xref: /dpdk/examples/l3fwd-power/main.c (revision d51d53b459142b4f41f7cc7b531ec6741bb0a2e7)
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 #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_memory.h>
51 #include <rte_memcpy.h>
52 #include <rte_memzone.h>
53 #include <rte_eal.h>
54 #include <rte_per_lcore.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_ring.h>
69 #include <rte_mempool.h>
70 #include <rte_mbuf.h>
71 #include <rte_ip.h>
72 #include <rte_tcp.h>
73 #include <rte_udp.h>
74 #include <rte_string_fns.h>
75 #include <rte_timer.h>
76 #include <rte_power.h>
77 #include <rte_eal.h>
78 #include <rte_spinlock.h>
79 
80 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
81 
82 #define MAX_PKT_BURST 32
83 
84 #define MIN_ZERO_POLL_COUNT 10
85 
86 /* around 100ms at 2 Ghz */
87 #define TIMER_RESOLUTION_CYCLES           200000000ULL
88 /* 100 ms interval */
89 #define TIMER_NUMBER_PER_SECOND           10
90 /* 100000 us */
91 #define SCALING_PERIOD                    (1000000/TIMER_NUMBER_PER_SECOND)
92 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
93 
94 #define APP_LOOKUP_EXACT_MATCH          0
95 #define APP_LOOKUP_LPM                  1
96 #define DO_RFC_1812_CHECKS
97 
98 #ifndef APP_LOOKUP_METHOD
99 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
100 #endif
101 
102 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
103 #include <rte_hash.h>
104 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
105 #include <rte_lpm.h>
106 #else
107 #error "APP_LOOKUP_METHOD set to incorrect value"
108 #endif
109 
110 #ifndef IPv6_BYTES
111 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
112                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
113 #define IPv6_BYTES(addr) \
114 	addr[0],  addr[1], addr[2],  addr[3], \
115 	addr[4],  addr[5], addr[6],  addr[7], \
116 	addr[8],  addr[9], addr[10], addr[11],\
117 	addr[12], addr[13],addr[14], addr[15]
118 #endif
119 
120 #define MAX_JUMBO_PKT_LEN  9600
121 
122 #define IPV6_ADDR_LEN 16
123 
124 #define MEMPOOL_CACHE_SIZE 256
125 
126 /*
127  * This expression is used to calculate the number of mbufs needed depending on
128  * user input, taking into account memory for rx and tx hardware rings, cache
129  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
130  * NB_MBUF never goes below a minimum value of 8192.
131  */
132 
133 #define NB_MBUF RTE_MAX	( \
134 	(nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT + \
135 	nb_ports*nb_lcores*MAX_PKT_BURST + \
136 	nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT + \
137 	nb_lcores*MEMPOOL_CACHE_SIZE), \
138 	(unsigned)8192)
139 
140 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
141 
142 #define NB_SOCKETS 8
143 
144 /* Configure how many packets ahead to prefetch, when reading packets */
145 #define PREFETCH_OFFSET	3
146 
147 /*
148  * Configurable number of RX/TX ring descriptors
149  */
150 #define RTE_TEST_RX_DESC_DEFAULT 128
151 #define RTE_TEST_TX_DESC_DEFAULT 512
152 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
153 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
154 
155 /* ethernet addresses of ports */
156 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
157 
158 /* ethernet addresses of ports */
159 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
160 
161 /* mask of enabled ports */
162 static uint32_t enabled_port_mask = 0;
163 /* Ports set in promiscuous mode off by default. */
164 static int promiscuous_on = 0;
165 /* NUMA is enabled by default. */
166 static int numa_on = 1;
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 mbuf_table {
177 	uint16_t len;
178 	struct rte_mbuf *m_table[MAX_PKT_BURST];
179 };
180 
181 struct lcore_rx_queue {
182 	uint8_t port_id;
183 	uint8_t queue_id;
184 	enum freq_scale_hint_t freq_up_hint;
185 	uint32_t zero_rx_packet_count;
186 	uint32_t idle_hint;
187 } __rte_cache_aligned;
188 
189 #define MAX_RX_QUEUE_PER_LCORE 16
190 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
191 #define MAX_RX_QUEUE_PER_PORT 128
192 
193 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
194 
195 
196 #define MAX_LCORE_PARAMS 1024
197 struct lcore_params {
198 	uint8_t port_id;
199 	uint8_t queue_id;
200 	uint8_t lcore_id;
201 } __rte_cache_aligned;
202 
203 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
204 static struct lcore_params lcore_params_array_default[] = {
205 	{0, 0, 2},
206 	{0, 1, 2},
207 	{0, 2, 2},
208 	{1, 0, 2},
209 	{1, 1, 2},
210 	{1, 2, 2},
211 	{2, 0, 2},
212 	{3, 0, 3},
213 	{3, 1, 3},
214 };
215 
216 static struct lcore_params * lcore_params = lcore_params_array_default;
217 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
218 				sizeof(lcore_params_array_default[0]);
219 
220 static struct rte_eth_conf port_conf = {
221 	.rxmode = {
222 		.mq_mode        = ETH_MQ_RX_RSS,
223 		.max_rx_pkt_len = ETHER_MAX_LEN,
224 		.split_hdr_size = 0,
225 		.header_split   = 0, /**< Header Split disabled */
226 		.hw_ip_checksum = 1, /**< IP checksum offload enabled */
227 		.hw_vlan_filter = 0, /**< VLAN filtering disabled */
228 		.jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
229 		.hw_strip_crc   = 0, /**< CRC stripped by hardware */
230 	},
231 	.rx_adv_conf = {
232 		.rss_conf = {
233 			.rss_key = NULL,
234 			.rss_hf = ETH_RSS_UDP,
235 		},
236 	},
237 	.txmode = {
238 		.mq_mode = ETH_MQ_TX_NONE,
239 	},
240 	.intr_conf = {
241 		.lsc = 1,
242 #ifdef RTE_NEXT_ABI
243 		.rxq = 1,
244 #endif
245 	},
246 };
247 
248 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
249 
250 
251 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
252 
253 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
254 #include <rte_hash_crc.h>
255 #define DEFAULT_HASH_FUNC       rte_hash_crc
256 #else
257 #include <rte_jhash.h>
258 #define DEFAULT_HASH_FUNC       rte_jhash
259 #endif
260 
261 struct ipv4_5tuple {
262 	uint32_t ip_dst;
263 	uint32_t ip_src;
264 	uint16_t port_dst;
265 	uint16_t port_src;
266 	uint8_t  proto;
267 } __attribute__((__packed__));
268 
269 struct ipv6_5tuple {
270 	uint8_t  ip_dst[IPV6_ADDR_LEN];
271 	uint8_t  ip_src[IPV6_ADDR_LEN];
272 	uint16_t port_dst;
273 	uint16_t port_src;
274 	uint8_t  proto;
275 } __attribute__((__packed__));
276 
277 struct ipv4_l3fwd_route {
278 	struct ipv4_5tuple key;
279 	uint8_t if_out;
280 };
281 
282 struct ipv6_l3fwd_route {
283 	struct ipv6_5tuple key;
284 	uint8_t if_out;
285 };
286 
287 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
288 	{{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
289 	{{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
290 	{{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
291 	{{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
292 };
293 
294 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
295 	{
296 		{
297 			{0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
298 			 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
299 			{0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
300 			 0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
301 			 1, 10, IPPROTO_UDP
302 		}, 4
303 	},
304 };
305 
306 typedef struct rte_hash lookup_struct_t;
307 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
308 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
309 
310 #define L3FWD_HASH_ENTRIES	1024
311 
312 #define IPV4_L3FWD_NUM_ROUTES \
313 	(sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
314 
315 #define IPV6_L3FWD_NUM_ROUTES \
316 	(sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
317 
318 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
319 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
320 #endif
321 
322 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
323 struct ipv4_l3fwd_route {
324 	uint32_t ip;
325 	uint8_t  depth;
326 	uint8_t  if_out;
327 };
328 
329 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
330 	{IPv4(1,1,1,0), 24, 0},
331 	{IPv4(2,1,1,0), 24, 1},
332 	{IPv4(3,1,1,0), 24, 2},
333 	{IPv4(4,1,1,0), 24, 3},
334 	{IPv4(5,1,1,0), 24, 4},
335 	{IPv4(6,1,1,0), 24, 5},
336 	{IPv4(7,1,1,0), 24, 6},
337 	{IPv4(8,1,1,0), 24, 7},
338 };
339 
340 #define IPV4_L3FWD_NUM_ROUTES \
341 	(sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
342 
343 #define IPV4_L3FWD_LPM_MAX_RULES     1024
344 
345 typedef struct rte_lpm lookup_struct_t;
346 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
347 #endif
348 
349 struct lcore_conf {
350 	uint16_t n_rx_queue;
351 	struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
352 	uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
353 	struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
354 	lookup_struct_t * ipv4_lookup_struct;
355 	lookup_struct_t * ipv6_lookup_struct;
356 } __rte_cache_aligned;
357 
358 struct lcore_stats {
359 	/* total sleep time in ms since last frequency scaling down */
360 	uint32_t sleep_time;
361 	/* number of long sleep recently */
362 	uint32_t nb_long_sleep;
363 	/* freq. scaling up trend */
364 	uint32_t trend;
365 	/* total packet processed recently */
366 	uint64_t nb_rx_processed;
367 	/* total iterations looped recently */
368 	uint64_t nb_iteration_looped;
369 	uint32_t padding[9];
370 } __rte_cache_aligned;
371 
372 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
373 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
374 static struct rte_timer power_timers[RTE_MAX_LCORE];
375 
376 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
377 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
378 			unsigned lcore_id, uint8_t port_id, uint16_t queue_id);
379 
380 /* exit signal handler */
381 static void
382 signal_exit_now(int sigtype)
383 {
384 	unsigned lcore_id;
385 	int ret;
386 
387 	if (sigtype == SIGINT) {
388 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
389 			if (rte_lcore_is_enabled(lcore_id) == 0)
390 				continue;
391 
392 			/* init power management library */
393 			ret = rte_power_exit(lcore_id);
394 			if (ret)
395 				rte_exit(EXIT_FAILURE, "Power management "
396 					"library de-initialization failed on "
397 							"core%u\n", lcore_id);
398 		}
399 	}
400 
401 	rte_exit(EXIT_SUCCESS, "User forced exit\n");
402 }
403 
404 /*  Freqency scale down timer callback */
405 static void
406 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
407 			  __attribute__((unused)) void *arg)
408 {
409 	uint64_t hz;
410 	float sleep_time_ratio;
411 	unsigned lcore_id = rte_lcore_id();
412 
413 	/* accumulate total execution time in us when callback is invoked */
414 	sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
415 					(float)SCALING_PERIOD;
416 	/**
417 	 * check whether need to scale down frequency a step if it sleep a lot.
418 	 */
419 	if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
420 		if (rte_power_freq_down)
421 			rte_power_freq_down(lcore_id);
422 	}
423 	else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
424 		stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
425 		/**
426 		 * scale down a step if average packet per iteration less
427 		 * than expectation.
428 		 */
429 		if (rte_power_freq_down)
430 			rte_power_freq_down(lcore_id);
431 	}
432 
433 	/**
434 	 * initialize another timer according to current frequency to ensure
435 	 * timer interval is relatively fixed.
436 	 */
437 	hz = rte_get_timer_hz();
438 	rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
439 				SINGLE, lcore_id, power_timer_cb, NULL);
440 
441 	stats[lcore_id].nb_rx_processed = 0;
442 	stats[lcore_id].nb_iteration_looped = 0;
443 
444 	stats[lcore_id].sleep_time = 0;
445 }
446 
447 /* Send burst of packets on an output interface */
448 static inline int
449 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
450 {
451 	struct rte_mbuf **m_table;
452 	int ret;
453 	uint16_t queueid;
454 
455 	queueid = qconf->tx_queue_id[port];
456 	m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
457 
458 	ret = rte_eth_tx_burst(port, queueid, m_table, n);
459 	if (unlikely(ret < n)) {
460 		do {
461 			rte_pktmbuf_free(m_table[ret]);
462 		} while (++ret < n);
463 	}
464 
465 	return 0;
466 }
467 
468 /* Enqueue a single packet, and send burst if queue is filled */
469 static inline int
470 send_single_packet(struct rte_mbuf *m, uint8_t port)
471 {
472 	uint32_t lcore_id;
473 	uint16_t len;
474 	struct lcore_conf *qconf;
475 
476 	lcore_id = rte_lcore_id();
477 
478 	qconf = &lcore_conf[lcore_id];
479 	len = qconf->tx_mbufs[port].len;
480 	qconf->tx_mbufs[port].m_table[len] = m;
481 	len++;
482 
483 	/* enough pkts to be sent */
484 	if (unlikely(len == MAX_PKT_BURST)) {
485 		send_burst(qconf, MAX_PKT_BURST, port);
486 		len = 0;
487 	}
488 
489 	qconf->tx_mbufs[port].len = len;
490 	return 0;
491 }
492 
493 #ifdef DO_RFC_1812_CHECKS
494 static inline int
495 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
496 {
497 	/* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
498 	/*
499 	 * 1. The packet length reported by the Link Layer must be large
500 	 * enough to hold the minimum length legal IP datagram (20 bytes).
501 	 */
502 	if (link_len < sizeof(struct ipv4_hdr))
503 		return -1;
504 
505 	/* 2. The IP checksum must be correct. */
506 	/* this is checked in H/W */
507 
508 	/*
509 	 * 3. The IP version number must be 4. If the version number is not 4
510 	 * then the packet may be another version of IP, such as IPng or
511 	 * ST-II.
512 	 */
513 	if (((pkt->version_ihl) >> 4) != 4)
514 		return -3;
515 	/*
516 	 * 4. The IP header length field must be large enough to hold the
517 	 * minimum length legal IP datagram (20 bytes = 5 words).
518 	 */
519 	if ((pkt->version_ihl & 0xf) < 5)
520 		return -4;
521 
522 	/*
523 	 * 5. The IP total length field must be large enough to hold the IP
524 	 * datagram header, whose length is specified in the IP header length
525 	 * field.
526 	 */
527 	if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
528 		return -5;
529 
530 	return 0;
531 }
532 #endif
533 
534 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
535 static void
536 print_ipv4_key(struct ipv4_5tuple key)
537 {
538 	printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
539 		"proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
540 				key.port_dst, key.port_src, key.proto);
541 }
542 static void
543 print_ipv6_key(struct ipv6_5tuple key)
544 {
545 	printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
546 	        "port dst = %d, port src = %d, proto = %d\n",
547 	        IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
548 	        key.port_dst, key.port_src, key.proto);
549 }
550 
551 static inline uint8_t
552 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
553 		lookup_struct_t * ipv4_l3fwd_lookup_struct)
554 {
555 	struct ipv4_5tuple key;
556 	struct tcp_hdr *tcp;
557 	struct udp_hdr *udp;
558 	int ret = 0;
559 
560 	key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
561 	key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
562 	key.proto = ipv4_hdr->next_proto_id;
563 
564 	switch (ipv4_hdr->next_proto_id) {
565 	case IPPROTO_TCP:
566 		tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
567 					sizeof(struct ipv4_hdr));
568 		key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
569 		key.port_src = rte_be_to_cpu_16(tcp->src_port);
570 		break;
571 
572 	case IPPROTO_UDP:
573 		udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
574 					sizeof(struct ipv4_hdr));
575 		key.port_dst = rte_be_to_cpu_16(udp->dst_port);
576 		key.port_src = rte_be_to_cpu_16(udp->src_port);
577 		break;
578 
579 	default:
580 		key.port_dst = 0;
581 		key.port_src = 0;
582 		break;
583 	}
584 
585 	/* Find destination port */
586 	ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
587 	return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
588 }
589 
590 static inline uint8_t
591 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr,  uint8_t portid,
592 			lookup_struct_t *ipv6_l3fwd_lookup_struct)
593 {
594 	struct ipv6_5tuple key;
595 	struct tcp_hdr *tcp;
596 	struct udp_hdr *udp;
597 	int ret = 0;
598 
599 	memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
600 	memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
601 
602 	key.proto = ipv6_hdr->proto;
603 
604 	switch (ipv6_hdr->proto) {
605 	case IPPROTO_TCP:
606 		tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
607 					sizeof(struct ipv6_hdr));
608 		key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
609 		key.port_src = rte_be_to_cpu_16(tcp->src_port);
610 		break;
611 
612 	case IPPROTO_UDP:
613 		udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
614 					sizeof(struct ipv6_hdr));
615 		key.port_dst = rte_be_to_cpu_16(udp->dst_port);
616 		key.port_src = rte_be_to_cpu_16(udp->src_port);
617 		break;
618 
619 	default:
620 		key.port_dst = 0;
621 		key.port_src = 0;
622 		break;
623 	}
624 
625 	/* Find destination port */
626 	ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
627 	return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
628 }
629 #endif
630 
631 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
632 static inline uint8_t
633 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
634 		lookup_struct_t *ipv4_l3fwd_lookup_struct)
635 {
636 	uint8_t next_hop;
637 
638 	return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
639 			rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
640 			next_hop : portid);
641 }
642 #endif
643 
644 static inline void
645 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
646 				struct lcore_conf *qconf)
647 {
648 	struct ether_hdr *eth_hdr;
649 	struct ipv4_hdr *ipv4_hdr;
650 	void *d_addr_bytes;
651 	uint8_t dst_port;
652 
653 	eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
654 
655 #ifdef RTE_NEXT_ABI
656 	if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
657 #else
658 	if (m->ol_flags & PKT_RX_IPV4_HDR) {
659 #endif
660 		/* Handle IPv4 headers.*/
661 		ipv4_hdr =
662 			rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
663 						sizeof(struct ether_hdr));
664 
665 #ifdef DO_RFC_1812_CHECKS
666 		/* Check to make sure the packet is valid (RFC1812) */
667 		if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
668 			rte_pktmbuf_free(m);
669 			return;
670 		}
671 #endif
672 
673 		dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
674 					qconf->ipv4_lookup_struct);
675 		if (dst_port >= RTE_MAX_ETHPORTS ||
676 				(enabled_port_mask & 1 << dst_port) == 0)
677 			dst_port = portid;
678 
679 		/* 02:00:00:00:00:xx */
680 		d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
681 		*((uint64_t *)d_addr_bytes) =
682 			0x000000000002 + ((uint64_t)dst_port << 40);
683 
684 #ifdef DO_RFC_1812_CHECKS
685 		/* Update time to live and header checksum */
686 		--(ipv4_hdr->time_to_live);
687 		++(ipv4_hdr->hdr_checksum);
688 #endif
689 
690 		/* src addr */
691 		ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
692 
693 		send_single_packet(m, dst_port);
694 #ifdef RTE_NEXT_ABI
695 	} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
696 #else
697 	}
698 	else {
699 #endif
700 		/* Handle IPv6 headers.*/
701 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
702 		struct ipv6_hdr *ipv6_hdr;
703 
704 		ipv6_hdr =
705 			rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
706 						sizeof(struct ether_hdr));
707 
708 		dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
709 					qconf->ipv6_lookup_struct);
710 
711 		if (dst_port >= RTE_MAX_ETHPORTS ||
712 				(enabled_port_mask & 1 << dst_port) == 0)
713 			dst_port = portid;
714 
715 		/* 02:00:00:00:00:xx */
716 		d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
717 		*((uint64_t *)d_addr_bytes) =
718 			0x000000000002 + ((uint64_t)dst_port << 40);
719 
720 		/* src addr */
721 		ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
722 
723 		send_single_packet(m, dst_port);
724 #else
725 		/* We don't currently handle IPv6 packets in LPM mode. */
726 		rte_pktmbuf_free(m);
727 #endif
728 	}
729 
730 }
731 
732 #define MINIMUM_SLEEP_TIME         1
733 #define SUSPEND_THRESHOLD          300
734 
735 static inline uint32_t
736 power_idle_heuristic(uint32_t zero_rx_packet_count)
737 {
738 	/* If zero count is less than 100,  sleep 1us */
739 	if (zero_rx_packet_count < SUSPEND_THRESHOLD)
740 		return MINIMUM_SLEEP_TIME;
741 	/* If zero count is less than 1000, sleep 100 us which is the
742 		minimum latency switching from C3/C6 to C0
743 	*/
744 	else
745 		return SUSPEND_THRESHOLD;
746 
747 	return 0;
748 }
749 
750 static inline enum freq_scale_hint_t
751 power_freq_scaleup_heuristic(unsigned lcore_id,
752 			     uint8_t port_id,
753 			     uint16_t queue_id)
754 {
755 /**
756  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
757  * per iteration
758  */
759 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
760 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
761 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
762 #define FREQ_UP_TREND1_ACC   1
763 #define FREQ_UP_TREND2_ACC   100
764 #define FREQ_UP_THRESHOLD    10000
765 
766 	if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
767 			FREQ_GEAR3_RX_PACKET_THRESHOLD) > 0)) {
768 		stats[lcore_id].trend = 0;
769 		return FREQ_HIGHEST;
770 	} else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
771 			FREQ_GEAR2_RX_PACKET_THRESHOLD) > 0))
772 		stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
773 	else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
774 			FREQ_GEAR1_RX_PACKET_THRESHOLD) > 0))
775 		stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
776 
777 	if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
778 		stats[lcore_id].trend = 0;
779 		return FREQ_HIGHER;
780 	}
781 
782 	return FREQ_CURRENT;
783 }
784 
785 /**
786  * force polling thread sleep until one-shot rx interrupt triggers
787  * @param port_id
788  *  Port id.
789  * @param queue_id
790  *  Rx queue id.
791  * @return
792  *  0 on success
793  */
794 static int
795 sleep_until_rx_interrupt(int num)
796 {
797 	struct rte_epoll_event event[num];
798 	int n, i;
799 	uint8_t port_id, queue_id;
800 	void *data;
801 
802 	RTE_LOG(INFO, L3FWD_POWER,
803 		"lcore %u sleeps until interrupt triggers\n",
804 		rte_lcore_id());
805 
806 	n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
807 	for (i = 0; i < n; i++) {
808 		data = event[i].epdata.data;
809 		port_id = ((uintptr_t)data) >> CHAR_BIT;
810 		queue_id = ((uintptr_t)data) &
811 			RTE_LEN2MASK(CHAR_BIT, uint8_t);
812 		RTE_LOG(INFO, L3FWD_POWER,
813 			"lcore %u is waked up from rx interrupt on"
814 			" port %d queue %d\n",
815 			rte_lcore_id(), port_id, queue_id);
816 	}
817 
818 	return 0;
819 }
820 
821 static void turn_on_intr(struct lcore_conf *qconf)
822 {
823 	int i;
824 	struct lcore_rx_queue *rx_queue;
825 	uint8_t port_id, queue_id;
826 
827 	for (i = 0; i < qconf->n_rx_queue; ++i) {
828 		rx_queue = &(qconf->rx_queue_list[i]);
829 		port_id = rx_queue->port_id;
830 		queue_id = rx_queue->queue_id;
831 
832 		rte_spinlock_lock(&(locks[port_id]));
833 		rte_eth_dev_rx_intr_enable(port_id, queue_id);
834 		rte_spinlock_unlock(&(locks[port_id]));
835 	}
836 }
837 
838 static int event_register(struct lcore_conf *qconf)
839 {
840 	struct lcore_rx_queue *rx_queue;
841 	uint8_t portid, queueid;
842 	uint32_t data;
843 	int ret;
844 	int i;
845 
846 	for (i = 0; i < qconf->n_rx_queue; ++i) {
847 		rx_queue = &(qconf->rx_queue_list[i]);
848 		portid = rx_queue->port_id;
849 		queueid = rx_queue->queue_id;
850 		data = portid << CHAR_BIT | queueid;
851 
852 		ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
853 						RTE_EPOLL_PER_THREAD,
854 						RTE_INTR_EVENT_ADD,
855 						(void *)((uintptr_t)data));
856 		if (ret)
857 			return ret;
858 	}
859 
860 	return 0;
861 }
862 
863 /* main processing loop */
864 static int
865 main_loop(__attribute__((unused)) void *dummy)
866 {
867 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
868 	unsigned lcore_id;
869 	uint64_t prev_tsc, diff_tsc, cur_tsc;
870 	uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
871 	int i, j, nb_rx;
872 	uint8_t portid, queueid;
873 	struct lcore_conf *qconf;
874 	struct lcore_rx_queue *rx_queue;
875 	enum freq_scale_hint_t lcore_scaleup_hint;
876 	uint32_t lcore_rx_idle_count = 0;
877 	uint32_t lcore_idle_hint = 0;
878 	int intr_en = 0;
879 
880 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
881 
882 	prev_tsc = 0;
883 
884 	lcore_id = rte_lcore_id();
885 	qconf = &lcore_conf[lcore_id];
886 
887 	if (qconf->n_rx_queue == 0) {
888 		RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
889 		return 0;
890 	}
891 
892 	RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
893 
894 	for (i = 0; i < qconf->n_rx_queue; i++) {
895 		portid = qconf->rx_queue_list[i].port_id;
896 		queueid = qconf->rx_queue_list[i].queue_id;
897 		RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%hhu "
898 			"rxqueueid=%hhu\n", lcore_id, portid, queueid);
899 	}
900 
901 	/* add into event wait list */
902 	if (event_register(qconf) == 0)
903 		intr_en = 1;
904 	else
905 		RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
906 
907 	while (1) {
908 		stats[lcore_id].nb_iteration_looped++;
909 
910 		cur_tsc = rte_rdtsc();
911 		cur_tsc_power = cur_tsc;
912 
913 		/*
914 		 * TX burst queue drain
915 		 */
916 		diff_tsc = cur_tsc - prev_tsc;
917 		if (unlikely(diff_tsc > drain_tsc)) {
918 
919 			/*
920 			 * This could be optimized (use queueid instead of
921 			 * portid), but it is not called so often
922 			 */
923 			for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
924 				if (qconf->tx_mbufs[portid].len == 0)
925 					continue;
926 				send_burst(&lcore_conf[lcore_id],
927 					qconf->tx_mbufs[portid].len,
928 					portid);
929 				qconf->tx_mbufs[portid].len = 0;
930 			}
931 
932 			prev_tsc = cur_tsc;
933 		}
934 
935 		diff_tsc_power = cur_tsc_power - prev_tsc_power;
936 		if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
937 			rte_timer_manage();
938 			prev_tsc_power = cur_tsc_power;
939 		}
940 
941 start_rx:
942 		/*
943 		 * Read packet from RX queues
944 		 */
945 		lcore_scaleup_hint = FREQ_CURRENT;
946 		lcore_rx_idle_count = 0;
947 		for (i = 0; i < qconf->n_rx_queue; ++i) {
948 			rx_queue = &(qconf->rx_queue_list[i]);
949 			rx_queue->idle_hint = 0;
950 			portid = rx_queue->port_id;
951 			queueid = rx_queue->queue_id;
952 
953 			nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
954 								MAX_PKT_BURST);
955 
956 			stats[lcore_id].nb_rx_processed += nb_rx;
957 			if (unlikely(nb_rx == 0)) {
958 				/**
959 				 * no packet received from rx queue, try to
960 				 * sleep for a while forcing CPU enter deeper
961 				 * C states.
962 				 */
963 				rx_queue->zero_rx_packet_count++;
964 
965 				if (rx_queue->zero_rx_packet_count <=
966 							MIN_ZERO_POLL_COUNT)
967 					continue;
968 
969 				rx_queue->idle_hint = power_idle_heuristic(\
970 					rx_queue->zero_rx_packet_count);
971 				lcore_rx_idle_count++;
972 			} else {
973 				rx_queue->zero_rx_packet_count = 0;
974 
975 				/**
976 				 * do not scale up frequency immediately as
977 				 * user to kernel space communication is costly
978 				 * which might impact packet I/O for received
979 				 * packets.
980 				 */
981 				rx_queue->freq_up_hint =
982 					power_freq_scaleup_heuristic(lcore_id,
983 							portid, queueid);
984 			}
985 
986 			/* Prefetch first packets */
987 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
988 				rte_prefetch0(rte_pktmbuf_mtod(
989 						pkts_burst[j], void *));
990 			}
991 
992 			/* Prefetch and forward already prefetched packets */
993 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
994 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
995 						j + PREFETCH_OFFSET], void *));
996 				l3fwd_simple_forward(pkts_burst[j], portid,
997 								qconf);
998 			}
999 
1000 			/* Forward remaining prefetched packets */
1001 			for (; j < nb_rx; j++) {
1002 				l3fwd_simple_forward(pkts_burst[j], portid,
1003 								qconf);
1004 			}
1005 		}
1006 
1007 		if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
1008 			for (i = 1, lcore_scaleup_hint =
1009 				qconf->rx_queue_list[0].freq_up_hint;
1010 					i < qconf->n_rx_queue; ++i) {
1011 				rx_queue = &(qconf->rx_queue_list[i]);
1012 				if (rx_queue->freq_up_hint >
1013 						lcore_scaleup_hint)
1014 					lcore_scaleup_hint =
1015 						rx_queue->freq_up_hint;
1016 			}
1017 
1018 			if (lcore_scaleup_hint == FREQ_HIGHEST) {
1019 				if (rte_power_freq_max)
1020 					rte_power_freq_max(lcore_id);
1021 			} else if (lcore_scaleup_hint == FREQ_HIGHER) {
1022 				if (rte_power_freq_up)
1023 					rte_power_freq_up(lcore_id);
1024 			}
1025 		} else {
1026 			/**
1027 			 * All Rx queues empty in recent consecutive polls,
1028 			 * sleep in a conservative manner, meaning sleep as
1029 			 * less as possible.
1030 			 */
1031 			for (i = 1, lcore_idle_hint =
1032 				qconf->rx_queue_list[0].idle_hint;
1033 					i < qconf->n_rx_queue; ++i) {
1034 				rx_queue = &(qconf->rx_queue_list[i]);
1035 				if (rx_queue->idle_hint < lcore_idle_hint)
1036 					lcore_idle_hint = rx_queue->idle_hint;
1037 			}
1038 
1039 			if (lcore_idle_hint < SUSPEND_THRESHOLD)
1040 				/**
1041 				 * execute "pause" instruction to avoid context
1042 				 * switch which generally take hundred of
1043 				 * microseconds for short sleep.
1044 				 */
1045 				rte_delay_us(lcore_idle_hint);
1046 			else {
1047 				/* suspend until rx interrupt trigges */
1048 				if (intr_en) {
1049 					turn_on_intr(qconf);
1050 					sleep_until_rx_interrupt(
1051 						qconf->n_rx_queue);
1052 				}
1053 				/* start receiving packets immediately */
1054 				goto start_rx;
1055 			}
1056 			stats[lcore_id].sleep_time += lcore_idle_hint;
1057 		}
1058 	}
1059 }
1060 
1061 static int
1062 check_lcore_params(void)
1063 {
1064 	uint8_t queue, lcore;
1065 	uint16_t i;
1066 	int socketid;
1067 
1068 	for (i = 0; i < nb_lcore_params; ++i) {
1069 		queue = lcore_params[i].queue_id;
1070 		if (queue >= MAX_RX_QUEUE_PER_PORT) {
1071 			printf("invalid queue number: %hhu\n", queue);
1072 			return -1;
1073 		}
1074 		lcore = lcore_params[i].lcore_id;
1075 		if (!rte_lcore_is_enabled(lcore)) {
1076 			printf("error: lcore %hhu is not enabled in lcore "
1077 							"mask\n", lcore);
1078 			return -1;
1079 		}
1080 		if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1081 							(numa_on == 0)) {
1082 			printf("warning: lcore %hhu is on socket %d with numa "
1083 						"off\n", lcore, socketid);
1084 		}
1085 	}
1086 	return 0;
1087 }
1088 
1089 static int
1090 check_port_config(const unsigned nb_ports)
1091 {
1092 	unsigned portid;
1093 	uint16_t i;
1094 
1095 	for (i = 0; i < nb_lcore_params; ++i) {
1096 		portid = lcore_params[i].port_id;
1097 		if ((enabled_port_mask & (1 << portid)) == 0) {
1098 			printf("port %u is not enabled in port mask\n",
1099 								portid);
1100 			return -1;
1101 		}
1102 		if (portid >= nb_ports) {
1103 			printf("port %u is not present on the board\n",
1104 								portid);
1105 			return -1;
1106 		}
1107 	}
1108 	return 0;
1109 }
1110 
1111 static uint8_t
1112 get_port_n_rx_queues(const uint8_t port)
1113 {
1114 	int queue = -1;
1115 	uint16_t i;
1116 
1117 	for (i = 0; i < nb_lcore_params; ++i) {
1118 		if (lcore_params[i].port_id == port &&
1119 				lcore_params[i].queue_id > queue)
1120 			queue = lcore_params[i].queue_id;
1121 	}
1122 	return (uint8_t)(++queue);
1123 }
1124 
1125 static int
1126 init_lcore_rx_queues(void)
1127 {
1128 	uint16_t i, nb_rx_queue;
1129 	uint8_t lcore;
1130 
1131 	for (i = 0; i < nb_lcore_params; ++i) {
1132 		lcore = lcore_params[i].lcore_id;
1133 		nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1134 		if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1135 			printf("error: too many queues (%u) for lcore: %u\n",
1136 				(unsigned)nb_rx_queue + 1, (unsigned)lcore);
1137 			return -1;
1138 		} else {
1139 			lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1140 				lcore_params[i].port_id;
1141 			lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1142 				lcore_params[i].queue_id;
1143 			lcore_conf[lcore].n_rx_queue++;
1144 		}
1145 	}
1146 	return 0;
1147 }
1148 
1149 /* display usage */
1150 static void
1151 print_usage(const char *prgname)
1152 {
1153 	printf ("%s [EAL options] -- -p PORTMASK -P"
1154 		"  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1155 		"  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1156 		"  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1157 		"  -P : enable promiscuous mode\n"
1158 		"  --config (port,queue,lcore): rx queues configuration\n"
1159 		"  --no-numa: optional, disable numa awareness\n"
1160 		"  --enable-jumbo: enable jumbo frame"
1161 		" which max packet len is PKTLEN in decimal (64-9600)\n",
1162 		prgname);
1163 }
1164 
1165 static int parse_max_pkt_len(const char *pktlen)
1166 {
1167 	char *end = NULL;
1168 	unsigned long len;
1169 
1170 	/* parse decimal string */
1171 	len = strtoul(pktlen, &end, 10);
1172 	if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1173 		return -1;
1174 
1175 	if (len == 0)
1176 		return -1;
1177 
1178 	return len;
1179 }
1180 
1181 static int
1182 parse_portmask(const char *portmask)
1183 {
1184 	char *end = NULL;
1185 	unsigned long pm;
1186 
1187 	/* parse hexadecimal string */
1188 	pm = strtoul(portmask, &end, 16);
1189 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1190 		return -1;
1191 
1192 	if (pm == 0)
1193 		return -1;
1194 
1195 	return pm;
1196 }
1197 
1198 static int
1199 parse_config(const char *q_arg)
1200 {
1201 	char s[256];
1202 	const char *p, *p0 = q_arg;
1203 	char *end;
1204 	enum fieldnames {
1205 		FLD_PORT = 0,
1206 		FLD_QUEUE,
1207 		FLD_LCORE,
1208 		_NUM_FLD
1209 	};
1210 	unsigned long int_fld[_NUM_FLD];
1211 	char *str_fld[_NUM_FLD];
1212 	int i;
1213 	unsigned size;
1214 
1215 	nb_lcore_params = 0;
1216 
1217 	while ((p = strchr(p0,'(')) != NULL) {
1218 		++p;
1219 		if((p0 = strchr(p,')')) == NULL)
1220 			return -1;
1221 
1222 		size = p0 - p;
1223 		if(size >= sizeof(s))
1224 			return -1;
1225 
1226 		snprintf(s, sizeof(s), "%.*s", size, p);
1227 		if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1228 								_NUM_FLD)
1229 			return -1;
1230 		for (i = 0; i < _NUM_FLD; i++){
1231 			errno = 0;
1232 			int_fld[i] = strtoul(str_fld[i], &end, 0);
1233 			if (errno != 0 || end == str_fld[i] || int_fld[i] >
1234 									255)
1235 				return -1;
1236 		}
1237 		if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1238 			printf("exceeded max number of lcore params: %hu\n",
1239 				nb_lcore_params);
1240 			return -1;
1241 		}
1242 		lcore_params_array[nb_lcore_params].port_id =
1243 				(uint8_t)int_fld[FLD_PORT];
1244 		lcore_params_array[nb_lcore_params].queue_id =
1245 				(uint8_t)int_fld[FLD_QUEUE];
1246 		lcore_params_array[nb_lcore_params].lcore_id =
1247 				(uint8_t)int_fld[FLD_LCORE];
1248 		++nb_lcore_params;
1249 	}
1250 	lcore_params = lcore_params_array;
1251 
1252 	return 0;
1253 }
1254 
1255 /* Parse the argument given in the command line of the application */
1256 static int
1257 parse_args(int argc, char **argv)
1258 {
1259 	int opt, ret;
1260 	char **argvopt;
1261 	int option_index;
1262 	char *prgname = argv[0];
1263 	static struct option lgopts[] = {
1264 		{"config", 1, 0, 0},
1265 		{"no-numa", 0, 0, 0},
1266 		{"enable-jumbo", 0, 0, 0},
1267 		{NULL, 0, 0, 0}
1268 	};
1269 
1270 	argvopt = argv;
1271 
1272 	while ((opt = getopt_long(argc, argvopt, "p:P",
1273 				lgopts, &option_index)) != EOF) {
1274 
1275 		switch (opt) {
1276 		/* portmask */
1277 		case 'p':
1278 			enabled_port_mask = parse_portmask(optarg);
1279 			if (enabled_port_mask == 0) {
1280 				printf("invalid portmask\n");
1281 				print_usage(prgname);
1282 				return -1;
1283 			}
1284 			break;
1285 		case 'P':
1286 			printf("Promiscuous mode selected\n");
1287 			promiscuous_on = 1;
1288 			break;
1289 
1290 		/* long options */
1291 		case 0:
1292 			if (!strncmp(lgopts[option_index].name, "config", 6)) {
1293 				ret = parse_config(optarg);
1294 				if (ret) {
1295 					printf("invalid config\n");
1296 					print_usage(prgname);
1297 					return -1;
1298 				}
1299 			}
1300 
1301 			if (!strncmp(lgopts[option_index].name,
1302 						"no-numa", 7)) {
1303 				printf("numa is disabled \n");
1304 				numa_on = 0;
1305 			}
1306 
1307 			if (!strncmp(lgopts[option_index].name,
1308 					"enable-jumbo", 12)) {
1309 				struct option lenopts =
1310 					{"max-pkt-len", required_argument, \
1311 									0, 0};
1312 
1313 				printf("jumbo frame is enabled \n");
1314 				port_conf.rxmode.jumbo_frame = 1;
1315 
1316 				/**
1317 				 * if no max-pkt-len set, use the default value
1318 				 * ETHER_MAX_LEN
1319 				 */
1320 				if (0 == getopt_long(argc, argvopt, "",
1321 						&lenopts, &option_index)) {
1322 					ret = parse_max_pkt_len(optarg);
1323 					if ((ret < 64) ||
1324 						(ret > MAX_JUMBO_PKT_LEN)){
1325 						printf("invalid packet "
1326 								"length\n");
1327 						print_usage(prgname);
1328 						return -1;
1329 					}
1330 					port_conf.rxmode.max_rx_pkt_len = ret;
1331 				}
1332 				printf("set jumbo frame "
1333 					"max packet length to %u\n",
1334 				(unsigned int)port_conf.rxmode.max_rx_pkt_len);
1335 			}
1336 
1337 			break;
1338 
1339 		default:
1340 			print_usage(prgname);
1341 			return -1;
1342 		}
1343 	}
1344 
1345 	if (optind >= 0)
1346 		argv[optind-1] = prgname;
1347 
1348 	ret = optind-1;
1349 	optind = 0; /* reset getopt lib */
1350 	return ret;
1351 }
1352 
1353 static void
1354 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1355 {
1356 	char buf[ETHER_ADDR_FMT_SIZE];
1357 	ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1358 	printf("%s%s", name, buf);
1359 }
1360 
1361 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1362 static void
1363 setup_hash(int socketid)
1364 {
1365 	struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1366 		.name = NULL,
1367 		.entries = L3FWD_HASH_ENTRIES,
1368 		.key_len = sizeof(struct ipv4_5tuple),
1369 		.hash_func = DEFAULT_HASH_FUNC,
1370 		.hash_func_init_val = 0,
1371 	};
1372 
1373 	struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1374 		.name = NULL,
1375 		.entries = L3FWD_HASH_ENTRIES,
1376 		.key_len = sizeof(struct ipv6_5tuple),
1377 		.hash_func = DEFAULT_HASH_FUNC,
1378 		.hash_func_init_val = 0,
1379 	};
1380 
1381 	unsigned i;
1382 	int ret;
1383 	char s[64];
1384 
1385 	/* create ipv4 hash */
1386 	rte_snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1387 	ipv4_l3fwd_hash_params.name = s;
1388 	ipv4_l3fwd_hash_params.socket_id = socketid;
1389 	ipv4_l3fwd_lookup_struct[socketid] =
1390 		rte_hash_create(&ipv4_l3fwd_hash_params);
1391 	if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1392 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1393 				"socket %d\n", socketid);
1394 
1395 	/* create ipv6 hash */
1396 	rte_snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1397 	ipv6_l3fwd_hash_params.name = s;
1398 	ipv6_l3fwd_hash_params.socket_id = socketid;
1399 	ipv6_l3fwd_lookup_struct[socketid] =
1400 		rte_hash_create(&ipv6_l3fwd_hash_params);
1401 	if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1402 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1403 				"socket %d\n", socketid);
1404 
1405 
1406 	/* populate the ipv4 hash */
1407 	for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1408 		ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1409 				(void *) &ipv4_l3fwd_route_array[i].key);
1410 		if (ret < 0) {
1411 			rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1412 				"l3fwd hash on socket %d\n", i, socketid);
1413 		}
1414 		ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1415 		printf("Hash: Adding key\n");
1416 		print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1417 	}
1418 
1419 	/* populate the ipv6 hash */
1420 	for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1421 		ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1422 				(void *) &ipv6_l3fwd_route_array[i].key);
1423 		if (ret < 0) {
1424 			rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1425 				"l3fwd hash on socket %d\n", i, socketid);
1426 		}
1427 		ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1428 		printf("Hash: Adding key\n");
1429 		print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1430 	}
1431 }
1432 #endif
1433 
1434 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1435 static void
1436 setup_lpm(int socketid)
1437 {
1438 	unsigned i;
1439 	int ret;
1440 	char s[64];
1441 
1442 	/* create the LPM table */
1443 	snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1444 	ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1445 				IPV4_L3FWD_LPM_MAX_RULES, 0);
1446 	if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1447 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1448 				" on socket %d\n", socketid);
1449 
1450 	/* populate the LPM table */
1451 	for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1452 		ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1453 			ipv4_l3fwd_route_array[i].ip,
1454 			ipv4_l3fwd_route_array[i].depth,
1455 			ipv4_l3fwd_route_array[i].if_out);
1456 
1457 		if (ret < 0) {
1458 			rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1459 				"l3fwd LPM table on socket %d\n",
1460 				i, socketid);
1461 		}
1462 
1463 		printf("LPM: Adding route 0x%08x / %d (%d)\n",
1464 			(unsigned)ipv4_l3fwd_route_array[i].ip,
1465 			ipv4_l3fwd_route_array[i].depth,
1466 			ipv4_l3fwd_route_array[i].if_out);
1467 	}
1468 }
1469 #endif
1470 
1471 static int
1472 init_mem(unsigned nb_mbuf)
1473 {
1474 	struct lcore_conf *qconf;
1475 	int socketid;
1476 	unsigned lcore_id;
1477 	char s[64];
1478 
1479 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1480 		if (rte_lcore_is_enabled(lcore_id) == 0)
1481 			continue;
1482 
1483 		if (numa_on)
1484 			socketid = rte_lcore_to_socket_id(lcore_id);
1485 		else
1486 			socketid = 0;
1487 
1488 		if (socketid >= NB_SOCKETS) {
1489 			rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1490 					"out of range %d\n", socketid,
1491 						lcore_id, NB_SOCKETS);
1492 		}
1493 		if (pktmbuf_pool[socketid] == NULL) {
1494 			snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1495 			pktmbuf_pool[socketid] =
1496 				rte_pktmbuf_pool_create(s, nb_mbuf,
1497 					MEMPOOL_CACHE_SIZE, 0,
1498 					RTE_MBUF_DEFAULT_BUF_SIZE,
1499 					socketid);
1500 			if (pktmbuf_pool[socketid] == NULL)
1501 				rte_exit(EXIT_FAILURE,
1502 					"Cannot init mbuf pool on socket %d\n",
1503 								socketid);
1504 			else
1505 				printf("Allocated mbuf pool on socket %d\n",
1506 								socketid);
1507 
1508 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1509 			setup_lpm(socketid);
1510 #else
1511 			setup_hash(socketid);
1512 #endif
1513 		}
1514 		qconf = &lcore_conf[lcore_id];
1515 		qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1516 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1517 		qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1518 #endif
1519 	}
1520 	return 0;
1521 }
1522 
1523 /* Check the link status of all ports in up to 9s, and print them finally */
1524 static void
1525 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1526 {
1527 #define CHECK_INTERVAL 100 /* 100ms */
1528 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1529 	uint8_t portid, count, all_ports_up, print_flag = 0;
1530 	struct rte_eth_link link;
1531 
1532 	printf("\nChecking link status");
1533 	fflush(stdout);
1534 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
1535 		all_ports_up = 1;
1536 		for (portid = 0; portid < port_num; portid++) {
1537 			if ((port_mask & (1 << portid)) == 0)
1538 				continue;
1539 			memset(&link, 0, sizeof(link));
1540 			rte_eth_link_get_nowait(portid, &link);
1541 			/* print link status if flag set */
1542 			if (print_flag == 1) {
1543 				if (link.link_status)
1544 					printf("Port %d Link Up - speed %u "
1545 						"Mbps - %s\n", (uint8_t)portid,
1546 						(unsigned)link.link_speed,
1547 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1548 					("full-duplex") : ("half-duplex\n"));
1549 				else
1550 					printf("Port %d Link Down\n",
1551 						(uint8_t)portid);
1552 				continue;
1553 			}
1554 			/* clear all_ports_up flag if any link down */
1555 			if (link.link_status == 0) {
1556 				all_ports_up = 0;
1557 				break;
1558 			}
1559 		}
1560 		/* after finally printing all link status, get out */
1561 		if (print_flag == 1)
1562 			break;
1563 
1564 		if (all_ports_up == 0) {
1565 			printf(".");
1566 			fflush(stdout);
1567 			rte_delay_ms(CHECK_INTERVAL);
1568 		}
1569 
1570 		/* set the print_flag if all ports up or timeout */
1571 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1572 			print_flag = 1;
1573 			printf("done\n");
1574 		}
1575 	}
1576 }
1577 
1578 int
1579 main(int argc, char **argv)
1580 {
1581 	struct lcore_conf *qconf;
1582 	struct rte_eth_dev_info dev_info;
1583 	struct rte_eth_txconf *txconf;
1584 	int ret;
1585 	unsigned nb_ports;
1586 	uint16_t queueid;
1587 	unsigned lcore_id;
1588 	uint64_t hz;
1589 	uint32_t n_tx_queue, nb_lcores;
1590 	uint32_t dev_rxq_num, dev_txq_num;
1591 	uint8_t portid, nb_rx_queue, queue, socketid;
1592 
1593 	/* catch SIGINT and restore cpufreq governor to ondemand */
1594 	signal(SIGINT, signal_exit_now);
1595 
1596 	/* init EAL */
1597 	ret = rte_eal_init(argc, argv);
1598 	if (ret < 0)
1599 		rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1600 	argc -= ret;
1601 	argv += ret;
1602 
1603 	/* init RTE timer library to be used late */
1604 	rte_timer_subsystem_init();
1605 
1606 	/* parse application arguments (after the EAL ones) */
1607 	ret = parse_args(argc, argv);
1608 	if (ret < 0)
1609 		rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1610 
1611 	if (check_lcore_params() < 0)
1612 		rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1613 
1614 	ret = init_lcore_rx_queues();
1615 	if (ret < 0)
1616 		rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1617 
1618 
1619 	nb_ports = rte_eth_dev_count();
1620 	if (nb_ports > RTE_MAX_ETHPORTS)
1621 		nb_ports = RTE_MAX_ETHPORTS;
1622 
1623 	if (check_port_config(nb_ports) < 0)
1624 		rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1625 
1626 	nb_lcores = rte_lcore_count();
1627 
1628 	/* initialize all ports */
1629 	for (portid = 0; portid < nb_ports; portid++) {
1630 		/* skip ports that are not enabled */
1631 		if ((enabled_port_mask & (1 << portid)) == 0) {
1632 			printf("\nSkipping disabled port %d\n", portid);
1633 			continue;
1634 		}
1635 
1636 		/* init port */
1637 		printf("Initializing port %d ... ", portid );
1638 		fflush(stdout);
1639 
1640 		rte_eth_dev_info_get(portid, &dev_info);
1641 		dev_rxq_num = dev_info.max_rx_queues;
1642 		dev_txq_num = dev_info.max_tx_queues;
1643 
1644 		nb_rx_queue = get_port_n_rx_queues(portid);
1645 		if (nb_rx_queue > dev_rxq_num)
1646 			rte_exit(EXIT_FAILURE,
1647 				"Cannot configure not existed rxq: "
1648 				"port=%d\n", portid);
1649 
1650 		n_tx_queue = nb_lcores;
1651 		if (n_tx_queue > dev_txq_num)
1652 			n_tx_queue = dev_txq_num;
1653 		printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1654 			nb_rx_queue, (unsigned)n_tx_queue );
1655 		ret = rte_eth_dev_configure(portid, nb_rx_queue,
1656 					(uint16_t)n_tx_queue, &port_conf);
1657 		if (ret < 0)
1658 			rte_exit(EXIT_FAILURE, "Cannot configure device: "
1659 					"err=%d, port=%d\n", ret, portid);
1660 
1661 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1662 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
1663 		printf(", ");
1664 
1665 		/* init memory */
1666 		ret = init_mem(NB_MBUF);
1667 		if (ret < 0)
1668 			rte_exit(EXIT_FAILURE, "init_mem failed\n");
1669 
1670 		/* init one TX queue per couple (lcore,port) */
1671 		queueid = 0;
1672 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1673 			if (rte_lcore_is_enabled(lcore_id) == 0)
1674 				continue;
1675 
1676 			if (queueid >= dev_txq_num)
1677 				continue;
1678 
1679 			if (numa_on)
1680 				socketid = \
1681 				(uint8_t)rte_lcore_to_socket_id(lcore_id);
1682 			else
1683 				socketid = 0;
1684 
1685 			printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1686 			fflush(stdout);
1687 
1688 			rte_eth_dev_info_get(portid, &dev_info);
1689 			txconf = &dev_info.default_txconf;
1690 			if (port_conf.rxmode.jumbo_frame)
1691 				txconf->txq_flags = 0;
1692 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1693 						     socketid, txconf);
1694 			if (ret < 0)
1695 				rte_exit(EXIT_FAILURE,
1696 					"rte_eth_tx_queue_setup: err=%d, "
1697 						"port=%d\n", ret, portid);
1698 
1699 			qconf = &lcore_conf[lcore_id];
1700 			qconf->tx_queue_id[portid] = queueid;
1701 			queueid++;
1702 		}
1703 		printf("\n");
1704 	}
1705 
1706 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1707 		if (rte_lcore_is_enabled(lcore_id) == 0)
1708 			continue;
1709 
1710 		/* init power management library */
1711 		ret = rte_power_init(lcore_id);
1712 		if (ret)
1713 			rte_log(RTE_LOG_ERR, RTE_LOGTYPE_POWER,
1714 				"Power management library initialization "
1715 				"failed on core%u", lcore_id);
1716 
1717 		/* init timer structures for each enabled lcore */
1718 		rte_timer_init(&power_timers[lcore_id]);
1719 		hz = rte_get_timer_hz();
1720 		rte_timer_reset(&power_timers[lcore_id],
1721 			hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1722 						power_timer_cb, NULL);
1723 
1724 		qconf = &lcore_conf[lcore_id];
1725 		printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1726 		fflush(stdout);
1727 		/* init RX queues */
1728 		for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1729 			portid = qconf->rx_queue_list[queue].port_id;
1730 			queueid = qconf->rx_queue_list[queue].queue_id;
1731 
1732 			if (numa_on)
1733 				socketid = \
1734 				(uint8_t)rte_lcore_to_socket_id(lcore_id);
1735 			else
1736 				socketid = 0;
1737 
1738 			printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1739 			fflush(stdout);
1740 
1741 			ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1742 				socketid, NULL,
1743 				pktmbuf_pool[socketid]);
1744 			if (ret < 0)
1745 				rte_exit(EXIT_FAILURE,
1746 					"rte_eth_rx_queue_setup: err=%d, "
1747 						"port=%d\n", ret, portid);
1748 		}
1749 	}
1750 
1751 	printf("\n");
1752 
1753 	/* start ports */
1754 	for (portid = 0; portid < nb_ports; portid++) {
1755 		if ((enabled_port_mask & (1 << portid)) == 0) {
1756 			continue;
1757 		}
1758 		/* Start device */
1759 		ret = rte_eth_dev_start(portid);
1760 		if (ret < 0)
1761 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1762 						"port=%d\n", ret, portid);
1763 		/*
1764 		 * If enabled, put device in promiscuous mode.
1765 		 * This allows IO forwarding mode to forward packets
1766 		 * to itself through 2 cross-connected  ports of the
1767 		 * target machine.
1768 		 */
1769 		if (promiscuous_on)
1770 			rte_eth_promiscuous_enable(portid);
1771 		/* initialize spinlock for each port */
1772 		rte_spinlock_init(&(locks[portid]));
1773 	}
1774 
1775 	check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1776 
1777 	/* launch per-lcore init on every lcore */
1778 	rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1779 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1780 		if (rte_eal_wait_lcore(lcore_id) < 0)
1781 			return -1;
1782 	}
1783 
1784 	return 0;
1785 }
1786