xref: /dpdk/app/test/test_pmd_perf.c (revision fc1f2750a3ec6da919e3c86e59d56f34ec97154b)
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, 0, 0);
239 	pkt_eth_hdr.ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv4);
240 
241 	pktlen = initialize_ipv4_header(&pkt_ipv4_hdr,
242 					IPV4_ADDR(10, 0, 0, 1),
243 					IPV4_ADDR(10, 0, 0, 2), 26);
244 	printf("IPv4 pktlen %u\n", pktlen);
245 
246 	pktlen = initialize_udp_header(&pkt_udp_hdr, 0, 0, 18);
247 
248 	printf("UDP pktlen %u\n", pktlen);
249 
250 	return generate_packet_burst(mp, pkts_burst, &pkt_eth_hdr,
251 				     0, &pkt_ipv4_hdr, 1,
252 				     &pkt_udp_hdr, burst_size,
253 				     PACKET_BURST_GEN_PKT_LEN, 1);
254 }
255 
256 static int
257 init_lcores(void)
258 {
259 	unsigned lcore_id;
260 
261 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
262 		lcore_conf[lcore_id].socketid =
263 			rte_lcore_to_socket_id(lcore_id);
264 		if (rte_lcore_is_enabled(lcore_id) == 0) {
265 			lcore_conf[lcore_id].status = LCORE_INVALID;
266 			continue;
267 		} else
268 			lcore_conf[lcore_id].status = LCORE_AVAIL;
269 	}
270 	return 0;
271 }
272 
273 static int
274 init_mbufpool(unsigned nb_mbuf)
275 {
276 	int socketid;
277 	unsigned lcore_id;
278 	char s[64];
279 
280 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
281 		if (rte_lcore_is_enabled(lcore_id) == 0)
282 			continue;
283 
284 		socketid = rte_lcore_to_socket_id(lcore_id);
285 		if (socketid >= NB_SOCKETS) {
286 			rte_exit(EXIT_FAILURE,
287 				"Socket %d of lcore %u is out of range %d\n",
288 				socketid, lcore_id, NB_SOCKETS);
289 		}
290 		if (mbufpool[socketid] == NULL) {
291 			snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
292 			mbufpool[socketid] =
293 				rte_mempool_create(s, nb_mbuf, MBUF_SIZE,
294 					MEMPOOL_CACHE_SIZE,
295 					sizeof(struct rte_pktmbuf_pool_private),
296 					rte_pktmbuf_pool_init, NULL,
297 					rte_pktmbuf_init, NULL,
298 					socketid, 0);
299 			if (mbufpool[socketid] == NULL)
300 				rte_exit(EXIT_FAILURE,
301 					"Cannot init mbuf pool on socket %d\n",
302 					socketid);
303 			else
304 				printf("Allocated mbuf pool on socket %d\n",
305 					socketid);
306 		}
307 	}
308 	return 0;
309 }
310 
311 static uint16_t
312 alloc_lcore(uint16_t socketid)
313 {
314 	unsigned lcore_id;
315 
316 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
317 		if (LCORE_AVAIL != lcore_conf[lcore_id].status ||
318 		    lcore_conf[lcore_id].socketid != socketid ||
319 		    lcore_id == rte_get_master_lcore())
320 			continue;
321 		lcore_conf[lcore_id].status = LCORE_USED;
322 		lcore_conf[lcore_id].nb_ports = 0;
323 		return lcore_id;
324 	}
325 
326 	return (uint16_t)-1;
327 }
328 
329 volatile uint64_t stop;
330 uint64_t count;
331 uint64_t drop;
332 uint64_t idle;
333 
334 static void
335 reset_count(void)
336 {
337 	count = 0;
338 	drop = 0;
339 	idle = 0;
340 }
341 
342 static void
343 stats_display(uint8_t port_id)
344 {
345 	struct rte_eth_stats stats;
346 	rte_eth_stats_get(port_id, &stats);
347 
348 	printf("  RX-packets: %-10"PRIu64" RX-missed: %-10"PRIu64" RX-bytes:  "
349 	       "%-"PRIu64"\n",
350 	       stats.ipackets, stats.imissed, stats.ibytes);
351 	printf("  RX-badcrc:  %-10"PRIu64" RX-badlen: %-10"PRIu64" RX-errors: "
352 	       "%-"PRIu64"\n",
353 	       stats.ibadcrc, stats.ibadlen, stats.ierrors);
354 	printf("  RX-nombuf:  %-10"PRIu64"\n",
355 	       stats.rx_nombuf);
356 	printf("  TX-packets: %-10"PRIu64" TX-errors: %-10"PRIu64" TX-bytes:  "
357 	       "%-"PRIu64"\n",
358 	       stats.opackets, stats.oerrors, stats.obytes);
359 }
360 
361 static void
362 signal_handler(int signum)
363 {
364 	/*  USR1 signal, stop testing */
365 	if (signum == SIGUSR1) {
366 		printf("Force Stop!\n");
367 		stop = 1;
368 	}
369 
370 	/*  USR2 signal, print stats */
371 	if (signum == SIGUSR2)
372 		stats_display(0);
373 }
374 
375 struct rte_mbuf **tx_burst;
376 
377 uint64_t (*do_measure)(struct lcore_conf *conf,
378 		       struct rte_mbuf *pkts_burst[],
379 		       uint64_t total_pkts);
380 
381 static uint64_t
382 measure_rxtx(struct lcore_conf *conf,
383 	     struct rte_mbuf *pkts_burst[],
384 	     uint64_t total_pkts)
385 {
386 	unsigned i, portid, nb_rx, nb_tx;
387 	uint64_t prev_tsc, cur_tsc;
388 
389 	prev_tsc = rte_rdtsc();
390 
391 	while (likely(!stop)) {
392 		for (i = 0; i < conf->nb_ports; i++) {
393 			portid = conf->portlist[i];
394 			nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
395 						 pkts_burst, MAX_PKT_BURST);
396 			if (unlikely(nb_rx == 0)) {
397 				idle++;
398 				continue;
399 			}
400 
401 			count += nb_rx;
402 			nb_tx = rte_eth_tx_burst(portid, 0, pkts_burst, nb_rx);
403 			if (unlikely(nb_tx < nb_rx)) {
404 				drop += (nb_rx - nb_tx);
405 				do {
406 					rte_pktmbuf_free(pkts_burst[nb_tx]);
407 				} while (++nb_tx < nb_rx);
408 			}
409 		}
410 		if (unlikely(count >= total_pkts))
411 			break;
412 	}
413 
414 	cur_tsc = rte_rdtsc();
415 
416 	return cur_tsc - prev_tsc;
417 }
418 
419 static uint64_t
420 measure_rxonly(struct lcore_conf *conf,
421 	       struct rte_mbuf *pkts_burst[],
422 	       uint64_t total_pkts)
423 {
424 	unsigned i, portid, nb_rx, nb_tx;
425 	uint64_t diff_tsc, cur_tsc;
426 
427 	diff_tsc = 0;
428 	while (likely(!stop)) {
429 		for (i = 0; i < conf->nb_ports; i++) {
430 			portid = conf->portlist[i];
431 
432 			cur_tsc = rte_rdtsc();
433 			nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
434 						 pkts_burst, MAX_PKT_BURST);
435 			if (unlikely(nb_rx == 0)) {
436 				idle++;
437 				continue;
438 			}
439 			diff_tsc += rte_rdtsc() - cur_tsc;
440 
441 			count += nb_rx;
442 			nb_tx = rte_eth_tx_burst(portid, 0, pkts_burst, nb_rx);
443 			if (unlikely(nb_tx < nb_rx)) {
444 				drop += (nb_rx - nb_tx);
445 				do {
446 					rte_pktmbuf_free(pkts_burst[nb_tx]);
447 				} while (++nb_tx < nb_rx);
448 			}
449 		}
450 		if (unlikely(count >= total_pkts))
451 			break;
452 	}
453 
454 	return diff_tsc;
455 }
456 
457 static uint64_t
458 measure_txonly(struct lcore_conf *conf,
459 	       struct rte_mbuf *pkts_burst[],
460 	       uint64_t total_pkts)
461 {
462 	unsigned i, portid, nb_rx, nb_tx;
463 	uint64_t diff_tsc, cur_tsc;
464 
465 	printf("do tx measure\n");
466 	diff_tsc = 0;
467 	while (likely(!stop)) {
468 		for (i = 0; i < conf->nb_ports; i++) {
469 			portid = conf->portlist[i];
470 			nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
471 						 pkts_burst, MAX_PKT_BURST);
472 			if (unlikely(nb_rx == 0)) {
473 				idle++;
474 				continue;
475 			}
476 
477 			count += nb_rx;
478 
479 			cur_tsc = rte_rdtsc();
480 			nb_tx = rte_eth_tx_burst(portid, 0, pkts_burst, nb_rx);
481 			if (unlikely(nb_tx < nb_rx)) {
482 				drop += (nb_rx - nb_tx);
483 				do {
484 					rte_pktmbuf_free(pkts_burst[nb_tx]);
485 				} while (++nb_tx < nb_rx);
486 			}
487 			diff_tsc += rte_rdtsc() - cur_tsc;
488 		}
489 		if (unlikely(count >= total_pkts))
490 			break;
491 	}
492 
493 	return diff_tsc;
494 }
495 
496 /* main processing loop */
497 static int
498 main_loop(__rte_unused void *args)
499 {
500 #define PACKET_SIZE 64
501 #define FRAME_GAP 12
502 #define MAC_PREAMBLE 8
503 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
504 	unsigned lcore_id;
505 	unsigned i, portid, nb_rx = 0, nb_tx = 0;
506 	struct lcore_conf *conf;
507 	int pkt_per_port;
508 	uint64_t diff_tsc;
509 	uint64_t packets_per_second, total_packets;
510 
511 	lcore_id = rte_lcore_id();
512 	conf = &lcore_conf[lcore_id];
513 	if (conf->status != LCORE_USED)
514 		return 0;
515 
516 	pkt_per_port = MAX_TRAFFIC_BURST;
517 
518 	int idx = 0;
519 	for (i = 0; i < conf->nb_ports; i++) {
520 		int num = pkt_per_port;
521 		portid = conf->portlist[i];
522 		printf("inject %d packet to port %d\n", num, portid);
523 		while (num) {
524 			nb_tx = RTE_MIN(MAX_PKT_BURST, num);
525 			nb_tx = rte_eth_tx_burst(portid, 0,
526 						&tx_burst[idx], nb_tx);
527 			num -= nb_tx;
528 			idx += nb_tx;
529 		}
530 	}
531 	printf("Total packets inject to prime ports = %u\n", idx);
532 
533 	packets_per_second = (link_mbps * 1000 * 1000) /
534 		((PACKET_SIZE + FRAME_GAP + MAC_PREAMBLE) * CHAR_BIT);
535 	printf("Each port will do %"PRIu64" packets per second\n",
536 	       packets_per_second);
537 
538 	total_packets = RTE_TEST_DURATION * conf->nb_ports * packets_per_second;
539 	printf("Test will stop after at least %"PRIu64" packets received\n",
540 		+ total_packets);
541 
542 	diff_tsc = do_measure(conf, pkts_burst, total_packets);
543 
544 	for (i = 0; i < conf->nb_ports; i++) {
545 		portid = conf->portlist[i];
546 		int nb_free = pkt_per_port;
547 		do { /* dry out */
548 			nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
549 						 pkts_burst, MAX_PKT_BURST);
550 			nb_tx = 0;
551 			while (nb_tx < nb_rx)
552 				rte_pktmbuf_free(pkts_burst[nb_tx++]);
553 			nb_free -= nb_rx;
554 		} while (nb_free != 0);
555 		printf("free %d mbuf left in port %u\n", pkt_per_port, portid);
556 	}
557 
558 	if (count == 0)
559 		return -1;
560 
561 	printf("%"PRIu64" packet, %"PRIu64" drop, %"PRIu64" idle\n",
562 	       count, drop, idle);
563 	printf("Result: %"PRIu64" cycles per packet\n", diff_tsc / count);
564 
565 	return 0;
566 }
567 
568 rte_atomic64_t start;
569 
570 static inline int
571 poll_burst(void *args)
572 {
573 #define MAX_IDLE           (10000)
574 	unsigned lcore_id;
575 	struct rte_mbuf **pkts_burst;
576 	uint64_t diff_tsc, cur_tsc;
577 	uint16_t next[RTE_MAX_ETHPORTS];
578 	struct lcore_conf *conf;
579 	uint32_t pkt_per_port = *((uint32_t *)args);
580 	unsigned i, portid, nb_rx = 0;
581 	uint64_t total;
582 	uint64_t timeout = MAX_IDLE;
583 
584 	lcore_id = rte_lcore_id();
585 	conf = &lcore_conf[lcore_id];
586 	if (conf->status != LCORE_USED)
587 		return 0;
588 
589 	total = pkt_per_port * conf->nb_ports;
590 	printf("start to receive total expect %"PRIu64"\n", total);
591 
592 	pkts_burst = (struct rte_mbuf **)
593 		rte_calloc_socket("poll_burst",
594 				  total, sizeof(void *),
595 				  CACHE_LINE_SIZE, conf->socketid);
596 	if (!pkts_burst)
597 		return -1;
598 
599 	for (i = 0; i < conf->nb_ports; i++) {
600 		portid = conf->portlist[i];
601 		next[portid] = i * pkt_per_port;
602 	}
603 
604 	while (!rte_atomic64_read(&start))
605 		;
606 
607 	cur_tsc = rte_rdtsc();
608 	while (total) {
609 		for (i = 0; i < conf->nb_ports; i++) {
610 			portid = conf->portlist[i];
611 			nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
612 						 &pkts_burst[next[portid]],
613 						 MAX_PKT_BURST);
614 			if (unlikely(nb_rx == 0)) {
615 				timeout--;
616 				if (unlikely(timeout == 0))
617 					goto timeout;
618 				continue;
619 			}
620 			next[portid] += nb_rx;
621 			total -= nb_rx;
622 		}
623 	}
624 timeout:
625 	diff_tsc = rte_rdtsc() - cur_tsc;
626 
627 	printf("%"PRIu64" packets lost, IDLE %"PRIu64" times\n",
628 	       total, MAX_IDLE - timeout);
629 
630 	/* clean up */
631 	total = pkt_per_port * conf->nb_ports - total;
632 	for (i = 0; i < total; i++)
633 		rte_pktmbuf_free(pkts_burst[i]);
634 
635 	rte_free(pkts_burst);
636 
637 	return diff_tsc / total;
638 }
639 
640 static int
641 exec_burst(uint32_t flags, int lcore)
642 {
643 	unsigned i, portid, nb_tx = 0;
644 	struct lcore_conf *conf;
645 	uint32_t pkt_per_port;
646 	int num, idx = 0;
647 	int diff_tsc;
648 
649 	conf = &lcore_conf[lcore];
650 
651 	pkt_per_port = MAX_TRAFFIC_BURST;
652 	num = pkt_per_port;
653 
654 	rte_atomic64_init(&start);
655 
656 	/* start polling thread, but not actually poll yet */
657 	rte_eal_remote_launch(poll_burst,
658 			      (void *)&pkt_per_port, lcore);
659 
660 	/* Only when polling first */
661 	if (flags == SC_BURST_POLL_FIRST)
662 		rte_atomic64_set(&start, 1);
663 
664 	/* start xmit */
665 	while (num) {
666 		nb_tx = RTE_MIN(MAX_PKT_BURST, num);
667 		for (i = 0; i < conf->nb_ports; i++) {
668 			portid = conf->portlist[i];
669 			rte_eth_tx_burst(portid, 0,
670 					 &tx_burst[idx], nb_tx);
671 			idx += nb_tx;
672 		}
673 		num -= nb_tx;
674 	}
675 
676 	sleep(5);
677 
678 	/* only when polling second  */
679 	if (flags == SC_BURST_XMIT_FIRST)
680 		rte_atomic64_set(&start, 1);
681 
682 	/* wait for polling finished */
683 	diff_tsc = rte_eal_wait_lcore(lcore);
684 	if (diff_tsc < 0)
685 		return -1;
686 
687 	printf("Result: %d cycles per packet\n", diff_tsc);
688 
689 	return 0;
690 }
691 
692 static int
693 test_pmd_perf(void)
694 {
695 	uint16_t nb_ports, num, nb_lcores, slave_id = (uint16_t)-1;
696 	uint16_t nb_rxd = MAX_TRAFFIC_BURST;
697 	uint16_t nb_txd = MAX_TRAFFIC_BURST;
698 	uint16_t portid;
699 	uint16_t nb_rx_queue = 1, nb_tx_queue = 1;
700 	int socketid = -1;
701 	int ret;
702 
703 	printf("Start PMD RXTX cycles cost test.\n");
704 
705 	signal(SIGUSR1, signal_handler);
706 	signal(SIGUSR2, signal_handler);
707 
708 	nb_ports = rte_eth_dev_count();
709 	if (nb_ports < NB_ETHPORTS_USED) {
710 		printf("At least %u port(s) used for perf. test\n",
711 		       NB_ETHPORTS_USED);
712 		return -1;
713 	}
714 
715 	if (nb_ports > RTE_MAX_ETHPORTS)
716 		nb_ports = RTE_MAX_ETHPORTS;
717 
718 	nb_lcores = rte_lcore_count();
719 
720 	memset(lcore_conf, 0, sizeof(lcore_conf));
721 	init_lcores();
722 
723 	init_mbufpool(NB_MBUF);
724 
725 	if (sc_flag == SC_CONTINUOUS) {
726 		nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
727 		nb_txd = RTE_TEST_TX_DESC_DEFAULT;
728 	}
729 	printf("CONFIG RXD=%d TXD=%d\n", nb_rxd, nb_txd);
730 
731 	reset_count();
732 	num = 0;
733 	for (portid = 0; portid < nb_ports; portid++) {
734 		if (socketid == -1) {
735 			socketid = rte_eth_dev_socket_id(portid);
736 			slave_id = alloc_lcore(socketid);
737 			if (slave_id == (uint16_t)-1) {
738 				printf("No avail lcore to run test\n");
739 				return -1;
740 			}
741 			printf("Performance test runs on lcore %u socket %u\n",
742 			       slave_id, socketid);
743 		}
744 
745 		if (socketid != rte_eth_dev_socket_id(portid)) {
746 			printf("Skip port %d\n", portid);
747 			continue;
748 		}
749 
750 		/* port configure */
751 		ret = rte_eth_dev_configure(portid, nb_rx_queue,
752 					    nb_tx_queue, &port_conf);
753 		if (ret < 0)
754 			rte_exit(EXIT_FAILURE,
755 				"Cannot configure device: err=%d, port=%d\n",
756 				 ret, portid);
757 
758 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
759 		printf("Port %u ", portid);
760 		print_ethaddr("Address:", &ports_eth_addr[portid]);
761 		printf("\n");
762 
763 		/* tx queue setup */
764 		ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
765 					     socketid, &tx_conf);
766 		if (ret < 0)
767 			rte_exit(EXIT_FAILURE,
768 				"rte_eth_tx_queue_setup: err=%d, "
769 				"port=%d\n", ret, portid);
770 
771 		/* rx queue steup */
772 		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
773 						socketid, &rx_conf,
774 						mbufpool[socketid]);
775 		if (ret < 0)
776 			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
777 				 "port=%d\n", ret, portid);
778 
779 		/* Start device */
780 		stop = 0;
781 		ret = rte_eth_dev_start(portid);
782 		if (ret < 0)
783 			rte_exit(EXIT_FAILURE,
784 				"rte_eth_dev_start: err=%d, port=%d\n",
785 				ret, portid);
786 
787 		/* always eanble promiscuous */
788 		rte_eth_promiscuous_enable(portid);
789 
790 		lcore_conf[slave_id].portlist[num++] = portid;
791 		lcore_conf[slave_id].nb_ports++;
792 	}
793 	check_all_ports_link_status(nb_ports, RTE_PORT_ALL);
794 
795 	if (tx_burst == NULL) {
796 		tx_burst = (struct rte_mbuf **)
797 			rte_calloc_socket("tx_buff",
798 					  MAX_TRAFFIC_BURST * nb_ports,
799 					  sizeof(void *),
800 					  CACHE_LINE_SIZE, socketid);
801 		if (!tx_burst)
802 			return -1;
803 	}
804 
805 	init_traffic(mbufpool[socketid],
806 		     tx_burst, MAX_TRAFFIC_BURST * nb_ports);
807 
808 	printf("Generate %d packets @socket %d\n",
809 	       MAX_TRAFFIC_BURST * nb_ports, socketid);
810 
811 	if (sc_flag == SC_CONTINUOUS) {
812 		/* do both rxtx by default */
813 		if (NULL == do_measure)
814 			do_measure = measure_rxtx;
815 
816 		rte_eal_remote_launch(main_loop, NULL, slave_id);
817 
818 		if (rte_eal_wait_lcore(slave_id) < 0)
819 			return -1;
820 	} else if (sc_flag == SC_BURST_POLL_FIRST ||
821 		   sc_flag == SC_BURST_XMIT_FIRST)
822 		exec_burst(sc_flag, slave_id);
823 
824 	/* port tear down */
825 	for (portid = 0; portid < nb_ports; portid++) {
826 		if (socketid != rte_eth_dev_socket_id(portid))
827 			continue;
828 
829 		rte_eth_dev_stop(portid);
830 	}
831 
832 	return 0;
833 }
834 
835 int
836 test_set_rxtx_conf(cmdline_fixed_string_t mode)
837 {
838 	printf("mode switch to %s\n", mode);
839 
840 	if (!strcmp(mode, "vector")) {
841 		/* vector rx, tx */
842 		tx_conf.txq_flags = 0xf01;
843 		tx_conf.tx_rs_thresh = 32;
844 		tx_conf.tx_free_thresh = 32;
845 		port_conf.rxmode.hw_ip_checksum = 0;
846 		port_conf.rxmode.enable_scatter = 0;
847 		return 0;
848 	} else if (!strcmp(mode, "scalar")) {
849 		/* bulk alloc rx, simple tx */
850 		tx_conf.txq_flags = 0xf01;
851 		tx_conf.tx_rs_thresh = 128;
852 		tx_conf.tx_free_thresh = 128;
853 		port_conf.rxmode.hw_ip_checksum = 1;
854 		port_conf.rxmode.enable_scatter = 0;
855 		return 0;
856 	} else if (!strcmp(mode, "hybrid")) {
857 		/* bulk alloc rx, vector tx
858 		 * when vec macro not define,
859 		 * using the same rx/tx as scalar
860 		 */
861 		tx_conf.txq_flags = 0xf01;
862 		tx_conf.tx_rs_thresh = 32;
863 		tx_conf.tx_free_thresh = 32;
864 		port_conf.rxmode.hw_ip_checksum = 1;
865 		port_conf.rxmode.enable_scatter = 0;
866 		return 0;
867 	} else if (!strcmp(mode, "full")) {
868 		/* full feature rx,tx pair */
869 		tx_conf.txq_flags = 0x0;   /* must condition */
870 		tx_conf.tx_rs_thresh = 32;
871 		tx_conf.tx_free_thresh = 32;
872 		port_conf.rxmode.hw_ip_checksum = 0;
873 		port_conf.rxmode.enable_scatter = 1; /* must condition */
874 		return 0;
875 	}
876 
877 	return -1;
878 }
879 
880 int
881 test_set_rxtx_anchor(cmdline_fixed_string_t type)
882 {
883 	printf("type switch to %s\n", type);
884 
885 	if (!strcmp(type, "rxtx")) {
886 		do_measure = measure_rxtx;
887 		return 0;
888 	} else if (!strcmp(type, "rxonly")) {
889 		do_measure = measure_rxonly;
890 		return 0;
891 	} else if (!strcmp(type, "txonly")) {
892 		do_measure = measure_txonly;
893 		return 0;
894 	}
895 
896 	return -1;
897 }
898 
899 int
900 test_set_rxtx_sc(cmdline_fixed_string_t type)
901 {
902 	printf("stream control switch to %s\n", type);
903 
904 	if (!strcmp(type, "continuous")) {
905 		sc_flag = SC_CONTINUOUS;
906 		return 0;
907 	} else if (!strcmp(type, "poll_before_xmit")) {
908 		sc_flag = SC_BURST_POLL_FIRST;
909 		return 0;
910 	} else if (!strcmp(type, "poll_after_xmit")) {
911 		sc_flag = SC_BURST_XMIT_FIRST;
912 		return 0;
913 	}
914 
915 	return -1;
916 }
917 
918 static struct test_command pmd_perf_cmd = {
919 	.command = "pmd_perf_autotest",
920 	.callback = test_pmd_perf,
921 };
922 REGISTER_TEST_COMMAND(pmd_perf_cmd);
923