xref: /dpdk/app/test/test_pmd_perf.c (revision f8dbaebbf1c9efcbb2e2354b341ed62175466a57)
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 static void
300 stats_display(uint16_t port_id)
301 {
302 	struct rte_eth_stats stats;
303 	rte_eth_stats_get(port_id, &stats);
304 
305 	printf("  RX-packets: %-10"PRIu64" RX-missed: %-10"PRIu64" RX-bytes:  "
306 	       "%-"PRIu64"\n",
307 	       stats.ipackets, stats.imissed, stats.ibytes);
308 	printf("  RX-errors: %-10"PRIu64" RX-nombuf:  %-10"PRIu64"\n",
309 	       stats.ierrors, stats.rx_nombuf);
310 	printf("  TX-packets: %-10"PRIu64" TX-errors: %-10"PRIu64" TX-bytes:  "
311 	       "%-"PRIu64"\n",
312 	       stats.opackets, stats.oerrors, stats.obytes);
313 }
314 
315 static void
316 signal_handler(int signum)
317 {
318 	/*  USR1 signal, stop testing */
319 	if (signum == SIGUSR1) {
320 		printf("Force Stop!\n");
321 		stop = 1;
322 	}
323 
324 	/*  USR2 signal, print stats */
325 	if (signum == SIGUSR2)
326 		stats_display(0);
327 }
328 
329 struct rte_mbuf **tx_burst;
330 
331 uint64_t (*do_measure)(struct lcore_conf *conf,
332 		       struct rte_mbuf *pkts_burst[],
333 		       uint64_t total_pkts);
334 
335 static uint64_t
336 measure_rxtx(struct lcore_conf *conf,
337 	     struct rte_mbuf *pkts_burst[],
338 	     uint64_t total_pkts)
339 {
340 	unsigned i, portid, nb_rx, nb_tx;
341 	uint64_t prev_tsc, cur_tsc;
342 
343 	prev_tsc = rte_rdtsc();
344 
345 	while (likely(!stop)) {
346 		for (i = 0; i < conf->nb_ports; i++) {
347 			portid = conf->portlist[i];
348 			nb_rx = rte_eth_rx_burst(portid, 0,
349 						 pkts_burst, MAX_PKT_BURST);
350 			if (unlikely(nb_rx == 0)) {
351 				idle++;
352 				continue;
353 			}
354 
355 			count += nb_rx;
356 			nb_tx = rte_eth_tx_burst(portid, 0, pkts_burst, nb_rx);
357 			if (unlikely(nb_tx < nb_rx)) {
358 				drop += (nb_rx - nb_tx);
359 				do {
360 					rte_pktmbuf_free(pkts_burst[nb_tx]);
361 				} while (++nb_tx < nb_rx);
362 			}
363 		}
364 		if (unlikely(count >= total_pkts))
365 			break;
366 	}
367 
368 	cur_tsc = rte_rdtsc();
369 
370 	return cur_tsc - prev_tsc;
371 }
372 
373 static uint64_t
374 measure_rxonly(struct lcore_conf *conf,
375 	       struct rte_mbuf *pkts_burst[],
376 	       uint64_t total_pkts)
377 {
378 	unsigned i, portid, nb_rx, nb_tx;
379 	uint64_t diff_tsc, cur_tsc;
380 
381 	diff_tsc = 0;
382 	while (likely(!stop)) {
383 		for (i = 0; i < conf->nb_ports; i++) {
384 			portid = conf->portlist[i];
385 
386 			cur_tsc = rte_rdtsc();
387 			nb_rx = rte_eth_rx_burst(portid, 0,
388 						 pkts_burst, MAX_PKT_BURST);
389 			if (unlikely(nb_rx == 0)) {
390 				idle++;
391 				continue;
392 			}
393 			diff_tsc += rte_rdtsc() - cur_tsc;
394 
395 			count += nb_rx;
396 			nb_tx = rte_eth_tx_burst(portid, 0, pkts_burst, nb_rx);
397 			if (unlikely(nb_tx < nb_rx)) {
398 				drop += (nb_rx - nb_tx);
399 				do {
400 					rte_pktmbuf_free(pkts_burst[nb_tx]);
401 				} while (++nb_tx < nb_rx);
402 			}
403 		}
404 		if (unlikely(count >= total_pkts))
405 			break;
406 	}
407 
408 	return diff_tsc;
409 }
410 
411 static uint64_t
412 measure_txonly(struct lcore_conf *conf,
413 	       struct rte_mbuf *pkts_burst[],
414 	       uint64_t total_pkts)
415 {
416 	unsigned i, portid, nb_rx, nb_tx;
417 	uint64_t diff_tsc, cur_tsc;
418 
419 	printf("do tx measure\n");
420 	diff_tsc = 0;
421 	while (likely(!stop)) {
422 		for (i = 0; i < conf->nb_ports; i++) {
423 			portid = conf->portlist[i];
424 			nb_rx = rte_eth_rx_burst(portid, 0,
425 						 pkts_burst, MAX_PKT_BURST);
426 			if (unlikely(nb_rx == 0)) {
427 				idle++;
428 				continue;
429 			}
430 
431 			count += nb_rx;
432 
433 			cur_tsc = rte_rdtsc();
434 			nb_tx = rte_eth_tx_burst(portid, 0, pkts_burst, nb_rx);
435 			if (unlikely(nb_tx < nb_rx)) {
436 				drop += (nb_rx - nb_tx);
437 				do {
438 					rte_pktmbuf_free(pkts_burst[nb_tx]);
439 				} while (++nb_tx < nb_rx);
440 			}
441 			diff_tsc += rte_rdtsc() - cur_tsc;
442 		}
443 		if (unlikely(count >= total_pkts))
444 			break;
445 	}
446 
447 	return diff_tsc;
448 }
449 
450 /* main processing loop */
451 static int
452 main_loop(__rte_unused void *args)
453 {
454 #define PACKET_SIZE 64
455 #define FRAME_GAP 12
456 #define MAC_PREAMBLE 8
457 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
458 	unsigned lcore_id;
459 	unsigned i, portid, nb_rx = 0, nb_tx = 0;
460 	struct lcore_conf *conf;
461 	int pkt_per_port;
462 	uint64_t diff_tsc;
463 	uint64_t packets_per_second, total_packets;
464 
465 	lcore_id = rte_lcore_id();
466 	conf = &lcore_conf[lcore_id];
467 	if (conf->status != LCORE_USED)
468 		return 0;
469 
470 	pkt_per_port = MAX_TRAFFIC_BURST;
471 
472 	int idx = 0;
473 	for (i = 0; i < conf->nb_ports; i++) {
474 		int num = pkt_per_port;
475 		portid = conf->portlist[i];
476 		printf("inject %d packet to port %d\n", num, portid);
477 		while (num) {
478 			nb_tx = RTE_MIN(MAX_PKT_BURST, num);
479 			nb_tx = rte_eth_tx_burst(portid, 0,
480 						&tx_burst[idx], nb_tx);
481 			num -= nb_tx;
482 			idx += nb_tx;
483 		}
484 	}
485 	printf("Total packets inject to prime ports = %u\n", idx);
486 
487 	packets_per_second = (link_mbps * 1000 * 1000) /
488 		((PACKET_SIZE + FRAME_GAP + MAC_PREAMBLE) * CHAR_BIT);
489 	printf("Each port will do %"PRIu64" packets per second\n",
490 	       packets_per_second);
491 
492 	total_packets = RTE_TEST_DURATION * conf->nb_ports * packets_per_second;
493 	printf("Test will stop after at least %"PRIu64" packets received\n",
494 		+ total_packets);
495 
496 	diff_tsc = do_measure(conf, pkts_burst, total_packets);
497 
498 	for (i = 0; i < conf->nb_ports; i++) {
499 		portid = conf->portlist[i];
500 		int nb_free = 0;
501 		uint64_t timeout = 10000;
502 		do { /* dry out */
503 			nb_rx = rte_eth_rx_burst(portid, 0,
504 						 pkts_burst, MAX_PKT_BURST);
505 			nb_tx = 0;
506 			while (nb_tx < nb_rx)
507 				rte_pktmbuf_free(pkts_burst[nb_tx++]);
508 			nb_free += nb_rx;
509 
510 			if (unlikely(nb_rx == 0))
511 				timeout--;
512 		} while (nb_free != pkt_per_port && timeout != 0);
513 		printf("free %d (expected %d) mbuf left in port %u\n", nb_free,
514 		       pkt_per_port, portid);
515 	}
516 
517 	if (count == 0)
518 		return -1;
519 
520 	printf("%"PRIu64" packet, %"PRIu64" drop, %"PRIu64" idle\n",
521 	       count, drop, idle);
522 	printf("Result: %"PRIu64" cycles per packet\n", diff_tsc / count);
523 
524 	return 0;
525 }
526 
527 static uint64_t start;
528 
529 static inline int
530 poll_burst(void *args)
531 {
532 #define MAX_IDLE           (10000)
533 	unsigned lcore_id;
534 	struct rte_mbuf **pkts_burst;
535 	uint64_t diff_tsc, cur_tsc;
536 	uint16_t next[RTE_MAX_ETHPORTS];
537 	struct lcore_conf *conf;
538 	uint32_t pkt_per_port = *((uint32_t *)args);
539 	unsigned i, portid, nb_rx = 0;
540 	uint64_t total;
541 	uint64_t timeout = MAX_IDLE;
542 	int num[RTE_MAX_ETHPORTS];
543 
544 	lcore_id = rte_lcore_id();
545 	conf = &lcore_conf[lcore_id];
546 	if (conf->status != LCORE_USED)
547 		return 0;
548 
549 	total = pkt_per_port * conf->nb_ports;
550 	printf("start to receive total expect %"PRIu64"\n", total);
551 
552 	pkts_burst = (struct rte_mbuf **)
553 		rte_calloc_socket("poll_burst",
554 				  total, sizeof(void *),
555 				  RTE_CACHE_LINE_SIZE, conf->socketid);
556 	if (!pkts_burst)
557 		return -1;
558 
559 	for (i = 0; i < conf->nb_ports; i++) {
560 		portid = conf->portlist[i];
561 		next[portid] = i * pkt_per_port;
562 		num[portid] = pkt_per_port;
563 	}
564 
565 	rte_wait_until_equal_64(&start, 1, __ATOMIC_ACQUIRE);
566 
567 	cur_tsc = rte_rdtsc();
568 	while (total) {
569 		for (i = 0; i < conf->nb_ports; i++) {
570 			portid = conf->portlist[i];
571 			nb_rx = rte_eth_rx_burst(portid, 0,
572 					&pkts_burst[next[portid]],
573 					RTE_MIN(MAX_PKT_BURST, num[portid]));
574 			if (unlikely(nb_rx == 0)) {
575 				timeout--;
576 				if (unlikely(timeout == 0))
577 					goto timeout;
578 				continue;
579 			}
580 			next[portid] += nb_rx;
581 			num[portid] -= nb_rx;
582 			total -= nb_rx;
583 		}
584 	}
585 timeout:
586 	diff_tsc = rte_rdtsc() - cur_tsc;
587 
588 	printf("%"PRIu64" packets lost, IDLE %"PRIu64" times\n",
589 	       total, MAX_IDLE - timeout);
590 	/* clean up */
591 	total = pkt_per_port * conf->nb_ports - total;
592 	for (i = 0; i < total; i++)
593 		rte_pktmbuf_free(pkts_burst[i]);
594 
595 	rte_free(pkts_burst);
596 
597 	if (total > 0)
598 		return diff_tsc / total;
599 	else
600 		return -1;
601 }
602 
603 static int
604 exec_burst(uint32_t flags, int lcore)
605 {
606 	unsigned int portid, nb_tx = 0;
607 	struct lcore_conf *conf;
608 	uint32_t pkt_per_port;
609 	int num, i, idx = 0;
610 	int diff_tsc;
611 
612 	conf = &lcore_conf[lcore];
613 
614 	pkt_per_port = MAX_TRAFFIC_BURST;
615 	num = pkt_per_port * conf->nb_ports;
616 
617 	/* only when polling first */
618 	if (flags == SC_BURST_POLL_FIRST)
619 		__atomic_store_n(&start, 1, __ATOMIC_RELAXED);
620 	else
621 		__atomic_store_n(&start, 0, __ATOMIC_RELAXED);
622 
623 	/* start polling thread
624 	 * if in POLL_FIRST mode, poll once launched;
625 	 * otherwise, not actually poll yet
626 	 */
627 	rte_eal_remote_launch(poll_burst,
628 			      (void *)&pkt_per_port, lcore);
629 
630 	/* start xmit */
631 	i = 0;
632 	while (num) {
633 		nb_tx = RTE_MIN(MAX_PKT_BURST, num);
634 		portid = conf->portlist[i];
635 		nb_tx = rte_eth_tx_burst(portid, 0, &tx_burst[idx], nb_tx);
636 		idx += nb_tx;
637 		num -= nb_tx;
638 		i = (i >= conf->nb_ports - 1) ? 0 : (i + 1);
639 	}
640 
641 	sleep(5);
642 
643 	/* only when polling second  */
644 	if (flags == SC_BURST_XMIT_FIRST)
645 		__atomic_store_n(&start, 1, __ATOMIC_RELEASE);
646 
647 	/* wait for polling finished */
648 	diff_tsc = rte_eal_wait_lcore(lcore);
649 	if (diff_tsc < 0) {
650 		printf("exec_burst: Failed to measure cycles per packet\n");
651 		return -1;
652 	}
653 
654 	printf("Result: %d cycles per packet\n", diff_tsc);
655 
656 	return 0;
657 }
658 
659 static int
660 test_pmd_perf(void)
661 {
662 	uint16_t nb_ports, num, nb_lcores, worker_id = (uint16_t)-1;
663 	uint16_t nb_rxd = MAX_TRAFFIC_BURST;
664 	uint16_t nb_txd = MAX_TRAFFIC_BURST;
665 	uint16_t portid;
666 	uint16_t nb_rx_queue = 1, nb_tx_queue = 1;
667 	int socketid = -1;
668 	int ret;
669 
670 	printf("Start PMD RXTX cycles cost test.\n");
671 
672 	signal(SIGUSR1, signal_handler);
673 	signal(SIGUSR2, signal_handler);
674 
675 	nb_ports = rte_eth_dev_count_avail();
676 	if (nb_ports < NB_ETHPORTS_USED) {
677 		printf("At least %u port(s) used for perf. test\n",
678 		       NB_ETHPORTS_USED);
679 		return -1;
680 	}
681 
682 	nb_lcores = rte_lcore_count();
683 
684 	memset(lcore_conf, 0, sizeof(lcore_conf));
685 	init_lcores();
686 
687 	init_mbufpool(NB_MBUF);
688 
689 	if (sc_flag == SC_CONTINUOUS) {
690 		nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
691 		nb_txd = RTE_TEST_TX_DESC_DEFAULT;
692 	}
693 	printf("CONFIG RXD=%d TXD=%d\n", nb_rxd, nb_txd);
694 
695 	reset_count();
696 	num = 0;
697 	RTE_ETH_FOREACH_DEV(portid) {
698 		if (socketid == -1) {
699 			socketid = rte_eth_dev_socket_id(portid);
700 			worker_id = alloc_lcore(socketid);
701 			if (worker_id == (uint16_t)-1) {
702 				printf("No avail lcore to run test\n");
703 				return -1;
704 			}
705 			printf("Performance test runs on lcore %u socket %u\n",
706 			       worker_id, socketid);
707 		}
708 
709 		if (socketid != rte_eth_dev_socket_id(portid)) {
710 			printf("Skip port %d\n", portid);
711 			continue;
712 		}
713 
714 		/* port configure */
715 		ret = rte_eth_dev_configure(portid, nb_rx_queue,
716 					    nb_tx_queue, &port_conf);
717 		if (ret < 0)
718 			rte_exit(EXIT_FAILURE,
719 				"Cannot configure device: err=%d, port=%d\n",
720 				 ret, portid);
721 
722 		ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
723 		if (ret < 0)
724 			rte_exit(EXIT_FAILURE,
725 				"Cannot get mac address: err=%d, port=%d\n",
726 				 ret, portid);
727 
728 		printf("Port %u ", portid);
729 		print_ethaddr("Address:", &ports_eth_addr[portid]);
730 		printf("\n");
731 
732 		/* tx queue setup */
733 		ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
734 					     socketid, &tx_conf);
735 		if (ret < 0)
736 			rte_exit(EXIT_FAILURE,
737 				"rte_eth_tx_queue_setup: err=%d, "
738 				"port=%d\n", ret, portid);
739 
740 		/* rx queue steup */
741 		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
742 						socketid, &rx_conf,
743 						mbufpool[socketid]);
744 		if (ret < 0)
745 			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
746 				 "port=%d\n", ret, portid);
747 
748 		/* Start device */
749 		stop = 0;
750 		ret = rte_eth_dev_start(portid);
751 		if (ret < 0)
752 			rte_exit(EXIT_FAILURE,
753 				"rte_eth_dev_start: err=%d, port=%d\n",
754 				ret, portid);
755 
756 		/* always enable promiscuous */
757 		ret = rte_eth_promiscuous_enable(portid);
758 		if (ret != 0)
759 			rte_exit(EXIT_FAILURE,
760 				 "rte_eth_promiscuous_enable: err=%s, port=%d\n",
761 				 rte_strerror(-ret), portid);
762 
763 		lcore_conf[worker_id].portlist[num++] = portid;
764 		lcore_conf[worker_id].nb_ports++;
765 	}
766 	check_all_ports_link_status(nb_ports, RTE_PORT_ALL);
767 
768 	if (tx_burst == NULL) {
769 		tx_burst = (struct rte_mbuf **)
770 			rte_calloc_socket("tx_buff",
771 					  MAX_TRAFFIC_BURST * nb_ports,
772 					  sizeof(void *),
773 					  RTE_CACHE_LINE_SIZE, socketid);
774 		if (!tx_burst)
775 			return -1;
776 	}
777 
778 	init_traffic(mbufpool[socketid],
779 		     tx_burst, MAX_TRAFFIC_BURST * nb_ports);
780 
781 	printf("Generate %d packets @socket %d\n",
782 	       MAX_TRAFFIC_BURST * nb_ports, socketid);
783 
784 	if (sc_flag == SC_CONTINUOUS) {
785 		/* do both rxtx by default */
786 		if (NULL == do_measure)
787 			do_measure = measure_rxtx;
788 
789 		rte_eal_remote_launch(main_loop, NULL, worker_id);
790 
791 		if (rte_eal_wait_lcore(worker_id) < 0)
792 			return -1;
793 	} else if (sc_flag == SC_BURST_POLL_FIRST ||
794 		   sc_flag == SC_BURST_XMIT_FIRST)
795 		if (exec_burst(sc_flag, worker_id) < 0)
796 			return -1;
797 
798 	/* port tear down */
799 	RTE_ETH_FOREACH_DEV(portid) {
800 		if (socketid != rte_eth_dev_socket_id(portid))
801 			continue;
802 
803 		ret = rte_eth_dev_stop(portid);
804 		if (ret != 0)
805 			printf("rte_eth_dev_stop: err=%s, port=%u\n",
806 			       rte_strerror(-ret), portid);
807 	}
808 
809 	return 0;
810 }
811 
812 int
813 test_set_rxtx_conf(cmdline_fixed_string_t mode)
814 {
815 	printf("mode switch to %s\n", mode);
816 
817 	if (!strcmp(mode, "vector")) {
818 		/* vector rx, tx */
819 		tx_conf.tx_rs_thresh = 32;
820 		tx_conf.tx_free_thresh = 32;
821 		return 0;
822 	} else if (!strcmp(mode, "scalar")) {
823 		/* bulk alloc rx, full-featured tx */
824 		tx_conf.tx_rs_thresh = 32;
825 		tx_conf.tx_free_thresh = 32;
826 		port_conf.rxmode.offloads |= RTE_ETH_RX_OFFLOAD_CHECKSUM;
827 		return 0;
828 	} else if (!strcmp(mode, "hybrid")) {
829 		/* bulk alloc rx, vector tx
830 		 * when vec macro not define,
831 		 * using the same rx/tx as scalar
832 		 */
833 		tx_conf.tx_rs_thresh = 32;
834 		tx_conf.tx_free_thresh = 32;
835 		port_conf.rxmode.offloads |= RTE_ETH_RX_OFFLOAD_CHECKSUM;
836 		return 0;
837 	} else if (!strcmp(mode, "full")) {
838 		/* full feature rx,tx pair */
839 		tx_conf.tx_rs_thresh = 32;
840 		tx_conf.tx_free_thresh = 32;
841 		port_conf.rxmode.offloads |= RTE_ETH_RX_OFFLOAD_SCATTER;
842 		return 0;
843 	}
844 
845 	return -1;
846 }
847 
848 int
849 test_set_rxtx_anchor(cmdline_fixed_string_t type)
850 {
851 	printf("type switch to %s\n", type);
852 
853 	if (!strcmp(type, "rxtx")) {
854 		do_measure = measure_rxtx;
855 		return 0;
856 	} else if (!strcmp(type, "rxonly")) {
857 		do_measure = measure_rxonly;
858 		return 0;
859 	} else if (!strcmp(type, "txonly")) {
860 		do_measure = measure_txonly;
861 		return 0;
862 	}
863 
864 	return -1;
865 }
866 
867 int
868 test_set_rxtx_sc(cmdline_fixed_string_t type)
869 {
870 	printf("stream control switch to %s\n", type);
871 
872 	if (!strcmp(type, "continuous")) {
873 		sc_flag = SC_CONTINUOUS;
874 		return 0;
875 	} else if (!strcmp(type, "poll_before_xmit")) {
876 		sc_flag = SC_BURST_POLL_FIRST;
877 		return 0;
878 	} else if (!strcmp(type, "poll_after_xmit")) {
879 		sc_flag = SC_BURST_XMIT_FIRST;
880 		return 0;
881 	}
882 
883 	return -1;
884 }
885 
886 REGISTER_TEST_COMMAND(pmd_perf_autotest, test_pmd_perf);
887