]> git.proxmox.com Git - ceph.git/blob - ceph/src/seastar/dpdk/examples/distributor/main.c
update sources to ceph Nautilus 14.2.1
[ceph.git] / ceph / src / seastar / dpdk / examples / distributor / main.c
1 /*-
2 * BSD LICENSE
3 *
4 * Copyright(c) 2010-2017 Intel Corporation. All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 *
10 * * Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * * Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in
14 * the documentation and/or other materials provided with the
15 * distribution.
16 * * Neither the name of Intel Corporation nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
23 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
24 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
25 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
26 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
27 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
28 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
29 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
30 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31 */
32
33 #include <stdint.h>
34 #include <inttypes.h>
35 #include <unistd.h>
36 #include <signal.h>
37 #include <getopt.h>
38
39 #include <rte_eal.h>
40 #include <rte_ethdev.h>
41 #include <rte_cycles.h>
42 #include <rte_malloc.h>
43 #include <rte_debug.h>
44 #include <rte_prefetch.h>
45 #include <rte_distributor.h>
46
47 #define RX_RING_SIZE 512
48 #define TX_RING_SIZE 512
49 #define NUM_MBUFS ((64*1024)-1)
50 #define MBUF_CACHE_SIZE 128
51 #define BURST_SIZE 64
52 #define SCHED_RX_RING_SZ 8192
53 #define SCHED_TX_RING_SZ 65536
54 #define BURST_SIZE_TX 32
55
56 #define RTE_LOGTYPE_DISTRAPP RTE_LOGTYPE_USER1
57
58 #define ANSI_COLOR_RED "\x1b[31m"
59 #define ANSI_COLOR_RESET "\x1b[0m"
60
61 /* mask of enabled ports */
62 static uint32_t enabled_port_mask;
63 volatile uint8_t quit_signal;
64 volatile uint8_t quit_signal_rx;
65 volatile uint8_t quit_signal_dist;
66 volatile uint8_t quit_signal_work;
67
68 static volatile struct app_stats {
69 struct {
70 uint64_t rx_pkts;
71 uint64_t returned_pkts;
72 uint64_t enqueued_pkts;
73 uint64_t enqdrop_pkts;
74 } rx __rte_cache_aligned;
75 int pad1 __rte_cache_aligned;
76
77 struct {
78 uint64_t in_pkts;
79 uint64_t ret_pkts;
80 uint64_t sent_pkts;
81 uint64_t enqdrop_pkts;
82 } dist __rte_cache_aligned;
83 int pad2 __rte_cache_aligned;
84
85 struct {
86 uint64_t dequeue_pkts;
87 uint64_t tx_pkts;
88 uint64_t enqdrop_pkts;
89 } tx __rte_cache_aligned;
90 int pad3 __rte_cache_aligned;
91
92 uint64_t worker_pkts[64] __rte_cache_aligned;
93
94 int pad4 __rte_cache_aligned;
95
96 uint64_t worker_bursts[64][8] __rte_cache_aligned;
97
98 int pad5 __rte_cache_aligned;
99
100 uint64_t port_rx_pkts[64] __rte_cache_aligned;
101 uint64_t port_tx_pkts[64] __rte_cache_aligned;
102 } app_stats;
103
104 struct app_stats prev_app_stats;
105
106 static const struct rte_eth_conf port_conf_default = {
107 .rxmode = {
108 .mq_mode = ETH_MQ_RX_RSS,
109 .max_rx_pkt_len = ETHER_MAX_LEN,
110 },
111 .txmode = {
112 .mq_mode = ETH_MQ_TX_NONE,
113 },
114 .rx_adv_conf = {
115 .rss_conf = {
116 .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
117 ETH_RSS_TCP | ETH_RSS_SCTP,
118 }
119 },
120 };
121
122 struct output_buffer {
123 unsigned count;
124 struct rte_mbuf *mbufs[BURST_SIZE];
125 };
126
127 static void print_stats(void);
128
129 /*
130 * Initialises a given port using global settings and with the rx buffers
131 * coming from the mbuf_pool passed as parameter
132 */
133 static inline int
134 port_init(uint8_t port, struct rte_mempool *mbuf_pool)
135 {
136 struct rte_eth_conf port_conf = port_conf_default;
137 const uint16_t rxRings = 1, txRings = rte_lcore_count() - 1;
138 int retval;
139 uint16_t q;
140
141 if (port >= rte_eth_dev_count())
142 return -1;
143
144 retval = rte_eth_dev_configure(port, rxRings, txRings, &port_conf);
145 if (retval != 0)
146 return retval;
147
148 for (q = 0; q < rxRings; q++) {
149 retval = rte_eth_rx_queue_setup(port, q, RX_RING_SIZE,
150 rte_eth_dev_socket_id(port),
151 NULL, mbuf_pool);
152 if (retval < 0)
153 return retval;
154 }
155
156 for (q = 0; q < txRings; q++) {
157 retval = rte_eth_tx_queue_setup(port, q, TX_RING_SIZE,
158 rte_eth_dev_socket_id(port),
159 NULL);
160 if (retval < 0)
161 return retval;
162 }
163
164 retval = rte_eth_dev_start(port);
165 if (retval < 0)
166 return retval;
167
168 struct rte_eth_link link;
169 rte_eth_link_get_nowait(port, &link);
170 while (!link.link_status) {
171 printf("Waiting for Link up on port %"PRIu8"\n", port);
172 sleep(1);
173 rte_eth_link_get_nowait(port, &link);
174 }
175
176 if (!link.link_status) {
177 printf("Link down on port %"PRIu8"\n", port);
178 return 0;
179 }
180
181 struct ether_addr addr;
182 rte_eth_macaddr_get(port, &addr);
183 printf("Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8
184 " %02"PRIx8" %02"PRIx8" %02"PRIx8"\n",
185 (unsigned)port,
186 addr.addr_bytes[0], addr.addr_bytes[1],
187 addr.addr_bytes[2], addr.addr_bytes[3],
188 addr.addr_bytes[4], addr.addr_bytes[5]);
189
190 rte_eth_promiscuous_enable(port);
191
192 return 0;
193 }
194
195 struct lcore_params {
196 unsigned worker_id;
197 struct rte_distributor *d;
198 struct rte_ring *rx_dist_ring;
199 struct rte_ring *dist_tx_ring;
200 struct rte_mempool *mem_pool;
201 };
202
203 static int
204 lcore_rx(struct lcore_params *p)
205 {
206 const uint8_t nb_ports = rte_eth_dev_count();
207 const int socket_id = rte_socket_id();
208 uint8_t port;
209 struct rte_mbuf *bufs[BURST_SIZE*2];
210
211 for (port = 0; port < nb_ports; port++) {
212 /* skip ports that are not enabled */
213 if ((enabled_port_mask & (1 << port)) == 0)
214 continue;
215
216 if (rte_eth_dev_socket_id(port) > 0 &&
217 rte_eth_dev_socket_id(port) != socket_id)
218 printf("WARNING, port %u is on remote NUMA node to "
219 "RX thread.\n\tPerformance will not "
220 "be optimal.\n", port);
221 }
222
223 printf("\nCore %u doing packet RX.\n", rte_lcore_id());
224 port = 0;
225 while (!quit_signal_rx) {
226
227 /* skip ports that are not enabled */
228 if ((enabled_port_mask & (1 << port)) == 0) {
229 if (++port == nb_ports)
230 port = 0;
231 continue;
232 }
233 const uint16_t nb_rx = rte_eth_rx_burst(port, 0, bufs,
234 BURST_SIZE);
235 if (unlikely(nb_rx == 0)) {
236 if (++port == nb_ports)
237 port = 0;
238 continue;
239 }
240 app_stats.rx.rx_pkts += nb_rx;
241
242 /*
243 * You can run the distributor on the rx core with this code. Returned
244 * packets are then send straight to the tx core.
245 */
246 #if 0
247 rte_distributor_process(d, bufs, nb_rx);
248 const uint16_t nb_ret = rte_distributor_returned_pktsd,
249 bufs, BURST_SIZE*2);
250
251 app_stats.rx.returned_pkts += nb_ret;
252 if (unlikely(nb_ret == 0)) {
253 if (++port == nb_ports)
254 port = 0;
255 continue;
256 }
257
258 struct rte_ring *tx_ring = p->dist_tx_ring;
259 uint16_t sent = rte_ring_enqueue_burst(tx_ring,
260 (void *)bufs, nb_ret, NULL);
261 #else
262 uint16_t nb_ret = nb_rx;
263 /*
264 * Swap the following two lines if you want the rx traffic
265 * to go directly to tx, no distribution.
266 */
267 struct rte_ring *out_ring = p->rx_dist_ring;
268 /* struct rte_ring *out_ring = p->dist_tx_ring; */
269
270 uint16_t sent = rte_ring_enqueue_burst(out_ring,
271 (void *)bufs, nb_ret, NULL);
272 #endif
273
274 app_stats.rx.enqueued_pkts += sent;
275 if (unlikely(sent < nb_ret)) {
276 app_stats.rx.enqdrop_pkts += nb_ret - sent;
277 RTE_LOG_DP(DEBUG, DISTRAPP,
278 "%s:Packet loss due to full ring\n", __func__);
279 while (sent < nb_ret)
280 rte_pktmbuf_free(bufs[sent++]);
281 }
282 if (++port == nb_ports)
283 port = 0;
284 }
285 /* set worker & tx threads quit flag */
286 printf("\nCore %u exiting rx task.\n", rte_lcore_id());
287 quit_signal = 1;
288 return 0;
289 }
290
291 static inline void
292 flush_one_port(struct output_buffer *outbuf, uint8_t outp)
293 {
294 unsigned int nb_tx = rte_eth_tx_burst(outp, 0,
295 outbuf->mbufs, outbuf->count);
296 app_stats.tx.tx_pkts += outbuf->count;
297
298 if (unlikely(nb_tx < outbuf->count)) {
299 app_stats.tx.enqdrop_pkts += outbuf->count - nb_tx;
300 do {
301 rte_pktmbuf_free(outbuf->mbufs[nb_tx]);
302 } while (++nb_tx < outbuf->count);
303 }
304 outbuf->count = 0;
305 }
306
307 static inline void
308 flush_all_ports(struct output_buffer *tx_buffers, uint8_t nb_ports)
309 {
310 uint8_t outp;
311
312 for (outp = 0; outp < nb_ports; outp++) {
313 /* skip ports that are not enabled */
314 if ((enabled_port_mask & (1 << outp)) == 0)
315 continue;
316
317 if (tx_buffers[outp].count == 0)
318 continue;
319
320 flush_one_port(&tx_buffers[outp], outp);
321 }
322 }
323
324
325
326 static int
327 lcore_distributor(struct lcore_params *p)
328 {
329 struct rte_ring *in_r = p->rx_dist_ring;
330 struct rte_ring *out_r = p->dist_tx_ring;
331 struct rte_mbuf *bufs[BURST_SIZE * 4];
332 struct rte_distributor *d = p->d;
333
334 printf("\nCore %u acting as distributor core.\n", rte_lcore_id());
335 while (!quit_signal_dist) {
336 const uint16_t nb_rx = rte_ring_dequeue_burst(in_r,
337 (void *)bufs, BURST_SIZE*1, NULL);
338 if (nb_rx) {
339 app_stats.dist.in_pkts += nb_rx;
340
341 /* Distribute the packets */
342 rte_distributor_process(d, bufs, nb_rx);
343 /* Handle Returns */
344 const uint16_t nb_ret =
345 rte_distributor_returned_pkts(d,
346 bufs, BURST_SIZE*2);
347
348 if (unlikely(nb_ret == 0))
349 continue;
350 app_stats.dist.ret_pkts += nb_ret;
351
352 uint16_t sent = rte_ring_enqueue_burst(out_r,
353 (void *)bufs, nb_ret, NULL);
354 app_stats.dist.sent_pkts += sent;
355 if (unlikely(sent < nb_ret)) {
356 app_stats.dist.enqdrop_pkts += nb_ret - sent;
357 RTE_LOG(DEBUG, DISTRAPP,
358 "%s:Packet loss due to full out ring\n",
359 __func__);
360 while (sent < nb_ret)
361 rte_pktmbuf_free(bufs[sent++]);
362 }
363 }
364 }
365 printf("\nCore %u exiting distributor task.\n", rte_lcore_id());
366 quit_signal_work = 1;
367
368 rte_distributor_flush(d);
369 /* Unblock any returns so workers can exit */
370 rte_distributor_clear_returns(d);
371 quit_signal_rx = 1;
372 return 0;
373 }
374
375
376 static int
377 lcore_tx(struct rte_ring *in_r)
378 {
379 static struct output_buffer tx_buffers[RTE_MAX_ETHPORTS];
380 const uint8_t nb_ports = rte_eth_dev_count();
381 const int socket_id = rte_socket_id();
382 uint8_t port;
383
384 for (port = 0; port < nb_ports; port++) {
385 /* skip ports that are not enabled */
386 if ((enabled_port_mask & (1 << port)) == 0)
387 continue;
388
389 if (rte_eth_dev_socket_id(port) > 0 &&
390 rte_eth_dev_socket_id(port) != socket_id)
391 printf("WARNING, port %u is on remote NUMA node to "
392 "TX thread.\n\tPerformance will not "
393 "be optimal.\n", port);
394 }
395
396 printf("\nCore %u doing packet TX.\n", rte_lcore_id());
397 while (!quit_signal) {
398
399 for (port = 0; port < nb_ports; port++) {
400 /* skip ports that are not enabled */
401 if ((enabled_port_mask & (1 << port)) == 0)
402 continue;
403
404 struct rte_mbuf *bufs[BURST_SIZE_TX];
405 const uint16_t nb_rx = rte_ring_dequeue_burst(in_r,
406 (void *)bufs, BURST_SIZE_TX, NULL);
407 app_stats.tx.dequeue_pkts += nb_rx;
408
409 /* if we get no traffic, flush anything we have */
410 if (unlikely(nb_rx == 0)) {
411 flush_all_ports(tx_buffers, nb_ports);
412 continue;
413 }
414
415 /* for traffic we receive, queue it up for transmit */
416 uint16_t i;
417 rte_prefetch_non_temporal((void *)bufs[0]);
418 rte_prefetch_non_temporal((void *)bufs[1]);
419 rte_prefetch_non_temporal((void *)bufs[2]);
420 for (i = 0; i < nb_rx; i++) {
421 struct output_buffer *outbuf;
422 uint8_t outp;
423 rte_prefetch_non_temporal((void *)bufs[i + 3]);
424 /*
425 * workers should update in_port to hold the
426 * output port value
427 */
428 outp = bufs[i]->port;
429 /* skip ports that are not enabled */
430 if ((enabled_port_mask & (1 << outp)) == 0)
431 continue;
432
433 outbuf = &tx_buffers[outp];
434 outbuf->mbufs[outbuf->count++] = bufs[i];
435 if (outbuf->count == BURST_SIZE_TX)
436 flush_one_port(outbuf, outp);
437 }
438 }
439 }
440 printf("\nCore %u exiting tx task.\n", rte_lcore_id());
441 return 0;
442 }
443
444 static void
445 int_handler(int sig_num)
446 {
447 printf("Exiting on signal %d\n", sig_num);
448 /* set quit flag for rx thread to exit */
449 quit_signal_dist = 1;
450 }
451
452 static void
453 print_stats(void)
454 {
455 struct rte_eth_stats eth_stats;
456 unsigned int i, j;
457 const unsigned int num_workers = rte_lcore_count() - 4;
458
459 for (i = 0; i < rte_eth_dev_count(); i++) {
460 rte_eth_stats_get(i, &eth_stats);
461 app_stats.port_rx_pkts[i] = eth_stats.ipackets;
462 app_stats.port_tx_pkts[i] = eth_stats.opackets;
463 }
464
465 printf("\n\nRX Thread:\n");
466 for (i = 0; i < rte_eth_dev_count(); i++) {
467 printf("Port %u Pktsin : %5.2f\n", i,
468 (app_stats.port_rx_pkts[i] -
469 prev_app_stats.port_rx_pkts[i])/1000000.0);
470 prev_app_stats.port_rx_pkts[i] = app_stats.port_rx_pkts[i];
471 }
472 printf(" - Received: %5.2f\n",
473 (app_stats.rx.rx_pkts -
474 prev_app_stats.rx.rx_pkts)/1000000.0);
475 printf(" - Returned: %5.2f\n",
476 (app_stats.rx.returned_pkts -
477 prev_app_stats.rx.returned_pkts)/1000000.0);
478 printf(" - Enqueued: %5.2f\n",
479 (app_stats.rx.enqueued_pkts -
480 prev_app_stats.rx.enqueued_pkts)/1000000.0);
481 printf(" - Dropped: %s%5.2f%s\n", ANSI_COLOR_RED,
482 (app_stats.rx.enqdrop_pkts -
483 prev_app_stats.rx.enqdrop_pkts)/1000000.0,
484 ANSI_COLOR_RESET);
485
486 printf("Distributor thread:\n");
487 printf(" - In: %5.2f\n",
488 (app_stats.dist.in_pkts -
489 prev_app_stats.dist.in_pkts)/1000000.0);
490 printf(" - Returned: %5.2f\n",
491 (app_stats.dist.ret_pkts -
492 prev_app_stats.dist.ret_pkts)/1000000.0);
493 printf(" - Sent: %5.2f\n",
494 (app_stats.dist.sent_pkts -
495 prev_app_stats.dist.sent_pkts)/1000000.0);
496 printf(" - Dropped %s%5.2f%s\n", ANSI_COLOR_RED,
497 (app_stats.dist.enqdrop_pkts -
498 prev_app_stats.dist.enqdrop_pkts)/1000000.0,
499 ANSI_COLOR_RESET);
500
501 printf("TX thread:\n");
502 printf(" - Dequeued: %5.2f\n",
503 (app_stats.tx.dequeue_pkts -
504 prev_app_stats.tx.dequeue_pkts)/1000000.0);
505 for (i = 0; i < rte_eth_dev_count(); i++) {
506 printf("Port %u Pktsout: %5.2f\n",
507 i, (app_stats.port_tx_pkts[i] -
508 prev_app_stats.port_tx_pkts[i])/1000000.0);
509 prev_app_stats.port_tx_pkts[i] = app_stats.port_tx_pkts[i];
510 }
511 printf(" - Transmitted: %5.2f\n",
512 (app_stats.tx.tx_pkts -
513 prev_app_stats.tx.tx_pkts)/1000000.0);
514 printf(" - Dropped: %s%5.2f%s\n", ANSI_COLOR_RED,
515 (app_stats.tx.enqdrop_pkts -
516 prev_app_stats.tx.enqdrop_pkts)/1000000.0,
517 ANSI_COLOR_RESET);
518
519 prev_app_stats.rx.rx_pkts = app_stats.rx.rx_pkts;
520 prev_app_stats.rx.returned_pkts = app_stats.rx.returned_pkts;
521 prev_app_stats.rx.enqueued_pkts = app_stats.rx.enqueued_pkts;
522 prev_app_stats.rx.enqdrop_pkts = app_stats.rx.enqdrop_pkts;
523 prev_app_stats.dist.in_pkts = app_stats.dist.in_pkts;
524 prev_app_stats.dist.ret_pkts = app_stats.dist.ret_pkts;
525 prev_app_stats.dist.sent_pkts = app_stats.dist.sent_pkts;
526 prev_app_stats.dist.enqdrop_pkts = app_stats.dist.enqdrop_pkts;
527 prev_app_stats.tx.dequeue_pkts = app_stats.tx.dequeue_pkts;
528 prev_app_stats.tx.tx_pkts = app_stats.tx.tx_pkts;
529 prev_app_stats.tx.enqdrop_pkts = app_stats.tx.enqdrop_pkts;
530
531 for (i = 0; i < num_workers; i++) {
532 printf("Worker %02u Pkts: %5.2f. Bursts(1-8): ", i,
533 (app_stats.worker_pkts[i] -
534 prev_app_stats.worker_pkts[i])/1000000.0);
535 for (j = 0; j < 8; j++) {
536 printf("%"PRIu64" ", app_stats.worker_bursts[i][j]);
537 app_stats.worker_bursts[i][j] = 0;
538 }
539 printf("\n");
540 prev_app_stats.worker_pkts[i] = app_stats.worker_pkts[i];
541 }
542 }
543
544 static int
545 lcore_worker(struct lcore_params *p)
546 {
547 struct rte_distributor *d = p->d;
548 const unsigned id = p->worker_id;
549 unsigned int num = 0;
550 unsigned int i;
551
552 /*
553 * for single port, xor_val will be zero so we won't modify the output
554 * port, otherwise we send traffic from 0 to 1, 2 to 3, and vice versa
555 */
556 const unsigned xor_val = (rte_eth_dev_count() > 1);
557 struct rte_mbuf *buf[8] __rte_cache_aligned;
558
559 for (i = 0; i < 8; i++)
560 buf[i] = NULL;
561
562 app_stats.worker_pkts[p->worker_id] = 1;
563
564 printf("\nCore %u acting as worker core.\n", rte_lcore_id());
565 while (!quit_signal_work) {
566 num = rte_distributor_get_pkt(d, id, buf, buf, num);
567 /* Do a little bit of work for each packet */
568 for (i = 0; i < num; i++) {
569 uint64_t t = rte_rdtsc()+100;
570
571 while (rte_rdtsc() < t)
572 rte_pause();
573 buf[i]->port ^= xor_val;
574 }
575
576 app_stats.worker_pkts[p->worker_id] += num;
577 if (num > 0)
578 app_stats.worker_bursts[p->worker_id][num-1]++;
579 }
580 return 0;
581 }
582
583 /* display usage */
584 static void
585 print_usage(const char *prgname)
586 {
587 printf("%s [EAL options] -- -p PORTMASK\n"
588 " -p PORTMASK: hexadecimal bitmask of ports to configure\n",
589 prgname);
590 }
591
592 static int
593 parse_portmask(const char *portmask)
594 {
595 char *end = NULL;
596 unsigned long pm;
597
598 /* parse hexadecimal string */
599 pm = strtoul(portmask, &end, 16);
600 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
601 return -1;
602
603 if (pm == 0)
604 return -1;
605
606 return pm;
607 }
608
609 /* Parse the argument given in the command line of the application */
610 static int
611 parse_args(int argc, char **argv)
612 {
613 int opt;
614 char **argvopt;
615 int option_index;
616 char *prgname = argv[0];
617 static struct option lgopts[] = {
618 {NULL, 0, 0, 0}
619 };
620
621 argvopt = argv;
622
623 while ((opt = getopt_long(argc, argvopt, "p:",
624 lgopts, &option_index)) != EOF) {
625
626 switch (opt) {
627 /* portmask */
628 case 'p':
629 enabled_port_mask = parse_portmask(optarg);
630 if (enabled_port_mask == 0) {
631 printf("invalid portmask\n");
632 print_usage(prgname);
633 return -1;
634 }
635 break;
636
637 default:
638 print_usage(prgname);
639 return -1;
640 }
641 }
642
643 if (optind <= 1) {
644 print_usage(prgname);
645 return -1;
646 }
647
648 argv[optind-1] = prgname;
649
650 optind = 1; /* reset getopt lib */
651 return 0;
652 }
653
654 /* Main function, does initialization and calls the per-lcore functions */
655 int
656 main(int argc, char *argv[])
657 {
658 struct rte_mempool *mbuf_pool;
659 struct rte_distributor *d;
660 struct rte_ring *dist_tx_ring;
661 struct rte_ring *rx_dist_ring;
662 unsigned lcore_id, worker_id = 0;
663 unsigned nb_ports;
664 uint8_t portid;
665 uint8_t nb_ports_available;
666 uint64_t t, freq;
667
668 /* catch ctrl-c so we can print on exit */
669 signal(SIGINT, int_handler);
670
671 /* init EAL */
672 int ret = rte_eal_init(argc, argv);
673 if (ret < 0)
674 rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
675 argc -= ret;
676 argv += ret;
677
678 /* parse application arguments (after the EAL ones) */
679 ret = parse_args(argc, argv);
680 if (ret < 0)
681 rte_exit(EXIT_FAILURE, "Invalid distributor parameters\n");
682
683 if (rte_lcore_count() < 5)
684 rte_exit(EXIT_FAILURE, "Error, This application needs at "
685 "least 5 logical cores to run:\n"
686 "1 lcore for stats (can be core 0)\n"
687 "1 lcore for packet RX\n"
688 "1 lcore for distribution\n"
689 "1 lcore for packet TX\n"
690 "and at least 1 lcore for worker threads\n");
691
692 nb_ports = rte_eth_dev_count();
693 if (nb_ports == 0)
694 rte_exit(EXIT_FAILURE, "Error: no ethernet ports detected\n");
695 if (nb_ports != 1 && (nb_ports & 1))
696 rte_exit(EXIT_FAILURE, "Error: number of ports must be even, except "
697 "when using a single port\n");
698
699 mbuf_pool = rte_pktmbuf_pool_create("MBUF_POOL",
700 NUM_MBUFS * nb_ports, MBUF_CACHE_SIZE, 0,
701 RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
702 if (mbuf_pool == NULL)
703 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
704 nb_ports_available = nb_ports;
705
706 /* initialize all ports */
707 for (portid = 0; portid < nb_ports; portid++) {
708 /* skip ports that are not enabled */
709 if ((enabled_port_mask & (1 << portid)) == 0) {
710 printf("\nSkipping disabled port %d\n", portid);
711 nb_ports_available--;
712 continue;
713 }
714 /* init port */
715 printf("Initializing port %u... done\n", (unsigned) portid);
716
717 if (port_init(portid, mbuf_pool) != 0)
718 rte_exit(EXIT_FAILURE, "Cannot initialize port %"PRIu8"\n",
719 portid);
720 }
721
722 if (!nb_ports_available) {
723 rte_exit(EXIT_FAILURE,
724 "All available ports are disabled. Please set portmask.\n");
725 }
726
727 d = rte_distributor_create("PKT_DIST", rte_socket_id(),
728 rte_lcore_count() - 4,
729 RTE_DIST_ALG_BURST);
730 if (d == NULL)
731 rte_exit(EXIT_FAILURE, "Cannot create distributor\n");
732
733 /*
734 * scheduler ring is read by the transmitter core, and written to
735 * by scheduler core
736 */
737 dist_tx_ring = rte_ring_create("Output_ring", SCHED_TX_RING_SZ,
738 rte_socket_id(), RING_F_SC_DEQ | RING_F_SP_ENQ);
739 if (dist_tx_ring == NULL)
740 rte_exit(EXIT_FAILURE, "Cannot create output ring\n");
741
742 rx_dist_ring = rte_ring_create("Input_ring", SCHED_RX_RING_SZ,
743 rte_socket_id(), RING_F_SC_DEQ | RING_F_SP_ENQ);
744 if (rx_dist_ring == NULL)
745 rte_exit(EXIT_FAILURE, "Cannot create output ring\n");
746
747 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
748 if (worker_id == rte_lcore_count() - 3) {
749 printf("Starting distributor on lcore_id %d\n",
750 lcore_id);
751 /* distributor core */
752 struct lcore_params *p =
753 rte_malloc(NULL, sizeof(*p), 0);
754 if (!p)
755 rte_panic("malloc failure\n");
756 *p = (struct lcore_params){worker_id, d,
757 rx_dist_ring, dist_tx_ring, mbuf_pool};
758 rte_eal_remote_launch(
759 (lcore_function_t *)lcore_distributor,
760 p, lcore_id);
761 } else if (worker_id == rte_lcore_count() - 4) {
762 printf("Starting tx on worker_id %d, lcore_id %d\n",
763 worker_id, lcore_id);
764 /* tx core */
765 rte_eal_remote_launch((lcore_function_t *)lcore_tx,
766 dist_tx_ring, lcore_id);
767 } else if (worker_id == rte_lcore_count() - 2) {
768 printf("Starting rx on worker_id %d, lcore_id %d\n",
769 worker_id, lcore_id);
770 /* rx core */
771 struct lcore_params *p =
772 rte_malloc(NULL, sizeof(*p), 0);
773 if (!p)
774 rte_panic("malloc failure\n");
775 *p = (struct lcore_params){worker_id, d, rx_dist_ring,
776 dist_tx_ring, mbuf_pool};
777 rte_eal_remote_launch((lcore_function_t *)lcore_rx,
778 p, lcore_id);
779 } else {
780 printf("Starting worker on worker_id %d, lcore_id %d\n",
781 worker_id, lcore_id);
782 struct lcore_params *p =
783 rte_malloc(NULL, sizeof(*p), 0);
784 if (!p)
785 rte_panic("malloc failure\n");
786 *p = (struct lcore_params){worker_id, d, rx_dist_ring,
787 dist_tx_ring, mbuf_pool};
788
789 rte_eal_remote_launch((lcore_function_t *)lcore_worker,
790 p, lcore_id);
791 }
792 worker_id++;
793 }
794
795 freq = rte_get_timer_hz();
796 t = rte_rdtsc() + freq;
797 while (!quit_signal_dist) {
798 if (t < rte_rdtsc()) {
799 print_stats();
800 t = rte_rdtsc() + freq;
801 }
802 usleep(1000);
803 }
804
805 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
806 if (rte_eal_wait_lcore(lcore_id) < 0)
807 return -1;
808 }
809
810 print_stats();
811 return 0;
812 }