1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2014 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 <sys/param.h>
17
18 #include <rte_common.h>
19 #include <rte_byteorder.h>
20 #include <rte_log.h>
21 #include <rte_memory.h>
22 #include <rte_memcpy.h>
23 #include <rte_eal.h>
24 #include <rte_launch.h>
25 #include <rte_atomic.h>
26 #include <rte_cycles.h>
27 #include <rte_prefetch.h>
28 #include <rte_lcore.h>
29 #include <rte_per_lcore.h>
30 #include <rte_branch_prediction.h>
31 #include <rte_interrupts.h>
32 #include <rte_random.h>
33 #include <rte_debug.h>
34 #include <rte_ether.h>
35 #include <rte_ethdev.h>
36 #include <rte_mempool.h>
37 #include <rte_mbuf.h>
38 #include <rte_malloc.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_lpm.h>
44 #include <rte_lpm6.h>
45
46 #include <rte_ip_frag.h>
47
48 #define MAX_PKT_BURST 32
49
50
51 #define RTE_LOGTYPE_IP_RSMBL RTE_LOGTYPE_USER1
52
53 #define MAX_JUMBO_PKT_LEN 9600
54
55 #define BUF_SIZE RTE_MBUF_DEFAULT_DATAROOM
56 #define MBUF_DATA_SIZE RTE_MBUF_DEFAULT_BUF_SIZE
57
58 #define NB_MBUF 8192
59 #define MEMPOOL_CACHE_SIZE 256
60
61 /* allow max jumbo frame 9.5 KB */
62 #define JUMBO_FRAME_MAX_SIZE 0x2600
63
64 #define MAX_FLOW_NUM UINT16_MAX
65 #define MIN_FLOW_NUM 1
66 #define DEF_FLOW_NUM 0x1000
67
68 /* TTL numbers are in ms. */
69 #define MAX_FLOW_TTL (3600 * MS_PER_S)
70 #define MIN_FLOW_TTL 1
71 #define DEF_FLOW_TTL MS_PER_S
72
73 #define MAX_FRAG_NUM RTE_LIBRTE_IP_FRAG_MAX_FRAG
74
75 /* Should be power of two. */
76 #define IP_FRAG_TBL_BUCKET_ENTRIES 16
77
78 static uint32_t max_flow_num = DEF_FLOW_NUM;
79 static uint32_t max_flow_ttl = DEF_FLOW_TTL;
80
81 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
82
83 #define NB_SOCKETS 8
84
85 /* Configure how many packets ahead to prefetch, when reading packets */
86 #define PREFETCH_OFFSET 3
87
88 /*
89 * Configurable number of RX/TX ring descriptors
90 */
91 #define RTE_TEST_RX_DESC_DEFAULT 1024
92 #define RTE_TEST_TX_DESC_DEFAULT 1024
93
94 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
95 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
96
97 /* ethernet addresses of ports */
98 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
99
100 #ifndef IPv4_BYTES
101 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
102 #define IPv4_BYTES(addr) \
103 (uint8_t) (((addr) >> 24) & 0xFF),\
104 (uint8_t) (((addr) >> 16) & 0xFF),\
105 (uint8_t) (((addr) >> 8) & 0xFF),\
106 (uint8_t) ((addr) & 0xFF)
107 #endif
108
109 #ifndef IPv6_BYTES
110 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
111 "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
112 #define IPv6_BYTES(addr) \
113 addr[0], addr[1], addr[2], addr[3], \
114 addr[4], addr[5], addr[6], addr[7], \
115 addr[8], addr[9], addr[10], addr[11],\
116 addr[12], addr[13],addr[14], addr[15]
117 #endif
118
119 #define IPV6_ADDR_LEN 16
120
121 /* mask of enabled ports */
122 static uint32_t enabled_port_mask = 0;
123
124 static int rx_queue_per_lcore = 1;
125
126 struct mbuf_table {
127 uint32_t len;
128 uint32_t head;
129 uint32_t tail;
130 struct rte_mbuf *m_table[0];
131 };
132
133 struct rx_queue {
134 struct rte_ip_frag_tbl *frag_tbl;
135 struct rte_mempool *pool;
136 struct rte_lpm *lpm;
137 struct rte_lpm6 *lpm6;
138 uint16_t portid;
139 };
140
141 struct tx_lcore_stat {
142 uint64_t call;
143 uint64_t drop;
144 uint64_t queue;
145 uint64_t send;
146 };
147
148 #define MAX_RX_QUEUE_PER_LCORE 16
149 #define MAX_TX_QUEUE_PER_PORT 16
150 #define MAX_RX_QUEUE_PER_PORT 128
151
152 struct lcore_queue_conf {
153 uint16_t n_rx_queue;
154 struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
155 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
156 struct rte_ip_frag_death_row death_row;
157 struct mbuf_table *tx_mbufs[RTE_MAX_ETHPORTS];
158 struct tx_lcore_stat tx_stat;
159 } __rte_cache_aligned;
160 static struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
161
162 static struct rte_eth_conf port_conf = {
163 .rxmode = {
164 .mq_mode = ETH_MQ_RX_RSS,
165 .max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
166 .split_hdr_size = 0,
167 .offloads = (DEV_RX_OFFLOAD_CHECKSUM |
168 DEV_RX_OFFLOAD_JUMBO_FRAME),
169 },
170 .rx_adv_conf = {
171 .rss_conf = {
172 .rss_key = NULL,
173 .rss_hf = ETH_RSS_IP,
174 },
175 },
176 .txmode = {
177 .mq_mode = ETH_MQ_TX_NONE,
178 .offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM |
179 DEV_TX_OFFLOAD_MULTI_SEGS),
180 },
181 };
182
183 /*
184 * IPv4 forwarding table
185 */
186 struct l3fwd_ipv4_route {
187 uint32_t ip;
188 uint8_t depth;
189 uint8_t if_out;
190 };
191
192 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
193 {RTE_IPV4(100,10,0,0), 16, 0},
194 {RTE_IPV4(100,20,0,0), 16, 1},
195 {RTE_IPV4(100,30,0,0), 16, 2},
196 {RTE_IPV4(100,40,0,0), 16, 3},
197 {RTE_IPV4(100,50,0,0), 16, 4},
198 {RTE_IPV4(100,60,0,0), 16, 5},
199 {RTE_IPV4(100,70,0,0), 16, 6},
200 {RTE_IPV4(100,80,0,0), 16, 7},
201 };
202
203 /*
204 * IPv6 forwarding table
205 */
206
207 struct l3fwd_ipv6_route {
208 uint8_t ip[IPV6_ADDR_LEN];
209 uint8_t depth;
210 uint8_t if_out;
211 };
212
213 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
214 {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
215 {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
216 {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
217 {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
218 {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
219 {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
220 {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
221 {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
222 };
223
224 #define LPM_MAX_RULES 1024
225 #define LPM6_MAX_RULES 1024
226 #define LPM6_NUMBER_TBL8S (1 << 16)
227
228 struct rte_lpm6_config lpm6_config = {
229 .max_rules = LPM6_MAX_RULES,
230 .number_tbl8s = LPM6_NUMBER_TBL8S,
231 .flags = 0
232 };
233
234 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
235 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
236
237 #ifdef RTE_LIBRTE_IP_FRAG_TBL_STAT
238 #define TX_LCORE_STAT_UPDATE(s, f, v) ((s)->f += (v))
239 #else
240 #define TX_LCORE_STAT_UPDATE(s, f, v) do {} while (0)
241 #endif /* RTE_LIBRTE_IP_FRAG_TBL_STAT */
242
243 /*
244 * If number of queued packets reached given threahold, then
245 * send burst of packets on an output interface.
246 */
247 static inline uint32_t
send_burst(struct lcore_queue_conf * qconf,uint32_t thresh,uint16_t port)248 send_burst(struct lcore_queue_conf *qconf, uint32_t thresh, uint16_t port)
249 {
250 uint32_t fill, len, k, n;
251 struct mbuf_table *txmb;
252
253 txmb = qconf->tx_mbufs[port];
254 len = txmb->len;
255
256 if ((int32_t)(fill = txmb->head - txmb->tail) < 0)
257 fill += len;
258
259 if (fill >= thresh) {
260 n = RTE_MIN(len - txmb->tail, fill);
261
262 k = rte_eth_tx_burst(port, qconf->tx_queue_id[port],
263 txmb->m_table + txmb->tail, (uint16_t)n);
264
265 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, call, 1);
266 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, send, k);
267
268 fill -= k;
269 if ((txmb->tail += k) == len)
270 txmb->tail = 0;
271 }
272
273 return fill;
274 }
275
276 /* Enqueue a single packet, and send burst if queue is filled */
277 static inline int
send_single_packet(struct rte_mbuf * m,uint16_t port)278 send_single_packet(struct rte_mbuf *m, uint16_t port)
279 {
280 uint32_t fill, lcore_id, len;
281 struct lcore_queue_conf *qconf;
282 struct mbuf_table *txmb;
283
284 lcore_id = rte_lcore_id();
285 qconf = &lcore_queue_conf[lcore_id];
286
287 txmb = qconf->tx_mbufs[port];
288 len = txmb->len;
289
290 fill = send_burst(qconf, MAX_PKT_BURST, port);
291
292 if (fill == len - 1) {
293 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, drop, 1);
294 rte_pktmbuf_free(txmb->m_table[txmb->tail]);
295 if (++txmb->tail == len)
296 txmb->tail = 0;
297 }
298
299 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, queue, 1);
300 txmb->m_table[txmb->head] = m;
301 if(++txmb->head == len)
302 txmb->head = 0;
303
304 return 0;
305 }
306
307 static inline void
reassemble(struct rte_mbuf * m,uint16_t portid,uint32_t queue,struct lcore_queue_conf * qconf,uint64_t tms)308 reassemble(struct rte_mbuf *m, uint16_t portid, uint32_t queue,
309 struct lcore_queue_conf *qconf, uint64_t tms)
310 {
311 struct rte_ether_hdr *eth_hdr;
312 struct rte_ip_frag_tbl *tbl;
313 struct rte_ip_frag_death_row *dr;
314 struct rx_queue *rxq;
315 void *d_addr_bytes;
316 uint32_t next_hop;
317 uint16_t dst_port;
318
319 rxq = &qconf->rx_queue_list[queue];
320
321 eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
322
323 dst_port = portid;
324
325 /* if packet is IPv4 */
326 if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
327 struct rte_ipv4_hdr *ip_hdr;
328 uint32_t ip_dst;
329
330 ip_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
331
332 /* if it is a fragmented packet, then try to reassemble. */
333 if (rte_ipv4_frag_pkt_is_fragmented(ip_hdr)) {
334 struct rte_mbuf *mo;
335
336 tbl = rxq->frag_tbl;
337 dr = &qconf->death_row;
338
339 /* prepare mbuf: setup l2_len/l3_len. */
340 m->l2_len = sizeof(*eth_hdr);
341 m->l3_len = sizeof(*ip_hdr);
342
343 /* process this fragment. */
344 mo = rte_ipv4_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr);
345 if (mo == NULL)
346 /* no packet to send out. */
347 return;
348
349 /* we have our packet reassembled. */
350 if (mo != m) {
351 m = mo;
352 eth_hdr = rte_pktmbuf_mtod(m,
353 struct rte_ether_hdr *);
354 ip_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
355 }
356
357 /* update offloading flags */
358 m->ol_flags |= (PKT_TX_IPV4 | PKT_TX_IP_CKSUM);
359 }
360 ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
361
362 /* Find destination port */
363 if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
364 (enabled_port_mask & 1 << next_hop) != 0) {
365 dst_port = next_hop;
366 }
367
368 eth_hdr->ether_type = rte_be_to_cpu_16(RTE_ETHER_TYPE_IPV4);
369 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
370 /* if packet is IPv6 */
371 struct ipv6_extension_fragment *frag_hdr;
372 struct rte_ipv6_hdr *ip_hdr;
373
374 ip_hdr = (struct rte_ipv6_hdr *)(eth_hdr + 1);
375
376 frag_hdr = rte_ipv6_frag_get_ipv6_fragment_header(ip_hdr);
377
378 if (frag_hdr != NULL) {
379 struct rte_mbuf *mo;
380
381 tbl = rxq->frag_tbl;
382 dr = &qconf->death_row;
383
384 /* prepare mbuf: setup l2_len/l3_len. */
385 m->l2_len = sizeof(*eth_hdr);
386 m->l3_len = sizeof(*ip_hdr) + sizeof(*frag_hdr);
387
388 mo = rte_ipv6_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr, frag_hdr);
389 if (mo == NULL)
390 return;
391
392 if (mo != m) {
393 m = mo;
394 eth_hdr = rte_pktmbuf_mtod(m,
395 struct rte_ether_hdr *);
396 ip_hdr = (struct rte_ipv6_hdr *)(eth_hdr + 1);
397 }
398 }
399
400 /* Find destination port */
401 if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr,
402 &next_hop) == 0 &&
403 (enabled_port_mask & 1 << next_hop) != 0) {
404 dst_port = next_hop;
405 }
406
407 eth_hdr->ether_type = rte_be_to_cpu_16(RTE_ETHER_TYPE_IPV6);
408 }
409 /* if packet wasn't IPv4 or IPv6, it's forwarded to the port it came from */
410
411 /* 02:00:00:00:00:xx */
412 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0];
413 *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
414
415 /* src addr */
416 rte_ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr);
417
418 send_single_packet(m, dst_port);
419 }
420
421 /* main processing loop */
422 static int
main_loop(__rte_unused void * dummy)423 main_loop(__rte_unused void *dummy)
424 {
425 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
426 unsigned lcore_id;
427 uint64_t diff_tsc, cur_tsc, prev_tsc;
428 int i, j, nb_rx;
429 uint16_t portid;
430 struct lcore_queue_conf *qconf;
431 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
432
433 prev_tsc = 0;
434
435 lcore_id = rte_lcore_id();
436 qconf = &lcore_queue_conf[lcore_id];
437
438 if (qconf->n_rx_queue == 0) {
439 RTE_LOG(INFO, IP_RSMBL, "lcore %u has nothing to do\n", lcore_id);
440 return 0;
441 }
442
443 RTE_LOG(INFO, IP_RSMBL, "entering main loop on lcore %u\n", lcore_id);
444
445 for (i = 0; i < qconf->n_rx_queue; i++) {
446
447 portid = qconf->rx_queue_list[i].portid;
448 RTE_LOG(INFO, IP_RSMBL, " -- lcoreid=%u portid=%u\n", lcore_id,
449 portid);
450 }
451
452 while (1) {
453
454 cur_tsc = rte_rdtsc();
455
456 /*
457 * TX burst queue drain
458 */
459 diff_tsc = cur_tsc - prev_tsc;
460 if (unlikely(diff_tsc > drain_tsc)) {
461
462 /*
463 * This could be optimized (use queueid instead of
464 * portid), but it is not called so often
465 */
466 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
467 if ((enabled_port_mask & (1 << portid)) != 0)
468 send_burst(qconf, 1, portid);
469 }
470
471 prev_tsc = cur_tsc;
472 }
473
474 /*
475 * Read packet from RX queues
476 */
477 for (i = 0; i < qconf->n_rx_queue; ++i) {
478
479 portid = qconf->rx_queue_list[i].portid;
480
481 nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
482 MAX_PKT_BURST);
483
484 /* Prefetch first packets */
485 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
486 rte_prefetch0(rte_pktmbuf_mtod(
487 pkts_burst[j], void *));
488 }
489
490 /* Prefetch and forward already prefetched packets */
491 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
492 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
493 j + PREFETCH_OFFSET], void *));
494 reassemble(pkts_burst[j], portid,
495 i, qconf, cur_tsc);
496 }
497
498 /* Forward remaining prefetched packets */
499 for (; j < nb_rx; j++) {
500 reassemble(pkts_burst[j], portid,
501 i, qconf, cur_tsc);
502 }
503
504 rte_ip_frag_free_death_row(&qconf->death_row,
505 PREFETCH_OFFSET);
506 }
507 }
508 }
509
510 /* display usage */
511 static void
print_usage(const char * prgname)512 print_usage(const char *prgname)
513 {
514 printf("%s [EAL options] -- -p PORTMASK [-q NQ]"
515 " [--max-pkt-len PKTLEN]"
516 " [--maxflows=<flows>] [--flowttl=<ttl>[(s|ms)]]\n"
517 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
518 " -q NQ: number of RX queues per lcore\n"
519 " --maxflows=<flows>: optional, maximum number of flows "
520 "supported\n"
521 " --flowttl=<ttl>[(s|ms)]: optional, maximum TTL for each "
522 "flow\n",
523 prgname);
524 }
525
526 static uint32_t
parse_flow_num(const char * str,uint32_t min,uint32_t max,uint32_t * val)527 parse_flow_num(const char *str, uint32_t min, uint32_t max, uint32_t *val)
528 {
529 char *end;
530 uint64_t v;
531
532 /* parse decimal string */
533 errno = 0;
534 v = strtoul(str, &end, 10);
535 if (errno != 0 || *end != '\0')
536 return -EINVAL;
537
538 if (v < min || v > max)
539 return -EINVAL;
540
541 *val = (uint32_t)v;
542 return 0;
543 }
544
545 static int
parse_flow_ttl(const char * str,uint32_t min,uint32_t max,uint32_t * val)546 parse_flow_ttl(const char *str, uint32_t min, uint32_t max, uint32_t *val)
547 {
548 char *end;
549 uint64_t v;
550
551 static const char frmt_sec[] = "s";
552 static const char frmt_msec[] = "ms";
553
554 /* parse decimal string */
555 errno = 0;
556 v = strtoul(str, &end, 10);
557 if (errno != 0)
558 return -EINVAL;
559
560 if (*end != '\0') {
561 if (strncmp(frmt_sec, end, sizeof(frmt_sec)) == 0)
562 v *= MS_PER_S;
563 else if (strncmp(frmt_msec, end, sizeof (frmt_msec)) != 0)
564 return -EINVAL;
565 }
566
567 if (v < min || v > max)
568 return -EINVAL;
569
570 *val = (uint32_t)v;
571 return 0;
572 }
573
574 static int
parse_portmask(const char * portmask)575 parse_portmask(const char *portmask)
576 {
577 char *end = NULL;
578 unsigned long pm;
579
580 /* parse hexadecimal string */
581 pm = strtoul(portmask, &end, 16);
582 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
583 return 0;
584
585 return pm;
586 }
587
588 static int
parse_nqueue(const char * q_arg)589 parse_nqueue(const char *q_arg)
590 {
591 char *end = NULL;
592 unsigned long n;
593
594 printf("%p\n", q_arg);
595
596 /* parse hexadecimal string */
597 n = strtoul(q_arg, &end, 10);
598 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
599 return -1;
600 if (n == 0)
601 return -1;
602 if (n >= MAX_RX_QUEUE_PER_LCORE)
603 return -1;
604
605 return n;
606 }
607
608 /* Parse the argument given in the command line of the application */
609 static int
parse_args(int argc,char ** argv)610 parse_args(int argc, char **argv)
611 {
612 int opt, ret;
613 char **argvopt;
614 int option_index;
615 char *prgname = argv[0];
616 static struct option lgopts[] = {
617 {"max-pkt-len", 1, 0, 0},
618 {"maxflows", 1, 0, 0},
619 {"flowttl", 1, 0, 0},
620 {NULL, 0, 0, 0}
621 };
622
623 argvopt = argv;
624
625 while ((opt = getopt_long(argc, argvopt, "p:q:",
626 lgopts, &option_index)) != EOF) {
627
628 switch (opt) {
629 /* portmask */
630 case 'p':
631 enabled_port_mask = parse_portmask(optarg);
632 if (enabled_port_mask == 0) {
633 printf("invalid portmask\n");
634 print_usage(prgname);
635 return -1;
636 }
637 break;
638
639 /* nqueue */
640 case 'q':
641 rx_queue_per_lcore = parse_nqueue(optarg);
642 if (rx_queue_per_lcore < 0) {
643 printf("invalid queue number\n");
644 print_usage(prgname);
645 return -1;
646 }
647 break;
648
649 /* long options */
650 case 0:
651 if (!strncmp(lgopts[option_index].name,
652 "maxflows", 8)) {
653 if ((ret = parse_flow_num(optarg, MIN_FLOW_NUM,
654 MAX_FLOW_NUM,
655 &max_flow_num)) != 0) {
656 printf("invalid value: \"%s\" for "
657 "parameter %s\n",
658 optarg,
659 lgopts[option_index].name);
660 print_usage(prgname);
661 return ret;
662 }
663 }
664
665 if (!strncmp(lgopts[option_index].name, "flowttl", 7)) {
666 if ((ret = parse_flow_ttl(optarg, MIN_FLOW_TTL,
667 MAX_FLOW_TTL,
668 &max_flow_ttl)) != 0) {
669 printf("invalid value: \"%s\" for "
670 "parameter %s\n",
671 optarg,
672 lgopts[option_index].name);
673 print_usage(prgname);
674 return ret;
675 }
676 }
677
678 break;
679
680 default:
681 print_usage(prgname);
682 return -1;
683 }
684 }
685
686 if (optind >= 0)
687 argv[optind-1] = prgname;
688
689 ret = optind-1;
690 optind = 1; /* reset getopt lib */
691 return ret;
692 }
693
694 static void
print_ethaddr(const char * name,const struct rte_ether_addr * eth_addr)695 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
696 {
697 char buf[RTE_ETHER_ADDR_FMT_SIZE];
698 rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
699 printf("%s%s", name, buf);
700 }
701
702 /* Check the link status of all ports in up to 9s, and print them finally */
703 static void
check_all_ports_link_status(uint32_t port_mask)704 check_all_ports_link_status(uint32_t port_mask)
705 {
706 #define CHECK_INTERVAL 100 /* 100ms */
707 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
708 uint16_t portid;
709 uint8_t count, all_ports_up, print_flag = 0;
710 struct rte_eth_link link;
711 int ret;
712 char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
713
714 printf("\nChecking link status");
715 fflush(stdout);
716 for (count = 0; count <= MAX_CHECK_TIME; count++) {
717 all_ports_up = 1;
718 RTE_ETH_FOREACH_DEV(portid) {
719 if ((port_mask & (1 << portid)) == 0)
720 continue;
721 memset(&link, 0, sizeof(link));
722 ret = rte_eth_link_get_nowait(portid, &link);
723 if (ret < 0) {
724 all_ports_up = 0;
725 if (print_flag == 1)
726 printf("Port %u link get failed: %s\n",
727 portid, rte_strerror(-ret));
728 continue;
729 }
730 /* print link status if flag set */
731 if (print_flag == 1) {
732 rte_eth_link_to_str(link_status_text,
733 sizeof(link_status_text), &link);
734 printf("Port %d %s\n", portid,
735 link_status_text);
736 continue;
737 }
738 /* clear all_ports_up flag if any link down */
739 if (link.link_status == ETH_LINK_DOWN) {
740 all_ports_up = 0;
741 break;
742 }
743 }
744 /* after finally printing all link status, get out */
745 if (print_flag == 1)
746 break;
747
748 if (all_ports_up == 0) {
749 printf(".");
750 fflush(stdout);
751 rte_delay_ms(CHECK_INTERVAL);
752 }
753
754 /* set the print_flag if all ports up or timeout */
755 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
756 print_flag = 1;
757 printf("\ndone\n");
758 }
759 }
760 }
761
762 static int
init_routing_table(void)763 init_routing_table(void)
764 {
765 struct rte_lpm *lpm;
766 struct rte_lpm6 *lpm6;
767 int socket, ret;
768 unsigned i;
769
770 for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
771 if (socket_lpm[socket]) {
772 lpm = socket_lpm[socket];
773 /* populate the LPM table */
774 for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
775 ret = rte_lpm_add(lpm,
776 l3fwd_ipv4_route_array[i].ip,
777 l3fwd_ipv4_route_array[i].depth,
778 l3fwd_ipv4_route_array[i].if_out);
779
780 if (ret < 0) {
781 RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
782 "LPM table\n", i);
783 return -1;
784 }
785
786 RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv4_BYTES_FMT
787 "/%d (port %d)\n",
788 socket,
789 IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
790 l3fwd_ipv4_route_array[i].depth,
791 l3fwd_ipv4_route_array[i].if_out);
792 }
793 }
794
795 if (socket_lpm6[socket]) {
796 lpm6 = socket_lpm6[socket];
797 /* populate the LPM6 table */
798 for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
799 ret = rte_lpm6_add(lpm6,
800 l3fwd_ipv6_route_array[i].ip,
801 l3fwd_ipv6_route_array[i].depth,
802 l3fwd_ipv6_route_array[i].if_out);
803
804 if (ret < 0) {
805 RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
806 "LPM6 table\n", i);
807 return -1;
808 }
809
810 RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv6_BYTES_FMT
811 "/%d (port %d)\n",
812 socket,
813 IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
814 l3fwd_ipv6_route_array[i].depth,
815 l3fwd_ipv6_route_array[i].if_out);
816 }
817 }
818 }
819 return 0;
820 }
821
822 static int
setup_port_tbl(struct lcore_queue_conf * qconf,uint32_t lcore,int socket,uint32_t port)823 setup_port_tbl(struct lcore_queue_conf *qconf, uint32_t lcore, int socket,
824 uint32_t port)
825 {
826 struct mbuf_table *mtb;
827 uint32_t n;
828 size_t sz;
829
830 n = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST);
831 sz = sizeof (*mtb) + sizeof (mtb->m_table[0]) * n;
832
833 if ((mtb = rte_zmalloc_socket(__func__, sz, RTE_CACHE_LINE_SIZE,
834 socket)) == NULL) {
835 RTE_LOG(ERR, IP_RSMBL, "%s() for lcore: %u, port: %u "
836 "failed to allocate %zu bytes\n",
837 __func__, lcore, port, sz);
838 return -1;
839 }
840
841 mtb->len = n;
842 qconf->tx_mbufs[port] = mtb;
843
844 return 0;
845 }
846
847 static int
setup_queue_tbl(struct rx_queue * rxq,uint32_t lcore,uint32_t queue)848 setup_queue_tbl(struct rx_queue *rxq, uint32_t lcore, uint32_t queue)
849 {
850 int socket;
851 uint32_t nb_mbuf;
852 uint64_t frag_cycles;
853 char buf[RTE_MEMPOOL_NAMESIZE];
854
855 socket = rte_lcore_to_socket_id(lcore);
856 if (socket == SOCKET_ID_ANY)
857 socket = 0;
858
859 frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) / MS_PER_S *
860 max_flow_ttl;
861
862 if ((rxq->frag_tbl = rte_ip_frag_table_create(max_flow_num,
863 IP_FRAG_TBL_BUCKET_ENTRIES, max_flow_num, frag_cycles,
864 socket)) == NULL) {
865 RTE_LOG(ERR, IP_RSMBL, "ip_frag_tbl_create(%u) on "
866 "lcore: %u for queue: %u failed\n",
867 max_flow_num, lcore, queue);
868 return -1;
869 }
870
871 /*
872 * At any given moment up to <max_flow_num * (MAX_FRAG_NUM)>
873 * mbufs could be stored int the fragment table.
874 * Plus, each TX queue can hold up to <max_flow_num> packets.
875 */
876
877 nb_mbuf = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST) * MAX_FRAG_NUM;
878 nb_mbuf *= (port_conf.rxmode.max_rx_pkt_len + BUF_SIZE - 1) / BUF_SIZE;
879 nb_mbuf *= 2; /* ipv4 and ipv6 */
880 nb_mbuf += nb_rxd + nb_txd;
881
882 nb_mbuf = RTE_MAX(nb_mbuf, (uint32_t)NB_MBUF);
883
884 snprintf(buf, sizeof(buf), "mbuf_pool_%u_%u", lcore, queue);
885
886 rxq->pool = rte_pktmbuf_pool_create(buf, nb_mbuf, MEMPOOL_CACHE_SIZE, 0,
887 MBUF_DATA_SIZE, socket);
888 if (rxq->pool == NULL) {
889 RTE_LOG(ERR, IP_RSMBL,
890 "rte_pktmbuf_pool_create(%s) failed", buf);
891 return -1;
892 }
893
894 return 0;
895 }
896
897 static int
init_mem(void)898 init_mem(void)
899 {
900 char buf[PATH_MAX];
901 struct rte_lpm *lpm;
902 struct rte_lpm6 *lpm6;
903 struct rte_lpm_config lpm_config;
904 int socket;
905 unsigned lcore_id;
906
907 /* traverse through lcores and initialize structures on each socket */
908
909 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
910
911 if (rte_lcore_is_enabled(lcore_id) == 0)
912 continue;
913
914 socket = rte_lcore_to_socket_id(lcore_id);
915
916 if (socket == SOCKET_ID_ANY)
917 socket = 0;
918
919 if (socket_lpm[socket] == NULL) {
920 RTE_LOG(INFO, IP_RSMBL, "Creating LPM table on socket %i\n", socket);
921 snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
922
923 lpm_config.max_rules = LPM_MAX_RULES;
924 lpm_config.number_tbl8s = 256;
925 lpm_config.flags = 0;
926
927 lpm = rte_lpm_create(buf, socket, &lpm_config);
928 if (lpm == NULL) {
929 RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
930 return -1;
931 }
932 socket_lpm[socket] = lpm;
933 }
934
935 if (socket_lpm6[socket] == NULL) {
936 RTE_LOG(INFO, IP_RSMBL, "Creating LPM6 table on socket %i\n", socket);
937 snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
938
939 lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
940 if (lpm6 == NULL) {
941 RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
942 return -1;
943 }
944 socket_lpm6[socket] = lpm6;
945 }
946 }
947
948 return 0;
949 }
950
951 static void
queue_dump_stat(void)952 queue_dump_stat(void)
953 {
954 uint32_t i, lcore;
955 const struct lcore_queue_conf *qconf;
956
957 for (lcore = 0; lcore < RTE_MAX_LCORE; lcore++) {
958 if (rte_lcore_is_enabled(lcore) == 0)
959 continue;
960
961 qconf = &lcore_queue_conf[lcore];
962 for (i = 0; i < qconf->n_rx_queue; i++) {
963
964 fprintf(stdout, " -- lcoreid=%u portid=%u "
965 "frag tbl stat:\n",
966 lcore, qconf->rx_queue_list[i].portid);
967 rte_ip_frag_table_statistics_dump(stdout,
968 qconf->rx_queue_list[i].frag_tbl);
969 fprintf(stdout, "TX bursts:\t%" PRIu64 "\n"
970 "TX packets _queued:\t%" PRIu64 "\n"
971 "TX packets dropped:\t%" PRIu64 "\n"
972 "TX packets send:\t%" PRIu64 "\n",
973 qconf->tx_stat.call,
974 qconf->tx_stat.queue,
975 qconf->tx_stat.drop,
976 qconf->tx_stat.send);
977 }
978 }
979 }
980
981 static void
signal_handler(int signum)982 signal_handler(int signum)
983 {
984 queue_dump_stat();
985 if (signum != SIGUSR1)
986 rte_exit(0, "received signal: %d, exiting\n", signum);
987 }
988
989 int
main(int argc,char ** argv)990 main(int argc, char **argv)
991 {
992 struct lcore_queue_conf *qconf;
993 struct rte_eth_dev_info dev_info;
994 struct rte_eth_txconf *txconf;
995 struct rx_queue *rxq;
996 int ret, socket;
997 unsigned nb_ports;
998 uint16_t queueid;
999 unsigned lcore_id = 0, rx_lcore_id = 0;
1000 uint32_t n_tx_queue, nb_lcores;
1001 uint16_t portid;
1002
1003 /* init EAL */
1004 ret = rte_eal_init(argc, argv);
1005 if (ret < 0)
1006 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1007 argc -= ret;
1008 argv += ret;
1009
1010 /* parse application arguments (after the EAL ones) */
1011 ret = parse_args(argc, argv);
1012 if (ret < 0)
1013 rte_exit(EXIT_FAILURE, "Invalid IP reassembly parameters\n");
1014
1015 nb_ports = rte_eth_dev_count_avail();
1016 if (nb_ports == 0)
1017 rte_exit(EXIT_FAILURE, "No ports found!\n");
1018
1019 nb_lcores = rte_lcore_count();
1020
1021 /* initialize structures (mempools, lpm etc.) */
1022 if (init_mem() < 0)
1023 rte_panic("Cannot initialize memory structures!\n");
1024
1025 /* check if portmask has non-existent ports */
1026 if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
1027 rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
1028
1029 /* initialize all ports */
1030 RTE_ETH_FOREACH_DEV(portid) {
1031 struct rte_eth_rxconf rxq_conf;
1032 struct rte_eth_conf local_port_conf = port_conf;
1033
1034 /* skip ports that are not enabled */
1035 if ((enabled_port_mask & (1 << portid)) == 0) {
1036 printf("\nSkipping disabled port %d\n", portid);
1037 continue;
1038 }
1039
1040 qconf = &lcore_queue_conf[rx_lcore_id];
1041
1042 /* limit the frame size to the maximum supported by NIC */
1043 ret = rte_eth_dev_info_get(portid, &dev_info);
1044 if (ret != 0)
1045 rte_exit(EXIT_FAILURE,
1046 "Error during getting device (port %u) info: %s\n",
1047 portid, strerror(-ret));
1048
1049 local_port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
1050 dev_info.max_rx_pktlen,
1051 local_port_conf.rxmode.max_rx_pkt_len);
1052
1053 /* get the lcore_id for this port */
1054 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
1055 qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
1056
1057 rx_lcore_id++;
1058 if (rx_lcore_id >= RTE_MAX_LCORE)
1059 rte_exit(EXIT_FAILURE, "Not enough cores\n");
1060
1061 qconf = &lcore_queue_conf[rx_lcore_id];
1062 }
1063
1064 socket = rte_lcore_to_socket_id(portid);
1065 if (socket == SOCKET_ID_ANY)
1066 socket = 0;
1067
1068 queueid = qconf->n_rx_queue;
1069 rxq = &qconf->rx_queue_list[queueid];
1070 rxq->portid = portid;
1071 rxq->lpm = socket_lpm[socket];
1072 rxq->lpm6 = socket_lpm6[socket];
1073
1074 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1075 &nb_txd);
1076 if (ret < 0)
1077 rte_exit(EXIT_FAILURE,
1078 "Cannot adjust number of descriptors: err=%d, port=%d\n",
1079 ret, portid);
1080
1081 if (setup_queue_tbl(rxq, rx_lcore_id, queueid) < 0)
1082 rte_exit(EXIT_FAILURE, "Failed to set up queue table\n");
1083 qconf->n_rx_queue++;
1084
1085 /* init port */
1086 printf("Initializing port %d ... ", portid );
1087 fflush(stdout);
1088
1089 n_tx_queue = nb_lcores;
1090 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1091 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1092 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
1093 local_port_conf.txmode.offloads |=
1094 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1095
1096 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
1097 dev_info.flow_type_rss_offloads;
1098 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
1099 port_conf.rx_adv_conf.rss_conf.rss_hf) {
1100 printf("Port %u modified RSS hash function based on hardware support,"
1101 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
1102 portid,
1103 port_conf.rx_adv_conf.rss_conf.rss_hf,
1104 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
1105 }
1106
1107 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
1108 &local_port_conf);
1109 if (ret < 0) {
1110 printf("\n");
1111 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1112 "err=%d, port=%d\n",
1113 ret, portid);
1114 }
1115
1116 /* init one RX queue */
1117 rxq_conf = dev_info.default_rxconf;
1118 rxq_conf.offloads = local_port_conf.rxmode.offloads;
1119 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
1120 socket, &rxq_conf,
1121 rxq->pool);
1122 if (ret < 0) {
1123 printf("\n");
1124 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
1125 "err=%d, port=%d\n",
1126 ret, portid);
1127 }
1128
1129 ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1130 if (ret < 0) {
1131 printf("\n");
1132 rte_exit(EXIT_FAILURE,
1133 "rte_eth_macaddr_get: err=%d, port=%d\n",
1134 ret, portid);
1135 }
1136
1137 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1138 printf("\n");
1139
1140 /* init one TX queue per couple (lcore,port) */
1141 queueid = 0;
1142 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1143 if (rte_lcore_is_enabled(lcore_id) == 0)
1144 continue;
1145
1146 socket = (int) rte_lcore_to_socket_id(lcore_id);
1147
1148 printf("txq=%u,%d,%d ", lcore_id, queueid, socket);
1149 fflush(stdout);
1150
1151 txconf = &dev_info.default_txconf;
1152 txconf->offloads = local_port_conf.txmode.offloads;
1153
1154 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1155 socket, txconf);
1156 if (ret < 0)
1157 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1158 "port=%d\n", ret, portid);
1159
1160 qconf = &lcore_queue_conf[lcore_id];
1161 qconf->tx_queue_id[portid] = queueid;
1162 setup_port_tbl(qconf, lcore_id, socket, portid);
1163 queueid++;
1164 }
1165 printf("\n");
1166 }
1167
1168 printf("\n");
1169
1170 /* start ports */
1171 RTE_ETH_FOREACH_DEV(portid) {
1172 if ((enabled_port_mask & (1 << portid)) == 0) {
1173 continue;
1174 }
1175 /* Start device */
1176 ret = rte_eth_dev_start(portid);
1177 if (ret < 0)
1178 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1179 ret, portid);
1180
1181 ret = rte_eth_promiscuous_enable(portid);
1182 if (ret != 0)
1183 rte_exit(EXIT_FAILURE,
1184 "rte_eth_promiscuous_enable: err=%s, port=%d\n",
1185 rte_strerror(-ret), portid);
1186 }
1187
1188 if (init_routing_table() < 0)
1189 rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1190
1191 check_all_ports_link_status(enabled_port_mask);
1192
1193 signal(SIGUSR1, signal_handler);
1194 signal(SIGTERM, signal_handler);
1195 signal(SIGINT, signal_handler);
1196
1197 /* launch per-lcore init on every lcore */
1198 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MAIN);
1199 RTE_LCORE_FOREACH_WORKER(lcore_id) {
1200 if (rte_eal_wait_lcore(lcore_id) < 0)
1201 return -1;
1202 }
1203
1204 return 0;
1205 }
1206