xref: /dpdk/app/test/test_pmd_perf.c (revision 6e6b34fb26ddc8cf3762bcd0bbb99d9864741f7b)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 
35 #include <stdio.h>
36 #include <inttypes.h>
37 #include <signal.h>
38 #include <unistd.h>
39 #include <rte_cycles.h>
40 #include <rte_ethdev.h>
41 #include <rte_byteorder.h>
42 #include <rte_atomic.h>
43 #include <rte_malloc.h>
44 #include "packet_burst_generator.h"
45 #include "test.h"
46 
47 #define NB_ETHPORTS_USED                (1)
48 #define NB_SOCKETS                      (2)
49 #define MEMPOOL_CACHE_SIZE 250
50 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
51 #define MAX_PKT_BURST                   (32)
52 #define RTE_TEST_RX_DESC_DEFAULT        (128)
53 #define RTE_TEST_TX_DESC_DEFAULT        (512)
54 #define RTE_PORT_ALL            (~(uint8_t)0x0)
55 
56 /* how long test would take at full line rate */
57 #define RTE_TEST_DURATION                (2)
58 
59 /*
60  * RX and TX Prefetch, Host, and Write-back threshold values should be
61  * carefully set for optimal performance. Consult the network
62  * controller's datasheet and supporting DPDK documentation for guidance
63  * on how these parameters should be set.
64  */
65 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
66 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
67 #define RX_WTHRESH 0 /**< Default values of RX write-back threshold reg. */
68 
69 /*
70  * These default values are optimized for use with the Intel(R) 82599 10 GbE
71  * Controller and the DPDK ixgbe PMD. Consider using other values for other
72  * network controllers and/or network drivers.
73  */
74 #define TX_PTHRESH 32 /**< Default values of TX prefetch threshold reg. */
75 #define TX_HTHRESH 0  /**< Default values of TX host threshold reg. */
76 #define TX_WTHRESH 0  /**< Default values of TX write-back threshold reg. */
77 
78 #define MAX_TRAFFIC_BURST              2048
79 
80 #define NB_MBUF RTE_MAX(						\
81 		(unsigned)(nb_ports*nb_rx_queue*nb_rxd +		\
82 			   nb_ports*nb_lcores*MAX_PKT_BURST +		\
83 			   nb_ports*nb_tx_queue*nb_txd +		\
84 			   nb_lcores*MEMPOOL_CACHE_SIZE +		\
85 			   nb_ports*MAX_TRAFFIC_BURST),			\
86 			(unsigned)8192)
87 
88 
89 static struct rte_mempool *mbufpool[NB_SOCKETS];
90 /* ethernet addresses of ports */
91 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
92 
93 static struct rte_eth_conf port_conf = {
94 	.rxmode = {
95 		.mq_mode = ETH_MQ_RX_NONE,
96 		.max_rx_pkt_len = ETHER_MAX_LEN,
97 		.split_hdr_size = 0,
98 		.header_split   = 0, /**< Header Split disabled */
99 		.hw_ip_checksum = 0, /**< IP checksum offload enabled */
100 		.hw_vlan_filter = 0, /**< VLAN filtering disabled */
101 		.hw_vlan_strip  = 0, /**< VLAN strip enabled. */
102 		.hw_vlan_extend = 0, /**< Extended VLAN disabled. */
103 		.jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
104 		.hw_strip_crc   = 0, /**< CRC stripped by hardware */
105 		.enable_scatter = 0, /**< scatter rx disabled */
106 	},
107 	.txmode = {
108 		.mq_mode = ETH_MQ_TX_NONE,
109 	},
110 	.lpbk_mode = 1,  /* enable loopback */
111 };
112 
113 static struct rte_eth_rxconf rx_conf = {
114 	.rx_thresh = {
115 		.pthresh = RX_PTHRESH,
116 		.hthresh = RX_HTHRESH,
117 		.wthresh = RX_WTHRESH,
118 	},
119 	.rx_free_thresh = 32,
120 };
121 
122 static struct rte_eth_txconf tx_conf = {
123 	.tx_thresh = {
124 		.pthresh = TX_PTHRESH,
125 		.hthresh = TX_HTHRESH,
126 		.wthresh = TX_WTHRESH,
127 	},
128 	.tx_free_thresh = 32, /* Use PMD default values */
129 	.tx_rs_thresh = 32, /* Use PMD default values */
130 	.txq_flags = (ETH_TXQ_FLAGS_NOMULTSEGS |
131 		      ETH_TXQ_FLAGS_NOVLANOFFL |
132 		      ETH_TXQ_FLAGS_NOXSUMSCTP |
133 		      ETH_TXQ_FLAGS_NOXSUMUDP |
134 		      ETH_TXQ_FLAGS_NOXSUMTCP)
135 };
136 
137 enum {
138 	LCORE_INVALID = 0,
139 	LCORE_AVAIL,
140 	LCORE_USED,
141 };
142 
143 struct lcore_conf {
144 	uint8_t status;
145 	uint8_t socketid;
146 	uint16_t nb_ports;
147 	uint8_t portlist[RTE_MAX_ETHPORTS];
148 } __rte_cache_aligned;
149 
150 struct lcore_conf lcore_conf[RTE_MAX_LCORE];
151 
152 static uint64_t link_mbps;
153 
154 enum {
155 	SC_CONTINUOUS = 0,
156 	SC_BURST_POLL_FIRST,
157 	SC_BURST_XMIT_FIRST,
158 };
159 
160 static uint32_t sc_flag;
161 
162 /* Check the link status of all ports in up to 3s, and print them finally */
163 static void
164 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
165 {
166 #define CHECK_INTERVAL 100 /* 100ms */
167 #define MAX_CHECK_TIME 30 /* 3s (30 * 100ms) in total */
168 	uint8_t portid, count, all_ports_up, print_flag = 0;
169 	struct rte_eth_link link;
170 
171 	printf("Checking link statuses...\n");
172 	fflush(stdout);
173 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
174 		all_ports_up = 1;
175 		for (portid = 0; portid < port_num; portid++) {
176 			if ((port_mask & (1 << portid)) == 0)
177 				continue;
178 			memset(&link, 0, sizeof(link));
179 			rte_eth_link_get_nowait(portid, &link);
180 			/* print link status if flag set */
181 			if (print_flag == 1) {
182 				if (link.link_status) {
183 					printf("Port %d Link Up - speed %u "
184 						"Mbps - %s\n", (uint8_t)portid,
185 						(unsigned)link.link_speed,
186 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
187 					("full-duplex") : ("half-duplex\n"));
188 					if (link_mbps == 0)
189 						link_mbps = link.link_speed;
190 				} else
191 					printf("Port %d Link Down\n",
192 						(uint8_t)portid);
193 				continue;
194 			}
195 			/* clear all_ports_up flag if any link down */
196 			if (link.link_status == 0) {
197 				all_ports_up = 0;
198 				break;
199 			}
200 		}
201 		/* after finally printing all link status, get out */
202 		if (print_flag == 1)
203 			break;
204 
205 		if (all_ports_up == 0) {
206 			fflush(stdout);
207 			rte_delay_ms(CHECK_INTERVAL);
208 		}
209 
210 		/* set the print_flag if all ports up or timeout */
211 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1))
212 			print_flag = 1;
213 	}
214 }
215 
216 static void
217 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
218 {
219 	char buf[ETHER_ADDR_FMT_SIZE];
220 	ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
221 	printf("%s%s", name, buf);
222 }
223 
224 static int
225 init_traffic(struct rte_mempool *mp,
226 	     struct rte_mbuf **pkts_burst, uint32_t burst_size)
227 {
228 	struct ether_hdr pkt_eth_hdr;
229 	struct ipv4_hdr pkt_ipv4_hdr;
230 	struct udp_hdr pkt_udp_hdr;
231 	uint32_t pktlen;
232 	static uint8_t src_mac[] = { 0x00, 0xFF, 0xAA, 0xFF, 0xAA, 0xFF };
233 	static uint8_t dst_mac[] = { 0x00, 0xAA, 0xFF, 0xAA, 0xFF, 0xAA };
234 
235 
236 	initialize_eth_header(&pkt_eth_hdr,
237 		(struct ether_addr *)src_mac,
238 		(struct ether_addr *)dst_mac, ETHER_TYPE_IPv4, 0, 0);
239 
240 	pktlen = initialize_ipv4_header(&pkt_ipv4_hdr,
241 					IPV4_ADDR(10, 0, 0, 1),
242 					IPV4_ADDR(10, 0, 0, 2), 26);
243 	printf("IPv4 pktlen %u\n", pktlen);
244 
245 	pktlen = initialize_udp_header(&pkt_udp_hdr, 0, 0, 18);
246 
247 	printf("UDP pktlen %u\n", pktlen);
248 
249 	return generate_packet_burst(mp, pkts_burst, &pkt_eth_hdr,
250 				     0, &pkt_ipv4_hdr, 1,
251 				     &pkt_udp_hdr, burst_size,
252 				     PACKET_BURST_GEN_PKT_LEN, 1);
253 }
254 
255 static int
256 init_lcores(void)
257 {
258 	unsigned lcore_id;
259 
260 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
261 		lcore_conf[lcore_id].socketid =
262 			rte_lcore_to_socket_id(lcore_id);
263 		if (rte_lcore_is_enabled(lcore_id) == 0) {
264 			lcore_conf[lcore_id].status = LCORE_INVALID;
265 			continue;
266 		} else
267 			lcore_conf[lcore_id].status = LCORE_AVAIL;
268 	}
269 	return 0;
270 }
271 
272 static int
273 init_mbufpool(unsigned nb_mbuf)
274 {
275 	int socketid;
276 	unsigned lcore_id;
277 	char s[64];
278 
279 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
280 		if (rte_lcore_is_enabled(lcore_id) == 0)
281 			continue;
282 
283 		socketid = rte_lcore_to_socket_id(lcore_id);
284 		if (socketid >= NB_SOCKETS) {
285 			rte_exit(EXIT_FAILURE,
286 				"Socket %d of lcore %u is out of range %d\n",
287 				socketid, lcore_id, NB_SOCKETS);
288 		}
289 		if (mbufpool[socketid] == NULL) {
290 			snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
291 			mbufpool[socketid] =
292 				rte_mempool_create(s, nb_mbuf, MBUF_SIZE,
293 					MEMPOOL_CACHE_SIZE,
294 					sizeof(struct rte_pktmbuf_pool_private),
295 					rte_pktmbuf_pool_init, NULL,
296 					rte_pktmbuf_init, NULL,
297 					socketid, 0);
298 			if (mbufpool[socketid] == NULL)
299 				rte_exit(EXIT_FAILURE,
300 					"Cannot init mbuf pool on socket %d\n",
301 					socketid);
302 			else
303 				printf("Allocated mbuf pool on socket %d\n",
304 					socketid);
305 		}
306 	}
307 	return 0;
308 }
309 
310 static uint16_t
311 alloc_lcore(uint16_t socketid)
312 {
313 	unsigned lcore_id;
314 
315 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
316 		if (LCORE_AVAIL != lcore_conf[lcore_id].status ||
317 		    lcore_conf[lcore_id].socketid != socketid ||
318 		    lcore_id == rte_get_master_lcore())
319 			continue;
320 		lcore_conf[lcore_id].status = LCORE_USED;
321 		lcore_conf[lcore_id].nb_ports = 0;
322 		return lcore_id;
323 	}
324 
325 	return (uint16_t)-1;
326 }
327 
328 volatile uint64_t stop;
329 uint64_t count;
330 uint64_t drop;
331 uint64_t idle;
332 
333 static void
334 reset_count(void)
335 {
336 	count = 0;
337 	drop = 0;
338 	idle = 0;
339 }
340 
341 static void
342 stats_display(uint8_t port_id)
343 {
344 	struct rte_eth_stats stats;
345 	rte_eth_stats_get(port_id, &stats);
346 
347 	printf("  RX-packets: %-10"PRIu64" RX-missed: %-10"PRIu64" RX-bytes:  "
348 	       "%-"PRIu64"\n",
349 	       stats.ipackets, stats.imissed, stats.ibytes);
350 	printf("  RX-badcrc:  %-10"PRIu64" RX-badlen: %-10"PRIu64" RX-errors: "
351 	       "%-"PRIu64"\n",
352 	       stats.ibadcrc, stats.ibadlen, stats.ierrors);
353 	printf("  RX-nombuf:  %-10"PRIu64"\n",
354 	       stats.rx_nombuf);
355 	printf("  TX-packets: %-10"PRIu64" TX-errors: %-10"PRIu64" TX-bytes:  "
356 	       "%-"PRIu64"\n",
357 	       stats.opackets, stats.oerrors, stats.obytes);
358 }
359 
360 static void
361 signal_handler(int signum)
362 {
363 	/*  USR1 signal, stop testing */
364 	if (signum == SIGUSR1) {
365 		printf("Force Stop!\n");
366 		stop = 1;
367 	}
368 
369 	/*  USR2 signal, print stats */
370 	if (signum == SIGUSR2)
371 		stats_display(0);
372 }
373 
374 struct rte_mbuf **tx_burst;
375 
376 uint64_t (*do_measure)(struct lcore_conf *conf,
377 		       struct rte_mbuf *pkts_burst[],
378 		       uint64_t total_pkts);
379 
380 static uint64_t
381 measure_rxtx(struct lcore_conf *conf,
382 	     struct rte_mbuf *pkts_burst[],
383 	     uint64_t total_pkts)
384 {
385 	unsigned i, portid, nb_rx, nb_tx;
386 	uint64_t prev_tsc, cur_tsc;
387 
388 	prev_tsc = rte_rdtsc();
389 
390 	while (likely(!stop)) {
391 		for (i = 0; i < conf->nb_ports; i++) {
392 			portid = conf->portlist[i];
393 			nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
394 						 pkts_burst, MAX_PKT_BURST);
395 			if (unlikely(nb_rx == 0)) {
396 				idle++;
397 				continue;
398 			}
399 
400 			count += nb_rx;
401 			nb_tx = rte_eth_tx_burst(portid, 0, pkts_burst, nb_rx);
402 			if (unlikely(nb_tx < nb_rx)) {
403 				drop += (nb_rx - nb_tx);
404 				do {
405 					rte_pktmbuf_free(pkts_burst[nb_tx]);
406 				} while (++nb_tx < nb_rx);
407 			}
408 		}
409 		if (unlikely(count >= total_pkts))
410 			break;
411 	}
412 
413 	cur_tsc = rte_rdtsc();
414 
415 	return cur_tsc - prev_tsc;
416 }
417 
418 static uint64_t
419 measure_rxonly(struct lcore_conf *conf,
420 	       struct rte_mbuf *pkts_burst[],
421 	       uint64_t total_pkts)
422 {
423 	unsigned i, portid, nb_rx, nb_tx;
424 	uint64_t diff_tsc, cur_tsc;
425 
426 	diff_tsc = 0;
427 	while (likely(!stop)) {
428 		for (i = 0; i < conf->nb_ports; i++) {
429 			portid = conf->portlist[i];
430 
431 			cur_tsc = rte_rdtsc();
432 			nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
433 						 pkts_burst, MAX_PKT_BURST);
434 			if (unlikely(nb_rx == 0)) {
435 				idle++;
436 				continue;
437 			}
438 			diff_tsc += rte_rdtsc() - cur_tsc;
439 
440 			count += nb_rx;
441 			nb_tx = rte_eth_tx_burst(portid, 0, pkts_burst, nb_rx);
442 			if (unlikely(nb_tx < nb_rx)) {
443 				drop += (nb_rx - nb_tx);
444 				do {
445 					rte_pktmbuf_free(pkts_burst[nb_tx]);
446 				} while (++nb_tx < nb_rx);
447 			}
448 		}
449 		if (unlikely(count >= total_pkts))
450 			break;
451 	}
452 
453 	return diff_tsc;
454 }
455 
456 static uint64_t
457 measure_txonly(struct lcore_conf *conf,
458 	       struct rte_mbuf *pkts_burst[],
459 	       uint64_t total_pkts)
460 {
461 	unsigned i, portid, nb_rx, nb_tx;
462 	uint64_t diff_tsc, cur_tsc;
463 
464 	printf("do tx measure\n");
465 	diff_tsc = 0;
466 	while (likely(!stop)) {
467 		for (i = 0; i < conf->nb_ports; i++) {
468 			portid = conf->portlist[i];
469 			nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
470 						 pkts_burst, MAX_PKT_BURST);
471 			if (unlikely(nb_rx == 0)) {
472 				idle++;
473 				continue;
474 			}
475 
476 			count += nb_rx;
477 
478 			cur_tsc = rte_rdtsc();
479 			nb_tx = rte_eth_tx_burst(portid, 0, pkts_burst, nb_rx);
480 			if (unlikely(nb_tx < nb_rx)) {
481 				drop += (nb_rx - nb_tx);
482 				do {
483 					rte_pktmbuf_free(pkts_burst[nb_tx]);
484 				} while (++nb_tx < nb_rx);
485 			}
486 			diff_tsc += rte_rdtsc() - cur_tsc;
487 		}
488 		if (unlikely(count >= total_pkts))
489 			break;
490 	}
491 
492 	return diff_tsc;
493 }
494 
495 /* main processing loop */
496 static int
497 main_loop(__rte_unused void *args)
498 {
499 #define PACKET_SIZE 64
500 #define FRAME_GAP 12
501 #define MAC_PREAMBLE 8
502 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
503 	unsigned lcore_id;
504 	unsigned i, portid, nb_rx = 0, nb_tx = 0;
505 	struct lcore_conf *conf;
506 	int pkt_per_port;
507 	uint64_t diff_tsc;
508 	uint64_t packets_per_second, total_packets;
509 
510 	lcore_id = rte_lcore_id();
511 	conf = &lcore_conf[lcore_id];
512 	if (conf->status != LCORE_USED)
513 		return 0;
514 
515 	pkt_per_port = MAX_TRAFFIC_BURST;
516 
517 	int idx = 0;
518 	for (i = 0; i < conf->nb_ports; i++) {
519 		int num = pkt_per_port;
520 		portid = conf->portlist[i];
521 		printf("inject %d packet to port %d\n", num, portid);
522 		while (num) {
523 			nb_tx = RTE_MIN(MAX_PKT_BURST, num);
524 			nb_tx = rte_eth_tx_burst(portid, 0,
525 						&tx_burst[idx], nb_tx);
526 			num -= nb_tx;
527 			idx += nb_tx;
528 		}
529 	}
530 	printf("Total packets inject to prime ports = %u\n", idx);
531 
532 	packets_per_second = (link_mbps * 1000 * 1000) /
533 		((PACKET_SIZE + FRAME_GAP + MAC_PREAMBLE) * CHAR_BIT);
534 	printf("Each port will do %"PRIu64" packets per second\n",
535 	       packets_per_second);
536 
537 	total_packets = RTE_TEST_DURATION * conf->nb_ports * packets_per_second;
538 	printf("Test will stop after at least %"PRIu64" packets received\n",
539 		+ total_packets);
540 
541 	diff_tsc = do_measure(conf, pkts_burst, total_packets);
542 
543 	for (i = 0; i < conf->nb_ports; i++) {
544 		portid = conf->portlist[i];
545 		int nb_free = pkt_per_port;
546 		do { /* dry out */
547 			nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
548 						 pkts_burst, MAX_PKT_BURST);
549 			nb_tx = 0;
550 			while (nb_tx < nb_rx)
551 				rte_pktmbuf_free(pkts_burst[nb_tx++]);
552 			nb_free -= nb_rx;
553 		} while (nb_free != 0);
554 		printf("free %d mbuf left in port %u\n", pkt_per_port, portid);
555 	}
556 
557 	if (count == 0)
558 		return -1;
559 
560 	printf("%"PRIu64" packet, %"PRIu64" drop, %"PRIu64" idle\n",
561 	       count, drop, idle);
562 	printf("Result: %"PRIu64" cycles per packet\n", diff_tsc / count);
563 
564 	return 0;
565 }
566 
567 rte_atomic64_t start;
568 
569 static inline int
570 poll_burst(void *args)
571 {
572 #define MAX_IDLE           (10000)
573 	unsigned lcore_id;
574 	struct rte_mbuf **pkts_burst;
575 	uint64_t diff_tsc, cur_tsc;
576 	uint16_t next[RTE_MAX_ETHPORTS];
577 	struct lcore_conf *conf;
578 	uint32_t pkt_per_port = *((uint32_t *)args);
579 	unsigned i, portid, nb_rx = 0;
580 	uint64_t total;
581 	uint64_t timeout = MAX_IDLE;
582 
583 	lcore_id = rte_lcore_id();
584 	conf = &lcore_conf[lcore_id];
585 	if (conf->status != LCORE_USED)
586 		return 0;
587 
588 	total = pkt_per_port * conf->nb_ports;
589 	printf("start to receive total expect %"PRIu64"\n", total);
590 
591 	pkts_burst = (struct rte_mbuf **)
592 		rte_calloc_socket("poll_burst",
593 				  total, sizeof(void *),
594 				  RTE_CACHE_LINE_SIZE, conf->socketid);
595 	if (!pkts_burst)
596 		return -1;
597 
598 	for (i = 0; i < conf->nb_ports; i++) {
599 		portid = conf->portlist[i];
600 		next[portid] = i * pkt_per_port;
601 	}
602 
603 	while (!rte_atomic64_read(&start))
604 		;
605 
606 	cur_tsc = rte_rdtsc();
607 	while (total) {
608 		for (i = 0; i < conf->nb_ports; i++) {
609 			portid = conf->portlist[i];
610 			nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
611 						 &pkts_burst[next[portid]],
612 						 MAX_PKT_BURST);
613 			if (unlikely(nb_rx == 0)) {
614 				timeout--;
615 				if (unlikely(timeout == 0))
616 					goto timeout;
617 				continue;
618 			}
619 			next[portid] += nb_rx;
620 			total -= nb_rx;
621 		}
622 	}
623 timeout:
624 	diff_tsc = rte_rdtsc() - cur_tsc;
625 
626 	printf("%"PRIu64" packets lost, IDLE %"PRIu64" times\n",
627 	       total, MAX_IDLE - timeout);
628 
629 	/* clean up */
630 	total = pkt_per_port * conf->nb_ports - total;
631 	for (i = 0; i < total; i++)
632 		rte_pktmbuf_free(pkts_burst[i]);
633 
634 	rte_free(pkts_burst);
635 
636 	return diff_tsc / total;
637 }
638 
639 static int
640 exec_burst(uint32_t flags, int lcore)
641 {
642 	unsigned i, portid, nb_tx = 0;
643 	struct lcore_conf *conf;
644 	uint32_t pkt_per_port;
645 	int num, idx = 0;
646 	int diff_tsc;
647 
648 	conf = &lcore_conf[lcore];
649 
650 	pkt_per_port = MAX_TRAFFIC_BURST;
651 	num = pkt_per_port;
652 
653 	rte_atomic64_init(&start);
654 
655 	/* start polling thread, but not actually poll yet */
656 	rte_eal_remote_launch(poll_burst,
657 			      (void *)&pkt_per_port, lcore);
658 
659 	/* Only when polling first */
660 	if (flags == SC_BURST_POLL_FIRST)
661 		rte_atomic64_set(&start, 1);
662 
663 	/* start xmit */
664 	while (num) {
665 		nb_tx = RTE_MIN(MAX_PKT_BURST, num);
666 		for (i = 0; i < conf->nb_ports; i++) {
667 			portid = conf->portlist[i];
668 			rte_eth_tx_burst(portid, 0,
669 					 &tx_burst[idx], nb_tx);
670 			idx += nb_tx;
671 		}
672 		num -= nb_tx;
673 	}
674 
675 	sleep(5);
676 
677 	/* only when polling second  */
678 	if (flags == SC_BURST_XMIT_FIRST)
679 		rte_atomic64_set(&start, 1);
680 
681 	/* wait for polling finished */
682 	diff_tsc = rte_eal_wait_lcore(lcore);
683 	if (diff_tsc < 0)
684 		return -1;
685 
686 	printf("Result: %d cycles per packet\n", diff_tsc);
687 
688 	return 0;
689 }
690 
691 static int
692 test_pmd_perf(void)
693 {
694 	uint16_t nb_ports, num, nb_lcores, slave_id = (uint16_t)-1;
695 	uint16_t nb_rxd = MAX_TRAFFIC_BURST;
696 	uint16_t nb_txd = MAX_TRAFFIC_BURST;
697 	uint16_t portid;
698 	uint16_t nb_rx_queue = 1, nb_tx_queue = 1;
699 	int socketid = -1;
700 	int ret;
701 
702 	printf("Start PMD RXTX cycles cost test.\n");
703 
704 	signal(SIGUSR1, signal_handler);
705 	signal(SIGUSR2, signal_handler);
706 
707 	nb_ports = rte_eth_dev_count();
708 	if (nb_ports < NB_ETHPORTS_USED) {
709 		printf("At least %u port(s) used for perf. test\n",
710 		       NB_ETHPORTS_USED);
711 		return -1;
712 	}
713 
714 	if (nb_ports > RTE_MAX_ETHPORTS)
715 		nb_ports = RTE_MAX_ETHPORTS;
716 
717 	nb_lcores = rte_lcore_count();
718 
719 	memset(lcore_conf, 0, sizeof(lcore_conf));
720 	init_lcores();
721 
722 	init_mbufpool(NB_MBUF);
723 
724 	if (sc_flag == SC_CONTINUOUS) {
725 		nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
726 		nb_txd = RTE_TEST_TX_DESC_DEFAULT;
727 	}
728 	printf("CONFIG RXD=%d TXD=%d\n", nb_rxd, nb_txd);
729 
730 	reset_count();
731 	num = 0;
732 	for (portid = 0; portid < nb_ports; portid++) {
733 		if (socketid == -1) {
734 			socketid = rte_eth_dev_socket_id(portid);
735 			slave_id = alloc_lcore(socketid);
736 			if (slave_id == (uint16_t)-1) {
737 				printf("No avail lcore to run test\n");
738 				return -1;
739 			}
740 			printf("Performance test runs on lcore %u socket %u\n",
741 			       slave_id, socketid);
742 		}
743 
744 		if (socketid != rte_eth_dev_socket_id(portid)) {
745 			printf("Skip port %d\n", portid);
746 			continue;
747 		}
748 
749 		/* port configure */
750 		ret = rte_eth_dev_configure(portid, nb_rx_queue,
751 					    nb_tx_queue, &port_conf);
752 		if (ret < 0)
753 			rte_exit(EXIT_FAILURE,
754 				"Cannot configure device: err=%d, port=%d\n",
755 				 ret, portid);
756 
757 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
758 		printf("Port %u ", portid);
759 		print_ethaddr("Address:", &ports_eth_addr[portid]);
760 		printf("\n");
761 
762 		/* tx queue setup */
763 		ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
764 					     socketid, &tx_conf);
765 		if (ret < 0)
766 			rte_exit(EXIT_FAILURE,
767 				"rte_eth_tx_queue_setup: err=%d, "
768 				"port=%d\n", ret, portid);
769 
770 		/* rx queue steup */
771 		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
772 						socketid, &rx_conf,
773 						mbufpool[socketid]);
774 		if (ret < 0)
775 			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
776 				 "port=%d\n", ret, portid);
777 
778 		/* Start device */
779 		stop = 0;
780 		ret = rte_eth_dev_start(portid);
781 		if (ret < 0)
782 			rte_exit(EXIT_FAILURE,
783 				"rte_eth_dev_start: err=%d, port=%d\n",
784 				ret, portid);
785 
786 		/* always eanble promiscuous */
787 		rte_eth_promiscuous_enable(portid);
788 
789 		lcore_conf[slave_id].portlist[num++] = portid;
790 		lcore_conf[slave_id].nb_ports++;
791 	}
792 	check_all_ports_link_status(nb_ports, RTE_PORT_ALL);
793 
794 	if (tx_burst == NULL) {
795 		tx_burst = (struct rte_mbuf **)
796 			rte_calloc_socket("tx_buff",
797 					  MAX_TRAFFIC_BURST * nb_ports,
798 					  sizeof(void *),
799 					  RTE_CACHE_LINE_SIZE, socketid);
800 		if (!tx_burst)
801 			return -1;
802 	}
803 
804 	init_traffic(mbufpool[socketid],
805 		     tx_burst, MAX_TRAFFIC_BURST * nb_ports);
806 
807 	printf("Generate %d packets @socket %d\n",
808 	       MAX_TRAFFIC_BURST * nb_ports, socketid);
809 
810 	if (sc_flag == SC_CONTINUOUS) {
811 		/* do both rxtx by default */
812 		if (NULL == do_measure)
813 			do_measure = measure_rxtx;
814 
815 		rte_eal_remote_launch(main_loop, NULL, slave_id);
816 
817 		if (rte_eal_wait_lcore(slave_id) < 0)
818 			return -1;
819 	} else if (sc_flag == SC_BURST_POLL_FIRST ||
820 		   sc_flag == SC_BURST_XMIT_FIRST)
821 		exec_burst(sc_flag, slave_id);
822 
823 	/* port tear down */
824 	for (portid = 0; portid < nb_ports; portid++) {
825 		if (socketid != rte_eth_dev_socket_id(portid))
826 			continue;
827 
828 		rte_eth_dev_stop(portid);
829 	}
830 
831 	return 0;
832 }
833 
834 int
835 test_set_rxtx_conf(cmdline_fixed_string_t mode)
836 {
837 	printf("mode switch to %s\n", mode);
838 
839 	if (!strcmp(mode, "vector")) {
840 		/* vector rx, tx */
841 		tx_conf.txq_flags = 0xf01;
842 		tx_conf.tx_rs_thresh = 32;
843 		tx_conf.tx_free_thresh = 32;
844 		port_conf.rxmode.hw_ip_checksum = 0;
845 		port_conf.rxmode.enable_scatter = 0;
846 		return 0;
847 	} else if (!strcmp(mode, "scalar")) {
848 		/* bulk alloc rx, simple tx */
849 		tx_conf.txq_flags = 0xf01;
850 		tx_conf.tx_rs_thresh = 128;
851 		tx_conf.tx_free_thresh = 128;
852 		port_conf.rxmode.hw_ip_checksum = 1;
853 		port_conf.rxmode.enable_scatter = 0;
854 		return 0;
855 	} else if (!strcmp(mode, "hybrid")) {
856 		/* bulk alloc rx, vector tx
857 		 * when vec macro not define,
858 		 * using the same rx/tx as scalar
859 		 */
860 		tx_conf.txq_flags = 0xf01;
861 		tx_conf.tx_rs_thresh = 32;
862 		tx_conf.tx_free_thresh = 32;
863 		port_conf.rxmode.hw_ip_checksum = 1;
864 		port_conf.rxmode.enable_scatter = 0;
865 		return 0;
866 	} else if (!strcmp(mode, "full")) {
867 		/* full feature rx,tx pair */
868 		tx_conf.txq_flags = 0x0;   /* must condition */
869 		tx_conf.tx_rs_thresh = 32;
870 		tx_conf.tx_free_thresh = 32;
871 		port_conf.rxmode.hw_ip_checksum = 0;
872 		port_conf.rxmode.enable_scatter = 1; /* must condition */
873 		return 0;
874 	}
875 
876 	return -1;
877 }
878 
879 int
880 test_set_rxtx_anchor(cmdline_fixed_string_t type)
881 {
882 	printf("type switch to %s\n", type);
883 
884 	if (!strcmp(type, "rxtx")) {
885 		do_measure = measure_rxtx;
886 		return 0;
887 	} else if (!strcmp(type, "rxonly")) {
888 		do_measure = measure_rxonly;
889 		return 0;
890 	} else if (!strcmp(type, "txonly")) {
891 		do_measure = measure_txonly;
892 		return 0;
893 	}
894 
895 	return -1;
896 }
897 
898 int
899 test_set_rxtx_sc(cmdline_fixed_string_t type)
900 {
901 	printf("stream control switch to %s\n", type);
902 
903 	if (!strcmp(type, "continuous")) {
904 		sc_flag = SC_CONTINUOUS;
905 		return 0;
906 	} else if (!strcmp(type, "poll_before_xmit")) {
907 		sc_flag = SC_BURST_POLL_FIRST;
908 		return 0;
909 	} else if (!strcmp(type, "poll_after_xmit")) {
910 		sc_flag = SC_BURST_XMIT_FIRST;
911 		return 0;
912 	}
913 
914 	return -1;
915 }
916 
917 static struct test_command pmd_perf_cmd = {
918 	.command = "pmd_perf_autotest",
919 	.callback = test_pmd_perf,
920 };
921 REGISTER_TEST_COMMAND(pmd_perf_cmd);
922