xref: /dpdk/examples/l3fwd/main.c (revision 2f45703c17acb943aaded9f79676fd56a72542b2)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2016 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 #include <stdio.h>
35 #include <stdlib.h>
36 #include <stdint.h>
37 #include <inttypes.h>
38 #include <sys/types.h>
39 #include <string.h>
40 #include <sys/queue.h>
41 #include <stdarg.h>
42 #include <errno.h>
43 #include <getopt.h>
44 #include <signal.h>
45 #include <stdbool.h>
46 
47 #include <rte_common.h>
48 #include <rte_vect.h>
49 #include <rte_byteorder.h>
50 #include <rte_log.h>
51 #include <rte_memory.h>
52 #include <rte_memcpy.h>
53 #include <rte_memzone.h>
54 #include <rte_eal.h>
55 #include <rte_per_lcore.h>
56 #include <rte_launch.h>
57 #include <rte_atomic.h>
58 #include <rte_cycles.h>
59 #include <rte_prefetch.h>
60 #include <rte_lcore.h>
61 #include <rte_per_lcore.h>
62 #include <rte_branch_prediction.h>
63 #include <rte_interrupts.h>
64 #include <rte_pci.h>
65 #include <rte_random.h>
66 #include <rte_debug.h>
67 #include <rte_ether.h>
68 #include <rte_ethdev.h>
69 #include <rte_mempool.h>
70 #include <rte_mbuf.h>
71 #include <rte_ip.h>
72 #include <rte_tcp.h>
73 #include <rte_udp.h>
74 #include <rte_string_fns.h>
75 #include <rte_cpuflags.h>
76 
77 #include <cmdline_parse.h>
78 #include <cmdline_parse_etheraddr.h>
79 
80 #include "l3fwd.h"
81 
82 /*
83  * Configurable number of RX/TX ring descriptors
84  */
85 #define RTE_TEST_RX_DESC_DEFAULT 128
86 #define RTE_TEST_TX_DESC_DEFAULT 512
87 
88 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
89 #define MAX_RX_QUEUE_PER_PORT 128
90 
91 #define MAX_LCORE_PARAMS 1024
92 
93 /* Static global variables used within this file. */
94 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
95 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
96 
97 /**< Ports set in promiscuous mode off by default. */
98 static int promiscuous_on;
99 
100 /* Select Longest-Prefix or Exact match. */
101 static int l3fwd_lpm_on;
102 static int l3fwd_em_on;
103 
104 static int numa_on = 1; /**< NUMA is enabled by default. */
105 static int parse_ptype; /**< Parse packet type using rx callback, and */
106 			/**< disabled by default */
107 
108 /* Global variables. */
109 
110 volatile bool force_quit;
111 
112 /* ethernet addresses of ports */
113 uint64_t dest_eth_addr[RTE_MAX_ETHPORTS];
114 struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
115 
116 xmm_t val_eth[RTE_MAX_ETHPORTS];
117 
118 /* mask of enabled ports */
119 uint32_t enabled_port_mask;
120 
121 /* Used only in exact match mode. */
122 int ipv6; /**< ipv6 is false by default. */
123 uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
124 
125 struct lcore_conf lcore_conf[RTE_MAX_LCORE];
126 
127 struct lcore_params {
128 	uint8_t port_id;
129 	uint8_t queue_id;
130 	uint8_t lcore_id;
131 } __rte_cache_aligned;
132 
133 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
134 static struct lcore_params lcore_params_array_default[] = {
135 	{0, 0, 2},
136 	{0, 1, 2},
137 	{0, 2, 2},
138 	{1, 0, 2},
139 	{1, 1, 2},
140 	{1, 2, 2},
141 	{2, 0, 2},
142 	{3, 0, 3},
143 	{3, 1, 3},
144 };
145 
146 static struct lcore_params * lcore_params = lcore_params_array_default;
147 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
148 				sizeof(lcore_params_array_default[0]);
149 
150 static struct rte_eth_conf port_conf = {
151 	.rxmode = {
152 		.mq_mode = ETH_MQ_RX_RSS,
153 		.max_rx_pkt_len = ETHER_MAX_LEN,
154 		.split_hdr_size = 0,
155 		.header_split   = 0, /**< Header Split disabled */
156 		.hw_ip_checksum = 1, /**< IP checksum offload enabled */
157 		.hw_vlan_filter = 0, /**< VLAN filtering disabled */
158 		.jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
159 		.hw_strip_crc   = 0, /**< CRC stripped by hardware */
160 	},
161 	.rx_adv_conf = {
162 		.rss_conf = {
163 			.rss_key = NULL,
164 			.rss_hf = ETH_RSS_IP,
165 		},
166 	},
167 	.txmode = {
168 		.mq_mode = ETH_MQ_TX_NONE,
169 	},
170 };
171 
172 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
173 
174 struct l3fwd_lkp_mode {
175 	void  (*setup)(int);
176 	int   (*check_ptype)(int);
177 	rte_rx_callback_fn cb_parse_ptype;
178 	int   (*main_loop)(void *);
179 	void* (*get_ipv4_lookup_struct)(int);
180 	void* (*get_ipv6_lookup_struct)(int);
181 };
182 
183 static struct l3fwd_lkp_mode l3fwd_lkp;
184 
185 static struct l3fwd_lkp_mode l3fwd_em_lkp = {
186 	.setup                  = setup_hash,
187 	.check_ptype		= em_check_ptype,
188 	.cb_parse_ptype		= em_cb_parse_ptype,
189 	.main_loop              = em_main_loop,
190 	.get_ipv4_lookup_struct = em_get_ipv4_l3fwd_lookup_struct,
191 	.get_ipv6_lookup_struct = em_get_ipv6_l3fwd_lookup_struct,
192 };
193 
194 static struct l3fwd_lkp_mode l3fwd_lpm_lkp = {
195 	.setup                  = setup_lpm,
196 	.check_ptype		= lpm_check_ptype,
197 	.cb_parse_ptype		= lpm_cb_parse_ptype,
198 	.main_loop              = lpm_main_loop,
199 	.get_ipv4_lookup_struct = lpm_get_ipv4_l3fwd_lookup_struct,
200 	.get_ipv6_lookup_struct = lpm_get_ipv6_l3fwd_lookup_struct,
201 };
202 
203 /*
204  * Setup lookup methods for forwarding.
205  * Currently exact-match and longest-prefix-match
206  * are supported ones.
207  */
208 static void
209 setup_l3fwd_lookup_tables(void)
210 {
211 	/* Setup HASH lookup functions. */
212 	if (l3fwd_em_on)
213 		l3fwd_lkp = l3fwd_em_lkp;
214 	/* Setup LPM lookup functions. */
215 	else
216 		l3fwd_lkp = l3fwd_lpm_lkp;
217 }
218 
219 static int
220 check_lcore_params(void)
221 {
222 	uint8_t queue, lcore;
223 	uint16_t i;
224 	int socketid;
225 
226 	for (i = 0; i < nb_lcore_params; ++i) {
227 		queue = lcore_params[i].queue_id;
228 		if (queue >= MAX_RX_QUEUE_PER_PORT) {
229 			printf("invalid queue number: %hhu\n", queue);
230 			return -1;
231 		}
232 		lcore = lcore_params[i].lcore_id;
233 		if (!rte_lcore_is_enabled(lcore)) {
234 			printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
235 			return -1;
236 		}
237 		if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
238 			(numa_on == 0)) {
239 			printf("warning: lcore %hhu is on socket %d with numa off \n",
240 				lcore, socketid);
241 		}
242 	}
243 	return 0;
244 }
245 
246 static int
247 check_port_config(const unsigned nb_ports)
248 {
249 	unsigned portid;
250 	uint16_t i;
251 
252 	for (i = 0; i < nb_lcore_params; ++i) {
253 		portid = lcore_params[i].port_id;
254 		if ((enabled_port_mask & (1 << portid)) == 0) {
255 			printf("port %u is not enabled in port mask\n", portid);
256 			return -1;
257 		}
258 		if (portid >= nb_ports) {
259 			printf("port %u is not present on the board\n", portid);
260 			return -1;
261 		}
262 	}
263 	return 0;
264 }
265 
266 static uint8_t
267 get_port_n_rx_queues(const uint8_t port)
268 {
269 	int queue = -1;
270 	uint16_t i;
271 
272 	for (i = 0; i < nb_lcore_params; ++i) {
273 		if (lcore_params[i].port_id == port) {
274 			if (lcore_params[i].queue_id == queue+1)
275 				queue = lcore_params[i].queue_id;
276 			else
277 				rte_exit(EXIT_FAILURE, "queue ids of the port %d must be"
278 						" in sequence and must start with 0\n",
279 						lcore_params[i].port_id);
280 		}
281 	}
282 	return (uint8_t)(++queue);
283 }
284 
285 static int
286 init_lcore_rx_queues(void)
287 {
288 	uint16_t i, nb_rx_queue;
289 	uint8_t lcore;
290 
291 	for (i = 0; i < nb_lcore_params; ++i) {
292 		lcore = lcore_params[i].lcore_id;
293 		nb_rx_queue = lcore_conf[lcore].n_rx_queue;
294 		if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
295 			printf("error: too many queues (%u) for lcore: %u\n",
296 				(unsigned)nb_rx_queue + 1, (unsigned)lcore);
297 			return -1;
298 		} else {
299 			lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
300 				lcore_params[i].port_id;
301 			lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
302 				lcore_params[i].queue_id;
303 			lcore_conf[lcore].n_rx_queue++;
304 		}
305 	}
306 	return 0;
307 }
308 
309 /* display usage */
310 static void
311 print_usage(const char *prgname)
312 {
313 	printf("%s [EAL options] --"
314 		" -p PORTMASK"
315 		" [-P]"
316 		" [-E]"
317 		" [-L]"
318 		" --config (port,queue,lcore)[,(port,queue,lcore)]"
319 		" [--eth-dest=X,MM:MM:MM:MM:MM:MM]"
320 		" [--enable-jumbo [--max-pkt-len PKTLEN]]"
321 		" [--no-numa]"
322 		" [--hash-entry-num]"
323 		" [--ipv6]"
324 		" [--parse-ptype]\n\n"
325 
326 		"  -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
327 		"  -P : Enable promiscuous mode\n"
328 		"  -E : Enable exact match\n"
329 		"  -L : Enable longest prefix match (default)\n"
330 		"  --config (port,queue,lcore): Rx queue configuration\n"
331 		"  --eth-dest=X,MM:MM:MM:MM:MM:MM: Ethernet destination for port X\n"
332 		"  --enable-jumbo: Enable jumbo frames\n"
333 		"  --max-pkt-len: Under the premise of enabling jumbo,\n"
334 		"                 maximum packet length in decimal (64-9600)\n"
335 		"  --no-numa: Disable numa awareness\n"
336 		"  --hash-entry-num: Specify the hash entry number in hexadecimal to be setup\n"
337 		"  --ipv6: Set if running ipv6 packets\n"
338 		"  --parse-ptype: Set to use software to analyze packet type\n\n",
339 		prgname);
340 }
341 
342 static int
343 parse_max_pkt_len(const char *pktlen)
344 {
345 	char *end = NULL;
346 	unsigned long len;
347 
348 	/* parse decimal string */
349 	len = strtoul(pktlen, &end, 10);
350 	if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
351 		return -1;
352 
353 	if (len == 0)
354 		return -1;
355 
356 	return len;
357 }
358 
359 static int
360 parse_portmask(const char *portmask)
361 {
362 	char *end = NULL;
363 	unsigned long pm;
364 
365 	/* parse hexadecimal string */
366 	pm = strtoul(portmask, &end, 16);
367 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
368 		return -1;
369 
370 	if (pm == 0)
371 		return -1;
372 
373 	return pm;
374 }
375 
376 static int
377 parse_hash_entry_number(const char *hash_entry_num)
378 {
379 	char *end = NULL;
380 	unsigned long hash_en;
381 	/* parse hexadecimal string */
382 	hash_en = strtoul(hash_entry_num, &end, 16);
383 	if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
384 		return -1;
385 
386 	if (hash_en == 0)
387 		return -1;
388 
389 	return hash_en;
390 }
391 
392 static int
393 parse_config(const char *q_arg)
394 {
395 	char s[256];
396 	const char *p, *p0 = q_arg;
397 	char *end;
398 	enum fieldnames {
399 		FLD_PORT = 0,
400 		FLD_QUEUE,
401 		FLD_LCORE,
402 		_NUM_FLD
403 	};
404 	unsigned long int_fld[_NUM_FLD];
405 	char *str_fld[_NUM_FLD];
406 	int i;
407 	unsigned size;
408 
409 	nb_lcore_params = 0;
410 
411 	while ((p = strchr(p0,'(')) != NULL) {
412 		++p;
413 		if((p0 = strchr(p,')')) == NULL)
414 			return -1;
415 
416 		size = p0 - p;
417 		if(size >= sizeof(s))
418 			return -1;
419 
420 		snprintf(s, sizeof(s), "%.*s", size, p);
421 		if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
422 			return -1;
423 		for (i = 0; i < _NUM_FLD; i++){
424 			errno = 0;
425 			int_fld[i] = strtoul(str_fld[i], &end, 0);
426 			if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
427 				return -1;
428 		}
429 		if (nb_lcore_params >= MAX_LCORE_PARAMS) {
430 			printf("exceeded max number of lcore params: %hu\n",
431 				nb_lcore_params);
432 			return -1;
433 		}
434 		lcore_params_array[nb_lcore_params].port_id =
435 			(uint8_t)int_fld[FLD_PORT];
436 		lcore_params_array[nb_lcore_params].queue_id =
437 			(uint8_t)int_fld[FLD_QUEUE];
438 		lcore_params_array[nb_lcore_params].lcore_id =
439 			(uint8_t)int_fld[FLD_LCORE];
440 		++nb_lcore_params;
441 	}
442 	lcore_params = lcore_params_array;
443 	return 0;
444 }
445 
446 static void
447 parse_eth_dest(const char *optarg)
448 {
449 	uint8_t portid;
450 	char *port_end;
451 	uint8_t c, *dest, peer_addr[6];
452 
453 	errno = 0;
454 	portid = strtoul(optarg, &port_end, 10);
455 	if (errno != 0 || port_end == optarg || *port_end++ != ',')
456 		rte_exit(EXIT_FAILURE,
457 		"Invalid eth-dest: %s", optarg);
458 	if (portid >= RTE_MAX_ETHPORTS)
459 		rte_exit(EXIT_FAILURE,
460 		"eth-dest: port %d >= RTE_MAX_ETHPORTS(%d)\n",
461 		portid, RTE_MAX_ETHPORTS);
462 
463 	if (cmdline_parse_etheraddr(NULL, port_end,
464 		&peer_addr, sizeof(peer_addr)) < 0)
465 		rte_exit(EXIT_FAILURE,
466 		"Invalid ethernet address: %s\n",
467 		port_end);
468 	dest = (uint8_t *)&dest_eth_addr[portid];
469 	for (c = 0; c < 6; c++)
470 		dest[c] = peer_addr[c];
471 	*(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
472 }
473 
474 #define MAX_JUMBO_PKT_LEN  9600
475 #define MEMPOOL_CACHE_SIZE 256
476 
477 #define CMD_LINE_OPT_CONFIG "config"
478 #define CMD_LINE_OPT_ETH_DEST "eth-dest"
479 #define CMD_LINE_OPT_NO_NUMA "no-numa"
480 #define CMD_LINE_OPT_IPV6 "ipv6"
481 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
482 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
483 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
484 
485 /*
486  * This expression is used to calculate the number of mbufs needed
487  * depending on user input, taking  into account memory for rx and
488  * tx hardware rings, cache per lcore and mtable per port per lcore.
489  * RTE_MAX is used to ensure that NB_MBUF never goes below a minimum
490  * value of 8192
491  */
492 #define NB_MBUF RTE_MAX(	\
493 	(nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT +	\
494 	nb_ports*nb_lcores*MAX_PKT_BURST +			\
495 	nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT +		\
496 	nb_lcores*MEMPOOL_CACHE_SIZE),				\
497 	(unsigned)8192)
498 
499 /* Parse the argument given in the command line of the application */
500 static int
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 		{CMD_LINE_OPT_CONFIG, 1, 0, 0},
509 		{CMD_LINE_OPT_ETH_DEST, 1, 0, 0},
510 		{CMD_LINE_OPT_NO_NUMA, 0, 0, 0},
511 		{CMD_LINE_OPT_IPV6, 0, 0, 0},
512 		{CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, 0},
513 		{CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, 0},
514 		{CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
515 		{NULL, 0, 0, 0}
516 	};
517 
518 	argvopt = argv;
519 
520 	/* Error or normal output strings. */
521 	const char *str1 = "L3FWD: Invalid portmask";
522 	const char *str2 = "L3FWD: Promiscuous mode selected";
523 	const char *str3 = "L3FWD: Exact match selected";
524 	const char *str4 = "L3FWD: Longest-prefix match selected";
525 	const char *str5 = "L3FWD: Invalid config";
526 	const char *str6 = "L3FWD: NUMA is disabled";
527 	const char *str7 = "L3FWD: IPV6 is specified";
528 	const char *str8 =
529 		"L3FWD: Jumbo frame is enabled - disabling simple TX path";
530 	const char *str9 = "L3FWD: Invalid packet length";
531 	const char *str10 = "L3FWD: Set jumbo frame max packet len to ";
532 	const char *str11 = "L3FWD: Invalid hash entry number";
533 	const char *str12 =
534 		"L3FWD: LPM and EM are mutually exclusive, select only one";
535 	const char *str13 = "L3FWD: LPM or EM none selected, default LPM on";
536 
537 	while ((opt = getopt_long(argc, argvopt, "p:PLE",
538 				lgopts, &option_index)) != EOF) {
539 
540 		switch (opt) {
541 		/* portmask */
542 		case 'p':
543 			enabled_port_mask = parse_portmask(optarg);
544 			if (enabled_port_mask == 0) {
545 				printf("%s\n", str1);
546 				print_usage(prgname);
547 				return -1;
548 			}
549 			break;
550 		case 'P':
551 			printf("%s\n", str2);
552 			promiscuous_on = 1;
553 			break;
554 
555 		case 'E':
556 			printf("%s\n", str3);
557 			l3fwd_em_on = 1;
558 			break;
559 
560 		case 'L':
561 			printf("%s\n", str4);
562 			l3fwd_lpm_on = 1;
563 			break;
564 
565 		/* long options */
566 		case 0:
567 			if (!strncmp(lgopts[option_index].name,
568 					CMD_LINE_OPT_CONFIG,
569 					sizeof(CMD_LINE_OPT_CONFIG))) {
570 
571 				ret = parse_config(optarg);
572 				if (ret) {
573 					printf("%s\n", str5);
574 					print_usage(prgname);
575 					return -1;
576 				}
577 			}
578 
579 			if (!strncmp(lgopts[option_index].name,
580 					CMD_LINE_OPT_ETH_DEST,
581 					sizeof(CMD_LINE_OPT_ETH_DEST))) {
582 					parse_eth_dest(optarg);
583 			}
584 
585 			if (!strncmp(lgopts[option_index].name,
586 					CMD_LINE_OPT_NO_NUMA,
587 					sizeof(CMD_LINE_OPT_NO_NUMA))) {
588 				printf("%s\n", str6);
589 				numa_on = 0;
590 			}
591 
592 			if (!strncmp(lgopts[option_index].name,
593 				CMD_LINE_OPT_IPV6,
594 				sizeof(CMD_LINE_OPT_IPV6))) {
595 				printf("%sn", str7);
596 				ipv6 = 1;
597 			}
598 
599 			if (!strncmp(lgopts[option_index].name,
600 					CMD_LINE_OPT_ENABLE_JUMBO,
601 					sizeof(CMD_LINE_OPT_ENABLE_JUMBO))) {
602 				struct option lenopts = {
603 					"max-pkt-len", required_argument, 0, 0
604 				};
605 
606 				printf("%s\n", str8);
607 				port_conf.rxmode.jumbo_frame = 1;
608 
609 				/*
610 				 * if no max-pkt-len set, use the default
611 				 * value ETHER_MAX_LEN.
612 				 */
613 				if (0 == getopt_long(argc, argvopt, "",
614 						&lenopts, &option_index)) {
615 					ret = parse_max_pkt_len(optarg);
616 					if ((ret < 64) ||
617 						(ret > MAX_JUMBO_PKT_LEN)) {
618 						printf("%s\n", str9);
619 						print_usage(prgname);
620 						return -1;
621 					}
622 					port_conf.rxmode.max_rx_pkt_len = ret;
623 				}
624 				printf("%s %u\n", str10,
625 				(unsigned int)port_conf.rxmode.max_rx_pkt_len);
626 			}
627 
628 			if (!strncmp(lgopts[option_index].name,
629 				CMD_LINE_OPT_HASH_ENTRY_NUM,
630 				sizeof(CMD_LINE_OPT_HASH_ENTRY_NUM))) {
631 
632 				ret = parse_hash_entry_number(optarg);
633 				if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
634 					hash_entry_number = ret;
635 				} else {
636 					printf("%s\n", str11);
637 					print_usage(prgname);
638 					return -1;
639 				}
640 			}
641 
642 			if (!strncmp(lgopts[option_index].name,
643 				     CMD_LINE_OPT_PARSE_PTYPE,
644 				     sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
645 				printf("soft parse-ptype is enabled\n");
646 				parse_ptype = 1;
647 			}
648 
649 			break;
650 
651 		default:
652 			print_usage(prgname);
653 			return -1;
654 		}
655 	}
656 
657 	/* If both LPM and EM are selected, return error. */
658 	if (l3fwd_lpm_on && l3fwd_em_on) {
659 		printf("%s\n", str12);
660 		return -1;
661 	}
662 
663 	/*
664 	 * Nothing is selected, pick longest-prefix match
665 	 * as default match.
666 	 */
667 	if (!l3fwd_lpm_on && !l3fwd_em_on) {
668 		l3fwd_lpm_on = 1;
669 		printf("%s\n", str13);
670 	}
671 
672 	/*
673 	 * ipv6 and hash flags are valid only for
674 	 * exact macth, reset them to default for
675 	 * longest-prefix match.
676 	 */
677 	if (l3fwd_lpm_on) {
678 		ipv6 = 0;
679 		hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
680 	}
681 
682 	if (optind >= 0)
683 		argv[optind-1] = prgname;
684 
685 	ret = optind-1;
686 	optind = 0; /* reset getopt lib */
687 	return ret;
688 }
689 
690 static void
691 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
692 {
693 	char buf[ETHER_ADDR_FMT_SIZE];
694 	ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
695 	printf("%s%s", name, buf);
696 }
697 
698 static int
699 init_mem(unsigned nb_mbuf)
700 {
701 	struct lcore_conf *qconf;
702 	int socketid;
703 	unsigned lcore_id;
704 	char s[64];
705 
706 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
707 		if (rte_lcore_is_enabled(lcore_id) == 0)
708 			continue;
709 
710 		if (numa_on)
711 			socketid = rte_lcore_to_socket_id(lcore_id);
712 		else
713 			socketid = 0;
714 
715 		if (socketid >= NB_SOCKETS) {
716 			rte_exit(EXIT_FAILURE,
717 				"Socket %d of lcore %u is out of range %d\n",
718 				socketid, lcore_id, NB_SOCKETS);
719 		}
720 
721 		if (pktmbuf_pool[socketid] == NULL) {
722 			snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
723 			pktmbuf_pool[socketid] =
724 				rte_pktmbuf_pool_create(s, nb_mbuf,
725 					MEMPOOL_CACHE_SIZE, 0,
726 					RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
727 			if (pktmbuf_pool[socketid] == NULL)
728 				rte_exit(EXIT_FAILURE,
729 					"Cannot init mbuf pool on socket %d\n",
730 					socketid);
731 			else
732 				printf("Allocated mbuf pool on socket %d\n",
733 					socketid);
734 
735 			/* Setup either LPM or EM(f.e Hash).  */
736 			l3fwd_lkp.setup(socketid);
737 		}
738 		qconf = &lcore_conf[lcore_id];
739 		qconf->ipv4_lookup_struct =
740 			l3fwd_lkp.get_ipv4_lookup_struct(socketid);
741 		qconf->ipv6_lookup_struct =
742 			l3fwd_lkp.get_ipv6_lookup_struct(socketid);
743 	}
744 	return 0;
745 }
746 
747 /* Check the link status of all ports in up to 9s, and print them finally */
748 static void
749 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
750 {
751 #define CHECK_INTERVAL 100 /* 100ms */
752 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
753 	uint8_t portid, count, all_ports_up, print_flag = 0;
754 	struct rte_eth_link link;
755 
756 	printf("\nChecking link status");
757 	fflush(stdout);
758 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
759 		if (force_quit)
760 			return;
761 		all_ports_up = 1;
762 		for (portid = 0; portid < port_num; portid++) {
763 			if (force_quit)
764 				return;
765 			if ((port_mask & (1 << portid)) == 0)
766 				continue;
767 			memset(&link, 0, sizeof(link));
768 			rte_eth_link_get_nowait(portid, &link);
769 			/* print link status if flag set */
770 			if (print_flag == 1) {
771 				if (link.link_status)
772 					printf("Port %d Link Up - speed %u "
773 						"Mbps - %s\n", (uint8_t)portid,
774 						(unsigned)link.link_speed,
775 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
776 					("full-duplex") : ("half-duplex\n"));
777 				else
778 					printf("Port %d Link Down\n",
779 						(uint8_t)portid);
780 				continue;
781 			}
782 			/* clear all_ports_up flag if any link down */
783 			if (link.link_status == ETH_LINK_DOWN) {
784 				all_ports_up = 0;
785 				break;
786 			}
787 		}
788 		/* after finally printing all link status, get out */
789 		if (print_flag == 1)
790 			break;
791 
792 		if (all_ports_up == 0) {
793 			printf(".");
794 			fflush(stdout);
795 			rte_delay_ms(CHECK_INTERVAL);
796 		}
797 
798 		/* set the print_flag if all ports up or timeout */
799 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
800 			print_flag = 1;
801 			printf("done\n");
802 		}
803 	}
804 }
805 
806 static void
807 signal_handler(int signum)
808 {
809 	if (signum == SIGINT || signum == SIGTERM) {
810 		printf("\n\nSignal %d received, preparing to exit...\n",
811 				signum);
812 		force_quit = true;
813 	}
814 }
815 
816 static int
817 prepare_ptype_parser(uint8_t portid, uint16_t queueid)
818 {
819 	if (parse_ptype) {
820 		printf("Port %d: softly parse packet type info\n", portid);
821 		if (rte_eth_add_rx_callback(portid, queueid,
822 					    l3fwd_lkp.cb_parse_ptype,
823 					    NULL))
824 			return 1;
825 
826 		printf("Failed to add rx callback: port=%d\n", portid);
827 		return 0;
828 	}
829 
830 	if (l3fwd_lkp.check_ptype(portid))
831 		return 1;
832 
833 	printf("port %d cannot parse packet type, please add --%s\n",
834 	       portid, CMD_LINE_OPT_PARSE_PTYPE);
835 	return 0;
836 }
837 
838 int
839 main(int argc, char **argv)
840 {
841 	struct lcore_conf *qconf;
842 	struct rte_eth_dev_info dev_info;
843 	struct rte_eth_txconf *txconf;
844 	int ret;
845 	unsigned nb_ports;
846 	uint16_t queueid;
847 	unsigned lcore_id;
848 	uint32_t n_tx_queue, nb_lcores;
849 	uint8_t portid, nb_rx_queue, queue, socketid;
850 
851 	/* init EAL */
852 	ret = rte_eal_init(argc, argv);
853 	if (ret < 0)
854 		rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
855 	argc -= ret;
856 	argv += ret;
857 
858 	force_quit = false;
859 	signal(SIGINT, signal_handler);
860 	signal(SIGTERM, signal_handler);
861 
862 	/* pre-init dst MACs for all ports to 02:00:00:00:00:xx */
863 	for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
864 		dest_eth_addr[portid] =
865 			ETHER_LOCAL_ADMIN_ADDR + ((uint64_t)portid << 40);
866 		*(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
867 	}
868 
869 	/* parse application arguments (after the EAL ones) */
870 	ret = parse_args(argc, argv);
871 	if (ret < 0)
872 		rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
873 
874 	if (check_lcore_params() < 0)
875 		rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
876 
877 	ret = init_lcore_rx_queues();
878 	if (ret < 0)
879 		rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
880 
881 	nb_ports = rte_eth_dev_count();
882 
883 	if (check_port_config(nb_ports) < 0)
884 		rte_exit(EXIT_FAILURE, "check_port_config failed\n");
885 
886 	nb_lcores = rte_lcore_count();
887 
888 	/* Setup function pointers for lookup method. */
889 	setup_l3fwd_lookup_tables();
890 
891 	/* initialize all ports */
892 	for (portid = 0; portid < nb_ports; portid++) {
893 		/* skip ports that are not enabled */
894 		if ((enabled_port_mask & (1 << portid)) == 0) {
895 			printf("\nSkipping disabled port %d\n", portid);
896 			continue;
897 		}
898 
899 		/* init port */
900 		printf("Initializing port %d ... ", portid );
901 		fflush(stdout);
902 
903 		nb_rx_queue = get_port_n_rx_queues(portid);
904 		n_tx_queue = nb_lcores;
905 		if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
906 			n_tx_queue = MAX_TX_QUEUE_PER_PORT;
907 		printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
908 			nb_rx_queue, (unsigned)n_tx_queue );
909 		ret = rte_eth_dev_configure(portid, nb_rx_queue,
910 					(uint16_t)n_tx_queue, &port_conf);
911 		if (ret < 0)
912 			rte_exit(EXIT_FAILURE,
913 				"Cannot configure device: err=%d, port=%d\n",
914 				ret, portid);
915 
916 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
917 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
918 		printf(", ");
919 		print_ethaddr("Destination:",
920 			(const struct ether_addr *)&dest_eth_addr[portid]);
921 		printf(", ");
922 
923 		/*
924 		 * prepare src MACs for each port.
925 		 */
926 		ether_addr_copy(&ports_eth_addr[portid],
927 			(struct ether_addr *)(val_eth + portid) + 1);
928 
929 		/* init memory */
930 		ret = init_mem(NB_MBUF);
931 		if (ret < 0)
932 			rte_exit(EXIT_FAILURE, "init_mem failed\n");
933 
934 		/* init one TX queue per couple (lcore,port) */
935 		queueid = 0;
936 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
937 			if (rte_lcore_is_enabled(lcore_id) == 0)
938 				continue;
939 
940 			if (numa_on)
941 				socketid =
942 				(uint8_t)rte_lcore_to_socket_id(lcore_id);
943 			else
944 				socketid = 0;
945 
946 			printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
947 			fflush(stdout);
948 
949 			rte_eth_dev_info_get(portid, &dev_info);
950 			txconf = &dev_info.default_txconf;
951 			if (port_conf.rxmode.jumbo_frame)
952 				txconf->txq_flags = 0;
953 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
954 						     socketid, txconf);
955 			if (ret < 0)
956 				rte_exit(EXIT_FAILURE,
957 					"rte_eth_tx_queue_setup: err=%d, "
958 					"port=%d\n", ret, portid);
959 
960 			qconf = &lcore_conf[lcore_id];
961 			qconf->tx_queue_id[portid] = queueid;
962 			queueid++;
963 
964 			qconf->tx_port_id[qconf->n_tx_port] = portid;
965 			qconf->n_tx_port++;
966 		}
967 		printf("\n");
968 	}
969 
970 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
971 		if (rte_lcore_is_enabled(lcore_id) == 0)
972 			continue;
973 		qconf = &lcore_conf[lcore_id];
974 		printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
975 		fflush(stdout);
976 		/* init RX queues */
977 		for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
978 			portid = qconf->rx_queue_list[queue].port_id;
979 			queueid = qconf->rx_queue_list[queue].queue_id;
980 
981 			if (numa_on)
982 				socketid =
983 				(uint8_t)rte_lcore_to_socket_id(lcore_id);
984 			else
985 				socketid = 0;
986 
987 			printf("rxq=%d,%d,%d ", portid, queueid, socketid);
988 			fflush(stdout);
989 
990 			ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
991 					socketid,
992 					NULL,
993 					pktmbuf_pool[socketid]);
994 			if (ret < 0)
995 				rte_exit(EXIT_FAILURE,
996 				"rte_eth_rx_queue_setup: err=%d, port=%d\n",
997 				ret, portid);
998 		}
999 	}
1000 
1001 	printf("\n");
1002 
1003 	/* start ports */
1004 	for (portid = 0; portid < nb_ports; portid++) {
1005 		if ((enabled_port_mask & (1 << portid)) == 0) {
1006 			continue;
1007 		}
1008 		/* Start device */
1009 		ret = rte_eth_dev_start(portid);
1010 		if (ret < 0)
1011 			rte_exit(EXIT_FAILURE,
1012 				"rte_eth_dev_start: err=%d, port=%d\n",
1013 				ret, portid);
1014 
1015 		/*
1016 		 * If enabled, put device in promiscuous mode.
1017 		 * This allows IO forwarding mode to forward packets
1018 		 * to itself through 2 cross-connected  ports of the
1019 		 * target machine.
1020 		 */
1021 		if (promiscuous_on)
1022 			rte_eth_promiscuous_enable(portid);
1023 	}
1024 
1025 	printf("\n");
1026 
1027 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1028 		if (rte_lcore_is_enabled(lcore_id) == 0)
1029 			continue;
1030 		qconf = &lcore_conf[lcore_id];
1031 		for (queue = 0; queue < qconf->n_rx_queue; ++queue) {
1032 			portid = qconf->rx_queue_list[queue].port_id;
1033 			queueid = qconf->rx_queue_list[queue].queue_id;
1034 			if (prepare_ptype_parser(portid, queueid) == 0)
1035 				rte_exit(EXIT_FAILURE, "ptype check fails\n");
1036 		}
1037 	}
1038 
1039 
1040 	check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1041 
1042 	ret = 0;
1043 	/* launch per-lcore init on every lcore */
1044 	rte_eal_mp_remote_launch(l3fwd_lkp.main_loop, NULL, CALL_MASTER);
1045 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1046 		if (rte_eal_wait_lcore(lcore_id) < 0) {
1047 			ret = -1;
1048 			break;
1049 		}
1050 	}
1051 
1052 	/* stop ports */
1053 	for (portid = 0; portid < nb_ports; portid++) {
1054 		if ((enabled_port_mask & (1 << portid)) == 0)
1055 			continue;
1056 		printf("Closing port %d...", portid);
1057 		rte_eth_dev_stop(portid);
1058 		rte_eth_dev_close(portid);
1059 		printf(" Done\n");
1060 	}
1061 	printf("Bye...\n");
1062 
1063 	return ret;
1064 }
1065