d\n", portid);
731 			continue;
732 		}
733 
734 		qconf = &lcore_queue_conf[rx_lcore_id];
735 
736 		/* get the lcore_id for this port */
737 		while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
738 		       qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
739 
740 			rx_lcore_id ++;
741 			qconf = &lcore_queue_conf[rx_lcore_id];
742 
743 			if (rx_lcore_id >= RTE_MAX_LCORE)
744 				rte_exit(EXIT_FAILURE, "Not enough cores\n");
745 		}
746 		qconf->rx_queue_list[qconf->n_rx_queue] = portid;
747 		qconf->n_rx_queue++;
748 
749 		/* init port */
750 		printf("Initializing port %d on lcore %u... ", portid,
751 		       rx_lcore_id);
752 		fflush(stdout);
753 
754 		n_tx_queue = nb_lcores;
755 		if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
756 			n_tx_queue = MAX_TX_QUEUE_PER_PORT;
757 		ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
758 					    &port_conf);
759 		if (ret < 0)
760 			rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
761 				  ret, portid);
762 
763 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
764 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
765 		printf(", ");
766 
767 		/* init one RX queue */
768 		queueid = 0;
769 		printf("rxq=%hu ", queueid);
770 		fflush(stdout);
771 		ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
772 					     SOCKET0, &rx_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 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
787 						     SOCKET0, &tx_conf);
788 			if (ret < 0)
789 				rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
790 					  "port=%d\n", ret, portid);
791 
792 			qconf = &lcore_queue_conf[lcore_id];
793 			qconf->tx_queue_id[portid] = queueid;
794 			queueid++;
795 		}
796 
797 		/* Start device */
798 		ret = rte_eth_dev_start(portid);
799 		if (ret < 0)
800 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
801 				  ret, portid);
802 
803 		printf("done: ");
804 
805 		/* get link status */
806 		rte_eth_link_get(portid, &link);
807 		if (link.link_status) {
808 			printf(" Link Up - speed %u Mbps - %s\n",
809 			       (uint32_t) link.link_speed,
810 			       (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
811 			       ("full-duplex") : ("half-duplex\n"));
812 			rte_eth_promiscuous_enable(portid);
813 			rte_eth_allmulticast_enable(portid);
814 		} else {
815 			printf(" Link Down\n");
816 		}
817 	}
818 
819 
820 	/* initialize the multicast hash */
821 	int retval = init_mcast_hash();
822 	if (retval != 0)
823 		rte_exit(EXIT_FAILURE, "Cannot build the multicast hash\n");
824 
825 	/* launch per-lcore init on every lcore */
826 	rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
827 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
828 		if (rte_eal_wait_lcore(lcore_id) < 0)
829 			return -1;
830 	}
831 
832 	return 0;
833 }
834