xref: /dpdk/app/test/test_pmd_perf.c (revision 0dff3f26d6faad4e51f75e5245f0387ee9bb0c6d)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2014 Intel Corporation
3  */
4 
5 
6 #include <stdio.h>
7 #include <inttypes.h>
8 #include <signal.h>
9 #include <unistd.h>
10 #include <rte_cycles.h>
11 #include <rte_ethdev.h>
12 #include <rte_byteorder.h>
13 #include <rte_malloc.h>
14 #include "packet_burst_generator.h"
15 #include "test.h"
16 
17 #define NB_ETHPORTS_USED                (1)
18 #define NB_SOCKETS                      (2)
19 #define MEMPOOL_CACHE_SIZE 250
20 #define MAX_PKT_BURST                   (32)
21 #define RTE_TEST_RX_DESC_DEFAULT        (1024)
22 #define RTE_TEST_TX_DESC_DEFAULT        (1024)
23 #define RTE_PORT_ALL            (~(uint16_t)0x0)
24 
25 /* how long test would take at full line rate */
26 #define RTE_TEST_DURATION                (2)
27 
28 /*
29  * RX and TX Prefetch, Host, and Write-back threshold values should be
30  * carefully set for optimal performance. Consult the network
31  * controller's datasheet and supporting DPDK documentation for guidance
32  * on how these parameters should be set.
33  */
34 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
35 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
36 #define RX_WTHRESH 0 /**< Default values of RX write-back threshold reg. */
37 
38 /*
39  * These default values are optimized for use with the Intel(R) 82599 10 GbE
40  * Controller and the DPDK ixgbe PMD. Consider using other values for other
41  * network controllers and/or network drivers.
42  */
43 #define TX_PTHRESH 32 /**< Default values of TX prefetch threshold reg. */
44 #define TX_HTHRESH 0  /**< Default values of TX host threshold reg. */
45 #define TX_WTHRESH 0  /**< Default values of TX write-back threshold reg. */
46 
47 #define MAX_TRAFFIC_BURST              2048
48 
49 #define NB_MBUF RTE_MAX(						\
50 		(unsigned)(nb_ports*nb_rx_queue*nb_rxd +		\
51 			   nb_ports*nb_lcores*MAX_PKT_BURST +		\
52 			   nb_ports*nb_tx_queue*nb_txd +		\
53 			   nb_lcores*MEMPOOL_CACHE_SIZE +		\
54 			   nb_ports*MAX_TRAFFIC_BURST),			\
55 			(unsigned)8192)
56 
57 
58 static struct rte_mempool *mbufpool[NB_SOCKETS];
59 /* ethernet addresses of ports */
60 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
61 
62 static struct rte_eth_conf port_conf = {
63 	.rxmode = {
64 		.mq_mode = RTE_ETH_MQ_RX_NONE,
65 		.split_hdr_size = 0,
66 	},
67 	.txmode = {
68 		.mq_mode = RTE_ETH_MQ_TX_NONE,
69 	},
70 	.lpbk_mode = 1,  /* enable loopback */
71 };
72 
73 static struct rte_eth_rxconf rx_conf = {
74 	.rx_thresh = {
75 		.pthresh = RX_PTHRESH,
76 		.hthresh = RX_HTHRESH,
77 		.wthresh = RX_WTHRESH,
78 	},
79 	.rx_free_thresh = 32,
80 };
81 
82 static struct rte_eth_txconf tx_conf = {
83 	.tx_thresh = {
84 		.pthresh = TX_PTHRESH,
85 		.hthresh = TX_HTHRESH,
86 		.wthresh = TX_WTHRESH,
87 	},
88 	.tx_free_thresh = 32, /* Use PMD default values */
89 	.tx_rs_thresh = 32, /* Use PMD default values */
90 };
91 
92 enum {
93 	LCORE_INVALID = 0,
94 	LCORE_AVAIL,
95 	LCORE_USED,
96 };
97 
98 struct lcore_conf {
99 	uint8_t status;
100 	uint8_t socketid;
101 	uint16_t nb_ports;
102 	uint16_t portlist[RTE_MAX_ETHPORTS];
103 } __rte_cache_aligned;
104 
105 struct lcore_conf lcore_conf[RTE_MAX_LCORE];
106 
107 static uint64_t link_mbps;
108 
109 enum {
110 	SC_CONTINUOUS = 0,
111 	SC_BURST_POLL_FIRST,
112 	SC_BURST_XMIT_FIRST,
113 };
114 
115 static uint32_t sc_flag;
116 
117 /* Check the link status of all ports in up to 3s, and print them finally */
118 static void
119 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask)
120 {
121 #define CHECK_INTERVAL 100 /* 100ms */
122 #define MAX_CHECK_TIME 30 /* 3s (30 * 100ms) in total */
123 	uint16_t portid;
124 	uint8_t count, all_ports_up, print_flag = 0;
125 	struct rte_eth_link link;
126 	int ret;
127 	char link_status[RTE_ETH_LINK_MAX_STR_LEN];
128 
129 	printf("Checking link statuses...\n");
130 	fflush(stdout);
131 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
132 		all_ports_up = 1;
133 		for (portid = 0; portid < port_num; portid++) {
134 			if ((port_mask & (1 << portid)) == 0)
135 				continue;
136 			memset(&link, 0, sizeof(link));
137 			ret = rte_eth_link_get_nowait(portid, &link);
138 			if (ret < 0) {
139 				all_ports_up = 0;
140 				if (print_flag == 1)
141 					printf("Port %u link get failed: %s\n",
142 						portid, rte_strerror(-ret));
143 				continue;
144 			}
145 
146 			/* print link status if flag set */
147 			if (print_flag == 1) {
148 				if (link.link_status && link_mbps == 0)
149 					link_mbps = link.link_speed;
150 
151 				rte_eth_link_to_str(link_status,
152 					sizeof(link_status), &link);
153 				printf("Port %d %s\n", portid, link_status);
154 				continue;
155 			}
156 			/* clear all_ports_up flag if any link down */
157 			if (link.link_status == RTE_ETH_LINK_DOWN) {
158 				all_ports_up = 0;
159 				break;
160 			}
161 		}
162 		/* after finally printing all link status, get out */
163 		if (print_flag == 1)
164 			break;
165 
166 		if (all_ports_up == 0) {
167 			fflush(stdout);
168 			rte_delay_ms(CHECK_INTERVAL);
169 		}
170 
171 		/* set the print_flag if all ports up or timeout */
172 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1))
173 			print_flag = 1;
174 	}
175 }
176 
177 static void
178 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
179 {
180 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
181 	rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
182 	printf("%s%s", name, buf);
183 }
184 
185 static int
186 init_traffic(struct rte_mempool *mp,
187 	     struct rte_mbuf **pkts_burst, uint32_t burst_size)
188 {
189 	struct rte_ether_hdr pkt_eth_hdr;
190 	struct rte_ipv4_hdr pkt_ipv4_hdr;
191 	struct rte_udp_hdr pkt_udp_hdr;
192 	uint32_t pktlen;
193 	static uint8_t src_mac[] = { 0x00, 0xFF, 0xAA, 0xFF, 0xAA, 0xFF };
194 	static uint8_t dst_mac[] = { 0x00, 0xAA, 0xFF, 0xAA, 0xFF, 0xAA };
195 
196 
197 	initialize_eth_header(&pkt_eth_hdr,
198 		(struct rte_ether_addr *)src_mac,
199 		(struct rte_ether_addr *)dst_mac, RTE_ETHER_TYPE_IPV4, 0, 0);
200 
201 	pktlen = initialize_ipv4_header(&pkt_ipv4_hdr,
202 					IPV4_ADDR(10, 0, 0, 1),
203 					IPV4_ADDR(10, 0, 0, 2), 26);
204 	printf("IPv4 pktlen %u\n", pktlen);
205 
206 	pktlen = initialize_udp_header(&pkt_udp_hdr, 0, 0, 18);
207 
208 	printf("UDP pktlen %u\n", pktlen);
209 
210 	return generate_packet_burst(mp, pkts_burst, &pkt_eth_hdr,
211 				     0, &pkt_ipv4_hdr, 1,
212 				     &pkt_udp_hdr, burst_size,
213 				     PACKET_BURST_GEN_PKT_LEN, 1);
214 }
215 
216 static int
217 init_lcores(void)
218 {
219 	unsigned lcore_id;
220 
221 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
222 		lcore_conf[lcore_id].socketid =
223 			rte_lcore_to_socket_id(lcore_id);
224 		if (rte_lcore_is_enabled(lcore_id) == 0) {
225 			lcore_conf[lcore_id].status = LCORE_INVALID;
226 			continue;
227 		} else
228 			lcore_conf[lcore_id].status = LCORE_AVAIL;
229 	}
230 	return 0;
231 }
232 
233 static int
234 init_mbufpool(unsigned nb_mbuf)
235 {
236 	int socketid;
237 	unsigned lcore_id;
238 	char s[64];
239 
240 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
241 		if (rte_lcore_is_enabled(lcore_id) == 0)
242 			continue;
243 
244 		socketid = rte_lcore_to_socket_id(lcore_id);
245 		if (socketid >= NB_SOCKETS) {
246 			rte_exit(EXIT_FAILURE,
247 				"Socket %d of lcore %u is out of range %d\n",
248 				socketid, lcore_id, NB_SOCKETS);
249 		}
250 		if (mbufpool[socketid] == NULL) {
251 			snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
252 			mbufpool[socketid] =
253 				rte_pktmbuf_pool_create(s, nb_mbuf,
254 					MEMPOOL_CACHE_SIZE, 0,
255 					RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
256 			if (mbufpool[socketid] == NULL)
257 				rte_exit(EXIT_FAILURE,
258 					"Cannot init mbuf pool on socket %d\n",
259 					socketid);
260 			else
261 				printf("Allocated mbuf pool on socket %d\n",
262 					socketid);
263 		}
264 	}
265 	return 0;
266 }
267 
268 static uint16_t
269 alloc_lcore(uint16_t socketid)
270 {
271 	unsigned lcore_id;
272 
273 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
274 		if (LCORE_AVAIL != lcore_conf[lcore_id].status ||
275 		    lcore_conf[lcore_id].socketid != socketid ||
276 		    lcore_id == rte_get_main_lcore())
277 			continue;
278 		lcore_conf[lcore_id].status = LCORE_USED;
279 		lcore_conf[lcore_id].nb_ports = 0;
280 		return lcore_id;
281 	}
282 
283 	return (uint16_t)-1;
284 }
285 
286 static volatile uint64_t stop;
287 static uint64_t count;
288 static uint64_t drop;
289 static uint64_t idle;
290 
291 static void
292 reset_count(void)
293 {
294 	count = 0;
295 	drop = 0;
296 	idle = 0;
297 }
298 
299 #ifndef RTE_EXEC_ENV_WINDOWS
300 static void
301 stats_display(uint16_t port_id)
302 {
303 	struct rte_eth_stats stats;
304 	rte_eth_stats_get(port_id, &stats);
305 
306 	printf("  RX-packets: %-10"PRIu64" RX-missed: %-10"PRIu64" RX-bytes:  "
307 	       "%-"PRIu64"\n",
308 	       stats.ipackets, stats.imissed, stats.ibytes);
309 	printf("  RX-errors: %-10"PRIu64" RX-nombuf:  %-10"PRIu64"\n",
310 	       stats.ierrors, stats.rx_nombuf);
311 	printf("  TX-packets: %-10"PRIu64" TX-errors: %-10"PRIu64" TX-bytes:  "
312 	       "%-"PRIu64"\n",
313 	       stats.opackets, stats.oerrors, stats.obytes);
314 }
315 
316 static void
317 signal_handler(int signum)
318 {
319 	/*  USR1 signal, stop testing */
320 	if (signum == SIGUSR1) {
321 		printf("Force Stop!\n");
322 		stop = 1;
323 	}
324 
325 	/*  USR2 signal, print stats */
326 	if (signum == SIGUSR2)
327 		stats_display(0);
328 }
329 #endif
330 
331 struct rte_mbuf **tx_burst;
332 
333 uint64_t (*do_measure)(struct lcore_conf *conf,
334 		       struct rte_mbuf *pkts_burst[],
335 		       uint64_t total_pkts);
336 
337 static uint64_t
338 measure_rxtx(struct lcore_conf *conf,
339 	     struct rte_mbuf *pkts_burst[],
340 	     uint64_t total_pkts)
341 {
342 	unsigned i, portid, nb_rx, nb_tx;
343 	uint64_t prev_tsc, cur_tsc;
344 
345 	prev_tsc = rte_rdtsc();
346 
347 	while (likely(!stop)) {
348 		for (i = 0; i < conf->nb_ports; i++) {
349 			portid = conf->portlist[i];
350 			nb_rx = rte_eth_rx_burst(portid, 0,
351 						 pkts_burst, MAX_PKT_BURST);
352 			if (unlikely(nb_rx == 0)) {
353 				idle++;
354 				continue;
355 			}
356 
357 			count += nb_rx;
358 			nb_tx = rte_eth_tx_burst(portid, 0, pkts_burst, nb_rx);
359 			if (unlikely(nb_tx < nb_rx)) {
360 				drop += (nb_rx - nb_tx);
361 				do {
362 					rte_pktmbuf_free(pkts_burst[nb_tx]);
363 				} while (++nb_tx < nb_rx);
364 			}
365 		}
366 		if (unlikely(count >= total_pkts))
367 			break;
368 	}
369 
370 	cur_tsc = rte_rdtsc();
371 
372 	return cur_tsc - prev_tsc;
373 }
374 
375 static uint64_t
376 measure_rxonly(struct lcore_conf *conf,
377 	       struct rte_mbuf *pkts_burst[],
378 	       uint64_t total_pkts)
379 {
380 	unsigned i, portid, nb_rx, nb_tx;
381 	uint64_t diff_tsc, cur_tsc;
382 
383 	diff_tsc = 0;
384 	while (likely(!stop)) {
385 		for (i = 0; i < conf->nb_ports; i++) {
386 			portid = conf->portlist[i];
387 
388 			cur_tsc = rte_rdtsc();
389 			nb_rx = rte_eth_rx_burst(portid, 0,
390 						 pkts_burst, MAX_PKT_BURST);
391 			if (unlikely(nb_rx == 0)) {
392 				idle++;
393 				continue;
394 			}
395 			diff_tsc += rte_rdtsc() - cur_tsc;
396 
397 			count += nb_rx;
398 			nb_tx = rte_eth_tx_burst(portid, 0, pkts_burst, nb_rx);
399 			if (unlikely(nb_tx < nb_rx)) {
400 				drop += (nb_rx - nb_tx);
401 				do {
402 					rte_pktmbuf_free(pkts_burst[nb_tx]);
403 				} while (++nb_tx < nb_rx);
404 			}
405 		}
406 		if (unlikely(count >= total_pkts))
407 			break;
408 	}
409 
410 	return diff_tsc;
411 }
412 
413 static uint64_t
414 measure_txonly(struct lcore_conf *conf,
415 	       struct rte_mbuf *pkts_burst[],
416 	       uint64_t total_pkts)
417 {
418 	unsigned i, portid, nb_rx, nb_tx;
419 	uint64_t diff_tsc, cur_tsc;
420 
421 	printf("do tx measure\n");
422 	diff_tsc = 0;
423 	while (likely(!stop)) {
424 		for (i = 0; i < conf->nb_ports; i++) {
425 			portid = conf->portlist[i];
426 			nb_rx = rte_eth_rx_burst(portid, 0,
427 						 pkts_burst, MAX_PKT_BURST);
428 			if (unlikely(nb_rx == 0)) {
429 				idle++;
430 				continue;
431 			}
432 
433 			count += nb_rx;
434 
435 			cur_tsc = rte_rdtsc();
436 			nb_tx = rte_eth_tx_burst(portid, 0, pkts_burst, nb_rx);
437 			if (unlikely(nb_tx < nb_rx)) {
438 				drop += (nb_rx - nb_tx);
439 				do {
440 					rte_pktmbuf_free(pkts_burst[nb_tx]);
441 				} while (++nb_tx < nb_rx);
442 			}
443 			diff_tsc += rte_rdtsc() - cur_tsc;
444 		}
445 		if (unlikely(count >= total_pkts))
446 			break;
447 	}
448 
449 	return diff_tsc;
450 }
451 
452 /* main processing loop */
453 static int
454 main_loop(__rte_unused void *args)
455 {
456 #define PACKET_SIZE 64
457 #define FRAME_GAP 12
458 #define MAC_PREAMBLE 8
459 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
460 	unsigned lcore_id;
461 	unsigned i, portid, nb_rx = 0, nb_tx = 0;
462 	struct lcore_conf *conf;
463 	int pkt_per_port;
464 	uint64_t diff_tsc;
465 	uint64_t packets_per_second, total_packets;
466 
467 	lcore_id = rte_lcore_id();
468 	conf = &lcore_conf[lcore_id];
469 	if (conf->status != LCORE_USED)
470 		return 0;
471 
472 	pkt_per_port = MAX_TRAFFIC_BURST;
473 
474 	int idx = 0;
475 	for (i = 0; i < conf->nb_ports; i++) {
476 		int num = pkt_per_port;
477 		portid = conf->portlist[i];
478 		printf("inject %d packet to port %d\n", num, portid);
479 		while (num) {
480 			nb_tx = RTE_MIN(MAX_PKT_BURST, num);
481 			nb_tx = rte_eth_tx_burst(portid, 0,
482 						&tx_burst[idx], nb_tx);
483 			num -= nb_tx;
484 			idx += nb_tx;
485 		}
486 	}
487 	printf("Total packets inject to prime ports = %u\n", idx);
488 
489 	packets_per_second = (link_mbps * 1000 * 1000) /
490 		((PACKET_SIZE + FRAME_GAP + MAC_PREAMBLE) * CHAR_BIT);
491 	printf("Each port will do %"PRIu64" packets per second\n",
492 	       packets_per_second);
493 
494 	total_packets = RTE_TEST_DURATION * conf->nb_ports * packets_per_second;
495 	printf("Test will stop after at least %"PRIu64" packets received\n",
496 		+ total_packets);
497 
498 	diff_tsc = do_measure(conf, pkts_burst, total_packets);
499 
500 	for (i = 0; i < conf->nb_ports; i++) {
501 		portid = conf->portlist[i];
502 		int nb_free = 0;
503 		uint64_t timeout = 10000;
504 		do { /* dry out */
505 			nb_rx = rte_eth_rx_burst(portid, 0,
506 						 pkts_burst, MAX_PKT_BURST);
507 			nb_tx = 0;
508 			while (nb_tx < nb_rx)
509 				rte_pktmbuf_free(pkts_burst[nb_tx++]);
510 			nb_free += nb_rx;
511 
512 			if (unlikely(nb_rx == 0))
513 				timeout--;
514 		} while (nb_free != pkt_per_port && timeout != 0);
515 		printf("free %d (expected %d) mbuf left in port %u\n", nb_free,
516 		       pkt_per_port, portid);
517 	}
518 
519 	if (count == 0)
520 		return -1;
521 
522 	printf("%"PRIu64" packet, %"PRIu64" drop, %"PRIu64" idle\n",
523 	       count, drop, idle);
524 	printf("Result: %"PRIu64" cycles per packet\n", diff_tsc / count);
525 
526 	return 0;
527 }
528 
529 static uint64_t start;
530 
531 static inline int
532 poll_burst(void *args)
533 {
534 #define MAX_IDLE           (10000)
535 	unsigned lcore_id;
536 	struct rte_mbuf **pkts_burst;
537 	uint64_t diff_tsc, cur_tsc;
538 	uint16_t next[RTE_MAX_ETHPORTS];
539 	struct lcore_conf *conf;
540 	uint32_t pkt_per_port = *((uint32_t *)args);
541 	unsigned i, portid, nb_rx = 0;
542 	uint64_t total;
543 	uint64_t timeout = MAX_IDLE;
544 	int num[RTE_MAX_ETHPORTS];
545 
546 	lcore_id = rte_lcore_id();
547 	conf = &lcore_conf[lcore_id];
548 	if (conf->status != LCORE_USED)
549 		return 0;
550 
551 	total = pkt_per_port * conf->nb_ports;
552 	printf("start to receive total expect %"PRIu64"\n", total);
553 
554 	pkts_burst = (struct rte_mbuf **)
555 		rte_calloc_socket("poll_burst",
556 				  total, sizeof(void *),
557 				  RTE_CACHE_LINE_SIZE, conf->socketid);
558 	if (!pkts_burst)
559 		return -1;
560 
561 	for (i = 0; i < conf->nb_ports; i++) {
562 		portid = conf->portlist[i];
563 		next[portid] = i * pkt_per_port;
564 		num[portid] = pkt_per_port;
565 	}
566 
567 	rte_wait_until_equal_64(&start, 1, __ATOMIC_ACQUIRE);
568 
569 	cur_tsc = rte_rdtsc();
570 	while (total) {
571 		for (i = 0; i < conf->nb_ports; i++) {
572 			portid = conf->portlist[i];
573 			nb_rx = rte_eth_rx_burst(portid, 0,
574 					&pkts_burst[next[portid]],
575 					RTE_MIN(MAX_PKT_BURST, num[portid]));
576 			if (unlikely(nb_rx == 0)) {
577 				timeout--;
578 				if (unlikely(timeout == 0))
579 					goto timeout;
580 				continue;
581 			}
582 			next[portid] += nb_rx;
583 			num[portid] -= nb_rx;
584 			total -= nb_rx;
585 		}
586 	}
587 timeout:
588 	diff_tsc = rte_rdtsc() - cur_tsc;
589 
590 	printf("%"PRIu64" packets lost, IDLE %"PRIu64" times\n",
591 	       total, MAX_IDLE - timeout);
592 	/* clean up */
593 	total = pkt_per_port * conf->nb_ports - total;
594 	for (i = 0; i < total; i++)
595 		rte_pktmbuf_free(pkts_burst[i]);
596 
597 	rte_free(pkts_burst);
598 
599 	if (total > 0)
600 		return diff_tsc / total;
601 	else
602 		return -1;
603 }
604 
605 static int
606 exec_burst(uint32_t flags, int lcore)
607 {
608 	unsigned int portid, nb_tx = 0;
609 	struct lcore_conf *conf;
610 	uint32_t pkt_per_port;
611 	int num, i, idx = 0;
612 	int diff_tsc;
613 
614 	conf = &lcore_conf[lcore];
615 
616 	pkt_per_port = MAX_TRAFFIC_BURST;
617 	num = pkt_per_port * conf->nb_ports;
618 
619 	/* only when polling first */
620 	if (flags == SC_BURST_POLL_FIRST)
621 		__atomic_store_n(&start, 1, __ATOMIC_RELAXED);
622 	else
623 		__atomic_store_n(&start, 0, __ATOMIC_RELAXED);
624 
625 	/* start polling thread
626 	 * if in POLL_FIRST mode, poll once launched;
627 	 * otherwise, not actually poll yet
628 	 */
629 	rte_eal_remote_launch(poll_burst,
630 			      (void *)&pkt_per_port, lcore);
631 
632 	/* start xmit */
633 	i = 0;
634 	while (num) {
635 		nb_tx = RTE_MIN(MAX_PKT_BURST, num);
636 		portid = conf->portlist[i];
637 		nb_tx = rte_eth_tx_burst(portid, 0, &tx_burst[idx], nb_tx);
638 		idx += nb_tx;
639 		num -= nb_tx;
640 		i = (i >= conf->nb_ports - 1) ? 0 : (i + 1);
641 	}
642 
643 	rte_delay_us(5 * US_PER_S);
644 
645 	/* only when polling second  */
646 	if (flags == SC_BURST_XMIT_FIRST)
647 		__atomic_store_n(&start, 1, __ATOMIC_RELEASE);
648 
649 	/* wait for polling finished */
650 	diff_tsc = rte_eal_wait_lcore(lcore);
651 	if (diff_tsc < 0) {
652 		printf("exec_burst: Failed to measure cycles per packet\n");
653 		return -1;
654 	}
655 
656 	printf("Result: %d cycles per packet\n", diff_tsc);
657 
658 	return 0;
659 }
660 
661 static int
662 test_pmd_perf(void)
663 {
664 	uint16_t nb_ports, num, nb_lcores, worker_id = (uint16_t)-1;
665 	uint16_t nb_rxd = MAX_TRAFFIC_BURST;
666 	uint16_t nb_txd = MAX_TRAFFIC_BURST;
667 	uint16_t portid;
668 	uint16_t nb_rx_queue = 1, nb_tx_queue = 1;
669 	int socketid = -1;
670 	int ret;
671 
672 	printf("Start PMD RXTX cycles cost test.\n");
673 
674 #ifndef RTE_EXEC_ENV_WINDOWS
675 	signal(SIGUSR1, signal_handler);
676 	signal(SIGUSR2, signal_handler);
677 #endif
678 
679 	nb_ports = rte_eth_dev_count_avail();
680 	if (nb_ports < NB_ETHPORTS_USED) {
681 		printf("At least %u port(s) used for perf. test\n",
682 		       NB_ETHPORTS_USED);
683 		return -1;
684 	}
685 
686 	nb_lcores = rte_lcore_count();
687 
688 	memset(lcore_conf, 0, sizeof(lcore_conf));
689 	init_lcores();
690 
691 	init_mbufpool(NB_MBUF);
692 
693 	if (sc_flag == SC_CONTINUOUS) {
694 		nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
695 		nb_txd = RTE_TEST_TX_DESC_DEFAULT;
696 	}
697 	printf("CONFIG RXD=%d TXD=%d\n", nb_rxd, nb_txd);
698 
699 	reset_count();
700 	num = 0;
701 	RTE_ETH_FOREACH_DEV(portid) {
702 		if (socketid == -1) {
703 			socketid = rte_eth_dev_socket_id(portid);
704 			worker_id = alloc_lcore(socketid);
705 			if (worker_id == (uint16_t)-1) {
706 				printf("No avail lcore to run test\n");
707 				return -1;
708 			}
709 			printf("Performance test runs on lcore %u socket %u\n",
710 			       worker_id, socketid);
711 		}
712 
713 		if (socketid != rte_eth_dev_socket_id(portid)) {
714 			printf("Skip port %d\n", portid);
715 			continue;
716 		}
717 
718 		/* port configure */
719 		ret = rte_eth_dev_configure(portid, nb_rx_queue,
720 					    nb_tx_queue, &port_conf);
721 		if (ret < 0)
722 			rte_exit(EXIT_FAILURE,
723 				"Cannot configure device: err=%d, port=%d\n",
724 				 ret, portid);
725 
726 		ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
727 		if (ret < 0)
728 			rte_exit(EXIT_FAILURE,
729 				"Cannot get mac address: err=%d, port=%d\n",
730 				 ret, portid);
731 
732 		printf("Port %u ", portid);
733 		print_ethaddr("Address:", &ports_eth_addr[portid]);
734 		printf("\n");
735 
736 		/* tx queue setup */
737 		ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
738 					     socketid, &tx_conf);
739 		if (ret < 0)
740 			rte_exit(EXIT_FAILURE,
741 				"rte_eth_tx_queue_setup: err=%d, "
742 				"port=%d\n", ret, portid);
743 
744 		/* rx queue steup */
745 		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
746 						socketid, &rx_conf,
747 						mbufpool[socketid]);
748 		if (ret < 0)
749 			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
750 				 "port=%d\n", ret, portid);
751 
752 		/* Start device */
753 		stop = 0;
754 		ret = rte_eth_dev_start(portid);
755 		if (ret < 0)
756 			rte_exit(EXIT_FAILURE,
757 				"rte_eth_dev_start: err=%d, port=%d\n",
758 				ret, portid);
759 
760 		/* always enable promiscuous */
761 		ret = rte_eth_promiscuous_enable(portid);
762 		if (ret != 0)
763 			rte_exit(EXIT_FAILURE,
764 				 "rte_eth_promiscuous_enable: err=%s, port=%d\n",
765 				 rte_strerror(-ret), portid);
766 
767 		lcore_conf[worker_id].portlist[num++] = portid;
768 		lcore_conf[worker_id].nb_ports++;
769 	}
770 	check_all_ports_link_status(nb_ports, RTE_PORT_ALL);
771 
772 	if (tx_burst == NULL) {
773 		tx_burst = (struct rte_mbuf **)
774 			rte_calloc_socket("tx_buff",
775 					  MAX_TRAFFIC_BURST * nb_ports,
776 					  sizeof(void *),
777 					  RTE_CACHE_LINE_SIZE, socketid);
778 		if (!tx_burst)
779 			return -1;
780 	}
781 
782 	init_traffic(mbufpool[socketid],
783 		     tx_burst, MAX_TRAFFIC_BURST * nb_ports);
784 
785 	printf("Generate %d packets @socket %d\n",
786 	       MAX_TRAFFIC_BURST * nb_ports, socketid);
787 
788 	if (sc_flag == SC_CONTINUOUS) {
789 		/* do both rxtx by default */
790 		if (NULL == do_measure)
791 			do_measure = measure_rxtx;
792 
793 		rte_eal_remote_launch(main_loop, NULL, worker_id);
794 
795 		if (rte_eal_wait_lcore(worker_id) < 0)
796 			return -1;
797 	} else if (sc_flag == SC_BURST_POLL_FIRST ||
798 		   sc_flag == SC_BURST_XMIT_FIRST)
799 		if (exec_burst(sc_flag, worker_id) < 0)
800 			return -1;
801 
802 	/* port tear down */
803 	RTE_ETH_FOREACH_DEV(portid) {
804 		if (socketid != rte_eth_dev_socket_id(portid))
805 			continue;
806 
807 		ret = rte_eth_dev_stop(portid);
808 		if (ret != 0)
809 			printf("rte_eth_dev_stop: err=%s, port=%u\n",
810 			       rte_strerror(-ret), portid);
811 	}
812 
813 	return 0;
814 }
815 
816 int
817 test_set_rxtx_conf(cmdline_fixed_string_t mode)
818 {
819 	printf("mode switch to %s\n", mode);
820 
821 	if (!strcmp(mode, "vector")) {
822 		/* vector rx, tx */
823 		tx_conf.tx_rs_thresh = 32;
824 		tx_conf.tx_free_thresh = 32;
825 		return 0;
826 	} else if (!strcmp(mode, "scalar")) {
827 		/* bulk alloc rx, full-featured tx */
828 		tx_conf.tx_rs_thresh = 32;
829 		tx_conf.tx_free_thresh = 32;
830 		port_conf.rxmode.offloads |= RTE_ETH_RX_OFFLOAD_CHECKSUM;
831 		return 0;
832 	} else if (!strcmp(mode, "hybrid")) {
833 		/* bulk alloc rx, vector tx
834 		 * when vec macro not define,
835 		 * using the same rx/tx as scalar
836 		 */
837 		tx_conf.tx_rs_thresh = 32;
838 		tx_conf.tx_free_thresh = 32;
839 		port_conf.rxmode.offloads |= RTE_ETH_RX_OFFLOAD_CHECKSUM;
840 		return 0;
841 	} else if (!strcmp(mode, "full")) {
842 		/* full feature rx,tx pair */
843 		tx_conf.tx_rs_thresh = 32;
844 		tx_conf.tx_free_thresh = 32;
845 		port_conf.rxmode.offloads |= RTE_ETH_RX_OFFLOAD_SCATTER;
846 		return 0;
847 	}
848 
849 	return -1;
850 }
851 
852 int
853 test_set_rxtx_anchor(cmdline_fixed_string_t type)
854 {
855 	printf("type switch to %s\n", type);
856 
857 	if (!strcmp(type, "rxtx")) {
858 		do_measure = measure_rxtx;
859 		return 0;
860 	} else if (!strcmp(type, "rxonly")) {
861 		do_measure = measure_rxonly;
862 		return 0;
863 	} else if (!strcmp(type, "txonly")) {
864 		do_measure = measure_txonly;
865 		return 0;
866 	}
867 
868 	return -1;
869 }
870 
871 int
872 test_set_rxtx_sc(cmdline_fixed_string_t type)
873 {
874 	printf("stream control switch to %s\n", type);
875 
876 	if (!strcmp(type, "continuous")) {
877 		sc_flag = SC_CONTINUOUS;
878 		return 0;
879 	} else if (!strcmp(type, "poll_before_xmit")) {
880 		sc_flag = SC_BURST_POLL_FIRST;
881 		return 0;
882 	} else if (!strcmp(type, "poll_after_xmit")) {
883 		sc_flag = SC_BURST_XMIT_FIRST;
884 		return 0;
885 	}
886 
887 	return -1;
888 }
889 
890 REGISTER_TEST_COMMAND(pmd_perf_autotest, test_pmd_perf);
891