net: add rte prefix to ether structures
[dpdk.git] / 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 /* Global variables. */
73
74 static int numa_on = 1; /**< NUMA is enabled by default. */
75 static int parse_ptype; /**< Parse packet type using rx callback, and */
76                         /**< disabled by default */
77 static int per_port_pool; /**< Use separate buffer pools per port; disabled */
78                           /**< by default */
79
80 volatile bool force_quit;
81
82 /* ethernet addresses of ports */
83 uint64_t dest_eth_addr[RTE_MAX_ETHPORTS];
84 struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
85
86 xmm_t val_eth[RTE_MAX_ETHPORTS];
87
88 /* mask of enabled ports */
89 uint32_t enabled_port_mask;
90
91 /* Used only in exact match mode. */
92 int ipv6; /**< ipv6 is false by default. */
93 uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
94
95 struct lcore_conf lcore_conf[RTE_MAX_LCORE];
96
97 struct lcore_params {
98         uint16_t port_id;
99         uint8_t queue_id;
100         uint8_t lcore_id;
101 } __rte_cache_aligned;
102
103 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
104 static struct lcore_params lcore_params_array_default[] = {
105         {0, 0, 2},
106         {0, 1, 2},
107         {0, 2, 2},
108         {1, 0, 2},
109         {1, 1, 2},
110         {1, 2, 2},
111         {2, 0, 2},
112         {3, 0, 3},
113         {3, 1, 3},
114 };
115
116 static struct lcore_params * lcore_params = lcore_params_array_default;
117 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
118                                 sizeof(lcore_params_array_default[0]);
119
120 static struct rte_eth_conf port_conf = {
121         .rxmode = {
122                 .mq_mode = ETH_MQ_RX_RSS,
123                 .max_rx_pkt_len = ETHER_MAX_LEN,
124                 .split_hdr_size = 0,
125                 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
126         },
127         .rx_adv_conf = {
128                 .rss_conf = {
129                         .rss_key = NULL,
130                         .rss_hf = ETH_RSS_IP,
131                 },
132         },
133         .txmode = {
134                 .mq_mode = ETH_MQ_TX_NONE,
135         },
136 };
137
138 static struct rte_mempool *pktmbuf_pool[RTE_MAX_ETHPORTS][NB_SOCKETS];
139 static uint8_t lkp_per_socket[NB_SOCKETS];
140
141 struct l3fwd_lkp_mode {
142         void  (*setup)(int);
143         int   (*check_ptype)(int);
144         rte_rx_callback_fn cb_parse_ptype;
145         int   (*main_loop)(void *);
146         void* (*get_ipv4_lookup_struct)(int);
147         void* (*get_ipv6_lookup_struct)(int);
148 };
149
150 static struct l3fwd_lkp_mode l3fwd_lkp;
151
152 static struct l3fwd_lkp_mode l3fwd_em_lkp = {
153         .setup                  = setup_hash,
154         .check_ptype            = em_check_ptype,
155         .cb_parse_ptype         = em_cb_parse_ptype,
156         .main_loop              = em_main_loop,
157         .get_ipv4_lookup_struct = em_get_ipv4_l3fwd_lookup_struct,
158         .get_ipv6_lookup_struct = em_get_ipv6_l3fwd_lookup_struct,
159 };
160
161 static struct l3fwd_lkp_mode l3fwd_lpm_lkp = {
162         .setup                  = setup_lpm,
163         .check_ptype            = lpm_check_ptype,
164         .cb_parse_ptype         = lpm_cb_parse_ptype,
165         .main_loop              = lpm_main_loop,
166         .get_ipv4_lookup_struct = lpm_get_ipv4_l3fwd_lookup_struct,
167         .get_ipv6_lookup_struct = lpm_get_ipv6_l3fwd_lookup_struct,
168 };
169
170 /*
171  * Setup lookup methods for forwarding.
172  * Currently exact-match and longest-prefix-match
173  * are supported ones.
174  */
175 static void
176 setup_l3fwd_lookup_tables(void)
177 {
178         /* Setup HASH lookup functions. */
179         if (l3fwd_em_on)
180                 l3fwd_lkp = l3fwd_em_lkp;
181         /* Setup LPM lookup functions. */
182         else
183                 l3fwd_lkp = l3fwd_lpm_lkp;
184 }
185
186 static int
187 check_lcore_params(void)
188 {
189         uint8_t queue, lcore;
190         uint16_t i;
191         int socketid;
192
193         for (i = 0; i < nb_lcore_params; ++i) {
194                 queue = lcore_params[i].queue_id;
195                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
196                         printf("invalid queue number: %hhu\n", queue);
197                         return -1;
198                 }
199                 lcore = lcore_params[i].lcore_id;
200                 if (!rte_lcore_is_enabled(lcore)) {
201                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
202                         return -1;
203                 }
204                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
205                         (numa_on == 0)) {
206                         printf("warning: lcore %hhu is on socket %d with numa off \n",
207                                 lcore, socketid);
208                 }
209         }
210         return 0;
211 }
212
213 static int
214 check_port_config(void)
215 {
216         uint16_t portid;
217         uint16_t i;
218
219         for (i = 0; i < nb_lcore_params; ++i) {
220                 portid = lcore_params[i].port_id;
221                 if ((enabled_port_mask & (1 << portid)) == 0) {
222                         printf("port %u is not enabled in port mask\n", portid);
223                         return -1;
224                 }
225                 if (!rte_eth_dev_is_valid_port(portid)) {
226                         printf("port %u is not present on the board\n", portid);
227                         return -1;
228                 }
229         }
230         return 0;
231 }
232
233 static uint8_t
234 get_port_n_rx_queues(const uint16_t port)
235 {
236         int queue = -1;
237         uint16_t i;
238
239         for (i = 0; i < nb_lcore_params; ++i) {
240                 if (lcore_params[i].port_id == port) {
241                         if (lcore_params[i].queue_id == queue+1)
242                                 queue = lcore_params[i].queue_id;
243                         else
244                                 rte_exit(EXIT_FAILURE, "queue ids of the port %d must be"
245                                                 " in sequence and must start with 0\n",
246                                                 lcore_params[i].port_id);
247                 }
248         }
249         return (uint8_t)(++queue);
250 }
251
252 static int
253 init_lcore_rx_queues(void)
254 {
255         uint16_t i, nb_rx_queue;
256         uint8_t lcore;
257
258         for (i = 0; i < nb_lcore_params; ++i) {
259                 lcore = lcore_params[i].lcore_id;
260                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
261                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
262                         printf("error: too many queues (%u) for lcore: %u\n",
263                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
264                         return -1;
265                 } else {
266                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
267                                 lcore_params[i].port_id;
268                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
269                                 lcore_params[i].queue_id;
270                         lcore_conf[lcore].n_rx_queue++;
271                 }
272         }
273         return 0;
274 }
275
276 /* display usage */
277 static void
278 print_usage(const char *prgname)
279 {
280         fprintf(stderr, "%s [EAL options] --"
281                 " -p PORTMASK"
282                 " [-P]"
283                 " [-E]"
284                 " [-L]"
285                 " --config (port,queue,lcore)[,(port,queue,lcore)]"
286                 " [--eth-dest=X,MM:MM:MM:MM:MM:MM]"
287                 " [--enable-jumbo [--max-pkt-len PKTLEN]]"
288                 " [--no-numa]"
289                 " [--hash-entry-num]"
290                 " [--ipv6]"
291                 " [--parse-ptype]"
292                 " [--per-port-pool]\n\n"
293
294                 "  -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
295                 "  -P : Enable promiscuous mode\n"
296                 "  -E : Enable exact match\n"
297                 "  -L : Enable longest prefix match (default)\n"
298                 "  --config (port,queue,lcore): Rx queue configuration\n"
299                 "  --eth-dest=X,MM:MM:MM:MM:MM:MM: Ethernet destination for port X\n"
300                 "  --enable-jumbo: Enable jumbo frames\n"
301                 "  --max-pkt-len: Under the premise of enabling jumbo,\n"
302                 "                 maximum packet length in decimal (64-9600)\n"
303                 "  --no-numa: Disable numa awareness\n"
304                 "  --hash-entry-num: Specify the hash entry number in hexadecimal to be setup\n"
305                 "  --ipv6: Set if running ipv6 packets\n"
306                 "  --parse-ptype: Set to use software to analyze packet type\n"
307                 "  --per-port-pool: Use separate buffer pool per port\n\n",
308                 prgname);
309 }
310
311 static int
312 parse_max_pkt_len(const char *pktlen)
313 {
314         char *end = NULL;
315         unsigned long len;
316
317         /* parse decimal string */
318         len = strtoul(pktlen, &end, 10);
319         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
320                 return -1;
321
322         if (len == 0)
323                 return -1;
324
325         return len;
326 }
327
328 static int
329 parse_portmask(const char *portmask)
330 {
331         char *end = NULL;
332         unsigned long pm;
333
334         /* parse hexadecimal string */
335         pm = strtoul(portmask, &end, 16);
336         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
337                 return -1;
338
339         if (pm == 0)
340                 return -1;
341
342         return pm;
343 }
344
345 static int
346 parse_hash_entry_number(const char *hash_entry_num)
347 {
348         char *end = NULL;
349         unsigned long hash_en;
350         /* parse hexadecimal string */
351         hash_en = strtoul(hash_entry_num, &end, 16);
352         if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
353                 return -1;
354
355         if (hash_en == 0)
356                 return -1;
357
358         return hash_en;
359 }
360
361 static int
362 parse_config(const char *q_arg)
363 {
364         char s[256];
365         const char *p, *p0 = q_arg;
366         char *end;
367         enum fieldnames {
368                 FLD_PORT = 0,
369                 FLD_QUEUE,
370                 FLD_LCORE,
371                 _NUM_FLD
372         };
373         unsigned long int_fld[_NUM_FLD];
374         char *str_fld[_NUM_FLD];
375         int i;
376         unsigned size;
377
378         nb_lcore_params = 0;
379
380         while ((p = strchr(p0,'(')) != NULL) {
381                 ++p;
382                 if((p0 = strchr(p,')')) == NULL)
383                         return -1;
384
385                 size = p0 - p;
386                 if(size >= sizeof(s))
387                         return -1;
388
389                 snprintf(s, sizeof(s), "%.*s", size, p);
390                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
391                         return -1;
392                 for (i = 0; i < _NUM_FLD; i++){
393                         errno = 0;
394                         int_fld[i] = strtoul(str_fld[i], &end, 0);
395                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
396                                 return -1;
397                 }
398                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
399                         printf("exceeded max number of lcore params: %hu\n",
400                                 nb_lcore_params);
401                         return -1;
402                 }
403                 lcore_params_array[nb_lcore_params].port_id =
404                         (uint8_t)int_fld[FLD_PORT];
405                 lcore_params_array[nb_lcore_params].queue_id =
406                         (uint8_t)int_fld[FLD_QUEUE];
407                 lcore_params_array[nb_lcore_params].lcore_id =
408                         (uint8_t)int_fld[FLD_LCORE];
409                 ++nb_lcore_params;
410         }
411         lcore_params = lcore_params_array;
412         return 0;
413 }
414
415 static void
416 parse_eth_dest(const char *optarg)
417 {
418         uint16_t portid;
419         char *port_end;
420         uint8_t c, *dest, peer_addr[6];
421
422         errno = 0;
423         portid = strtoul(optarg, &port_end, 10);
424         if (errno != 0 || port_end == optarg || *port_end++ != ',')
425                 rte_exit(EXIT_FAILURE,
426                 "Invalid eth-dest: %s", optarg);
427         if (portid >= RTE_MAX_ETHPORTS)
428                 rte_exit(EXIT_FAILURE,
429                 "eth-dest: port %d >= RTE_MAX_ETHPORTS(%d)\n",
430                 portid, RTE_MAX_ETHPORTS);
431
432         if (cmdline_parse_etheraddr(NULL, port_end,
433                 &peer_addr, sizeof(peer_addr)) < 0)
434                 rte_exit(EXIT_FAILURE,
435                 "Invalid ethernet address: %s\n",
436                 port_end);
437         dest = (uint8_t *)&dest_eth_addr[portid];
438         for (c = 0; c < 6; c++)
439                 dest[c] = peer_addr[c];
440         *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
441 }
442
443 #define MAX_JUMBO_PKT_LEN  9600
444 #define MEMPOOL_CACHE_SIZE 256
445
446 static const char short_options[] =
447         "p:"  /* portmask */
448         "P"   /* promiscuous */
449         "L"   /* enable long prefix match */
450         "E"   /* enable exact match */
451         ;
452
453 #define CMD_LINE_OPT_CONFIG "config"
454 #define CMD_LINE_OPT_ETH_DEST "eth-dest"
455 #define CMD_LINE_OPT_NO_NUMA "no-numa"
456 #define CMD_LINE_OPT_IPV6 "ipv6"
457 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
458 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
459 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
460 #define CMD_LINE_OPT_PER_PORT_POOL "per-port-pool"
461 enum {
462         /* long options mapped to a short option */
463
464         /* first long only option value must be >= 256, so that we won't
465          * conflict with short options */
466         CMD_LINE_OPT_MIN_NUM = 256,
467         CMD_LINE_OPT_CONFIG_NUM,
468         CMD_LINE_OPT_ETH_DEST_NUM,
469         CMD_LINE_OPT_NO_NUMA_NUM,
470         CMD_LINE_OPT_IPV6_NUM,
471         CMD_LINE_OPT_ENABLE_JUMBO_NUM,
472         CMD_LINE_OPT_HASH_ENTRY_NUM_NUM,
473         CMD_LINE_OPT_PARSE_PTYPE_NUM,
474         CMD_LINE_OPT_PARSE_PER_PORT_POOL,
475 };
476
477 static const struct option lgopts[] = {
478         {CMD_LINE_OPT_CONFIG, 1, 0, CMD_LINE_OPT_CONFIG_NUM},
479         {CMD_LINE_OPT_ETH_DEST, 1, 0, CMD_LINE_OPT_ETH_DEST_NUM},
480         {CMD_LINE_OPT_NO_NUMA, 0, 0, CMD_LINE_OPT_NO_NUMA_NUM},
481         {CMD_LINE_OPT_IPV6, 0, 0, CMD_LINE_OPT_IPV6_NUM},
482         {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, CMD_LINE_OPT_ENABLE_JUMBO_NUM},
483         {CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, CMD_LINE_OPT_HASH_ENTRY_NUM_NUM},
484         {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, CMD_LINE_OPT_PARSE_PTYPE_NUM},
485         {CMD_LINE_OPT_PER_PORT_POOL, 0, 0, CMD_LINE_OPT_PARSE_PER_PORT_POOL},
486         {NULL, 0, 0, 0}
487 };
488
489 /*
490  * This expression is used to calculate the number of mbufs needed
491  * depending on user input, taking  into account memory for rx and
492  * tx hardware rings, cache per lcore and mtable per port per lcore.
493  * RTE_MAX is used to ensure that NB_MBUF never goes below a minimum
494  * value of 8192
495  */
496 #define NB_MBUF(nports) RTE_MAX(        \
497         (nports*nb_rx_queue*nb_rxd +            \
498         nports*nb_lcores*MAX_PKT_BURST +        \
499         nports*n_tx_queue*nb_txd +              \
500         nb_lcores*MEMPOOL_CACHE_SIZE),          \
501         (unsigned)8192)
502
503 /* Parse the argument given in the command line of the application */
504 static int
505 parse_args(int argc, char **argv)
506 {
507         int opt, ret;
508         char **argvopt;
509         int option_index;
510         char *prgname = argv[0];
511
512         argvopt = argv;
513
514         /* Error or normal output strings. */
515         while ((opt = getopt_long(argc, argvopt, short_options,
516                                 lgopts, &option_index)) != EOF) {
517
518                 switch (opt) {
519                 /* portmask */
520                 case 'p':
521                         enabled_port_mask = parse_portmask(optarg);
522                         if (enabled_port_mask == 0) {
523                                 fprintf(stderr, "Invalid portmask\n");
524                                 print_usage(prgname);
525                                 return -1;
526                         }
527                         break;
528
529                 case 'P':
530                         promiscuous_on = 1;
531                         break;
532
533                 case 'E':
534                         l3fwd_em_on = 1;
535                         break;
536
537                 case 'L':
538                         l3fwd_lpm_on = 1;
539                         break;
540
541                 /* long options */
542                 case CMD_LINE_OPT_CONFIG_NUM:
543                         ret = parse_config(optarg);
544                         if (ret) {
545                                 fprintf(stderr, "Invalid config\n");
546                                 print_usage(prgname);
547                                 return -1;
548                         }
549                         break;
550
551                 case CMD_LINE_OPT_ETH_DEST_NUM:
552                         parse_eth_dest(optarg);
553                         break;
554
555                 case CMD_LINE_OPT_NO_NUMA_NUM:
556                         numa_on = 0;
557                         break;
558
559                 case CMD_LINE_OPT_IPV6_NUM:
560                         ipv6 = 1;
561                         break;
562
563                 case CMD_LINE_OPT_ENABLE_JUMBO_NUM: {
564                         const struct option lenopts = {
565                                 "max-pkt-len", required_argument, 0, 0
566                         };
567
568                         port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
569                         port_conf.txmode.offloads |= DEV_TX_OFFLOAD_MULTI_SEGS;
570
571                         /*
572                          * if no max-pkt-len set, use the default
573                          * value ETHER_MAX_LEN.
574                          */
575                         if (getopt_long(argc, argvopt, "",
576                                         &lenopts, &option_index) == 0) {
577                                 ret = parse_max_pkt_len(optarg);
578                                 if (ret < 64 || ret > MAX_JUMBO_PKT_LEN) {
579                                         fprintf(stderr,
580                                                 "invalid maximum packet length\n");
581                                         print_usage(prgname);
582                                         return -1;
583                                 }
584                                 port_conf.rxmode.max_rx_pkt_len = ret;
585                         }
586                         break;
587                 }
588
589                 case CMD_LINE_OPT_HASH_ENTRY_NUM_NUM:
590                         ret = parse_hash_entry_number(optarg);
591                         if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
592                                 hash_entry_number = ret;
593                         } else {
594                                 fprintf(stderr, "invalid hash entry number\n");
595                                 print_usage(prgname);
596                                 return -1;
597                         }
598                         break;
599
600                 case CMD_LINE_OPT_PARSE_PTYPE_NUM:
601                         printf("soft parse-ptype is enabled\n");
602                         parse_ptype = 1;
603                         break;
604
605                 case CMD_LINE_OPT_PARSE_PER_PORT_POOL:
606                         printf("per port buffer pool is enabled\n");
607                         per_port_pool = 1;
608                         break;
609
610                 default:
611                         print_usage(prgname);
612                         return -1;
613                 }
614         }
615
616         /* If both LPM and EM are selected, return error. */
617         if (l3fwd_lpm_on && l3fwd_em_on) {
618                 fprintf(stderr, "LPM and EM are mutually exclusive, select only one\n");
619                 return -1;
620         }
621
622         /*
623          * Nothing is selected, pick longest-prefix match
624          * as default match.
625          */
626         if (!l3fwd_lpm_on && !l3fwd_em_on) {
627                 fprintf(stderr, "LPM or EM none selected, default LPM on\n");
628                 l3fwd_lpm_on = 1;
629         }
630
631         /*
632          * ipv6 and hash flags are valid only for
633          * exact macth, reset them to default for
634          * longest-prefix match.
635          */
636         if (l3fwd_lpm_on) {
637                 ipv6 = 0;
638                 hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
639         }
640
641         if (optind >= 0)
642                 argv[optind-1] = prgname;
643
644         ret = optind-1;
645         optind = 1; /* reset getopt lib */
646         return ret;
647 }
648
649 static void
650 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
651 {
652         char buf[ETHER_ADDR_FMT_SIZE];
653         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
654         printf("%s%s", name, buf);
655 }
656
657 static int
658 init_mem(uint16_t portid, unsigned int nb_mbuf)
659 {
660         struct lcore_conf *qconf;
661         int socketid;
662         unsigned lcore_id;
663         char s[64];
664
665         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
666                 if (rte_lcore_is_enabled(lcore_id) == 0)
667                         continue;
668
669                 if (numa_on)
670                         socketid = rte_lcore_to_socket_id(lcore_id);
671                 else
672                         socketid = 0;
673
674                 if (socketid >= NB_SOCKETS) {
675                         rte_exit(EXIT_FAILURE,
676                                 "Socket %d of lcore %u is out of range %d\n",
677                                 socketid, lcore_id, NB_SOCKETS);
678                 }
679
680                 if (pktmbuf_pool[portid][socketid] == NULL) {
681                         snprintf(s, sizeof(s), "mbuf_pool_%d:%d",
682                                  portid, socketid);
683                         pktmbuf_pool[portid][socketid] =
684                                 rte_pktmbuf_pool_create(s, nb_mbuf,
685                                         MEMPOOL_CACHE_SIZE, 0,
686                                         RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
687                         if (pktmbuf_pool[portid][socketid] == NULL)
688                                 rte_exit(EXIT_FAILURE,
689                                         "Cannot init mbuf pool on socket %d\n",
690                                         socketid);
691                         else
692                                 printf("Allocated mbuf pool on socket %d\n",
693                                         socketid);
694
695                         /* Setup either LPM or EM(f.e Hash). But, only once per
696                          * available socket.
697                          */
698                         if (!lkp_per_socket[socketid]) {
699                                 l3fwd_lkp.setup(socketid);
700                                 lkp_per_socket[socketid] = 1;
701                         }
702                 }
703                 qconf = &lcore_conf[lcore_id];
704                 qconf->ipv4_lookup_struct =
705                         l3fwd_lkp.get_ipv4_lookup_struct(socketid);
706                 qconf->ipv6_lookup_struct =
707                         l3fwd_lkp.get_ipv6_lookup_struct(socketid);
708         }
709         return 0;
710 }
711
712 /* Check the link status of all ports in up to 9s, and print them finally */
713 static void
714 check_all_ports_link_status(uint32_t port_mask)
715 {
716 #define CHECK_INTERVAL 100 /* 100ms */
717 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
718         uint16_t portid;
719         uint8_t count, all_ports_up, print_flag = 0;
720         struct rte_eth_link link;
721
722         printf("\nChecking link status");
723         fflush(stdout);
724         for (count = 0; count <= MAX_CHECK_TIME; count++) {
725                 if (force_quit)
726                         return;
727                 all_ports_up = 1;
728                 RTE_ETH_FOREACH_DEV(portid) {
729                         if (force_quit)
730                                 return;
731                         if ((port_mask & (1 << portid)) == 0)
732                                 continue;
733                         memset(&link, 0, sizeof(link));
734                         rte_eth_link_get_nowait(portid, &link);
735                         /* print link status if flag set */
736                         if (print_flag == 1) {
737                                 if (link.link_status)
738                                         printf(
739                                         "Port%d Link Up. Speed %u Mbps -%s\n",
740                                                 portid, link.link_speed,
741                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
742                                         ("full-duplex") : ("half-duplex\n"));
743                                 else
744                                         printf("Port %d Link Down\n", portid);
745                                 continue;
746                         }
747                         /* clear all_ports_up flag if any link down */
748                         if (link.link_status == ETH_LINK_DOWN) {
749                                 all_ports_up = 0;
750                                 break;
751                         }
752                 }
753                 /* after finally printing all link status, get out */
754                 if (print_flag == 1)
755                         break;
756
757                 if (all_ports_up == 0) {
758                         printf(".");
759                         fflush(stdout);
760                         rte_delay_ms(CHECK_INTERVAL);
761                 }
762
763                 /* set the print_flag if all ports up or timeout */
764                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
765                         print_flag = 1;
766                         printf("done\n");
767                 }
768         }
769 }
770
771 static void
772 signal_handler(int signum)
773 {
774         if (signum == SIGINT || signum == SIGTERM) {
775                 printf("\n\nSignal %d received, preparing to exit...\n",
776                                 signum);
777                 force_quit = true;
778         }
779 }
780
781 static int
782 prepare_ptype_parser(uint16_t portid, uint16_t queueid)
783 {
784         if (parse_ptype) {
785                 printf("Port %d: softly parse packet type info\n", portid);
786                 if (rte_eth_add_rx_callback(portid, queueid,
787                                             l3fwd_lkp.cb_parse_ptype,
788                                             NULL))
789                         return 1;
790
791                 printf("Failed to add rx callback: port=%d\n", portid);
792                 return 0;
793         }
794
795         if (l3fwd_lkp.check_ptype(portid))
796                 return 1;
797
798         printf("port %d cannot parse packet type, please add --%s\n",
799                portid, CMD_LINE_OPT_PARSE_PTYPE);
800         return 0;
801 }
802
803 int
804 main(int argc, char **argv)
805 {
806         struct lcore_conf *qconf;
807         struct rte_eth_dev_info dev_info;
808         struct rte_eth_txconf *txconf;
809         int ret;
810         unsigned nb_ports;
811         uint16_t queueid, portid;
812         unsigned lcore_id;
813         uint32_t n_tx_queue, nb_lcores;
814         uint8_t nb_rx_queue, queue, socketid;
815
816         /* init EAL */
817         ret = rte_eal_init(argc, argv);
818         if (ret < 0)
819                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
820         argc -= ret;
821         argv += ret;
822
823         force_quit = false;
824         signal(SIGINT, signal_handler);
825         signal(SIGTERM, signal_handler);
826
827         /* pre-init dst MACs for all ports to 02:00:00:00:00:xx */
828         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
829                 dest_eth_addr[portid] =
830                         ETHER_LOCAL_ADMIN_ADDR + ((uint64_t)portid << 40);
831                 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
832         }
833
834         /* parse application arguments (after the EAL ones) */
835         ret = parse_args(argc, argv);
836         if (ret < 0)
837                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
838
839         if (check_lcore_params() < 0)
840                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
841
842         ret = init_lcore_rx_queues();
843         if (ret < 0)
844                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
845
846         nb_ports = rte_eth_dev_count_avail();
847
848         if (check_port_config() < 0)
849                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
850
851         nb_lcores = rte_lcore_count();
852
853         /* Setup function pointers for lookup method. */
854         setup_l3fwd_lookup_tables();
855
856         /* initialize all ports */
857         RTE_ETH_FOREACH_DEV(portid) {
858                 struct rte_eth_conf local_port_conf = port_conf;
859
860                 /* skip ports that are not enabled */
861                 if ((enabled_port_mask & (1 << portid)) == 0) {
862                         printf("\nSkipping disabled port %d\n", portid);
863                         continue;
864                 }
865
866                 /* init port */
867                 printf("Initializing port %d ... ", portid );
868                 fflush(stdout);
869
870                 nb_rx_queue = get_port_n_rx_queues(portid);
871                 n_tx_queue = nb_lcores;
872                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
873                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
874                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
875                         nb_rx_queue, (unsigned)n_tx_queue );
876
877                 rte_eth_dev_info_get(portid, &dev_info);
878                 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
879                         local_port_conf.txmode.offloads |=
880                                 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
881
882                 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
883                         dev_info.flow_type_rss_offloads;
884                 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
885                                 port_conf.rx_adv_conf.rss_conf.rss_hf) {
886                         printf("Port %u modified RSS hash function based on hardware support,"
887                                 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
888                                 portid,
889                                 port_conf.rx_adv_conf.rss_conf.rss_hf,
890                                 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
891                 }
892
893                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
894                                         (uint16_t)n_tx_queue, &local_port_conf);
895                 if (ret < 0)
896                         rte_exit(EXIT_FAILURE,
897                                 "Cannot configure device: err=%d, port=%d\n",
898                                 ret, portid);
899
900                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
901                                                        &nb_txd);
902                 if (ret < 0)
903                         rte_exit(EXIT_FAILURE,
904                                  "Cannot adjust number of descriptors: err=%d, "
905                                  "port=%d\n", ret, portid);
906
907                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
908                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
909                 printf(", ");
910                 print_ethaddr("Destination:",
911                         (const struct rte_ether_addr *)&dest_eth_addr[portid]);
912                 printf(", ");
913
914                 /*
915                  * prepare src MACs for each port.
916                  */
917                 ether_addr_copy(&ports_eth_addr[portid],
918                         (struct rte_ether_addr *)(val_eth + portid) + 1);
919
920                 /* init memory */
921                 if (!per_port_pool) {
922                         /* portid = 0; this is *not* signifying the first port,
923                          * rather, it signifies that portid is ignored.
924                          */
925                         ret = init_mem(0, NB_MBUF(nb_ports));
926                 } else {
927                         ret = init_mem(portid, NB_MBUF(1));
928                 }
929                 if (ret < 0)
930                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
931
932                 /* init one TX queue per couple (lcore,port) */
933                 queueid = 0;
934                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
935                         if (rte_lcore_is_enabled(lcore_id) == 0)
936                                 continue;
937
938                         if (numa_on)
939                                 socketid =
940                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
941                         else
942                                 socketid = 0;
943
944                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
945                         fflush(stdout);
946
947                         txconf = &dev_info.default_txconf;
948                         txconf->offloads = local_port_conf.txmode.offloads;
949                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
950                                                      socketid, txconf);
951                         if (ret < 0)
952                                 rte_exit(EXIT_FAILURE,
953                                         "rte_eth_tx_queue_setup: err=%d, "
954                                         "port=%d\n", ret, portid);
955
956                         qconf = &lcore_conf[lcore_id];
957                         qconf->tx_queue_id[portid] = queueid;
958                         queueid++;
959
960                         qconf->tx_port_id[qconf->n_tx_port] = portid;
961                         qconf->n_tx_port++;
962                 }
963                 printf("\n");
964         }
965
966         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
967                 if (rte_lcore_is_enabled(lcore_id) == 0)
968                         continue;
969                 qconf = &lcore_conf[lcore_id];
970                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
971                 fflush(stdout);
972                 /* init RX queues */
973                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
974                         struct rte_eth_dev *dev;
975                         struct rte_eth_conf *conf;
976                         struct rte_eth_rxconf rxq_conf;
977
978                         portid = qconf->rx_queue_list[queue].port_id;
979                         queueid = qconf->rx_queue_list[queue].queue_id;
980                         dev = &rte_eth_devices[portid];
981                         conf = &dev->data->dev_conf;
982
983                         if (numa_on)
984                                 socketid =
985                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
986                         else
987                                 socketid = 0;
988
989                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
990                         fflush(stdout);
991
992                         rte_eth_dev_info_get(portid, &dev_info);
993                         rxq_conf = dev_info.default_rxconf;
994                         rxq_conf.offloads = conf->rxmode.offloads;
995                         if (!per_port_pool)
996                                 ret = rte_eth_rx_queue_setup(portid, queueid,
997                                                 nb_rxd, socketid,
998                                                 &rxq_conf,
999                                                 pktmbuf_pool[0][socketid]);
1000                         else
1001                                 ret = rte_eth_rx_queue_setup(portid, queueid,
1002                                                 nb_rxd, socketid,
1003                                                 &rxq_conf,
1004                                                 pktmbuf_pool[portid][socketid]);
1005                         if (ret < 0)
1006                                 rte_exit(EXIT_FAILURE,
1007                                 "rte_eth_rx_queue_setup: err=%d, port=%d\n",
1008                                 ret, portid);
1009                 }
1010         }
1011
1012         printf("\n");
1013
1014         /* start ports */
1015         RTE_ETH_FOREACH_DEV(portid) {
1016                 if ((enabled_port_mask & (1 << portid)) == 0) {
1017                         continue;
1018                 }
1019                 /* Start device */
1020                 ret = rte_eth_dev_start(portid);
1021                 if (ret < 0)
1022                         rte_exit(EXIT_FAILURE,
1023                                 "rte_eth_dev_start: err=%d, port=%d\n",
1024                                 ret, portid);
1025
1026                 /*
1027                  * If enabled, put device in promiscuous mode.
1028                  * This allows IO forwarding mode to forward packets
1029                  * to itself through 2 cross-connected  ports of the
1030                  * target machine.
1031                  */
1032                 if (promiscuous_on)
1033                         rte_eth_promiscuous_enable(portid);
1034         }
1035
1036         printf("\n");
1037
1038         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1039                 if (rte_lcore_is_enabled(lcore_id) == 0)
1040                         continue;
1041                 qconf = &lcore_conf[lcore_id];
1042                 for (queue = 0; queue < qconf->n_rx_queue; ++queue) {
1043                         portid = qconf->rx_queue_list[queue].port_id;
1044                         queueid = qconf->rx_queue_list[queue].queue_id;
1045                         if (prepare_ptype_parser(portid, queueid) == 0)
1046                                 rte_exit(EXIT_FAILURE, "ptype check fails\n");
1047                 }
1048         }
1049
1050
1051         check_all_ports_link_status(enabled_port_mask);
1052
1053         ret = 0;
1054         /* launch per-lcore init on every lcore */
1055         rte_eal_mp_remote_launch(l3fwd_lkp.main_loop, NULL, CALL_MASTER);
1056         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1057                 if (rte_eal_wait_lcore(lcore_id) < 0) {
1058                         ret = -1;
1059                         break;
1060                 }
1061         }
1062
1063         /* stop ports */
1064         RTE_ETH_FOREACH_DEV(portid) {
1065                 if ((enabled_port_mask & (1 << portid)) == 0)
1066                         continue;
1067                 printf("Closing port %d...", portid);
1068                 rte_eth_dev_stop(portid);
1069                 rte_eth_dev_close(portid);
1070                 printf(" Done\n");
1071         }
1072         printf("Bye...\n");
1073
1074         return ret;
1075 }