]> git.proxmox.com Git - ceph.git/blob - ceph/src/spdk/dpdk/examples/l3fwd/main.c
update sources to ceph Nautilus 14.2.1
[ceph.git] / ceph / src / spdk / dpdk / examples / l3fwd / main.c
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_CRC_STRIP |
124 DEV_RX_OFFLOAD_CHECKSUM),
125 },
126 .rx_adv_conf = {
127 .rss_conf = {
128 .rss_key = NULL,
129 .rss_hf = ETH_RSS_IP,
130 },
131 },
132 .txmode = {
133 .mq_mode = ETH_MQ_TX_NONE,
134 },
135 };
136
137 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
138
139 struct l3fwd_lkp_mode {
140 void (*setup)(int);
141 int (*check_ptype)(int);
142 rte_rx_callback_fn cb_parse_ptype;
143 int (*main_loop)(void *);
144 void* (*get_ipv4_lookup_struct)(int);
145 void* (*get_ipv6_lookup_struct)(int);
146 };
147
148 static struct l3fwd_lkp_mode l3fwd_lkp;
149
150 static struct l3fwd_lkp_mode l3fwd_em_lkp = {
151 .setup = setup_hash,
152 .check_ptype = em_check_ptype,
153 .cb_parse_ptype = em_cb_parse_ptype,
154 .main_loop = em_main_loop,
155 .get_ipv4_lookup_struct = em_get_ipv4_l3fwd_lookup_struct,
156 .get_ipv6_lookup_struct = em_get_ipv6_l3fwd_lookup_struct,
157 };
158
159 static struct l3fwd_lkp_mode l3fwd_lpm_lkp = {
160 .setup = setup_lpm,
161 .check_ptype = lpm_check_ptype,
162 .cb_parse_ptype = lpm_cb_parse_ptype,
163 .main_loop = lpm_main_loop,
164 .get_ipv4_lookup_struct = lpm_get_ipv4_l3fwd_lookup_struct,
165 .get_ipv6_lookup_struct = lpm_get_ipv6_l3fwd_lookup_struct,
166 };
167
168 /*
169 * Setup lookup methods for forwarding.
170 * Currently exact-match and longest-prefix-match
171 * are supported ones.
172 */
173 static void
174 setup_l3fwd_lookup_tables(void)
175 {
176 /* Setup HASH lookup functions. */
177 if (l3fwd_em_on)
178 l3fwd_lkp = l3fwd_em_lkp;
179 /* Setup LPM lookup functions. */
180 else
181 l3fwd_lkp = l3fwd_lpm_lkp;
182 }
183
184 static int
185 check_lcore_params(void)
186 {
187 uint8_t queue, lcore;
188 uint16_t i;
189 int socketid;
190
191 for (i = 0; i < nb_lcore_params; ++i) {
192 queue = lcore_params[i].queue_id;
193 if (queue >= MAX_RX_QUEUE_PER_PORT) {
194 printf("invalid queue number: %hhu\n", queue);
195 return -1;
196 }
197 lcore = lcore_params[i].lcore_id;
198 if (!rte_lcore_is_enabled(lcore)) {
199 printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
200 return -1;
201 }
202 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
203 (numa_on == 0)) {
204 printf("warning: lcore %hhu is on socket %d with numa off \n",
205 lcore, socketid);
206 }
207 }
208 return 0;
209 }
210
211 static int
212 check_port_config(void)
213 {
214 uint16_t portid;
215 uint16_t i;
216
217 for (i = 0; i < nb_lcore_params; ++i) {
218 portid = lcore_params[i].port_id;
219 if ((enabled_port_mask & (1 << portid)) == 0) {
220 printf("port %u is not enabled in port mask\n", portid);
221 return -1;
222 }
223 if (!rte_eth_dev_is_valid_port(portid)) {
224 printf("port %u is not present on the board\n", portid);
225 return -1;
226 }
227 }
228 return 0;
229 }
230
231 static uint8_t
232 get_port_n_rx_queues(const uint16_t port)
233 {
234 int queue = -1;
235 uint16_t i;
236
237 for (i = 0; i < nb_lcore_params; ++i) {
238 if (lcore_params[i].port_id == port) {
239 if (lcore_params[i].queue_id == queue+1)
240 queue = lcore_params[i].queue_id;
241 else
242 rte_exit(EXIT_FAILURE, "queue ids of the port %d must be"
243 " in sequence and must start with 0\n",
244 lcore_params[i].port_id);
245 }
246 }
247 return (uint8_t)(++queue);
248 }
249
250 static int
251 init_lcore_rx_queues(void)
252 {
253 uint16_t i, nb_rx_queue;
254 uint8_t lcore;
255
256 for (i = 0; i < nb_lcore_params; ++i) {
257 lcore = lcore_params[i].lcore_id;
258 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
259 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
260 printf("error: too many queues (%u) for lcore: %u\n",
261 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
262 return -1;
263 } else {
264 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
265 lcore_params[i].port_id;
266 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
267 lcore_params[i].queue_id;
268 lcore_conf[lcore].n_rx_queue++;
269 }
270 }
271 return 0;
272 }
273
274 /* display usage */
275 static void
276 print_usage(const char *prgname)
277 {
278 fprintf(stderr, "%s [EAL options] --"
279 " -p PORTMASK"
280 " [-P]"
281 " [-E]"
282 " [-L]"
283 " --config (port,queue,lcore)[,(port,queue,lcore)]"
284 " [--eth-dest=X,MM:MM:MM:MM:MM:MM]"
285 " [--enable-jumbo [--max-pkt-len PKTLEN]]"
286 " [--no-numa]"
287 " [--hash-entry-num]"
288 " [--ipv6]"
289 " [--parse-ptype]\n\n"
290
291 " -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
292 " -P : Enable promiscuous mode\n"
293 " -E : Enable exact match\n"
294 " -L : Enable longest prefix match (default)\n"
295 " --config (port,queue,lcore): Rx queue configuration\n"
296 " --eth-dest=X,MM:MM:MM:MM:MM:MM: Ethernet destination for port X\n"
297 " --enable-jumbo: Enable jumbo frames\n"
298 " --max-pkt-len: Under the premise of enabling jumbo,\n"
299 " maximum packet length in decimal (64-9600)\n"
300 " --no-numa: Disable numa awareness\n"
301 " --hash-entry-num: Specify the hash entry number in hexadecimal to be setup\n"
302 " --ipv6: Set if running ipv6 packets\n"
303 " --parse-ptype: Set to use software to analyze packet type\n\n",
304 prgname);
305 }
306
307 static int
308 parse_max_pkt_len(const char *pktlen)
309 {
310 char *end = NULL;
311 unsigned long len;
312
313 /* parse decimal string */
314 len = strtoul(pktlen, &end, 10);
315 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
316 return -1;
317
318 if (len == 0)
319 return -1;
320
321 return len;
322 }
323
324 static int
325 parse_portmask(const char *portmask)
326 {
327 char *end = NULL;
328 unsigned long pm;
329
330 /* parse hexadecimal string */
331 pm = strtoul(portmask, &end, 16);
332 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
333 return -1;
334
335 if (pm == 0)
336 return -1;
337
338 return pm;
339 }
340
341 static int
342 parse_hash_entry_number(const char *hash_entry_num)
343 {
344 char *end = NULL;
345 unsigned long hash_en;
346 /* parse hexadecimal string */
347 hash_en = strtoul(hash_entry_num, &end, 16);
348 if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
349 return -1;
350
351 if (hash_en == 0)
352 return -1;
353
354 return hash_en;
355 }
356
357 static int
358 parse_config(const char *q_arg)
359 {
360 char s[256];
361 const char *p, *p0 = q_arg;
362 char *end;
363 enum fieldnames {
364 FLD_PORT = 0,
365 FLD_QUEUE,
366 FLD_LCORE,
367 _NUM_FLD
368 };
369 unsigned long int_fld[_NUM_FLD];
370 char *str_fld[_NUM_FLD];
371 int i;
372 unsigned size;
373
374 nb_lcore_params = 0;
375
376 while ((p = strchr(p0,'(')) != NULL) {
377 ++p;
378 if((p0 = strchr(p,')')) == NULL)
379 return -1;
380
381 size = p0 - p;
382 if(size >= sizeof(s))
383 return -1;
384
385 snprintf(s, sizeof(s), "%.*s", size, p);
386 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
387 return -1;
388 for (i = 0; i < _NUM_FLD; i++){
389 errno = 0;
390 int_fld[i] = strtoul(str_fld[i], &end, 0);
391 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
392 return -1;
393 }
394 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
395 printf("exceeded max number of lcore params: %hu\n",
396 nb_lcore_params);
397 return -1;
398 }
399 lcore_params_array[nb_lcore_params].port_id =
400 (uint8_t)int_fld[FLD_PORT];
401 lcore_params_array[nb_lcore_params].queue_id =
402 (uint8_t)int_fld[FLD_QUEUE];
403 lcore_params_array[nb_lcore_params].lcore_id =
404 (uint8_t)int_fld[FLD_LCORE];
405 ++nb_lcore_params;
406 }
407 lcore_params = lcore_params_array;
408 return 0;
409 }
410
411 static void
412 parse_eth_dest(const char *optarg)
413 {
414 uint16_t portid;
415 char *port_end;
416 uint8_t c, *dest, peer_addr[6];
417
418 errno = 0;
419 portid = strtoul(optarg, &port_end, 10);
420 if (errno != 0 || port_end == optarg || *port_end++ != ',')
421 rte_exit(EXIT_FAILURE,
422 "Invalid eth-dest: %s", optarg);
423 if (portid >= RTE_MAX_ETHPORTS)
424 rte_exit(EXIT_FAILURE,
425 "eth-dest: port %d >= RTE_MAX_ETHPORTS(%d)\n",
426 portid, RTE_MAX_ETHPORTS);
427
428 if (cmdline_parse_etheraddr(NULL, port_end,
429 &peer_addr, sizeof(peer_addr)) < 0)
430 rte_exit(EXIT_FAILURE,
431 "Invalid ethernet address: %s\n",
432 port_end);
433 dest = (uint8_t *)&dest_eth_addr[portid];
434 for (c = 0; c < 6; c++)
435 dest[c] = peer_addr[c];
436 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
437 }
438
439 #define MAX_JUMBO_PKT_LEN 9600
440 #define MEMPOOL_CACHE_SIZE 256
441
442 static const char short_options[] =
443 "p:" /* portmask */
444 "P" /* promiscuous */
445 "L" /* enable long prefix match */
446 "E" /* enable exact match */
447 ;
448
449 #define CMD_LINE_OPT_CONFIG "config"
450 #define CMD_LINE_OPT_ETH_DEST "eth-dest"
451 #define CMD_LINE_OPT_NO_NUMA "no-numa"
452 #define CMD_LINE_OPT_IPV6 "ipv6"
453 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
454 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
455 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
456 enum {
457 /* long options mapped to a short option */
458
459 /* first long only option value must be >= 256, so that we won't
460 * conflict with short options */
461 CMD_LINE_OPT_MIN_NUM = 256,
462 CMD_LINE_OPT_CONFIG_NUM,
463 CMD_LINE_OPT_ETH_DEST_NUM,
464 CMD_LINE_OPT_NO_NUMA_NUM,
465 CMD_LINE_OPT_IPV6_NUM,
466 CMD_LINE_OPT_ENABLE_JUMBO_NUM,
467 CMD_LINE_OPT_HASH_ENTRY_NUM_NUM,
468 CMD_LINE_OPT_PARSE_PTYPE_NUM,
469 };
470
471 static const struct option lgopts[] = {
472 {CMD_LINE_OPT_CONFIG, 1, 0, CMD_LINE_OPT_CONFIG_NUM},
473 {CMD_LINE_OPT_ETH_DEST, 1, 0, CMD_LINE_OPT_ETH_DEST_NUM},
474 {CMD_LINE_OPT_NO_NUMA, 0, 0, CMD_LINE_OPT_NO_NUMA_NUM},
475 {CMD_LINE_OPT_IPV6, 0, 0, CMD_LINE_OPT_IPV6_NUM},
476 {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, CMD_LINE_OPT_ENABLE_JUMBO_NUM},
477 {CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, CMD_LINE_OPT_HASH_ENTRY_NUM_NUM},
478 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, CMD_LINE_OPT_PARSE_PTYPE_NUM},
479 {NULL, 0, 0, 0}
480 };
481
482 /*
483 * This expression is used to calculate the number of mbufs needed
484 * depending on user input, taking into account memory for rx and
485 * tx hardware rings, cache per lcore and mtable per port per lcore.
486 * RTE_MAX is used to ensure that NB_MBUF never goes below a minimum
487 * value of 8192
488 */
489 #define NB_MBUF RTE_MAX( \
490 (nb_ports*nb_rx_queue*nb_rxd + \
491 nb_ports*nb_lcores*MAX_PKT_BURST + \
492 nb_ports*n_tx_queue*nb_txd + \
493 nb_lcores*MEMPOOL_CACHE_SIZE), \
494 (unsigned)8192)
495
496 /* Parse the argument given in the command line of the application */
497 static int
498 parse_args(int argc, char **argv)
499 {
500 int opt, ret;
501 char **argvopt;
502 int option_index;
503 char *prgname = argv[0];
504
505 argvopt = argv;
506
507 /* Error or normal output strings. */
508 while ((opt = getopt_long(argc, argvopt, short_options,
509 lgopts, &option_index)) != EOF) {
510
511 switch (opt) {
512 /* portmask */
513 case 'p':
514 enabled_port_mask = parse_portmask(optarg);
515 if (enabled_port_mask == 0) {
516 fprintf(stderr, "Invalid portmask\n");
517 print_usage(prgname);
518 return -1;
519 }
520 break;
521
522 case 'P':
523 promiscuous_on = 1;
524 break;
525
526 case 'E':
527 l3fwd_em_on = 1;
528 break;
529
530 case 'L':
531 l3fwd_lpm_on = 1;
532 break;
533
534 /* long options */
535 case CMD_LINE_OPT_CONFIG_NUM:
536 ret = parse_config(optarg);
537 if (ret) {
538 fprintf(stderr, "Invalid config\n");
539 print_usage(prgname);
540 return -1;
541 }
542 break;
543
544 case CMD_LINE_OPT_ETH_DEST_NUM:
545 parse_eth_dest(optarg);
546 break;
547
548 case CMD_LINE_OPT_NO_NUMA_NUM:
549 numa_on = 0;
550 break;
551
552 case CMD_LINE_OPT_IPV6_NUM:
553 ipv6 = 1;
554 break;
555
556 case CMD_LINE_OPT_ENABLE_JUMBO_NUM: {
557 const struct option lenopts = {
558 "max-pkt-len", required_argument, 0, 0
559 };
560
561 port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
562 port_conf.txmode.offloads |= DEV_TX_OFFLOAD_MULTI_SEGS;
563
564 /*
565 * if no max-pkt-len set, use the default
566 * value ETHER_MAX_LEN.
567 */
568 if (getopt_long(argc, argvopt, "",
569 &lenopts, &option_index) == 0) {
570 ret = parse_max_pkt_len(optarg);
571 if (ret < 64 || ret > MAX_JUMBO_PKT_LEN) {
572 fprintf(stderr,
573 "invalid maximum packet length\n");
574 print_usage(prgname);
575 return -1;
576 }
577 port_conf.rxmode.max_rx_pkt_len = ret;
578 }
579 break;
580 }
581
582 case CMD_LINE_OPT_HASH_ENTRY_NUM_NUM:
583 ret = parse_hash_entry_number(optarg);
584 if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
585 hash_entry_number = ret;
586 } else {
587 fprintf(stderr, "invalid hash entry number\n");
588 print_usage(prgname);
589 return -1;
590 }
591 break;
592
593 case CMD_LINE_OPT_PARSE_PTYPE_NUM:
594 printf("soft parse-ptype is enabled\n");
595 parse_ptype = 1;
596 break;
597
598 default:
599 print_usage(prgname);
600 return -1;
601 }
602 }
603
604 /* If both LPM and EM are selected, return error. */
605 if (l3fwd_lpm_on && l3fwd_em_on) {
606 fprintf(stderr, "LPM and EM are mutually exclusive, select only one\n");
607 return -1;
608 }
609
610 /*
611 * Nothing is selected, pick longest-prefix match
612 * as default match.
613 */
614 if (!l3fwd_lpm_on && !l3fwd_em_on) {
615 fprintf(stderr, "LPM or EM none selected, default LPM on\n");
616 l3fwd_lpm_on = 1;
617 }
618
619 /*
620 * ipv6 and hash flags are valid only for
621 * exact macth, reset them to default for
622 * longest-prefix match.
623 */
624 if (l3fwd_lpm_on) {
625 ipv6 = 0;
626 hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
627 }
628
629 if (optind >= 0)
630 argv[optind-1] = prgname;
631
632 ret = optind-1;
633 optind = 1; /* reset getopt lib */
634 return ret;
635 }
636
637 static void
638 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
639 {
640 char buf[ETHER_ADDR_FMT_SIZE];
641 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
642 printf("%s%s", name, buf);
643 }
644
645 static int
646 init_mem(unsigned nb_mbuf)
647 {
648 struct lcore_conf *qconf;
649 int socketid;
650 unsigned lcore_id;
651 char s[64];
652
653 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
654 if (rte_lcore_is_enabled(lcore_id) == 0)
655 continue;
656
657 if (numa_on)
658 socketid = rte_lcore_to_socket_id(lcore_id);
659 else
660 socketid = 0;
661
662 if (socketid >= NB_SOCKETS) {
663 rte_exit(EXIT_FAILURE,
664 "Socket %d of lcore %u is out of range %d\n",
665 socketid, lcore_id, NB_SOCKETS);
666 }
667
668 if (pktmbuf_pool[socketid] == NULL) {
669 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
670 pktmbuf_pool[socketid] =
671 rte_pktmbuf_pool_create(s, nb_mbuf,
672 MEMPOOL_CACHE_SIZE, 0,
673 RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
674 if (pktmbuf_pool[socketid] == NULL)
675 rte_exit(EXIT_FAILURE,
676 "Cannot init mbuf pool on socket %d\n",
677 socketid);
678 else
679 printf("Allocated mbuf pool on socket %d\n",
680 socketid);
681
682 /* Setup either LPM or EM(f.e Hash). */
683 l3fwd_lkp.setup(socketid);
684 }
685 qconf = &lcore_conf[lcore_id];
686 qconf->ipv4_lookup_struct =
687 l3fwd_lkp.get_ipv4_lookup_struct(socketid);
688 qconf->ipv6_lookup_struct =
689 l3fwd_lkp.get_ipv6_lookup_struct(socketid);
690 }
691 return 0;
692 }
693
694 /* Check the link status of all ports in up to 9s, and print them finally */
695 static void
696 check_all_ports_link_status(uint32_t port_mask)
697 {
698 #define CHECK_INTERVAL 100 /* 100ms */
699 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
700 uint16_t portid;
701 uint8_t count, all_ports_up, print_flag = 0;
702 struct rte_eth_link link;
703
704 printf("\nChecking link status");
705 fflush(stdout);
706 for (count = 0; count <= MAX_CHECK_TIME; count++) {
707 if (force_quit)
708 return;
709 all_ports_up = 1;
710 RTE_ETH_FOREACH_DEV(portid) {
711 if (force_quit)
712 return;
713 if ((port_mask & (1 << portid)) == 0)
714 continue;
715 memset(&link, 0, sizeof(link));
716 rte_eth_link_get_nowait(portid, &link);
717 /* print link status if flag set */
718 if (print_flag == 1) {
719 if (link.link_status)
720 printf(
721 "Port%d Link Up. Speed %u Mbps -%s\n",
722 portid, link.link_speed,
723 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
724 ("full-duplex") : ("half-duplex\n"));
725 else
726 printf("Port %d Link Down\n", portid);
727 continue;
728 }
729 /* clear all_ports_up flag if any link down */
730 if (link.link_status == ETH_LINK_DOWN) {
731 all_ports_up = 0;
732 break;
733 }
734 }
735 /* after finally printing all link status, get out */
736 if (print_flag == 1)
737 break;
738
739 if (all_ports_up == 0) {
740 printf(".");
741 fflush(stdout);
742 rte_delay_ms(CHECK_INTERVAL);
743 }
744
745 /* set the print_flag if all ports up or timeout */
746 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
747 print_flag = 1;
748 printf("done\n");
749 }
750 }
751 }
752
753 static void
754 signal_handler(int signum)
755 {
756 if (signum == SIGINT || signum == SIGTERM) {
757 printf("\n\nSignal %d received, preparing to exit...\n",
758 signum);
759 force_quit = true;
760 }
761 }
762
763 static int
764 prepare_ptype_parser(uint16_t portid, uint16_t queueid)
765 {
766 if (parse_ptype) {
767 printf("Port %d: softly parse packet type info\n", portid);
768 if (rte_eth_add_rx_callback(portid, queueid,
769 l3fwd_lkp.cb_parse_ptype,
770 NULL))
771 return 1;
772
773 printf("Failed to add rx callback: port=%d\n", portid);
774 return 0;
775 }
776
777 if (l3fwd_lkp.check_ptype(portid))
778 return 1;
779
780 printf("port %d cannot parse packet type, please add --%s\n",
781 portid, CMD_LINE_OPT_PARSE_PTYPE);
782 return 0;
783 }
784
785 int
786 main(int argc, char **argv)
787 {
788 struct lcore_conf *qconf;
789 struct rte_eth_dev_info dev_info;
790 struct rte_eth_txconf *txconf;
791 int ret;
792 unsigned nb_ports;
793 uint16_t queueid, portid;
794 unsigned lcore_id;
795 uint32_t n_tx_queue, nb_lcores;
796 uint8_t nb_rx_queue, queue, socketid;
797
798 /* init EAL */
799 ret = rte_eal_init(argc, argv);
800 if (ret < 0)
801 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
802 argc -= ret;
803 argv += ret;
804
805 force_quit = false;
806 signal(SIGINT, signal_handler);
807 signal(SIGTERM, signal_handler);
808
809 /* pre-init dst MACs for all ports to 02:00:00:00:00:xx */
810 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
811 dest_eth_addr[portid] =
812 ETHER_LOCAL_ADMIN_ADDR + ((uint64_t)portid << 40);
813 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
814 }
815
816 /* parse application arguments (after the EAL ones) */
817 ret = parse_args(argc, argv);
818 if (ret < 0)
819 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
820
821 if (check_lcore_params() < 0)
822 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
823
824 ret = init_lcore_rx_queues();
825 if (ret < 0)
826 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
827
828 nb_ports = rte_eth_dev_count_avail();
829
830 if (check_port_config() < 0)
831 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
832
833 nb_lcores = rte_lcore_count();
834
835 /* Setup function pointers for lookup method. */
836 setup_l3fwd_lookup_tables();
837
838 /* initialize all ports */
839 RTE_ETH_FOREACH_DEV(portid) {
840 struct rte_eth_conf local_port_conf = port_conf;
841
842 /* skip ports that are not enabled */
843 if ((enabled_port_mask & (1 << portid)) == 0) {
844 printf("\nSkipping disabled port %d\n", portid);
845 continue;
846 }
847
848 /* init port */
849 printf("Initializing port %d ... ", portid );
850 fflush(stdout);
851
852 nb_rx_queue = get_port_n_rx_queues(portid);
853 n_tx_queue = nb_lcores;
854 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
855 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
856 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
857 nb_rx_queue, (unsigned)n_tx_queue );
858
859 rte_eth_dev_info_get(portid, &dev_info);
860 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
861 local_port_conf.txmode.offloads |=
862 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
863
864 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
865 dev_info.flow_type_rss_offloads;
866 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
867 port_conf.rx_adv_conf.rss_conf.rss_hf) {
868 printf("Port %u modified RSS hash function based on hardware support,"
869 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
870 portid,
871 port_conf.rx_adv_conf.rss_conf.rss_hf,
872 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
873 }
874
875 ret = rte_eth_dev_configure(portid, nb_rx_queue,
876 (uint16_t)n_tx_queue, &local_port_conf);
877 if (ret < 0)
878 rte_exit(EXIT_FAILURE,
879 "Cannot configure device: err=%d, port=%d\n",
880 ret, portid);
881
882 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
883 &nb_txd);
884 if (ret < 0)
885 rte_exit(EXIT_FAILURE,
886 "Cannot adjust number of descriptors: err=%d, "
887 "port=%d\n", ret, portid);
888
889 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
890 print_ethaddr(" Address:", &ports_eth_addr[portid]);
891 printf(", ");
892 print_ethaddr("Destination:",
893 (const struct ether_addr *)&dest_eth_addr[portid]);
894 printf(", ");
895
896 /*
897 * prepare src MACs for each port.
898 */
899 ether_addr_copy(&ports_eth_addr[portid],
900 (struct ether_addr *)(val_eth + portid) + 1);
901
902 /* init memory */
903 ret = init_mem(NB_MBUF);
904 if (ret < 0)
905 rte_exit(EXIT_FAILURE, "init_mem failed\n");
906
907 /* init one TX queue per couple (lcore,port) */
908 queueid = 0;
909 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
910 if (rte_lcore_is_enabled(lcore_id) == 0)
911 continue;
912
913 if (numa_on)
914 socketid =
915 (uint8_t)rte_lcore_to_socket_id(lcore_id);
916 else
917 socketid = 0;
918
919 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
920 fflush(stdout);
921
922 txconf = &dev_info.default_txconf;
923 txconf->offloads = local_port_conf.txmode.offloads;
924 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
925 socketid, txconf);
926 if (ret < 0)
927 rte_exit(EXIT_FAILURE,
928 "rte_eth_tx_queue_setup: err=%d, "
929 "port=%d\n", ret, portid);
930
931 qconf = &lcore_conf[lcore_id];
932 qconf->tx_queue_id[portid] = queueid;
933 queueid++;
934
935 qconf->tx_port_id[qconf->n_tx_port] = portid;
936 qconf->n_tx_port++;
937 }
938 printf("\n");
939 }
940
941 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
942 if (rte_lcore_is_enabled(lcore_id) == 0)
943 continue;
944 qconf = &lcore_conf[lcore_id];
945 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
946 fflush(stdout);
947 /* init RX queues */
948 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
949 struct rte_eth_dev *dev;
950 struct rte_eth_conf *conf;
951 struct rte_eth_rxconf rxq_conf;
952
953 portid = qconf->rx_queue_list[queue].port_id;
954 queueid = qconf->rx_queue_list[queue].queue_id;
955 dev = &rte_eth_devices[portid];
956 conf = &dev->data->dev_conf;
957
958 if (numa_on)
959 socketid =
960 (uint8_t)rte_lcore_to_socket_id(lcore_id);
961 else
962 socketid = 0;
963
964 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
965 fflush(stdout);
966
967 rte_eth_dev_info_get(portid, &dev_info);
968 rxq_conf = dev_info.default_rxconf;
969 rxq_conf.offloads = conf->rxmode.offloads;
970 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
971 socketid,
972 &rxq_conf,
973 pktmbuf_pool[socketid]);
974 if (ret < 0)
975 rte_exit(EXIT_FAILURE,
976 "rte_eth_rx_queue_setup: err=%d, port=%d\n",
977 ret, portid);
978 }
979 }
980
981 printf("\n");
982
983 /* start ports */
984 RTE_ETH_FOREACH_DEV(portid) {
985 if ((enabled_port_mask & (1 << portid)) == 0) {
986 continue;
987 }
988 /* Start device */
989 ret = rte_eth_dev_start(portid);
990 if (ret < 0)
991 rte_exit(EXIT_FAILURE,
992 "rte_eth_dev_start: err=%d, port=%d\n",
993 ret, portid);
994
995 /*
996 * If enabled, put device in promiscuous mode.
997 * This allows IO forwarding mode to forward packets
998 * to itself through 2 cross-connected ports of the
999 * target machine.
1000 */
1001 if (promiscuous_on)
1002 rte_eth_promiscuous_enable(portid);
1003 }
1004
1005 printf("\n");
1006
1007 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1008 if (rte_lcore_is_enabled(lcore_id) == 0)
1009 continue;
1010 qconf = &lcore_conf[lcore_id];
1011 for (queue = 0; queue < qconf->n_rx_queue; ++queue) {
1012 portid = qconf->rx_queue_list[queue].port_id;
1013 queueid = qconf->rx_queue_list[queue].queue_id;
1014 if (prepare_ptype_parser(portid, queueid) == 0)
1015 rte_exit(EXIT_FAILURE, "ptype check fails\n");
1016 }
1017 }
1018
1019
1020 check_all_ports_link_status(enabled_port_mask);
1021
1022 ret = 0;
1023 /* launch per-lcore init on every lcore */
1024 rte_eal_mp_remote_launch(l3fwd_lkp.main_loop, NULL, CALL_MASTER);
1025 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1026 if (rte_eal_wait_lcore(lcore_id) < 0) {
1027 ret = -1;
1028 break;
1029 }
1030 }
1031
1032 /* stop ports */
1033 RTE_ETH_FOREACH_DEV(portid) {
1034 if ((enabled_port_mask & (1 << portid)) == 0)
1035 continue;
1036 printf("Closing port %d...", portid);
1037 rte_eth_dev_stop(portid);
1038 rte_eth_dev_close(portid);
1039 printf(" Done\n");
1040 }
1041 printf("Bye...\n");
1042
1043 return ret;
1044 }