xref: /dpdk/app/test/test_pmd_perf.c (revision b6a7e6852e9ab82ae0e05e2d2a0b83abca17de3b)
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 RX_DESC_DEFAULT        (1024)
22 #define 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 	},
66 	.txmode = {
67 		.mq_mode = RTE_ETH_MQ_TX_NONE,
68 	},
69 	.lpbk_mode = 1,  /* enable loopback */
70 };
71 
72 static struct rte_eth_rxconf rx_conf = {
73 	.rx_thresh = {
74 		.pthresh = RX_PTHRESH,
75 		.hthresh = RX_HTHRESH,
76 		.wthresh = RX_WTHRESH,
77 	},
78 	.rx_free_thresh = 32,
79 };
80 
81 static struct rte_eth_txconf tx_conf = {
82 	.tx_thresh = {
83 		.pthresh = TX_PTHRESH,
84 		.hthresh = TX_HTHRESH,
85 		.wthresh = TX_WTHRESH,
86 	},
87 	.tx_free_thresh = 32, /* Use PMD default values */
88 	.tx_rs_thresh = 32, /* Use PMD default values */
89 };
90 
91 enum {
92 	LCORE_INVALID = 0,
93 	LCORE_AVAIL,
94 	LCORE_USED,
95 };
96 
97 struct __rte_cache_aligned lcore_conf {
98 	uint8_t status;
99 	uint8_t socketid;
100 	uint16_t nb_ports;
101 	uint16_t portlist[RTE_MAX_ETHPORTS];
102 };
103 
104 struct lcore_conf lcore_conf[RTE_MAX_LCORE];
105 
106 static uint64_t link_mbps;
107 
108 enum {
109 	SC_CONTINUOUS = 0,
110 	SC_BURST_POLL_FIRST,
111 	SC_BURST_XMIT_FIRST,
112 };
113 
114 static uint32_t sc_flag;
115 
116 /* Check the link status of all ports in up to 3s, and print them finally */
117 static void
check_all_ports_link_status(uint16_t port_num,uint32_t port_mask)118 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask)
119 {
120 #define CHECK_INTERVAL 100 /* 100ms */
121 #define MAX_CHECK_TIME 30 /* 3s (30 * 100ms) in total */
122 	uint16_t portid;
123 	uint8_t count, all_ports_up, print_flag = 0;
124 	struct rte_eth_link link;
125 	int ret;
126 	char link_status[RTE_ETH_LINK_MAX_STR_LEN];
127 
128 	printf("Checking link statuses...\n");
129 	fflush(stdout);
130 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
131 		all_ports_up = 1;
132 		for (portid = 0; portid < port_num; portid++) {
133 			if ((port_mask & (1 << portid)) == 0)
134 				continue;
135 			memset(&link, 0, sizeof(link));
136 			ret = rte_eth_link_get_nowait(portid, &link);
137 			if (ret < 0) {
138 				all_ports_up = 0;
139 				if (print_flag == 1)
140 					printf("Port %u link get failed: %s\n",
141 						portid, rte_strerror(-ret));
142 				continue;
143 			}
144 
145 			/* print link status if flag set */
146 			if (print_flag == 1) {
147 				if (link.link_status && link_mbps == 0)
148 					link_mbps = link.link_speed;
149 
150 				rte_eth_link_to_str(link_status,
151 					sizeof(link_status), &link);
152 				printf("Port %d %s\n", portid, link_status);
153 				continue;
154 			}
155 			/* clear all_ports_up flag if any link down */
156 			if (link.link_status == RTE_ETH_LINK_DOWN) {
157 				all_ports_up = 0;
158 				break;
159 			}
160 		}
161 		/* after finally printing all link status, get out */
162 		if (print_flag == 1)
163 			break;
164 
165 		if (all_ports_up == 0) {
166 			fflush(stdout);
167 			rte_delay_ms(CHECK_INTERVAL);
168 		}
169 
170 		/* set the print_flag if all ports up or timeout */
171 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1))
172 			print_flag = 1;
173 	}
174 }
175 
176 static void
print_ethaddr(const char * name,const struct rte_ether_addr * eth_addr)177 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
178 {
179 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
180 	rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
181 	printf("%s%s", name, buf);
182 }
183 
184 static int
init_traffic(struct rte_mempool * mp,struct rte_mbuf ** pkts_burst,uint32_t burst_size)185 init_traffic(struct rte_mempool *mp,
186 	     struct rte_mbuf **pkts_burst, uint32_t burst_size)
187 {
188 	struct rte_ether_hdr pkt_eth_hdr;
189 	struct rte_ipv4_hdr pkt_ipv4_hdr;
190 	struct rte_udp_hdr pkt_udp_hdr;
191 	uint32_t pktlen;
192 	static uint8_t src_mac[] = { 0x00, 0xFF, 0xAA, 0xFF, 0xAA, 0xFF };
193 	static uint8_t dst_mac[] = { 0x00, 0xAA, 0xFF, 0xAA, 0xFF, 0xAA };
194 
195 
196 	initialize_eth_header(&pkt_eth_hdr,
197 		(struct rte_ether_addr *)src_mac,
198 		(struct rte_ether_addr *)dst_mac, RTE_ETHER_TYPE_IPV4, 0, 0);
199 
200 	pktlen = initialize_ipv4_header(&pkt_ipv4_hdr,
201 					IPV4_ADDR(10, 0, 0, 1),
202 					IPV4_ADDR(10, 0, 0, 2), 26);
203 	printf("IPv4 pktlen %u\n", pktlen);
204 
205 	pktlen = initialize_udp_header(&pkt_udp_hdr, 0, 0, 18);
206 
207 	printf("UDP pktlen %u\n", pktlen);
208 
209 	return generate_packet_burst(mp, pkts_burst, &pkt_eth_hdr,
210 				     0, &pkt_ipv4_hdr, 1,
211 				     &pkt_udp_hdr, burst_size,
212 				     PACKET_BURST_GEN_PKT_LEN, 1);
213 }
214 
215 static int
init_lcores(void)216 init_lcores(void)
217 {
218 	unsigned lcore_id;
219 
220 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
221 		lcore_conf[lcore_id].socketid =
222 			rte_lcore_to_socket_id(lcore_id);
223 		if (rte_lcore_is_enabled(lcore_id) == 0) {
224 			lcore_conf[lcore_id].status = LCORE_INVALID;
225 			continue;
226 		} else
227 			lcore_conf[lcore_id].status = LCORE_AVAIL;
228 	}
229 	return 0;
230 }
231 
232 static int
init_mbufpool(unsigned nb_mbuf)233 init_mbufpool(unsigned nb_mbuf)
234 {
235 	int socketid;
236 	unsigned lcore_id;
237 	char s[64];
238 
239 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
240 		if (rte_lcore_is_enabled(lcore_id) == 0)
241 			continue;
242 
243 		socketid = rte_lcore_to_socket_id(lcore_id);
244 		if (socketid >= NB_SOCKETS) {
245 			rte_exit(EXIT_FAILURE,
246 				"Socket %d of lcore %u is out of range %d\n",
247 				socketid, lcore_id, NB_SOCKETS);
248 		}
249 		if (mbufpool[socketid] == NULL) {
250 			snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
251 			mbufpool[socketid] =
252 				rte_pktmbuf_pool_create(s, nb_mbuf,
253 					MEMPOOL_CACHE_SIZE, 0,
254 					RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
255 			if (mbufpool[socketid] == NULL)
256 				rte_exit(EXIT_FAILURE,
257 					"Cannot init mbuf pool on socket %d\n",
258 					socketid);
259 			else
260 				printf("Allocated mbuf pool on socket %d\n",
261 					socketid);
262 		}
263 	}
264 	return 0;
265 }
266 
267 static uint16_t
alloc_lcore(int socketid)268 alloc_lcore(int socketid)
269 {
270 	unsigned lcore_id;
271 
272 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
273 		if (LCORE_AVAIL != lcore_conf[lcore_id].status ||
274 		    (socketid != SOCKET_ID_ANY &&
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
reset_count(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
stats_display(uint16_t port_id)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
signal_handler(int signum)317 signal_handler(int signum)
318 {
319 	/*  USR1 signal, stop testing */
320 	if (signum == SIGUSR1) {
321 		stop = 1;
322 	}
323 
324 	/*  USR2 signal, print stats */
325 	if (signum == SIGUSR2)
326 		stats_display(0);
327 }
328 #endif
329 
330 struct rte_mbuf **tx_burst;
331 
332 uint64_t (*do_measure)(struct lcore_conf *conf,
333 		       struct rte_mbuf *pkts_burst[],
334 		       uint64_t total_pkts);
335 
336 static uint64_t
measure_rxtx(struct lcore_conf * conf,struct rte_mbuf * pkts_burst[],uint64_t total_pkts)337 measure_rxtx(struct lcore_conf *conf,
338 	     struct rte_mbuf *pkts_burst[],
339 	     uint64_t total_pkts)
340 {
341 	unsigned i, portid, nb_rx, nb_tx;
342 	uint64_t prev_tsc, cur_tsc;
343 
344 	prev_tsc = rte_rdtsc();
345 
346 	while (likely(!stop)) {
347 		for (i = 0; i < conf->nb_ports; i++) {
348 			portid = conf->portlist[i];
349 			nb_rx = rte_eth_rx_burst(portid, 0,
350 						 pkts_burst, MAX_PKT_BURST);
351 			if (unlikely(nb_rx == 0)) {
352 				idle++;
353 				continue;
354 			}
355 
356 			count += nb_rx;
357 			nb_tx = rte_eth_tx_burst(portid, 0, pkts_burst, nb_rx);
358 			if (unlikely(nb_tx < nb_rx)) {
359 				drop += (nb_rx - nb_tx);
360 				do {
361 					rte_pktmbuf_free(pkts_burst[nb_tx]);
362 				} while (++nb_tx < nb_rx);
363 			}
364 		}
365 		if (unlikely(count >= total_pkts))
366 			break;
367 	}
368 
369 	cur_tsc = rte_rdtsc();
370 
371 	return cur_tsc - prev_tsc;
372 }
373 
374 static uint64_t
measure_rxonly(struct lcore_conf * conf,struct rte_mbuf * pkts_burst[],uint64_t total_pkts)375 measure_rxonly(struct lcore_conf *conf,
376 	       struct rte_mbuf *pkts_burst[],
377 	       uint64_t total_pkts)
378 {
379 	unsigned i, portid, nb_rx, nb_tx;
380 	uint64_t diff_tsc, cur_tsc;
381 
382 	diff_tsc = 0;
383 	while (likely(!stop)) {
384 		for (i = 0; i < conf->nb_ports; i++) {
385 			portid = conf->portlist[i];
386 
387 			cur_tsc = rte_rdtsc();
388 			nb_rx = rte_eth_rx_burst(portid, 0,
389 						 pkts_burst, MAX_PKT_BURST);
390 			if (unlikely(nb_rx == 0)) {
391 				idle++;
392 				continue;
393 			}
394 			diff_tsc += rte_rdtsc() - cur_tsc;
395 
396 			count += nb_rx;
397 			nb_tx = rte_eth_tx_burst(portid, 0, pkts_burst, nb_rx);
398 			if (unlikely(nb_tx < nb_rx)) {
399 				drop += (nb_rx - nb_tx);
400 				do {
401 					rte_pktmbuf_free(pkts_burst[nb_tx]);
402 				} while (++nb_tx < nb_rx);
403 			}
404 		}
405 		if (unlikely(count >= total_pkts))
406 			break;
407 	}
408 
409 	return diff_tsc;
410 }
411 
412 static uint64_t
measure_txonly(struct lcore_conf * conf,struct rte_mbuf * pkts_burst[],uint64_t total_pkts)413 measure_txonly(struct lcore_conf *conf,
414 	       struct rte_mbuf *pkts_burst[],
415 	       uint64_t total_pkts)
416 {
417 	unsigned i, portid, nb_rx, nb_tx;
418 	uint64_t diff_tsc, cur_tsc;
419 
420 	printf("do tx measure\n");
421 	diff_tsc = 0;
422 	while (likely(!stop)) {
423 		for (i = 0; i < conf->nb_ports; i++) {
424 			portid = conf->portlist[i];
425 			nb_rx = rte_eth_rx_burst(portid, 0,
426 						 pkts_burst, MAX_PKT_BURST);
427 			if (unlikely(nb_rx == 0)) {
428 				idle++;
429 				continue;
430 			}
431 
432 			count += nb_rx;
433 
434 			cur_tsc = rte_rdtsc();
435 			nb_tx = rte_eth_tx_burst(portid, 0, pkts_burst, nb_rx);
436 			if (unlikely(nb_tx < nb_rx)) {
437 				drop += (nb_rx - nb_tx);
438 				do {
439 					rte_pktmbuf_free(pkts_burst[nb_tx]);
440 				} while (++nb_tx < nb_rx);
441 			}
442 			diff_tsc += rte_rdtsc() - cur_tsc;
443 		}
444 		if (unlikely(count >= total_pkts))
445 			break;
446 	}
447 
448 	return diff_tsc;
449 }
450 
451 /* main processing loop */
452 static int
main_loop(__rte_unused void * args)453 main_loop(__rte_unused void *args)
454 {
455 #define PACKET_SIZE 64
456 #define FRAME_GAP 12
457 #define MAC_PREAMBLE 8
458 #define MAX_RETRY_COUNT 5
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 	int retry_cnt = 0;
467 	int free_pkt = 0;
468 
469 	lcore_id = rte_lcore_id();
470 	conf = &lcore_conf[lcore_id];
471 	if (conf->status != LCORE_USED)
472 		return 0;
473 
474 	pkt_per_port = MAX_TRAFFIC_BURST;
475 
476 	int idx = 0;
477 	for (i = 0; i < conf->nb_ports; i++) {
478 		int num = pkt_per_port;
479 		portid = conf->portlist[i];
480 		printf("inject %d packet to port %d\n", num, portid);
481 		while (num) {
482 			nb_tx = RTE_MIN(MAX_PKT_BURST, num);
483 			nb_tx = rte_eth_tx_burst(portid, 0,
484 						&tx_burst[idx], nb_tx);
485 			if (nb_tx == 0)
486 				retry_cnt++;
487 			num -= nb_tx;
488 			idx += nb_tx;
489 			if (retry_cnt == MAX_RETRY_COUNT) {
490 				retry_cnt = 0;
491 				break;
492 			}
493 		}
494 	}
495 	for (free_pkt = idx; free_pkt < (MAX_TRAFFIC_BURST * conf->nb_ports);
496 			free_pkt++)
497 		rte_pktmbuf_free(tx_burst[free_pkt]);
498 	printf("Total packets inject to prime ports = %u\n", idx);
499 
500 	packets_per_second = (link_mbps * 1000 * 1000) /
501 		((PACKET_SIZE + FRAME_GAP + MAC_PREAMBLE) * CHAR_BIT);
502 	printf("Each port will do %"PRIu64" packets per second\n",
503 	       packets_per_second);
504 
505 	total_packets = RTE_TEST_DURATION * conf->nb_ports * packets_per_second;
506 	printf("Test will stop after at least %"PRIu64" packets received\n",
507 		+ total_packets);
508 
509 	diff_tsc = do_measure(conf, pkts_burst, total_packets);
510 
511 	for (i = 0; i < conf->nb_ports; i++) {
512 		portid = conf->portlist[i];
513 		int nb_free = 0;
514 		uint64_t timeout = 10000;
515 		do { /* dry out */
516 			nb_rx = rte_eth_rx_burst(portid, 0,
517 						 pkts_burst, MAX_PKT_BURST);
518 			nb_tx = 0;
519 			while (nb_tx < nb_rx)
520 				rte_pktmbuf_free(pkts_burst[nb_tx++]);
521 			nb_free += nb_rx;
522 
523 			if (unlikely(nb_rx == 0))
524 				timeout--;
525 		} while (nb_free != pkt_per_port && timeout != 0);
526 		printf("free %d (expected %d) mbuf left in port %u\n", nb_free,
527 		       pkt_per_port, portid);
528 	}
529 
530 	if (count == 0)
531 		return -1;
532 
533 	printf("%"PRIu64" packet, %"PRIu64" drop, %"PRIu64" idle\n",
534 	       count, drop, idle);
535 	printf("Result: %"PRIu64" cycles per packet\n", diff_tsc / count);
536 
537 	return 0;
538 }
539 
540 static RTE_ATOMIC(uint64_t) start;
541 
542 static inline int
poll_burst(void * args)543 poll_burst(void *args)
544 {
545 #define MAX_IDLE           (10000)
546 	unsigned lcore_id;
547 	struct rte_mbuf **pkts_burst;
548 	uint64_t diff_tsc, cur_tsc;
549 	uint16_t next[RTE_MAX_ETHPORTS];
550 	struct lcore_conf *conf;
551 	uint32_t pkt_per_port = *((uint32_t *)args);
552 	unsigned i, portid, nb_rx = 0;
553 	uint64_t total;
554 	uint64_t timeout = MAX_IDLE;
555 	int num[RTE_MAX_ETHPORTS];
556 
557 	lcore_id = rte_lcore_id();
558 	conf = &lcore_conf[lcore_id];
559 	if (conf->status != LCORE_USED)
560 		return 0;
561 
562 	total = pkt_per_port * conf->nb_ports;
563 	printf("start to receive total expect %"PRIu64"\n", total);
564 
565 	pkts_burst = (struct rte_mbuf **)
566 		rte_calloc_socket("poll_burst",
567 				  total, sizeof(void *),
568 				  RTE_CACHE_LINE_SIZE, conf->socketid);
569 	if (!pkts_burst)
570 		return -1;
571 
572 	for (i = 0; i < conf->nb_ports; i++) {
573 		portid = conf->portlist[i];
574 		next[portid] = i * pkt_per_port;
575 		num[portid] = pkt_per_port;
576 	}
577 
578 	rte_wait_until_equal_64((uint64_t *)(uintptr_t)&start, 1, rte_memory_order_acquire);
579 
580 	cur_tsc = rte_rdtsc();
581 	while (total) {
582 		for (i = 0; i < conf->nb_ports; i++) {
583 			portid = conf->portlist[i];
584 			nb_rx = rte_eth_rx_burst(portid, 0,
585 					&pkts_burst[next[portid]],
586 					RTE_MIN(MAX_PKT_BURST, num[portid]));
587 			if (unlikely(nb_rx == 0)) {
588 				timeout--;
589 				if (unlikely(timeout == 0))
590 					goto timeout;
591 				continue;
592 			}
593 			next[portid] += nb_rx;
594 			num[portid] -= nb_rx;
595 			total -= nb_rx;
596 		}
597 	}
598 timeout:
599 	diff_tsc = rte_rdtsc() - cur_tsc;
600 
601 	printf("%"PRIu64" packets lost, IDLE %"PRIu64" times\n",
602 	       total, MAX_IDLE - timeout);
603 	/* clean up */
604 	total = pkt_per_port * conf->nb_ports - total;
605 	for (i = 0; i < total; i++)
606 		rte_pktmbuf_free(pkts_burst[i]);
607 
608 	rte_free(pkts_burst);
609 
610 	if (total > 0)
611 		return diff_tsc / total;
612 	else
613 		return -1;
614 }
615 
616 static int
exec_burst(uint32_t flags,int lcore)617 exec_burst(uint32_t flags, int lcore)
618 {
619 	unsigned int portid, nb_tx = 0;
620 	struct lcore_conf *conf;
621 	uint32_t pkt_per_port;
622 	int num, i, idx = 0;
623 	int diff_tsc;
624 
625 	conf = &lcore_conf[lcore];
626 
627 	pkt_per_port = MAX_TRAFFIC_BURST;
628 	num = pkt_per_port * conf->nb_ports;
629 
630 	/* only when polling first */
631 	if (flags == SC_BURST_POLL_FIRST)
632 		rte_atomic_store_explicit(&start, 1, rte_memory_order_relaxed);
633 	else
634 		rte_atomic_store_explicit(&start, 0, rte_memory_order_relaxed);
635 
636 	/* start polling thread
637 	 * if in POLL_FIRST mode, poll once launched;
638 	 * otherwise, not actually poll yet
639 	 */
640 	rte_eal_remote_launch(poll_burst,
641 			      (void *)&pkt_per_port, lcore);
642 
643 	/* start xmit */
644 	i = 0;
645 	while (num) {
646 		nb_tx = RTE_MIN(MAX_PKT_BURST, num);
647 		portid = conf->portlist[i];
648 		nb_tx = rte_eth_tx_burst(portid, 0, &tx_burst[idx], nb_tx);
649 		idx += nb_tx;
650 		num -= nb_tx;
651 		i = (i >= conf->nb_ports - 1) ? 0 : (i + 1);
652 	}
653 
654 	rte_delay_us(5 * US_PER_S);
655 
656 	/* only when polling second  */
657 	if (flags == SC_BURST_XMIT_FIRST)
658 		rte_atomic_store_explicit(&start, 1, rte_memory_order_release);
659 
660 	/* wait for polling finished */
661 	diff_tsc = rte_eal_wait_lcore(lcore);
662 	if (diff_tsc < 0) {
663 		printf("exec_burst: Failed to measure cycles per packet\n");
664 		return -1;
665 	}
666 
667 	printf("Result: %d cycles per packet\n", diff_tsc);
668 
669 	return 0;
670 }
671 
672 static int
test_pmd_perf(void)673 test_pmd_perf(void)
674 {
675 	uint16_t nb_ports, num, nb_lcores, worker_id = (uint16_t)-1;
676 	uint16_t nb_rxd = MAX_TRAFFIC_BURST;
677 	uint16_t nb_txd = MAX_TRAFFIC_BURST;
678 	uint16_t portid;
679 	uint16_t nb_rx_queue = 1, nb_tx_queue = 1;
680 	int socketid = -1;
681 	int ret;
682 
683 	printf("Start PMD RXTX cycles cost test.\n");
684 
685 #ifndef RTE_EXEC_ENV_WINDOWS
686 	signal(SIGUSR1, signal_handler);
687 	signal(SIGUSR2, signal_handler);
688 #endif
689 
690 	nb_ports = rte_eth_dev_count_avail();
691 	if (nb_ports < NB_ETHPORTS_USED) {
692 		printf("At least %u port(s) used for perf. test\n",
693 		       NB_ETHPORTS_USED);
694 		return -1;
695 	}
696 
697 	nb_lcores = rte_lcore_count();
698 
699 	memset(lcore_conf, 0, sizeof(lcore_conf));
700 	init_lcores();
701 
702 	init_mbufpool(NB_MBUF);
703 
704 	if (sc_flag == SC_CONTINUOUS) {
705 		nb_rxd = RX_DESC_DEFAULT;
706 		nb_txd = TX_DESC_DEFAULT;
707 	}
708 	printf("CONFIG RXD=%d TXD=%d\n", nb_rxd, nb_txd);
709 
710 	reset_count();
711 	num = 0;
712 	RTE_ETH_FOREACH_DEV(portid) {
713 		if (socketid == -1) {
714 			worker_id = alloc_lcore(rte_eth_dev_socket_id(portid));
715 			if (worker_id == (uint16_t)-1) {
716 				printf("No avail lcore to run test\n");
717 				return -1;
718 			}
719 			socketid = rte_lcore_to_socket_id(worker_id);
720 			printf("Performance test runs on lcore %u socket %u\n",
721 			       worker_id, socketid);
722 		}
723 
724 		if (socketid != rte_eth_dev_socket_id(portid) &&
725 		    rte_eth_dev_socket_id(portid) != SOCKET_ID_ANY) {
726 			printf("Skip port %d\n", portid);
727 			continue;
728 		}
729 
730 		/* port configure */
731 		ret = rte_eth_dev_configure(portid, nb_rx_queue,
732 					    nb_tx_queue, &port_conf);
733 		if (ret < 0)
734 			rte_exit(EXIT_FAILURE,
735 				"Cannot configure device: err=%d, port=%d\n",
736 				 ret, portid);
737 
738 		ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
739 		if (ret < 0)
740 			rte_exit(EXIT_FAILURE,
741 				"Cannot get mac address: err=%d, port=%d\n",
742 				 ret, portid);
743 
744 		printf("Port %u ", portid);
745 		print_ethaddr("Address:", &ports_eth_addr[portid]);
746 		printf("\n");
747 
748 		/* tx queue setup */
749 		ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
750 					     socketid, &tx_conf);
751 		if (ret < 0)
752 			rte_exit(EXIT_FAILURE,
753 				"rte_eth_tx_queue_setup: err=%d, "
754 				"port=%d\n", ret, portid);
755 
756 		/* rx queue steup */
757 		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
758 						socketid, &rx_conf,
759 						mbufpool[socketid]);
760 		if (ret < 0)
761 			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
762 				 "port=%d\n", ret, portid);
763 
764 		/* Start device */
765 		stop = 0;
766 		ret = rte_eth_dev_start(portid);
767 		if (ret < 0)
768 			rte_exit(EXIT_FAILURE,
769 				"rte_eth_dev_start: err=%d, port=%d\n",
770 				ret, portid);
771 
772 		/* always enable promiscuous */
773 		ret = rte_eth_promiscuous_enable(portid);
774 		if (ret != 0)
775 			rte_exit(EXIT_FAILURE,
776 				 "rte_eth_promiscuous_enable: err=%s, port=%d\n",
777 				 rte_strerror(-ret), portid);
778 
779 		lcore_conf[worker_id].portlist[num++] = portid;
780 		lcore_conf[worker_id].nb_ports++;
781 	}
782 	check_all_ports_link_status(nb_ports, RTE_PORT_ALL);
783 
784 	if (tx_burst == NULL) {
785 		tx_burst = (struct rte_mbuf **)
786 			rte_calloc_socket("tx_buff",
787 					  MAX_TRAFFIC_BURST * nb_ports,
788 					  sizeof(void *),
789 					  RTE_CACHE_LINE_SIZE, socketid);
790 		if (!tx_burst)
791 			return -1;
792 	}
793 
794 	init_traffic(mbufpool[socketid],
795 		     tx_burst, MAX_TRAFFIC_BURST * nb_ports);
796 
797 	printf("Generate %d packets @socket %d\n",
798 	       MAX_TRAFFIC_BURST * nb_ports, socketid);
799 
800 	if (sc_flag == SC_CONTINUOUS) {
801 		/* do both rxtx by default */
802 		if (NULL == do_measure)
803 			do_measure = measure_rxtx;
804 
805 		rte_eal_remote_launch(main_loop, NULL, worker_id);
806 
807 		if (rte_eal_wait_lcore(worker_id) < 0)
808 			return -1;
809 	} else if (sc_flag == SC_BURST_POLL_FIRST ||
810 		   sc_flag == SC_BURST_XMIT_FIRST)
811 		if (exec_burst(sc_flag, worker_id) < 0)
812 			return -1;
813 
814 	/* port tear down */
815 	RTE_ETH_FOREACH_DEV(portid) {
816 		if (socketid != rte_eth_dev_socket_id(portid))
817 			continue;
818 
819 		ret = rte_eth_dev_stop(portid);
820 		if (ret != 0)
821 			printf("rte_eth_dev_stop: err=%s, port=%u\n",
822 			       rte_strerror(-ret), portid);
823 	}
824 
825 	return 0;
826 }
827 
828 int
test_set_rxtx_conf(cmdline_fixed_string_t mode)829 test_set_rxtx_conf(cmdline_fixed_string_t mode)
830 {
831 	printf("mode switch to %s\n", mode);
832 
833 	if (!strcmp(mode, "vector")) {
834 		/* vector rx, tx */
835 		tx_conf.tx_rs_thresh = 32;
836 		tx_conf.tx_free_thresh = 32;
837 		return 0;
838 	} else if (!strcmp(mode, "scalar")) {
839 		/* bulk alloc rx, full-featured tx */
840 		tx_conf.tx_rs_thresh = 32;
841 		tx_conf.tx_free_thresh = 32;
842 		port_conf.rxmode.offloads |= RTE_ETH_RX_OFFLOAD_CHECKSUM;
843 		return 0;
844 	} else if (!strcmp(mode, "hybrid")) {
845 		/* bulk alloc rx, vector tx
846 		 * when vec macro not define,
847 		 * using the same rx/tx as scalar
848 		 */
849 		tx_conf.tx_rs_thresh = 32;
850 		tx_conf.tx_free_thresh = 32;
851 		port_conf.rxmode.offloads |= RTE_ETH_RX_OFFLOAD_CHECKSUM;
852 		return 0;
853 	} else if (!strcmp(mode, "full")) {
854 		/* full feature rx,tx pair */
855 		tx_conf.tx_rs_thresh = 32;
856 		tx_conf.tx_free_thresh = 32;
857 		port_conf.rxmode.offloads |= RTE_ETH_RX_OFFLOAD_SCATTER;
858 		return 0;
859 	}
860 
861 	return -1;
862 }
863 
864 int
test_set_rxtx_anchor(cmdline_fixed_string_t type)865 test_set_rxtx_anchor(cmdline_fixed_string_t type)
866 {
867 	printf("type switch to %s\n", type);
868 
869 	if (!strcmp(type, "rxtx")) {
870 		do_measure = measure_rxtx;
871 		return 0;
872 	} else if (!strcmp(type, "rxonly")) {
873 		do_measure = measure_rxonly;
874 		return 0;
875 	} else if (!strcmp(type, "txonly")) {
876 		do_measure = measure_txonly;
877 		return 0;
878 	}
879 
880 	return -1;
881 }
882 
883 int
test_set_rxtx_sc(cmdline_fixed_string_t type)884 test_set_rxtx_sc(cmdline_fixed_string_t type)
885 {
886 	printf("stream control switch to %s\n", type);
887 
888 	if (!strcmp(type, "continuous")) {
889 		sc_flag = SC_CONTINUOUS;
890 		return 0;
891 	} else if (!strcmp(type, "poll_before_xmit")) {
892 		sc_flag = SC_BURST_POLL_FIRST;
893 		return 0;
894 	} else if (!strcmp(type, "poll_after_xmit")) {
895 		sc_flag = SC_BURST_XMIT_FIRST;
896 		return 0;
897 	}
898 
899 	return -1;
900 }
901 
902 REGISTER_PERF_TEST(pmd_perf_autotest, test_pmd_perf);
903