xref: /dpdk/examples/l3fwd-power/main.c (revision a103a97e7191179ad6a451ce85182df2ecb10c26)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2016 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 #include <stdio.h>
35 #include <stdlib.h>
36 #include <stdint.h>
37 #include <inttypes.h>
38 #include <sys/types.h>
39 #include <string.h>
40 #include <sys/queue.h>
41 #include <stdarg.h>
42 #include <errno.h>
43 #include <getopt.h>
44 #include <unistd.h>
45 #include <signal.h>
46 
47 #include <rte_common.h>
48 #include <rte_byteorder.h>
49 #include <rte_log.h>
50 #include <rte_malloc.h>
51 #include <rte_memory.h>
52 #include <rte_memcpy.h>
53 #include <rte_memzone.h>
54 #include <rte_eal.h>
55 #include <rte_launch.h>
56 #include <rte_atomic.h>
57 #include <rte_cycles.h>
58 #include <rte_prefetch.h>
59 #include <rte_lcore.h>
60 #include <rte_per_lcore.h>
61 #include <rte_branch_prediction.h>
62 #include <rte_interrupts.h>
63 #include <rte_pci.h>
64 #include <rte_random.h>
65 #include <rte_debug.h>
66 #include <rte_ether.h>
67 #include <rte_ethdev.h>
68 #include <rte_mempool.h>
69 #include <rte_mbuf.h>
70 #include <rte_ip.h>
71 #include <rte_tcp.h>
72 #include <rte_udp.h>
73 #include <rte_string_fns.h>
74 #include <rte_timer.h>
75 #include <rte_power.h>
76 #include <rte_spinlock.h>
77 
78 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
79 
80 #define MAX_PKT_BURST 32
81 
82 #define MIN_ZERO_POLL_COUNT 10
83 
84 /* around 100ms at 2 Ghz */
85 #define TIMER_RESOLUTION_CYCLES           200000000ULL
86 /* 100 ms interval */
87 #define TIMER_NUMBER_PER_SECOND           10
88 /* 100000 us */
89 #define SCALING_PERIOD                    (1000000/TIMER_NUMBER_PER_SECOND)
90 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
91 
92 #define APP_LOOKUP_EXACT_MATCH          0
93 #define APP_LOOKUP_LPM                  1
94 #define DO_RFC_1812_CHECKS
95 
96 #ifndef APP_LOOKUP_METHOD
97 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
98 #endif
99 
100 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
101 #include <rte_hash.h>
102 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
103 #include <rte_lpm.h>
104 #else
105 #error "APP_LOOKUP_METHOD set to incorrect value"
106 #endif
107 
108 #ifndef IPv6_BYTES
109 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
110                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
111 #define IPv6_BYTES(addr) \
112 	addr[0],  addr[1], addr[2],  addr[3], \
113 	addr[4],  addr[5], addr[6],  addr[7], \
114 	addr[8],  addr[9], addr[10], addr[11],\
115 	addr[12], addr[13],addr[14], addr[15]
116 #endif
117 
118 #define MAX_JUMBO_PKT_LEN  9600
119 
120 #define IPV6_ADDR_LEN 16
121 
122 #define MEMPOOL_CACHE_SIZE 256
123 
124 /*
125  * This expression is used to calculate the number of mbufs needed depending on
126  * user input, taking into account memory for rx and tx hardware rings, cache
127  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
128  * NB_MBUF never goes below a minimum value of 8192.
129  */
130 
131 #define NB_MBUF RTE_MAX	( \
132 	(nb_ports*nb_rx_queue*nb_rxd + \
133 	nb_ports*nb_lcores*MAX_PKT_BURST + \
134 	nb_ports*n_tx_queue*nb_txd + \
135 	nb_lcores*MEMPOOL_CACHE_SIZE), \
136 	(unsigned)8192)
137 
138 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
139 
140 #define NB_SOCKETS 8
141 
142 /* Configure how many packets ahead to prefetch, when reading packets */
143 #define PREFETCH_OFFSET	3
144 
145 /*
146  * Configurable number of RX/TX ring descriptors
147  */
148 #define RTE_TEST_RX_DESC_DEFAULT 512
149 #define RTE_TEST_TX_DESC_DEFAULT 512
150 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
151 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
152 
153 /* ethernet addresses of ports */
154 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
155 
156 /* ethernet addresses of ports */
157 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
158 
159 /* mask of enabled ports */
160 static uint32_t enabled_port_mask = 0;
161 /* Ports set in promiscuous mode off by default. */
162 static int promiscuous_on = 0;
163 /* NUMA is enabled by default. */
164 static int numa_on = 1;
165 static int parse_ptype; /**< Parse packet type using rx callback, and */
166 			/**< disabled by default */
167 
168 enum freq_scale_hint_t
169 {
170 	FREQ_LOWER    =      -1,
171 	FREQ_CURRENT  =       0,
172 	FREQ_HIGHER   =       1,
173 	FREQ_HIGHEST  =       2
174 };
175 
176 struct lcore_rx_queue {
177 	uint16_t port_id;
178 	uint8_t queue_id;
179 	enum freq_scale_hint_t freq_up_hint;
180 	uint32_t zero_rx_packet_count;
181 	uint32_t idle_hint;
182 } __rte_cache_aligned;
183 
184 #define MAX_RX_QUEUE_PER_LCORE 16
185 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
186 #define MAX_RX_QUEUE_PER_PORT 128
187 
188 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
189 
190 
191 #define MAX_LCORE_PARAMS 1024
192 struct lcore_params {
193 	uint16_t port_id;
194 	uint8_t queue_id;
195 	uint8_t lcore_id;
196 } __rte_cache_aligned;
197 
198 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
199 static struct lcore_params lcore_params_array_default[] = {
200 	{0, 0, 2},
201 	{0, 1, 2},
202 	{0, 2, 2},
203 	{1, 0, 2},
204 	{1, 1, 2},
205 	{1, 2, 2},
206 	{2, 0, 2},
207 	{3, 0, 3},
208 	{3, 1, 3},
209 };
210 
211 static struct lcore_params * lcore_params = lcore_params_array_default;
212 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
213 				sizeof(lcore_params_array_default[0]);
214 
215 static struct rte_eth_conf port_conf = {
216 	.rxmode = {
217 		.mq_mode        = ETH_MQ_RX_RSS,
218 		.max_rx_pkt_len = ETHER_MAX_LEN,
219 		.split_hdr_size = 0,
220 		.header_split   = 0, /**< Header Split disabled */
221 		.hw_ip_checksum = 1, /**< IP checksum offload enabled */
222 		.hw_vlan_filter = 0, /**< VLAN filtering disabled */
223 		.jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
224 		.hw_strip_crc   = 1, /**< CRC stripped by hardware */
225 	},
226 	.rx_adv_conf = {
227 		.rss_conf = {
228 			.rss_key = NULL,
229 			.rss_hf = ETH_RSS_UDP,
230 		},
231 	},
232 	.txmode = {
233 		.mq_mode = ETH_MQ_TX_NONE,
234 	},
235 	.intr_conf = {
236 		.lsc = 1,
237 		.rxq = 1,
238 	},
239 };
240 
241 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
242 
243 
244 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
245 
246 #ifdef RTE_ARCH_X86
247 #include <rte_hash_crc.h>
248 #define DEFAULT_HASH_FUNC       rte_hash_crc
249 #else
250 #include <rte_jhash.h>
251 #define DEFAULT_HASH_FUNC       rte_jhash
252 #endif
253 
254 struct ipv4_5tuple {
255 	uint32_t ip_dst;
256 	uint32_t ip_src;
257 	uint16_t port_dst;
258 	uint16_t port_src;
259 	uint8_t  proto;
260 } __attribute__((__packed__));
261 
262 struct ipv6_5tuple {
263 	uint8_t  ip_dst[IPV6_ADDR_LEN];
264 	uint8_t  ip_src[IPV6_ADDR_LEN];
265 	uint16_t port_dst;
266 	uint16_t port_src;
267 	uint8_t  proto;
268 } __attribute__((__packed__));
269 
270 struct ipv4_l3fwd_route {
271 	struct ipv4_5tuple key;
272 	uint8_t if_out;
273 };
274 
275 struct ipv6_l3fwd_route {
276 	struct ipv6_5tuple key;
277 	uint8_t if_out;
278 };
279 
280 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
281 	{{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
282 	{{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
283 	{{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
284 	{{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
285 };
286 
287 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
288 	{
289 		{
290 			{0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
291 			 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
292 			{0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
293 			 0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
294 			 1, 10, IPPROTO_UDP
295 		}, 4
296 	},
297 };
298 
299 typedef struct rte_hash lookup_struct_t;
300 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
301 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
302 
303 #define L3FWD_HASH_ENTRIES	1024
304 
305 #define IPV4_L3FWD_NUM_ROUTES \
306 	(sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
307 
308 #define IPV6_L3FWD_NUM_ROUTES \
309 	(sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
310 
311 static uint16_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
312 static uint16_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
313 #endif
314 
315 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
316 struct ipv4_l3fwd_route {
317 	uint32_t ip;
318 	uint8_t  depth;
319 	uint8_t  if_out;
320 };
321 
322 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
323 	{IPv4(1,1,1,0), 24, 0},
324 	{IPv4(2,1,1,0), 24, 1},
325 	{IPv4(3,1,1,0), 24, 2},
326 	{IPv4(4,1,1,0), 24, 3},
327 	{IPv4(5,1,1,0), 24, 4},
328 	{IPv4(6,1,1,0), 24, 5},
329 	{IPv4(7,1,1,0), 24, 6},
330 	{IPv4(8,1,1,0), 24, 7},
331 };
332 
333 #define IPV4_L3FWD_NUM_ROUTES \
334 	(sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
335 
336 #define IPV4_L3FWD_LPM_MAX_RULES     1024
337 
338 typedef struct rte_lpm lookup_struct_t;
339 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
340 #endif
341 
342 struct lcore_conf {
343 	uint16_t n_rx_queue;
344 	struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
345 	uint16_t n_tx_port;
346 	uint16_t tx_port_id[RTE_MAX_ETHPORTS];
347 	uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
348 	struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
349 	lookup_struct_t * ipv4_lookup_struct;
350 	lookup_struct_t * ipv6_lookup_struct;
351 } __rte_cache_aligned;
352 
353 struct lcore_stats {
354 	/* total sleep time in ms since last frequency scaling down */
355 	uint32_t sleep_time;
356 	/* number of long sleep recently */
357 	uint32_t nb_long_sleep;
358 	/* freq. scaling up trend */
359 	uint32_t trend;
360 	/* total packet processed recently */
361 	uint64_t nb_rx_processed;
362 	/* total iterations looped recently */
363 	uint64_t nb_iteration_looped;
364 	uint32_t padding[9];
365 } __rte_cache_aligned;
366 
367 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
368 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
369 static struct rte_timer power_timers[RTE_MAX_LCORE];
370 
371 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
372 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
373 		unsigned int lcore_id, uint16_t port_id, uint16_t queue_id);
374 
375 /* exit signal handler */
376 static void
377 signal_exit_now(int sigtype)
378 {
379 	unsigned lcore_id;
380 	unsigned int portid, nb_ports;
381 	int ret;
382 
383 	if (sigtype == SIGINT) {
384 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
385 			if (rte_lcore_is_enabled(lcore_id) == 0)
386 				continue;
387 
388 			/* init power management library */
389 			ret = rte_power_exit(lcore_id);
390 			if (ret)
391 				rte_exit(EXIT_FAILURE, "Power management "
392 					"library de-initialization failed on "
393 							"core%u\n", lcore_id);
394 		}
395 
396 		nb_ports = rte_eth_dev_count();
397 		for (portid = 0; portid < nb_ports; portid++) {
398 			if ((enabled_port_mask & (1 << portid)) == 0)
399 				continue;
400 
401 			rte_eth_dev_stop(portid);
402 			rte_eth_dev_close(portid);
403 		}
404 	}
405 
406 	rte_exit(EXIT_SUCCESS, "User forced exit\n");
407 }
408 
409 /*  Freqency scale down timer callback */
410 static void
411 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
412 			  __attribute__((unused)) void *arg)
413 {
414 	uint64_t hz;
415 	float sleep_time_ratio;
416 	unsigned lcore_id = rte_lcore_id();
417 
418 	/* accumulate total execution time in us when callback is invoked */
419 	sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
420 					(float)SCALING_PERIOD;
421 	/**
422 	 * check whether need to scale down frequency a step if it sleep a lot.
423 	 */
424 	if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
425 		if (rte_power_freq_down)
426 			rte_power_freq_down(lcore_id);
427 	}
428 	else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
429 		stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
430 		/**
431 		 * scale down a step if average packet per iteration less
432 		 * than expectation.
433 		 */
434 		if (rte_power_freq_down)
435 			rte_power_freq_down(lcore_id);
436 	}
437 
438 	/**
439 	 * initialize another timer according to current frequency to ensure
440 	 * timer interval is relatively fixed.
441 	 */
442 	hz = rte_get_timer_hz();
443 	rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
444 				SINGLE, lcore_id, power_timer_cb, NULL);
445 
446 	stats[lcore_id].nb_rx_processed = 0;
447 	stats[lcore_id].nb_iteration_looped = 0;
448 
449 	stats[lcore_id].sleep_time = 0;
450 }
451 
452 /* Enqueue a single packet, and send burst if queue is filled */
453 static inline int
454 send_single_packet(struct rte_mbuf *m, uint8_t port)
455 {
456 	uint32_t lcore_id;
457 	struct lcore_conf *qconf;
458 
459 	lcore_id = rte_lcore_id();
460 	qconf = &lcore_conf[lcore_id];
461 
462 	rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
463 			qconf->tx_buffer[port], m);
464 
465 	return 0;
466 }
467 
468 #ifdef DO_RFC_1812_CHECKS
469 static inline int
470 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
471 {
472 	/* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
473 	/*
474 	 * 1. The packet length reported by the Link Layer must be large
475 	 * enough to hold the minimum length legal IP datagram (20 bytes).
476 	 */
477 	if (link_len < sizeof(struct ipv4_hdr))
478 		return -1;
479 
480 	/* 2. The IP checksum must be correct. */
481 	/* this is checked in H/W */
482 
483 	/*
484 	 * 3. The IP version number must be 4. If the version number is not 4
485 	 * then the packet may be another version of IP, such as IPng or
486 	 * ST-II.
487 	 */
488 	if (((pkt->version_ihl) >> 4) != 4)
489 		return -3;
490 	/*
491 	 * 4. The IP header length field must be large enough to hold the
492 	 * minimum length legal IP datagram (20 bytes = 5 words).
493 	 */
494 	if ((pkt->version_ihl & 0xf) < 5)
495 		return -4;
496 
497 	/*
498 	 * 5. The IP total length field must be large enough to hold the IP
499 	 * datagram header, whose length is specified in the IP header length
500 	 * field.
501 	 */
502 	if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
503 		return -5;
504 
505 	return 0;
506 }
507 #endif
508 
509 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
510 static void
511 print_ipv4_key(struct ipv4_5tuple key)
512 {
513 	printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
514 		"proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
515 				key.port_dst, key.port_src, key.proto);
516 }
517 static void
518 print_ipv6_key(struct ipv6_5tuple key)
519 {
520 	printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
521 	        "port dst = %d, port src = %d, proto = %d\n",
522 	        IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
523 	        key.port_dst, key.port_src, key.proto);
524 }
525 
526 static inline uint16_t
527 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
528 		lookup_struct_t * ipv4_l3fwd_lookup_struct)
529 {
530 	struct ipv4_5tuple key;
531 	struct tcp_hdr *tcp;
532 	struct udp_hdr *udp;
533 	int ret = 0;
534 
535 	key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
536 	key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
537 	key.proto = ipv4_hdr->next_proto_id;
538 
539 	switch (ipv4_hdr->next_proto_id) {
540 	case IPPROTO_TCP:
541 		tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
542 					sizeof(struct ipv4_hdr));
543 		key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
544 		key.port_src = rte_be_to_cpu_16(tcp->src_port);
545 		break;
546 
547 	case IPPROTO_UDP:
548 		udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
549 					sizeof(struct ipv4_hdr));
550 		key.port_dst = rte_be_to_cpu_16(udp->dst_port);
551 		key.port_src = rte_be_to_cpu_16(udp->src_port);
552 		break;
553 
554 	default:
555 		key.port_dst = 0;
556 		key.port_src = 0;
557 		break;
558 	}
559 
560 	/* Find destination port */
561 	ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
562 	return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
563 }
564 
565 static inline uint16_t
566 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint16_t portid,
567 			lookup_struct_t *ipv6_l3fwd_lookup_struct)
568 {
569 	struct ipv6_5tuple key;
570 	struct tcp_hdr *tcp;
571 	struct udp_hdr *udp;
572 	int ret = 0;
573 
574 	memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
575 	memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
576 
577 	key.proto = ipv6_hdr->proto;
578 
579 	switch (ipv6_hdr->proto) {
580 	case IPPROTO_TCP:
581 		tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
582 					sizeof(struct ipv6_hdr));
583 		key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
584 		key.port_src = rte_be_to_cpu_16(tcp->src_port);
585 		break;
586 
587 	case IPPROTO_UDP:
588 		udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
589 					sizeof(struct ipv6_hdr));
590 		key.port_dst = rte_be_to_cpu_16(udp->dst_port);
591 		key.port_src = rte_be_to_cpu_16(udp->src_port);
592 		break;
593 
594 	default:
595 		key.port_dst = 0;
596 		key.port_src = 0;
597 		break;
598 	}
599 
600 	/* Find destination port */
601 	ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
602 	return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
603 }
604 #endif
605 
606 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
607 static inline uint16_t
608 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
609 		lookup_struct_t *ipv4_l3fwd_lookup_struct)
610 {
611 	uint32_t next_hop;
612 
613 	return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
614 			rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
615 			next_hop : portid);
616 }
617 #endif
618 
619 static inline void
620 parse_ptype_one(struct rte_mbuf *m)
621 {
622 	struct ether_hdr *eth_hdr;
623 	uint32_t packet_type = RTE_PTYPE_UNKNOWN;
624 	uint16_t ether_type;
625 
626 	eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
627 	ether_type = eth_hdr->ether_type;
628 	if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
629 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
630 	else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
631 		packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
632 
633 	m->packet_type = packet_type;
634 }
635 
636 static uint16_t
637 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
638 	       struct rte_mbuf *pkts[], uint16_t nb_pkts,
639 	       uint16_t max_pkts __rte_unused,
640 	       void *user_param __rte_unused)
641 {
642 	unsigned int i;
643 
644 	for (i = 0; i < nb_pkts; ++i)
645 		parse_ptype_one(pkts[i]);
646 
647 	return nb_pkts;
648 }
649 
650 static int
651 add_cb_parse_ptype(uint16_t portid, uint16_t queueid)
652 {
653 	printf("Port %d: softly parse packet type info\n", portid);
654 	if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL))
655 		return 0;
656 
657 	printf("Failed to add rx callback: port=%d\n", portid);
658 	return -1;
659 }
660 
661 static inline void
662 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
663 				struct lcore_conf *qconf)
664 {
665 	struct ether_hdr *eth_hdr;
666 	struct ipv4_hdr *ipv4_hdr;
667 	void *d_addr_bytes;
668 	uint16_t dst_port;
669 
670 	eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
671 
672 	if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
673 		/* Handle IPv4 headers.*/
674 		ipv4_hdr =
675 			rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
676 						sizeof(struct ether_hdr));
677 
678 #ifdef DO_RFC_1812_CHECKS
679 		/* Check to make sure the packet is valid (RFC1812) */
680 		if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
681 			rte_pktmbuf_free(m);
682 			return;
683 		}
684 #endif
685 
686 		dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
687 					qconf->ipv4_lookup_struct);
688 		if (dst_port >= RTE_MAX_ETHPORTS ||
689 				(enabled_port_mask & 1 << dst_port) == 0)
690 			dst_port = portid;
691 
692 		/* 02:00:00:00:00:xx */
693 		d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
694 		*((uint64_t *)d_addr_bytes) =
695 			0x000000000002 + ((uint64_t)dst_port << 40);
696 
697 #ifdef DO_RFC_1812_CHECKS
698 		/* Update time to live and header checksum */
699 		--(ipv4_hdr->time_to_live);
700 		++(ipv4_hdr->hdr_checksum);
701 #endif
702 
703 		/* src addr */
704 		ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
705 
706 		send_single_packet(m, dst_port);
707 	} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
708 		/* Handle IPv6 headers.*/
709 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
710 		struct ipv6_hdr *ipv6_hdr;
711 
712 		ipv6_hdr =
713 			rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
714 						sizeof(struct ether_hdr));
715 
716 		dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
717 					qconf->ipv6_lookup_struct);
718 
719 		if (dst_port >= RTE_MAX_ETHPORTS ||
720 				(enabled_port_mask & 1 << dst_port) == 0)
721 			dst_port = portid;
722 
723 		/* 02:00:00:00:00:xx */
724 		d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
725 		*((uint64_t *)d_addr_bytes) =
726 			0x000000000002 + ((uint64_t)dst_port << 40);
727 
728 		/* src addr */
729 		ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
730 
731 		send_single_packet(m, dst_port);
732 #else
733 		/* We don't currently handle IPv6 packets in LPM mode. */
734 		rte_pktmbuf_free(m);
735 #endif
736 	} else
737 		rte_pktmbuf_free(m);
738 
739 }
740 
741 #define MINIMUM_SLEEP_TIME         1
742 #define SUSPEND_THRESHOLD          300
743 
744 static inline uint32_t
745 power_idle_heuristic(uint32_t zero_rx_packet_count)
746 {
747 	/* If zero count is less than 100,  sleep 1us */
748 	if (zero_rx_packet_count < SUSPEND_THRESHOLD)
749 		return MINIMUM_SLEEP_TIME;
750 	/* If zero count is less than 1000, sleep 100 us which is the
751 		minimum latency switching from C3/C6 to C0
752 	*/
753 	else
754 		return SUSPEND_THRESHOLD;
755 
756 	return 0;
757 }
758 
759 static inline enum freq_scale_hint_t
760 power_freq_scaleup_heuristic(unsigned lcore_id,
761 			     uint16_t port_id,
762 			     uint16_t queue_id)
763 {
764 /**
765  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
766  * per iteration
767  */
768 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
769 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
770 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
771 #define FREQ_UP_TREND1_ACC   1
772 #define FREQ_UP_TREND2_ACC   100
773 #define FREQ_UP_THRESHOLD    10000
774 
775 	if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
776 			FREQ_GEAR3_RX_PACKET_THRESHOLD) > 0)) {
777 		stats[lcore_id].trend = 0;
778 		return FREQ_HIGHEST;
779 	} else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
780 			FREQ_GEAR2_RX_PACKET_THRESHOLD) > 0))
781 		stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
782 	else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
783 			FREQ_GEAR1_RX_PACKET_THRESHOLD) > 0))
784 		stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
785 
786 	if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
787 		stats[lcore_id].trend = 0;
788 		return FREQ_HIGHER;
789 	}
790 
791 	return FREQ_CURRENT;
792 }
793 
794 /**
795  * force polling thread sleep until one-shot rx interrupt triggers
796  * @param port_id
797  *  Port id.
798  * @param queue_id
799  *  Rx queue id.
800  * @return
801  *  0 on success
802  */
803 static int
804 sleep_until_rx_interrupt(int num)
805 {
806 	struct rte_epoll_event event[num];
807 	int n, i;
808 	uint16_t port_id;
809 	uint8_t queue_id;
810 	void *data;
811 
812 	RTE_LOG(INFO, L3FWD_POWER,
813 		"lcore %u sleeps until interrupt triggers\n",
814 		rte_lcore_id());
815 
816 	n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
817 	for (i = 0; i < n; i++) {
818 		data = event[i].epdata.data;
819 		port_id = ((uintptr_t)data) >> CHAR_BIT;
820 		queue_id = ((uintptr_t)data) &
821 			RTE_LEN2MASK(CHAR_BIT, uint8_t);
822 		rte_eth_dev_rx_intr_disable(port_id, queue_id);
823 		RTE_LOG(INFO, L3FWD_POWER,
824 			"lcore %u is waked up from rx interrupt on"
825 			" port %d queue %d\n",
826 			rte_lcore_id(), port_id, queue_id);
827 	}
828 
829 	return 0;
830 }
831 
832 static void turn_on_intr(struct lcore_conf *qconf)
833 {
834 	int i;
835 	struct lcore_rx_queue *rx_queue;
836 	uint8_t queue_id;
837 	uint16_t port_id;
838 
839 	for (i = 0; i < qconf->n_rx_queue; ++i) {
840 		rx_queue = &(qconf->rx_queue_list[i]);
841 		port_id = rx_queue->port_id;
842 		queue_id = rx_queue->queue_id;
843 
844 		rte_spinlock_lock(&(locks[port_id]));
845 		rte_eth_dev_rx_intr_enable(port_id, queue_id);
846 		rte_spinlock_unlock(&(locks[port_id]));
847 	}
848 }
849 
850 static int event_register(struct lcore_conf *qconf)
851 {
852 	struct lcore_rx_queue *rx_queue;
853 	uint8_t queueid;
854 	uint16_t portid;
855 	uint32_t data;
856 	int ret;
857 	int i;
858 
859 	for (i = 0; i < qconf->n_rx_queue; ++i) {
860 		rx_queue = &(qconf->rx_queue_list[i]);
861 		portid = rx_queue->port_id;
862 		queueid = rx_queue->queue_id;
863 		data = portid << CHAR_BIT | queueid;
864 
865 		ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
866 						RTE_EPOLL_PER_THREAD,
867 						RTE_INTR_EVENT_ADD,
868 						(void *)((uintptr_t)data));
869 		if (ret)
870 			return ret;
871 	}
872 
873 	return 0;
874 }
875 
876 /* main processing loop */
877 static int
878 main_loop(__attribute__((unused)) void *dummy)
879 {
880 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
881 	unsigned lcore_id;
882 	uint64_t prev_tsc, diff_tsc, cur_tsc;
883 	uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
884 	int i, j, nb_rx;
885 	uint8_t queueid;
886 	uint16_t portid;
887 	struct lcore_conf *qconf;
888 	struct lcore_rx_queue *rx_queue;
889 	enum freq_scale_hint_t lcore_scaleup_hint;
890 	uint32_t lcore_rx_idle_count = 0;
891 	uint32_t lcore_idle_hint = 0;
892 	int intr_en = 0;
893 
894 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
895 
896 	prev_tsc = 0;
897 
898 	lcore_id = rte_lcore_id();
899 	qconf = &lcore_conf[lcore_id];
900 
901 	if (qconf->n_rx_queue == 0) {
902 		RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
903 		return 0;
904 	}
905 
906 	RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
907 
908 	for (i = 0; i < qconf->n_rx_queue; i++) {
909 		portid = qconf->rx_queue_list[i].port_id;
910 		queueid = qconf->rx_queue_list[i].queue_id;
911 		RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
912 			"rxqueueid=%hhu\n", lcore_id, portid, queueid);
913 	}
914 
915 	/* add into event wait list */
916 	if (event_register(qconf) == 0)
917 		intr_en = 1;
918 	else
919 		RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
920 
921 	while (1) {
922 		stats[lcore_id].nb_iteration_looped++;
923 
924 		cur_tsc = rte_rdtsc();
925 		cur_tsc_power = cur_tsc;
926 
927 		/*
928 		 * TX burst queue drain
929 		 */
930 		diff_tsc = cur_tsc - prev_tsc;
931 		if (unlikely(diff_tsc > drain_tsc)) {
932 			for (i = 0; i < qconf->n_tx_port; ++i) {
933 				portid = qconf->tx_port_id[i];
934 				rte_eth_tx_buffer_flush(portid,
935 						qconf->tx_queue_id[portid],
936 						qconf->tx_buffer[portid]);
937 			}
938 			prev_tsc = cur_tsc;
939 		}
940 
941 		diff_tsc_power = cur_tsc_power - prev_tsc_power;
942 		if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
943 			rte_timer_manage();
944 			prev_tsc_power = cur_tsc_power;
945 		}
946 
947 start_rx:
948 		/*
949 		 * Read packet from RX queues
950 		 */
951 		lcore_scaleup_hint = FREQ_CURRENT;
952 		lcore_rx_idle_count = 0;
953 		for (i = 0; i < qconf->n_rx_queue; ++i) {
954 			rx_queue = &(qconf->rx_queue_list[i]);
955 			rx_queue->idle_hint = 0;
956 			portid = rx_queue->port_id;
957 			queueid = rx_queue->queue_id;
958 
959 			nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
960 								MAX_PKT_BURST);
961 
962 			stats[lcore_id].nb_rx_processed += nb_rx;
963 			if (unlikely(nb_rx == 0)) {
964 				/**
965 				 * no packet received from rx queue, try to
966 				 * sleep for a while forcing CPU enter deeper
967 				 * C states.
968 				 */
969 				rx_queue->zero_rx_packet_count++;
970 
971 				if (rx_queue->zero_rx_packet_count <=
972 							MIN_ZERO_POLL_COUNT)
973 					continue;
974 
975 				rx_queue->idle_hint = power_idle_heuristic(\
976 					rx_queue->zero_rx_packet_count);
977 				lcore_rx_idle_count++;
978 			} else {
979 				rx_queue->zero_rx_packet_count = 0;
980 
981 				/**
982 				 * do not scale up frequency immediately as
983 				 * user to kernel space communication is costly
984 				 * which might impact packet I/O for received
985 				 * packets.
986 				 */
987 				rx_queue->freq_up_hint =
988 					power_freq_scaleup_heuristic(lcore_id,
989 							portid, queueid);
990 			}
991 
992 			/* Prefetch first packets */
993 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
994 				rte_prefetch0(rte_pktmbuf_mtod(
995 						pkts_burst[j], void *));
996 			}
997 
998 			/* Prefetch and forward already prefetched packets */
999 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1000 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1001 						j + PREFETCH_OFFSET], void *));
1002 				l3fwd_simple_forward(pkts_burst[j], portid,
1003 								qconf);
1004 			}
1005 
1006 			/* Forward remaining prefetched packets */
1007 			for (; j < nb_rx; j++) {
1008 				l3fwd_simple_forward(pkts_burst[j], portid,
1009 								qconf);
1010 			}
1011 		}
1012 
1013 		if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
1014 			for (i = 1, lcore_scaleup_hint =
1015 				qconf->rx_queue_list[0].freq_up_hint;
1016 					i < qconf->n_rx_queue; ++i) {
1017 				rx_queue = &(qconf->rx_queue_list[i]);
1018 				if (rx_queue->freq_up_hint >
1019 						lcore_scaleup_hint)
1020 					lcore_scaleup_hint =
1021 						rx_queue->freq_up_hint;
1022 			}
1023 
1024 			if (lcore_scaleup_hint == FREQ_HIGHEST) {
1025 				if (rte_power_freq_max)
1026 					rte_power_freq_max(lcore_id);
1027 			} else if (lcore_scaleup_hint == FREQ_HIGHER) {
1028 				if (rte_power_freq_up)
1029 					rte_power_freq_up(lcore_id);
1030 			}
1031 		} else {
1032 			/**
1033 			 * All Rx queues empty in recent consecutive polls,
1034 			 * sleep in a conservative manner, meaning sleep as
1035 			 * less as possible.
1036 			 */
1037 			for (i = 1, lcore_idle_hint =
1038 				qconf->rx_queue_list[0].idle_hint;
1039 					i < qconf->n_rx_queue; ++i) {
1040 				rx_queue = &(qconf->rx_queue_list[i]);
1041 				if (rx_queue->idle_hint < lcore_idle_hint)
1042 					lcore_idle_hint = rx_queue->idle_hint;
1043 			}
1044 
1045 			if (lcore_idle_hint < SUSPEND_THRESHOLD)
1046 				/**
1047 				 * execute "pause" instruction to avoid context
1048 				 * switch which generally take hundred of
1049 				 * microseconds for short sleep.
1050 				 */
1051 				rte_delay_us(lcore_idle_hint);
1052 			else {
1053 				/* suspend until rx interrupt trigges */
1054 				if (intr_en) {
1055 					turn_on_intr(qconf);
1056 					sleep_until_rx_interrupt(
1057 						qconf->n_rx_queue);
1058 				}
1059 				/* start receiving packets immediately */
1060 				goto start_rx;
1061 			}
1062 			stats[lcore_id].sleep_time += lcore_idle_hint;
1063 		}
1064 	}
1065 }
1066 
1067 static int
1068 check_lcore_params(void)
1069 {
1070 	uint8_t queue, lcore;
1071 	uint16_t i;
1072 	int socketid;
1073 
1074 	for (i = 0; i < nb_lcore_params; ++i) {
1075 		queue = lcore_params[i].queue_id;
1076 		if (queue >= MAX_RX_QUEUE_PER_PORT) {
1077 			printf("invalid queue number: %hhu\n", queue);
1078 			return -1;
1079 		}
1080 		lcore = lcore_params[i].lcore_id;
1081 		if (!rte_lcore_is_enabled(lcore)) {
1082 			printf("error: lcore %hhu is not enabled in lcore "
1083 							"mask\n", lcore);
1084 			return -1;
1085 		}
1086 		if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1087 							(numa_on == 0)) {
1088 			printf("warning: lcore %hhu is on socket %d with numa "
1089 						"off\n", lcore, socketid);
1090 		}
1091 	}
1092 	return 0;
1093 }
1094 
1095 static int
1096 check_port_config(const unsigned nb_ports)
1097 {
1098 	unsigned portid;
1099 	uint16_t i;
1100 
1101 	for (i = 0; i < nb_lcore_params; ++i) {
1102 		portid = lcore_params[i].port_id;
1103 		if ((enabled_port_mask & (1 << portid)) == 0) {
1104 			printf("port %u is not enabled in port mask\n",
1105 								portid);
1106 			return -1;
1107 		}
1108 		if (portid >= nb_ports) {
1109 			printf("port %u is not present on the board\n",
1110 								portid);
1111 			return -1;
1112 		}
1113 	}
1114 	return 0;
1115 }
1116 
1117 static uint8_t
1118 get_port_n_rx_queues(const uint8_t port)
1119 {
1120 	int queue = -1;
1121 	uint16_t i;
1122 
1123 	for (i = 0; i < nb_lcore_params; ++i) {
1124 		if (lcore_params[i].port_id == port &&
1125 				lcore_params[i].queue_id > queue)
1126 			queue = lcore_params[i].queue_id;
1127 	}
1128 	return (uint8_t)(++queue);
1129 }
1130 
1131 static int
1132 init_lcore_rx_queues(void)
1133 {
1134 	uint16_t i, nb_rx_queue;
1135 	uint8_t lcore;
1136 
1137 	for (i = 0; i < nb_lcore_params; ++i) {
1138 		lcore = lcore_params[i].lcore_id;
1139 		nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1140 		if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1141 			printf("error: too many queues (%u) for lcore: %u\n",
1142 				(unsigned)nb_rx_queue + 1, (unsigned)lcore);
1143 			return -1;
1144 		} else {
1145 			lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1146 				lcore_params[i].port_id;
1147 			lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1148 				lcore_params[i].queue_id;
1149 			lcore_conf[lcore].n_rx_queue++;
1150 		}
1151 	}
1152 	return 0;
1153 }
1154 
1155 /* display usage */
1156 static void
1157 print_usage(const char *prgname)
1158 {
1159 	printf ("%s [EAL options] -- -p PORTMASK -P"
1160 		"  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1161 		"  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1162 		"  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1163 		"  -P : enable promiscuous mode\n"
1164 		"  --config (port,queue,lcore): rx queues configuration\n"
1165 		"  --no-numa: optional, disable numa awareness\n"
1166 		"  --enable-jumbo: enable jumbo frame"
1167 		" which max packet len is PKTLEN in decimal (64-9600)\n"
1168 		"  --parse-ptype: parse packet type by software\n",
1169 		prgname);
1170 }
1171 
1172 static int parse_max_pkt_len(const char *pktlen)
1173 {
1174 	char *end = NULL;
1175 	unsigned long len;
1176 
1177 	/* parse decimal string */
1178 	len = strtoul(pktlen, &end, 10);
1179 	if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1180 		return -1;
1181 
1182 	if (len == 0)
1183 		return -1;
1184 
1185 	return len;
1186 }
1187 
1188 static int
1189 parse_portmask(const char *portmask)
1190 {
1191 	char *end = NULL;
1192 	unsigned long pm;
1193 
1194 	/* parse hexadecimal string */
1195 	pm = strtoul(portmask, &end, 16);
1196 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1197 		return -1;
1198 
1199 	if (pm == 0)
1200 		return -1;
1201 
1202 	return pm;
1203 }
1204 
1205 static int
1206 parse_config(const char *q_arg)
1207 {
1208 	char s[256];
1209 	const char *p, *p0 = q_arg;
1210 	char *end;
1211 	enum fieldnames {
1212 		FLD_PORT = 0,
1213 		FLD_QUEUE,
1214 		FLD_LCORE,
1215 		_NUM_FLD
1216 	};
1217 	unsigned long int_fld[_NUM_FLD];
1218 	char *str_fld[_NUM_FLD];
1219 	int i;
1220 	unsigned size;
1221 
1222 	nb_lcore_params = 0;
1223 
1224 	while ((p = strchr(p0,'(')) != NULL) {
1225 		++p;
1226 		if((p0 = strchr(p,')')) == NULL)
1227 			return -1;
1228 
1229 		size = p0 - p;
1230 		if(size >= sizeof(s))
1231 			return -1;
1232 
1233 		snprintf(s, sizeof(s), "%.*s", size, p);
1234 		if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1235 								_NUM_FLD)
1236 			return -1;
1237 		for (i = 0; i < _NUM_FLD; i++){
1238 			errno = 0;
1239 			int_fld[i] = strtoul(str_fld[i], &end, 0);
1240 			if (errno != 0 || end == str_fld[i] || int_fld[i] >
1241 									255)
1242 				return -1;
1243 		}
1244 		if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1245 			printf("exceeded max number of lcore params: %hu\n",
1246 				nb_lcore_params);
1247 			return -1;
1248 		}
1249 		lcore_params_array[nb_lcore_params].port_id =
1250 				(uint8_t)int_fld[FLD_PORT];
1251 		lcore_params_array[nb_lcore_params].queue_id =
1252 				(uint8_t)int_fld[FLD_QUEUE];
1253 		lcore_params_array[nb_lcore_params].lcore_id =
1254 				(uint8_t)int_fld[FLD_LCORE];
1255 		++nb_lcore_params;
1256 	}
1257 	lcore_params = lcore_params_array;
1258 
1259 	return 0;
1260 }
1261 
1262 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1263 
1264 /* Parse the argument given in the command line of the application */
1265 static int
1266 parse_args(int argc, char **argv)
1267 {
1268 	int opt, ret;
1269 	char **argvopt;
1270 	int option_index;
1271 	char *prgname = argv[0];
1272 	static struct option lgopts[] = {
1273 		{"config", 1, 0, 0},
1274 		{"no-numa", 0, 0, 0},
1275 		{"enable-jumbo", 0, 0, 0},
1276 		{CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
1277 		{NULL, 0, 0, 0}
1278 	};
1279 
1280 	argvopt = argv;
1281 
1282 	while ((opt = getopt_long(argc, argvopt, "p:P",
1283 				lgopts, &option_index)) != EOF) {
1284 
1285 		switch (opt) {
1286 		/* portmask */
1287 		case 'p':
1288 			enabled_port_mask = parse_portmask(optarg);
1289 			if (enabled_port_mask == 0) {
1290 				printf("invalid portmask\n");
1291 				print_usage(prgname);
1292 				return -1;
1293 			}
1294 			break;
1295 		case 'P':
1296 			printf("Promiscuous mode selected\n");
1297 			promiscuous_on = 1;
1298 			break;
1299 
1300 		/* long options */
1301 		case 0:
1302 			if (!strncmp(lgopts[option_index].name, "config", 6)) {
1303 				ret = parse_config(optarg);
1304 				if (ret) {
1305 					printf("invalid config\n");
1306 					print_usage(prgname);
1307 					return -1;
1308 				}
1309 			}
1310 
1311 			if (!strncmp(lgopts[option_index].name,
1312 						"no-numa", 7)) {
1313 				printf("numa is disabled \n");
1314 				numa_on = 0;
1315 			}
1316 
1317 			if (!strncmp(lgopts[option_index].name,
1318 					"enable-jumbo", 12)) {
1319 				struct option lenopts =
1320 					{"max-pkt-len", required_argument, \
1321 									0, 0};
1322 
1323 				printf("jumbo frame is enabled \n");
1324 				port_conf.rxmode.jumbo_frame = 1;
1325 
1326 				/**
1327 				 * if no max-pkt-len set, use the default value
1328 				 * ETHER_MAX_LEN
1329 				 */
1330 				if (0 == getopt_long(argc, argvopt, "",
1331 						&lenopts, &option_index)) {
1332 					ret = parse_max_pkt_len(optarg);
1333 					if ((ret < 64) ||
1334 						(ret > MAX_JUMBO_PKT_LEN)){
1335 						printf("invalid packet "
1336 								"length\n");
1337 						print_usage(prgname);
1338 						return -1;
1339 					}
1340 					port_conf.rxmode.max_rx_pkt_len = ret;
1341 				}
1342 				printf("set jumbo frame "
1343 					"max packet length to %u\n",
1344 				(unsigned int)port_conf.rxmode.max_rx_pkt_len);
1345 			}
1346 
1347 			if (!strncmp(lgopts[option_index].name,
1348 				     CMD_LINE_OPT_PARSE_PTYPE,
1349 				     sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
1350 				printf("soft parse-ptype is enabled\n");
1351 				parse_ptype = 1;
1352 			}
1353 
1354 			break;
1355 
1356 		default:
1357 			print_usage(prgname);
1358 			return -1;
1359 		}
1360 	}
1361 
1362 	if (optind >= 0)
1363 		argv[optind-1] = prgname;
1364 
1365 	ret = optind-1;
1366 	optind = 1; /* reset getopt lib */
1367 	return ret;
1368 }
1369 
1370 static void
1371 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1372 {
1373 	char buf[ETHER_ADDR_FMT_SIZE];
1374 	ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1375 	printf("%s%s", name, buf);
1376 }
1377 
1378 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1379 static void
1380 setup_hash(int socketid)
1381 {
1382 	struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1383 		.name = NULL,
1384 		.entries = L3FWD_HASH_ENTRIES,
1385 		.key_len = sizeof(struct ipv4_5tuple),
1386 		.hash_func = DEFAULT_HASH_FUNC,
1387 		.hash_func_init_val = 0,
1388 	};
1389 
1390 	struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1391 		.name = NULL,
1392 		.entries = L3FWD_HASH_ENTRIES,
1393 		.key_len = sizeof(struct ipv6_5tuple),
1394 		.hash_func = DEFAULT_HASH_FUNC,
1395 		.hash_func_init_val = 0,
1396 	};
1397 
1398 	unsigned i;
1399 	int ret;
1400 	char s[64];
1401 
1402 	/* create ipv4 hash */
1403 	snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1404 	ipv4_l3fwd_hash_params.name = s;
1405 	ipv4_l3fwd_hash_params.socket_id = socketid;
1406 	ipv4_l3fwd_lookup_struct[socketid] =
1407 		rte_hash_create(&ipv4_l3fwd_hash_params);
1408 	if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1409 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1410 				"socket %d\n", socketid);
1411 
1412 	/* create ipv6 hash */
1413 	snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1414 	ipv6_l3fwd_hash_params.name = s;
1415 	ipv6_l3fwd_hash_params.socket_id = socketid;
1416 	ipv6_l3fwd_lookup_struct[socketid] =
1417 		rte_hash_create(&ipv6_l3fwd_hash_params);
1418 	if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1419 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1420 				"socket %d\n", socketid);
1421 
1422 
1423 	/* populate the ipv4 hash */
1424 	for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1425 		ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1426 				(void *) &ipv4_l3fwd_route_array[i].key);
1427 		if (ret < 0) {
1428 			rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1429 				"l3fwd hash on socket %d\n", i, socketid);
1430 		}
1431 		ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1432 		printf("Hash: Adding key\n");
1433 		print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1434 	}
1435 
1436 	/* populate the ipv6 hash */
1437 	for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1438 		ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1439 				(void *) &ipv6_l3fwd_route_array[i].key);
1440 		if (ret < 0) {
1441 			rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1442 				"l3fwd hash on socket %d\n", i, socketid);
1443 		}
1444 		ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1445 		printf("Hash: Adding key\n");
1446 		print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1447 	}
1448 }
1449 #endif
1450 
1451 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1452 static void
1453 setup_lpm(int socketid)
1454 {
1455 	unsigned i;
1456 	int ret;
1457 	char s[64];
1458 
1459 	/* create the LPM table */
1460 	struct rte_lpm_config lpm_ipv4_config;
1461 
1462 	lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
1463 	lpm_ipv4_config.number_tbl8s = 256;
1464 	lpm_ipv4_config.flags = 0;
1465 
1466 	snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1467 	ipv4_l3fwd_lookup_struct[socketid] =
1468 			rte_lpm_create(s, socketid, &lpm_ipv4_config);
1469 	if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1470 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1471 				" on socket %d\n", socketid);
1472 
1473 	/* populate the LPM table */
1474 	for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1475 		ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1476 			ipv4_l3fwd_route_array[i].ip,
1477 			ipv4_l3fwd_route_array[i].depth,
1478 			ipv4_l3fwd_route_array[i].if_out);
1479 
1480 		if (ret < 0) {
1481 			rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1482 				"l3fwd LPM table on socket %d\n",
1483 				i, socketid);
1484 		}
1485 
1486 		printf("LPM: Adding route 0x%08x / %d (%d)\n",
1487 			(unsigned)ipv4_l3fwd_route_array[i].ip,
1488 			ipv4_l3fwd_route_array[i].depth,
1489 			ipv4_l3fwd_route_array[i].if_out);
1490 	}
1491 }
1492 #endif
1493 
1494 static int
1495 init_mem(unsigned nb_mbuf)
1496 {
1497 	struct lcore_conf *qconf;
1498 	int socketid;
1499 	unsigned lcore_id;
1500 	char s[64];
1501 
1502 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1503 		if (rte_lcore_is_enabled(lcore_id) == 0)
1504 			continue;
1505 
1506 		if (numa_on)
1507 			socketid = rte_lcore_to_socket_id(lcore_id);
1508 		else
1509 			socketid = 0;
1510 
1511 		if (socketid >= NB_SOCKETS) {
1512 			rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1513 					"out of range %d\n", socketid,
1514 						lcore_id, NB_SOCKETS);
1515 		}
1516 		if (pktmbuf_pool[socketid] == NULL) {
1517 			snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1518 			pktmbuf_pool[socketid] =
1519 				rte_pktmbuf_pool_create(s, nb_mbuf,
1520 					MEMPOOL_CACHE_SIZE, 0,
1521 					RTE_MBUF_DEFAULT_BUF_SIZE,
1522 					socketid);
1523 			if (pktmbuf_pool[socketid] == NULL)
1524 				rte_exit(EXIT_FAILURE,
1525 					"Cannot init mbuf pool on socket %d\n",
1526 								socketid);
1527 			else
1528 				printf("Allocated mbuf pool on socket %d\n",
1529 								socketid);
1530 
1531 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1532 			setup_lpm(socketid);
1533 #else
1534 			setup_hash(socketid);
1535 #endif
1536 		}
1537 		qconf = &lcore_conf[lcore_id];
1538 		qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1539 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1540 		qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1541 #endif
1542 	}
1543 	return 0;
1544 }
1545 
1546 /* Check the link status of all ports in up to 9s, and print them finally */
1547 static void
1548 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask)
1549 {
1550 #define CHECK_INTERVAL 100 /* 100ms */
1551 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1552 	uint8_t count, all_ports_up, print_flag = 0;
1553 	uint16_t portid;
1554 	struct rte_eth_link link;
1555 
1556 	printf("\nChecking link status");
1557 	fflush(stdout);
1558 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
1559 		all_ports_up = 1;
1560 		for (portid = 0; portid < port_num; portid++) {
1561 			if ((port_mask & (1 << portid)) == 0)
1562 				continue;
1563 			memset(&link, 0, sizeof(link));
1564 			rte_eth_link_get_nowait(portid, &link);
1565 			/* print link status if flag set */
1566 			if (print_flag == 1) {
1567 				if (link.link_status)
1568 					printf("Port %d Link Up - speed %u "
1569 						"Mbps - %s\n", (uint8_t)portid,
1570 						(unsigned)link.link_speed,
1571 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1572 					("full-duplex") : ("half-duplex\n"));
1573 				else
1574 					printf("Port %d Link Down\n",
1575 						(uint8_t)portid);
1576 				continue;
1577 			}
1578 			/* clear all_ports_up flag if any link down */
1579 			if (link.link_status == ETH_LINK_DOWN) {
1580 				all_ports_up = 0;
1581 				break;
1582 			}
1583 		}
1584 		/* after finally printing all link status, get out */
1585 		if (print_flag == 1)
1586 			break;
1587 
1588 		if (all_ports_up == 0) {
1589 			printf(".");
1590 			fflush(stdout);
1591 			rte_delay_ms(CHECK_INTERVAL);
1592 		}
1593 
1594 		/* set the print_flag if all ports up or timeout */
1595 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1596 			print_flag = 1;
1597 			printf("done\n");
1598 		}
1599 	}
1600 }
1601 
1602 static int check_ptype(uint8_t portid)
1603 {
1604 	int i, ret;
1605 	int ptype_l3_ipv4 = 0;
1606 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1607 	int ptype_l3_ipv6 = 0;
1608 #endif
1609 	uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
1610 
1611 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
1612 	if (ret <= 0)
1613 		return 0;
1614 
1615 	uint32_t ptypes[ret];
1616 
1617 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
1618 	for (i = 0; i < ret; ++i) {
1619 		if (ptypes[i] & RTE_PTYPE_L3_IPV4)
1620 			ptype_l3_ipv4 = 1;
1621 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1622 		if (ptypes[i] & RTE_PTYPE_L3_IPV6)
1623 			ptype_l3_ipv6 = 1;
1624 #endif
1625 	}
1626 
1627 	if (ptype_l3_ipv4 == 0)
1628 		printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
1629 
1630 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1631 	if (ptype_l3_ipv6 == 0)
1632 		printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
1633 #endif
1634 
1635 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1636 	if (ptype_l3_ipv4)
1637 #else /* APP_LOOKUP_EXACT_MATCH */
1638 	if (ptype_l3_ipv4 && ptype_l3_ipv6)
1639 #endif
1640 		return 1;
1641 
1642 	return 0;
1643 
1644 }
1645 
1646 int
1647 main(int argc, char **argv)
1648 {
1649 	struct lcore_conf *qconf;
1650 	struct rte_eth_dev_info dev_info;
1651 	struct rte_eth_txconf *txconf;
1652 	int ret;
1653 	unsigned nb_ports;
1654 	uint16_t queueid;
1655 	unsigned lcore_id;
1656 	uint64_t hz;
1657 	uint32_t n_tx_queue, nb_lcores;
1658 	uint32_t dev_rxq_num, dev_txq_num;
1659 	uint8_t nb_rx_queue, queue, socketid;
1660 	uint16_t portid;
1661 	uint16_t org_rxq_intr = port_conf.intr_conf.rxq;
1662 
1663 	/* catch SIGINT and restore cpufreq governor to ondemand */
1664 	signal(SIGINT, signal_exit_now);
1665 
1666 	/* init EAL */
1667 	ret = rte_eal_init(argc, argv);
1668 	if (ret < 0)
1669 		rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1670 	argc -= ret;
1671 	argv += ret;
1672 
1673 	/* init RTE timer library to be used late */
1674 	rte_timer_subsystem_init();
1675 
1676 	/* parse application arguments (after the EAL ones) */
1677 	ret = parse_args(argc, argv);
1678 	if (ret < 0)
1679 		rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1680 
1681 	if (check_lcore_params() < 0)
1682 		rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1683 
1684 	ret = init_lcore_rx_queues();
1685 	if (ret < 0)
1686 		rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1687 
1688 	nb_ports = rte_eth_dev_count();
1689 
1690 	if (check_port_config(nb_ports) < 0)
1691 		rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1692 
1693 	nb_lcores = rte_lcore_count();
1694 
1695 	/* initialize all ports */
1696 	for (portid = 0; portid < nb_ports; portid++) {
1697 		/* skip ports that are not enabled */
1698 		if ((enabled_port_mask & (1 << portid)) == 0) {
1699 			printf("\nSkipping disabled port %d\n", portid);
1700 			continue;
1701 		}
1702 
1703 		/* init port */
1704 		printf("Initializing port %d ... ", portid );
1705 		fflush(stdout);
1706 
1707 		rte_eth_dev_info_get(portid, &dev_info);
1708 		dev_rxq_num = dev_info.max_rx_queues;
1709 		dev_txq_num = dev_info.max_tx_queues;
1710 
1711 		nb_rx_queue = get_port_n_rx_queues(portid);
1712 		if (nb_rx_queue > dev_rxq_num)
1713 			rte_exit(EXIT_FAILURE,
1714 				"Cannot configure not existed rxq: "
1715 				"port=%d\n", portid);
1716 
1717 		n_tx_queue = nb_lcores;
1718 		if (n_tx_queue > dev_txq_num)
1719 			n_tx_queue = dev_txq_num;
1720 		printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1721 			nb_rx_queue, (unsigned)n_tx_queue );
1722 		/* If number of Rx queue is 0, no need to enable Rx interrupt */
1723 		if (nb_rx_queue == 0)
1724 			port_conf.intr_conf.rxq = 0;
1725 		ret = rte_eth_dev_configure(portid, nb_rx_queue,
1726 					(uint16_t)n_tx_queue, &port_conf);
1727 		/* Revert to original value */
1728 		port_conf.intr_conf.rxq = org_rxq_intr;
1729 		if (ret < 0)
1730 			rte_exit(EXIT_FAILURE, "Cannot configure device: "
1731 					"err=%d, port=%d\n", ret, portid);
1732 
1733 		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1734 						       &nb_txd);
1735 		if (ret < 0)
1736 			rte_exit(EXIT_FAILURE,
1737 				 "Cannot adjust number of descriptors: err=%d, port=%d\n",
1738 				 ret, portid);
1739 
1740 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1741 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
1742 		printf(", ");
1743 
1744 		/* init memory */
1745 		ret = init_mem(NB_MBUF);
1746 		if (ret < 0)
1747 			rte_exit(EXIT_FAILURE, "init_mem failed\n");
1748 
1749 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1750 			if (rte_lcore_is_enabled(lcore_id) == 0)
1751 				continue;
1752 
1753 			/* Initialize TX buffers */
1754 			qconf = &lcore_conf[lcore_id];
1755 			qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
1756 				RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
1757 				rte_eth_dev_socket_id(portid));
1758 			if (qconf->tx_buffer[portid] == NULL)
1759 				rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
1760 						 portid);
1761 
1762 			rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
1763 		}
1764 
1765 		/* init one TX queue per couple (lcore,port) */
1766 		queueid = 0;
1767 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1768 			if (rte_lcore_is_enabled(lcore_id) == 0)
1769 				continue;
1770 
1771 			if (queueid >= dev_txq_num)
1772 				continue;
1773 
1774 			if (numa_on)
1775 				socketid = \
1776 				(uint8_t)rte_lcore_to_socket_id(lcore_id);
1777 			else
1778 				socketid = 0;
1779 
1780 			printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1781 			fflush(stdout);
1782 
1783 			rte_eth_dev_info_get(portid, &dev_info);
1784 			txconf = &dev_info.default_txconf;
1785 			if (port_conf.rxmode.jumbo_frame)
1786 				txconf->txq_flags = 0;
1787 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1788 						     socketid, txconf);
1789 			if (ret < 0)
1790 				rte_exit(EXIT_FAILURE,
1791 					"rte_eth_tx_queue_setup: err=%d, "
1792 						"port=%d\n", ret, portid);
1793 
1794 			qconf = &lcore_conf[lcore_id];
1795 			qconf->tx_queue_id[portid] = queueid;
1796 			queueid++;
1797 
1798 			qconf->tx_port_id[qconf->n_tx_port] = portid;
1799 			qconf->n_tx_port++;
1800 		}
1801 		printf("\n");
1802 	}
1803 
1804 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1805 		if (rte_lcore_is_enabled(lcore_id) == 0)
1806 			continue;
1807 
1808 		/* init power management library */
1809 		ret = rte_power_init(lcore_id);
1810 		if (ret)
1811 			RTE_LOG(ERR, POWER,
1812 				"Library initialization failed on core %u\n", lcore_id);
1813 
1814 		/* init timer structures for each enabled lcore */
1815 		rte_timer_init(&power_timers[lcore_id]);
1816 		hz = rte_get_timer_hz();
1817 		rte_timer_reset(&power_timers[lcore_id],
1818 			hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1819 						power_timer_cb, NULL);
1820 
1821 		qconf = &lcore_conf[lcore_id];
1822 		printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1823 		fflush(stdout);
1824 		/* init RX queues */
1825 		for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1826 			portid = qconf->rx_queue_list[queue].port_id;
1827 			queueid = qconf->rx_queue_list[queue].queue_id;
1828 
1829 			if (numa_on)
1830 				socketid = \
1831 				(uint8_t)rte_lcore_to_socket_id(lcore_id);
1832 			else
1833 				socketid = 0;
1834 
1835 			printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1836 			fflush(stdout);
1837 
1838 			ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1839 				socketid, NULL,
1840 				pktmbuf_pool[socketid]);
1841 			if (ret < 0)
1842 				rte_exit(EXIT_FAILURE,
1843 					"rte_eth_rx_queue_setup: err=%d, "
1844 						"port=%d\n", ret, portid);
1845 
1846 			if (parse_ptype) {
1847 				if (add_cb_parse_ptype(portid, queueid) < 0)
1848 					rte_exit(EXIT_FAILURE,
1849 						 "Fail to add ptype cb\n");
1850 			} else if (!check_ptype(portid))
1851 				rte_exit(EXIT_FAILURE,
1852 					 "PMD can not provide needed ptypes\n");
1853 		}
1854 	}
1855 
1856 	printf("\n");
1857 
1858 	/* start ports */
1859 	for (portid = 0; portid < nb_ports; portid++) {
1860 		if ((enabled_port_mask & (1 << portid)) == 0) {
1861 			continue;
1862 		}
1863 		/* Start device */
1864 		ret = rte_eth_dev_start(portid);
1865 		if (ret < 0)
1866 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1867 						"port=%d\n", ret, portid);
1868 		/*
1869 		 * If enabled, put device in promiscuous mode.
1870 		 * This allows IO forwarding mode to forward packets
1871 		 * to itself through 2 cross-connected  ports of the
1872 		 * target machine.
1873 		 */
1874 		if (promiscuous_on)
1875 			rte_eth_promiscuous_enable(portid);
1876 		/* initialize spinlock for each port */
1877 		rte_spinlock_init(&(locks[portid]));
1878 	}
1879 
1880 	check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1881 
1882 	/* launch per-lcore init on every lcore */
1883 	rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1884 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1885 		if (rte_eal_wait_lcore(lcore_id) < 0)
1886 			return -1;
1887 	}
1888 
1889 	return 0;
1890 }
1891