xref: /dpdk/examples/l3fwd-power/main.c (revision 8205e241b2b01c05f2cffe5158c053d614d1f68c)
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 		.rxq = 1,
243 	},
244 };
245 
246 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
247 
248 
249 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
250 
251 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
252 #include <rte_hash_crc.h>
253 #define DEFAULT_HASH_FUNC       rte_hash_crc
254 #else
255 #include <rte_jhash.h>
256 #define DEFAULT_HASH_FUNC       rte_jhash
257 #endif
258 
259 struct ipv4_5tuple {
260 	uint32_t ip_dst;
261 	uint32_t ip_src;
262 	uint16_t port_dst;
263 	uint16_t port_src;
264 	uint8_t  proto;
265 } __attribute__((__packed__));
266 
267 struct ipv6_5tuple {
268 	uint8_t  ip_dst[IPV6_ADDR_LEN];
269 	uint8_t  ip_src[IPV6_ADDR_LEN];
270 	uint16_t port_dst;
271 	uint16_t port_src;
272 	uint8_t  proto;
273 } __attribute__((__packed__));
274 
275 struct ipv4_l3fwd_route {
276 	struct ipv4_5tuple key;
277 	uint8_t if_out;
278 };
279 
280 struct ipv6_l3fwd_route {
281 	struct ipv6_5tuple key;
282 	uint8_t if_out;
283 };
284 
285 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
286 	{{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
287 	{{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
288 	{{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
289 	{{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
290 };
291 
292 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
293 	{
294 		{
295 			{0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
296 			 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
297 			{0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
298 			 0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
299 			 1, 10, IPPROTO_UDP
300 		}, 4
301 	},
302 };
303 
304 typedef struct rte_hash lookup_struct_t;
305 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
306 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
307 
308 #define L3FWD_HASH_ENTRIES	1024
309 
310 #define IPV4_L3FWD_NUM_ROUTES \
311 	(sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
312 
313 #define IPV6_L3FWD_NUM_ROUTES \
314 	(sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
315 
316 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
317 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
318 #endif
319 
320 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
321 struct ipv4_l3fwd_route {
322 	uint32_t ip;
323 	uint8_t  depth;
324 	uint8_t  if_out;
325 };
326 
327 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
328 	{IPv4(1,1,1,0), 24, 0},
329 	{IPv4(2,1,1,0), 24, 1},
330 	{IPv4(3,1,1,0), 24, 2},
331 	{IPv4(4,1,1,0), 24, 3},
332 	{IPv4(5,1,1,0), 24, 4},
333 	{IPv4(6,1,1,0), 24, 5},
334 	{IPv4(7,1,1,0), 24, 6},
335 	{IPv4(8,1,1,0), 24, 7},
336 };
337 
338 #define IPV4_L3FWD_NUM_ROUTES \
339 	(sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
340 
341 #define IPV4_L3FWD_LPM_MAX_RULES     1024
342 
343 typedef struct rte_lpm lookup_struct_t;
344 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
345 #endif
346 
347 struct lcore_conf {
348 	uint16_t n_rx_queue;
349 	struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
350 	uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
351 	struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
352 	lookup_struct_t * ipv4_lookup_struct;
353 	lookup_struct_t * ipv6_lookup_struct;
354 } __rte_cache_aligned;
355 
356 struct lcore_stats {
357 	/* total sleep time in ms since last frequency scaling down */
358 	uint32_t sleep_time;
359 	/* number of long sleep recently */
360 	uint32_t nb_long_sleep;
361 	/* freq. scaling up trend */
362 	uint32_t trend;
363 	/* total packet processed recently */
364 	uint64_t nb_rx_processed;
365 	/* total iterations looped recently */
366 	uint64_t nb_iteration_looped;
367 	uint32_t padding[9];
368 } __rte_cache_aligned;
369 
370 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
371 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
372 static struct rte_timer power_timers[RTE_MAX_LCORE];
373 
374 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
375 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
376 			unsigned lcore_id, uint8_t port_id, uint16_t queue_id);
377 
378 /* exit signal handler */
379 static void
380 signal_exit_now(int sigtype)
381 {
382 	unsigned lcore_id;
383 	int ret;
384 
385 	if (sigtype == SIGINT) {
386 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
387 			if (rte_lcore_is_enabled(lcore_id) == 0)
388 				continue;
389 
390 			/* init power management library */
391 			ret = rte_power_exit(lcore_id);
392 			if (ret)
393 				rte_exit(EXIT_FAILURE, "Power management "
394 					"library de-initialization failed on "
395 							"core%u\n", lcore_id);
396 		}
397 	}
398 
399 	rte_exit(EXIT_SUCCESS, "User forced exit\n");
400 }
401 
402 /*  Freqency scale down timer callback */
403 static void
404 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
405 			  __attribute__((unused)) void *arg)
406 {
407 	uint64_t hz;
408 	float sleep_time_ratio;
409 	unsigned lcore_id = rte_lcore_id();
410 
411 	/* accumulate total execution time in us when callback is invoked */
412 	sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
413 					(float)SCALING_PERIOD;
414 	/**
415 	 * check whether need to scale down frequency a step if it sleep a lot.
416 	 */
417 	if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
418 		if (rte_power_freq_down)
419 			rte_power_freq_down(lcore_id);
420 	}
421 	else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
422 		stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
423 		/**
424 		 * scale down a step if average packet per iteration less
425 		 * than expectation.
426 		 */
427 		if (rte_power_freq_down)
428 			rte_power_freq_down(lcore_id);
429 	}
430 
431 	/**
432 	 * initialize another timer according to current frequency to ensure
433 	 * timer interval is relatively fixed.
434 	 */
435 	hz = rte_get_timer_hz();
436 	rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
437 				SINGLE, lcore_id, power_timer_cb, NULL);
438 
439 	stats[lcore_id].nb_rx_processed = 0;
440 	stats[lcore_id].nb_iteration_looped = 0;
441 
442 	stats[lcore_id].sleep_time = 0;
443 }
444 
445 /* Send burst of packets on an output interface */
446 static inline int
447 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
448 {
449 	struct rte_mbuf **m_table;
450 	int ret;
451 	uint16_t queueid;
452 
453 	queueid = qconf->tx_queue_id[port];
454 	m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
455 
456 	ret = rte_eth_tx_burst(port, queueid, m_table, n);
457 	if (unlikely(ret < n)) {
458 		do {
459 			rte_pktmbuf_free(m_table[ret]);
460 		} while (++ret < n);
461 	}
462 
463 	return 0;
464 }
465 
466 /* Enqueue a single packet, and send burst if queue is filled */
467 static inline int
468 send_single_packet(struct rte_mbuf *m, uint8_t port)
469 {
470 	uint32_t lcore_id;
471 	uint16_t len;
472 	struct lcore_conf *qconf;
473 
474 	lcore_id = rte_lcore_id();
475 
476 	qconf = &lcore_conf[lcore_id];
477 	len = qconf->tx_mbufs[port].len;
478 	qconf->tx_mbufs[port].m_table[len] = m;
479 	len++;
480 
481 	/* enough pkts to be sent */
482 	if (unlikely(len == MAX_PKT_BURST)) {
483 		send_burst(qconf, MAX_PKT_BURST, port);
484 		len = 0;
485 	}
486 
487 	qconf->tx_mbufs[port].len = len;
488 	return 0;
489 }
490 
491 #ifdef DO_RFC_1812_CHECKS
492 static inline int
493 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
494 {
495 	/* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
496 	/*
497 	 * 1. The packet length reported by the Link Layer must be large
498 	 * enough to hold the minimum length legal IP datagram (20 bytes).
499 	 */
500 	if (link_len < sizeof(struct ipv4_hdr))
501 		return -1;
502 
503 	/* 2. The IP checksum must be correct. */
504 	/* this is checked in H/W */
505 
506 	/*
507 	 * 3. The IP version number must be 4. If the version number is not 4
508 	 * then the packet may be another version of IP, such as IPng or
509 	 * ST-II.
510 	 */
511 	if (((pkt->version_ihl) >> 4) != 4)
512 		return -3;
513 	/*
514 	 * 4. The IP header length field must be large enough to hold the
515 	 * minimum length legal IP datagram (20 bytes = 5 words).
516 	 */
517 	if ((pkt->version_ihl & 0xf) < 5)
518 		return -4;
519 
520 	/*
521 	 * 5. The IP total length field must be large enough to hold the IP
522 	 * datagram header, whose length is specified in the IP header length
523 	 * field.
524 	 */
525 	if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
526 		return -5;
527 
528 	return 0;
529 }
530 #endif
531 
532 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
533 static void
534 print_ipv4_key(struct ipv4_5tuple key)
535 {
536 	printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
537 		"proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
538 				key.port_dst, key.port_src, key.proto);
539 }
540 static void
541 print_ipv6_key(struct ipv6_5tuple key)
542 {
543 	printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
544 	        "port dst = %d, port src = %d, proto = %d\n",
545 	        IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
546 	        key.port_dst, key.port_src, key.proto);
547 }
548 
549 static inline uint8_t
550 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
551 		lookup_struct_t * ipv4_l3fwd_lookup_struct)
552 {
553 	struct ipv4_5tuple key;
554 	struct tcp_hdr *tcp;
555 	struct udp_hdr *udp;
556 	int ret = 0;
557 
558 	key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
559 	key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
560 	key.proto = ipv4_hdr->next_proto_id;
561 
562 	switch (ipv4_hdr->next_proto_id) {
563 	case IPPROTO_TCP:
564 		tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
565 					sizeof(struct ipv4_hdr));
566 		key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
567 		key.port_src = rte_be_to_cpu_16(tcp->src_port);
568 		break;
569 
570 	case IPPROTO_UDP:
571 		udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
572 					sizeof(struct ipv4_hdr));
573 		key.port_dst = rte_be_to_cpu_16(udp->dst_port);
574 		key.port_src = rte_be_to_cpu_16(udp->src_port);
575 		break;
576 
577 	default:
578 		key.port_dst = 0;
579 		key.port_src = 0;
580 		break;
581 	}
582 
583 	/* Find destination port */
584 	ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
585 	return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
586 }
587 
588 static inline uint8_t
589 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr,  uint8_t portid,
590 			lookup_struct_t *ipv6_l3fwd_lookup_struct)
591 {
592 	struct ipv6_5tuple key;
593 	struct tcp_hdr *tcp;
594 	struct udp_hdr *udp;
595 	int ret = 0;
596 
597 	memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
598 	memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
599 
600 	key.proto = ipv6_hdr->proto;
601 
602 	switch (ipv6_hdr->proto) {
603 	case IPPROTO_TCP:
604 		tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
605 					sizeof(struct ipv6_hdr));
606 		key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
607 		key.port_src = rte_be_to_cpu_16(tcp->src_port);
608 		break;
609 
610 	case IPPROTO_UDP:
611 		udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
612 					sizeof(struct ipv6_hdr));
613 		key.port_dst = rte_be_to_cpu_16(udp->dst_port);
614 		key.port_src = rte_be_to_cpu_16(udp->src_port);
615 		break;
616 
617 	default:
618 		key.port_dst = 0;
619 		key.port_src = 0;
620 		break;
621 	}
622 
623 	/* Find destination port */
624 	ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
625 	return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
626 }
627 #endif
628 
629 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
630 static inline uint8_t
631 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
632 		lookup_struct_t *ipv4_l3fwd_lookup_struct)
633 {
634 	uint8_t next_hop;
635 
636 	return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
637 			rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
638 			next_hop : portid);
639 }
640 #endif
641 
642 static inline void
643 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
644 				struct lcore_conf *qconf)
645 {
646 	struct ether_hdr *eth_hdr;
647 	struct ipv4_hdr *ipv4_hdr;
648 	void *d_addr_bytes;
649 	uint8_t dst_port;
650 
651 	eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
652 
653 	if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
654 		/* Handle IPv4 headers.*/
655 		ipv4_hdr =
656 			rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
657 						sizeof(struct ether_hdr));
658 
659 #ifdef DO_RFC_1812_CHECKS
660 		/* Check to make sure the packet is valid (RFC1812) */
661 		if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
662 			rte_pktmbuf_free(m);
663 			return;
664 		}
665 #endif
666 
667 		dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
668 					qconf->ipv4_lookup_struct);
669 		if (dst_port >= RTE_MAX_ETHPORTS ||
670 				(enabled_port_mask & 1 << dst_port) == 0)
671 			dst_port = portid;
672 
673 		/* 02:00:00:00:00:xx */
674 		d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
675 		*((uint64_t *)d_addr_bytes) =
676 			0x000000000002 + ((uint64_t)dst_port << 40);
677 
678 #ifdef DO_RFC_1812_CHECKS
679 		/* Update time to live and header checksum */
680 		--(ipv4_hdr->time_to_live);
681 		++(ipv4_hdr->hdr_checksum);
682 #endif
683 
684 		/* src addr */
685 		ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
686 
687 		send_single_packet(m, dst_port);
688 	} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
689 		/* Handle IPv6 headers.*/
690 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
691 		struct ipv6_hdr *ipv6_hdr;
692 
693 		ipv6_hdr =
694 			rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
695 						sizeof(struct ether_hdr));
696 
697 		dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
698 					qconf->ipv6_lookup_struct);
699 
700 		if (dst_port >= RTE_MAX_ETHPORTS ||
701 				(enabled_port_mask & 1 << dst_port) == 0)
702 			dst_port = portid;
703 
704 		/* 02:00:00:00:00:xx */
705 		d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
706 		*((uint64_t *)d_addr_bytes) =
707 			0x000000000002 + ((uint64_t)dst_port << 40);
708 
709 		/* src addr */
710 		ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
711 
712 		send_single_packet(m, dst_port);
713 #else
714 		/* We don't currently handle IPv6 packets in LPM mode. */
715 		rte_pktmbuf_free(m);
716 #endif
717 	}
718 
719 }
720 
721 #define MINIMUM_SLEEP_TIME         1
722 #define SUSPEND_THRESHOLD          300
723 
724 static inline uint32_t
725 power_idle_heuristic(uint32_t zero_rx_packet_count)
726 {
727 	/* If zero count is less than 100,  sleep 1us */
728 	if (zero_rx_packet_count < SUSPEND_THRESHOLD)
729 		return MINIMUM_SLEEP_TIME;
730 	/* If zero count is less than 1000, sleep 100 us which is the
731 		minimum latency switching from C3/C6 to C0
732 	*/
733 	else
734 		return SUSPEND_THRESHOLD;
735 
736 	return 0;
737 }
738 
739 static inline enum freq_scale_hint_t
740 power_freq_scaleup_heuristic(unsigned lcore_id,
741 			     uint8_t port_id,
742 			     uint16_t queue_id)
743 {
744 /**
745  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
746  * per iteration
747  */
748 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
749 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
750 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
751 #define FREQ_UP_TREND1_ACC   1
752 #define FREQ_UP_TREND2_ACC   100
753 #define FREQ_UP_THRESHOLD    10000
754 
755 	if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
756 			FREQ_GEAR3_RX_PACKET_THRESHOLD) > 0)) {
757 		stats[lcore_id].trend = 0;
758 		return FREQ_HIGHEST;
759 	} else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
760 			FREQ_GEAR2_RX_PACKET_THRESHOLD) > 0))
761 		stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
762 	else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
763 			FREQ_GEAR1_RX_PACKET_THRESHOLD) > 0))
764 		stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
765 
766 	if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
767 		stats[lcore_id].trend = 0;
768 		return FREQ_HIGHER;
769 	}
770 
771 	return FREQ_CURRENT;
772 }
773 
774 /**
775  * force polling thread sleep until one-shot rx interrupt triggers
776  * @param port_id
777  *  Port id.
778  * @param queue_id
779  *  Rx queue id.
780  * @return
781  *  0 on success
782  */
783 static int
784 sleep_until_rx_interrupt(int num)
785 {
786 	struct rte_epoll_event event[num];
787 	int n, i;
788 	uint8_t port_id, queue_id;
789 	void *data;
790 
791 	RTE_LOG(INFO, L3FWD_POWER,
792 		"lcore %u sleeps until interrupt triggers\n",
793 		rte_lcore_id());
794 
795 	n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
796 	for (i = 0; i < n; i++) {
797 		data = event[i].epdata.data;
798 		port_id = ((uintptr_t)data) >> CHAR_BIT;
799 		queue_id = ((uintptr_t)data) &
800 			RTE_LEN2MASK(CHAR_BIT, uint8_t);
801 		rte_eth_dev_rx_intr_disable(port_id, queue_id);
802 		RTE_LOG(INFO, L3FWD_POWER,
803 			"lcore %u is waked up from rx interrupt on"
804 			" port %d queue %d\n",
805 			rte_lcore_id(), port_id, queue_id);
806 	}
807 
808 	return 0;
809 }
810 
811 static void turn_on_intr(struct lcore_conf *qconf)
812 {
813 	int i;
814 	struct lcore_rx_queue *rx_queue;
815 	uint8_t port_id, queue_id;
816 
817 	for (i = 0; i < qconf->n_rx_queue; ++i) {
818 		rx_queue = &(qconf->rx_queue_list[i]);
819 		port_id = rx_queue->port_id;
820 		queue_id = rx_queue->queue_id;
821 
822 		rte_spinlock_lock(&(locks[port_id]));
823 		rte_eth_dev_rx_intr_enable(port_id, queue_id);
824 		rte_spinlock_unlock(&(locks[port_id]));
825 	}
826 }
827 
828 static int event_register(struct lcore_conf *qconf)
829 {
830 	struct lcore_rx_queue *rx_queue;
831 	uint8_t portid, queueid;
832 	uint32_t data;
833 	int ret;
834 	int i;
835 
836 	for (i = 0; i < qconf->n_rx_queue; ++i) {
837 		rx_queue = &(qconf->rx_queue_list[i]);
838 		portid = rx_queue->port_id;
839 		queueid = rx_queue->queue_id;
840 		data = portid << CHAR_BIT | queueid;
841 
842 		ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
843 						RTE_EPOLL_PER_THREAD,
844 						RTE_INTR_EVENT_ADD,
845 						(void *)((uintptr_t)data));
846 		if (ret)
847 			return ret;
848 	}
849 
850 	return 0;
851 }
852 
853 /* main processing loop */
854 static int
855 main_loop(__attribute__((unused)) void *dummy)
856 {
857 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
858 	unsigned lcore_id;
859 	uint64_t prev_tsc, diff_tsc, cur_tsc;
860 	uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
861 	int i, j, nb_rx;
862 	uint8_t portid, queueid;
863 	struct lcore_conf *qconf;
864 	struct lcore_rx_queue *rx_queue;
865 	enum freq_scale_hint_t lcore_scaleup_hint;
866 	uint32_t lcore_rx_idle_count = 0;
867 	uint32_t lcore_idle_hint = 0;
868 	int intr_en = 0;
869 
870 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
871 
872 	prev_tsc = 0;
873 
874 	lcore_id = rte_lcore_id();
875 	qconf = &lcore_conf[lcore_id];
876 
877 	if (qconf->n_rx_queue == 0) {
878 		RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
879 		return 0;
880 	}
881 
882 	RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
883 
884 	for (i = 0; i < qconf->n_rx_queue; i++) {
885 		portid = qconf->rx_queue_list[i].port_id;
886 		queueid = qconf->rx_queue_list[i].queue_id;
887 		RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%hhu "
888 			"rxqueueid=%hhu\n", lcore_id, portid, queueid);
889 	}
890 
891 	/* add into event wait list */
892 	if (event_register(qconf) == 0)
893 		intr_en = 1;
894 	else
895 		RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
896 
897 	while (1) {
898 		stats[lcore_id].nb_iteration_looped++;
899 
900 		cur_tsc = rte_rdtsc();
901 		cur_tsc_power = cur_tsc;
902 
903 		/*
904 		 * TX burst queue drain
905 		 */
906 		diff_tsc = cur_tsc - prev_tsc;
907 		if (unlikely(diff_tsc > drain_tsc)) {
908 
909 			/*
910 			 * This could be optimized (use queueid instead of
911 			 * portid), but it is not called so often
912 			 */
913 			for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
914 				if (qconf->tx_mbufs[portid].len == 0)
915 					continue;
916 				send_burst(&lcore_conf[lcore_id],
917 					qconf->tx_mbufs[portid].len,
918 					portid);
919 				qconf->tx_mbufs[portid].len = 0;
920 			}
921 
922 			prev_tsc = cur_tsc;
923 		}
924 
925 		diff_tsc_power = cur_tsc_power - prev_tsc_power;
926 		if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
927 			rte_timer_manage();
928 			prev_tsc_power = cur_tsc_power;
929 		}
930 
931 start_rx:
932 		/*
933 		 * Read packet from RX queues
934 		 */
935 		lcore_scaleup_hint = FREQ_CURRENT;
936 		lcore_rx_idle_count = 0;
937 		for (i = 0; i < qconf->n_rx_queue; ++i) {
938 			rx_queue = &(qconf->rx_queue_list[i]);
939 			rx_queue->idle_hint = 0;
940 			portid = rx_queue->port_id;
941 			queueid = rx_queue->queue_id;
942 
943 			nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
944 								MAX_PKT_BURST);
945 
946 			stats[lcore_id].nb_rx_processed += nb_rx;
947 			if (unlikely(nb_rx == 0)) {
948 				/**
949 				 * no packet received from rx queue, try to
950 				 * sleep for a while forcing CPU enter deeper
951 				 * C states.
952 				 */
953 				rx_queue->zero_rx_packet_count++;
954 
955 				if (rx_queue->zero_rx_packet_count <=
956 							MIN_ZERO_POLL_COUNT)
957 					continue;
958 
959 				rx_queue->idle_hint = power_idle_heuristic(\
960 					rx_queue->zero_rx_packet_count);
961 				lcore_rx_idle_count++;
962 			} else {
963 				rx_queue->zero_rx_packet_count = 0;
964 
965 				/**
966 				 * do not scale up frequency immediately as
967 				 * user to kernel space communication is costly
968 				 * which might impact packet I/O for received
969 				 * packets.
970 				 */
971 				rx_queue->freq_up_hint =
972 					power_freq_scaleup_heuristic(lcore_id,
973 							portid, queueid);
974 			}
975 
976 			/* Prefetch first packets */
977 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
978 				rte_prefetch0(rte_pktmbuf_mtod(
979 						pkts_burst[j], void *));
980 			}
981 
982 			/* Prefetch and forward already prefetched packets */
983 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
984 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
985 						j + PREFETCH_OFFSET], void *));
986 				l3fwd_simple_forward(pkts_burst[j], portid,
987 								qconf);
988 			}
989 
990 			/* Forward remaining prefetched packets */
991 			for (; j < nb_rx; j++) {
992 				l3fwd_simple_forward(pkts_burst[j], portid,
993 								qconf);
994 			}
995 		}
996 
997 		if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
998 			for (i = 1, lcore_scaleup_hint =
999 				qconf->rx_queue_list[0].freq_up_hint;
1000 					i < qconf->n_rx_queue; ++i) {
1001 				rx_queue = &(qconf->rx_queue_list[i]);
1002 				if (rx_queue->freq_up_hint >
1003 						lcore_scaleup_hint)
1004 					lcore_scaleup_hint =
1005 						rx_queue->freq_up_hint;
1006 			}
1007 
1008 			if (lcore_scaleup_hint == FREQ_HIGHEST) {
1009 				if (rte_power_freq_max)
1010 					rte_power_freq_max(lcore_id);
1011 			} else if (lcore_scaleup_hint == FREQ_HIGHER) {
1012 				if (rte_power_freq_up)
1013 					rte_power_freq_up(lcore_id);
1014 			}
1015 		} else {
1016 			/**
1017 			 * All Rx queues empty in recent consecutive polls,
1018 			 * sleep in a conservative manner, meaning sleep as
1019 			 * less as possible.
1020 			 */
1021 			for (i = 1, lcore_idle_hint =
1022 				qconf->rx_queue_list[0].idle_hint;
1023 					i < qconf->n_rx_queue; ++i) {
1024 				rx_queue = &(qconf->rx_queue_list[i]);
1025 				if (rx_queue->idle_hint < lcore_idle_hint)
1026 					lcore_idle_hint = rx_queue->idle_hint;
1027 			}
1028 
1029 			if (lcore_idle_hint < SUSPEND_THRESHOLD)
1030 				/**
1031 				 * execute "pause" instruction to avoid context
1032 				 * switch which generally take hundred of
1033 				 * microseconds for short sleep.
1034 				 */
1035 				rte_delay_us(lcore_idle_hint);
1036 			else {
1037 				/* suspend until rx interrupt trigges */
1038 				if (intr_en) {
1039 					turn_on_intr(qconf);
1040 					sleep_until_rx_interrupt(
1041 						qconf->n_rx_queue);
1042 				}
1043 				/* start receiving packets immediately */
1044 				goto start_rx;
1045 			}
1046 			stats[lcore_id].sleep_time += lcore_idle_hint;
1047 		}
1048 	}
1049 }
1050 
1051 static int
1052 check_lcore_params(void)
1053 {
1054 	uint8_t queue, lcore;
1055 	uint16_t i;
1056 	int socketid;
1057 
1058 	for (i = 0; i < nb_lcore_params; ++i) {
1059 		queue = lcore_params[i].queue_id;
1060 		if (queue >= MAX_RX_QUEUE_PER_PORT) {
1061 			printf("invalid queue number: %hhu\n", queue);
1062 			return -1;
1063 		}
1064 		lcore = lcore_params[i].lcore_id;
1065 		if (!rte_lcore_is_enabled(lcore)) {
1066 			printf("error: lcore %hhu is not enabled in lcore "
1067 							"mask\n", lcore);
1068 			return -1;
1069 		}
1070 		if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1071 							(numa_on == 0)) {
1072 			printf("warning: lcore %hhu is on socket %d with numa "
1073 						"off\n", lcore, socketid);
1074 		}
1075 	}
1076 	return 0;
1077 }
1078 
1079 static int
1080 check_port_config(const unsigned nb_ports)
1081 {
1082 	unsigned portid;
1083 	uint16_t i;
1084 
1085 	for (i = 0; i < nb_lcore_params; ++i) {
1086 		portid = lcore_params[i].port_id;
1087 		if ((enabled_port_mask & (1 << portid)) == 0) {
1088 			printf("port %u is not enabled in port mask\n",
1089 								portid);
1090 			return -1;
1091 		}
1092 		if (portid >= nb_ports) {
1093 			printf("port %u is not present on the board\n",
1094 								portid);
1095 			return -1;
1096 		}
1097 	}
1098 	return 0;
1099 }
1100 
1101 static uint8_t
1102 get_port_n_rx_queues(const uint8_t port)
1103 {
1104 	int queue = -1;
1105 	uint16_t i;
1106 
1107 	for (i = 0; i < nb_lcore_params; ++i) {
1108 		if (lcore_params[i].port_id == port &&
1109 				lcore_params[i].queue_id > queue)
1110 			queue = lcore_params[i].queue_id;
1111 	}
1112 	return (uint8_t)(++queue);
1113 }
1114 
1115 static int
1116 init_lcore_rx_queues(void)
1117 {
1118 	uint16_t i, nb_rx_queue;
1119 	uint8_t lcore;
1120 
1121 	for (i = 0; i < nb_lcore_params; ++i) {
1122 		lcore = lcore_params[i].lcore_id;
1123 		nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1124 		if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1125 			printf("error: too many queues (%u) for lcore: %u\n",
1126 				(unsigned)nb_rx_queue + 1, (unsigned)lcore);
1127 			return -1;
1128 		} else {
1129 			lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1130 				lcore_params[i].port_id;
1131 			lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1132 				lcore_params[i].queue_id;
1133 			lcore_conf[lcore].n_rx_queue++;
1134 		}
1135 	}
1136 	return 0;
1137 }
1138 
1139 /* display usage */
1140 static void
1141 print_usage(const char *prgname)
1142 {
1143 	printf ("%s [EAL options] -- -p PORTMASK -P"
1144 		"  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1145 		"  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1146 		"  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1147 		"  -P : enable promiscuous mode\n"
1148 		"  --config (port,queue,lcore): rx queues configuration\n"
1149 		"  --no-numa: optional, disable numa awareness\n"
1150 		"  --enable-jumbo: enable jumbo frame"
1151 		" which max packet len is PKTLEN in decimal (64-9600)\n",
1152 		prgname);
1153 }
1154 
1155 static int parse_max_pkt_len(const char *pktlen)
1156 {
1157 	char *end = NULL;
1158 	unsigned long len;
1159 
1160 	/* parse decimal string */
1161 	len = strtoul(pktlen, &end, 10);
1162 	if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1163 		return -1;
1164 
1165 	if (len == 0)
1166 		return -1;
1167 
1168 	return len;
1169 }
1170 
1171 static int
1172 parse_portmask(const char *portmask)
1173 {
1174 	char *end = NULL;
1175 	unsigned long pm;
1176 
1177 	/* parse hexadecimal string */
1178 	pm = strtoul(portmask, &end, 16);
1179 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1180 		return -1;
1181 
1182 	if (pm == 0)
1183 		return -1;
1184 
1185 	return pm;
1186 }
1187 
1188 static int
1189 parse_config(const char *q_arg)
1190 {
1191 	char s[256];
1192 	const char *p, *p0 = q_arg;
1193 	char *end;
1194 	enum fieldnames {
1195 		FLD_PORT = 0,
1196 		FLD_QUEUE,
1197 		FLD_LCORE,
1198 		_NUM_FLD
1199 	};
1200 	unsigned long int_fld[_NUM_FLD];
1201 	char *str_fld[_NUM_FLD];
1202 	int i;
1203 	unsigned size;
1204 
1205 	nb_lcore_params = 0;
1206 
1207 	while ((p = strchr(p0,'(')) != NULL) {
1208 		++p;
1209 		if((p0 = strchr(p,')')) == NULL)
1210 			return -1;
1211 
1212 		size = p0 - p;
1213 		if(size >= sizeof(s))
1214 			return -1;
1215 
1216 		snprintf(s, sizeof(s), "%.*s", size, p);
1217 		if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1218 								_NUM_FLD)
1219 			return -1;
1220 		for (i = 0; i < _NUM_FLD; i++){
1221 			errno = 0;
1222 			int_fld[i] = strtoul(str_fld[i], &end, 0);
1223 			if (errno != 0 || end == str_fld[i] || int_fld[i] >
1224 									255)
1225 				return -1;
1226 		}
1227 		if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1228 			printf("exceeded max number of lcore params: %hu\n",
1229 				nb_lcore_params);
1230 			return -1;
1231 		}
1232 		lcore_params_array[nb_lcore_params].port_id =
1233 				(uint8_t)int_fld[FLD_PORT];
1234 		lcore_params_array[nb_lcore_params].queue_id =
1235 				(uint8_t)int_fld[FLD_QUEUE];
1236 		lcore_params_array[nb_lcore_params].lcore_id =
1237 				(uint8_t)int_fld[FLD_LCORE];
1238 		++nb_lcore_params;
1239 	}
1240 	lcore_params = lcore_params_array;
1241 
1242 	return 0;
1243 }
1244 
1245 /* Parse the argument given in the command line of the application */
1246 static int
1247 parse_args(int argc, char **argv)
1248 {
1249 	int opt, ret;
1250 	char **argvopt;
1251 	int option_index;
1252 	char *prgname = argv[0];
1253 	static struct option lgopts[] = {
1254 		{"config", 1, 0, 0},
1255 		{"no-numa", 0, 0, 0},
1256 		{"enable-jumbo", 0, 0, 0},
1257 		{NULL, 0, 0, 0}
1258 	};
1259 
1260 	argvopt = argv;
1261 
1262 	while ((opt = getopt_long(argc, argvopt, "p:P",
1263 				lgopts, &option_index)) != EOF) {
1264 
1265 		switch (opt) {
1266 		/* portmask */
1267 		case 'p':
1268 			enabled_port_mask = parse_portmask(optarg);
1269 			if (enabled_port_mask == 0) {
1270 				printf("invalid portmask\n");
1271 				print_usage(prgname);
1272 				return -1;
1273 			}
1274 			break;
1275 		case 'P':
1276 			printf("Promiscuous mode selected\n");
1277 			promiscuous_on = 1;
1278 			break;
1279 
1280 		/* long options */
1281 		case 0:
1282 			if (!strncmp(lgopts[option_index].name, "config", 6)) {
1283 				ret = parse_config(optarg);
1284 				if (ret) {
1285 					printf("invalid config\n");
1286 					print_usage(prgname);
1287 					return -1;
1288 				}
1289 			}
1290 
1291 			if (!strncmp(lgopts[option_index].name,
1292 						"no-numa", 7)) {
1293 				printf("numa is disabled \n");
1294 				numa_on = 0;
1295 			}
1296 
1297 			if (!strncmp(lgopts[option_index].name,
1298 					"enable-jumbo", 12)) {
1299 				struct option lenopts =
1300 					{"max-pkt-len", required_argument, \
1301 									0, 0};
1302 
1303 				printf("jumbo frame is enabled \n");
1304 				port_conf.rxmode.jumbo_frame = 1;
1305 
1306 				/**
1307 				 * if no max-pkt-len set, use the default value
1308 				 * ETHER_MAX_LEN
1309 				 */
1310 				if (0 == getopt_long(argc, argvopt, "",
1311 						&lenopts, &option_index)) {
1312 					ret = parse_max_pkt_len(optarg);
1313 					if ((ret < 64) ||
1314 						(ret > MAX_JUMBO_PKT_LEN)){
1315 						printf("invalid packet "
1316 								"length\n");
1317 						print_usage(prgname);
1318 						return -1;
1319 					}
1320 					port_conf.rxmode.max_rx_pkt_len = ret;
1321 				}
1322 				printf("set jumbo frame "
1323 					"max packet length to %u\n",
1324 				(unsigned int)port_conf.rxmode.max_rx_pkt_len);
1325 			}
1326 
1327 			break;
1328 
1329 		default:
1330 			print_usage(prgname);
1331 			return -1;
1332 		}
1333 	}
1334 
1335 	if (optind >= 0)
1336 		argv[optind-1] = prgname;
1337 
1338 	ret = optind-1;
1339 	optind = 0; /* reset getopt lib */
1340 	return ret;
1341 }
1342 
1343 static void
1344 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1345 {
1346 	char buf[ETHER_ADDR_FMT_SIZE];
1347 	ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1348 	printf("%s%s", name, buf);
1349 }
1350 
1351 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1352 static void
1353 setup_hash(int socketid)
1354 {
1355 	struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1356 		.name = NULL,
1357 		.entries = L3FWD_HASH_ENTRIES,
1358 		.key_len = sizeof(struct ipv4_5tuple),
1359 		.hash_func = DEFAULT_HASH_FUNC,
1360 		.hash_func_init_val = 0,
1361 	};
1362 
1363 	struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1364 		.name = NULL,
1365 		.entries = L3FWD_HASH_ENTRIES,
1366 		.key_len = sizeof(struct ipv6_5tuple),
1367 		.hash_func = DEFAULT_HASH_FUNC,
1368 		.hash_func_init_val = 0,
1369 	};
1370 
1371 	unsigned i;
1372 	int ret;
1373 	char s[64];
1374 
1375 	/* create ipv4 hash */
1376 	rte_snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1377 	ipv4_l3fwd_hash_params.name = s;
1378 	ipv4_l3fwd_hash_params.socket_id = socketid;
1379 	ipv4_l3fwd_lookup_struct[socketid] =
1380 		rte_hash_create(&ipv4_l3fwd_hash_params);
1381 	if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1382 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1383 				"socket %d\n", socketid);
1384 
1385 	/* create ipv6 hash */
1386 	rte_snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1387 	ipv6_l3fwd_hash_params.name = s;
1388 	ipv6_l3fwd_hash_params.socket_id = socketid;
1389 	ipv6_l3fwd_lookup_struct[socketid] =
1390 		rte_hash_create(&ipv6_l3fwd_hash_params);
1391 	if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1392 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1393 				"socket %d\n", socketid);
1394 
1395 
1396 	/* populate the ipv4 hash */
1397 	for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1398 		ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1399 				(void *) &ipv4_l3fwd_route_array[i].key);
1400 		if (ret < 0) {
1401 			rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1402 				"l3fwd hash on socket %d\n", i, socketid);
1403 		}
1404 		ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1405 		printf("Hash: Adding key\n");
1406 		print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1407 	}
1408 
1409 	/* populate the ipv6 hash */
1410 	for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1411 		ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1412 				(void *) &ipv6_l3fwd_route_array[i].key);
1413 		if (ret < 0) {
1414 			rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1415 				"l3fwd hash on socket %d\n", i, socketid);
1416 		}
1417 		ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1418 		printf("Hash: Adding key\n");
1419 		print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1420 	}
1421 }
1422 #endif
1423 
1424 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1425 static void
1426 setup_lpm(int socketid)
1427 {
1428 	unsigned i;
1429 	int ret;
1430 	char s[64];
1431 
1432 	/* create the LPM table */
1433 	snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1434 	ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1435 				IPV4_L3FWD_LPM_MAX_RULES, 0);
1436 	if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1437 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1438 				" on socket %d\n", socketid);
1439 
1440 	/* populate the LPM table */
1441 	for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1442 		ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1443 			ipv4_l3fwd_route_array[i].ip,
1444 			ipv4_l3fwd_route_array[i].depth,
1445 			ipv4_l3fwd_route_array[i].if_out);
1446 
1447 		if (ret < 0) {
1448 			rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1449 				"l3fwd LPM table on socket %d\n",
1450 				i, socketid);
1451 		}
1452 
1453 		printf("LPM: Adding route 0x%08x / %d (%d)\n",
1454 			(unsigned)ipv4_l3fwd_route_array[i].ip,
1455 			ipv4_l3fwd_route_array[i].depth,
1456 			ipv4_l3fwd_route_array[i].if_out);
1457 	}
1458 }
1459 #endif
1460 
1461 static int
1462 init_mem(unsigned nb_mbuf)
1463 {
1464 	struct lcore_conf *qconf;
1465 	int socketid;
1466 	unsigned lcore_id;
1467 	char s[64];
1468 
1469 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1470 		if (rte_lcore_is_enabled(lcore_id) == 0)
1471 			continue;
1472 
1473 		if (numa_on)
1474 			socketid = rte_lcore_to_socket_id(lcore_id);
1475 		else
1476 			socketid = 0;
1477 
1478 		if (socketid >= NB_SOCKETS) {
1479 			rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1480 					"out of range %d\n", socketid,
1481 						lcore_id, NB_SOCKETS);
1482 		}
1483 		if (pktmbuf_pool[socketid] == NULL) {
1484 			snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1485 			pktmbuf_pool[socketid] =
1486 				rte_pktmbuf_pool_create(s, nb_mbuf,
1487 					MEMPOOL_CACHE_SIZE, 0,
1488 					RTE_MBUF_DEFAULT_BUF_SIZE,
1489 					socketid);
1490 			if (pktmbuf_pool[socketid] == NULL)
1491 				rte_exit(EXIT_FAILURE,
1492 					"Cannot init mbuf pool on socket %d\n",
1493 								socketid);
1494 			else
1495 				printf("Allocated mbuf pool on socket %d\n",
1496 								socketid);
1497 
1498 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1499 			setup_lpm(socketid);
1500 #else
1501 			setup_hash(socketid);
1502 #endif
1503 		}
1504 		qconf = &lcore_conf[lcore_id];
1505 		qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1506 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1507 		qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1508 #endif
1509 	}
1510 	return 0;
1511 }
1512 
1513 /* Check the link status of all ports in up to 9s, and print them finally */
1514 static void
1515 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1516 {
1517 #define CHECK_INTERVAL 100 /* 100ms */
1518 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1519 	uint8_t portid, count, all_ports_up, print_flag = 0;
1520 	struct rte_eth_link link;
1521 
1522 	printf("\nChecking link status");
1523 	fflush(stdout);
1524 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
1525 		all_ports_up = 1;
1526 		for (portid = 0; portid < port_num; portid++) {
1527 			if ((port_mask & (1 << portid)) == 0)
1528 				continue;
1529 			memset(&link, 0, sizeof(link));
1530 			rte_eth_link_get_nowait(portid, &link);
1531 			/* print link status if flag set */
1532 			if (print_flag == 1) {
1533 				if (link.link_status)
1534 					printf("Port %d Link Up - speed %u "
1535 						"Mbps - %s\n", (uint8_t)portid,
1536 						(unsigned)link.link_speed,
1537 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1538 					("full-duplex") : ("half-duplex\n"));
1539 				else
1540 					printf("Port %d Link Down\n",
1541 						(uint8_t)portid);
1542 				continue;
1543 			}
1544 			/* clear all_ports_up flag if any link down */
1545 			if (link.link_status == 0) {
1546 				all_ports_up = 0;
1547 				break;
1548 			}
1549 		}
1550 		/* after finally printing all link status, get out */
1551 		if (print_flag == 1)
1552 			break;
1553 
1554 		if (all_ports_up == 0) {
1555 			printf(".");
1556 			fflush(stdout);
1557 			rte_delay_ms(CHECK_INTERVAL);
1558 		}
1559 
1560 		/* set the print_flag if all ports up or timeout */
1561 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1562 			print_flag = 1;
1563 			printf("done\n");
1564 		}
1565 	}
1566 }
1567 
1568 int
1569 main(int argc, char **argv)
1570 {
1571 	struct lcore_conf *qconf;
1572 	struct rte_eth_dev_info dev_info;
1573 	struct rte_eth_txconf *txconf;
1574 	int ret;
1575 	unsigned nb_ports;
1576 	uint16_t queueid;
1577 	unsigned lcore_id;
1578 	uint64_t hz;
1579 	uint32_t n_tx_queue, nb_lcores;
1580 	uint32_t dev_rxq_num, dev_txq_num;
1581 	uint8_t portid, nb_rx_queue, queue, socketid;
1582 
1583 	/* catch SIGINT and restore cpufreq governor to ondemand */
1584 	signal(SIGINT, signal_exit_now);
1585 
1586 	/* init EAL */
1587 	ret = rte_eal_init(argc, argv);
1588 	if (ret < 0)
1589 		rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1590 	argc -= ret;
1591 	argv += ret;
1592 
1593 	/* init RTE timer library to be used late */
1594 	rte_timer_subsystem_init();
1595 
1596 	/* parse application arguments (after the EAL ones) */
1597 	ret = parse_args(argc, argv);
1598 	if (ret < 0)
1599 		rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1600 
1601 	if (check_lcore_params() < 0)
1602 		rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1603 
1604 	ret = init_lcore_rx_queues();
1605 	if (ret < 0)
1606 		rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1607 
1608 
1609 	nb_ports = rte_eth_dev_count();
1610 	if (nb_ports > RTE_MAX_ETHPORTS)
1611 		nb_ports = RTE_MAX_ETHPORTS;
1612 
1613 	if (check_port_config(nb_ports) < 0)
1614 		rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1615 
1616 	nb_lcores = rte_lcore_count();
1617 
1618 	/* initialize all ports */
1619 	for (portid = 0; portid < nb_ports; portid++) {
1620 		/* skip ports that are not enabled */
1621 		if ((enabled_port_mask & (1 << portid)) == 0) {
1622 			printf("\nSkipping disabled port %d\n", portid);
1623 			continue;
1624 		}
1625 
1626 		/* init port */
1627 		printf("Initializing port %d ... ", portid );
1628 		fflush(stdout);
1629 
1630 		rte_eth_dev_info_get(portid, &dev_info);
1631 		dev_rxq_num = dev_info.max_rx_queues;
1632 		dev_txq_num = dev_info.max_tx_queues;
1633 
1634 		nb_rx_queue = get_port_n_rx_queues(portid);
1635 		if (nb_rx_queue > dev_rxq_num)
1636 			rte_exit(EXIT_FAILURE,
1637 				"Cannot configure not existed rxq: "
1638 				"port=%d\n", portid);
1639 
1640 		n_tx_queue = nb_lcores;
1641 		if (n_tx_queue > dev_txq_num)
1642 			n_tx_queue = dev_txq_num;
1643 		printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1644 			nb_rx_queue, (unsigned)n_tx_queue );
1645 		ret = rte_eth_dev_configure(portid, nb_rx_queue,
1646 					(uint16_t)n_tx_queue, &port_conf);
1647 		if (ret < 0)
1648 			rte_exit(EXIT_FAILURE, "Cannot configure device: "
1649 					"err=%d, port=%d\n", ret, portid);
1650 
1651 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1652 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
1653 		printf(", ");
1654 
1655 		/* init memory */
1656 		ret = init_mem(NB_MBUF);
1657 		if (ret < 0)
1658 			rte_exit(EXIT_FAILURE, "init_mem failed\n");
1659 
1660 		/* init one TX queue per couple (lcore,port) */
1661 		queueid = 0;
1662 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1663 			if (rte_lcore_is_enabled(lcore_id) == 0)
1664 				continue;
1665 
1666 			if (queueid >= dev_txq_num)
1667 				continue;
1668 
1669 			if (numa_on)
1670 				socketid = \
1671 				(uint8_t)rte_lcore_to_socket_id(lcore_id);
1672 			else
1673 				socketid = 0;
1674 
1675 			printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1676 			fflush(stdout);
1677 
1678 			rte_eth_dev_info_get(portid, &dev_info);
1679 			txconf = &dev_info.default_txconf;
1680 			if (port_conf.rxmode.jumbo_frame)
1681 				txconf->txq_flags = 0;
1682 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1683 						     socketid, txconf);
1684 			if (ret < 0)
1685 				rte_exit(EXIT_FAILURE,
1686 					"rte_eth_tx_queue_setup: err=%d, "
1687 						"port=%d\n", ret, portid);
1688 
1689 			qconf = &lcore_conf[lcore_id];
1690 			qconf->tx_queue_id[portid] = queueid;
1691 			queueid++;
1692 		}
1693 		printf("\n");
1694 	}
1695 
1696 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1697 		if (rte_lcore_is_enabled(lcore_id) == 0)
1698 			continue;
1699 
1700 		/* init power management library */
1701 		ret = rte_power_init(lcore_id);
1702 		if (ret)
1703 			RTE_LOG(ERR, POWER,
1704 				"Library initialization failed on core %u\n", lcore_id);
1705 
1706 		/* init timer structures for each enabled lcore */
1707 		rte_timer_init(&power_timers[lcore_id]);
1708 		hz = rte_get_timer_hz();
1709 		rte_timer_reset(&power_timers[lcore_id],
1710 			hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1711 						power_timer_cb, NULL);
1712 
1713 		qconf = &lcore_conf[lcore_id];
1714 		printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1715 		fflush(stdout);
1716 		/* init RX queues */
1717 		for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1718 			portid = qconf->rx_queue_list[queue].port_id;
1719 			queueid = qconf->rx_queue_list[queue].queue_id;
1720 
1721 			if (numa_on)
1722 				socketid = \
1723 				(uint8_t)rte_lcore_to_socket_id(lcore_id);
1724 			else
1725 				socketid = 0;
1726 
1727 			printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1728 			fflush(stdout);
1729 
1730 			ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1731 				socketid, NULL,
1732 				pktmbuf_pool[socketid]);
1733 			if (ret < 0)
1734 				rte_exit(EXIT_FAILURE,
1735 					"rte_eth_rx_queue_setup: err=%d, "
1736 						"port=%d\n", ret, portid);
1737 		}
1738 	}
1739 
1740 	printf("\n");
1741 
1742 	/* start ports */
1743 	for (portid = 0; portid < nb_ports; portid++) {
1744 		if ((enabled_port_mask & (1 << portid)) == 0) {
1745 			continue;
1746 		}
1747 		/* Start device */
1748 		ret = rte_eth_dev_start(portid);
1749 		if (ret < 0)
1750 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1751 						"port=%d\n", ret, portid);
1752 		/*
1753 		 * If enabled, put device in promiscuous mode.
1754 		 * This allows IO forwarding mode to forward packets
1755 		 * to itself through 2 cross-connected  ports of the
1756 		 * target machine.
1757 		 */
1758 		if (promiscuous_on)
1759 			rte_eth_promiscuous_enable(portid);
1760 		/* initialize spinlock for each port */
1761 		rte_spinlock_init(&(locks[portid]));
1762 	}
1763 
1764 	check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1765 
1766 	/* launch per-lcore init on every lcore */
1767 	rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1768 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1769 		if (rte_eal_wait_lcore(lcore_id) < 0)
1770 			return -1;
1771 	}
1772 
1773 	return 0;
1774 }
1775