xref: /dpdk/examples/l3fwd-power/main.c (revision ceb1ccd5d50c1a89ba8bdd97cc199e7f07422b98)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <stdio.h>
35 #include <stdlib.h>
36 #include <stdint.h>
37 #include <inttypes.h>
38 #include <sys/types.h>
39 #include <string.h>
40 #include <sys/queue.h>
41 #include <stdarg.h>
42 #include <errno.h>
43 #include <getopt.h>
44 #include <unistd.h>
45 #include <signal.h>
46 
47 #include <rte_common.h>
48 #include <rte_byteorder.h>
49 #include <rte_log.h>
50 #include <rte_malloc.h>
51 #include <rte_memory.h>
52 #include <rte_memcpy.h>
53 #include <rte_memzone.h>
54 #include <rte_eal.h>
55 #include <rte_per_lcore.h>
56 #include <rte_launch.h>
57 #include <rte_atomic.h>
58 #include <rte_cycles.h>
59 #include <rte_prefetch.h>
60 #include <rte_lcore.h>
61 #include <rte_per_lcore.h>
62 #include <rte_branch_prediction.h>
63 #include <rte_interrupts.h>
64 #include <rte_pci.h>
65 #include <rte_random.h>
66 #include <rte_debug.h>
67 #include <rte_ether.h>
68 #include <rte_ethdev.h>
69 #include <rte_ring.h>
70 #include <rte_mempool.h>
71 #include <rte_mbuf.h>
72 #include <rte_ip.h>
73 #include <rte_tcp.h>
74 #include <rte_udp.h>
75 #include <rte_string_fns.h>
76 #include <rte_timer.h>
77 #include <rte_power.h>
78 #include <rte_eal.h>
79 #include <rte_spinlock.h>
80 
81 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
82 
83 #define MAX_PKT_BURST 32
84 
85 #define MIN_ZERO_POLL_COUNT 10
86 
87 /* around 100ms at 2 Ghz */
88 #define TIMER_RESOLUTION_CYCLES           200000000ULL
89 /* 100 ms interval */
90 #define TIMER_NUMBER_PER_SECOND           10
91 /* 100000 us */
92 #define SCALING_PERIOD                    (1000000/TIMER_NUMBER_PER_SECOND)
93 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
94 
95 #define APP_LOOKUP_EXACT_MATCH          0
96 #define APP_LOOKUP_LPM                  1
97 #define DO_RFC_1812_CHECKS
98 
99 #ifndef APP_LOOKUP_METHOD
100 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
101 #endif
102 
103 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
104 #include <rte_hash.h>
105 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
106 #include <rte_lpm.h>
107 #else
108 #error "APP_LOOKUP_METHOD set to incorrect value"
109 #endif
110 
111 #ifndef IPv6_BYTES
112 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
113                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
114 #define IPv6_BYTES(addr) \
115 	addr[0],  addr[1], addr[2],  addr[3], \
116 	addr[4],  addr[5], addr[6],  addr[7], \
117 	addr[8],  addr[9], addr[10], addr[11],\
118 	addr[12], addr[13],addr[14], addr[15]
119 #endif
120 
121 #define MAX_JUMBO_PKT_LEN  9600
122 
123 #define IPV6_ADDR_LEN 16
124 
125 #define MEMPOOL_CACHE_SIZE 256
126 
127 /*
128  * This expression is used to calculate the number of mbufs needed depending on
129  * user input, taking into account memory for rx and tx hardware rings, cache
130  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
131  * NB_MBUF never goes below a minimum value of 8192.
132  */
133 
134 #define NB_MBUF RTE_MAX	( \
135 	(nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT + \
136 	nb_ports*nb_lcores*MAX_PKT_BURST + \
137 	nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT + \
138 	nb_lcores*MEMPOOL_CACHE_SIZE), \
139 	(unsigned)8192)
140 
141 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
142 
143 #define NB_SOCKETS 8
144 
145 /* Configure how many packets ahead to prefetch, when reading packets */
146 #define PREFETCH_OFFSET	3
147 
148 /*
149  * Configurable number of RX/TX ring descriptors
150  */
151 #define RTE_TEST_RX_DESC_DEFAULT 128
152 #define RTE_TEST_TX_DESC_DEFAULT 512
153 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
154 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
155 
156 /* ethernet addresses of ports */
157 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
158 
159 /* ethernet addresses of ports */
160 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
161 
162 /* mask of enabled ports */
163 static uint32_t enabled_port_mask = 0;
164 /* Ports set in promiscuous mode off by default. */
165 static int promiscuous_on = 0;
166 /* NUMA is enabled by default. */
167 static int numa_on = 1;
168 
169 enum freq_scale_hint_t
170 {
171 	FREQ_LOWER    =      -1,
172 	FREQ_CURRENT  =       0,
173 	FREQ_HIGHER   =       1,
174 	FREQ_HIGHEST  =       2
175 };
176 
177 struct lcore_rx_queue {
178 	uint8_t port_id;
179 	uint8_t queue_id;
180 	enum freq_scale_hint_t freq_up_hint;
181 	uint32_t zero_rx_packet_count;
182 	uint32_t idle_hint;
183 } __rte_cache_aligned;
184 
185 #define MAX_RX_QUEUE_PER_LCORE 16
186 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
187 #define MAX_RX_QUEUE_PER_PORT 128
188 
189 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
190 
191 
192 #define MAX_LCORE_PARAMS 1024
193 struct lcore_params {
194 	uint8_t port_id;
195 	uint8_t queue_id;
196 	uint8_t lcore_id;
197 } __rte_cache_aligned;
198 
199 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
200 static struct lcore_params lcore_params_array_default[] = {
201 	{0, 0, 2},
202 	{0, 1, 2},
203 	{0, 2, 2},
204 	{1, 0, 2},
205 	{1, 1, 2},
206 	{1, 2, 2},
207 	{2, 0, 2},
208 	{3, 0, 3},
209 	{3, 1, 3},
210 };
211 
212 static struct lcore_params * lcore_params = lcore_params_array_default;
213 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
214 				sizeof(lcore_params_array_default[0]);
215 
216 static struct rte_eth_conf port_conf = {
217 	.rxmode = {
218 		.mq_mode        = ETH_MQ_RX_RSS,
219 		.max_rx_pkt_len = ETHER_MAX_LEN,
220 		.split_hdr_size = 0,
221 		.header_split   = 0, /**< Header Split disabled */
222 		.hw_ip_checksum = 1, /**< IP checksum offload enabled */
223 		.hw_vlan_filter = 0, /**< VLAN filtering disabled */
224 		.jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
225 		.hw_strip_crc   = 0, /**< CRC stripped by hardware */
226 	},
227 	.rx_adv_conf = {
228 		.rss_conf = {
229 			.rss_key = NULL,
230 			.rss_hf = ETH_RSS_UDP,
231 		},
232 	},
233 	.txmode = {
234 		.mq_mode = ETH_MQ_TX_NONE,
235 	},
236 	.intr_conf = {
237 		.lsc = 1,
238 		.rxq = 1,
239 	},
240 };
241 
242 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
243 
244 
245 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
246 
247 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
248 #include <rte_hash_crc.h>
249 #define DEFAULT_HASH_FUNC       rte_hash_crc
250 #else
251 #include <rte_jhash.h>
252 #define DEFAULT_HASH_FUNC       rte_jhash
253 #endif
254 
255 struct ipv4_5tuple {
256 	uint32_t ip_dst;
257 	uint32_t ip_src;
258 	uint16_t port_dst;
259 	uint16_t port_src;
260 	uint8_t  proto;
261 } __attribute__((__packed__));
262 
263 struct ipv6_5tuple {
264 	uint8_t  ip_dst[IPV6_ADDR_LEN];
265 	uint8_t  ip_src[IPV6_ADDR_LEN];
266 	uint16_t port_dst;
267 	uint16_t port_src;
268 	uint8_t  proto;
269 } __attribute__((__packed__));
270 
271 struct ipv4_l3fwd_route {
272 	struct ipv4_5tuple key;
273 	uint8_t if_out;
274 };
275 
276 struct ipv6_l3fwd_route {
277 	struct ipv6_5tuple key;
278 	uint8_t if_out;
279 };
280 
281 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
282 	{{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
283 	{{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
284 	{{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
285 	{{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
286 };
287 
288 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
289 	{
290 		{
291 			{0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
292 			 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
293 			{0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
294 			 0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
295 			 1, 10, IPPROTO_UDP
296 		}, 4
297 	},
298 };
299 
300 typedef struct rte_hash lookup_struct_t;
301 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
302 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
303 
304 #define L3FWD_HASH_ENTRIES	1024
305 
306 #define IPV4_L3FWD_NUM_ROUTES \
307 	(sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
308 
309 #define IPV6_L3FWD_NUM_ROUTES \
310 	(sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
311 
312 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
313 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
314 #endif
315 
316 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
317 struct ipv4_l3fwd_route {
318 	uint32_t ip;
319 	uint8_t  depth;
320 	uint8_t  if_out;
321 };
322 
323 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
324 	{IPv4(1,1,1,0), 24, 0},
325 	{IPv4(2,1,1,0), 24, 1},
326 	{IPv4(3,1,1,0), 24, 2},
327 	{IPv4(4,1,1,0), 24, 3},
328 	{IPv4(5,1,1,0), 24, 4},
329 	{IPv4(6,1,1,0), 24, 5},
330 	{IPv4(7,1,1,0), 24, 6},
331 	{IPv4(8,1,1,0), 24, 7},
332 };
333 
334 #define IPV4_L3FWD_NUM_ROUTES \
335 	(sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
336 
337 #define IPV4_L3FWD_LPM_MAX_RULES     1024
338 
339 typedef struct rte_lpm lookup_struct_t;
340 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
341 #endif
342 
343 struct lcore_conf {
344 	uint16_t n_rx_queue;
345 	struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
346 	uint16_t n_tx_port;
347 	uint16_t tx_port_id[RTE_MAX_ETHPORTS];
348 	uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
349 	struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
350 	lookup_struct_t * ipv4_lookup_struct;
351 	lookup_struct_t * ipv6_lookup_struct;
352 } __rte_cache_aligned;
353 
354 struct lcore_stats {
355 	/* total sleep time in ms since last frequency scaling down */
356 	uint32_t sleep_time;
357 	/* number of long sleep recently */
358 	uint32_t nb_long_sleep;
359 	/* freq. scaling up trend */
360 	uint32_t trend;
361 	/* total packet processed recently */
362 	uint64_t nb_rx_processed;
363 	/* total iterations looped recently */
364 	uint64_t nb_iteration_looped;
365 	uint32_t padding[9];
366 } __rte_cache_aligned;
367 
368 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
369 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
370 static struct rte_timer power_timers[RTE_MAX_LCORE];
371 
372 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
373 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
374 			unsigned lcore_id, uint8_t port_id, uint16_t queue_id);
375 
376 /* exit signal handler */
377 static void
378 signal_exit_now(int sigtype)
379 {
380 	unsigned lcore_id;
381 	int ret;
382 
383 	if (sigtype == SIGINT) {
384 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
385 			if (rte_lcore_is_enabled(lcore_id) == 0)
386 				continue;
387 
388 			/* init power management library */
389 			ret = rte_power_exit(lcore_id);
390 			if (ret)
391 				rte_exit(EXIT_FAILURE, "Power management "
392 					"library de-initialization failed on "
393 							"core%u\n", lcore_id);
394 		}
395 	}
396 
397 	rte_exit(EXIT_SUCCESS, "User forced exit\n");
398 }
399 
400 /*  Freqency scale down timer callback */
401 static void
402 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
403 			  __attribute__((unused)) void *arg)
404 {
405 	uint64_t hz;
406 	float sleep_time_ratio;
407 	unsigned lcore_id = rte_lcore_id();
408 
409 	/* accumulate total execution time in us when callback is invoked */
410 	sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
411 					(float)SCALING_PERIOD;
412 	/**
413 	 * check whether need to scale down frequency a step if it sleep a lot.
414 	 */
415 	if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
416 		if (rte_power_freq_down)
417 			rte_power_freq_down(lcore_id);
418 	}
419 	else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
420 		stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
421 		/**
422 		 * scale down a step if average packet per iteration less
423 		 * than expectation.
424 		 */
425 		if (rte_power_freq_down)
426 			rte_power_freq_down(lcore_id);
427 	}
428 
429 	/**
430 	 * initialize another timer according to current frequency to ensure
431 	 * timer interval is relatively fixed.
432 	 */
433 	hz = rte_get_timer_hz();
434 	rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
435 				SINGLE, lcore_id, power_timer_cb, NULL);
436 
437 	stats[lcore_id].nb_rx_processed = 0;
438 	stats[lcore_id].nb_iteration_looped = 0;
439 
440 	stats[lcore_id].sleep_time = 0;
441 }
442 
443 /* Enqueue a single packet, and send burst if queue is filled */
444 static inline int
445 send_single_packet(struct rte_mbuf *m, uint8_t port)
446 {
447 	uint32_t lcore_id;
448 	struct lcore_conf *qconf;
449 
450 	lcore_id = rte_lcore_id();
451 	qconf = &lcore_conf[lcore_id];
452 
453 	rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
454 			qconf->tx_buffer[port], m);
455 
456 	return 0;
457 }
458 
459 #ifdef DO_RFC_1812_CHECKS
460 static inline int
461 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
462 {
463 	/* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
464 	/*
465 	 * 1. The packet length reported by the Link Layer must be large
466 	 * enough to hold the minimum length legal IP datagram (20 bytes).
467 	 */
468 	if (link_len < sizeof(struct ipv4_hdr))
469 		return -1;
470 
471 	/* 2. The IP checksum must be correct. */
472 	/* this is checked in H/W */
473 
474 	/*
475 	 * 3. The IP version number must be 4. If the version number is not 4
476 	 * then the packet may be another version of IP, such as IPng or
477 	 * ST-II.
478 	 */
479 	if (((pkt->version_ihl) >> 4) != 4)
480 		return -3;
481 	/*
482 	 * 4. The IP header length field must be large enough to hold the
483 	 * minimum length legal IP datagram (20 bytes = 5 words).
484 	 */
485 	if ((pkt->version_ihl & 0xf) < 5)
486 		return -4;
487 
488 	/*
489 	 * 5. The IP total length field must be large enough to hold the IP
490 	 * datagram header, whose length is specified in the IP header length
491 	 * field.
492 	 */
493 	if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
494 		return -5;
495 
496 	return 0;
497 }
498 #endif
499 
500 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
501 static void
502 print_ipv4_key(struct ipv4_5tuple key)
503 {
504 	printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
505 		"proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
506 				key.port_dst, key.port_src, key.proto);
507 }
508 static void
509 print_ipv6_key(struct ipv6_5tuple key)
510 {
511 	printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
512 	        "port dst = %d, port src = %d, proto = %d\n",
513 	        IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
514 	        key.port_dst, key.port_src, key.proto);
515 }
516 
517 static inline uint8_t
518 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
519 		lookup_struct_t * ipv4_l3fwd_lookup_struct)
520 {
521 	struct ipv4_5tuple key;
522 	struct tcp_hdr *tcp;
523 	struct udp_hdr *udp;
524 	int ret = 0;
525 
526 	key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
527 	key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
528 	key.proto = ipv4_hdr->next_proto_id;
529 
530 	switch (ipv4_hdr->next_proto_id) {
531 	case IPPROTO_TCP:
532 		tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
533 					sizeof(struct ipv4_hdr));
534 		key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
535 		key.port_src = rte_be_to_cpu_16(tcp->src_port);
536 		break;
537 
538 	case IPPROTO_UDP:
539 		udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
540 					sizeof(struct ipv4_hdr));
541 		key.port_dst = rte_be_to_cpu_16(udp->dst_port);
542 		key.port_src = rte_be_to_cpu_16(udp->src_port);
543 		break;
544 
545 	default:
546 		key.port_dst = 0;
547 		key.port_src = 0;
548 		break;
549 	}
550 
551 	/* Find destination port */
552 	ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
553 	return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
554 }
555 
556 static inline uint8_t
557 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr,  uint8_t portid,
558 			lookup_struct_t *ipv6_l3fwd_lookup_struct)
559 {
560 	struct ipv6_5tuple key;
561 	struct tcp_hdr *tcp;
562 	struct udp_hdr *udp;
563 	int ret = 0;
564 
565 	memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
566 	memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
567 
568 	key.proto = ipv6_hdr->proto;
569 
570 	switch (ipv6_hdr->proto) {
571 	case IPPROTO_TCP:
572 		tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
573 					sizeof(struct ipv6_hdr));
574 		key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
575 		key.port_src = rte_be_to_cpu_16(tcp->src_port);
576 		break;
577 
578 	case IPPROTO_UDP:
579 		udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
580 					sizeof(struct ipv6_hdr));
581 		key.port_dst = rte_be_to_cpu_16(udp->dst_port);
582 		key.port_src = rte_be_to_cpu_16(udp->src_port);
583 		break;
584 
585 	default:
586 		key.port_dst = 0;
587 		key.port_src = 0;
588 		break;
589 	}
590 
591 	/* Find destination port */
592 	ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
593 	return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
594 }
595 #endif
596 
597 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
598 static inline uint8_t
599 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
600 		lookup_struct_t *ipv4_l3fwd_lookup_struct)
601 {
602 	uint32_t next_hop;
603 
604 	return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
605 			rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
606 			next_hop : portid);
607 }
608 #endif
609 
610 static inline void
611 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
612 				struct lcore_conf *qconf)
613 {
614 	struct ether_hdr *eth_hdr;
615 	struct ipv4_hdr *ipv4_hdr;
616 	void *d_addr_bytes;
617 	uint8_t dst_port;
618 
619 	eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
620 
621 	if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
622 		/* Handle IPv4 headers.*/
623 		ipv4_hdr =
624 			rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
625 						sizeof(struct ether_hdr));
626 
627 #ifdef DO_RFC_1812_CHECKS
628 		/* Check to make sure the packet is valid (RFC1812) */
629 		if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
630 			rte_pktmbuf_free(m);
631 			return;
632 		}
633 #endif
634 
635 		dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
636 					qconf->ipv4_lookup_struct);
637 		if (dst_port >= RTE_MAX_ETHPORTS ||
638 				(enabled_port_mask & 1 << dst_port) == 0)
639 			dst_port = portid;
640 
641 		/* 02:00:00:00:00:xx */
642 		d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
643 		*((uint64_t *)d_addr_bytes) =
644 			0x000000000002 + ((uint64_t)dst_port << 40);
645 
646 #ifdef DO_RFC_1812_CHECKS
647 		/* Update time to live and header checksum */
648 		--(ipv4_hdr->time_to_live);
649 		++(ipv4_hdr->hdr_checksum);
650 #endif
651 
652 		/* src addr */
653 		ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
654 
655 		send_single_packet(m, dst_port);
656 	} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
657 		/* Handle IPv6 headers.*/
658 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
659 		struct ipv6_hdr *ipv6_hdr;
660 
661 		ipv6_hdr =
662 			rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
663 						sizeof(struct ether_hdr));
664 
665 		dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
666 					qconf->ipv6_lookup_struct);
667 
668 		if (dst_port >= RTE_MAX_ETHPORTS ||
669 				(enabled_port_mask & 1 << dst_port) == 0)
670 			dst_port = portid;
671 
672 		/* 02:00:00:00:00:xx */
673 		d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
674 		*((uint64_t *)d_addr_bytes) =
675 			0x000000000002 + ((uint64_t)dst_port << 40);
676 
677 		/* src addr */
678 		ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
679 
680 		send_single_packet(m, dst_port);
681 #else
682 		/* We don't currently handle IPv6 packets in LPM mode. */
683 		rte_pktmbuf_free(m);
684 #endif
685 	} else
686 		rte_pktmbuf_free(m);
687 
688 }
689 
690 #define MINIMUM_SLEEP_TIME         1
691 #define SUSPEND_THRESHOLD          300
692 
693 static inline uint32_t
694 power_idle_heuristic(uint32_t zero_rx_packet_count)
695 {
696 	/* If zero count is less than 100,  sleep 1us */
697 	if (zero_rx_packet_count < SUSPEND_THRESHOLD)
698 		return MINIMUM_SLEEP_TIME;
699 	/* If zero count is less than 1000, sleep 100 us which is the
700 		minimum latency switching from C3/C6 to C0
701 	*/
702 	else
703 		return SUSPEND_THRESHOLD;
704 
705 	return 0;
706 }
707 
708 static inline enum freq_scale_hint_t
709 power_freq_scaleup_heuristic(unsigned lcore_id,
710 			     uint8_t port_id,
711 			     uint16_t queue_id)
712 {
713 /**
714  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
715  * per iteration
716  */
717 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
718 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
719 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
720 #define FREQ_UP_TREND1_ACC   1
721 #define FREQ_UP_TREND2_ACC   100
722 #define FREQ_UP_THRESHOLD    10000
723 
724 	if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
725 			FREQ_GEAR3_RX_PACKET_THRESHOLD) > 0)) {
726 		stats[lcore_id].trend = 0;
727 		return FREQ_HIGHEST;
728 	} else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
729 			FREQ_GEAR2_RX_PACKET_THRESHOLD) > 0))
730 		stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
731 	else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
732 			FREQ_GEAR1_RX_PACKET_THRESHOLD) > 0))
733 		stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
734 
735 	if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
736 		stats[lcore_id].trend = 0;
737 		return FREQ_HIGHER;
738 	}
739 
740 	return FREQ_CURRENT;
741 }
742 
743 /**
744  * force polling thread sleep until one-shot rx interrupt triggers
745  * @param port_id
746  *  Port id.
747  * @param queue_id
748  *  Rx queue id.
749  * @return
750  *  0 on success
751  */
752 static int
753 sleep_until_rx_interrupt(int num)
754 {
755 	struct rte_epoll_event event[num];
756 	int n, i;
757 	uint8_t port_id, queue_id;
758 	void *data;
759 
760 	RTE_LOG(INFO, L3FWD_POWER,
761 		"lcore %u sleeps until interrupt triggers\n",
762 		rte_lcore_id());
763 
764 	n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
765 	for (i = 0; i < n; i++) {
766 		data = event[i].epdata.data;
767 		port_id = ((uintptr_t)data) >> CHAR_BIT;
768 		queue_id = ((uintptr_t)data) &
769 			RTE_LEN2MASK(CHAR_BIT, uint8_t);
770 		rte_eth_dev_rx_intr_disable(port_id, queue_id);
771 		RTE_LOG(INFO, L3FWD_POWER,
772 			"lcore %u is waked up from rx interrupt on"
773 			" port %d queue %d\n",
774 			rte_lcore_id(), port_id, queue_id);
775 	}
776 
777 	return 0;
778 }
779 
780 static void turn_on_intr(struct lcore_conf *qconf)
781 {
782 	int i;
783 	struct lcore_rx_queue *rx_queue;
784 	uint8_t port_id, queue_id;
785 
786 	for (i = 0; i < qconf->n_rx_queue; ++i) {
787 		rx_queue = &(qconf->rx_queue_list[i]);
788 		port_id = rx_queue->port_id;
789 		queue_id = rx_queue->queue_id;
790 
791 		rte_spinlock_lock(&(locks[port_id]));
792 		rte_eth_dev_rx_intr_enable(port_id, queue_id);
793 		rte_spinlock_unlock(&(locks[port_id]));
794 	}
795 }
796 
797 static int event_register(struct lcore_conf *qconf)
798 {
799 	struct lcore_rx_queue *rx_queue;
800 	uint8_t portid, queueid;
801 	uint32_t data;
802 	int ret;
803 	int i;
804 
805 	for (i = 0; i < qconf->n_rx_queue; ++i) {
806 		rx_queue = &(qconf->rx_queue_list[i]);
807 		portid = rx_queue->port_id;
808 		queueid = rx_queue->queue_id;
809 		data = portid << CHAR_BIT | queueid;
810 
811 		ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
812 						RTE_EPOLL_PER_THREAD,
813 						RTE_INTR_EVENT_ADD,
814 						(void *)((uintptr_t)data));
815 		if (ret)
816 			return ret;
817 	}
818 
819 	return 0;
820 }
821 
822 /* main processing loop */
823 static int
824 main_loop(__attribute__((unused)) void *dummy)
825 {
826 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
827 	unsigned lcore_id;
828 	uint64_t prev_tsc, diff_tsc, cur_tsc;
829 	uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
830 	int i, j, nb_rx;
831 	uint8_t portid, queueid;
832 	struct lcore_conf *qconf;
833 	struct lcore_rx_queue *rx_queue;
834 	enum freq_scale_hint_t lcore_scaleup_hint;
835 	uint32_t lcore_rx_idle_count = 0;
836 	uint32_t lcore_idle_hint = 0;
837 	int intr_en = 0;
838 
839 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
840 
841 	prev_tsc = 0;
842 
843 	lcore_id = rte_lcore_id();
844 	qconf = &lcore_conf[lcore_id];
845 
846 	if (qconf->n_rx_queue == 0) {
847 		RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
848 		return 0;
849 	}
850 
851 	RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
852 
853 	for (i = 0; i < qconf->n_rx_queue; i++) {
854 		portid = qconf->rx_queue_list[i].port_id;
855 		queueid = qconf->rx_queue_list[i].queue_id;
856 		RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%hhu "
857 			"rxqueueid=%hhu\n", lcore_id, portid, queueid);
858 	}
859 
860 	/* add into event wait list */
861 	if (event_register(qconf) == 0)
862 		intr_en = 1;
863 	else
864 		RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
865 
866 	while (1) {
867 		stats[lcore_id].nb_iteration_looped++;
868 
869 		cur_tsc = rte_rdtsc();
870 		cur_tsc_power = cur_tsc;
871 
872 		/*
873 		 * TX burst queue drain
874 		 */
875 		diff_tsc = cur_tsc - prev_tsc;
876 		if (unlikely(diff_tsc > drain_tsc)) {
877 			for (i = 0; i < qconf->n_tx_port; ++i) {
878 				portid = qconf->tx_port_id[i];
879 				rte_eth_tx_buffer_flush(portid,
880 						qconf->tx_queue_id[portid],
881 						qconf->tx_buffer[portid]);
882 			}
883 			prev_tsc = cur_tsc;
884 		}
885 
886 		diff_tsc_power = cur_tsc_power - prev_tsc_power;
887 		if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
888 			rte_timer_manage();
889 			prev_tsc_power = cur_tsc_power;
890 		}
891 
892 start_rx:
893 		/*
894 		 * Read packet from RX queues
895 		 */
896 		lcore_scaleup_hint = FREQ_CURRENT;
897 		lcore_rx_idle_count = 0;
898 		for (i = 0; i < qconf->n_rx_queue; ++i) {
899 			rx_queue = &(qconf->rx_queue_list[i]);
900 			rx_queue->idle_hint = 0;
901 			portid = rx_queue->port_id;
902 			queueid = rx_queue->queue_id;
903 
904 			nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
905 								MAX_PKT_BURST);
906 
907 			stats[lcore_id].nb_rx_processed += nb_rx;
908 			if (unlikely(nb_rx == 0)) {
909 				/**
910 				 * no packet received from rx queue, try to
911 				 * sleep for a while forcing CPU enter deeper
912 				 * C states.
913 				 */
914 				rx_queue->zero_rx_packet_count++;
915 
916 				if (rx_queue->zero_rx_packet_count <=
917 							MIN_ZERO_POLL_COUNT)
918 					continue;
919 
920 				rx_queue->idle_hint = power_idle_heuristic(\
921 					rx_queue->zero_rx_packet_count);
922 				lcore_rx_idle_count++;
923 			} else {
924 				rx_queue->zero_rx_packet_count = 0;
925 
926 				/**
927 				 * do not scale up frequency immediately as
928 				 * user to kernel space communication is costly
929 				 * which might impact packet I/O for received
930 				 * packets.
931 				 */
932 				rx_queue->freq_up_hint =
933 					power_freq_scaleup_heuristic(lcore_id,
934 							portid, queueid);
935 			}
936 
937 			/* Prefetch first packets */
938 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
939 				rte_prefetch0(rte_pktmbuf_mtod(
940 						pkts_burst[j], void *));
941 			}
942 
943 			/* Prefetch and forward already prefetched packets */
944 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
945 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
946 						j + PREFETCH_OFFSET], void *));
947 				l3fwd_simple_forward(pkts_burst[j], portid,
948 								qconf);
949 			}
950 
951 			/* Forward remaining prefetched packets */
952 			for (; j < nb_rx; j++) {
953 				l3fwd_simple_forward(pkts_burst[j], portid,
954 								qconf);
955 			}
956 		}
957 
958 		if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
959 			for (i = 1, lcore_scaleup_hint =
960 				qconf->rx_queue_list[0].freq_up_hint;
961 					i < qconf->n_rx_queue; ++i) {
962 				rx_queue = &(qconf->rx_queue_list[i]);
963 				if (rx_queue->freq_up_hint >
964 						lcore_scaleup_hint)
965 					lcore_scaleup_hint =
966 						rx_queue->freq_up_hint;
967 			}
968 
969 			if (lcore_scaleup_hint == FREQ_HIGHEST) {
970 				if (rte_power_freq_max)
971 					rte_power_freq_max(lcore_id);
972 			} else if (lcore_scaleup_hint == FREQ_HIGHER) {
973 				if (rte_power_freq_up)
974 					rte_power_freq_up(lcore_id);
975 			}
976 		} else {
977 			/**
978 			 * All Rx queues empty in recent consecutive polls,
979 			 * sleep in a conservative manner, meaning sleep as
980 			 * less as possible.
981 			 */
982 			for (i = 1, lcore_idle_hint =
983 				qconf->rx_queue_list[0].idle_hint;
984 					i < qconf->n_rx_queue; ++i) {
985 				rx_queue = &(qconf->rx_queue_list[i]);
986 				if (rx_queue->idle_hint < lcore_idle_hint)
987 					lcore_idle_hint = rx_queue->idle_hint;
988 			}
989 
990 			if (lcore_idle_hint < SUSPEND_THRESHOLD)
991 				/**
992 				 * execute "pause" instruction to avoid context
993 				 * switch which generally take hundred of
994 				 * microseconds for short sleep.
995 				 */
996 				rte_delay_us(lcore_idle_hint);
997 			else {
998 				/* suspend until rx interrupt trigges */
999 				if (intr_en) {
1000 					turn_on_intr(qconf);
1001 					sleep_until_rx_interrupt(
1002 						qconf->n_rx_queue);
1003 				}
1004 				/* start receiving packets immediately */
1005 				goto start_rx;
1006 			}
1007 			stats[lcore_id].sleep_time += lcore_idle_hint;
1008 		}
1009 	}
1010 }
1011 
1012 static int
1013 check_lcore_params(void)
1014 {
1015 	uint8_t queue, lcore;
1016 	uint16_t i;
1017 	int socketid;
1018 
1019 	for (i = 0; i < nb_lcore_params; ++i) {
1020 		queue = lcore_params[i].queue_id;
1021 		if (queue >= MAX_RX_QUEUE_PER_PORT) {
1022 			printf("invalid queue number: %hhu\n", queue);
1023 			return -1;
1024 		}
1025 		lcore = lcore_params[i].lcore_id;
1026 		if (!rte_lcore_is_enabled(lcore)) {
1027 			printf("error: lcore %hhu is not enabled in lcore "
1028 							"mask\n", lcore);
1029 			return -1;
1030 		}
1031 		if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1032 							(numa_on == 0)) {
1033 			printf("warning: lcore %hhu is on socket %d with numa "
1034 						"off\n", lcore, socketid);
1035 		}
1036 	}
1037 	return 0;
1038 }
1039 
1040 static int
1041 check_port_config(const unsigned nb_ports)
1042 {
1043 	unsigned portid;
1044 	uint16_t i;
1045 
1046 	for (i = 0; i < nb_lcore_params; ++i) {
1047 		portid = lcore_params[i].port_id;
1048 		if ((enabled_port_mask & (1 << portid)) == 0) {
1049 			printf("port %u is not enabled in port mask\n",
1050 								portid);
1051 			return -1;
1052 		}
1053 		if (portid >= nb_ports) {
1054 			printf("port %u is not present on the board\n",
1055 								portid);
1056 			return -1;
1057 		}
1058 	}
1059 	return 0;
1060 }
1061 
1062 static uint8_t
1063 get_port_n_rx_queues(const uint8_t port)
1064 {
1065 	int queue = -1;
1066 	uint16_t i;
1067 
1068 	for (i = 0; i < nb_lcore_params; ++i) {
1069 		if (lcore_params[i].port_id == port &&
1070 				lcore_params[i].queue_id > queue)
1071 			queue = lcore_params[i].queue_id;
1072 	}
1073 	return (uint8_t)(++queue);
1074 }
1075 
1076 static int
1077 init_lcore_rx_queues(void)
1078 {
1079 	uint16_t i, nb_rx_queue;
1080 	uint8_t lcore;
1081 
1082 	for (i = 0; i < nb_lcore_params; ++i) {
1083 		lcore = lcore_params[i].lcore_id;
1084 		nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1085 		if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1086 			printf("error: too many queues (%u) for lcore: %u\n",
1087 				(unsigned)nb_rx_queue + 1, (unsigned)lcore);
1088 			return -1;
1089 		} else {
1090 			lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1091 				lcore_params[i].port_id;
1092 			lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1093 				lcore_params[i].queue_id;
1094 			lcore_conf[lcore].n_rx_queue++;
1095 		}
1096 	}
1097 	return 0;
1098 }
1099 
1100 /* display usage */
1101 static void
1102 print_usage(const char *prgname)
1103 {
1104 	printf ("%s [EAL options] -- -p PORTMASK -P"
1105 		"  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1106 		"  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1107 		"  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1108 		"  -P : enable promiscuous mode\n"
1109 		"  --config (port,queue,lcore): rx queues configuration\n"
1110 		"  --no-numa: optional, disable numa awareness\n"
1111 		"  --enable-jumbo: enable jumbo frame"
1112 		" which max packet len is PKTLEN in decimal (64-9600)\n",
1113 		prgname);
1114 }
1115 
1116 static int parse_max_pkt_len(const char *pktlen)
1117 {
1118 	char *end = NULL;
1119 	unsigned long len;
1120 
1121 	/* parse decimal string */
1122 	len = strtoul(pktlen, &end, 10);
1123 	if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1124 		return -1;
1125 
1126 	if (len == 0)
1127 		return -1;
1128 
1129 	return len;
1130 }
1131 
1132 static int
1133 parse_portmask(const char *portmask)
1134 {
1135 	char *end = NULL;
1136 	unsigned long pm;
1137 
1138 	/* parse hexadecimal string */
1139 	pm = strtoul(portmask, &end, 16);
1140 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1141 		return -1;
1142 
1143 	if (pm == 0)
1144 		return -1;
1145 
1146 	return pm;
1147 }
1148 
1149 static int
1150 parse_config(const char *q_arg)
1151 {
1152 	char s[256];
1153 	const char *p, *p0 = q_arg;
1154 	char *end;
1155 	enum fieldnames {
1156 		FLD_PORT = 0,
1157 		FLD_QUEUE,
1158 		FLD_LCORE,
1159 		_NUM_FLD
1160 	};
1161 	unsigned long int_fld[_NUM_FLD];
1162 	char *str_fld[_NUM_FLD];
1163 	int i;
1164 	unsigned size;
1165 
1166 	nb_lcore_params = 0;
1167 
1168 	while ((p = strchr(p0,'(')) != NULL) {
1169 		++p;
1170 		if((p0 = strchr(p,')')) == NULL)
1171 			return -1;
1172 
1173 		size = p0 - p;
1174 		if(size >= sizeof(s))
1175 			return -1;
1176 
1177 		snprintf(s, sizeof(s), "%.*s", size, p);
1178 		if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1179 								_NUM_FLD)
1180 			return -1;
1181 		for (i = 0; i < _NUM_FLD; i++){
1182 			errno = 0;
1183 			int_fld[i] = strtoul(str_fld[i], &end, 0);
1184 			if (errno != 0 || end == str_fld[i] || int_fld[i] >
1185 									255)
1186 				return -1;
1187 		}
1188 		if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1189 			printf("exceeded max number of lcore params: %hu\n",
1190 				nb_lcore_params);
1191 			return -1;
1192 		}
1193 		lcore_params_array[nb_lcore_params].port_id =
1194 				(uint8_t)int_fld[FLD_PORT];
1195 		lcore_params_array[nb_lcore_params].queue_id =
1196 				(uint8_t)int_fld[FLD_QUEUE];
1197 		lcore_params_array[nb_lcore_params].lcore_id =
1198 				(uint8_t)int_fld[FLD_LCORE];
1199 		++nb_lcore_params;
1200 	}
1201 	lcore_params = lcore_params_array;
1202 
1203 	return 0;
1204 }
1205 
1206 /* Parse the argument given in the command line of the application */
1207 static int
1208 parse_args(int argc, char **argv)
1209 {
1210 	int opt, ret;
1211 	char **argvopt;
1212 	int option_index;
1213 	char *prgname = argv[0];
1214 	static struct option lgopts[] = {
1215 		{"config", 1, 0, 0},
1216 		{"no-numa", 0, 0, 0},
1217 		{"enable-jumbo", 0, 0, 0},
1218 		{NULL, 0, 0, 0}
1219 	};
1220 
1221 	argvopt = argv;
1222 
1223 	while ((opt = getopt_long(argc, argvopt, "p:P",
1224 				lgopts, &option_index)) != EOF) {
1225 
1226 		switch (opt) {
1227 		/* portmask */
1228 		case 'p':
1229 			enabled_port_mask = parse_portmask(optarg);
1230 			if (enabled_port_mask == 0) {
1231 				printf("invalid portmask\n");
1232 				print_usage(prgname);
1233 				return -1;
1234 			}
1235 			break;
1236 		case 'P':
1237 			printf("Promiscuous mode selected\n");
1238 			promiscuous_on = 1;
1239 			break;
1240 
1241 		/* long options */
1242 		case 0:
1243 			if (!strncmp(lgopts[option_index].name, "config", 6)) {
1244 				ret = parse_config(optarg);
1245 				if (ret) {
1246 					printf("invalid config\n");
1247 					print_usage(prgname);
1248 					return -1;
1249 				}
1250 			}
1251 
1252 			if (!strncmp(lgopts[option_index].name,
1253 						"no-numa", 7)) {
1254 				printf("numa is disabled \n");
1255 				numa_on = 0;
1256 			}
1257 
1258 			if (!strncmp(lgopts[option_index].name,
1259 					"enable-jumbo", 12)) {
1260 				struct option lenopts =
1261 					{"max-pkt-len", required_argument, \
1262 									0, 0};
1263 
1264 				printf("jumbo frame is enabled \n");
1265 				port_conf.rxmode.jumbo_frame = 1;
1266 
1267 				/**
1268 				 * if no max-pkt-len set, use the default value
1269 				 * ETHER_MAX_LEN
1270 				 */
1271 				if (0 == getopt_long(argc, argvopt, "",
1272 						&lenopts, &option_index)) {
1273 					ret = parse_max_pkt_len(optarg);
1274 					if ((ret < 64) ||
1275 						(ret > MAX_JUMBO_PKT_LEN)){
1276 						printf("invalid packet "
1277 								"length\n");
1278 						print_usage(prgname);
1279 						return -1;
1280 					}
1281 					port_conf.rxmode.max_rx_pkt_len = ret;
1282 				}
1283 				printf("set jumbo frame "
1284 					"max packet length to %u\n",
1285 				(unsigned int)port_conf.rxmode.max_rx_pkt_len);
1286 			}
1287 
1288 			break;
1289 
1290 		default:
1291 			print_usage(prgname);
1292 			return -1;
1293 		}
1294 	}
1295 
1296 	if (optind >= 0)
1297 		argv[optind-1] = prgname;
1298 
1299 	ret = optind-1;
1300 	optind = 0; /* reset getopt lib */
1301 	return ret;
1302 }
1303 
1304 static void
1305 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1306 {
1307 	char buf[ETHER_ADDR_FMT_SIZE];
1308 	ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1309 	printf("%s%s", name, buf);
1310 }
1311 
1312 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1313 static void
1314 setup_hash(int socketid)
1315 {
1316 	struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1317 		.name = NULL,
1318 		.entries = L3FWD_HASH_ENTRIES,
1319 		.key_len = sizeof(struct ipv4_5tuple),
1320 		.hash_func = DEFAULT_HASH_FUNC,
1321 		.hash_func_init_val = 0,
1322 	};
1323 
1324 	struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1325 		.name = NULL,
1326 		.entries = L3FWD_HASH_ENTRIES,
1327 		.key_len = sizeof(struct ipv6_5tuple),
1328 		.hash_func = DEFAULT_HASH_FUNC,
1329 		.hash_func_init_val = 0,
1330 	};
1331 
1332 	unsigned i;
1333 	int ret;
1334 	char s[64];
1335 
1336 	/* create ipv4 hash */
1337 	snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1338 	ipv4_l3fwd_hash_params.name = s;
1339 	ipv4_l3fwd_hash_params.socket_id = socketid;
1340 	ipv4_l3fwd_lookup_struct[socketid] =
1341 		rte_hash_create(&ipv4_l3fwd_hash_params);
1342 	if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1343 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1344 				"socket %d\n", socketid);
1345 
1346 	/* create ipv6 hash */
1347 	snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1348 	ipv6_l3fwd_hash_params.name = s;
1349 	ipv6_l3fwd_hash_params.socket_id = socketid;
1350 	ipv6_l3fwd_lookup_struct[socketid] =
1351 		rte_hash_create(&ipv6_l3fwd_hash_params);
1352 	if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1353 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1354 				"socket %d\n", socketid);
1355 
1356 
1357 	/* populate the ipv4 hash */
1358 	for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1359 		ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1360 				(void *) &ipv4_l3fwd_route_array[i].key);
1361 		if (ret < 0) {
1362 			rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1363 				"l3fwd hash on socket %d\n", i, socketid);
1364 		}
1365 		ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1366 		printf("Hash: Adding key\n");
1367 		print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1368 	}
1369 
1370 	/* populate the ipv6 hash */
1371 	for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1372 		ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1373 				(void *) &ipv6_l3fwd_route_array[i].key);
1374 		if (ret < 0) {
1375 			rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1376 				"l3fwd hash on socket %d\n", i, socketid);
1377 		}
1378 		ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1379 		printf("Hash: Adding key\n");
1380 		print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1381 	}
1382 }
1383 #endif
1384 
1385 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1386 static void
1387 setup_lpm(int socketid)
1388 {
1389 	unsigned i;
1390 	int ret;
1391 	char s[64];
1392 
1393 	/* create the LPM table */
1394 	struct rte_lpm_config lpm_ipv4_config;
1395 
1396 	lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
1397 	lpm_ipv4_config.number_tbl8s = 256;
1398 	lpm_ipv4_config.flags = 0;
1399 
1400 	snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1401 	ipv4_l3fwd_lookup_struct[socketid] =
1402 			rte_lpm_create(s, socketid, &lpm_ipv4_config);
1403 	if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1404 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1405 				" on socket %d\n", socketid);
1406 
1407 	/* populate the LPM table */
1408 	for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1409 		ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1410 			ipv4_l3fwd_route_array[i].ip,
1411 			ipv4_l3fwd_route_array[i].depth,
1412 			ipv4_l3fwd_route_array[i].if_out);
1413 
1414 		if (ret < 0) {
1415 			rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1416 				"l3fwd LPM table on socket %d\n",
1417 				i, socketid);
1418 		}
1419 
1420 		printf("LPM: Adding route 0x%08x / %d (%d)\n",
1421 			(unsigned)ipv4_l3fwd_route_array[i].ip,
1422 			ipv4_l3fwd_route_array[i].depth,
1423 			ipv4_l3fwd_route_array[i].if_out);
1424 	}
1425 }
1426 #endif
1427 
1428 static int
1429 init_mem(unsigned nb_mbuf)
1430 {
1431 	struct lcore_conf *qconf;
1432 	int socketid;
1433 	unsigned lcore_id;
1434 	char s[64];
1435 
1436 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1437 		if (rte_lcore_is_enabled(lcore_id) == 0)
1438 			continue;
1439 
1440 		if (numa_on)
1441 			socketid = rte_lcore_to_socket_id(lcore_id);
1442 		else
1443 			socketid = 0;
1444 
1445 		if (socketid >= NB_SOCKETS) {
1446 			rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1447 					"out of range %d\n", socketid,
1448 						lcore_id, NB_SOCKETS);
1449 		}
1450 		if (pktmbuf_pool[socketid] == NULL) {
1451 			snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1452 			pktmbuf_pool[socketid] =
1453 				rte_pktmbuf_pool_create(s, nb_mbuf,
1454 					MEMPOOL_CACHE_SIZE, 0,
1455 					RTE_MBUF_DEFAULT_BUF_SIZE,
1456 					socketid);
1457 			if (pktmbuf_pool[socketid] == NULL)
1458 				rte_exit(EXIT_FAILURE,
1459 					"Cannot init mbuf pool on socket %d\n",
1460 								socketid);
1461 			else
1462 				printf("Allocated mbuf pool on socket %d\n",
1463 								socketid);
1464 
1465 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1466 			setup_lpm(socketid);
1467 #else
1468 			setup_hash(socketid);
1469 #endif
1470 		}
1471 		qconf = &lcore_conf[lcore_id];
1472 		qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1473 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1474 		qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1475 #endif
1476 	}
1477 	return 0;
1478 }
1479 
1480 /* Check the link status of all ports in up to 9s, and print them finally */
1481 static void
1482 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1483 {
1484 #define CHECK_INTERVAL 100 /* 100ms */
1485 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1486 	uint8_t portid, count, all_ports_up, print_flag = 0;
1487 	struct rte_eth_link link;
1488 
1489 	printf("\nChecking link status");
1490 	fflush(stdout);
1491 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
1492 		all_ports_up = 1;
1493 		for (portid = 0; portid < port_num; portid++) {
1494 			if ((port_mask & (1 << portid)) == 0)
1495 				continue;
1496 			memset(&link, 0, sizeof(link));
1497 			rte_eth_link_get_nowait(portid, &link);
1498 			/* print link status if flag set */
1499 			if (print_flag == 1) {
1500 				if (link.link_status)
1501 					printf("Port %d Link Up - speed %u "
1502 						"Mbps - %s\n", (uint8_t)portid,
1503 						(unsigned)link.link_speed,
1504 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1505 					("full-duplex") : ("half-duplex\n"));
1506 				else
1507 					printf("Port %d Link Down\n",
1508 						(uint8_t)portid);
1509 				continue;
1510 			}
1511 			/* clear all_ports_up flag if any link down */
1512 			if (link.link_status == 0) {
1513 				all_ports_up = 0;
1514 				break;
1515 			}
1516 		}
1517 		/* after finally printing all link status, get out */
1518 		if (print_flag == 1)
1519 			break;
1520 
1521 		if (all_ports_up == 0) {
1522 			printf(".");
1523 			fflush(stdout);
1524 			rte_delay_ms(CHECK_INTERVAL);
1525 		}
1526 
1527 		/* set the print_flag if all ports up or timeout */
1528 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1529 			print_flag = 1;
1530 			printf("done\n");
1531 		}
1532 	}
1533 }
1534 
1535 int
1536 main(int argc, char **argv)
1537 {
1538 	struct lcore_conf *qconf;
1539 	struct rte_eth_dev_info dev_info;
1540 	struct rte_eth_txconf *txconf;
1541 	int ret;
1542 	unsigned nb_ports;
1543 	uint16_t queueid;
1544 	unsigned lcore_id;
1545 	uint64_t hz;
1546 	uint32_t n_tx_queue, nb_lcores;
1547 	uint32_t dev_rxq_num, dev_txq_num;
1548 	uint8_t portid, nb_rx_queue, queue, socketid;
1549 	uint8_t nb_tx_port;
1550 
1551 	/* catch SIGINT and restore cpufreq governor to ondemand */
1552 	signal(SIGINT, signal_exit_now);
1553 
1554 	/* init EAL */
1555 	ret = rte_eal_init(argc, argv);
1556 	if (ret < 0)
1557 		rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1558 	argc -= ret;
1559 	argv += ret;
1560 
1561 	/* init RTE timer library to be used late */
1562 	rte_timer_subsystem_init();
1563 
1564 	/* parse application arguments (after the EAL ones) */
1565 	ret = parse_args(argc, argv);
1566 	if (ret < 0)
1567 		rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1568 
1569 	if (check_lcore_params() < 0)
1570 		rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1571 
1572 	ret = init_lcore_rx_queues();
1573 	if (ret < 0)
1574 		rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1575 
1576 
1577 	nb_ports = rte_eth_dev_count();
1578 	if (nb_ports > RTE_MAX_ETHPORTS)
1579 		nb_ports = RTE_MAX_ETHPORTS;
1580 
1581 	if (check_port_config(nb_ports) < 0)
1582 		rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1583 
1584 	nb_lcores = rte_lcore_count();
1585 	nb_tx_port = 0;
1586 
1587 	/* initialize all ports */
1588 	for (portid = 0; portid < nb_ports; portid++) {
1589 		/* skip ports that are not enabled */
1590 		if ((enabled_port_mask & (1 << portid)) == 0) {
1591 			printf("\nSkipping disabled port %d\n", portid);
1592 			continue;
1593 		}
1594 
1595 		/* init port */
1596 		printf("Initializing port %d ... ", portid );
1597 		fflush(stdout);
1598 
1599 		rte_eth_dev_info_get(portid, &dev_info);
1600 		dev_rxq_num = dev_info.max_rx_queues;
1601 		dev_txq_num = dev_info.max_tx_queues;
1602 
1603 		nb_rx_queue = get_port_n_rx_queues(portid);
1604 		if (nb_rx_queue > dev_rxq_num)
1605 			rte_exit(EXIT_FAILURE,
1606 				"Cannot configure not existed rxq: "
1607 				"port=%d\n", portid);
1608 
1609 		n_tx_queue = nb_lcores;
1610 		if (n_tx_queue > dev_txq_num)
1611 			n_tx_queue = dev_txq_num;
1612 		printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1613 			nb_rx_queue, (unsigned)n_tx_queue );
1614 		ret = rte_eth_dev_configure(portid, nb_rx_queue,
1615 					(uint16_t)n_tx_queue, &port_conf);
1616 		if (ret < 0)
1617 			rte_exit(EXIT_FAILURE, "Cannot configure device: "
1618 					"err=%d, port=%d\n", ret, portid);
1619 
1620 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1621 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
1622 		printf(", ");
1623 
1624 		/* init memory */
1625 		ret = init_mem(NB_MBUF);
1626 		if (ret < 0)
1627 			rte_exit(EXIT_FAILURE, "init_mem failed\n");
1628 
1629 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1630 			if (rte_lcore_is_enabled(lcore_id) == 0)
1631 				continue;
1632 
1633 			/* Initialize TX buffers */
1634 			qconf = &lcore_conf[lcore_id];
1635 			qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
1636 				RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
1637 				rte_eth_dev_socket_id(portid));
1638 			if (qconf->tx_buffer[portid] == NULL)
1639 				rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
1640 						(unsigned) portid);
1641 
1642 			rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
1643 		}
1644 
1645 		/* init one TX queue per couple (lcore,port) */
1646 		queueid = 0;
1647 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1648 			if (rte_lcore_is_enabled(lcore_id) == 0)
1649 				continue;
1650 
1651 			if (queueid >= dev_txq_num)
1652 				continue;
1653 
1654 			if (numa_on)
1655 				socketid = \
1656 				(uint8_t)rte_lcore_to_socket_id(lcore_id);
1657 			else
1658 				socketid = 0;
1659 
1660 			printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1661 			fflush(stdout);
1662 
1663 			rte_eth_dev_info_get(portid, &dev_info);
1664 			txconf = &dev_info.default_txconf;
1665 			if (port_conf.rxmode.jumbo_frame)
1666 				txconf->txq_flags = 0;
1667 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1668 						     socketid, txconf);
1669 			if (ret < 0)
1670 				rte_exit(EXIT_FAILURE,
1671 					"rte_eth_tx_queue_setup: err=%d, "
1672 						"port=%d\n", ret, portid);
1673 
1674 			qconf = &lcore_conf[lcore_id];
1675 			qconf->tx_queue_id[portid] = queueid;
1676 			queueid++;
1677 
1678 			qconf->n_tx_port = nb_tx_port;
1679 			qconf->tx_port_id[qconf->n_tx_port] = portid;
1680 		}
1681 		printf("\n");
1682 
1683 		nb_tx_port++;
1684 	}
1685 
1686 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1687 		if (rte_lcore_is_enabled(lcore_id) == 0)
1688 			continue;
1689 
1690 		/* init power management library */
1691 		ret = rte_power_init(lcore_id);
1692 		if (ret)
1693 			RTE_LOG(ERR, POWER,
1694 				"Library initialization failed on core %u\n", lcore_id);
1695 
1696 		/* init timer structures for each enabled lcore */
1697 		rte_timer_init(&power_timers[lcore_id]);
1698 		hz = rte_get_timer_hz();
1699 		rte_timer_reset(&power_timers[lcore_id],
1700 			hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1701 						power_timer_cb, NULL);
1702 
1703 		qconf = &lcore_conf[lcore_id];
1704 		printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1705 		fflush(stdout);
1706 		/* init RX queues */
1707 		for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1708 			portid = qconf->rx_queue_list[queue].port_id;
1709 			queueid = qconf->rx_queue_list[queue].queue_id;
1710 
1711 			if (numa_on)
1712 				socketid = \
1713 				(uint8_t)rte_lcore_to_socket_id(lcore_id);
1714 			else
1715 				socketid = 0;
1716 
1717 			printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1718 			fflush(stdout);
1719 
1720 			ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1721 				socketid, NULL,
1722 				pktmbuf_pool[socketid]);
1723 			if (ret < 0)
1724 				rte_exit(EXIT_FAILURE,
1725 					"rte_eth_rx_queue_setup: err=%d, "
1726 						"port=%d\n", ret, portid);
1727 		}
1728 	}
1729 
1730 	printf("\n");
1731 
1732 	/* start ports */
1733 	for (portid = 0; portid < nb_ports; portid++) {
1734 		if ((enabled_port_mask & (1 << portid)) == 0) {
1735 			continue;
1736 		}
1737 		/* Start device */
1738 		ret = rte_eth_dev_start(portid);
1739 		if (ret < 0)
1740 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1741 						"port=%d\n", ret, portid);
1742 		/*
1743 		 * If enabled, put device in promiscuous mode.
1744 		 * This allows IO forwarding mode to forward packets
1745 		 * to itself through 2 cross-connected  ports of the
1746 		 * target machine.
1747 		 */
1748 		if (promiscuous_on)
1749 			rte_eth_promiscuous_enable(portid);
1750 		/* initialize spinlock for each port */
1751 		rte_spinlock_init(&(locks[portid]));
1752 	}
1753 
1754 	check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1755 
1756 	/* launch per-lcore init on every lcore */
1757 	rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1758 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1759 		if (rte_eal_wait_lcore(lcore_id) < 0)
1760 			return -1;
1761 	}
1762 
1763 	return 0;
1764 }
1765