xref: /dpdk/examples/distributor/main.c (revision c9902a15bd005b6d4fe072cf7b60fe4ee679155f)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2017 Intel Corporation
3  */
4 
5 #include <stdint.h>
6 #include <inttypes.h>
7 #include <unistd.h>
8 #include <signal.h>
9 #include <getopt.h>
10 
11 #include <rte_eal.h>
12 #include <rte_ethdev.h>
13 #include <rte_cycles.h>
14 #include <rte_malloc.h>
15 #include <rte_debug.h>
16 #include <rte_prefetch.h>
17 #include <rte_distributor.h>
18 #include <rte_pause.h>
19 #include <rte_power.h>
20 
21 #define RX_RING_SIZE 1024
22 #define TX_RING_SIZE 1024
23 #define NUM_MBUFS ((64*1024)-1)
24 #define MBUF_CACHE_SIZE 128
25 #define BURST_SIZE 64
26 #define SCHED_RX_RING_SZ 8192
27 #define SCHED_TX_RING_SZ 65536
28 #define BURST_SIZE_TX 32
29 
30 #define RTE_LOGTYPE_DISTRAPP RTE_LOGTYPE_USER1
31 
32 #define ANSI_COLOR_RED     "\x1b[31m"
33 #define ANSI_COLOR_RESET   "\x1b[0m"
34 
35 /* mask of enabled ports */
36 static uint32_t enabled_port_mask;
37 volatile uint8_t quit_signal;
38 volatile uint8_t quit_signal_rx;
39 volatile uint8_t quit_signal_dist;
40 volatile uint8_t quit_signal_work;
41 unsigned int power_lib_initialised;
42 
43 static volatile struct app_stats {
44 	struct {
45 		uint64_t rx_pkts;
46 		uint64_t returned_pkts;
47 		uint64_t enqueued_pkts;
48 		uint64_t enqdrop_pkts;
49 	} rx __rte_cache_aligned;
50 	int pad1 __rte_cache_aligned;
51 
52 	struct {
53 		uint64_t in_pkts;
54 		uint64_t ret_pkts;
55 		uint64_t sent_pkts;
56 		uint64_t enqdrop_pkts;
57 	} dist __rte_cache_aligned;
58 	int pad2 __rte_cache_aligned;
59 
60 	struct {
61 		uint64_t dequeue_pkts;
62 		uint64_t tx_pkts;
63 		uint64_t enqdrop_pkts;
64 	} tx __rte_cache_aligned;
65 	int pad3 __rte_cache_aligned;
66 
67 	uint64_t worker_pkts[64] __rte_cache_aligned;
68 
69 	int pad4 __rte_cache_aligned;
70 
71 	uint64_t worker_bursts[64][8] __rte_cache_aligned;
72 
73 	int pad5 __rte_cache_aligned;
74 
75 	uint64_t port_rx_pkts[64] __rte_cache_aligned;
76 	uint64_t port_tx_pkts[64] __rte_cache_aligned;
77 } app_stats;
78 
79 struct app_stats prev_app_stats;
80 
81 static const struct rte_eth_conf port_conf_default = {
82 	.rxmode = {
83 		.mq_mode = ETH_MQ_RX_RSS,
84 		.max_rx_pkt_len = RTE_ETHER_MAX_LEN,
85 	},
86 	.txmode = {
87 		.mq_mode = ETH_MQ_TX_NONE,
88 	},
89 	.rx_adv_conf = {
90 		.rss_conf = {
91 			.rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
92 				ETH_RSS_TCP | ETH_RSS_SCTP,
93 		}
94 	},
95 };
96 
97 struct output_buffer {
98 	unsigned count;
99 	struct rte_mbuf *mbufs[BURST_SIZE];
100 };
101 
102 static void print_stats(void);
103 
104 /*
105  * Initialises a given port using global settings and with the rx buffers
106  * coming from the mbuf_pool passed as parameter
107  */
108 static inline int
109 port_init(uint16_t port, struct rte_mempool *mbuf_pool)
110 {
111 	struct rte_eth_conf port_conf = port_conf_default;
112 	const uint16_t rxRings = 1, txRings = rte_lcore_count() - 1;
113 	int retval;
114 	uint16_t q;
115 	uint16_t nb_rxd = RX_RING_SIZE;
116 	uint16_t nb_txd = TX_RING_SIZE;
117 	struct rte_eth_dev_info dev_info;
118 	struct rte_eth_txconf txconf;
119 
120 	if (!rte_eth_dev_is_valid_port(port))
121 		return -1;
122 
123 	retval = rte_eth_dev_info_get(port, &dev_info);
124 	if (retval != 0) {
125 		printf("Error during getting device (port %u) info: %s\n",
126 				port, strerror(-retval));
127 		return retval;
128 	}
129 
130 	if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
131 		port_conf.txmode.offloads |=
132 			DEV_TX_OFFLOAD_MBUF_FAST_FREE;
133 
134 	port_conf.rx_adv_conf.rss_conf.rss_hf &=
135 		dev_info.flow_type_rss_offloads;
136 	if (port_conf.rx_adv_conf.rss_conf.rss_hf !=
137 			port_conf_default.rx_adv_conf.rss_conf.rss_hf) {
138 		printf("Port %u modified RSS hash function based on hardware support,"
139 			"requested:%#"PRIx64" configured:%#"PRIx64"\n",
140 			port,
141 			port_conf_default.rx_adv_conf.rss_conf.rss_hf,
142 			port_conf.rx_adv_conf.rss_conf.rss_hf);
143 	}
144 
145 	retval = rte_eth_dev_configure(port, rxRings, txRings, &port_conf);
146 	if (retval != 0)
147 		return retval;
148 
149 	retval = rte_eth_dev_adjust_nb_rx_tx_desc(port, &nb_rxd, &nb_txd);
150 	if (retval != 0)
151 		return retval;
152 
153 	for (q = 0; q < rxRings; q++) {
154 		retval = rte_eth_rx_queue_setup(port, q, nb_rxd,
155 						rte_eth_dev_socket_id(port),
156 						NULL, mbuf_pool);
157 		if (retval < 0)
158 			return retval;
159 	}
160 
161 	txconf = dev_info.default_txconf;
162 	txconf.offloads = port_conf.txmode.offloads;
163 	for (q = 0; q < txRings; q++) {
164 		retval = rte_eth_tx_queue_setup(port, q, nb_txd,
165 						rte_eth_dev_socket_id(port),
166 						&txconf);
167 		if (retval < 0)
168 			return retval;
169 	}
170 
171 	retval = rte_eth_dev_start(port);
172 	if (retval < 0)
173 		return retval;
174 
175 	struct rte_eth_link link;
176 	do {
177 		retval = rte_eth_link_get_nowait(port, &link);
178 		if (retval < 0) {
179 			printf("Failed link get (port %u): %s\n",
180 				port, rte_strerror(-retval));
181 			return retval;
182 		} else if (link.link_status)
183 			break;
184 
185 		printf("Waiting for Link up on port %"PRIu16"\n", port);
186 		sleep(1);
187 	} while (!link.link_status);
188 
189 	if (!link.link_status) {
190 		printf("Link down on port %"PRIu16"\n", port);
191 		return 0;
192 	}
193 
194 	struct rte_ether_addr addr;
195 	retval = rte_eth_macaddr_get(port, &addr);
196 	if (retval < 0) {
197 		printf("Failed to get MAC address (port %u): %s\n",
198 				port, rte_strerror(-retval));
199 		return retval;
200 	}
201 
202 	printf("Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8
203 			" %02"PRIx8" %02"PRIx8" %02"PRIx8"\n",
204 			port, RTE_ETHER_ADDR_BYTES(&addr));
205 
206 	retval = rte_eth_promiscuous_enable(port);
207 	if (retval != 0)
208 		return retval;
209 
210 	return 0;
211 }
212 
213 struct lcore_params {
214 	unsigned worker_id;
215 	struct rte_distributor *d;
216 	struct rte_ring *rx_dist_ring;
217 	struct rte_ring *dist_tx_ring;
218 	struct rte_mempool *mem_pool;
219 };
220 
221 static int
222 lcore_rx(struct lcore_params *p)
223 {
224 	const uint16_t nb_ports = rte_eth_dev_count_avail();
225 	const int socket_id = rte_socket_id();
226 	uint16_t port;
227 	struct rte_mbuf *bufs[BURST_SIZE*2];
228 
229 	RTE_ETH_FOREACH_DEV(port) {
230 		/* skip ports that are not enabled */
231 		if ((enabled_port_mask & (1 << port)) == 0)
232 			continue;
233 
234 		if (rte_eth_dev_socket_id(port) > 0 &&
235 				rte_eth_dev_socket_id(port) != socket_id)
236 			printf("WARNING, port %u is on remote NUMA node to "
237 					"RX thread.\n\tPerformance will not "
238 					"be optimal.\n", port);
239 	}
240 
241 	printf("\nCore %u doing packet RX.\n", rte_lcore_id());
242 	port = 0;
243 	while (!quit_signal_rx) {
244 
245 		/* skip ports that are not enabled */
246 		if ((enabled_port_mask & (1 << port)) == 0) {
247 			if (++port == nb_ports)
248 				port = 0;
249 			continue;
250 		}
251 		const uint16_t nb_rx = rte_eth_rx_burst(port, 0, bufs,
252 				BURST_SIZE);
253 		if (unlikely(nb_rx == 0)) {
254 			if (++port == nb_ports)
255 				port = 0;
256 			continue;
257 		}
258 		app_stats.rx.rx_pkts += nb_rx;
259 
260 /*
261  * You can run the distributor on the rx core with this code. Returned
262  * packets are then send straight to the tx core.
263  */
264 #if 0
265 	rte_distributor_process(d, bufs, nb_rx);
266 	const uint16_t nb_ret = rte_distributor_returned_pktsd,
267 			bufs, BURST_SIZE*2);
268 
269 		app_stats.rx.returned_pkts += nb_ret;
270 		if (unlikely(nb_ret == 0)) {
271 			if (++port == nb_ports)
272 				port = 0;
273 			continue;
274 		}
275 
276 		struct rte_ring *tx_ring = p->dist_tx_ring;
277 		uint16_t sent = rte_ring_enqueue_burst(tx_ring,
278 				(void *)bufs, nb_ret, NULL);
279 #else
280 		uint16_t nb_ret = nb_rx;
281 		/*
282 		 * Swap the following two lines if you want the rx traffic
283 		 * to go directly to tx, no distribution.
284 		 */
285 		struct rte_ring *out_ring = p->rx_dist_ring;
286 		/* struct rte_ring *out_ring = p->dist_tx_ring; */
287 
288 		uint16_t sent = rte_ring_enqueue_burst(out_ring,
289 				(void *)bufs, nb_ret, NULL);
290 #endif
291 
292 		app_stats.rx.enqueued_pkts += sent;
293 		if (unlikely(sent < nb_ret)) {
294 			app_stats.rx.enqdrop_pkts +=  nb_ret - sent;
295 			RTE_LOG_DP(DEBUG, DISTRAPP,
296 				"%s:Packet loss due to full ring\n", __func__);
297 			while (sent < nb_ret)
298 				rte_pktmbuf_free(bufs[sent++]);
299 		}
300 		if (++port == nb_ports)
301 			port = 0;
302 	}
303 	if (power_lib_initialised)
304 		rte_power_exit(rte_lcore_id());
305 	/* set worker & tx threads quit flag */
306 	printf("\nCore %u exiting rx task.\n", rte_lcore_id());
307 	quit_signal = 1;
308 	return 0;
309 }
310 
311 static inline void
312 flush_one_port(struct output_buffer *outbuf, uint8_t outp)
313 {
314 	unsigned int nb_tx = rte_eth_tx_burst(outp, 0,
315 			outbuf->mbufs, outbuf->count);
316 	app_stats.tx.tx_pkts += outbuf->count;
317 
318 	if (unlikely(nb_tx < outbuf->count)) {
319 		app_stats.tx.enqdrop_pkts +=  outbuf->count - nb_tx;
320 		do {
321 			rte_pktmbuf_free(outbuf->mbufs[nb_tx]);
322 		} while (++nb_tx < outbuf->count);
323 	}
324 	outbuf->count = 0;
325 }
326 
327 static inline void
328 flush_all_ports(struct output_buffer *tx_buffers)
329 {
330 	uint16_t outp;
331 
332 	RTE_ETH_FOREACH_DEV(outp) {
333 		/* skip ports that are not enabled */
334 		if ((enabled_port_mask & (1 << outp)) == 0)
335 			continue;
336 
337 		if (tx_buffers[outp].count == 0)
338 			continue;
339 
340 		flush_one_port(&tx_buffers[outp], outp);
341 	}
342 }
343 
344 
345 
346 static int
347 lcore_distributor(struct lcore_params *p)
348 {
349 	struct rte_ring *in_r = p->rx_dist_ring;
350 	struct rte_ring *out_r = p->dist_tx_ring;
351 	struct rte_mbuf *bufs[BURST_SIZE * 4];
352 	struct rte_distributor *d = p->d;
353 
354 	printf("\nCore %u acting as distributor core.\n", rte_lcore_id());
355 	while (!quit_signal_dist) {
356 		const uint16_t nb_rx = rte_ring_dequeue_burst(in_r,
357 				(void *)bufs, BURST_SIZE*1, NULL);
358 		if (nb_rx) {
359 			app_stats.dist.in_pkts += nb_rx;
360 
361 			/* Distribute the packets */
362 			rte_distributor_process(d, bufs, nb_rx);
363 			/* Handle Returns */
364 			const uint16_t nb_ret =
365 				rte_distributor_returned_pkts(d,
366 					bufs, BURST_SIZE*2);
367 
368 			if (unlikely(nb_ret == 0))
369 				continue;
370 			app_stats.dist.ret_pkts += nb_ret;
371 
372 			uint16_t sent = rte_ring_enqueue_burst(out_r,
373 					(void *)bufs, nb_ret, NULL);
374 			app_stats.dist.sent_pkts += sent;
375 			if (unlikely(sent < nb_ret)) {
376 				app_stats.dist.enqdrop_pkts += nb_ret - sent;
377 				RTE_LOG(DEBUG, DISTRAPP,
378 					"%s:Packet loss due to full out ring\n",
379 					__func__);
380 				while (sent < nb_ret)
381 					rte_pktmbuf_free(bufs[sent++]);
382 			}
383 		}
384 	}
385 	printf("\nCore %u exiting distributor task.\n", rte_lcore_id());
386 	quit_signal_work = 1;
387 	if (power_lib_initialised)
388 		rte_power_exit(rte_lcore_id());
389 	rte_distributor_flush(d);
390 	/* Unblock any returns so workers can exit */
391 	rte_distributor_clear_returns(d);
392 	quit_signal_rx = 1;
393 	return 0;
394 }
395 
396 
397 static int
398 lcore_tx(struct rte_ring *in_r)
399 {
400 	static struct output_buffer tx_buffers[RTE_MAX_ETHPORTS];
401 	const int socket_id = rte_socket_id();
402 	uint16_t port;
403 
404 	RTE_ETH_FOREACH_DEV(port) {
405 		/* skip ports that are not enabled */
406 		if ((enabled_port_mask & (1 << port)) == 0)
407 			continue;
408 
409 		if (rte_eth_dev_socket_id(port) > 0 &&
410 				rte_eth_dev_socket_id(port) != socket_id)
411 			printf("WARNING, port %u is on remote NUMA node to "
412 					"TX thread.\n\tPerformance will not "
413 					"be optimal.\n", port);
414 	}
415 
416 	printf("\nCore %u doing packet TX.\n", rte_lcore_id());
417 	while (!quit_signal) {
418 
419 		RTE_ETH_FOREACH_DEV(port) {
420 			/* skip ports that are not enabled */
421 			if ((enabled_port_mask & (1 << port)) == 0)
422 				continue;
423 
424 			struct rte_mbuf *bufs[BURST_SIZE_TX];
425 			const uint16_t nb_rx = rte_ring_dequeue_burst(in_r,
426 					(void *)bufs, BURST_SIZE_TX, NULL);
427 			app_stats.tx.dequeue_pkts += nb_rx;
428 
429 			/* if we get no traffic, flush anything we have */
430 			if (unlikely(nb_rx == 0)) {
431 				flush_all_ports(tx_buffers);
432 				continue;
433 			}
434 
435 			/* for traffic we receive, queue it up for transmit */
436 			uint16_t i;
437 			rte_prefetch_non_temporal((void *)bufs[0]);
438 			rte_prefetch_non_temporal((void *)bufs[1]);
439 			rte_prefetch_non_temporal((void *)bufs[2]);
440 			for (i = 0; i < nb_rx; i++) {
441 				struct output_buffer *outbuf;
442 				uint8_t outp;
443 				rte_prefetch_non_temporal((void *)bufs[i + 3]);
444 				/*
445 				 * workers should update in_port to hold the
446 				 * output port value
447 				 */
448 				outp = bufs[i]->port;
449 				/* skip ports that are not enabled */
450 				if ((enabled_port_mask & (1 << outp)) == 0)
451 					continue;
452 
453 				outbuf = &tx_buffers[outp];
454 				outbuf->mbufs[outbuf->count++] = bufs[i];
455 				if (outbuf->count == BURST_SIZE_TX)
456 					flush_one_port(outbuf, outp);
457 			}
458 		}
459 	}
460 	if (power_lib_initialised)
461 		rte_power_exit(rte_lcore_id());
462 	printf("\nCore %u exiting tx task.\n", rte_lcore_id());
463 	return 0;
464 }
465 
466 static void
467 int_handler(int sig_num)
468 {
469 	printf("Exiting on signal %d\n", sig_num);
470 	/* set quit flag for rx thread to exit */
471 	quit_signal_dist = 1;
472 }
473 
474 static void
475 print_stats(void)
476 {
477 	struct rte_eth_stats eth_stats;
478 	unsigned int i, j;
479 	const unsigned int num_workers = rte_lcore_count() - 4;
480 
481 	RTE_ETH_FOREACH_DEV(i) {
482 		rte_eth_stats_get(i, &eth_stats);
483 		app_stats.port_rx_pkts[i] = eth_stats.ipackets;
484 		app_stats.port_tx_pkts[i] = eth_stats.opackets;
485 	}
486 
487 	printf("\n\nRX Thread:\n");
488 	RTE_ETH_FOREACH_DEV(i) {
489 		printf("Port %u Pktsin : %5.2f\n", i,
490 				(app_stats.port_rx_pkts[i] -
491 				prev_app_stats.port_rx_pkts[i])/1000000.0);
492 		prev_app_stats.port_rx_pkts[i] = app_stats.port_rx_pkts[i];
493 	}
494 	printf(" - Received:    %5.2f\n",
495 			(app_stats.rx.rx_pkts -
496 			prev_app_stats.rx.rx_pkts)/1000000.0);
497 	printf(" - Returned:    %5.2f\n",
498 			(app_stats.rx.returned_pkts -
499 			prev_app_stats.rx.returned_pkts)/1000000.0);
500 	printf(" - Enqueued:    %5.2f\n",
501 			(app_stats.rx.enqueued_pkts -
502 			prev_app_stats.rx.enqueued_pkts)/1000000.0);
503 	printf(" - Dropped:     %s%5.2f%s\n", ANSI_COLOR_RED,
504 			(app_stats.rx.enqdrop_pkts -
505 			prev_app_stats.rx.enqdrop_pkts)/1000000.0,
506 			ANSI_COLOR_RESET);
507 
508 	printf("Distributor thread:\n");
509 	printf(" - In:          %5.2f\n",
510 			(app_stats.dist.in_pkts -
511 			prev_app_stats.dist.in_pkts)/1000000.0);
512 	printf(" - Returned:    %5.2f\n",
513 			(app_stats.dist.ret_pkts -
514 			prev_app_stats.dist.ret_pkts)/1000000.0);
515 	printf(" - Sent:        %5.2f\n",
516 			(app_stats.dist.sent_pkts -
517 			prev_app_stats.dist.sent_pkts)/1000000.0);
518 	printf(" - Dropped      %s%5.2f%s\n", ANSI_COLOR_RED,
519 			(app_stats.dist.enqdrop_pkts -
520 			prev_app_stats.dist.enqdrop_pkts)/1000000.0,
521 			ANSI_COLOR_RESET);
522 
523 	printf("TX thread:\n");
524 	printf(" - Dequeued:    %5.2f\n",
525 			(app_stats.tx.dequeue_pkts -
526 			prev_app_stats.tx.dequeue_pkts)/1000000.0);
527 	RTE_ETH_FOREACH_DEV(i) {
528 		printf("Port %u Pktsout: %5.2f\n",
529 				i, (app_stats.port_tx_pkts[i] -
530 				prev_app_stats.port_tx_pkts[i])/1000000.0);
531 		prev_app_stats.port_tx_pkts[i] = app_stats.port_tx_pkts[i];
532 	}
533 	printf(" - Transmitted: %5.2f\n",
534 			(app_stats.tx.tx_pkts -
535 			prev_app_stats.tx.tx_pkts)/1000000.0);
536 	printf(" - Dropped:     %s%5.2f%s\n", ANSI_COLOR_RED,
537 			(app_stats.tx.enqdrop_pkts -
538 			prev_app_stats.tx.enqdrop_pkts)/1000000.0,
539 			ANSI_COLOR_RESET);
540 
541 	prev_app_stats.rx.rx_pkts = app_stats.rx.rx_pkts;
542 	prev_app_stats.rx.returned_pkts = app_stats.rx.returned_pkts;
543 	prev_app_stats.rx.enqueued_pkts = app_stats.rx.enqueued_pkts;
544 	prev_app_stats.rx.enqdrop_pkts = app_stats.rx.enqdrop_pkts;
545 	prev_app_stats.dist.in_pkts = app_stats.dist.in_pkts;
546 	prev_app_stats.dist.ret_pkts = app_stats.dist.ret_pkts;
547 	prev_app_stats.dist.sent_pkts = app_stats.dist.sent_pkts;
548 	prev_app_stats.dist.enqdrop_pkts = app_stats.dist.enqdrop_pkts;
549 	prev_app_stats.tx.dequeue_pkts = app_stats.tx.dequeue_pkts;
550 	prev_app_stats.tx.tx_pkts = app_stats.tx.tx_pkts;
551 	prev_app_stats.tx.enqdrop_pkts = app_stats.tx.enqdrop_pkts;
552 
553 	for (i = 0; i < num_workers; i++) {
554 		printf("Worker %02u Pkts: %5.2f. Bursts(1-8): ", i,
555 				(app_stats.worker_pkts[i] -
556 				prev_app_stats.worker_pkts[i])/1000000.0);
557 		for (j = 0; j < 8; j++) {
558 			printf("%"PRIu64" ", app_stats.worker_bursts[i][j]);
559 			app_stats.worker_bursts[i][j] = 0;
560 		}
561 		printf("\n");
562 		prev_app_stats.worker_pkts[i] = app_stats.worker_pkts[i];
563 	}
564 }
565 
566 static int
567 lcore_worker(struct lcore_params *p)
568 {
569 	struct rte_distributor *d = p->d;
570 	const unsigned id = p->worker_id;
571 	unsigned int num = 0;
572 	unsigned int i;
573 
574 	/*
575 	 * for single port, xor_val will be zero so we won't modify the output
576 	 * port, otherwise we send traffic from 0 to 1, 2 to 3, and vice versa
577 	 */
578 	const unsigned xor_val = (rte_eth_dev_count_avail() > 1);
579 	struct rte_mbuf *buf[8] __rte_cache_aligned;
580 
581 	for (i = 0; i < 8; i++)
582 		buf[i] = NULL;
583 
584 	app_stats.worker_pkts[p->worker_id] = 1;
585 
586 	printf("\nCore %u acting as worker core.\n", rte_lcore_id());
587 	while (!quit_signal_work) {
588 		num = rte_distributor_get_pkt(d, id, buf, buf, num);
589 		/* Do a little bit of work for each packet */
590 		for (i = 0; i < num; i++) {
591 			uint64_t t = rte_rdtsc()+100;
592 
593 			while (rte_rdtsc() < t)
594 				rte_pause();
595 			buf[i]->port ^= xor_val;
596 		}
597 
598 		app_stats.worker_pkts[p->worker_id] += num;
599 		if (num > 0)
600 			app_stats.worker_bursts[p->worker_id][num-1]++;
601 	}
602 	if (power_lib_initialised)
603 		rte_power_exit(rte_lcore_id());
604 	rte_free(p);
605 	return 0;
606 }
607 
608 static int
609 init_power_library(void)
610 {
611 	int ret = 0, lcore_id;
612 	RTE_LCORE_FOREACH_WORKER(lcore_id) {
613 		/* init power management library */
614 		ret = rte_power_init(lcore_id);
615 		if (ret) {
616 			RTE_LOG(ERR, POWER,
617 				"Library initialization failed on core %u\n",
618 				lcore_id);
619 			/*
620 			 * Return on first failure, we'll fall back
621 			 * to non-power operation
622 			 */
623 			return ret;
624 		}
625 	}
626 	return ret;
627 }
628 
629 /* display usage */
630 static void
631 print_usage(const char *prgname)
632 {
633 	printf("%s [EAL options] -- -p PORTMASK\n"
634 			"  -p PORTMASK: hexadecimal bitmask of ports to configure\n",
635 			prgname);
636 }
637 
638 static int
639 parse_portmask(const char *portmask)
640 {
641 	char *end = NULL;
642 	unsigned long pm;
643 
644 	/* parse hexadecimal string */
645 	pm = strtoul(portmask, &end, 16);
646 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
647 		return 0;
648 
649 	return pm;
650 }
651 
652 /* Parse the argument given in the command line of the application */
653 static int
654 parse_args(int argc, char **argv)
655 {
656 	int opt;
657 	char **argvopt;
658 	int option_index;
659 	char *prgname = argv[0];
660 	static struct option lgopts[] = {
661 		{NULL, 0, 0, 0}
662 	};
663 
664 	argvopt = argv;
665 
666 	while ((opt = getopt_long(argc, argvopt, "p:",
667 			lgopts, &option_index)) != EOF) {
668 
669 		switch (opt) {
670 		/* portmask */
671 		case 'p':
672 			enabled_port_mask = parse_portmask(optarg);
673 			if (enabled_port_mask == 0) {
674 				printf("invalid portmask\n");
675 				print_usage(prgname);
676 				return -1;
677 			}
678 			break;
679 
680 		default:
681 			print_usage(prgname);
682 			return -1;
683 		}
684 	}
685 
686 	if (optind <= 1) {
687 		print_usage(prgname);
688 		return -1;
689 	}
690 
691 	argv[optind-1] = prgname;
692 
693 	optind = 1; /* reset getopt lib */
694 	return 0;
695 }
696 
697 /* Main function, does initialization and calls the per-lcore functions */
698 int
699 main(int argc, char *argv[])
700 {
701 	struct rte_mempool *mbuf_pool;
702 	struct rte_distributor *d;
703 	struct rte_ring *dist_tx_ring;
704 	struct rte_ring *rx_dist_ring;
705 	struct rte_power_core_capabilities lcore_cap;
706 	unsigned int lcore_id, worker_id = 0;
707 	int distr_core_id = -1, rx_core_id = -1, tx_core_id = -1;
708 	unsigned nb_ports;
709 	uint16_t portid;
710 	uint16_t nb_ports_available;
711 	uint64_t t, freq;
712 
713 	/* catch ctrl-c so we can print on exit */
714 	signal(SIGINT, int_handler);
715 
716 	/* init EAL */
717 	int ret = rte_eal_init(argc, argv);
718 	if (ret < 0)
719 		rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
720 	argc -= ret;
721 	argv += ret;
722 
723 	/* parse application arguments (after the EAL ones) */
724 	ret = parse_args(argc, argv);
725 	if (ret < 0)
726 		rte_exit(EXIT_FAILURE, "Invalid distributor parameters\n");
727 
728 	if (rte_lcore_count() < 5)
729 		rte_exit(EXIT_FAILURE, "Error, This application needs at "
730 				"least 5 logical cores to run:\n"
731 				"1 lcore for stats (can be core 0)\n"
732 				"1 lcore for packet RX\n"
733 				"1 lcore for distribution\n"
734 				"1 lcore for packet TX\n"
735 				"and at least 1 lcore for worker threads\n");
736 
737 	if (init_power_library() == 0)
738 		power_lib_initialised = 1;
739 
740 	nb_ports = rte_eth_dev_count_avail();
741 	if (nb_ports == 0)
742 		rte_exit(EXIT_FAILURE, "Error: no ethernet ports detected\n");
743 	if (nb_ports != 1 && (nb_ports & 1))
744 		rte_exit(EXIT_FAILURE, "Error: number of ports must be even, except "
745 				"when using a single port\n");
746 
747 	mbuf_pool = rte_pktmbuf_pool_create("MBUF_POOL",
748 		NUM_MBUFS * nb_ports, MBUF_CACHE_SIZE, 0,
749 		RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id());
750 	if (mbuf_pool == NULL)
751 		rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
752 	nb_ports_available = nb_ports;
753 
754 	/* initialize all ports */
755 	RTE_ETH_FOREACH_DEV(portid) {
756 		/* skip ports that are not enabled */
757 		if ((enabled_port_mask & (1 << portid)) == 0) {
758 			printf("\nSkipping disabled port %d\n", portid);
759 			nb_ports_available--;
760 			continue;
761 		}
762 		/* init port */
763 		printf("Initializing port %u... done\n", portid);
764 
765 		if (port_init(portid, mbuf_pool) != 0)
766 			rte_exit(EXIT_FAILURE, "Cannot initialize port %u\n",
767 					portid);
768 	}
769 
770 	if (!nb_ports_available) {
771 		rte_exit(EXIT_FAILURE,
772 				"All available ports are disabled. Please set portmask.\n");
773 	}
774 
775 	d = rte_distributor_create("PKT_DIST", rte_socket_id(),
776 			rte_lcore_count() - 4,
777 			RTE_DIST_ALG_BURST);
778 	if (d == NULL)
779 		rte_exit(EXIT_FAILURE, "Cannot create distributor\n");
780 
781 	/*
782 	 * scheduler ring is read by the transmitter core, and written to
783 	 * by scheduler core
784 	 */
785 	dist_tx_ring = rte_ring_create("Output_ring", SCHED_TX_RING_SZ,
786 			rte_socket_id(), RING_F_SC_DEQ | RING_F_SP_ENQ);
787 	if (dist_tx_ring == NULL)
788 		rte_exit(EXIT_FAILURE, "Cannot create output ring\n");
789 
790 	rx_dist_ring = rte_ring_create("Input_ring", SCHED_RX_RING_SZ,
791 			rte_socket_id(), RING_F_SC_DEQ | RING_F_SP_ENQ);
792 	if (rx_dist_ring == NULL)
793 		rte_exit(EXIT_FAILURE, "Cannot create output ring\n");
794 
795 	if (power_lib_initialised) {
796 		/*
797 		 * Here we'll pre-assign lcore ids to the rx, tx and
798 		 * distributor workloads if there's higher frequency
799 		 * on those cores e.g. if Turbo Boost is enabled.
800 		 * It's also worth mentioning that it will assign cores in a
801 		 * specific order, so that if there's less than three
802 		 * available, the higher frequency cores will go to the
803 		 * distributor first, then rx, then tx.
804 		 */
805 		RTE_LCORE_FOREACH_WORKER(lcore_id) {
806 
807 			rte_power_get_capabilities(lcore_id, &lcore_cap);
808 
809 			if (lcore_cap.priority != 1)
810 				continue;
811 
812 			if (distr_core_id < 0) {
813 				distr_core_id = lcore_id;
814 				printf("Distributor on priority core %d\n",
815 					lcore_id);
816 				continue;
817 			}
818 			if (rx_core_id < 0) {
819 				rx_core_id = lcore_id;
820 				printf("Rx on priority core %d\n",
821 					lcore_id);
822 				continue;
823 			}
824 			if (tx_core_id < 0) {
825 				tx_core_id = lcore_id;
826 				printf("Tx on priority core %d\n",
827 					lcore_id);
828 				continue;
829 			}
830 		}
831 	}
832 
833 	/*
834 	 * If there's any of the key workloads left without an lcore_id
835 	 * after the high performing core assignment above, pre-assign
836 	 * them here.
837 	 */
838 	RTE_LCORE_FOREACH_WORKER(lcore_id) {
839 		if (lcore_id == (unsigned int)distr_core_id ||
840 				lcore_id == (unsigned int)rx_core_id ||
841 				lcore_id == (unsigned int)tx_core_id)
842 			continue;
843 		if (distr_core_id < 0) {
844 			distr_core_id = lcore_id;
845 			printf("Distributor on core %d\n", lcore_id);
846 			continue;
847 		}
848 		if (rx_core_id < 0) {
849 			rx_core_id = lcore_id;
850 			printf("Rx on core %d\n", lcore_id);
851 			continue;
852 		}
853 		if (tx_core_id < 0) {
854 			tx_core_id = lcore_id;
855 			printf("Tx on core %d\n", lcore_id);
856 			continue;
857 		}
858 	}
859 
860 	printf(" tx id %d, dist id %d, rx id %d\n",
861 			tx_core_id,
862 			distr_core_id,
863 			rx_core_id);
864 
865 	/*
866 	 * Kick off all the worker threads first, avoiding the pre-assigned
867 	 * lcore_ids for tx, rx and distributor workloads.
868 	 */
869 	RTE_LCORE_FOREACH_WORKER(lcore_id) {
870 		if (lcore_id == (unsigned int)distr_core_id ||
871 				lcore_id == (unsigned int)rx_core_id ||
872 				lcore_id == (unsigned int)tx_core_id)
873 			continue;
874 		printf("Starting thread %d as worker, lcore_id %d\n",
875 				worker_id, lcore_id);
876 		struct lcore_params *p =
877 			rte_malloc(NULL, sizeof(*p), 0);
878 		if (!p)
879 			rte_panic("malloc failure\n");
880 		*p = (struct lcore_params){worker_id++, d, rx_dist_ring,
881 			dist_tx_ring, mbuf_pool};
882 
883 		rte_eal_remote_launch((lcore_function_t *)lcore_worker,
884 				p, lcore_id);
885 	}
886 
887 	/* Start tx core */
888 	rte_eal_remote_launch((lcore_function_t *)lcore_tx,
889 			dist_tx_ring, tx_core_id);
890 
891 	/* Start distributor core */
892 	struct lcore_params *pd =
893 		rte_malloc(NULL, sizeof(*pd), 0);
894 	if (!pd)
895 		rte_panic("malloc failure\n");
896 	*pd = (struct lcore_params){worker_id++, d,
897 		rx_dist_ring, dist_tx_ring, mbuf_pool};
898 	rte_eal_remote_launch(
899 			(lcore_function_t *)lcore_distributor,
900 			pd, distr_core_id);
901 
902 	/* Start rx core */
903 	struct lcore_params *pr =
904 		rte_malloc(NULL, sizeof(*pr), 0);
905 	if (!pr)
906 		rte_panic("malloc failure\n");
907 	*pr = (struct lcore_params){worker_id++, d, rx_dist_ring,
908 		dist_tx_ring, mbuf_pool};
909 	rte_eal_remote_launch((lcore_function_t *)lcore_rx,
910 			pr, rx_core_id);
911 
912 	freq = rte_get_timer_hz();
913 	t = rte_rdtsc() + freq;
914 	while (!quit_signal_dist) {
915 		if (t < rte_rdtsc()) {
916 			print_stats();
917 			t = rte_rdtsc() + freq;
918 		}
919 		usleep(1000);
920 	}
921 
922 	RTE_LCORE_FOREACH_WORKER(lcore_id) {
923 		if (rte_eal_wait_lcore(lcore_id) < 0)
924 			return -1;
925 	}
926 
927 	print_stats();
928 
929 	rte_free(pd);
930 	rte_free(pr);
931 
932 	/* clean up the EAL */
933 	rte_eal_cleanup();
934 
935 	return 0;
936 }
937