xref: /dpdk/examples/l3fwd-power/main.c (revision fdf7471cccb8be023037c218d1402c0549eb2c8e)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2018 Intel Corporation
3  */
4 
5 #include <stdio.h>
6 #include <stdlib.h>
7 #include <stdint.h>
8 #include <inttypes.h>
9 #include <sys/types.h>
10 #include <string.h>
11 #include <sys/queue.h>
12 #include <stdarg.h>
13 #include <errno.h>
14 #include <getopt.h>
15 #include <unistd.h>
16 #include <signal.h>
17 #include <math.h>
18 
19 #include <rte_common.h>
20 #include <rte_byteorder.h>
21 #include <rte_log.h>
22 #include <rte_malloc.h>
23 #include <rte_memory.h>
24 #include <rte_memcpy.h>
25 #include <rte_eal.h>
26 #include <rte_launch.h>
27 #include <rte_atomic.h>
28 #include <rte_cycles.h>
29 #include <rte_prefetch.h>
30 #include <rte_lcore.h>
31 #include <rte_per_lcore.h>
32 #include <rte_branch_prediction.h>
33 #include <rte_interrupts.h>
34 #include <rte_random.h>
35 #include <rte_debug.h>
36 #include <rte_ether.h>
37 #include <rte_ethdev.h>
38 #include <rte_mempool.h>
39 #include <rte_mbuf.h>
40 #include <rte_ip.h>
41 #include <rte_tcp.h>
42 #include <rte_udp.h>
43 #include <rte_string_fns.h>
44 #include <rte_timer.h>
45 #include <rte_power.h>
46 #include <rte_spinlock.h>
47 #include <rte_power_empty_poll.h>
48 #include <rte_metrics.h>
49 
50 #include "perf_core.h"
51 #include "main.h"
52 
53 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
54 
55 #define MAX_PKT_BURST 32
56 
57 #define MIN_ZERO_POLL_COUNT 10
58 
59 /* 100 ms interval */
60 #define TIMER_NUMBER_PER_SECOND           10
61 /* (10ms) */
62 #define INTERVALS_PER_SECOND             100
63 /* 100000 us */
64 #define SCALING_PERIOD                    (1000000/TIMER_NUMBER_PER_SECOND)
65 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
66 
67 #define APP_LOOKUP_EXACT_MATCH          0
68 #define APP_LOOKUP_LPM                  1
69 #define DO_RFC_1812_CHECKS
70 
71 #ifndef APP_LOOKUP_METHOD
72 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
73 #endif
74 
75 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
76 #include <rte_hash.h>
77 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
78 #include <rte_lpm.h>
79 #else
80 #error "APP_LOOKUP_METHOD set to incorrect value"
81 #endif
82 
83 #ifndef IPv6_BYTES
84 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
85                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
86 #define IPv6_BYTES(addr) \
87 	addr[0],  addr[1], addr[2],  addr[3], \
88 	addr[4],  addr[5], addr[6],  addr[7], \
89 	addr[8],  addr[9], addr[10], addr[11],\
90 	addr[12], addr[13],addr[14], addr[15]
91 #endif
92 
93 #define MAX_JUMBO_PKT_LEN  9600
94 
95 #define IPV6_ADDR_LEN 16
96 
97 #define MEMPOOL_CACHE_SIZE 256
98 
99 /*
100  * This expression is used to calculate the number of mbufs needed depending on
101  * user input, taking into account memory for rx and tx hardware rings, cache
102  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
103  * NB_MBUF never goes below a minimum value of 8192.
104  */
105 
106 #define NB_MBUF RTE_MAX	( \
107 	(nb_ports*nb_rx_queue*nb_rxd + \
108 	nb_ports*nb_lcores*MAX_PKT_BURST + \
109 	nb_ports*n_tx_queue*nb_txd + \
110 	nb_lcores*MEMPOOL_CACHE_SIZE), \
111 	(unsigned)8192)
112 
113 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
114 
115 #define NB_SOCKETS 8
116 
117 /* Configure how many packets ahead to prefetch, when reading packets */
118 #define PREFETCH_OFFSET	3
119 
120 /*
121  * Configurable number of RX/TX ring descriptors
122  */
123 #define RTE_TEST_RX_DESC_DEFAULT 1024
124 #define RTE_TEST_TX_DESC_DEFAULT 1024
125 
126 /*
127  * These two thresholds were decided on by running the training algorithm on
128  * a 2.5GHz Xeon. These defaults can be overridden by supplying non-zero values
129  * for the med_threshold and high_threshold parameters on the command line.
130  */
131 #define EMPTY_POLL_MED_THRESHOLD 350000UL
132 #define EMPTY_POLL_HGH_THRESHOLD 580000UL
133 
134 
135 
136 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
137 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
138 
139 /* ethernet addresses of ports */
140 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
141 
142 /* ethernet addresses of ports */
143 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
144 
145 /* mask of enabled ports */
146 static uint32_t enabled_port_mask = 0;
147 /* Ports set in promiscuous mode off by default. */
148 static int promiscuous_on = 0;
149 /* NUMA is enabled by default. */
150 static int numa_on = 1;
151 static bool empty_poll_stop;
152 static bool empty_poll_train;
153 volatile bool quit_signal;
154 static struct  ep_params *ep_params;
155 static struct  ep_policy policy;
156 static long  ep_med_edpi, ep_hgh_edpi;
157 /* timer to update telemetry every 500ms */
158 static struct rte_timer telemetry_timer;
159 
160 /* stats index returned by metrics lib */
161 int telstats_index;
162 
163 struct telstats_name {
164 	char name[RTE_ETH_XSTATS_NAME_SIZE];
165 };
166 
167 /* telemetry stats to be reported */
168 const struct telstats_name telstats_strings[] = {
169 	{"empty_poll"},
170 	{"full_poll"},
171 	{"busy_percent"}
172 };
173 
174 /* core busyness in percentage */
175 enum busy_rate {
176 	ZERO = 0,
177 	PARTIAL = 50,
178 	FULL = 100
179 };
180 
181 /* reference poll count to measure core busyness */
182 #define DEFAULT_COUNT 10000
183 /*
184  * reference CYCLES to be used to
185  * measure core busyness based on poll count
186  */
187 #define MIN_CYCLES  1500000ULL
188 #define MAX_CYCLES 22000000ULL
189 
190 /* (500ms) */
191 #define TELEMETRY_INTERVALS_PER_SEC 2
192 
193 static int parse_ptype; /**< Parse packet type using rx callback, and */
194 			/**< disabled by default */
195 
196 enum appmode {
197 	APP_MODE_LEGACY = 0,
198 	APP_MODE_EMPTY_POLL,
199 	APP_MODE_TELEMETRY
200 };
201 
202 enum appmode app_mode;
203 
204 enum freq_scale_hint_t
205 {
206 	FREQ_LOWER    =      -1,
207 	FREQ_CURRENT  =       0,
208 	FREQ_HIGHER   =       1,
209 	FREQ_HIGHEST  =       2
210 };
211 
212 struct lcore_rx_queue {
213 	uint16_t port_id;
214 	uint8_t queue_id;
215 	enum freq_scale_hint_t freq_up_hint;
216 	uint32_t zero_rx_packet_count;
217 	uint32_t idle_hint;
218 } __rte_cache_aligned;
219 
220 #define MAX_RX_QUEUE_PER_LCORE 16
221 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
222 #define MAX_RX_QUEUE_PER_PORT 128
223 
224 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
225 
226 
227 struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
228 static struct lcore_params lcore_params_array_default[] = {
229 	{0, 0, 2},
230 	{0, 1, 2},
231 	{0, 2, 2},
232 	{1, 0, 2},
233 	{1, 1, 2},
234 	{1, 2, 2},
235 	{2, 0, 2},
236 	{3, 0, 3},
237 	{3, 1, 3},
238 };
239 
240 struct lcore_params *lcore_params = lcore_params_array_default;
241 uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
242 				sizeof(lcore_params_array_default[0]);
243 
244 static struct rte_eth_conf port_conf = {
245 	.rxmode = {
246 		.mq_mode        = ETH_MQ_RX_RSS,
247 		.max_rx_pkt_len = RTE_ETHER_MAX_LEN,
248 		.split_hdr_size = 0,
249 		.offloads = DEV_RX_OFFLOAD_CHECKSUM,
250 	},
251 	.rx_adv_conf = {
252 		.rss_conf = {
253 			.rss_key = NULL,
254 			.rss_hf = ETH_RSS_UDP,
255 		},
256 	},
257 	.txmode = {
258 		.mq_mode = ETH_MQ_TX_NONE,
259 	},
260 	.intr_conf = {
261 		.rxq = 1,
262 	},
263 };
264 
265 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
266 
267 
268 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
269 
270 #ifdef RTE_ARCH_X86
271 #include <rte_hash_crc.h>
272 #define DEFAULT_HASH_FUNC       rte_hash_crc
273 #else
274 #include <rte_jhash.h>
275 #define DEFAULT_HASH_FUNC       rte_jhash
276 #endif
277 
278 struct ipv4_5tuple {
279 	uint32_t ip_dst;
280 	uint32_t ip_src;
281 	uint16_t port_dst;
282 	uint16_t port_src;
283 	uint8_t  proto;
284 } __attribute__((__packed__));
285 
286 struct ipv6_5tuple {
287 	uint8_t  ip_dst[IPV6_ADDR_LEN];
288 	uint8_t  ip_src[IPV6_ADDR_LEN];
289 	uint16_t port_dst;
290 	uint16_t port_src;
291 	uint8_t  proto;
292 } __attribute__((__packed__));
293 
294 struct ipv4_l3fwd_route {
295 	struct ipv4_5tuple key;
296 	uint8_t if_out;
297 };
298 
299 struct ipv6_l3fwd_route {
300 	struct ipv6_5tuple key;
301 	uint8_t if_out;
302 };
303 
304 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
305 	{{RTE_IPV4(100,10,0,1), RTE_IPV4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
306 	{{RTE_IPV4(100,20,0,2), RTE_IPV4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
307 	{{RTE_IPV4(100,30,0,3), RTE_IPV4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
308 	{{RTE_IPV4(100,40,0,4), RTE_IPV4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
309 };
310 
311 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
312 	{
313 		{
314 			{0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
315 			 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
316 			{0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
317 			 0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
318 			 1, 10, IPPROTO_UDP
319 		}, 4
320 	},
321 };
322 
323 typedef struct rte_hash lookup_struct_t;
324 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
325 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
326 
327 #define L3FWD_HASH_ENTRIES	1024
328 
329 #define IPV4_L3FWD_NUM_ROUTES \
330 	(sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
331 
332 #define IPV6_L3FWD_NUM_ROUTES \
333 	(sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
334 
335 static uint16_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
336 static uint16_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
337 #endif
338 
339 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
340 struct ipv4_l3fwd_route {
341 	uint32_t ip;
342 	uint8_t  depth;
343 	uint8_t  if_out;
344 };
345 
346 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
347 	{RTE_IPV4(1,1,1,0), 24, 0},
348 	{RTE_IPV4(2,1,1,0), 24, 1},
349 	{RTE_IPV4(3,1,1,0), 24, 2},
350 	{RTE_IPV4(4,1,1,0), 24, 3},
351 	{RTE_IPV4(5,1,1,0), 24, 4},
352 	{RTE_IPV4(6,1,1,0), 24, 5},
353 	{RTE_IPV4(7,1,1,0), 24, 6},
354 	{RTE_IPV4(8,1,1,0), 24, 7},
355 };
356 
357 #define IPV4_L3FWD_NUM_ROUTES \
358 	(sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
359 
360 #define IPV4_L3FWD_LPM_MAX_RULES     1024
361 
362 typedef struct rte_lpm lookup_struct_t;
363 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
364 #endif
365 
366 struct lcore_conf {
367 	uint16_t n_rx_queue;
368 	struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
369 	uint16_t n_tx_port;
370 	uint16_t tx_port_id[RTE_MAX_ETHPORTS];
371 	uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
372 	struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
373 	lookup_struct_t * ipv4_lookup_struct;
374 	lookup_struct_t * ipv6_lookup_struct;
375 } __rte_cache_aligned;
376 
377 struct lcore_stats {
378 	/* total sleep time in ms since last frequency scaling down */
379 	uint32_t sleep_time;
380 	/* number of long sleep recently */
381 	uint32_t nb_long_sleep;
382 	/* freq. scaling up trend */
383 	uint32_t trend;
384 	/* total packet processed recently */
385 	uint64_t nb_rx_processed;
386 	/* total iterations looped recently */
387 	uint64_t nb_iteration_looped;
388 	/*
389 	 * Represents empty and non empty polls
390 	 * of rte_eth_rx_burst();
391 	 * ep_nep[0] holds non empty polls
392 	 * i.e. 0 < nb_rx <= MAX_BURST
393 	 * ep_nep[1] holds empty polls.
394 	 * i.e. nb_rx == 0
395 	 */
396 	uint64_t ep_nep[2];
397 	/*
398 	 * Represents full and empty+partial
399 	 * polls of rte_eth_rx_burst();
400 	 * ep_nep[0] holds empty+partial polls.
401 	 * i.e. 0 <= nb_rx < MAX_BURST
402 	 * ep_nep[1] holds full polls
403 	 * i.e. nb_rx == MAX_BURST
404 	 */
405 	uint64_t fp_nfp[2];
406 	enum busy_rate br;
407 	rte_spinlock_t telemetry_lock;
408 } __rte_cache_aligned;
409 
410 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
411 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
412 static struct rte_timer power_timers[RTE_MAX_LCORE];
413 
414 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
415 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
416 		unsigned int lcore_id, uint16_t port_id, uint16_t queue_id);
417 
418 
419 /*
420  * These defaults are using the max frequency index (1), a medium index (9)
421  * and a typical low frequency index (14). These can be adjusted to use
422  * different indexes using the relevant command line parameters.
423  */
424 static uint8_t  freq_tlb[] = {14, 9, 1};
425 
426 static int is_done(void)
427 {
428 	return quit_signal;
429 }
430 
431 /* exit signal handler */
432 static void
433 signal_exit_now(int sigtype)
434 {
435 	unsigned lcore_id;
436 	unsigned int portid;
437 	int ret;
438 
439 	if (sigtype == SIGINT) {
440 		if (app_mode == APP_MODE_EMPTY_POLL ||
441 				app_mode == APP_MODE_TELEMETRY)
442 			quit_signal = true;
443 
444 
445 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
446 			if (rte_lcore_is_enabled(lcore_id) == 0)
447 				continue;
448 
449 			/* init power management library */
450 			ret = rte_power_exit(lcore_id);
451 			if (ret)
452 				rte_exit(EXIT_FAILURE, "Power management "
453 					"library de-initialization failed on "
454 							"core%u\n", lcore_id);
455 		}
456 
457 		if (app_mode != APP_MODE_EMPTY_POLL) {
458 			RTE_ETH_FOREACH_DEV(portid) {
459 				if ((enabled_port_mask & (1 << portid)) == 0)
460 					continue;
461 
462 				rte_eth_dev_stop(portid);
463 				rte_eth_dev_close(portid);
464 			}
465 		}
466 	}
467 
468 	if (app_mode != APP_MODE_EMPTY_POLL)
469 		rte_exit(EXIT_SUCCESS, "User forced exit\n");
470 }
471 
472 /*  Freqency scale down timer callback */
473 static void
474 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
475 			  __attribute__((unused)) void *arg)
476 {
477 	uint64_t hz;
478 	float sleep_time_ratio;
479 	unsigned lcore_id = rte_lcore_id();
480 
481 	/* accumulate total execution time in us when callback is invoked */
482 	sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
483 					(float)SCALING_PERIOD;
484 	/**
485 	 * check whether need to scale down frequency a step if it sleep a lot.
486 	 */
487 	if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
488 		if (rte_power_freq_down)
489 			rte_power_freq_down(lcore_id);
490 	}
491 	else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
492 		stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
493 		/**
494 		 * scale down a step if average packet per iteration less
495 		 * than expectation.
496 		 */
497 		if (rte_power_freq_down)
498 			rte_power_freq_down(lcore_id);
499 	}
500 
501 	/**
502 	 * initialize another timer according to current frequency to ensure
503 	 * timer interval is relatively fixed.
504 	 */
505 	hz = rte_get_timer_hz();
506 	rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
507 				SINGLE, lcore_id, power_timer_cb, NULL);
508 
509 	stats[lcore_id].nb_rx_processed = 0;
510 	stats[lcore_id].nb_iteration_looped = 0;
511 
512 	stats[lcore_id].sleep_time = 0;
513 }
514 
515 /* Enqueue a single packet, and send burst if queue is filled */
516 static inline int
517 send_single_packet(struct rte_mbuf *m, uint16_t port)
518 {
519 	uint32_t lcore_id;
520 	struct lcore_conf *qconf;
521 
522 	lcore_id = rte_lcore_id();
523 	qconf = &lcore_conf[lcore_id];
524 
525 	rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
526 			qconf->tx_buffer[port], m);
527 
528 	return 0;
529 }
530 
531 #ifdef DO_RFC_1812_CHECKS
532 static inline int
533 is_valid_ipv4_pkt(struct rte_ipv4_hdr *pkt, uint32_t link_len)
534 {
535 	/* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
536 	/*
537 	 * 1. The packet length reported by the Link Layer must be large
538 	 * enough to hold the minimum length legal IP datagram (20 bytes).
539 	 */
540 	if (link_len < sizeof(struct rte_ipv4_hdr))
541 		return -1;
542 
543 	/* 2. The IP checksum must be correct. */
544 	/* this is checked in H/W */
545 
546 	/*
547 	 * 3. The IP version number must be 4. If the version number is not 4
548 	 * then the packet may be another version of IP, such as IPng or
549 	 * ST-II.
550 	 */
551 	if (((pkt->version_ihl) >> 4) != 4)
552 		return -3;
553 	/*
554 	 * 4. The IP header length field must be large enough to hold the
555 	 * minimum length legal IP datagram (20 bytes = 5 words).
556 	 */
557 	if ((pkt->version_ihl & 0xf) < 5)
558 		return -4;
559 
560 	/*
561 	 * 5. The IP total length field must be large enough to hold the IP
562 	 * datagram header, whose length is specified in the IP header length
563 	 * field.
564 	 */
565 	if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct rte_ipv4_hdr))
566 		return -5;
567 
568 	return 0;
569 }
570 #endif
571 
572 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
573 static void
574 print_ipv4_key(struct ipv4_5tuple key)
575 {
576 	printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
577 		"proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
578 				key.port_dst, key.port_src, key.proto);
579 }
580 static void
581 print_ipv6_key(struct ipv6_5tuple key)
582 {
583 	printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
584 	        "port dst = %d, port src = %d, proto = %d\n",
585 	        IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
586 	        key.port_dst, key.port_src, key.proto);
587 }
588 
589 static inline uint16_t
590 get_ipv4_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
591 		lookup_struct_t * ipv4_l3fwd_lookup_struct)
592 {
593 	struct ipv4_5tuple key;
594 	struct rte_tcp_hdr *tcp;
595 	struct rte_udp_hdr *udp;
596 	int ret = 0;
597 
598 	key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
599 	key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
600 	key.proto = ipv4_hdr->next_proto_id;
601 
602 	switch (ipv4_hdr->next_proto_id) {
603 	case IPPROTO_TCP:
604 		tcp = (struct rte_tcp_hdr *)((unsigned char *)ipv4_hdr +
605 					sizeof(struct rte_ipv4_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 rte_udp_hdr *)((unsigned char *)ipv4_hdr +
612 					sizeof(struct rte_ipv4_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(ipv4_l3fwd_lookup_struct, (const void *)&key);
625 	return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
626 }
627 
628 static inline uint16_t
629 get_ipv6_dst_port(struct rte_ipv6_hdr *ipv6_hdr, uint16_t portid,
630 			lookup_struct_t *ipv6_l3fwd_lookup_struct)
631 {
632 	struct ipv6_5tuple key;
633 	struct rte_tcp_hdr *tcp;
634 	struct rte_udp_hdr *udp;
635 	int ret = 0;
636 
637 	memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
638 	memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
639 
640 	key.proto = ipv6_hdr->proto;
641 
642 	switch (ipv6_hdr->proto) {
643 	case IPPROTO_TCP:
644 		tcp = (struct rte_tcp_hdr *)((unsigned char *) ipv6_hdr +
645 					sizeof(struct rte_ipv6_hdr));
646 		key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
647 		key.port_src = rte_be_to_cpu_16(tcp->src_port);
648 		break;
649 
650 	case IPPROTO_UDP:
651 		udp = (struct rte_udp_hdr *)((unsigned char *) ipv6_hdr +
652 					sizeof(struct rte_ipv6_hdr));
653 		key.port_dst = rte_be_to_cpu_16(udp->dst_port);
654 		key.port_src = rte_be_to_cpu_16(udp->src_port);
655 		break;
656 
657 	default:
658 		key.port_dst = 0;
659 		key.port_src = 0;
660 		break;
661 	}
662 
663 	/* Find destination port */
664 	ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
665 	return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
666 }
667 #endif
668 
669 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
670 static inline uint16_t
671 get_ipv4_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
672 		lookup_struct_t *ipv4_l3fwd_lookup_struct)
673 {
674 	uint32_t next_hop;
675 
676 	return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
677 			rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
678 			next_hop : portid);
679 }
680 #endif
681 
682 static inline void
683 parse_ptype_one(struct rte_mbuf *m)
684 {
685 	struct rte_ether_hdr *eth_hdr;
686 	uint32_t packet_type = RTE_PTYPE_UNKNOWN;
687 	uint16_t ether_type;
688 
689 	eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
690 	ether_type = eth_hdr->ether_type;
691 	if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
692 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
693 	else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6))
694 		packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
695 
696 	m->packet_type = packet_type;
697 }
698 
699 static uint16_t
700 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
701 	       struct rte_mbuf *pkts[], uint16_t nb_pkts,
702 	       uint16_t max_pkts __rte_unused,
703 	       void *user_param __rte_unused)
704 {
705 	unsigned int i;
706 
707 	for (i = 0; i < nb_pkts; ++i)
708 		parse_ptype_one(pkts[i]);
709 
710 	return nb_pkts;
711 }
712 
713 static int
714 add_cb_parse_ptype(uint16_t portid, uint16_t queueid)
715 {
716 	printf("Port %d: softly parse packet type info\n", portid);
717 	if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL))
718 		return 0;
719 
720 	printf("Failed to add rx callback: port=%d\n", portid);
721 	return -1;
722 }
723 
724 static inline void
725 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
726 				struct lcore_conf *qconf)
727 {
728 	struct rte_ether_hdr *eth_hdr;
729 	struct rte_ipv4_hdr *ipv4_hdr;
730 	void *d_addr_bytes;
731 	uint16_t dst_port;
732 
733 	eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
734 
735 	if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
736 		/* Handle IPv4 headers.*/
737 		ipv4_hdr =
738 			rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
739 						sizeof(struct rte_ether_hdr));
740 
741 #ifdef DO_RFC_1812_CHECKS
742 		/* Check to make sure the packet is valid (RFC1812) */
743 		if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
744 			rte_pktmbuf_free(m);
745 			return;
746 		}
747 #endif
748 
749 		dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
750 					qconf->ipv4_lookup_struct);
751 		if (dst_port >= RTE_MAX_ETHPORTS ||
752 				(enabled_port_mask & 1 << dst_port) == 0)
753 			dst_port = portid;
754 
755 		/* 02:00:00:00:00:xx */
756 		d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
757 		*((uint64_t *)d_addr_bytes) =
758 			0x000000000002 + ((uint64_t)dst_port << 40);
759 
760 #ifdef DO_RFC_1812_CHECKS
761 		/* Update time to live and header checksum */
762 		--(ipv4_hdr->time_to_live);
763 		++(ipv4_hdr->hdr_checksum);
764 #endif
765 
766 		/* src addr */
767 		rte_ether_addr_copy(&ports_eth_addr[dst_port],
768 				&eth_hdr->s_addr);
769 
770 		send_single_packet(m, dst_port);
771 	} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
772 		/* Handle IPv6 headers.*/
773 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
774 		struct rte_ipv6_hdr *ipv6_hdr;
775 
776 		ipv6_hdr =
777 			rte_pktmbuf_mtod_offset(m, struct rte_ipv6_hdr *,
778 						sizeof(struct rte_ether_hdr));
779 
780 		dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
781 					qconf->ipv6_lookup_struct);
782 
783 		if (dst_port >= RTE_MAX_ETHPORTS ||
784 				(enabled_port_mask & 1 << dst_port) == 0)
785 			dst_port = portid;
786 
787 		/* 02:00:00:00:00:xx */
788 		d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
789 		*((uint64_t *)d_addr_bytes) =
790 			0x000000000002 + ((uint64_t)dst_port << 40);
791 
792 		/* src addr */
793 		rte_ether_addr_copy(&ports_eth_addr[dst_port],
794 				&eth_hdr->s_addr);
795 
796 		send_single_packet(m, dst_port);
797 #else
798 		/* We don't currently handle IPv6 packets in LPM mode. */
799 		rte_pktmbuf_free(m);
800 #endif
801 	} else
802 		rte_pktmbuf_free(m);
803 
804 }
805 
806 #define MINIMUM_SLEEP_TIME         1
807 #define SUSPEND_THRESHOLD          300
808 
809 static inline uint32_t
810 power_idle_heuristic(uint32_t zero_rx_packet_count)
811 {
812 	/* If zero count is less than 100,  sleep 1us */
813 	if (zero_rx_packet_count < SUSPEND_THRESHOLD)
814 		return MINIMUM_SLEEP_TIME;
815 	/* If zero count is less than 1000, sleep 100 us which is the
816 		minimum latency switching from C3/C6 to C0
817 	*/
818 	else
819 		return SUSPEND_THRESHOLD;
820 }
821 
822 static inline enum freq_scale_hint_t
823 power_freq_scaleup_heuristic(unsigned lcore_id,
824 			     uint16_t port_id,
825 			     uint16_t queue_id)
826 {
827 	uint32_t rxq_count = rte_eth_rx_queue_count(port_id, queue_id);
828 /**
829  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
830  * per iteration
831  */
832 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
833 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
834 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
835 #define FREQ_UP_TREND1_ACC   1
836 #define FREQ_UP_TREND2_ACC   100
837 #define FREQ_UP_THRESHOLD    10000
838 
839 	if (likely(rxq_count > FREQ_GEAR3_RX_PACKET_THRESHOLD)) {
840 		stats[lcore_id].trend = 0;
841 		return FREQ_HIGHEST;
842 	} else if (likely(rxq_count > FREQ_GEAR2_RX_PACKET_THRESHOLD))
843 		stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
844 	else if (likely(rxq_count > FREQ_GEAR1_RX_PACKET_THRESHOLD))
845 		stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
846 
847 	if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
848 		stats[lcore_id].trend = 0;
849 		return FREQ_HIGHER;
850 	}
851 
852 	return FREQ_CURRENT;
853 }
854 
855 /**
856  * force polling thread sleep until one-shot rx interrupt triggers
857  * @param port_id
858  *  Port id.
859  * @param queue_id
860  *  Rx queue id.
861  * @return
862  *  0 on success
863  */
864 static int
865 sleep_until_rx_interrupt(int num)
866 {
867 	struct rte_epoll_event event[num];
868 	int n, i;
869 	uint16_t port_id;
870 	uint8_t queue_id;
871 	void *data;
872 
873 	RTE_LOG(INFO, L3FWD_POWER,
874 		"lcore %u sleeps until interrupt triggers\n",
875 		rte_lcore_id());
876 
877 	n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
878 	for (i = 0; i < n; i++) {
879 		data = event[i].epdata.data;
880 		port_id = ((uintptr_t)data) >> CHAR_BIT;
881 		queue_id = ((uintptr_t)data) &
882 			RTE_LEN2MASK(CHAR_BIT, uint8_t);
883 		rte_spinlock_lock(&(locks[port_id]));
884 		rte_eth_dev_rx_intr_disable(port_id, queue_id);
885 		rte_spinlock_unlock(&(locks[port_id]));
886 		RTE_LOG(INFO, L3FWD_POWER,
887 			"lcore %u is waked up from rx interrupt on"
888 			" port %d queue %d\n",
889 			rte_lcore_id(), port_id, queue_id);
890 	}
891 
892 	return 0;
893 }
894 
895 static void turn_on_intr(struct lcore_conf *qconf)
896 {
897 	int i;
898 	struct lcore_rx_queue *rx_queue;
899 	uint8_t queue_id;
900 	uint16_t port_id;
901 
902 	for (i = 0; i < qconf->n_rx_queue; ++i) {
903 		rx_queue = &(qconf->rx_queue_list[i]);
904 		port_id = rx_queue->port_id;
905 		queue_id = rx_queue->queue_id;
906 
907 		rte_spinlock_lock(&(locks[port_id]));
908 		rte_eth_dev_rx_intr_enable(port_id, queue_id);
909 		rte_spinlock_unlock(&(locks[port_id]));
910 	}
911 }
912 
913 static int event_register(struct lcore_conf *qconf)
914 {
915 	struct lcore_rx_queue *rx_queue;
916 	uint8_t queueid;
917 	uint16_t portid;
918 	uint32_t data;
919 	int ret;
920 	int i;
921 
922 	for (i = 0; i < qconf->n_rx_queue; ++i) {
923 		rx_queue = &(qconf->rx_queue_list[i]);
924 		portid = rx_queue->port_id;
925 		queueid = rx_queue->queue_id;
926 		data = portid << CHAR_BIT | queueid;
927 
928 		ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
929 						RTE_EPOLL_PER_THREAD,
930 						RTE_INTR_EVENT_ADD,
931 						(void *)((uintptr_t)data));
932 		if (ret)
933 			return ret;
934 	}
935 
936 	return 0;
937 }
938 /* main processing loop */
939 static int
940 main_telemetry_loop(__attribute__((unused)) void *dummy)
941 {
942 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
943 	unsigned int lcore_id;
944 	uint64_t prev_tsc, diff_tsc, cur_tsc, prev_tel_tsc;
945 	int i, j, nb_rx;
946 	uint8_t queueid;
947 	uint16_t portid;
948 	struct lcore_conf *qconf;
949 	struct lcore_rx_queue *rx_queue;
950 	uint64_t ep_nep[2] = {0}, fp_nfp[2] = {0};
951 	uint64_t poll_count;
952 	enum busy_rate br;
953 
954 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
955 					US_PER_S * BURST_TX_DRAIN_US;
956 
957 	poll_count = 0;
958 	prev_tsc = 0;
959 	prev_tel_tsc = 0;
960 
961 	lcore_id = rte_lcore_id();
962 	qconf = &lcore_conf[lcore_id];
963 
964 	if (qconf->n_rx_queue == 0) {
965 		RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
966 			lcore_id);
967 		return 0;
968 	}
969 
970 	RTE_LOG(INFO, L3FWD_POWER, "entering main telemetry loop on lcore %u\n",
971 		lcore_id);
972 
973 	for (i = 0; i < qconf->n_rx_queue; i++) {
974 		portid = qconf->rx_queue_list[i].port_id;
975 		queueid = qconf->rx_queue_list[i].queue_id;
976 		RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
977 			"rxqueueid=%hhu\n", lcore_id, portid, queueid);
978 	}
979 
980 	while (!is_done()) {
981 
982 		cur_tsc = rte_rdtsc();
983 		/*
984 		 * TX burst queue drain
985 		 */
986 		diff_tsc = cur_tsc - prev_tsc;
987 		if (unlikely(diff_tsc > drain_tsc)) {
988 			for (i = 0; i < qconf->n_tx_port; ++i) {
989 				portid = qconf->tx_port_id[i];
990 				rte_eth_tx_buffer_flush(portid,
991 						qconf->tx_queue_id[portid],
992 						qconf->tx_buffer[portid]);
993 			}
994 			prev_tsc = cur_tsc;
995 		}
996 
997 		/*
998 		 * Read packet from RX queues
999 		 */
1000 		for (i = 0; i < qconf->n_rx_queue; ++i) {
1001 			rx_queue = &(qconf->rx_queue_list[i]);
1002 			portid = rx_queue->port_id;
1003 			queueid = rx_queue->queue_id;
1004 
1005 			nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1006 								MAX_PKT_BURST);
1007 			ep_nep[nb_rx == 0]++;
1008 			fp_nfp[nb_rx == MAX_PKT_BURST]++;
1009 			poll_count++;
1010 			if (unlikely(nb_rx == 0))
1011 				continue;
1012 
1013 			/* Prefetch first packets */
1014 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1015 				rte_prefetch0(rte_pktmbuf_mtod(
1016 						pkts_burst[j], void *));
1017 			}
1018 
1019 			/* Prefetch and forward already prefetched packets */
1020 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1021 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1022 						j + PREFETCH_OFFSET], void *));
1023 				l3fwd_simple_forward(pkts_burst[j], portid,
1024 								qconf);
1025 			}
1026 
1027 			/* Forward remaining prefetched packets */
1028 			for (; j < nb_rx; j++) {
1029 				l3fwd_simple_forward(pkts_burst[j], portid,
1030 								qconf);
1031 			}
1032 		}
1033 		if (unlikely(poll_count >= DEFAULT_COUNT)) {
1034 			diff_tsc = cur_tsc - prev_tel_tsc;
1035 			if (diff_tsc >= MAX_CYCLES) {
1036 				br = FULL;
1037 			} else if (diff_tsc > MIN_CYCLES &&
1038 					diff_tsc < MAX_CYCLES) {
1039 				br = (diff_tsc * 100) / MAX_CYCLES;
1040 			} else {
1041 				br = ZERO;
1042 			}
1043 			poll_count = 0;
1044 			prev_tel_tsc = cur_tsc;
1045 			/* update stats for telemetry */
1046 			rte_spinlock_lock(&stats[lcore_id].telemetry_lock);
1047 			stats[lcore_id].ep_nep[0] = ep_nep[0];
1048 			stats[lcore_id].ep_nep[1] = ep_nep[1];
1049 			stats[lcore_id].fp_nfp[0] = fp_nfp[0];
1050 			stats[lcore_id].fp_nfp[1] = fp_nfp[1];
1051 			stats[lcore_id].br = br;
1052 			rte_spinlock_unlock(&stats[lcore_id].telemetry_lock);
1053 		}
1054 	}
1055 
1056 	return 0;
1057 }
1058 /* main processing loop */
1059 static int
1060 main_empty_poll_loop(__attribute__((unused)) void *dummy)
1061 {
1062 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1063 	unsigned int lcore_id;
1064 	uint64_t prev_tsc, diff_tsc, cur_tsc;
1065 	int i, j, nb_rx;
1066 	uint8_t queueid;
1067 	uint16_t portid;
1068 	struct lcore_conf *qconf;
1069 	struct lcore_rx_queue *rx_queue;
1070 
1071 	const uint64_t drain_tsc =
1072 		(rte_get_tsc_hz() + US_PER_S - 1) /
1073 		US_PER_S * BURST_TX_DRAIN_US;
1074 
1075 	prev_tsc = 0;
1076 
1077 	lcore_id = rte_lcore_id();
1078 	qconf = &lcore_conf[lcore_id];
1079 
1080 	if (qconf->n_rx_queue == 0) {
1081 		RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
1082 			lcore_id);
1083 		return 0;
1084 	}
1085 
1086 	for (i = 0; i < qconf->n_rx_queue; i++) {
1087 		portid = qconf->rx_queue_list[i].port_id;
1088 		queueid = qconf->rx_queue_list[i].queue_id;
1089 		RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1090 				"rxqueueid=%hhu\n", lcore_id, portid, queueid);
1091 	}
1092 
1093 	while (!is_done()) {
1094 		stats[lcore_id].nb_iteration_looped++;
1095 
1096 		cur_tsc = rte_rdtsc();
1097 		/*
1098 		 * TX burst queue drain
1099 		 */
1100 		diff_tsc = cur_tsc - prev_tsc;
1101 		if (unlikely(diff_tsc > drain_tsc)) {
1102 			for (i = 0; i < qconf->n_tx_port; ++i) {
1103 				portid = qconf->tx_port_id[i];
1104 				rte_eth_tx_buffer_flush(portid,
1105 						qconf->tx_queue_id[portid],
1106 						qconf->tx_buffer[portid]);
1107 			}
1108 			prev_tsc = cur_tsc;
1109 		}
1110 
1111 		/*
1112 		 * Read packet from RX queues
1113 		 */
1114 		for (i = 0; i < qconf->n_rx_queue; ++i) {
1115 			rx_queue = &(qconf->rx_queue_list[i]);
1116 			rx_queue->idle_hint = 0;
1117 			portid = rx_queue->port_id;
1118 			queueid = rx_queue->queue_id;
1119 
1120 			nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1121 					MAX_PKT_BURST);
1122 
1123 			stats[lcore_id].nb_rx_processed += nb_rx;
1124 
1125 			if (nb_rx == 0) {
1126 
1127 				rte_power_empty_poll_stat_update(lcore_id);
1128 
1129 				continue;
1130 			} else {
1131 				rte_power_poll_stat_update(lcore_id, nb_rx);
1132 			}
1133 
1134 
1135 			/* Prefetch first packets */
1136 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1137 				rte_prefetch0(rte_pktmbuf_mtod(
1138 							pkts_burst[j], void *));
1139 			}
1140 
1141 			/* Prefetch and forward already prefetched packets */
1142 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1143 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1144 							j + PREFETCH_OFFSET],
1145 							void *));
1146 				l3fwd_simple_forward(pkts_burst[j], portid,
1147 						qconf);
1148 			}
1149 
1150 			/* Forward remaining prefetched packets */
1151 			for (; j < nb_rx; j++) {
1152 				l3fwd_simple_forward(pkts_burst[j], portid,
1153 						qconf);
1154 			}
1155 
1156 		}
1157 
1158 	}
1159 
1160 	return 0;
1161 }
1162 /* main processing loop */
1163 static int
1164 main_loop(__attribute__((unused)) void *dummy)
1165 {
1166 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1167 	unsigned lcore_id;
1168 	uint64_t prev_tsc, diff_tsc, cur_tsc, tim_res_tsc, hz;
1169 	uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
1170 	int i, j, nb_rx;
1171 	uint8_t queueid;
1172 	uint16_t portid;
1173 	struct lcore_conf *qconf;
1174 	struct lcore_rx_queue *rx_queue;
1175 	enum freq_scale_hint_t lcore_scaleup_hint;
1176 	uint32_t lcore_rx_idle_count = 0;
1177 	uint32_t lcore_idle_hint = 0;
1178 	int intr_en = 0;
1179 
1180 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
1181 
1182 	prev_tsc = 0;
1183 	hz = rte_get_timer_hz();
1184 	tim_res_tsc = hz/TIMER_NUMBER_PER_SECOND;
1185 
1186 	lcore_id = rte_lcore_id();
1187 	qconf = &lcore_conf[lcore_id];
1188 
1189 	if (qconf->n_rx_queue == 0) {
1190 		RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
1191 		return 0;
1192 	}
1193 
1194 	RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
1195 
1196 	for (i = 0; i < qconf->n_rx_queue; i++) {
1197 		portid = qconf->rx_queue_list[i].port_id;
1198 		queueid = qconf->rx_queue_list[i].queue_id;
1199 		RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1200 			"rxqueueid=%hhu\n", lcore_id, portid, queueid);
1201 	}
1202 
1203 	/* add into event wait list */
1204 	if (event_register(qconf) == 0)
1205 		intr_en = 1;
1206 	else
1207 		RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
1208 
1209 	while (1) {
1210 		stats[lcore_id].nb_iteration_looped++;
1211 
1212 		cur_tsc = rte_rdtsc();
1213 		cur_tsc_power = cur_tsc;
1214 
1215 		/*
1216 		 * TX burst queue drain
1217 		 */
1218 		diff_tsc = cur_tsc - prev_tsc;
1219 		if (unlikely(diff_tsc > drain_tsc)) {
1220 			for (i = 0; i < qconf->n_tx_port; ++i) {
1221 				portid = qconf->tx_port_id[i];
1222 				rte_eth_tx_buffer_flush(portid,
1223 						qconf->tx_queue_id[portid],
1224 						qconf->tx_buffer[portid]);
1225 			}
1226 			prev_tsc = cur_tsc;
1227 		}
1228 
1229 		diff_tsc_power = cur_tsc_power - prev_tsc_power;
1230 		if (diff_tsc_power > tim_res_tsc) {
1231 			rte_timer_manage();
1232 			prev_tsc_power = cur_tsc_power;
1233 		}
1234 
1235 start_rx:
1236 		/*
1237 		 * Read packet from RX queues
1238 		 */
1239 		lcore_scaleup_hint = FREQ_CURRENT;
1240 		lcore_rx_idle_count = 0;
1241 		for (i = 0; i < qconf->n_rx_queue; ++i) {
1242 			rx_queue = &(qconf->rx_queue_list[i]);
1243 			rx_queue->idle_hint = 0;
1244 			portid = rx_queue->port_id;
1245 			queueid = rx_queue->queue_id;
1246 
1247 			nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1248 								MAX_PKT_BURST);
1249 
1250 			stats[lcore_id].nb_rx_processed += nb_rx;
1251 			if (unlikely(nb_rx == 0)) {
1252 				/**
1253 				 * no packet received from rx queue, try to
1254 				 * sleep for a while forcing CPU enter deeper
1255 				 * C states.
1256 				 */
1257 				rx_queue->zero_rx_packet_count++;
1258 
1259 				if (rx_queue->zero_rx_packet_count <=
1260 							MIN_ZERO_POLL_COUNT)
1261 					continue;
1262 
1263 				rx_queue->idle_hint = power_idle_heuristic(\
1264 					rx_queue->zero_rx_packet_count);
1265 				lcore_rx_idle_count++;
1266 			} else {
1267 				rx_queue->zero_rx_packet_count = 0;
1268 
1269 				/**
1270 				 * do not scale up frequency immediately as
1271 				 * user to kernel space communication is costly
1272 				 * which might impact packet I/O for received
1273 				 * packets.
1274 				 */
1275 				rx_queue->freq_up_hint =
1276 					power_freq_scaleup_heuristic(lcore_id,
1277 							portid, queueid);
1278 			}
1279 
1280 			/* Prefetch first packets */
1281 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1282 				rte_prefetch0(rte_pktmbuf_mtod(
1283 						pkts_burst[j], void *));
1284 			}
1285 
1286 			/* Prefetch and forward already prefetched packets */
1287 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1288 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1289 						j + PREFETCH_OFFSET], void *));
1290 				l3fwd_simple_forward(pkts_burst[j], portid,
1291 								qconf);
1292 			}
1293 
1294 			/* Forward remaining prefetched packets */
1295 			for (; j < nb_rx; j++) {
1296 				l3fwd_simple_forward(pkts_burst[j], portid,
1297 								qconf);
1298 			}
1299 		}
1300 
1301 		if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
1302 			for (i = 1, lcore_scaleup_hint =
1303 				qconf->rx_queue_list[0].freq_up_hint;
1304 					i < qconf->n_rx_queue; ++i) {
1305 				rx_queue = &(qconf->rx_queue_list[i]);
1306 				if (rx_queue->freq_up_hint >
1307 						lcore_scaleup_hint)
1308 					lcore_scaleup_hint =
1309 						rx_queue->freq_up_hint;
1310 			}
1311 
1312 			if (lcore_scaleup_hint == FREQ_HIGHEST) {
1313 				if (rte_power_freq_max)
1314 					rte_power_freq_max(lcore_id);
1315 			} else if (lcore_scaleup_hint == FREQ_HIGHER) {
1316 				if (rte_power_freq_up)
1317 					rte_power_freq_up(lcore_id);
1318 			}
1319 		} else {
1320 			/**
1321 			 * All Rx queues empty in recent consecutive polls,
1322 			 * sleep in a conservative manner, meaning sleep as
1323 			 * less as possible.
1324 			 */
1325 			for (i = 1, lcore_idle_hint =
1326 				qconf->rx_queue_list[0].idle_hint;
1327 					i < qconf->n_rx_queue; ++i) {
1328 				rx_queue = &(qconf->rx_queue_list[i]);
1329 				if (rx_queue->idle_hint < lcore_idle_hint)
1330 					lcore_idle_hint = rx_queue->idle_hint;
1331 			}
1332 
1333 			if (lcore_idle_hint < SUSPEND_THRESHOLD)
1334 				/**
1335 				 * execute "pause" instruction to avoid context
1336 				 * switch which generally take hundred of
1337 				 * microseconds for short sleep.
1338 				 */
1339 				rte_delay_us(lcore_idle_hint);
1340 			else {
1341 				/* suspend until rx interrupt trigges */
1342 				if (intr_en) {
1343 					turn_on_intr(qconf);
1344 					sleep_until_rx_interrupt(
1345 						qconf->n_rx_queue);
1346 					/**
1347 					 * start receiving packets immediately
1348 					 */
1349 					goto start_rx;
1350 				}
1351 			}
1352 			stats[lcore_id].sleep_time += lcore_idle_hint;
1353 		}
1354 	}
1355 }
1356 
1357 static int
1358 check_lcore_params(void)
1359 {
1360 	uint8_t queue, lcore;
1361 	uint16_t i;
1362 	int socketid;
1363 
1364 	for (i = 0; i < nb_lcore_params; ++i) {
1365 		queue = lcore_params[i].queue_id;
1366 		if (queue >= MAX_RX_QUEUE_PER_PORT) {
1367 			printf("invalid queue number: %hhu\n", queue);
1368 			return -1;
1369 		}
1370 		lcore = lcore_params[i].lcore_id;
1371 		if (!rte_lcore_is_enabled(lcore)) {
1372 			printf("error: lcore %hhu is not enabled in lcore "
1373 							"mask\n", lcore);
1374 			return -1;
1375 		}
1376 		if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1377 							(numa_on == 0)) {
1378 			printf("warning: lcore %hhu is on socket %d with numa "
1379 						"off\n", lcore, socketid);
1380 		}
1381 		if (app_mode == APP_MODE_TELEMETRY && lcore == rte_lcore_id()) {
1382 			printf("cannot enable master core %d in config for telemetry mode\n",
1383 				rte_lcore_id());
1384 			return -1;
1385 		}
1386 	}
1387 	return 0;
1388 }
1389 
1390 static int
1391 check_port_config(void)
1392 {
1393 	unsigned portid;
1394 	uint16_t i;
1395 
1396 	for (i = 0; i < nb_lcore_params; ++i) {
1397 		portid = lcore_params[i].port_id;
1398 		if ((enabled_port_mask & (1 << portid)) == 0) {
1399 			printf("port %u is not enabled in port mask\n",
1400 								portid);
1401 			return -1;
1402 		}
1403 		if (!rte_eth_dev_is_valid_port(portid)) {
1404 			printf("port %u is not present on the board\n",
1405 								portid);
1406 			return -1;
1407 		}
1408 	}
1409 	return 0;
1410 }
1411 
1412 static uint8_t
1413 get_port_n_rx_queues(const uint16_t port)
1414 {
1415 	int queue = -1;
1416 	uint16_t i;
1417 
1418 	for (i = 0; i < nb_lcore_params; ++i) {
1419 		if (lcore_params[i].port_id == port &&
1420 				lcore_params[i].queue_id > queue)
1421 			queue = lcore_params[i].queue_id;
1422 	}
1423 	return (uint8_t)(++queue);
1424 }
1425 
1426 static int
1427 init_lcore_rx_queues(void)
1428 {
1429 	uint16_t i, nb_rx_queue;
1430 	uint8_t lcore;
1431 
1432 	for (i = 0; i < nb_lcore_params; ++i) {
1433 		lcore = lcore_params[i].lcore_id;
1434 		nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1435 		if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1436 			printf("error: too many queues (%u) for lcore: %u\n",
1437 				(unsigned)nb_rx_queue + 1, (unsigned)lcore);
1438 			return -1;
1439 		} else {
1440 			lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1441 				lcore_params[i].port_id;
1442 			lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1443 				lcore_params[i].queue_id;
1444 			lcore_conf[lcore].n_rx_queue++;
1445 		}
1446 	}
1447 	return 0;
1448 }
1449 
1450 /* display usage */
1451 static void
1452 print_usage(const char *prgname)
1453 {
1454 	printf ("%s [EAL options] -- -p PORTMASK -P"
1455 		"  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1456 		"  [--high-perf-cores CORELIST"
1457 		"  [--perf-config (port,queue,hi_perf,lcore_index)[,(port,queue,hi_perf,lcore_index]]"
1458 		"  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1459 		"  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1460 		"  -P : enable promiscuous mode\n"
1461 		"  --config (port,queue,lcore): rx queues configuration\n"
1462 		"  --high-perf-cores CORELIST: list of high performance cores\n"
1463 		"  --perf-config: similar as config, cores specified as indices"
1464 		" for bins containing high or regular performance cores\n"
1465 		"  --no-numa: optional, disable numa awareness\n"
1466 		"  --enable-jumbo: enable jumbo frame"
1467 		" which max packet len is PKTLEN in decimal (64-9600)\n"
1468 		"  --parse-ptype: parse packet type by software\n"
1469 		"  --empty-poll: enable empty poll detection"
1470 		" follow (training_flag, high_threshold, med_threshold)\n"
1471 		" --telemetry: enable telemetry mode, to update"
1472 		" empty polls, full polls, and core busyness to telemetry\n",
1473 		prgname);
1474 }
1475 
1476 static int parse_max_pkt_len(const char *pktlen)
1477 {
1478 	char *end = NULL;
1479 	unsigned long len;
1480 
1481 	/* parse decimal string */
1482 	len = strtoul(pktlen, &end, 10);
1483 	if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1484 		return -1;
1485 
1486 	if (len == 0)
1487 		return -1;
1488 
1489 	return len;
1490 }
1491 
1492 static int
1493 parse_portmask(const char *portmask)
1494 {
1495 	char *end = NULL;
1496 	unsigned long pm;
1497 
1498 	/* parse hexadecimal string */
1499 	pm = strtoul(portmask, &end, 16);
1500 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1501 		return -1;
1502 
1503 	if (pm == 0)
1504 		return -1;
1505 
1506 	return pm;
1507 }
1508 
1509 static int
1510 parse_config(const char *q_arg)
1511 {
1512 	char s[256];
1513 	const char *p, *p0 = q_arg;
1514 	char *end;
1515 	enum fieldnames {
1516 		FLD_PORT = 0,
1517 		FLD_QUEUE,
1518 		FLD_LCORE,
1519 		_NUM_FLD
1520 	};
1521 	unsigned long int_fld[_NUM_FLD];
1522 	char *str_fld[_NUM_FLD];
1523 	int i;
1524 	unsigned size;
1525 
1526 	nb_lcore_params = 0;
1527 
1528 	while ((p = strchr(p0,'(')) != NULL) {
1529 		++p;
1530 		if((p0 = strchr(p,')')) == NULL)
1531 			return -1;
1532 
1533 		size = p0 - p;
1534 		if(size >= sizeof(s))
1535 			return -1;
1536 
1537 		snprintf(s, sizeof(s), "%.*s", size, p);
1538 		if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1539 								_NUM_FLD)
1540 			return -1;
1541 		for (i = 0; i < _NUM_FLD; i++){
1542 			errno = 0;
1543 			int_fld[i] = strtoul(str_fld[i], &end, 0);
1544 			if (errno != 0 || end == str_fld[i] || int_fld[i] >
1545 									255)
1546 				return -1;
1547 		}
1548 		if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1549 			printf("exceeded max number of lcore params: %hu\n",
1550 				nb_lcore_params);
1551 			return -1;
1552 		}
1553 		lcore_params_array[nb_lcore_params].port_id =
1554 				(uint8_t)int_fld[FLD_PORT];
1555 		lcore_params_array[nb_lcore_params].queue_id =
1556 				(uint8_t)int_fld[FLD_QUEUE];
1557 		lcore_params_array[nb_lcore_params].lcore_id =
1558 				(uint8_t)int_fld[FLD_LCORE];
1559 		++nb_lcore_params;
1560 	}
1561 	lcore_params = lcore_params_array;
1562 
1563 	return 0;
1564 }
1565 static int
1566 parse_ep_config(const char *q_arg)
1567 {
1568 	char s[256];
1569 	const char *p = q_arg;
1570 	char *end;
1571 	int  num_arg;
1572 
1573 	char *str_fld[3];
1574 
1575 	int training_flag;
1576 	int med_edpi;
1577 	int hgh_edpi;
1578 
1579 	ep_med_edpi = EMPTY_POLL_MED_THRESHOLD;
1580 	ep_hgh_edpi = EMPTY_POLL_MED_THRESHOLD;
1581 
1582 	strlcpy(s, p, sizeof(s));
1583 
1584 	num_arg = rte_strsplit(s, sizeof(s), str_fld, 3, ',');
1585 
1586 	empty_poll_train = false;
1587 
1588 	if (num_arg == 0)
1589 		return 0;
1590 
1591 	if (num_arg == 3) {
1592 
1593 		training_flag = strtoul(str_fld[0], &end, 0);
1594 		med_edpi = strtoul(str_fld[1], &end, 0);
1595 		hgh_edpi = strtoul(str_fld[2], &end, 0);
1596 
1597 		if (training_flag == 1)
1598 			empty_poll_train = true;
1599 
1600 		if (med_edpi > 0)
1601 			ep_med_edpi = med_edpi;
1602 
1603 		if (med_edpi > 0)
1604 			ep_hgh_edpi = hgh_edpi;
1605 
1606 	} else {
1607 
1608 		return -1;
1609 	}
1610 
1611 	return 0;
1612 
1613 }
1614 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1615 #define CMD_LINE_OPT_TELEMETRY "telemetry"
1616 
1617 /* Parse the argument given in the command line of the application */
1618 static int
1619 parse_args(int argc, char **argv)
1620 {
1621 	int opt, ret;
1622 	char **argvopt;
1623 	int option_index;
1624 	uint32_t limit;
1625 	char *prgname = argv[0];
1626 	static struct option lgopts[] = {
1627 		{"config", 1, 0, 0},
1628 		{"perf-config", 1, 0, 0},
1629 		{"high-perf-cores", 1, 0, 0},
1630 		{"no-numa", 0, 0, 0},
1631 		{"enable-jumbo", 0, 0, 0},
1632 		{"empty-poll", 1, 0, 0},
1633 		{CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
1634 		{CMD_LINE_OPT_TELEMETRY, 0, 0, 0},
1635 		{NULL, 0, 0, 0}
1636 	};
1637 
1638 	argvopt = argv;
1639 
1640 	while ((opt = getopt_long(argc, argvopt, "p:l:m:h:P",
1641 				lgopts, &option_index)) != EOF) {
1642 
1643 		switch (opt) {
1644 		/* portmask */
1645 		case 'p':
1646 			enabled_port_mask = parse_portmask(optarg);
1647 			if (enabled_port_mask == 0) {
1648 				printf("invalid portmask\n");
1649 				print_usage(prgname);
1650 				return -1;
1651 			}
1652 			break;
1653 		case 'P':
1654 			printf("Promiscuous mode selected\n");
1655 			promiscuous_on = 1;
1656 			break;
1657 		case 'l':
1658 			limit = parse_max_pkt_len(optarg);
1659 			freq_tlb[LOW] = limit;
1660 			break;
1661 		case 'm':
1662 			limit = parse_max_pkt_len(optarg);
1663 			freq_tlb[MED] = limit;
1664 			break;
1665 		case 'h':
1666 			limit = parse_max_pkt_len(optarg);
1667 			freq_tlb[HGH] = limit;
1668 			break;
1669 		/* long options */
1670 		case 0:
1671 			if (!strncmp(lgopts[option_index].name, "config", 6)) {
1672 				ret = parse_config(optarg);
1673 				if (ret) {
1674 					printf("invalid config\n");
1675 					print_usage(prgname);
1676 					return -1;
1677 				}
1678 			}
1679 
1680 			if (!strncmp(lgopts[option_index].name,
1681 					"perf-config", 11)) {
1682 				ret = parse_perf_config(optarg);
1683 				if (ret) {
1684 					printf("invalid perf-config\n");
1685 					print_usage(prgname);
1686 					return -1;
1687 				}
1688 			}
1689 
1690 			if (!strncmp(lgopts[option_index].name,
1691 					"high-perf-cores", 15)) {
1692 				ret = parse_perf_core_list(optarg);
1693 				if (ret) {
1694 					printf("invalid high-perf-cores\n");
1695 					print_usage(prgname);
1696 					return -1;
1697 				}
1698 			}
1699 
1700 			if (!strncmp(lgopts[option_index].name,
1701 						"no-numa", 7)) {
1702 				printf("numa is disabled \n");
1703 				numa_on = 0;
1704 			}
1705 
1706 			if (!strncmp(lgopts[option_index].name,
1707 						"empty-poll", 10)) {
1708 				if (app_mode == APP_MODE_TELEMETRY) {
1709 					printf(" empty-poll cannot be enabled as telemetry mode is enabled\n");
1710 					return -1;
1711 				}
1712 				app_mode = APP_MODE_EMPTY_POLL;
1713 				ret = parse_ep_config(optarg);
1714 
1715 				if (ret) {
1716 					printf("invalid empty poll config\n");
1717 					print_usage(prgname);
1718 					return -1;
1719 				}
1720 				printf("empty-poll is enabled\n");
1721 			}
1722 
1723 			if (!strncmp(lgopts[option_index].name,
1724 					CMD_LINE_OPT_TELEMETRY,
1725 					sizeof(CMD_LINE_OPT_TELEMETRY))) {
1726 				if (app_mode == APP_MODE_EMPTY_POLL) {
1727 					printf("telemetry mode cannot be enabled as empty poll mode is enabled\n");
1728 					return -1;
1729 				}
1730 				app_mode = APP_MODE_TELEMETRY;
1731 				printf("telemetry mode is enabled\n");
1732 			}
1733 
1734 			if (!strncmp(lgopts[option_index].name,
1735 					"enable-jumbo", 12)) {
1736 				struct option lenopts =
1737 					{"max-pkt-len", required_argument, \
1738 									0, 0};
1739 
1740 				printf("jumbo frame is enabled \n");
1741 				port_conf.rxmode.offloads |=
1742 						DEV_RX_OFFLOAD_JUMBO_FRAME;
1743 				port_conf.txmode.offloads |=
1744 						DEV_TX_OFFLOAD_MULTI_SEGS;
1745 
1746 				/**
1747 				 * if no max-pkt-len set, use the default value
1748 				 * RTE_ETHER_MAX_LEN
1749 				 */
1750 				if (0 == getopt_long(argc, argvopt, "",
1751 						&lenopts, &option_index)) {
1752 					ret = parse_max_pkt_len(optarg);
1753 					if ((ret < 64) ||
1754 						(ret > MAX_JUMBO_PKT_LEN)){
1755 						printf("invalid packet "
1756 								"length\n");
1757 						print_usage(prgname);
1758 						return -1;
1759 					}
1760 					port_conf.rxmode.max_rx_pkt_len = ret;
1761 				}
1762 				printf("set jumbo frame "
1763 					"max packet length to %u\n",
1764 				(unsigned int)port_conf.rxmode.max_rx_pkt_len);
1765 			}
1766 
1767 			if (!strncmp(lgopts[option_index].name,
1768 				     CMD_LINE_OPT_PARSE_PTYPE,
1769 				     sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
1770 				printf("soft parse-ptype is enabled\n");
1771 				parse_ptype = 1;
1772 			}
1773 
1774 			break;
1775 
1776 		default:
1777 			print_usage(prgname);
1778 			return -1;
1779 		}
1780 	}
1781 
1782 	if (optind >= 0)
1783 		argv[optind-1] = prgname;
1784 
1785 	ret = optind-1;
1786 	optind = 1; /* reset getopt lib */
1787 	return ret;
1788 }
1789 
1790 static void
1791 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1792 {
1793 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1794 	rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
1795 	printf("%s%s", name, buf);
1796 }
1797 
1798 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1799 static void
1800 setup_hash(int socketid)
1801 {
1802 	struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1803 		.name = NULL,
1804 		.entries = L3FWD_HASH_ENTRIES,
1805 		.key_len = sizeof(struct ipv4_5tuple),
1806 		.hash_func = DEFAULT_HASH_FUNC,
1807 		.hash_func_init_val = 0,
1808 	};
1809 
1810 	struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1811 		.name = NULL,
1812 		.entries = L3FWD_HASH_ENTRIES,
1813 		.key_len = sizeof(struct ipv6_5tuple),
1814 		.hash_func = DEFAULT_HASH_FUNC,
1815 		.hash_func_init_val = 0,
1816 	};
1817 
1818 	unsigned i;
1819 	int ret;
1820 	char s[64];
1821 
1822 	/* create ipv4 hash */
1823 	snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1824 	ipv4_l3fwd_hash_params.name = s;
1825 	ipv4_l3fwd_hash_params.socket_id = socketid;
1826 	ipv4_l3fwd_lookup_struct[socketid] =
1827 		rte_hash_create(&ipv4_l3fwd_hash_params);
1828 	if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1829 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1830 				"socket %d\n", socketid);
1831 
1832 	/* create ipv6 hash */
1833 	snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1834 	ipv6_l3fwd_hash_params.name = s;
1835 	ipv6_l3fwd_hash_params.socket_id = socketid;
1836 	ipv6_l3fwd_lookup_struct[socketid] =
1837 		rte_hash_create(&ipv6_l3fwd_hash_params);
1838 	if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1839 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1840 				"socket %d\n", socketid);
1841 
1842 
1843 	/* populate the ipv4 hash */
1844 	for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1845 		ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1846 				(void *) &ipv4_l3fwd_route_array[i].key);
1847 		if (ret < 0) {
1848 			rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1849 				"l3fwd hash on socket %d\n", i, socketid);
1850 		}
1851 		ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1852 		printf("Hash: Adding key\n");
1853 		print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1854 	}
1855 
1856 	/* populate the ipv6 hash */
1857 	for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1858 		ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1859 				(void *) &ipv6_l3fwd_route_array[i].key);
1860 		if (ret < 0) {
1861 			rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1862 				"l3fwd hash on socket %d\n", i, socketid);
1863 		}
1864 		ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1865 		printf("Hash: Adding key\n");
1866 		print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1867 	}
1868 }
1869 #endif
1870 
1871 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1872 static void
1873 setup_lpm(int socketid)
1874 {
1875 	unsigned i;
1876 	int ret;
1877 	char s[64];
1878 
1879 	/* create the LPM table */
1880 	struct rte_lpm_config lpm_ipv4_config;
1881 
1882 	lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
1883 	lpm_ipv4_config.number_tbl8s = 256;
1884 	lpm_ipv4_config.flags = 0;
1885 
1886 	snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1887 	ipv4_l3fwd_lookup_struct[socketid] =
1888 			rte_lpm_create(s, socketid, &lpm_ipv4_config);
1889 	if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1890 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1891 				" on socket %d\n", socketid);
1892 
1893 	/* populate the LPM table */
1894 	for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1895 		ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1896 			ipv4_l3fwd_route_array[i].ip,
1897 			ipv4_l3fwd_route_array[i].depth,
1898 			ipv4_l3fwd_route_array[i].if_out);
1899 
1900 		if (ret < 0) {
1901 			rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1902 				"l3fwd LPM table on socket %d\n",
1903 				i, socketid);
1904 		}
1905 
1906 		printf("LPM: Adding route 0x%08x / %d (%d)\n",
1907 			(unsigned)ipv4_l3fwd_route_array[i].ip,
1908 			ipv4_l3fwd_route_array[i].depth,
1909 			ipv4_l3fwd_route_array[i].if_out);
1910 	}
1911 }
1912 #endif
1913 
1914 static int
1915 init_mem(unsigned nb_mbuf)
1916 {
1917 	struct lcore_conf *qconf;
1918 	int socketid;
1919 	unsigned lcore_id;
1920 	char s[64];
1921 
1922 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1923 		if (rte_lcore_is_enabled(lcore_id) == 0)
1924 			continue;
1925 
1926 		if (numa_on)
1927 			socketid = rte_lcore_to_socket_id(lcore_id);
1928 		else
1929 			socketid = 0;
1930 
1931 		if (socketid >= NB_SOCKETS) {
1932 			rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1933 					"out of range %d\n", socketid,
1934 						lcore_id, NB_SOCKETS);
1935 		}
1936 		if (pktmbuf_pool[socketid] == NULL) {
1937 			snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1938 			pktmbuf_pool[socketid] =
1939 				rte_pktmbuf_pool_create(s, nb_mbuf,
1940 					MEMPOOL_CACHE_SIZE, 0,
1941 					RTE_MBUF_DEFAULT_BUF_SIZE,
1942 					socketid);
1943 			if (pktmbuf_pool[socketid] == NULL)
1944 				rte_exit(EXIT_FAILURE,
1945 					"Cannot init mbuf pool on socket %d\n",
1946 								socketid);
1947 			else
1948 				printf("Allocated mbuf pool on socket %d\n",
1949 								socketid);
1950 
1951 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1952 			setup_lpm(socketid);
1953 #else
1954 			setup_hash(socketid);
1955 #endif
1956 		}
1957 		qconf = &lcore_conf[lcore_id];
1958 		qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1959 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1960 		qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1961 #endif
1962 	}
1963 	return 0;
1964 }
1965 
1966 /* Check the link status of all ports in up to 9s, and print them finally */
1967 static void
1968 check_all_ports_link_status(uint32_t port_mask)
1969 {
1970 #define CHECK_INTERVAL 100 /* 100ms */
1971 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1972 	uint8_t count, all_ports_up, print_flag = 0;
1973 	uint16_t portid;
1974 	struct rte_eth_link link;
1975 	int ret;
1976 
1977 	printf("\nChecking link status");
1978 	fflush(stdout);
1979 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
1980 		all_ports_up = 1;
1981 		RTE_ETH_FOREACH_DEV(portid) {
1982 			if ((port_mask & (1 << portid)) == 0)
1983 				continue;
1984 			memset(&link, 0, sizeof(link));
1985 			ret = rte_eth_link_get_nowait(portid, &link);
1986 			if (ret < 0) {
1987 				all_ports_up = 0;
1988 				if (print_flag == 1)
1989 					printf("Port %u link get failed: %s\n",
1990 						portid, rte_strerror(-ret));
1991 				continue;
1992 			}
1993 			/* print link status if flag set */
1994 			if (print_flag == 1) {
1995 				if (link.link_status)
1996 					printf("Port %d Link Up - speed %u "
1997 						"Mbps - %s\n", (uint8_t)portid,
1998 						(unsigned)link.link_speed,
1999 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
2000 					("full-duplex") : ("half-duplex\n"));
2001 				else
2002 					printf("Port %d Link Down\n",
2003 						(uint8_t)portid);
2004 				continue;
2005 			}
2006 			/* clear all_ports_up flag if any link down */
2007 			if (link.link_status == ETH_LINK_DOWN) {
2008 				all_ports_up = 0;
2009 				break;
2010 			}
2011 		}
2012 		/* after finally printing all link status, get out */
2013 		if (print_flag == 1)
2014 			break;
2015 
2016 		if (all_ports_up == 0) {
2017 			printf(".");
2018 			fflush(stdout);
2019 			rte_delay_ms(CHECK_INTERVAL);
2020 		}
2021 
2022 		/* set the print_flag if all ports up or timeout */
2023 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
2024 			print_flag = 1;
2025 			printf("done\n");
2026 		}
2027 	}
2028 }
2029 
2030 static int check_ptype(uint16_t portid)
2031 {
2032 	int i, ret;
2033 	int ptype_l3_ipv4 = 0;
2034 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2035 	int ptype_l3_ipv6 = 0;
2036 #endif
2037 	uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
2038 
2039 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
2040 	if (ret <= 0)
2041 		return 0;
2042 
2043 	uint32_t ptypes[ret];
2044 
2045 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
2046 	for (i = 0; i < ret; ++i) {
2047 		if (ptypes[i] & RTE_PTYPE_L3_IPV4)
2048 			ptype_l3_ipv4 = 1;
2049 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2050 		if (ptypes[i] & RTE_PTYPE_L3_IPV6)
2051 			ptype_l3_ipv6 = 1;
2052 #endif
2053 	}
2054 
2055 	if (ptype_l3_ipv4 == 0)
2056 		printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
2057 
2058 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2059 	if (ptype_l3_ipv6 == 0)
2060 		printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
2061 #endif
2062 
2063 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2064 	if (ptype_l3_ipv4)
2065 #else /* APP_LOOKUP_EXACT_MATCH */
2066 	if (ptype_l3_ipv4 && ptype_l3_ipv6)
2067 #endif
2068 		return 1;
2069 
2070 	return 0;
2071 
2072 }
2073 
2074 static int
2075 init_power_library(void)
2076 {
2077 	int ret = 0, lcore_id;
2078 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2079 		if (rte_lcore_is_enabled(lcore_id)) {
2080 			/* init power management library */
2081 			ret = rte_power_init(lcore_id);
2082 			if (ret)
2083 				RTE_LOG(ERR, POWER,
2084 				"Library initialization failed on core %u\n",
2085 				lcore_id);
2086 		}
2087 	}
2088 	return ret;
2089 }
2090 static void
2091 update_telemetry(__attribute__((unused)) struct rte_timer *tim,
2092 		__attribute__((unused)) void *arg)
2093 {
2094 	unsigned int lcore_id = rte_lcore_id();
2095 	struct lcore_conf *qconf;
2096 	uint64_t app_eps = 0, app_fps = 0, app_br = 0;
2097 	uint64_t values[3] = {0};
2098 	int ret;
2099 	uint64_t count = 0;
2100 
2101 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2102 		qconf = &lcore_conf[lcore_id];
2103 		if (qconf->n_rx_queue == 0)
2104 			continue;
2105 		count++;
2106 		rte_spinlock_lock(&stats[lcore_id].telemetry_lock);
2107 		app_eps += stats[lcore_id].ep_nep[1];
2108 		app_fps += stats[lcore_id].fp_nfp[1];
2109 		app_br += stats[lcore_id].br;
2110 		rte_spinlock_unlock(&stats[lcore_id].telemetry_lock);
2111 	}
2112 
2113 	if (count > 0) {
2114 		values[0] = app_eps/count;
2115 		values[1] = app_fps/count;
2116 		values[2] = app_br/count;
2117 	} else {
2118 		values[0] = 0;
2119 		values[1] = 0;
2120 		values[2] = 0;
2121 	}
2122 
2123 	ret = rte_metrics_update_values(RTE_METRICS_GLOBAL, telstats_index,
2124 					values, RTE_DIM(values));
2125 	if (ret < 0)
2126 		RTE_LOG(WARNING, POWER, "failed to update metrcis\n");
2127 }
2128 static void
2129 telemetry_setup_timer(void)
2130 {
2131 	int lcore_id = rte_lcore_id();
2132 	uint64_t hz = rte_get_timer_hz();
2133 	uint64_t ticks;
2134 
2135 	ticks = hz / TELEMETRY_INTERVALS_PER_SEC;
2136 	rte_timer_reset_sync(&telemetry_timer,
2137 			ticks,
2138 			PERIODICAL,
2139 			lcore_id,
2140 			update_telemetry,
2141 			NULL);
2142 }
2143 static void
2144 empty_poll_setup_timer(void)
2145 {
2146 	int lcore_id = rte_lcore_id();
2147 	uint64_t hz = rte_get_timer_hz();
2148 
2149 	struct  ep_params *ep_ptr = ep_params;
2150 
2151 	ep_ptr->interval_ticks = hz / INTERVALS_PER_SECOND;
2152 
2153 	rte_timer_reset_sync(&ep_ptr->timer0,
2154 			ep_ptr->interval_ticks,
2155 			PERIODICAL,
2156 			lcore_id,
2157 			rte_empty_poll_detection,
2158 			(void *)ep_ptr);
2159 
2160 }
2161 static int
2162 launch_timer(unsigned int lcore_id)
2163 {
2164 	int64_t prev_tsc = 0, cur_tsc, diff_tsc, cycles_10ms;
2165 
2166 	RTE_SET_USED(lcore_id);
2167 
2168 
2169 	if (rte_get_master_lcore() != lcore_id) {
2170 		rte_panic("timer on lcore:%d which is not master core:%d\n",
2171 				lcore_id,
2172 				rte_get_master_lcore());
2173 	}
2174 
2175 	RTE_LOG(INFO, POWER, "Bring up the Timer\n");
2176 
2177 	if (app_mode == APP_MODE_EMPTY_POLL)
2178 		empty_poll_setup_timer();
2179 	else
2180 		telemetry_setup_timer();
2181 
2182 	cycles_10ms = rte_get_timer_hz() / 100;
2183 
2184 	while (!is_done()) {
2185 		cur_tsc = rte_rdtsc();
2186 		diff_tsc = cur_tsc - prev_tsc;
2187 		if (diff_tsc > cycles_10ms) {
2188 			rte_timer_manage();
2189 			prev_tsc = cur_tsc;
2190 			cycles_10ms = rte_get_timer_hz() / 100;
2191 		}
2192 	}
2193 
2194 	RTE_LOG(INFO, POWER, "Timer_subsystem is done\n");
2195 
2196 	return 0;
2197 }
2198 
2199 
2200 int
2201 main(int argc, char **argv)
2202 {
2203 	struct lcore_conf *qconf;
2204 	struct rte_eth_dev_info dev_info;
2205 	struct rte_eth_txconf *txconf;
2206 	int ret;
2207 	uint16_t nb_ports;
2208 	uint16_t queueid;
2209 	unsigned lcore_id;
2210 	uint64_t hz;
2211 	uint32_t n_tx_queue, nb_lcores;
2212 	uint32_t dev_rxq_num, dev_txq_num;
2213 	uint8_t nb_rx_queue, queue, socketid;
2214 	uint16_t portid;
2215 	uint8_t num_telstats = RTE_DIM(telstats_strings);
2216 	const char *ptr_strings[num_telstats];
2217 
2218 	/* catch SIGINT and restore cpufreq governor to ondemand */
2219 	signal(SIGINT, signal_exit_now);
2220 
2221 	/* init EAL */
2222 	ret = rte_eal_init(argc, argv);
2223 	if (ret < 0)
2224 		rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2225 	argc -= ret;
2226 	argv += ret;
2227 
2228 	/* init RTE timer library to be used late */
2229 	rte_timer_subsystem_init();
2230 
2231 	/* parse application arguments (after the EAL ones) */
2232 	ret = parse_args(argc, argv);
2233 	if (ret < 0)
2234 		rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
2235 
2236 	if (init_power_library())
2237 		RTE_LOG(ERR, L3FWD_POWER, "init_power_library failed\n");
2238 
2239 	if (update_lcore_params() < 0)
2240 		rte_exit(EXIT_FAILURE, "update_lcore_params failed\n");
2241 
2242 	if (check_lcore_params() < 0)
2243 		rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
2244 
2245 	ret = init_lcore_rx_queues();
2246 	if (ret < 0)
2247 		rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2248 
2249 	nb_ports = rte_eth_dev_count_avail();
2250 
2251 	if (check_port_config() < 0)
2252 		rte_exit(EXIT_FAILURE, "check_port_config failed\n");
2253 
2254 	nb_lcores = rte_lcore_count();
2255 
2256 	/* initialize all ports */
2257 	RTE_ETH_FOREACH_DEV(portid) {
2258 		struct rte_eth_conf local_port_conf = port_conf;
2259 
2260 		/* skip ports that are not enabled */
2261 		if ((enabled_port_mask & (1 << portid)) == 0) {
2262 			printf("\nSkipping disabled port %d\n", portid);
2263 			continue;
2264 		}
2265 
2266 		/* init port */
2267 		printf("Initializing port %d ... ", portid );
2268 		fflush(stdout);
2269 
2270 		ret = rte_eth_dev_info_get(portid, &dev_info);
2271 		if (ret != 0)
2272 			rte_exit(EXIT_FAILURE,
2273 				"Error during getting device (port %u) info: %s\n",
2274 				portid, strerror(-ret));
2275 
2276 		dev_rxq_num = dev_info.max_rx_queues;
2277 		dev_txq_num = dev_info.max_tx_queues;
2278 
2279 		nb_rx_queue = get_port_n_rx_queues(portid);
2280 		if (nb_rx_queue > dev_rxq_num)
2281 			rte_exit(EXIT_FAILURE,
2282 				"Cannot configure not existed rxq: "
2283 				"port=%d\n", portid);
2284 
2285 		n_tx_queue = nb_lcores;
2286 		if (n_tx_queue > dev_txq_num)
2287 			n_tx_queue = dev_txq_num;
2288 		printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
2289 			nb_rx_queue, (unsigned)n_tx_queue );
2290 		/* If number of Rx queue is 0, no need to enable Rx interrupt */
2291 		if (nb_rx_queue == 0)
2292 			local_port_conf.intr_conf.rxq = 0;
2293 
2294 		ret = rte_eth_dev_info_get(portid, &dev_info);
2295 		if (ret != 0)
2296 			rte_exit(EXIT_FAILURE,
2297 				"Error during getting device (port %u) info: %s\n",
2298 				portid, strerror(-ret));
2299 
2300 		if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
2301 			local_port_conf.txmode.offloads |=
2302 				DEV_TX_OFFLOAD_MBUF_FAST_FREE;
2303 
2304 		local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2305 			dev_info.flow_type_rss_offloads;
2306 		if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2307 				port_conf.rx_adv_conf.rss_conf.rss_hf) {
2308 			printf("Port %u modified RSS hash function based on hardware support,"
2309 				"requested:%#"PRIx64" configured:%#"PRIx64"\n",
2310 				portid,
2311 				port_conf.rx_adv_conf.rss_conf.rss_hf,
2312 				local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2313 		}
2314 
2315 		ret = rte_eth_dev_configure(portid, nb_rx_queue,
2316 					(uint16_t)n_tx_queue, &local_port_conf);
2317 		if (ret < 0)
2318 			rte_exit(EXIT_FAILURE, "Cannot configure device: "
2319 					"err=%d, port=%d\n", ret, portid);
2320 
2321 		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
2322 						       &nb_txd);
2323 		if (ret < 0)
2324 			rte_exit(EXIT_FAILURE,
2325 				 "Cannot adjust number of descriptors: err=%d, port=%d\n",
2326 				 ret, portid);
2327 
2328 		ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
2329 		if (ret < 0)
2330 			rte_exit(EXIT_FAILURE,
2331 				 "Cannot get MAC address: err=%d, port=%d\n",
2332 				 ret, portid);
2333 
2334 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
2335 		printf(", ");
2336 
2337 		/* init memory */
2338 		ret = init_mem(NB_MBUF);
2339 		if (ret < 0)
2340 			rte_exit(EXIT_FAILURE, "init_mem failed\n");
2341 
2342 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2343 			if (rte_lcore_is_enabled(lcore_id) == 0)
2344 				continue;
2345 
2346 			/* Initialize TX buffers */
2347 			qconf = &lcore_conf[lcore_id];
2348 			qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
2349 				RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
2350 				rte_eth_dev_socket_id(portid));
2351 			if (qconf->tx_buffer[portid] == NULL)
2352 				rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
2353 						 portid);
2354 
2355 			rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
2356 		}
2357 
2358 		/* init one TX queue per couple (lcore,port) */
2359 		queueid = 0;
2360 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2361 			if (rte_lcore_is_enabled(lcore_id) == 0)
2362 				continue;
2363 
2364 			if (queueid >= dev_txq_num)
2365 				continue;
2366 
2367 			if (numa_on)
2368 				socketid = \
2369 				(uint8_t)rte_lcore_to_socket_id(lcore_id);
2370 			else
2371 				socketid = 0;
2372 
2373 			printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
2374 			fflush(stdout);
2375 
2376 			txconf = &dev_info.default_txconf;
2377 			txconf->offloads = local_port_conf.txmode.offloads;
2378 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
2379 						     socketid, txconf);
2380 			if (ret < 0)
2381 				rte_exit(EXIT_FAILURE,
2382 					"rte_eth_tx_queue_setup: err=%d, "
2383 						"port=%d\n", ret, portid);
2384 
2385 			qconf = &lcore_conf[lcore_id];
2386 			qconf->tx_queue_id[portid] = queueid;
2387 			queueid++;
2388 
2389 			qconf->tx_port_id[qconf->n_tx_port] = portid;
2390 			qconf->n_tx_port++;
2391 		}
2392 		printf("\n");
2393 	}
2394 
2395 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2396 		if (rte_lcore_is_enabled(lcore_id) == 0)
2397 			continue;
2398 
2399 		if (app_mode == APP_MODE_LEGACY) {
2400 			/* init timer structures for each enabled lcore */
2401 			rte_timer_init(&power_timers[lcore_id]);
2402 			hz = rte_get_timer_hz();
2403 			rte_timer_reset(&power_timers[lcore_id],
2404 					hz/TIMER_NUMBER_PER_SECOND,
2405 					SINGLE, lcore_id,
2406 					power_timer_cb, NULL);
2407 		}
2408 		qconf = &lcore_conf[lcore_id];
2409 		printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
2410 		fflush(stdout);
2411 		/* init RX queues */
2412 		for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
2413 			struct rte_eth_rxconf rxq_conf;
2414 
2415 			portid = qconf->rx_queue_list[queue].port_id;
2416 			queueid = qconf->rx_queue_list[queue].queue_id;
2417 
2418 			if (numa_on)
2419 				socketid = \
2420 				(uint8_t)rte_lcore_to_socket_id(lcore_id);
2421 			else
2422 				socketid = 0;
2423 
2424 			printf("rxq=%d,%d,%d ", portid, queueid, socketid);
2425 			fflush(stdout);
2426 
2427 			ret = rte_eth_dev_info_get(portid, &dev_info);
2428 			if (ret != 0)
2429 				rte_exit(EXIT_FAILURE,
2430 					"Error during getting device (port %u) info: %s\n",
2431 					portid, strerror(-ret));
2432 
2433 			rxq_conf = dev_info.default_rxconf;
2434 			rxq_conf.offloads = port_conf.rxmode.offloads;
2435 			ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
2436 				socketid, &rxq_conf,
2437 				pktmbuf_pool[socketid]);
2438 			if (ret < 0)
2439 				rte_exit(EXIT_FAILURE,
2440 					"rte_eth_rx_queue_setup: err=%d, "
2441 						"port=%d\n", ret, portid);
2442 
2443 			if (parse_ptype) {
2444 				if (add_cb_parse_ptype(portid, queueid) < 0)
2445 					rte_exit(EXIT_FAILURE,
2446 						 "Fail to add ptype cb\n");
2447 			} else if (!check_ptype(portid))
2448 				rte_exit(EXIT_FAILURE,
2449 					 "PMD can not provide needed ptypes\n");
2450 		}
2451 	}
2452 
2453 	printf("\n");
2454 
2455 	/* start ports */
2456 	RTE_ETH_FOREACH_DEV(portid) {
2457 		if ((enabled_port_mask & (1 << portid)) == 0) {
2458 			continue;
2459 		}
2460 		/* Start device */
2461 		ret = rte_eth_dev_start(portid);
2462 		if (ret < 0)
2463 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
2464 						"port=%d\n", ret, portid);
2465 		/*
2466 		 * If enabled, put device in promiscuous mode.
2467 		 * This allows IO forwarding mode to forward packets
2468 		 * to itself through 2 cross-connected  ports of the
2469 		 * target machine.
2470 		 */
2471 		if (promiscuous_on) {
2472 			ret = rte_eth_promiscuous_enable(portid);
2473 			if (ret != 0)
2474 				rte_exit(EXIT_FAILURE,
2475 					"rte_eth_promiscuous_enable: err=%s, port=%u\n",
2476 					rte_strerror(-ret), portid);
2477 		}
2478 		/* initialize spinlock for each port */
2479 		rte_spinlock_init(&(locks[portid]));
2480 	}
2481 
2482 	check_all_ports_link_status(enabled_port_mask);
2483 
2484 	if (app_mode == APP_MODE_EMPTY_POLL) {
2485 
2486 		if (empty_poll_train) {
2487 			policy.state = TRAINING;
2488 		} else {
2489 			policy.state = MED_NORMAL;
2490 			policy.med_base_edpi = ep_med_edpi;
2491 			policy.hgh_base_edpi = ep_hgh_edpi;
2492 		}
2493 
2494 		ret = rte_power_empty_poll_stat_init(&ep_params,
2495 				freq_tlb,
2496 				&policy);
2497 		if (ret < 0)
2498 			rte_exit(EXIT_FAILURE, "empty poll init failed");
2499 	}
2500 
2501 
2502 	/* launch per-lcore init on every lcore */
2503 	if (app_mode == APP_MODE_LEGACY) {
2504 		rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2505 	} else if (app_mode == APP_MODE_EMPTY_POLL) {
2506 		empty_poll_stop = false;
2507 		rte_eal_mp_remote_launch(main_empty_poll_loop, NULL,
2508 				SKIP_MASTER);
2509 	} else {
2510 		unsigned int i;
2511 
2512 		/* Init metrics library */
2513 		rte_metrics_init(rte_socket_id());
2514 		/** Register stats with metrics library */
2515 		for (i = 0; i < num_telstats; i++)
2516 			ptr_strings[i] = telstats_strings[i].name;
2517 
2518 		ret = rte_metrics_reg_names(ptr_strings, num_telstats);
2519 		if (ret >= 0)
2520 			telstats_index = ret;
2521 		else
2522 			rte_exit(EXIT_FAILURE, "failed to register metrics names");
2523 
2524 		RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2525 			rte_spinlock_init(&stats[lcore_id].telemetry_lock);
2526 		}
2527 		rte_timer_init(&telemetry_timer);
2528 		rte_eal_mp_remote_launch(main_telemetry_loop, NULL,
2529 						SKIP_MASTER);
2530 	}
2531 
2532 	if (app_mode == APP_MODE_EMPTY_POLL || app_mode == APP_MODE_TELEMETRY)
2533 		launch_timer(rte_lcore_id());
2534 
2535 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2536 		if (rte_eal_wait_lcore(lcore_id) < 0)
2537 			return -1;
2538 	}
2539 
2540 	if (app_mode == APP_MODE_EMPTY_POLL)
2541 		rte_power_empty_poll_stat_free();
2542 
2543 	return 0;
2544 }
2545