xref: /dpdk/examples/l3fwd/main.c (revision 25d11a86c56d50947af33d0b79ede622809bd8b9)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2016 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 #include <signal.h>
16 #include <stdbool.h>
17 
18 #include <rte_common.h>
19 #include <rte_vect.h>
20 #include <rte_byteorder.h>
21 #include <rte_log.h>
22 #include <rte_memory.h>
23 #include <rte_memcpy.h>
24 #include <rte_eal.h>
25 #include <rte_launch.h>
26 #include <rte_atomic.h>
27 #include <rte_cycles.h>
28 #include <rte_prefetch.h>
29 #include <rte_lcore.h>
30 #include <rte_per_lcore.h>
31 #include <rte_branch_prediction.h>
32 #include <rte_interrupts.h>
33 #include <rte_random.h>
34 #include <rte_debug.h>
35 #include <rte_ether.h>
36 #include <rte_ethdev.h>
37 #include <rte_mempool.h>
38 #include <rte_mbuf.h>
39 #include <rte_ip.h>
40 #include <rte_tcp.h>
41 #include <rte_udp.h>
42 #include <rte_string_fns.h>
43 #include <rte_cpuflags.h>
44 
45 #include <cmdline_parse.h>
46 #include <cmdline_parse_etheraddr.h>
47 
48 #include "l3fwd.h"
49 
50 /*
51  * Configurable number of RX/TX ring descriptors
52  */
53 #define RTE_TEST_RX_DESC_DEFAULT 1024
54 #define RTE_TEST_TX_DESC_DEFAULT 1024
55 
56 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
57 #define MAX_RX_QUEUE_PER_PORT 128
58 
59 #define MAX_LCORE_PARAMS 1024
60 
61 /* Static global variables used within this file. */
62 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
63 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
64 
65 /**< Ports set in promiscuous mode off by default. */
66 static int promiscuous_on;
67 
68 /* Select Longest-Prefix or Exact match. */
69 static int l3fwd_lpm_on;
70 static int l3fwd_em_on;
71 
72 static int numa_on = 1; /**< NUMA is enabled by default. */
73 static int parse_ptype; /**< Parse packet type using rx callback, and */
74 			/**< disabled by default */
75 
76 /* Global variables. */
77 
78 volatile bool force_quit;
79 
80 /* ethernet addresses of ports */
81 uint64_t dest_eth_addr[RTE_MAX_ETHPORTS];
82 struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
83 
84 xmm_t val_eth[RTE_MAX_ETHPORTS];
85 
86 /* mask of enabled ports */
87 uint32_t enabled_port_mask;
88 
89 /* Used only in exact match mode. */
90 int ipv6; /**< ipv6 is false by default. */
91 uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
92 
93 struct lcore_conf lcore_conf[RTE_MAX_LCORE];
94 
95 struct lcore_params {
96 	uint16_t port_id;
97 	uint8_t queue_id;
98 	uint8_t lcore_id;
99 } __rte_cache_aligned;
100 
101 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
102 static struct lcore_params lcore_params_array_default[] = {
103 	{0, 0, 2},
104 	{0, 1, 2},
105 	{0, 2, 2},
106 	{1, 0, 2},
107 	{1, 1, 2},
108 	{1, 2, 2},
109 	{2, 0, 2},
110 	{3, 0, 3},
111 	{3, 1, 3},
112 };
113 
114 static struct lcore_params * lcore_params = lcore_params_array_default;
115 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
116 				sizeof(lcore_params_array_default[0]);
117 
118 static struct rte_eth_conf port_conf = {
119 	.rxmode = {
120 		.mq_mode = ETH_MQ_RX_RSS,
121 		.max_rx_pkt_len = ETHER_MAX_LEN,
122 		.split_hdr_size = 0,
123 		.offloads = DEV_RX_OFFLOAD_CHECKSUM,
124 	},
125 	.rx_adv_conf = {
126 		.rss_conf = {
127 			.rss_key = NULL,
128 			.rss_hf = ETH_RSS_IP,
129 		},
130 	},
131 	.txmode = {
132 		.mq_mode = ETH_MQ_TX_NONE,
133 	},
134 };
135 
136 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
137 
138 struct l3fwd_lkp_mode {
139 	void  (*setup)(int);
140 	int   (*check_ptype)(int);
141 	rte_rx_callback_fn cb_parse_ptype;
142 	int   (*main_loop)(void *);
143 	void* (*get_ipv4_lookup_struct)(int);
144 	void* (*get_ipv6_lookup_struct)(int);
145 };
146 
147 static struct l3fwd_lkp_mode l3fwd_lkp;
148 
149 static struct l3fwd_lkp_mode l3fwd_em_lkp = {
150 	.setup                  = setup_hash,
151 	.check_ptype		= em_check_ptype,
152 	.cb_parse_ptype		= em_cb_parse_ptype,
153 	.main_loop              = em_main_loop,
154 	.get_ipv4_lookup_struct = em_get_ipv4_l3fwd_lookup_struct,
155 	.get_ipv6_lookup_struct = em_get_ipv6_l3fwd_lookup_struct,
156 };
157 
158 static struct l3fwd_lkp_mode l3fwd_lpm_lkp = {
159 	.setup                  = setup_lpm,
160 	.check_ptype		= lpm_check_ptype,
161 	.cb_parse_ptype		= lpm_cb_parse_ptype,
162 	.main_loop              = lpm_main_loop,
163 	.get_ipv4_lookup_struct = lpm_get_ipv4_l3fwd_lookup_struct,
164 	.get_ipv6_lookup_struct = lpm_get_ipv6_l3fwd_lookup_struct,
165 };
166 
167 /*
168  * Setup lookup methods for forwarding.
169  * Currently exact-match and longest-prefix-match
170  * are supported ones.
171  */
172 static void
173 setup_l3fwd_lookup_tables(void)
174 {
175 	/* Setup HASH lookup functions. */
176 	if (l3fwd_em_on)
177 		l3fwd_lkp = l3fwd_em_lkp;
178 	/* Setup LPM lookup functions. */
179 	else
180 		l3fwd_lkp = l3fwd_lpm_lkp;
181 }
182 
183 static int
184 check_lcore_params(void)
185 {
186 	uint8_t queue, lcore;
187 	uint16_t i;
188 	int socketid;
189 
190 	for (i = 0; i < nb_lcore_params; ++i) {
191 		queue = lcore_params[i].queue_id;
192 		if (queue >= MAX_RX_QUEUE_PER_PORT) {
193 			printf("invalid queue number: %hhu\n", queue);
194 			return -1;
195 		}
196 		lcore = lcore_params[i].lcore_id;
197 		if (!rte_lcore_is_enabled(lcore)) {
198 			printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
199 			return -1;
200 		}
201 		if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
202 			(numa_on == 0)) {
203 			printf("warning: lcore %hhu is on socket %d with numa off \n",
204 				lcore, socketid);
205 		}
206 	}
207 	return 0;
208 }
209 
210 static int
211 check_port_config(void)
212 {
213 	uint16_t portid;
214 	uint16_t i;
215 
216 	for (i = 0; i < nb_lcore_params; ++i) {
217 		portid = lcore_params[i].port_id;
218 		if ((enabled_port_mask & (1 << portid)) == 0) {
219 			printf("port %u is not enabled in port mask\n", portid);
220 			return -1;
221 		}
222 		if (!rte_eth_dev_is_valid_port(portid)) {
223 			printf("port %u is not present on the board\n", portid);
224 			return -1;
225 		}
226 	}
227 	return 0;
228 }
229 
230 static uint8_t
231 get_port_n_rx_queues(const uint16_t port)
232 {
233 	int queue = -1;
234 	uint16_t i;
235 
236 	for (i = 0; i < nb_lcore_params; ++i) {
237 		if (lcore_params[i].port_id == port) {
238 			if (lcore_params[i].queue_id == queue+1)
239 				queue = lcore_params[i].queue_id;
240 			else
241 				rte_exit(EXIT_FAILURE, "queue ids of the port %d must be"
242 						" in sequence and must start with 0\n",
243 						lcore_params[i].port_id);
244 		}
245 	}
246 	return (uint8_t)(++queue);
247 }
248 
249 static int
250 init_lcore_rx_queues(void)
251 {
252 	uint16_t i, nb_rx_queue;
253 	uint8_t lcore;
254 
255 	for (i = 0; i < nb_lcore_params; ++i) {
256 		lcore = lcore_params[i].lcore_id;
257 		nb_rx_queue = lcore_conf[lcore].n_rx_queue;
258 		if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
259 			printf("error: too many queues (%u) for lcore: %u\n",
260 				(unsigned)nb_rx_queue + 1, (unsigned)lcore);
261 			return -1;
262 		} else {
263 			lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
264 				lcore_params[i].port_id;
265 			lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
266 				lcore_params[i].queue_id;
267 			lcore_conf[lcore].n_rx_queue++;
268 		}
269 	}
270 	return 0;
271 }
272 
273 /* display usage */
274 static void
275 print_usage(const char *prgname)
276 {
277 	fprintf(stderr, "%s [EAL options] --"
278 		" -p PORTMASK"
279 		" [-P]"
280 		" [-E]"
281 		" [-L]"
282 		" --config (port,queue,lcore)[,(port,queue,lcore)]"
283 		" [--eth-dest=X,MM:MM:MM:MM:MM:MM]"
284 		" [--enable-jumbo [--max-pkt-len PKTLEN]]"
285 		" [--no-numa]"
286 		" [--hash-entry-num]"
287 		" [--ipv6]"
288 		" [--parse-ptype]\n\n"
289 
290 		"  -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
291 		"  -P : Enable promiscuous mode\n"
292 		"  -E : Enable exact match\n"
293 		"  -L : Enable longest prefix match (default)\n"
294 		"  --config (port,queue,lcore): Rx queue configuration\n"
295 		"  --eth-dest=X,MM:MM:MM:MM:MM:MM: Ethernet destination for port X\n"
296 		"  --enable-jumbo: Enable jumbo frames\n"
297 		"  --max-pkt-len: Under the premise of enabling jumbo,\n"
298 		"                 maximum packet length in decimal (64-9600)\n"
299 		"  --no-numa: Disable numa awareness\n"
300 		"  --hash-entry-num: Specify the hash entry number in hexadecimal to be setup\n"
301 		"  --ipv6: Set if running ipv6 packets\n"
302 		"  --parse-ptype: Set to use software to analyze packet type\n\n",
303 		prgname);
304 }
305 
306 static int
307 parse_max_pkt_len(const char *pktlen)
308 {
309 	char *end = NULL;
310 	unsigned long len;
311 
312 	/* parse decimal string */
313 	len = strtoul(pktlen, &end, 10);
314 	if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
315 		return -1;
316 
317 	if (len == 0)
318 		return -1;
319 
320 	return len;
321 }
322 
323 static int
324 parse_portmask(const char *portmask)
325 {
326 	char *end = NULL;
327 	unsigned long pm;
328 
329 	/* parse hexadecimal string */
330 	pm = strtoul(portmask, &end, 16);
331 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
332 		return -1;
333 
334 	if (pm == 0)
335 		return -1;
336 
337 	return pm;
338 }
339 
340 static int
341 parse_hash_entry_number(const char *hash_entry_num)
342 {
343 	char *end = NULL;
344 	unsigned long hash_en;
345 	/* parse hexadecimal string */
346 	hash_en = strtoul(hash_entry_num, &end, 16);
347 	if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
348 		return -1;
349 
350 	if (hash_en == 0)
351 		return -1;
352 
353 	return hash_en;
354 }
355 
356 static int
357 parse_config(const char *q_arg)
358 {
359 	char s[256];
360 	const char *p, *p0 = q_arg;
361 	char *end;
362 	enum fieldnames {
363 		FLD_PORT = 0,
364 		FLD_QUEUE,
365 		FLD_LCORE,
366 		_NUM_FLD
367 	};
368 	unsigned long int_fld[_NUM_FLD];
369 	char *str_fld[_NUM_FLD];
370 	int i;
371 	unsigned size;
372 
373 	nb_lcore_params = 0;
374 
375 	while ((p = strchr(p0,'(')) != NULL) {
376 		++p;
377 		if((p0 = strchr(p,')')) == NULL)
378 			return -1;
379 
380 		size = p0 - p;
381 		if(size >= sizeof(s))
382 			return -1;
383 
384 		snprintf(s, sizeof(s), "%.*s", size, p);
385 		if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
386 			return -1;
387 		for (i = 0; i < _NUM_FLD; i++){
388 			errno = 0;
389 			int_fld[i] = strtoul(str_fld[i], &end, 0);
390 			if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
391 				return -1;
392 		}
393 		if (nb_lcore_params >= MAX_LCORE_PARAMS) {
394 			printf("exceeded max number of lcore params: %hu\n",
395 				nb_lcore_params);
396 			return -1;
397 		}
398 		lcore_params_array[nb_lcore_params].port_id =
399 			(uint8_t)int_fld[FLD_PORT];
400 		lcore_params_array[nb_lcore_params].queue_id =
401 			(uint8_t)int_fld[FLD_QUEUE];
402 		lcore_params_array[nb_lcore_params].lcore_id =
403 			(uint8_t)int_fld[FLD_LCORE];
404 		++nb_lcore_params;
405 	}
406 	lcore_params = lcore_params_array;
407 	return 0;
408 }
409 
410 static void
411 parse_eth_dest(const char *optarg)
412 {
413 	uint16_t portid;
414 	char *port_end;
415 	uint8_t c, *dest, peer_addr[6];
416 
417 	errno = 0;
418 	portid = strtoul(optarg, &port_end, 10);
419 	if (errno != 0 || port_end == optarg || *port_end++ != ',')
420 		rte_exit(EXIT_FAILURE,
421 		"Invalid eth-dest: %s", optarg);
422 	if (portid >= RTE_MAX_ETHPORTS)
423 		rte_exit(EXIT_FAILURE,
424 		"eth-dest: port %d >= RTE_MAX_ETHPORTS(%d)\n",
425 		portid, RTE_MAX_ETHPORTS);
426 
427 	if (cmdline_parse_etheraddr(NULL, port_end,
428 		&peer_addr, sizeof(peer_addr)) < 0)
429 		rte_exit(EXIT_FAILURE,
430 		"Invalid ethernet address: %s\n",
431 		port_end);
432 	dest = (uint8_t *)&dest_eth_addr[portid];
433 	for (c = 0; c < 6; c++)
434 		dest[c] = peer_addr[c];
435 	*(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
436 }
437 
438 #define MAX_JUMBO_PKT_LEN  9600
439 #define MEMPOOL_CACHE_SIZE 256
440 
441 static const char short_options[] =
442 	"p:"  /* portmask */
443 	"P"   /* promiscuous */
444 	"L"   /* enable long prefix match */
445 	"E"   /* enable exact match */
446 	;
447 
448 #define CMD_LINE_OPT_CONFIG "config"
449 #define CMD_LINE_OPT_ETH_DEST "eth-dest"
450 #define CMD_LINE_OPT_NO_NUMA "no-numa"
451 #define CMD_LINE_OPT_IPV6 "ipv6"
452 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
453 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
454 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
455 enum {
456 	/* long options mapped to a short option */
457 
458 	/* first long only option value must be >= 256, so that we won't
459 	 * conflict with short options */
460 	CMD_LINE_OPT_MIN_NUM = 256,
461 	CMD_LINE_OPT_CONFIG_NUM,
462 	CMD_LINE_OPT_ETH_DEST_NUM,
463 	CMD_LINE_OPT_NO_NUMA_NUM,
464 	CMD_LINE_OPT_IPV6_NUM,
465 	CMD_LINE_OPT_ENABLE_JUMBO_NUM,
466 	CMD_LINE_OPT_HASH_ENTRY_NUM_NUM,
467 	CMD_LINE_OPT_PARSE_PTYPE_NUM,
468 };
469 
470 static const struct option lgopts[] = {
471 	{CMD_LINE_OPT_CONFIG, 1, 0, CMD_LINE_OPT_CONFIG_NUM},
472 	{CMD_LINE_OPT_ETH_DEST, 1, 0, CMD_LINE_OPT_ETH_DEST_NUM},
473 	{CMD_LINE_OPT_NO_NUMA, 0, 0, CMD_LINE_OPT_NO_NUMA_NUM},
474 	{CMD_LINE_OPT_IPV6, 0, 0, CMD_LINE_OPT_IPV6_NUM},
475 	{CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, CMD_LINE_OPT_ENABLE_JUMBO_NUM},
476 	{CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, CMD_LINE_OPT_HASH_ENTRY_NUM_NUM},
477 	{CMD_LINE_OPT_PARSE_PTYPE, 0, 0, CMD_LINE_OPT_PARSE_PTYPE_NUM},
478 	{NULL, 0, 0, 0}
479 };
480 
481 /*
482  * This expression is used to calculate the number of mbufs needed
483  * depending on user input, taking  into account memory for rx and
484  * tx hardware rings, cache per lcore and mtable per port per lcore.
485  * RTE_MAX is used to ensure that NB_MBUF never goes below a minimum
486  * value of 8192
487  */
488 #define NB_MBUF RTE_MAX(	\
489 	(nb_ports*nb_rx_queue*nb_rxd +		\
490 	nb_ports*nb_lcores*MAX_PKT_BURST +	\
491 	nb_ports*n_tx_queue*nb_txd +		\
492 	nb_lcores*MEMPOOL_CACHE_SIZE),		\
493 	(unsigned)8192)
494 
495 /* Parse the argument given in the command line of the application */
496 static int
497 parse_args(int argc, char **argv)
498 {
499 	int opt, ret;
500 	char **argvopt;
501 	int option_index;
502 	char *prgname = argv[0];
503 
504 	argvopt = argv;
505 
506 	/* Error or normal output strings. */
507 	while ((opt = getopt_long(argc, argvopt, short_options,
508 				lgopts, &option_index)) != EOF) {
509 
510 		switch (opt) {
511 		/* portmask */
512 		case 'p':
513 			enabled_port_mask = parse_portmask(optarg);
514 			if (enabled_port_mask == 0) {
515 				fprintf(stderr, "Invalid portmask\n");
516 				print_usage(prgname);
517 				return -1;
518 			}
519 			break;
520 
521 		case 'P':
522 			promiscuous_on = 1;
523 			break;
524 
525 		case 'E':
526 			l3fwd_em_on = 1;
527 			break;
528 
529 		case 'L':
530 			l3fwd_lpm_on = 1;
531 			break;
532 
533 		/* long options */
534 		case CMD_LINE_OPT_CONFIG_NUM:
535 			ret = parse_config(optarg);
536 			if (ret) {
537 				fprintf(stderr, "Invalid config\n");
538 				print_usage(prgname);
539 				return -1;
540 			}
541 			break;
542 
543 		case CMD_LINE_OPT_ETH_DEST_NUM:
544 			parse_eth_dest(optarg);
545 			break;
546 
547 		case CMD_LINE_OPT_NO_NUMA_NUM:
548 			numa_on = 0;
549 			break;
550 
551 		case CMD_LINE_OPT_IPV6_NUM:
552 			ipv6 = 1;
553 			break;
554 
555 		case CMD_LINE_OPT_ENABLE_JUMBO_NUM: {
556 			const struct option lenopts = {
557 				"max-pkt-len", required_argument, 0, 0
558 			};
559 
560 			port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
561 			port_conf.txmode.offloads |= DEV_TX_OFFLOAD_MULTI_SEGS;
562 
563 			/*
564 			 * if no max-pkt-len set, use the default
565 			 * value ETHER_MAX_LEN.
566 			 */
567 			if (getopt_long(argc, argvopt, "",
568 					&lenopts, &option_index) == 0) {
569 				ret = parse_max_pkt_len(optarg);
570 				if (ret < 64 || ret > MAX_JUMBO_PKT_LEN) {
571 					fprintf(stderr,
572 						"invalid maximum packet length\n");
573 					print_usage(prgname);
574 					return -1;
575 				}
576 				port_conf.rxmode.max_rx_pkt_len = ret;
577 			}
578 			break;
579 		}
580 
581 		case CMD_LINE_OPT_HASH_ENTRY_NUM_NUM:
582 			ret = parse_hash_entry_number(optarg);
583 			if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
584 				hash_entry_number = ret;
585 			} else {
586 				fprintf(stderr, "invalid hash entry number\n");
587 				print_usage(prgname);
588 				return -1;
589 			}
590 			break;
591 
592 		case CMD_LINE_OPT_PARSE_PTYPE_NUM:
593 			printf("soft parse-ptype is enabled\n");
594 			parse_ptype = 1;
595 			break;
596 
597 		default:
598 			print_usage(prgname);
599 			return -1;
600 		}
601 	}
602 
603 	/* If both LPM and EM are selected, return error. */
604 	if (l3fwd_lpm_on && l3fwd_em_on) {
605 		fprintf(stderr, "LPM and EM are mutually exclusive, select only one\n");
606 		return -1;
607 	}
608 
609 	/*
610 	 * Nothing is selected, pick longest-prefix match
611 	 * as default match.
612 	 */
613 	if (!l3fwd_lpm_on && !l3fwd_em_on) {
614 		fprintf(stderr, "LPM or EM none selected, default LPM on\n");
615 		l3fwd_lpm_on = 1;
616 	}
617 
618 	/*
619 	 * ipv6 and hash flags are valid only for
620 	 * exact macth, reset them to default for
621 	 * longest-prefix match.
622 	 */
623 	if (l3fwd_lpm_on) {
624 		ipv6 = 0;
625 		hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
626 	}
627 
628 	if (optind >= 0)
629 		argv[optind-1] = prgname;
630 
631 	ret = optind-1;
632 	optind = 1; /* reset getopt lib */
633 	return ret;
634 }
635 
636 static void
637 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
638 {
639 	char buf[ETHER_ADDR_FMT_SIZE];
640 	ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
641 	printf("%s%s", name, buf);
642 }
643 
644 static int
645 init_mem(unsigned nb_mbuf)
646 {
647 	struct lcore_conf *qconf;
648 	int socketid;
649 	unsigned lcore_id;
650 	char s[64];
651 
652 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
653 		if (rte_lcore_is_enabled(lcore_id) == 0)
654 			continue;
655 
656 		if (numa_on)
657 			socketid = rte_lcore_to_socket_id(lcore_id);
658 		else
659 			socketid = 0;
660 
661 		if (socketid >= NB_SOCKETS) {
662 			rte_exit(EXIT_FAILURE,
663 				"Socket %d of lcore %u is out of range %d\n",
664 				socketid, lcore_id, NB_SOCKETS);
665 		}
666 
667 		if (pktmbuf_pool[socketid] == NULL) {
668 			snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
669 			pktmbuf_pool[socketid] =
670 				rte_pktmbuf_pool_create(s, nb_mbuf,
671 					MEMPOOL_CACHE_SIZE, 0,
672 					RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
673 			if (pktmbuf_pool[socketid] == NULL)
674 				rte_exit(EXIT_FAILURE,
675 					"Cannot init mbuf pool on socket %d\n",
676 					socketid);
677 			else
678 				printf("Allocated mbuf pool on socket %d\n",
679 					socketid);
680 
681 			/* Setup either LPM or EM(f.e Hash).  */
682 			l3fwd_lkp.setup(socketid);
683 		}
684 		qconf = &lcore_conf[lcore_id];
685 		qconf->ipv4_lookup_struct =
686 			l3fwd_lkp.get_ipv4_lookup_struct(socketid);
687 		qconf->ipv6_lookup_struct =
688 			l3fwd_lkp.get_ipv6_lookup_struct(socketid);
689 	}
690 	return 0;
691 }
692 
693 /* Check the link status of all ports in up to 9s, and print them finally */
694 static void
695 check_all_ports_link_status(uint32_t port_mask)
696 {
697 #define CHECK_INTERVAL 100 /* 100ms */
698 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
699 	uint16_t portid;
700 	uint8_t count, all_ports_up, print_flag = 0;
701 	struct rte_eth_link link;
702 
703 	printf("\nChecking link status");
704 	fflush(stdout);
705 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
706 		if (force_quit)
707 			return;
708 		all_ports_up = 1;
709 		RTE_ETH_FOREACH_DEV(portid) {
710 			if (force_quit)
711 				return;
712 			if ((port_mask & (1 << portid)) == 0)
713 				continue;
714 			memset(&link, 0, sizeof(link));
715 			rte_eth_link_get_nowait(portid, &link);
716 			/* print link status if flag set */
717 			if (print_flag == 1) {
718 				if (link.link_status)
719 					printf(
720 					"Port%d Link Up. Speed %u Mbps -%s\n",
721 						portid, link.link_speed,
722 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
723 					("full-duplex") : ("half-duplex\n"));
724 				else
725 					printf("Port %d Link Down\n", portid);
726 				continue;
727 			}
728 			/* clear all_ports_up flag if any link down */
729 			if (link.link_status == ETH_LINK_DOWN) {
730 				all_ports_up = 0;
731 				break;
732 			}
733 		}
734 		/* after finally printing all link status, get out */
735 		if (print_flag == 1)
736 			break;
737 
738 		if (all_ports_up == 0) {
739 			printf(".");
740 			fflush(stdout);
741 			rte_delay_ms(CHECK_INTERVAL);
742 		}
743 
744 		/* set the print_flag if all ports up or timeout */
745 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
746 			print_flag = 1;
747 			printf("done\n");
748 		}
749 	}
750 }
751 
752 static void
753 signal_handler(int signum)
754 {
755 	if (signum == SIGINT || signum == SIGTERM) {
756 		printf("\n\nSignal %d received, preparing to exit...\n",
757 				signum);
758 		force_quit = true;
759 	}
760 }
761 
762 static int
763 prepare_ptype_parser(uint16_t portid, uint16_t queueid)
764 {
765 	if (parse_ptype) {
766 		printf("Port %d: softly parse packet type info\n", portid);
767 		if (rte_eth_add_rx_callback(portid, queueid,
768 					    l3fwd_lkp.cb_parse_ptype,
769 					    NULL))
770 			return 1;
771 
772 		printf("Failed to add rx callback: port=%d\n", portid);
773 		return 0;
774 	}
775 
776 	if (l3fwd_lkp.check_ptype(portid))
777 		return 1;
778 
779 	printf("port %d cannot parse packet type, please add --%s\n",
780 	       portid, CMD_LINE_OPT_PARSE_PTYPE);
781 	return 0;
782 }
783 
784 int
785 main(int argc, char **argv)
786 {
787 	struct lcore_conf *qconf;
788 	struct rte_eth_dev_info dev_info;
789 	struct rte_eth_txconf *txconf;
790 	int ret;
791 	unsigned nb_ports;
792 	uint16_t queueid, portid;
793 	unsigned lcore_id;
794 	uint32_t n_tx_queue, nb_lcores;
795 	uint8_t nb_rx_queue, queue, socketid;
796 
797 	/* init EAL */
798 	ret = rte_eal_init(argc, argv);
799 	if (ret < 0)
800 		rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
801 	argc -= ret;
802 	argv += ret;
803 
804 	force_quit = false;
805 	signal(SIGINT, signal_handler);
806 	signal(SIGTERM, signal_handler);
807 
808 	/* pre-init dst MACs for all ports to 02:00:00:00:00:xx */
809 	for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
810 		dest_eth_addr[portid] =
811 			ETHER_LOCAL_ADMIN_ADDR + ((uint64_t)portid << 40);
812 		*(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
813 	}
814 
815 	/* parse application arguments (after the EAL ones) */
816 	ret = parse_args(argc, argv);
817 	if (ret < 0)
818 		rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
819 
820 	if (check_lcore_params() < 0)
821 		rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
822 
823 	ret = init_lcore_rx_queues();
824 	if (ret < 0)
825 		rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
826 
827 	nb_ports = rte_eth_dev_count_avail();
828 
829 	if (check_port_config() < 0)
830 		rte_exit(EXIT_FAILURE, "check_port_config failed\n");
831 
832 	nb_lcores = rte_lcore_count();
833 
834 	/* Setup function pointers for lookup method. */
835 	setup_l3fwd_lookup_tables();
836 
837 	/* initialize all ports */
838 	RTE_ETH_FOREACH_DEV(portid) {
839 		struct rte_eth_conf local_port_conf = port_conf;
840 
841 		/* skip ports that are not enabled */
842 		if ((enabled_port_mask & (1 << portid)) == 0) {
843 			printf("\nSkipping disabled port %d\n", portid);
844 			continue;
845 		}
846 
847 		/* init port */
848 		printf("Initializing port %d ... ", portid );
849 		fflush(stdout);
850 
851 		nb_rx_queue = get_port_n_rx_queues(portid);
852 		n_tx_queue = nb_lcores;
853 		if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
854 			n_tx_queue = MAX_TX_QUEUE_PER_PORT;
855 		printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
856 			nb_rx_queue, (unsigned)n_tx_queue );
857 
858 		rte_eth_dev_info_get(portid, &dev_info);
859 		if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
860 			local_port_conf.txmode.offloads |=
861 				DEV_TX_OFFLOAD_MBUF_FAST_FREE;
862 
863 		local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
864 			dev_info.flow_type_rss_offloads;
865 		if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
866 				port_conf.rx_adv_conf.rss_conf.rss_hf) {
867 			printf("Port %u modified RSS hash function based on hardware support,"
868 				"requested:%#"PRIx64" configured:%#"PRIx64"\n",
869 				portid,
870 				port_conf.rx_adv_conf.rss_conf.rss_hf,
871 				local_port_conf.rx_adv_conf.rss_conf.rss_hf);
872 		}
873 
874 		ret = rte_eth_dev_configure(portid, nb_rx_queue,
875 					(uint16_t)n_tx_queue, &local_port_conf);
876 		if (ret < 0)
877 			rte_exit(EXIT_FAILURE,
878 				"Cannot configure device: err=%d, port=%d\n",
879 				ret, portid);
880 
881 		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
882 						       &nb_txd);
883 		if (ret < 0)
884 			rte_exit(EXIT_FAILURE,
885 				 "Cannot adjust number of descriptors: err=%d, "
886 				 "port=%d\n", ret, portid);
887 
888 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
889 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
890 		printf(", ");
891 		print_ethaddr("Destination:",
892 			(const struct ether_addr *)&dest_eth_addr[portid]);
893 		printf(", ");
894 
895 		/*
896 		 * prepare src MACs for each port.
897 		 */
898 		ether_addr_copy(&ports_eth_addr[portid],
899 			(struct ether_addr *)(val_eth + portid) + 1);
900 
901 		/* init memory */
902 		ret = init_mem(NB_MBUF);
903 		if (ret < 0)
904 			rte_exit(EXIT_FAILURE, "init_mem failed\n");
905 
906 		/* init one TX queue per couple (lcore,port) */
907 		queueid = 0;
908 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
909 			if (rte_lcore_is_enabled(lcore_id) == 0)
910 				continue;
911 
912 			if (numa_on)
913 				socketid =
914 				(uint8_t)rte_lcore_to_socket_id(lcore_id);
915 			else
916 				socketid = 0;
917 
918 			printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
919 			fflush(stdout);
920 
921 			txconf = &dev_info.default_txconf;
922 			txconf->offloads = local_port_conf.txmode.offloads;
923 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
924 						     socketid, txconf);
925 			if (ret < 0)
926 				rte_exit(EXIT_FAILURE,
927 					"rte_eth_tx_queue_setup: err=%d, "
928 					"port=%d\n", ret, portid);
929 
930 			qconf = &lcore_conf[lcore_id];
931 			qconf->tx_queue_id[portid] = queueid;
932 			queueid++;
933 
934 			qconf->tx_port_id[qconf->n_tx_port] = portid;
935 			qconf->n_tx_port++;
936 		}
937 		printf("\n");
938 	}
939 
940 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
941 		if (rte_lcore_is_enabled(lcore_id) == 0)
942 			continue;
943 		qconf = &lcore_conf[lcore_id];
944 		printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
945 		fflush(stdout);
946 		/* init RX queues */
947 		for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
948 			struct rte_eth_dev *dev;
949 			struct rte_eth_conf *conf;
950 			struct rte_eth_rxconf rxq_conf;
951 
952 			portid = qconf->rx_queue_list[queue].port_id;
953 			queueid = qconf->rx_queue_list[queue].queue_id;
954 			dev = &rte_eth_devices[portid];
955 			conf = &dev->data->dev_conf;
956 
957 			if (numa_on)
958 				socketid =
959 				(uint8_t)rte_lcore_to_socket_id(lcore_id);
960 			else
961 				socketid = 0;
962 
963 			printf("rxq=%d,%d,%d ", portid, queueid, socketid);
964 			fflush(stdout);
965 
966 			rte_eth_dev_info_get(portid, &dev_info);
967 			rxq_conf = dev_info.default_rxconf;
968 			rxq_conf.offloads = conf->rxmode.offloads;
969 			ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
970 					socketid,
971 					&rxq_conf,
972 					pktmbuf_pool[socketid]);
973 			if (ret < 0)
974 				rte_exit(EXIT_FAILURE,
975 				"rte_eth_rx_queue_setup: err=%d, port=%d\n",
976 				ret, portid);
977 		}
978 	}
979 
980 	printf("\n");
981 
982 	/* start ports */
983 	RTE_ETH_FOREACH_DEV(portid) {
984 		if ((enabled_port_mask & (1 << portid)) == 0) {
985 			continue;
986 		}
987 		/* Start device */
988 		ret = rte_eth_dev_start(portid);
989 		if (ret < 0)
990 			rte_exit(EXIT_FAILURE,
991 				"rte_eth_dev_start: err=%d, port=%d\n",
992 				ret, portid);
993 
994 		/*
995 		 * If enabled, put device in promiscuous mode.
996 		 * This allows IO forwarding mode to forward packets
997 		 * to itself through 2 cross-connected  ports of the
998 		 * target machine.
999 		 */
1000 		if (promiscuous_on)
1001 			rte_eth_promiscuous_enable(portid);
1002 	}
1003 
1004 	printf("\n");
1005 
1006 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1007 		if (rte_lcore_is_enabled(lcore_id) == 0)
1008 			continue;
1009 		qconf = &lcore_conf[lcore_id];
1010 		for (queue = 0; queue < qconf->n_rx_queue; ++queue) {
1011 			portid = qconf->rx_queue_list[queue].port_id;
1012 			queueid = qconf->rx_queue_list[queue].queue_id;
1013 			if (prepare_ptype_parser(portid, queueid) == 0)
1014 				rte_exit(EXIT_FAILURE, "ptype check fails\n");
1015 		}
1016 	}
1017 
1018 
1019 	check_all_ports_link_status(enabled_port_mask);
1020 
1021 	ret = 0;
1022 	/* launch per-lcore init on every lcore */
1023 	rte_eal_mp_remote_launch(l3fwd_lkp.main_loop, NULL, CALL_MASTER);
1024 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1025 		if (rte_eal_wait_lcore(lcore_id) < 0) {
1026 			ret = -1;
1027 			break;
1028 		}
1029 	}
1030 
1031 	/* stop ports */
1032 	RTE_ETH_FOREACH_DEV(portid) {
1033 		if ((enabled_port_mask & (1 << portid)) == 0)
1034 			continue;
1035 		printf("Closing port %d...", portid);
1036 		rte_eth_dev_stop(portid);
1037 		rte_eth_dev_close(portid);
1038 		printf(" Done\n");
1039 	}
1040 	printf("Bye...\n");
1041 
1042 	return ret;
1043 }
1044