xref: /dpdk/examples/l3fwd-power/main.c (revision 8809f78c7dd9f33a44a4f89c58fc91ded34296ed)
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 #include <rte_telemetry.h>
50 
51 #include "perf_core.h"
52 #include "main.h"
53 
54 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
55 
56 #define MAX_PKT_BURST 32
57 
58 #define MIN_ZERO_POLL_COUNT 10
59 
60 /* 100 ms interval */
61 #define TIMER_NUMBER_PER_SECOND           10
62 /* (10ms) */
63 #define INTERVALS_PER_SECOND             100
64 /* 100000 us */
65 #define SCALING_PERIOD                    (1000000/TIMER_NUMBER_PER_SECOND)
66 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
67 
68 #define APP_LOOKUP_EXACT_MATCH          0
69 #define APP_LOOKUP_LPM                  1
70 #define DO_RFC_1812_CHECKS
71 
72 #ifndef APP_LOOKUP_METHOD
73 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
74 #endif
75 
76 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
77 #include <rte_hash.h>
78 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
79 #include <rte_lpm.h>
80 #else
81 #error "APP_LOOKUP_METHOD set to incorrect value"
82 #endif
83 
84 #ifndef IPv6_BYTES
85 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
86                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
87 #define IPv6_BYTES(addr) \
88 	addr[0],  addr[1], addr[2],  addr[3], \
89 	addr[4],  addr[5], addr[6],  addr[7], \
90 	addr[8],  addr[9], addr[10], addr[11],\
91 	addr[12], addr[13],addr[14], addr[15]
92 #endif
93 
94 #define MAX_JUMBO_PKT_LEN  9600
95 
96 #define IPV6_ADDR_LEN 16
97 
98 #define MEMPOOL_CACHE_SIZE 256
99 
100 /*
101  * This expression is used to calculate the number of mbufs needed depending on
102  * user input, taking into account memory for rx and tx hardware rings, cache
103  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
104  * NB_MBUF never goes below a minimum value of 8192.
105  */
106 
107 #define NB_MBUF RTE_MAX	( \
108 	(nb_ports*nb_rx_queue*nb_rxd + \
109 	nb_ports*nb_lcores*MAX_PKT_BURST + \
110 	nb_ports*n_tx_queue*nb_txd + \
111 	nb_lcores*MEMPOOL_CACHE_SIZE), \
112 	(unsigned)8192)
113 
114 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
115 
116 #define NB_SOCKETS 8
117 
118 /* Configure how many packets ahead to prefetch, when reading packets */
119 #define PREFETCH_OFFSET	3
120 
121 /*
122  * Configurable number of RX/TX ring descriptors
123  */
124 #define RTE_TEST_RX_DESC_DEFAULT 1024
125 #define RTE_TEST_TX_DESC_DEFAULT 1024
126 
127 /*
128  * These two thresholds were decided on by running the training algorithm on
129  * a 2.5GHz Xeon. These defaults can be overridden by supplying non-zero values
130  * for the med_threshold and high_threshold parameters on the command line.
131  */
132 #define EMPTY_POLL_MED_THRESHOLD 350000UL
133 #define EMPTY_POLL_HGH_THRESHOLD 580000UL
134 
135 #define NUM_TELSTATS RTE_DIM(telstats_strings)
136 
137 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
138 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
139 
140 /* ethernet addresses of ports */
141 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
142 
143 /* ethernet addresses of ports */
144 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
145 
146 /* mask of enabled ports */
147 static uint32_t enabled_port_mask = 0;
148 /* Ports set in promiscuous mode off by default. */
149 static int promiscuous_on = 0;
150 /* NUMA is enabled by default. */
151 static int numa_on = 1;
152 static bool empty_poll_stop;
153 static bool empty_poll_train;
154 volatile bool quit_signal;
155 static struct  ep_params *ep_params;
156 static struct  ep_policy policy;
157 static long  ep_med_edpi, ep_hgh_edpi;
158 /* timer to update telemetry every 500ms */
159 static struct rte_timer telemetry_timer;
160 
161 /* stats index returned by metrics lib */
162 int telstats_index;
163 
164 struct telstats_name {
165 	char name[RTE_ETH_XSTATS_NAME_SIZE];
166 };
167 
168 /* telemetry stats to be reported */
169 const struct telstats_name telstats_strings[] = {
170 	{"empty_poll"},
171 	{"full_poll"},
172 	{"busy_percent"}
173 };
174 
175 /* core busyness in percentage */
176 enum busy_rate {
177 	ZERO = 0,
178 	PARTIAL = 50,
179 	FULL = 100
180 };
181 
182 /* reference poll count to measure core busyness */
183 #define DEFAULT_COUNT 10000
184 /*
185  * reference CYCLES to be used to
186  * measure core busyness based on poll count
187  */
188 #define MIN_CYCLES  1500000ULL
189 #define MAX_CYCLES 22000000ULL
190 
191 /* (500ms) */
192 #define TELEMETRY_INTERVALS_PER_SEC 2
193 
194 static int parse_ptype; /**< Parse packet type using rx callback, and */
195 			/**< disabled by default */
196 
197 enum appmode {
198 	APP_MODE_DEFAULT = 0,
199 	APP_MODE_LEGACY,
200 	APP_MODE_EMPTY_POLL,
201 	APP_MODE_TELEMETRY,
202 	APP_MODE_INTERRUPT
203 };
204 
205 enum appmode app_mode;
206 
207 enum freq_scale_hint_t
208 {
209 	FREQ_LOWER    =      -1,
210 	FREQ_CURRENT  =       0,
211 	FREQ_HIGHER   =       1,
212 	FREQ_HIGHEST  =       2
213 };
214 
215 struct lcore_rx_queue {
216 	uint16_t port_id;
217 	uint8_t queue_id;
218 	enum freq_scale_hint_t freq_up_hint;
219 	uint32_t zero_rx_packet_count;
220 	uint32_t idle_hint;
221 } __rte_cache_aligned;
222 
223 #define MAX_RX_QUEUE_PER_LCORE 16
224 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
225 #define MAX_RX_QUEUE_PER_PORT 128
226 
227 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
228 
229 
230 struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
231 static struct lcore_params lcore_params_array_default[] = {
232 	{0, 0, 2},
233 	{0, 1, 2},
234 	{0, 2, 2},
235 	{1, 0, 2},
236 	{1, 1, 2},
237 	{1, 2, 2},
238 	{2, 0, 2},
239 	{3, 0, 3},
240 	{3, 1, 3},
241 };
242 
243 struct lcore_params *lcore_params = lcore_params_array_default;
244 uint16_t nb_lcore_params = RTE_DIM(lcore_params_array_default);
245 
246 static struct rte_eth_conf port_conf = {
247 	.rxmode = {
248 		.mq_mode        = ETH_MQ_RX_RSS,
249 		.max_rx_pkt_len = RTE_ETHER_MAX_LEN,
250 		.split_hdr_size = 0,
251 		.offloads = DEV_RX_OFFLOAD_CHECKSUM,
252 	},
253 	.rx_adv_conf = {
254 		.rss_conf = {
255 			.rss_key = NULL,
256 			.rss_hf = ETH_RSS_UDP,
257 		},
258 	},
259 	.txmode = {
260 		.mq_mode = ETH_MQ_TX_NONE,
261 	}
262 };
263 
264 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
265 
266 
267 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
268 
269 #ifdef RTE_ARCH_X86
270 #include <rte_hash_crc.h>
271 #define DEFAULT_HASH_FUNC       rte_hash_crc
272 #else
273 #include <rte_jhash.h>
274 #define DEFAULT_HASH_FUNC       rte_jhash
275 #endif
276 
277 struct ipv4_5tuple {
278 	uint32_t ip_dst;
279 	uint32_t ip_src;
280 	uint16_t port_dst;
281 	uint16_t port_src;
282 	uint8_t  proto;
283 } __rte_packed;
284 
285 struct ipv6_5tuple {
286 	uint8_t  ip_dst[IPV6_ADDR_LEN];
287 	uint8_t  ip_src[IPV6_ADDR_LEN];
288 	uint16_t port_dst;
289 	uint16_t port_src;
290 	uint8_t  proto;
291 } __rte_packed;
292 
293 struct ipv4_l3fwd_route {
294 	struct ipv4_5tuple key;
295 	uint8_t if_out;
296 };
297 
298 struct ipv6_l3fwd_route {
299 	struct ipv6_5tuple key;
300 	uint8_t if_out;
301 };
302 
303 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
304 	{{RTE_IPV4(100,10,0,1), RTE_IPV4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
305 	{{RTE_IPV4(100,20,0,2), RTE_IPV4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
306 	{{RTE_IPV4(100,30,0,3), RTE_IPV4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
307 	{{RTE_IPV4(100,40,0,4), RTE_IPV4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
308 };
309 
310 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
311 	{
312 		{
313 			{0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
314 			 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
315 			{0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
316 			 0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
317 			 1, 10, IPPROTO_UDP
318 		}, 4
319 	},
320 };
321 
322 typedef struct rte_hash lookup_struct_t;
323 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
324 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
325 
326 #define L3FWD_HASH_ENTRIES	1024
327 
328 static uint16_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
329 static uint16_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
330 #endif
331 
332 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
333 struct ipv4_l3fwd_route {
334 	uint32_t ip;
335 	uint8_t  depth;
336 	uint8_t  if_out;
337 };
338 
339 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
340 	{RTE_IPV4(1,1,1,0), 24, 0},
341 	{RTE_IPV4(2,1,1,0), 24, 1},
342 	{RTE_IPV4(3,1,1,0), 24, 2},
343 	{RTE_IPV4(4,1,1,0), 24, 3},
344 	{RTE_IPV4(5,1,1,0), 24, 4},
345 	{RTE_IPV4(6,1,1,0), 24, 5},
346 	{RTE_IPV4(7,1,1,0), 24, 6},
347 	{RTE_IPV4(8,1,1,0), 24, 7},
348 };
349 
350 #define IPV4_L3FWD_LPM_MAX_RULES     1024
351 
352 typedef struct rte_lpm lookup_struct_t;
353 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
354 #endif
355 
356 struct lcore_conf {
357 	uint16_t n_rx_queue;
358 	struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
359 	uint16_t n_tx_port;
360 	uint16_t tx_port_id[RTE_MAX_ETHPORTS];
361 	uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
362 	struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
363 	lookup_struct_t * ipv4_lookup_struct;
364 	lookup_struct_t * ipv6_lookup_struct;
365 } __rte_cache_aligned;
366 
367 struct lcore_stats {
368 	/* total sleep time in ms since last frequency scaling down */
369 	uint32_t sleep_time;
370 	/* number of long sleep recently */
371 	uint32_t nb_long_sleep;
372 	/* freq. scaling up trend */
373 	uint32_t trend;
374 	/* total packet processed recently */
375 	uint64_t nb_rx_processed;
376 	/* total iterations looped recently */
377 	uint64_t nb_iteration_looped;
378 	/*
379 	 * Represents empty and non empty polls
380 	 * of rte_eth_rx_burst();
381 	 * ep_nep[0] holds non empty polls
382 	 * i.e. 0 < nb_rx <= MAX_BURST
383 	 * ep_nep[1] holds empty polls.
384 	 * i.e. nb_rx == 0
385 	 */
386 	uint64_t ep_nep[2];
387 	/*
388 	 * Represents full and empty+partial
389 	 * polls of rte_eth_rx_burst();
390 	 * ep_nep[0] holds empty+partial polls.
391 	 * i.e. 0 <= nb_rx < MAX_BURST
392 	 * ep_nep[1] holds full polls
393 	 * i.e. nb_rx == MAX_BURST
394 	 */
395 	uint64_t fp_nfp[2];
396 	enum busy_rate br;
397 	rte_spinlock_t telemetry_lock;
398 } __rte_cache_aligned;
399 
400 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
401 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
402 static struct rte_timer power_timers[RTE_MAX_LCORE];
403 
404 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
405 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
406 		unsigned int lcore_id, uint16_t port_id, uint16_t queue_id);
407 
408 
409 /*
410  * These defaults are using the max frequency index (1), a medium index (9)
411  * and a typical low frequency index (14). These can be adjusted to use
412  * different indexes using the relevant command line parameters.
413  */
414 static uint8_t  freq_tlb[] = {14, 9, 1};
415 
416 static int is_done(void)
417 {
418 	return quit_signal;
419 }
420 
421 /* exit signal handler */
422 static void
423 signal_exit_now(int sigtype)
424 {
425 
426 	if (sigtype == SIGINT)
427 		quit_signal = true;
428 
429 }
430 
431 /*  Freqency scale down timer callback */
432 static void
433 power_timer_cb(__rte_unused struct rte_timer *tim,
434 			  __rte_unused void *arg)
435 {
436 	uint64_t hz;
437 	float sleep_time_ratio;
438 	unsigned lcore_id = rte_lcore_id();
439 
440 	/* accumulate total execution time in us when callback is invoked */
441 	sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
442 					(float)SCALING_PERIOD;
443 	/**
444 	 * check whether need to scale down frequency a step if it sleep a lot.
445 	 */
446 	if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
447 		if (rte_power_freq_down)
448 			rte_power_freq_down(lcore_id);
449 	}
450 	else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
451 		stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
452 		/**
453 		 * scale down a step if average packet per iteration less
454 		 * than expectation.
455 		 */
456 		if (rte_power_freq_down)
457 			rte_power_freq_down(lcore_id);
458 	}
459 
460 	/**
461 	 * initialize another timer according to current frequency to ensure
462 	 * timer interval is relatively fixed.
463 	 */
464 	hz = rte_get_timer_hz();
465 	rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
466 				SINGLE, lcore_id, power_timer_cb, NULL);
467 
468 	stats[lcore_id].nb_rx_processed = 0;
469 	stats[lcore_id].nb_iteration_looped = 0;
470 
471 	stats[lcore_id].sleep_time = 0;
472 }
473 
474 /* Enqueue a single packet, and send burst if queue is filled */
475 static inline int
476 send_single_packet(struct rte_mbuf *m, uint16_t port)
477 {
478 	uint32_t lcore_id;
479 	struct lcore_conf *qconf;
480 
481 	lcore_id = rte_lcore_id();
482 	qconf = &lcore_conf[lcore_id];
483 
484 	rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
485 			qconf->tx_buffer[port], m);
486 
487 	return 0;
488 }
489 
490 #ifdef DO_RFC_1812_CHECKS
491 static inline int
492 is_valid_ipv4_pkt(struct rte_ipv4_hdr *pkt, uint32_t link_len)
493 {
494 	/* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
495 	/*
496 	 * 1. The packet length reported by the Link Layer must be large
497 	 * enough to hold the minimum length legal IP datagram (20 bytes).
498 	 */
499 	if (link_len < sizeof(struct rte_ipv4_hdr))
500 		return -1;
501 
502 	/* 2. The IP checksum must be correct. */
503 	/* this is checked in H/W */
504 
505 	/*
506 	 * 3. The IP version number must be 4. If the version number is not 4
507 	 * then the packet may be another version of IP, such as IPng or
508 	 * ST-II.
509 	 */
510 	if (((pkt->version_ihl) >> 4) != 4)
511 		return -3;
512 	/*
513 	 * 4. The IP header length field must be large enough to hold the
514 	 * minimum length legal IP datagram (20 bytes = 5 words).
515 	 */
516 	if ((pkt->version_ihl & 0xf) < 5)
517 		return -4;
518 
519 	/*
520 	 * 5. The IP total length field must be large enough to hold the IP
521 	 * datagram header, whose length is specified in the IP header length
522 	 * field.
523 	 */
524 	if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct rte_ipv4_hdr))
525 		return -5;
526 
527 	return 0;
528 }
529 #endif
530 
531 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
532 static void
533 print_ipv4_key(struct ipv4_5tuple key)
534 {
535 	printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
536 		"proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
537 				key.port_dst, key.port_src, key.proto);
538 }
539 static void
540 print_ipv6_key(struct ipv6_5tuple key)
541 {
542 	printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
543 	        "port dst = %d, port src = %d, proto = %d\n",
544 	        IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
545 	        key.port_dst, key.port_src, key.proto);
546 }
547 
548 static inline uint16_t
549 get_ipv4_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
550 		lookup_struct_t * ipv4_l3fwd_lookup_struct)
551 {
552 	struct ipv4_5tuple key;
553 	struct rte_tcp_hdr *tcp;
554 	struct rte_udp_hdr *udp;
555 	int ret = 0;
556 
557 	key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
558 	key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
559 	key.proto = ipv4_hdr->next_proto_id;
560 
561 	switch (ipv4_hdr->next_proto_id) {
562 	case IPPROTO_TCP:
563 		tcp = (struct rte_tcp_hdr *)((unsigned char *)ipv4_hdr +
564 					sizeof(struct rte_ipv4_hdr));
565 		key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
566 		key.port_src = rte_be_to_cpu_16(tcp->src_port);
567 		break;
568 
569 	case IPPROTO_UDP:
570 		udp = (struct rte_udp_hdr *)((unsigned char *)ipv4_hdr +
571 					sizeof(struct rte_ipv4_hdr));
572 		key.port_dst = rte_be_to_cpu_16(udp->dst_port);
573 		key.port_src = rte_be_to_cpu_16(udp->src_port);
574 		break;
575 
576 	default:
577 		key.port_dst = 0;
578 		key.port_src = 0;
579 		break;
580 	}
581 
582 	/* Find destination port */
583 	ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
584 	return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
585 }
586 
587 static inline uint16_t
588 get_ipv6_dst_port(struct rte_ipv6_hdr *ipv6_hdr, uint16_t portid,
589 			lookup_struct_t *ipv6_l3fwd_lookup_struct)
590 {
591 	struct ipv6_5tuple key;
592 	struct rte_tcp_hdr *tcp;
593 	struct rte_udp_hdr *udp;
594 	int ret = 0;
595 
596 	memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
597 	memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
598 
599 	key.proto = ipv6_hdr->proto;
600 
601 	switch (ipv6_hdr->proto) {
602 	case IPPROTO_TCP:
603 		tcp = (struct rte_tcp_hdr *)((unsigned char *) ipv6_hdr +
604 					sizeof(struct rte_ipv6_hdr));
605 		key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
606 		key.port_src = rte_be_to_cpu_16(tcp->src_port);
607 		break;
608 
609 	case IPPROTO_UDP:
610 		udp = (struct rte_udp_hdr *)((unsigned char *) ipv6_hdr +
611 					sizeof(struct rte_ipv6_hdr));
612 		key.port_dst = rte_be_to_cpu_16(udp->dst_port);
613 		key.port_src = rte_be_to_cpu_16(udp->src_port);
614 		break;
615 
616 	default:
617 		key.port_dst = 0;
618 		key.port_src = 0;
619 		break;
620 	}
621 
622 	/* Find destination port */
623 	ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
624 	return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
625 }
626 #endif
627 
628 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
629 static inline uint16_t
630 get_ipv4_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
631 		lookup_struct_t *ipv4_l3fwd_lookup_struct)
632 {
633 	uint32_t next_hop;
634 
635 	return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
636 			rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
637 			next_hop : portid);
638 }
639 #endif
640 
641 static inline void
642 parse_ptype_one(struct rte_mbuf *m)
643 {
644 	struct rte_ether_hdr *eth_hdr;
645 	uint32_t packet_type = RTE_PTYPE_UNKNOWN;
646 	uint16_t ether_type;
647 
648 	eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
649 	ether_type = eth_hdr->ether_type;
650 	if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
651 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
652 	else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6))
653 		packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
654 
655 	m->packet_type = packet_type;
656 }
657 
658 static uint16_t
659 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
660 	       struct rte_mbuf *pkts[], uint16_t nb_pkts,
661 	       uint16_t max_pkts __rte_unused,
662 	       void *user_param __rte_unused)
663 {
664 	unsigned int i;
665 
666 	for (i = 0; i < nb_pkts; ++i)
667 		parse_ptype_one(pkts[i]);
668 
669 	return nb_pkts;
670 }
671 
672 static int
673 add_cb_parse_ptype(uint16_t portid, uint16_t queueid)
674 {
675 	printf("Port %d: softly parse packet type info\n", portid);
676 	if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL))
677 		return 0;
678 
679 	printf("Failed to add rx callback: port=%d\n", portid);
680 	return -1;
681 }
682 
683 static inline void
684 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
685 				struct lcore_conf *qconf)
686 {
687 	struct rte_ether_hdr *eth_hdr;
688 	struct rte_ipv4_hdr *ipv4_hdr;
689 	void *d_addr_bytes;
690 	uint16_t dst_port;
691 
692 	eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
693 
694 	if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
695 		/* Handle IPv4 headers.*/
696 		ipv4_hdr =
697 			rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
698 						sizeof(struct rte_ether_hdr));
699 
700 #ifdef DO_RFC_1812_CHECKS
701 		/* Check to make sure the packet is valid (RFC1812) */
702 		if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
703 			rte_pktmbuf_free(m);
704 			return;
705 		}
706 #endif
707 
708 		dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
709 					qconf->ipv4_lookup_struct);
710 		if (dst_port >= RTE_MAX_ETHPORTS ||
711 				(enabled_port_mask & 1 << dst_port) == 0)
712 			dst_port = portid;
713 
714 		/* 02:00:00:00:00:xx */
715 		d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
716 		*((uint64_t *)d_addr_bytes) =
717 			0x000000000002 + ((uint64_t)dst_port << 40);
718 
719 #ifdef DO_RFC_1812_CHECKS
720 		/* Update time to live and header checksum */
721 		--(ipv4_hdr->time_to_live);
722 		++(ipv4_hdr->hdr_checksum);
723 #endif
724 
725 		/* src addr */
726 		rte_ether_addr_copy(&ports_eth_addr[dst_port],
727 				&eth_hdr->s_addr);
728 
729 		send_single_packet(m, dst_port);
730 	} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
731 		/* Handle IPv6 headers.*/
732 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
733 		struct rte_ipv6_hdr *ipv6_hdr;
734 
735 		ipv6_hdr =
736 			rte_pktmbuf_mtod_offset(m, struct rte_ipv6_hdr *,
737 						sizeof(struct rte_ether_hdr));
738 
739 		dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
740 					qconf->ipv6_lookup_struct);
741 
742 		if (dst_port >= RTE_MAX_ETHPORTS ||
743 				(enabled_port_mask & 1 << dst_port) == 0)
744 			dst_port = portid;
745 
746 		/* 02:00:00:00:00:xx */
747 		d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
748 		*((uint64_t *)d_addr_bytes) =
749 			0x000000000002 + ((uint64_t)dst_port << 40);
750 
751 		/* src addr */
752 		rte_ether_addr_copy(&ports_eth_addr[dst_port],
753 				&eth_hdr->s_addr);
754 
755 		send_single_packet(m, dst_port);
756 #else
757 		/* We don't currently handle IPv6 packets in LPM mode. */
758 		rte_pktmbuf_free(m);
759 #endif
760 	} else
761 		rte_pktmbuf_free(m);
762 
763 }
764 
765 #define MINIMUM_SLEEP_TIME         1
766 #define SUSPEND_THRESHOLD          300
767 
768 static inline uint32_t
769 power_idle_heuristic(uint32_t zero_rx_packet_count)
770 {
771 	/* If zero count is less than 100,  sleep 1us */
772 	if (zero_rx_packet_count < SUSPEND_THRESHOLD)
773 		return MINIMUM_SLEEP_TIME;
774 	/* If zero count is less than 1000, sleep 100 us which is the
775 		minimum latency switching from C3/C6 to C0
776 	*/
777 	else
778 		return SUSPEND_THRESHOLD;
779 }
780 
781 static inline enum freq_scale_hint_t
782 power_freq_scaleup_heuristic(unsigned lcore_id,
783 			     uint16_t port_id,
784 			     uint16_t queue_id)
785 {
786 	uint32_t rxq_count = rte_eth_rx_queue_count(port_id, queue_id);
787 /**
788  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
789  * per iteration
790  */
791 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
792 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
793 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
794 #define FREQ_UP_TREND1_ACC   1
795 #define FREQ_UP_TREND2_ACC   100
796 #define FREQ_UP_THRESHOLD    10000
797 
798 	if (likely(rxq_count > FREQ_GEAR3_RX_PACKET_THRESHOLD)) {
799 		stats[lcore_id].trend = 0;
800 		return FREQ_HIGHEST;
801 	} else if (likely(rxq_count > FREQ_GEAR2_RX_PACKET_THRESHOLD))
802 		stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
803 	else if (likely(rxq_count > FREQ_GEAR1_RX_PACKET_THRESHOLD))
804 		stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
805 
806 	if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
807 		stats[lcore_id].trend = 0;
808 		return FREQ_HIGHER;
809 	}
810 
811 	return FREQ_CURRENT;
812 }
813 
814 /**
815  * force polling thread sleep until one-shot rx interrupt triggers
816  * @param port_id
817  *  Port id.
818  * @param queue_id
819  *  Rx queue id.
820  * @return
821  *  0 on success
822  */
823 static int
824 sleep_until_rx_interrupt(int num, int lcore)
825 {
826 	/*
827 	 * we want to track when we are woken up by traffic so that we can go
828 	 * back to sleep again without log spamming. Avoid cache line sharing
829 	 * to prevent threads stepping on each others' toes.
830 	 */
831 	static struct {
832 		bool wakeup;
833 	} __rte_cache_aligned status[RTE_MAX_LCORE];
834 	struct rte_epoll_event event[num];
835 	int n, i;
836 	uint16_t port_id;
837 	uint8_t queue_id;
838 	void *data;
839 
840 	if (status[lcore].wakeup) {
841 		RTE_LOG(INFO, L3FWD_POWER,
842 				"lcore %u sleeps until interrupt triggers\n",
843 				rte_lcore_id());
844 	}
845 
846 	n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, 10);
847 	for (i = 0; i < n; i++) {
848 		data = event[i].epdata.data;
849 		port_id = ((uintptr_t)data) >> CHAR_BIT;
850 		queue_id = ((uintptr_t)data) &
851 			RTE_LEN2MASK(CHAR_BIT, uint8_t);
852 		RTE_LOG(INFO, L3FWD_POWER,
853 			"lcore %u is waked up from rx interrupt on"
854 			" port %d queue %d\n",
855 			rte_lcore_id(), port_id, queue_id);
856 	}
857 	status[lcore].wakeup = n != 0;
858 
859 	return 0;
860 }
861 
862 static void turn_on_off_intr(struct lcore_conf *qconf, bool on)
863 {
864 	int i;
865 	struct lcore_rx_queue *rx_queue;
866 	uint8_t queue_id;
867 	uint16_t port_id;
868 
869 	for (i = 0; i < qconf->n_rx_queue; ++i) {
870 		rx_queue = &(qconf->rx_queue_list[i]);
871 		port_id = rx_queue->port_id;
872 		queue_id = rx_queue->queue_id;
873 
874 		rte_spinlock_lock(&(locks[port_id]));
875 		if (on)
876 			rte_eth_dev_rx_intr_enable(port_id, queue_id);
877 		else
878 			rte_eth_dev_rx_intr_disable(port_id, queue_id);
879 		rte_spinlock_unlock(&(locks[port_id]));
880 	}
881 }
882 
883 static int event_register(struct lcore_conf *qconf)
884 {
885 	struct lcore_rx_queue *rx_queue;
886 	uint8_t queueid;
887 	uint16_t portid;
888 	uint32_t data;
889 	int ret;
890 	int i;
891 
892 	for (i = 0; i < qconf->n_rx_queue; ++i) {
893 		rx_queue = &(qconf->rx_queue_list[i]);
894 		portid = rx_queue->port_id;
895 		queueid = rx_queue->queue_id;
896 		data = portid << CHAR_BIT | queueid;
897 
898 		ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
899 						RTE_EPOLL_PER_THREAD,
900 						RTE_INTR_EVENT_ADD,
901 						(void *)((uintptr_t)data));
902 		if (ret)
903 			return ret;
904 	}
905 
906 	return 0;
907 }
908 
909 /* main processing loop */
910 static int main_intr_loop(__rte_unused void *dummy)
911 {
912 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
913 	unsigned int lcore_id;
914 	uint64_t prev_tsc, diff_tsc, cur_tsc;
915 	int i, j, nb_rx;
916 	uint8_t queueid;
917 	uint16_t portid;
918 	struct lcore_conf *qconf;
919 	struct lcore_rx_queue *rx_queue;
920 	uint32_t lcore_rx_idle_count = 0;
921 	uint32_t lcore_idle_hint = 0;
922 	int intr_en = 0;
923 
924 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
925 				   US_PER_S * BURST_TX_DRAIN_US;
926 
927 	prev_tsc = 0;
928 
929 	lcore_id = rte_lcore_id();
930 	qconf = &lcore_conf[lcore_id];
931 
932 	if (qconf->n_rx_queue == 0) {
933 		RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
934 				lcore_id);
935 		return 0;
936 	}
937 
938 	RTE_LOG(INFO, L3FWD_POWER, "entering main interrupt loop on lcore %u\n",
939 			lcore_id);
940 
941 	for (i = 0; i < qconf->n_rx_queue; i++) {
942 		portid = qconf->rx_queue_list[i].port_id;
943 		queueid = qconf->rx_queue_list[i].queue_id;
944 		RTE_LOG(INFO, L3FWD_POWER,
945 				" -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
946 				lcore_id, portid, queueid);
947 	}
948 
949 	/* add into event wait list */
950 	if (event_register(qconf) == 0)
951 		intr_en = 1;
952 	else
953 		RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
954 
955 	while (!is_done()) {
956 		stats[lcore_id].nb_iteration_looped++;
957 
958 		cur_tsc = rte_rdtsc();
959 
960 		/*
961 		 * TX burst queue drain
962 		 */
963 		diff_tsc = cur_tsc - prev_tsc;
964 		if (unlikely(diff_tsc > drain_tsc)) {
965 			for (i = 0; i < qconf->n_tx_port; ++i) {
966 				portid = qconf->tx_port_id[i];
967 				rte_eth_tx_buffer_flush(portid,
968 						qconf->tx_queue_id[portid],
969 						qconf->tx_buffer[portid]);
970 			}
971 			prev_tsc = cur_tsc;
972 		}
973 
974 start_rx:
975 		/*
976 		 * Read packet from RX queues
977 		 */
978 		lcore_rx_idle_count = 0;
979 		for (i = 0; i < qconf->n_rx_queue; ++i) {
980 			rx_queue = &(qconf->rx_queue_list[i]);
981 			rx_queue->idle_hint = 0;
982 			portid = rx_queue->port_id;
983 			queueid = rx_queue->queue_id;
984 
985 			nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
986 					MAX_PKT_BURST);
987 
988 			stats[lcore_id].nb_rx_processed += nb_rx;
989 			if (unlikely(nb_rx == 0)) {
990 				/**
991 				 * no packet received from rx queue, try to
992 				 * sleep for a while forcing CPU enter deeper
993 				 * C states.
994 				 */
995 				rx_queue->zero_rx_packet_count++;
996 
997 				if (rx_queue->zero_rx_packet_count <=
998 						MIN_ZERO_POLL_COUNT)
999 					continue;
1000 
1001 				rx_queue->idle_hint = power_idle_heuristic(
1002 						rx_queue->zero_rx_packet_count);
1003 				lcore_rx_idle_count++;
1004 			} else {
1005 				rx_queue->zero_rx_packet_count = 0;
1006 			}
1007 
1008 			/* Prefetch first packets */
1009 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1010 				rte_prefetch0(rte_pktmbuf_mtod(
1011 						pkts_burst[j], void *));
1012 			}
1013 
1014 			/* Prefetch and forward already prefetched packets */
1015 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1016 				rte_prefetch0(rte_pktmbuf_mtod(
1017 						pkts_burst[j + PREFETCH_OFFSET],
1018 						void *));
1019 				l3fwd_simple_forward(
1020 						pkts_burst[j], portid, qconf);
1021 			}
1022 
1023 			/* Forward remaining prefetched packets */
1024 			for (; j < nb_rx; j++) {
1025 				l3fwd_simple_forward(
1026 						pkts_burst[j], portid, qconf);
1027 			}
1028 		}
1029 
1030 		if (unlikely(lcore_rx_idle_count == qconf->n_rx_queue)) {
1031 			/**
1032 			 * All Rx queues empty in recent consecutive polls,
1033 			 * sleep in a conservative manner, meaning sleep as
1034 			 * less as possible.
1035 			 */
1036 			for (i = 1,
1037 			    lcore_idle_hint = qconf->rx_queue_list[0].idle_hint;
1038 					i < qconf->n_rx_queue; ++i) {
1039 				rx_queue = &(qconf->rx_queue_list[i]);
1040 				if (rx_queue->idle_hint < lcore_idle_hint)
1041 					lcore_idle_hint = rx_queue->idle_hint;
1042 			}
1043 
1044 			if (lcore_idle_hint < SUSPEND_THRESHOLD)
1045 				/**
1046 				 * execute "pause" instruction to avoid context
1047 				 * switch which generally take hundred of
1048 				 * microseconds for short sleep.
1049 				 */
1050 				rte_delay_us(lcore_idle_hint);
1051 			else {
1052 				/* suspend until rx interrupt triggers */
1053 				if (intr_en) {
1054 					turn_on_off_intr(qconf, 1);
1055 					sleep_until_rx_interrupt(
1056 							qconf->n_rx_queue,
1057 							lcore_id);
1058 					turn_on_off_intr(qconf, 0);
1059 					/**
1060 					 * start receiving packets immediately
1061 					 */
1062 					if (likely(!is_done()))
1063 						goto start_rx;
1064 				}
1065 			}
1066 			stats[lcore_id].sleep_time += lcore_idle_hint;
1067 		}
1068 	}
1069 
1070 	return 0;
1071 }
1072 
1073 /* main processing loop */
1074 static int
1075 main_telemetry_loop(__rte_unused void *dummy)
1076 {
1077 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1078 	unsigned int lcore_id;
1079 	uint64_t prev_tsc, diff_tsc, cur_tsc, prev_tel_tsc;
1080 	int i, j, nb_rx;
1081 	uint8_t queueid;
1082 	uint16_t portid;
1083 	struct lcore_conf *qconf;
1084 	struct lcore_rx_queue *rx_queue;
1085 	uint64_t ep_nep[2] = {0}, fp_nfp[2] = {0};
1086 	uint64_t poll_count;
1087 	enum busy_rate br;
1088 
1089 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1090 					US_PER_S * BURST_TX_DRAIN_US;
1091 
1092 	poll_count = 0;
1093 	prev_tsc = 0;
1094 	prev_tel_tsc = 0;
1095 
1096 	lcore_id = rte_lcore_id();
1097 	qconf = &lcore_conf[lcore_id];
1098 
1099 	if (qconf->n_rx_queue == 0) {
1100 		RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
1101 			lcore_id);
1102 		return 0;
1103 	}
1104 
1105 	RTE_LOG(INFO, L3FWD_POWER, "entering main telemetry loop on lcore %u\n",
1106 		lcore_id);
1107 
1108 	for (i = 0; i < qconf->n_rx_queue; i++) {
1109 		portid = qconf->rx_queue_list[i].port_id;
1110 		queueid = qconf->rx_queue_list[i].queue_id;
1111 		RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1112 			"rxqueueid=%hhu\n", lcore_id, portid, queueid);
1113 	}
1114 
1115 	while (!is_done()) {
1116 
1117 		cur_tsc = rte_rdtsc();
1118 		/*
1119 		 * TX burst queue drain
1120 		 */
1121 		diff_tsc = cur_tsc - prev_tsc;
1122 		if (unlikely(diff_tsc > drain_tsc)) {
1123 			for (i = 0; i < qconf->n_tx_port; ++i) {
1124 				portid = qconf->tx_port_id[i];
1125 				rte_eth_tx_buffer_flush(portid,
1126 						qconf->tx_queue_id[portid],
1127 						qconf->tx_buffer[portid]);
1128 			}
1129 			prev_tsc = cur_tsc;
1130 		}
1131 
1132 		/*
1133 		 * Read packet from RX queues
1134 		 */
1135 		for (i = 0; i < qconf->n_rx_queue; ++i) {
1136 			rx_queue = &(qconf->rx_queue_list[i]);
1137 			portid = rx_queue->port_id;
1138 			queueid = rx_queue->queue_id;
1139 
1140 			nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1141 								MAX_PKT_BURST);
1142 			ep_nep[nb_rx == 0]++;
1143 			fp_nfp[nb_rx == MAX_PKT_BURST]++;
1144 			poll_count++;
1145 			if (unlikely(nb_rx == 0))
1146 				continue;
1147 
1148 			/* Prefetch first packets */
1149 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1150 				rte_prefetch0(rte_pktmbuf_mtod(
1151 						pkts_burst[j], void *));
1152 			}
1153 
1154 			/* Prefetch and forward already prefetched packets */
1155 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1156 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1157 						j + PREFETCH_OFFSET], void *));
1158 				l3fwd_simple_forward(pkts_burst[j], portid,
1159 								qconf);
1160 			}
1161 
1162 			/* Forward remaining prefetched packets */
1163 			for (; j < nb_rx; j++) {
1164 				l3fwd_simple_forward(pkts_burst[j], portid,
1165 								qconf);
1166 			}
1167 		}
1168 		if (unlikely(poll_count >= DEFAULT_COUNT)) {
1169 			diff_tsc = cur_tsc - prev_tel_tsc;
1170 			if (diff_tsc >= MAX_CYCLES) {
1171 				br = FULL;
1172 			} else if (diff_tsc > MIN_CYCLES &&
1173 					diff_tsc < MAX_CYCLES) {
1174 				br = (diff_tsc * 100) / MAX_CYCLES;
1175 			} else {
1176 				br = ZERO;
1177 			}
1178 			poll_count = 0;
1179 			prev_tel_tsc = cur_tsc;
1180 			/* update stats for telemetry */
1181 			rte_spinlock_lock(&stats[lcore_id].telemetry_lock);
1182 			stats[lcore_id].ep_nep[0] = ep_nep[0];
1183 			stats[lcore_id].ep_nep[1] = ep_nep[1];
1184 			stats[lcore_id].fp_nfp[0] = fp_nfp[0];
1185 			stats[lcore_id].fp_nfp[1] = fp_nfp[1];
1186 			stats[lcore_id].br = br;
1187 			rte_spinlock_unlock(&stats[lcore_id].telemetry_lock);
1188 		}
1189 	}
1190 
1191 	return 0;
1192 }
1193 /* main processing loop */
1194 static int
1195 main_empty_poll_loop(__rte_unused void *dummy)
1196 {
1197 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1198 	unsigned int lcore_id;
1199 	uint64_t prev_tsc, diff_tsc, cur_tsc;
1200 	int i, j, nb_rx;
1201 	uint8_t queueid;
1202 	uint16_t portid;
1203 	struct lcore_conf *qconf;
1204 	struct lcore_rx_queue *rx_queue;
1205 
1206 	const uint64_t drain_tsc =
1207 		(rte_get_tsc_hz() + US_PER_S - 1) /
1208 		US_PER_S * BURST_TX_DRAIN_US;
1209 
1210 	prev_tsc = 0;
1211 
1212 	lcore_id = rte_lcore_id();
1213 	qconf = &lcore_conf[lcore_id];
1214 
1215 	if (qconf->n_rx_queue == 0) {
1216 		RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
1217 			lcore_id);
1218 		return 0;
1219 	}
1220 
1221 	for (i = 0; i < qconf->n_rx_queue; i++) {
1222 		portid = qconf->rx_queue_list[i].port_id;
1223 		queueid = qconf->rx_queue_list[i].queue_id;
1224 		RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1225 				"rxqueueid=%hhu\n", lcore_id, portid, queueid);
1226 	}
1227 
1228 	while (!is_done()) {
1229 		stats[lcore_id].nb_iteration_looped++;
1230 
1231 		cur_tsc = rte_rdtsc();
1232 		/*
1233 		 * TX burst queue drain
1234 		 */
1235 		diff_tsc = cur_tsc - prev_tsc;
1236 		if (unlikely(diff_tsc > drain_tsc)) {
1237 			for (i = 0; i < qconf->n_tx_port; ++i) {
1238 				portid = qconf->tx_port_id[i];
1239 				rte_eth_tx_buffer_flush(portid,
1240 						qconf->tx_queue_id[portid],
1241 						qconf->tx_buffer[portid]);
1242 			}
1243 			prev_tsc = cur_tsc;
1244 		}
1245 
1246 		/*
1247 		 * Read packet from RX queues
1248 		 */
1249 		for (i = 0; i < qconf->n_rx_queue; ++i) {
1250 			rx_queue = &(qconf->rx_queue_list[i]);
1251 			rx_queue->idle_hint = 0;
1252 			portid = rx_queue->port_id;
1253 			queueid = rx_queue->queue_id;
1254 
1255 			nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1256 					MAX_PKT_BURST);
1257 
1258 			stats[lcore_id].nb_rx_processed += nb_rx;
1259 
1260 			if (nb_rx == 0) {
1261 
1262 				rte_power_empty_poll_stat_update(lcore_id);
1263 
1264 				continue;
1265 			} else {
1266 				rte_power_poll_stat_update(lcore_id, nb_rx);
1267 			}
1268 
1269 
1270 			/* Prefetch first packets */
1271 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1272 				rte_prefetch0(rte_pktmbuf_mtod(
1273 							pkts_burst[j], void *));
1274 			}
1275 
1276 			/* Prefetch and forward already prefetched packets */
1277 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1278 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1279 							j + PREFETCH_OFFSET],
1280 							void *));
1281 				l3fwd_simple_forward(pkts_burst[j], portid,
1282 						qconf);
1283 			}
1284 
1285 			/* Forward remaining prefetched packets */
1286 			for (; j < nb_rx; j++) {
1287 				l3fwd_simple_forward(pkts_burst[j], portid,
1288 						qconf);
1289 			}
1290 
1291 		}
1292 
1293 	}
1294 
1295 	return 0;
1296 }
1297 /* main processing loop */
1298 static int
1299 main_legacy_loop(__rte_unused void *dummy)
1300 {
1301 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1302 	unsigned lcore_id;
1303 	uint64_t prev_tsc, diff_tsc, cur_tsc, tim_res_tsc, hz;
1304 	uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
1305 	int i, j, nb_rx;
1306 	uint8_t queueid;
1307 	uint16_t portid;
1308 	struct lcore_conf *qconf;
1309 	struct lcore_rx_queue *rx_queue;
1310 	enum freq_scale_hint_t lcore_scaleup_hint;
1311 	uint32_t lcore_rx_idle_count = 0;
1312 	uint32_t lcore_idle_hint = 0;
1313 	int intr_en = 0;
1314 
1315 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
1316 
1317 	prev_tsc = 0;
1318 	hz = rte_get_timer_hz();
1319 	tim_res_tsc = hz/TIMER_NUMBER_PER_SECOND;
1320 
1321 	lcore_id = rte_lcore_id();
1322 	qconf = &lcore_conf[lcore_id];
1323 
1324 	if (qconf->n_rx_queue == 0) {
1325 		RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
1326 		return 0;
1327 	}
1328 
1329 	RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
1330 
1331 	for (i = 0; i < qconf->n_rx_queue; i++) {
1332 		portid = qconf->rx_queue_list[i].port_id;
1333 		queueid = qconf->rx_queue_list[i].queue_id;
1334 		RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1335 			"rxqueueid=%hhu\n", lcore_id, portid, queueid);
1336 	}
1337 
1338 	/* add into event wait list */
1339 	if (event_register(qconf) == 0)
1340 		intr_en = 1;
1341 	else
1342 		RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
1343 
1344 	while (!is_done()) {
1345 		stats[lcore_id].nb_iteration_looped++;
1346 
1347 		cur_tsc = rte_rdtsc();
1348 		cur_tsc_power = cur_tsc;
1349 
1350 		/*
1351 		 * TX burst queue drain
1352 		 */
1353 		diff_tsc = cur_tsc - prev_tsc;
1354 		if (unlikely(diff_tsc > drain_tsc)) {
1355 			for (i = 0; i < qconf->n_tx_port; ++i) {
1356 				portid = qconf->tx_port_id[i];
1357 				rte_eth_tx_buffer_flush(portid,
1358 						qconf->tx_queue_id[portid],
1359 						qconf->tx_buffer[portid]);
1360 			}
1361 			prev_tsc = cur_tsc;
1362 		}
1363 
1364 		diff_tsc_power = cur_tsc_power - prev_tsc_power;
1365 		if (diff_tsc_power > tim_res_tsc) {
1366 			rte_timer_manage();
1367 			prev_tsc_power = cur_tsc_power;
1368 		}
1369 
1370 start_rx:
1371 		/*
1372 		 * Read packet from RX queues
1373 		 */
1374 		lcore_scaleup_hint = FREQ_CURRENT;
1375 		lcore_rx_idle_count = 0;
1376 		for (i = 0; i < qconf->n_rx_queue; ++i) {
1377 			rx_queue = &(qconf->rx_queue_list[i]);
1378 			rx_queue->idle_hint = 0;
1379 			portid = rx_queue->port_id;
1380 			queueid = rx_queue->queue_id;
1381 
1382 			nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1383 								MAX_PKT_BURST);
1384 
1385 			stats[lcore_id].nb_rx_processed += nb_rx;
1386 			if (unlikely(nb_rx == 0)) {
1387 				/**
1388 				 * no packet received from rx queue, try to
1389 				 * sleep for a while forcing CPU enter deeper
1390 				 * C states.
1391 				 */
1392 				rx_queue->zero_rx_packet_count++;
1393 
1394 				if (rx_queue->zero_rx_packet_count <=
1395 							MIN_ZERO_POLL_COUNT)
1396 					continue;
1397 
1398 				rx_queue->idle_hint = power_idle_heuristic(\
1399 					rx_queue->zero_rx_packet_count);
1400 				lcore_rx_idle_count++;
1401 			} else {
1402 				rx_queue->zero_rx_packet_count = 0;
1403 
1404 				/**
1405 				 * do not scale up frequency immediately as
1406 				 * user to kernel space communication is costly
1407 				 * which might impact packet I/O for received
1408 				 * packets.
1409 				 */
1410 				rx_queue->freq_up_hint =
1411 					power_freq_scaleup_heuristic(lcore_id,
1412 							portid, queueid);
1413 			}
1414 
1415 			/* Prefetch first packets */
1416 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1417 				rte_prefetch0(rte_pktmbuf_mtod(
1418 						pkts_burst[j], void *));
1419 			}
1420 
1421 			/* Prefetch and forward already prefetched packets */
1422 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1423 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1424 						j + PREFETCH_OFFSET], void *));
1425 				l3fwd_simple_forward(pkts_burst[j], portid,
1426 								qconf);
1427 			}
1428 
1429 			/* Forward remaining prefetched packets */
1430 			for (; j < nb_rx; j++) {
1431 				l3fwd_simple_forward(pkts_burst[j], portid,
1432 								qconf);
1433 			}
1434 		}
1435 
1436 		if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
1437 			for (i = 1, lcore_scaleup_hint =
1438 				qconf->rx_queue_list[0].freq_up_hint;
1439 					i < qconf->n_rx_queue; ++i) {
1440 				rx_queue = &(qconf->rx_queue_list[i]);
1441 				if (rx_queue->freq_up_hint >
1442 						lcore_scaleup_hint)
1443 					lcore_scaleup_hint =
1444 						rx_queue->freq_up_hint;
1445 			}
1446 
1447 			if (lcore_scaleup_hint == FREQ_HIGHEST) {
1448 				if (rte_power_freq_max)
1449 					rte_power_freq_max(lcore_id);
1450 			} else if (lcore_scaleup_hint == FREQ_HIGHER) {
1451 				if (rte_power_freq_up)
1452 					rte_power_freq_up(lcore_id);
1453 			}
1454 		} else {
1455 			/**
1456 			 * All Rx queues empty in recent consecutive polls,
1457 			 * sleep in a conservative manner, meaning sleep as
1458 			 * less as possible.
1459 			 */
1460 			for (i = 1, lcore_idle_hint =
1461 				qconf->rx_queue_list[0].idle_hint;
1462 					i < qconf->n_rx_queue; ++i) {
1463 				rx_queue = &(qconf->rx_queue_list[i]);
1464 				if (rx_queue->idle_hint < lcore_idle_hint)
1465 					lcore_idle_hint = rx_queue->idle_hint;
1466 			}
1467 
1468 			if (lcore_idle_hint < SUSPEND_THRESHOLD)
1469 				/**
1470 				 * execute "pause" instruction to avoid context
1471 				 * switch which generally take hundred of
1472 				 * microseconds for short sleep.
1473 				 */
1474 				rte_delay_us(lcore_idle_hint);
1475 			else {
1476 				/* suspend until rx interrupt triggers */
1477 				if (intr_en) {
1478 					turn_on_off_intr(qconf, 1);
1479 					sleep_until_rx_interrupt(
1480 							qconf->n_rx_queue,
1481 							lcore_id);
1482 					turn_on_off_intr(qconf, 0);
1483 					/**
1484 					 * start receiving packets immediately
1485 					 */
1486 					if (likely(!is_done()))
1487 						goto start_rx;
1488 				}
1489 			}
1490 			stats[lcore_id].sleep_time += lcore_idle_hint;
1491 		}
1492 	}
1493 
1494 	return 0;
1495 }
1496 
1497 static int
1498 check_lcore_params(void)
1499 {
1500 	uint8_t queue, lcore;
1501 	uint16_t i;
1502 	int socketid;
1503 
1504 	for (i = 0; i < nb_lcore_params; ++i) {
1505 		queue = lcore_params[i].queue_id;
1506 		if (queue >= MAX_RX_QUEUE_PER_PORT) {
1507 			printf("invalid queue number: %hhu\n", queue);
1508 			return -1;
1509 		}
1510 		lcore = lcore_params[i].lcore_id;
1511 		if (!rte_lcore_is_enabled(lcore)) {
1512 			printf("error: lcore %hhu is not enabled in lcore "
1513 							"mask\n", lcore);
1514 			return -1;
1515 		}
1516 		if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1517 							(numa_on == 0)) {
1518 			printf("warning: lcore %hhu is on socket %d with numa "
1519 						"off\n", lcore, socketid);
1520 		}
1521 		if (app_mode == APP_MODE_TELEMETRY && lcore == rte_lcore_id()) {
1522 			printf("cannot enable main core %d in config for telemetry mode\n",
1523 				rte_lcore_id());
1524 			return -1;
1525 		}
1526 	}
1527 	return 0;
1528 }
1529 
1530 static int
1531 check_port_config(void)
1532 {
1533 	unsigned portid;
1534 	uint16_t i;
1535 
1536 	for (i = 0; i < nb_lcore_params; ++i) {
1537 		portid = lcore_params[i].port_id;
1538 		if ((enabled_port_mask & (1 << portid)) == 0) {
1539 			printf("port %u is not enabled in port mask\n",
1540 								portid);
1541 			return -1;
1542 		}
1543 		if (!rte_eth_dev_is_valid_port(portid)) {
1544 			printf("port %u is not present on the board\n",
1545 								portid);
1546 			return -1;
1547 		}
1548 	}
1549 	return 0;
1550 }
1551 
1552 static uint8_t
1553 get_port_n_rx_queues(const uint16_t port)
1554 {
1555 	int queue = -1;
1556 	uint16_t i;
1557 
1558 	for (i = 0; i < nb_lcore_params; ++i) {
1559 		if (lcore_params[i].port_id == port &&
1560 				lcore_params[i].queue_id > queue)
1561 			queue = lcore_params[i].queue_id;
1562 	}
1563 	return (uint8_t)(++queue);
1564 }
1565 
1566 static int
1567 init_lcore_rx_queues(void)
1568 {
1569 	uint16_t i, nb_rx_queue;
1570 	uint8_t lcore;
1571 
1572 	for (i = 0; i < nb_lcore_params; ++i) {
1573 		lcore = lcore_params[i].lcore_id;
1574 		nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1575 		if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1576 			printf("error: too many queues (%u) for lcore: %u\n",
1577 				(unsigned)nb_rx_queue + 1, (unsigned)lcore);
1578 			return -1;
1579 		} else {
1580 			lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1581 				lcore_params[i].port_id;
1582 			lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1583 				lcore_params[i].queue_id;
1584 			lcore_conf[lcore].n_rx_queue++;
1585 		}
1586 	}
1587 	return 0;
1588 }
1589 
1590 /* display usage */
1591 static void
1592 print_usage(const char *prgname)
1593 {
1594 	printf ("%s [EAL options] -- -p PORTMASK -P"
1595 		"  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1596 		"  [--high-perf-cores CORELIST"
1597 		"  [--perf-config (port,queue,hi_perf,lcore_index)[,(port,queue,hi_perf,lcore_index]]"
1598 		"  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1599 		"  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1600 		"  -P : enable promiscuous mode\n"
1601 		"  --config (port,queue,lcore): rx queues configuration\n"
1602 		"  --high-perf-cores CORELIST: list of high performance cores\n"
1603 		"  --perf-config: similar as config, cores specified as indices"
1604 		" for bins containing high or regular performance cores\n"
1605 		"  --no-numa: optional, disable numa awareness\n"
1606 		"  --enable-jumbo: enable jumbo frame"
1607 		" which max packet len is PKTLEN in decimal (64-9600)\n"
1608 		"  --parse-ptype: parse packet type by software\n"
1609 		"  --legacy: use legacy interrupt-based scaling\n"
1610 		"  --empty-poll: enable empty poll detection"
1611 		" follow (training_flag, high_threshold, med_threshold)\n"
1612 		" --telemetry: enable telemetry mode, to update"
1613 		" empty polls, full polls, and core busyness to telemetry\n"
1614 		" --interrupt-only: enable interrupt-only mode\n",
1615 		prgname);
1616 }
1617 
1618 static int parse_max_pkt_len(const char *pktlen)
1619 {
1620 	char *end = NULL;
1621 	unsigned long len;
1622 
1623 	/* parse decimal string */
1624 	len = strtoul(pktlen, &end, 10);
1625 	if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1626 		return -1;
1627 
1628 	if (len == 0)
1629 		return -1;
1630 
1631 	return len;
1632 }
1633 
1634 static int
1635 parse_portmask(const char *portmask)
1636 {
1637 	char *end = NULL;
1638 	unsigned long pm;
1639 
1640 	/* parse hexadecimal string */
1641 	pm = strtoul(portmask, &end, 16);
1642 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1643 		return 0;
1644 
1645 	return pm;
1646 }
1647 
1648 static int
1649 parse_config(const char *q_arg)
1650 {
1651 	char s[256];
1652 	const char *p, *p0 = q_arg;
1653 	char *end;
1654 	enum fieldnames {
1655 		FLD_PORT = 0,
1656 		FLD_QUEUE,
1657 		FLD_LCORE,
1658 		_NUM_FLD
1659 	};
1660 	unsigned long int_fld[_NUM_FLD];
1661 	char *str_fld[_NUM_FLD];
1662 	int i;
1663 	unsigned size;
1664 
1665 	nb_lcore_params = 0;
1666 
1667 	while ((p = strchr(p0,'(')) != NULL) {
1668 		++p;
1669 		if((p0 = strchr(p,')')) == NULL)
1670 			return -1;
1671 
1672 		size = p0 - p;
1673 		if(size >= sizeof(s))
1674 			return -1;
1675 
1676 		snprintf(s, sizeof(s), "%.*s", size, p);
1677 		if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1678 								_NUM_FLD)
1679 			return -1;
1680 		for (i = 0; i < _NUM_FLD; i++){
1681 			errno = 0;
1682 			int_fld[i] = strtoul(str_fld[i], &end, 0);
1683 			if (errno != 0 || end == str_fld[i] || int_fld[i] >
1684 									255)
1685 				return -1;
1686 		}
1687 		if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1688 			printf("exceeded max number of lcore params: %hu\n",
1689 				nb_lcore_params);
1690 			return -1;
1691 		}
1692 		lcore_params_array[nb_lcore_params].port_id =
1693 				(uint8_t)int_fld[FLD_PORT];
1694 		lcore_params_array[nb_lcore_params].queue_id =
1695 				(uint8_t)int_fld[FLD_QUEUE];
1696 		lcore_params_array[nb_lcore_params].lcore_id =
1697 				(uint8_t)int_fld[FLD_LCORE];
1698 		++nb_lcore_params;
1699 	}
1700 	lcore_params = lcore_params_array;
1701 
1702 	return 0;
1703 }
1704 static int
1705 parse_ep_config(const char *q_arg)
1706 {
1707 	char s[256];
1708 	const char *p = q_arg;
1709 	char *end;
1710 	int  num_arg;
1711 
1712 	char *str_fld[3];
1713 
1714 	int training_flag;
1715 	int med_edpi;
1716 	int hgh_edpi;
1717 
1718 	ep_med_edpi = EMPTY_POLL_MED_THRESHOLD;
1719 	ep_hgh_edpi = EMPTY_POLL_MED_THRESHOLD;
1720 
1721 	strlcpy(s, p, sizeof(s));
1722 
1723 	num_arg = rte_strsplit(s, sizeof(s), str_fld, 3, ',');
1724 
1725 	empty_poll_train = false;
1726 
1727 	if (num_arg == 0)
1728 		return 0;
1729 
1730 	if (num_arg == 3) {
1731 
1732 		training_flag = strtoul(str_fld[0], &end, 0);
1733 		med_edpi = strtoul(str_fld[1], &end, 0);
1734 		hgh_edpi = strtoul(str_fld[2], &end, 0);
1735 
1736 		if (training_flag == 1)
1737 			empty_poll_train = true;
1738 
1739 		if (med_edpi > 0)
1740 			ep_med_edpi = med_edpi;
1741 
1742 		if (med_edpi > 0)
1743 			ep_hgh_edpi = hgh_edpi;
1744 
1745 	} else {
1746 
1747 		return -1;
1748 	}
1749 
1750 	return 0;
1751 
1752 }
1753 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1754 #define CMD_LINE_OPT_LEGACY "legacy"
1755 #define CMD_LINE_OPT_EMPTY_POLL "empty-poll"
1756 #define CMD_LINE_OPT_INTERRUPT_ONLY "interrupt-only"
1757 #define CMD_LINE_OPT_TELEMETRY "telemetry"
1758 
1759 /* Parse the argument given in the command line of the application */
1760 static int
1761 parse_args(int argc, char **argv)
1762 {
1763 	int opt, ret;
1764 	char **argvopt;
1765 	int option_index;
1766 	uint32_t limit;
1767 	char *prgname = argv[0];
1768 	static struct option lgopts[] = {
1769 		{"config", 1, 0, 0},
1770 		{"perf-config", 1, 0, 0},
1771 		{"high-perf-cores", 1, 0, 0},
1772 		{"no-numa", 0, 0, 0},
1773 		{"enable-jumbo", 0, 0, 0},
1774 		{CMD_LINE_OPT_EMPTY_POLL, 1, 0, 0},
1775 		{CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
1776 		{CMD_LINE_OPT_LEGACY, 0, 0, 0},
1777 		{CMD_LINE_OPT_TELEMETRY, 0, 0, 0},
1778 		{CMD_LINE_OPT_INTERRUPT_ONLY, 0, 0, 0},
1779 		{NULL, 0, 0, 0}
1780 	};
1781 
1782 	argvopt = argv;
1783 
1784 	while ((opt = getopt_long(argc, argvopt, "p:l:m:h:P",
1785 				lgopts, &option_index)) != EOF) {
1786 
1787 		switch (opt) {
1788 		/* portmask */
1789 		case 'p':
1790 			enabled_port_mask = parse_portmask(optarg);
1791 			if (enabled_port_mask == 0) {
1792 				printf("invalid portmask\n");
1793 				print_usage(prgname);
1794 				return -1;
1795 			}
1796 			break;
1797 		case 'P':
1798 			printf("Promiscuous mode selected\n");
1799 			promiscuous_on = 1;
1800 			break;
1801 		case 'l':
1802 			limit = parse_max_pkt_len(optarg);
1803 			freq_tlb[LOW] = limit;
1804 			break;
1805 		case 'm':
1806 			limit = parse_max_pkt_len(optarg);
1807 			freq_tlb[MED] = limit;
1808 			break;
1809 		case 'h':
1810 			limit = parse_max_pkt_len(optarg);
1811 			freq_tlb[HGH] = limit;
1812 			break;
1813 		/* long options */
1814 		case 0:
1815 			if (!strncmp(lgopts[option_index].name, "config", 6)) {
1816 				ret = parse_config(optarg);
1817 				if (ret) {
1818 					printf("invalid config\n");
1819 					print_usage(prgname);
1820 					return -1;
1821 				}
1822 			}
1823 
1824 			if (!strncmp(lgopts[option_index].name,
1825 					"perf-config", 11)) {
1826 				ret = parse_perf_config(optarg);
1827 				if (ret) {
1828 					printf("invalid perf-config\n");
1829 					print_usage(prgname);
1830 					return -1;
1831 				}
1832 			}
1833 
1834 			if (!strncmp(lgopts[option_index].name,
1835 					"high-perf-cores", 15)) {
1836 				ret = parse_perf_core_list(optarg);
1837 				if (ret) {
1838 					printf("invalid high-perf-cores\n");
1839 					print_usage(prgname);
1840 					return -1;
1841 				}
1842 			}
1843 
1844 			if (!strncmp(lgopts[option_index].name,
1845 						"no-numa", 7)) {
1846 				printf("numa is disabled \n");
1847 				numa_on = 0;
1848 			}
1849 
1850 			if (!strncmp(lgopts[option_index].name,
1851 					CMD_LINE_OPT_LEGACY,
1852 					sizeof(CMD_LINE_OPT_LEGACY))) {
1853 				if (app_mode != APP_MODE_DEFAULT) {
1854 					printf(" legacy mode is mutually exclusive with other modes\n");
1855 					return -1;
1856 				}
1857 				app_mode = APP_MODE_LEGACY;
1858 				printf("legacy mode is enabled\n");
1859 			}
1860 
1861 			if (!strncmp(lgopts[option_index].name,
1862 					CMD_LINE_OPT_EMPTY_POLL, 10)) {
1863 				if (app_mode != APP_MODE_DEFAULT) {
1864 					printf(" empty-poll mode is mutually exclusive with other modes\n");
1865 					return -1;
1866 				}
1867 				app_mode = APP_MODE_EMPTY_POLL;
1868 				ret = parse_ep_config(optarg);
1869 
1870 				if (ret) {
1871 					printf("invalid empty poll config\n");
1872 					print_usage(prgname);
1873 					return -1;
1874 				}
1875 				printf("empty-poll is enabled\n");
1876 			}
1877 
1878 			if (!strncmp(lgopts[option_index].name,
1879 					CMD_LINE_OPT_TELEMETRY,
1880 					sizeof(CMD_LINE_OPT_TELEMETRY))) {
1881 				if (app_mode != APP_MODE_DEFAULT) {
1882 					printf(" telemetry mode is mutually exclusive with other modes\n");
1883 					return -1;
1884 				}
1885 				app_mode = APP_MODE_TELEMETRY;
1886 				printf("telemetry mode is enabled\n");
1887 			}
1888 
1889 			if (!strncmp(lgopts[option_index].name,
1890 					CMD_LINE_OPT_INTERRUPT_ONLY,
1891 					sizeof(CMD_LINE_OPT_INTERRUPT_ONLY))) {
1892 				if (app_mode != APP_MODE_DEFAULT) {
1893 					printf(" interrupt-only mode is mutually exclusive with other modes\n");
1894 					return -1;
1895 				}
1896 				app_mode = APP_MODE_INTERRUPT;
1897 				printf("interrupt-only mode is enabled\n");
1898 			}
1899 
1900 			if (!strncmp(lgopts[option_index].name,
1901 					"enable-jumbo", 12)) {
1902 				struct option lenopts =
1903 					{"max-pkt-len", required_argument, \
1904 									0, 0};
1905 
1906 				printf("jumbo frame is enabled \n");
1907 				port_conf.rxmode.offloads |=
1908 						DEV_RX_OFFLOAD_JUMBO_FRAME;
1909 				port_conf.txmode.offloads |=
1910 						DEV_TX_OFFLOAD_MULTI_SEGS;
1911 
1912 				/**
1913 				 * if no max-pkt-len set, use the default value
1914 				 * RTE_ETHER_MAX_LEN
1915 				 */
1916 				if (0 == getopt_long(argc, argvopt, "",
1917 						&lenopts, &option_index)) {
1918 					ret = parse_max_pkt_len(optarg);
1919 					if ((ret < 64) ||
1920 						(ret > MAX_JUMBO_PKT_LEN)){
1921 						printf("invalid packet "
1922 								"length\n");
1923 						print_usage(prgname);
1924 						return -1;
1925 					}
1926 					port_conf.rxmode.max_rx_pkt_len = ret;
1927 				}
1928 				printf("set jumbo frame "
1929 					"max packet length to %u\n",
1930 				(unsigned int)port_conf.rxmode.max_rx_pkt_len);
1931 			}
1932 
1933 			if (!strncmp(lgopts[option_index].name,
1934 				     CMD_LINE_OPT_PARSE_PTYPE,
1935 				     sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
1936 				printf("soft parse-ptype is enabled\n");
1937 				parse_ptype = 1;
1938 			}
1939 
1940 			break;
1941 
1942 		default:
1943 			print_usage(prgname);
1944 			return -1;
1945 		}
1946 	}
1947 
1948 	if (optind >= 0)
1949 		argv[optind-1] = prgname;
1950 
1951 	ret = optind-1;
1952 	optind = 1; /* reset getopt lib */
1953 	return ret;
1954 }
1955 
1956 static void
1957 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1958 {
1959 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
1960 	rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
1961 	printf("%s%s", name, buf);
1962 }
1963 
1964 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1965 static void
1966 setup_hash(int socketid)
1967 {
1968 	struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1969 		.name = NULL,
1970 		.entries = L3FWD_HASH_ENTRIES,
1971 		.key_len = sizeof(struct ipv4_5tuple),
1972 		.hash_func = DEFAULT_HASH_FUNC,
1973 		.hash_func_init_val = 0,
1974 	};
1975 
1976 	struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1977 		.name = NULL,
1978 		.entries = L3FWD_HASH_ENTRIES,
1979 		.key_len = sizeof(struct ipv6_5tuple),
1980 		.hash_func = DEFAULT_HASH_FUNC,
1981 		.hash_func_init_val = 0,
1982 	};
1983 
1984 	unsigned i;
1985 	int ret;
1986 	char s[64];
1987 
1988 	/* create ipv4 hash */
1989 	snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1990 	ipv4_l3fwd_hash_params.name = s;
1991 	ipv4_l3fwd_hash_params.socket_id = socketid;
1992 	ipv4_l3fwd_lookup_struct[socketid] =
1993 		rte_hash_create(&ipv4_l3fwd_hash_params);
1994 	if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1995 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1996 				"socket %d\n", socketid);
1997 
1998 	/* create ipv6 hash */
1999 	snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
2000 	ipv6_l3fwd_hash_params.name = s;
2001 	ipv6_l3fwd_hash_params.socket_id = socketid;
2002 	ipv6_l3fwd_lookup_struct[socketid] =
2003 		rte_hash_create(&ipv6_l3fwd_hash_params);
2004 	if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
2005 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2006 				"socket %d\n", socketid);
2007 
2008 
2009 	/* populate the ipv4 hash */
2010 	for (i = 0; i < RTE_DIM(ipv4_l3fwd_route_array); i++) {
2011 		ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
2012 				(void *) &ipv4_l3fwd_route_array[i].key);
2013 		if (ret < 0) {
2014 			rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
2015 				"l3fwd hash on socket %d\n", i, socketid);
2016 		}
2017 		ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
2018 		printf("Hash: Adding key\n");
2019 		print_ipv4_key(ipv4_l3fwd_route_array[i].key);
2020 	}
2021 
2022 	/* populate the ipv6 hash */
2023 	for (i = 0; i < RTE_DIM(ipv6_l3fwd_route_array); i++) {
2024 		ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
2025 				(void *) &ipv6_l3fwd_route_array[i].key);
2026 		if (ret < 0) {
2027 			rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
2028 				"l3fwd hash on socket %d\n", i, socketid);
2029 		}
2030 		ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
2031 		printf("Hash: Adding key\n");
2032 		print_ipv6_key(ipv6_l3fwd_route_array[i].key);
2033 	}
2034 }
2035 #endif
2036 
2037 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2038 static void
2039 setup_lpm(int socketid)
2040 {
2041 	unsigned i;
2042 	int ret;
2043 	char s[64];
2044 
2045 	/* create the LPM table */
2046 	struct rte_lpm_config lpm_ipv4_config;
2047 
2048 	lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
2049 	lpm_ipv4_config.number_tbl8s = 256;
2050 	lpm_ipv4_config.flags = 0;
2051 
2052 	snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
2053 	ipv4_l3fwd_lookup_struct[socketid] =
2054 			rte_lpm_create(s, socketid, &lpm_ipv4_config);
2055 	if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2056 		rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
2057 				" on socket %d\n", socketid);
2058 
2059 	/* populate the LPM table */
2060 	for (i = 0; i < RTE_DIM(ipv4_l3fwd_route_array); i++) {
2061 		ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
2062 			ipv4_l3fwd_route_array[i].ip,
2063 			ipv4_l3fwd_route_array[i].depth,
2064 			ipv4_l3fwd_route_array[i].if_out);
2065 
2066 		if (ret < 0) {
2067 			rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
2068 				"l3fwd LPM table on socket %d\n",
2069 				i, socketid);
2070 		}
2071 
2072 		printf("LPM: Adding route 0x%08x / %d (%d)\n",
2073 			(unsigned)ipv4_l3fwd_route_array[i].ip,
2074 			ipv4_l3fwd_route_array[i].depth,
2075 			ipv4_l3fwd_route_array[i].if_out);
2076 	}
2077 }
2078 #endif
2079 
2080 static int
2081 init_mem(unsigned nb_mbuf)
2082 {
2083 	struct lcore_conf *qconf;
2084 	int socketid;
2085 	unsigned lcore_id;
2086 	char s[64];
2087 
2088 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2089 		if (rte_lcore_is_enabled(lcore_id) == 0)
2090 			continue;
2091 
2092 		if (numa_on)
2093 			socketid = rte_lcore_to_socket_id(lcore_id);
2094 		else
2095 			socketid = 0;
2096 
2097 		if (socketid >= NB_SOCKETS) {
2098 			rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
2099 					"out of range %d\n", socketid,
2100 						lcore_id, NB_SOCKETS);
2101 		}
2102 		if (pktmbuf_pool[socketid] == NULL) {
2103 			snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
2104 			pktmbuf_pool[socketid] =
2105 				rte_pktmbuf_pool_create(s, nb_mbuf,
2106 					MEMPOOL_CACHE_SIZE, 0,
2107 					RTE_MBUF_DEFAULT_BUF_SIZE,
2108 					socketid);
2109 			if (pktmbuf_pool[socketid] == NULL)
2110 				rte_exit(EXIT_FAILURE,
2111 					"Cannot init mbuf pool on socket %d\n",
2112 								socketid);
2113 			else
2114 				printf("Allocated mbuf pool on socket %d\n",
2115 								socketid);
2116 
2117 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2118 			setup_lpm(socketid);
2119 #else
2120 			setup_hash(socketid);
2121 #endif
2122 		}
2123 		qconf = &lcore_conf[lcore_id];
2124 		qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
2125 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2126 		qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
2127 #endif
2128 	}
2129 	return 0;
2130 }
2131 
2132 /* Check the link status of all ports in up to 9s, and print them finally */
2133 static void
2134 check_all_ports_link_status(uint32_t port_mask)
2135 {
2136 #define CHECK_INTERVAL 100 /* 100ms */
2137 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
2138 	uint8_t count, all_ports_up, print_flag = 0;
2139 	uint16_t portid;
2140 	struct rte_eth_link link;
2141 	int ret;
2142 	char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
2143 
2144 	printf("\nChecking link status");
2145 	fflush(stdout);
2146 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
2147 		all_ports_up = 1;
2148 		RTE_ETH_FOREACH_DEV(portid) {
2149 			if ((port_mask & (1 << portid)) == 0)
2150 				continue;
2151 			memset(&link, 0, sizeof(link));
2152 			ret = rte_eth_link_get_nowait(portid, &link);
2153 			if (ret < 0) {
2154 				all_ports_up = 0;
2155 				if (print_flag == 1)
2156 					printf("Port %u link get failed: %s\n",
2157 						portid, rte_strerror(-ret));
2158 				continue;
2159 			}
2160 			/* print link status if flag set */
2161 			if (print_flag == 1) {
2162 				rte_eth_link_to_str(link_status_text,
2163 					sizeof(link_status_text), &link);
2164 				printf("Port %d %s\n", portid,
2165 				       link_status_text);
2166 				continue;
2167 			}
2168 			/* clear all_ports_up flag if any link down */
2169 			if (link.link_status == ETH_LINK_DOWN) {
2170 				all_ports_up = 0;
2171 				break;
2172 			}
2173 		}
2174 		/* after finally printing all link status, get out */
2175 		if (print_flag == 1)
2176 			break;
2177 
2178 		if (all_ports_up == 0) {
2179 			printf(".");
2180 			fflush(stdout);
2181 			rte_delay_ms(CHECK_INTERVAL);
2182 		}
2183 
2184 		/* set the print_flag if all ports up or timeout */
2185 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
2186 			print_flag = 1;
2187 			printf("done\n");
2188 		}
2189 	}
2190 }
2191 
2192 static int check_ptype(uint16_t portid)
2193 {
2194 	int i, ret;
2195 	int ptype_l3_ipv4 = 0;
2196 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2197 	int ptype_l3_ipv6 = 0;
2198 #endif
2199 	uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
2200 
2201 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
2202 	if (ret <= 0)
2203 		return 0;
2204 
2205 	uint32_t ptypes[ret];
2206 
2207 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
2208 	for (i = 0; i < ret; ++i) {
2209 		if (ptypes[i] & RTE_PTYPE_L3_IPV4)
2210 			ptype_l3_ipv4 = 1;
2211 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2212 		if (ptypes[i] & RTE_PTYPE_L3_IPV6)
2213 			ptype_l3_ipv6 = 1;
2214 #endif
2215 	}
2216 
2217 	if (ptype_l3_ipv4 == 0)
2218 		printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
2219 
2220 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2221 	if (ptype_l3_ipv6 == 0)
2222 		printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
2223 #endif
2224 
2225 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2226 	if (ptype_l3_ipv4)
2227 #else /* APP_LOOKUP_EXACT_MATCH */
2228 	if (ptype_l3_ipv4 && ptype_l3_ipv6)
2229 #endif
2230 		return 1;
2231 
2232 	return 0;
2233 
2234 }
2235 
2236 static int
2237 init_power_library(void)
2238 {
2239 	enum power_management_env env;
2240 	unsigned int lcore_id;
2241 	int ret = 0;
2242 
2243 	RTE_LCORE_FOREACH(lcore_id) {
2244 		/* init power management library */
2245 		ret = rte_power_init(lcore_id);
2246 		if (ret) {
2247 			RTE_LOG(ERR, POWER,
2248 				"Library initialization failed on core %u\n",
2249 				lcore_id);
2250 			return ret;
2251 		}
2252 		/* we're not supporting the VM channel mode */
2253 		env = rte_power_get_env();
2254 		if (env != PM_ENV_ACPI_CPUFREQ &&
2255 				env != PM_ENV_PSTATE_CPUFREQ) {
2256 			RTE_LOG(ERR, POWER,
2257 				"Only ACPI and PSTATE mode are supported\n");
2258 			return -1;
2259 		}
2260 	}
2261 	return ret;
2262 }
2263 
2264 static int
2265 deinit_power_library(void)
2266 {
2267 	unsigned int lcore_id;
2268 	int ret = 0;
2269 
2270 	RTE_LCORE_FOREACH(lcore_id) {
2271 		/* deinit power management library */
2272 		ret = rte_power_exit(lcore_id);
2273 		if (ret) {
2274 			RTE_LOG(ERR, POWER,
2275 				"Library deinitialization failed on core %u\n",
2276 				lcore_id);
2277 			return ret;
2278 		}
2279 	}
2280 	return ret;
2281 }
2282 
2283 static void
2284 get_current_stat_values(uint64_t *values)
2285 {
2286 	unsigned int lcore_id = rte_lcore_id();
2287 	struct lcore_conf *qconf;
2288 	uint64_t app_eps = 0, app_fps = 0, app_br = 0;
2289 	uint64_t count = 0;
2290 
2291 	RTE_LCORE_FOREACH_WORKER(lcore_id) {
2292 		qconf = &lcore_conf[lcore_id];
2293 		if (qconf->n_rx_queue == 0)
2294 			continue;
2295 		count++;
2296 		rte_spinlock_lock(&stats[lcore_id].telemetry_lock);
2297 		app_eps += stats[lcore_id].ep_nep[1];
2298 		app_fps += stats[lcore_id].fp_nfp[1];
2299 		app_br += stats[lcore_id].br;
2300 		rte_spinlock_unlock(&stats[lcore_id].telemetry_lock);
2301 	}
2302 
2303 	if (count > 0) {
2304 		values[0] = app_eps/count;
2305 		values[1] = app_fps/count;
2306 		values[2] = app_br/count;
2307 	} else
2308 		memset(values, 0, sizeof(uint64_t) * NUM_TELSTATS);
2309 
2310 }
2311 
2312 static void
2313 update_telemetry(__rte_unused struct rte_timer *tim,
2314 		__rte_unused void *arg)
2315 {
2316 	int ret;
2317 	uint64_t values[NUM_TELSTATS] = {0};
2318 
2319 	get_current_stat_values(values);
2320 	ret = rte_metrics_update_values(RTE_METRICS_GLOBAL, telstats_index,
2321 					values, RTE_DIM(values));
2322 	if (ret < 0)
2323 		RTE_LOG(WARNING, POWER, "failed to update metrcis\n");
2324 }
2325 
2326 static int
2327 handle_app_stats(const char *cmd __rte_unused,
2328 		const char *params __rte_unused,
2329 		struct rte_tel_data *d)
2330 {
2331 	uint64_t values[NUM_TELSTATS] = {0};
2332 	uint32_t i;
2333 
2334 	rte_tel_data_start_dict(d);
2335 	get_current_stat_values(values);
2336 	for (i = 0; i < NUM_TELSTATS; i++)
2337 		rte_tel_data_add_dict_u64(d, telstats_strings[i].name,
2338 				values[i]);
2339 	return 0;
2340 }
2341 
2342 static void
2343 telemetry_setup_timer(void)
2344 {
2345 	int lcore_id = rte_lcore_id();
2346 	uint64_t hz = rte_get_timer_hz();
2347 	uint64_t ticks;
2348 
2349 	ticks = hz / TELEMETRY_INTERVALS_PER_SEC;
2350 	rte_timer_reset_sync(&telemetry_timer,
2351 			ticks,
2352 			PERIODICAL,
2353 			lcore_id,
2354 			update_telemetry,
2355 			NULL);
2356 }
2357 static void
2358 empty_poll_setup_timer(void)
2359 {
2360 	int lcore_id = rte_lcore_id();
2361 	uint64_t hz = rte_get_timer_hz();
2362 
2363 	struct  ep_params *ep_ptr = ep_params;
2364 
2365 	ep_ptr->interval_ticks = hz / INTERVALS_PER_SECOND;
2366 
2367 	rte_timer_reset_sync(&ep_ptr->timer0,
2368 			ep_ptr->interval_ticks,
2369 			PERIODICAL,
2370 			lcore_id,
2371 			rte_empty_poll_detection,
2372 			(void *)ep_ptr);
2373 
2374 }
2375 static int
2376 launch_timer(unsigned int lcore_id)
2377 {
2378 	int64_t prev_tsc = 0, cur_tsc, diff_tsc, cycles_10ms;
2379 
2380 	RTE_SET_USED(lcore_id);
2381 
2382 
2383 	if (rte_get_main_lcore() != lcore_id) {
2384 		rte_panic("timer on lcore:%d which is not main core:%d\n",
2385 				lcore_id,
2386 				rte_get_main_lcore());
2387 	}
2388 
2389 	RTE_LOG(INFO, POWER, "Bring up the Timer\n");
2390 
2391 	if (app_mode == APP_MODE_EMPTY_POLL)
2392 		empty_poll_setup_timer();
2393 	else
2394 		telemetry_setup_timer();
2395 
2396 	cycles_10ms = rte_get_timer_hz() / 100;
2397 
2398 	while (!is_done()) {
2399 		cur_tsc = rte_rdtsc();
2400 		diff_tsc = cur_tsc - prev_tsc;
2401 		if (diff_tsc > cycles_10ms) {
2402 			rte_timer_manage();
2403 			prev_tsc = cur_tsc;
2404 			cycles_10ms = rte_get_timer_hz() / 100;
2405 		}
2406 	}
2407 
2408 	RTE_LOG(INFO, POWER, "Timer_subsystem is done\n");
2409 
2410 	return 0;
2411 }
2412 
2413 static int
2414 autodetect_mode(void)
2415 {
2416 	RTE_LOG(NOTICE, L3FWD_POWER, "Operating mode not specified, probing frequency scaling support...\n");
2417 
2418 	/*
2419 	 * Empty poll and telemetry modes have to be specifically requested to
2420 	 * be enabled, but we can auto-detect between interrupt mode with or
2421 	 * without frequency scaling. Both ACPI and pstate can be used.
2422 	 */
2423 	if (rte_power_check_env_supported(PM_ENV_ACPI_CPUFREQ))
2424 		return APP_MODE_LEGACY;
2425 	if (rte_power_check_env_supported(PM_ENV_PSTATE_CPUFREQ))
2426 		return APP_MODE_LEGACY;
2427 
2428 	RTE_LOG(NOTICE, L3FWD_POWER, "Frequency scaling not supported, selecting interrupt-only mode\n");
2429 
2430 	return APP_MODE_INTERRUPT;
2431 }
2432 
2433 static const char *
2434 mode_to_str(enum appmode mode)
2435 {
2436 	switch (mode) {
2437 	case APP_MODE_LEGACY:
2438 		return "legacy";
2439 	case APP_MODE_EMPTY_POLL:
2440 		return "empty poll";
2441 	case APP_MODE_TELEMETRY:
2442 		return "telemetry";
2443 	case APP_MODE_INTERRUPT:
2444 		return "interrupt-only";
2445 	default:
2446 		return "invalid";
2447 	}
2448 }
2449 
2450 int
2451 main(int argc, char **argv)
2452 {
2453 	struct lcore_conf *qconf;
2454 	struct rte_eth_dev_info dev_info;
2455 	struct rte_eth_txconf *txconf;
2456 	int ret;
2457 	uint16_t nb_ports;
2458 	uint16_t queueid;
2459 	unsigned lcore_id;
2460 	uint64_t hz;
2461 	uint32_t n_tx_queue, nb_lcores;
2462 	uint32_t dev_rxq_num, dev_txq_num;
2463 	uint8_t nb_rx_queue, queue, socketid;
2464 	uint16_t portid;
2465 	const char *ptr_strings[NUM_TELSTATS];
2466 
2467 	/* catch SIGINT and restore cpufreq governor to ondemand */
2468 	signal(SIGINT, signal_exit_now);
2469 
2470 	/* init EAL */
2471 	ret = rte_eal_init(argc, argv);
2472 	if (ret < 0)
2473 		rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2474 	argc -= ret;
2475 	argv += ret;
2476 
2477 	/* init RTE timer library to be used late */
2478 	rte_timer_subsystem_init();
2479 
2480 	/* parse application arguments (after the EAL ones) */
2481 	ret = parse_args(argc, argv);
2482 	if (ret < 0)
2483 		rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
2484 
2485 	if (app_mode == APP_MODE_DEFAULT)
2486 		app_mode = autodetect_mode();
2487 
2488 	RTE_LOG(INFO, L3FWD_POWER, "Selected operation mode: %s\n",
2489 			mode_to_str(app_mode));
2490 
2491 	/* only legacy and empty poll mode rely on power library */
2492 	if ((app_mode == APP_MODE_LEGACY || app_mode == APP_MODE_EMPTY_POLL) &&
2493 			init_power_library())
2494 		rte_exit(EXIT_FAILURE, "init_power_library failed\n");
2495 
2496 	if (update_lcore_params() < 0)
2497 		rte_exit(EXIT_FAILURE, "update_lcore_params failed\n");
2498 
2499 	if (check_lcore_params() < 0)
2500 		rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
2501 
2502 	ret = init_lcore_rx_queues();
2503 	if (ret < 0)
2504 		rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2505 
2506 	nb_ports = rte_eth_dev_count_avail();
2507 
2508 	if (check_port_config() < 0)
2509 		rte_exit(EXIT_FAILURE, "check_port_config failed\n");
2510 
2511 	nb_lcores = rte_lcore_count();
2512 
2513 	/* initialize all ports */
2514 	RTE_ETH_FOREACH_DEV(portid) {
2515 		struct rte_eth_conf local_port_conf = port_conf;
2516 		/* not all app modes need interrupts */
2517 		bool need_intr = app_mode == APP_MODE_LEGACY ||
2518 				app_mode == APP_MODE_INTERRUPT;
2519 
2520 		/* skip ports that are not enabled */
2521 		if ((enabled_port_mask & (1 << portid)) == 0) {
2522 			printf("\nSkipping disabled port %d\n", portid);
2523 			continue;
2524 		}
2525 
2526 		/* init port */
2527 		printf("Initializing port %d ... ", portid );
2528 		fflush(stdout);
2529 
2530 		ret = rte_eth_dev_info_get(portid, &dev_info);
2531 		if (ret != 0)
2532 			rte_exit(EXIT_FAILURE,
2533 				"Error during getting device (port %u) info: %s\n",
2534 				portid, strerror(-ret));
2535 
2536 		dev_rxq_num = dev_info.max_rx_queues;
2537 		dev_txq_num = dev_info.max_tx_queues;
2538 
2539 		nb_rx_queue = get_port_n_rx_queues(portid);
2540 		if (nb_rx_queue > dev_rxq_num)
2541 			rte_exit(EXIT_FAILURE,
2542 				"Cannot configure not existed rxq: "
2543 				"port=%d\n", portid);
2544 
2545 		n_tx_queue = nb_lcores;
2546 		if (n_tx_queue > dev_txq_num)
2547 			n_tx_queue = dev_txq_num;
2548 		printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
2549 			nb_rx_queue, (unsigned)n_tx_queue );
2550 		/* If number of Rx queue is 0, no need to enable Rx interrupt */
2551 		if (nb_rx_queue == 0)
2552 			need_intr = false;
2553 
2554 		if (need_intr)
2555 			local_port_conf.intr_conf.rxq = 1;
2556 
2557 		ret = rte_eth_dev_info_get(portid, &dev_info);
2558 		if (ret != 0)
2559 			rte_exit(EXIT_FAILURE,
2560 				"Error during getting device (port %u) info: %s\n",
2561 				portid, strerror(-ret));
2562 
2563 		if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
2564 			local_port_conf.txmode.offloads |=
2565 				DEV_TX_OFFLOAD_MBUF_FAST_FREE;
2566 
2567 		local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2568 			dev_info.flow_type_rss_offloads;
2569 		if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2570 				port_conf.rx_adv_conf.rss_conf.rss_hf) {
2571 			printf("Port %u modified RSS hash function based on hardware support,"
2572 				"requested:%#"PRIx64" configured:%#"PRIx64"\n",
2573 				portid,
2574 				port_conf.rx_adv_conf.rss_conf.rss_hf,
2575 				local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2576 		}
2577 
2578 		ret = rte_eth_dev_configure(portid, nb_rx_queue,
2579 					(uint16_t)n_tx_queue, &local_port_conf);
2580 		if (ret < 0)
2581 			rte_exit(EXIT_FAILURE, "Cannot configure device: "
2582 					"err=%d, port=%d\n", ret, portid);
2583 
2584 		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
2585 						       &nb_txd);
2586 		if (ret < 0)
2587 			rte_exit(EXIT_FAILURE,
2588 				 "Cannot adjust number of descriptors: err=%d, port=%d\n",
2589 				 ret, portid);
2590 
2591 		ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
2592 		if (ret < 0)
2593 			rte_exit(EXIT_FAILURE,
2594 				 "Cannot get MAC address: err=%d, port=%d\n",
2595 				 ret, portid);
2596 
2597 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
2598 		printf(", ");
2599 
2600 		/* init memory */
2601 		ret = init_mem(NB_MBUF);
2602 		if (ret < 0)
2603 			rte_exit(EXIT_FAILURE, "init_mem failed\n");
2604 
2605 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2606 			if (rte_lcore_is_enabled(lcore_id) == 0)
2607 				continue;
2608 
2609 			/* Initialize TX buffers */
2610 			qconf = &lcore_conf[lcore_id];
2611 			qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
2612 				RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
2613 				rte_eth_dev_socket_id(portid));
2614 			if (qconf->tx_buffer[portid] == NULL)
2615 				rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
2616 						 portid);
2617 
2618 			rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
2619 		}
2620 
2621 		/* init one TX queue per couple (lcore,port) */
2622 		queueid = 0;
2623 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2624 			if (rte_lcore_is_enabled(lcore_id) == 0)
2625 				continue;
2626 
2627 			if (queueid >= dev_txq_num)
2628 				continue;
2629 
2630 			if (numa_on)
2631 				socketid = \
2632 				(uint8_t)rte_lcore_to_socket_id(lcore_id);
2633 			else
2634 				socketid = 0;
2635 
2636 			printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
2637 			fflush(stdout);
2638 
2639 			txconf = &dev_info.default_txconf;
2640 			txconf->offloads = local_port_conf.txmode.offloads;
2641 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
2642 						     socketid, txconf);
2643 			if (ret < 0)
2644 				rte_exit(EXIT_FAILURE,
2645 					"rte_eth_tx_queue_setup: err=%d, "
2646 						"port=%d\n", ret, portid);
2647 
2648 			qconf = &lcore_conf[lcore_id];
2649 			qconf->tx_queue_id[portid] = queueid;
2650 			queueid++;
2651 
2652 			qconf->tx_port_id[qconf->n_tx_port] = portid;
2653 			qconf->n_tx_port++;
2654 		}
2655 		printf("\n");
2656 	}
2657 
2658 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2659 		if (rte_lcore_is_enabled(lcore_id) == 0)
2660 			continue;
2661 
2662 		if (app_mode == APP_MODE_LEGACY) {
2663 			/* init timer structures for each enabled lcore */
2664 			rte_timer_init(&power_timers[lcore_id]);
2665 			hz = rte_get_timer_hz();
2666 			rte_timer_reset(&power_timers[lcore_id],
2667 					hz/TIMER_NUMBER_PER_SECOND,
2668 					SINGLE, lcore_id,
2669 					power_timer_cb, NULL);
2670 		}
2671 		qconf = &lcore_conf[lcore_id];
2672 		printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
2673 		fflush(stdout);
2674 		/* init RX queues */
2675 		for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
2676 			struct rte_eth_rxconf rxq_conf;
2677 
2678 			portid = qconf->rx_queue_list[queue].port_id;
2679 			queueid = qconf->rx_queue_list[queue].queue_id;
2680 
2681 			if (numa_on)
2682 				socketid = \
2683 				(uint8_t)rte_lcore_to_socket_id(lcore_id);
2684 			else
2685 				socketid = 0;
2686 
2687 			printf("rxq=%d,%d,%d ", portid, queueid, socketid);
2688 			fflush(stdout);
2689 
2690 			ret = rte_eth_dev_info_get(portid, &dev_info);
2691 			if (ret != 0)
2692 				rte_exit(EXIT_FAILURE,
2693 					"Error during getting device (port %u) info: %s\n",
2694 					portid, strerror(-ret));
2695 
2696 			rxq_conf = dev_info.default_rxconf;
2697 			rxq_conf.offloads = port_conf.rxmode.offloads;
2698 			ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
2699 				socketid, &rxq_conf,
2700 				pktmbuf_pool[socketid]);
2701 			if (ret < 0)
2702 				rte_exit(EXIT_FAILURE,
2703 					"rte_eth_rx_queue_setup: err=%d, "
2704 						"port=%d\n", ret, portid);
2705 
2706 			if (parse_ptype) {
2707 				if (add_cb_parse_ptype(portid, queueid) < 0)
2708 					rte_exit(EXIT_FAILURE,
2709 						 "Fail to add ptype cb\n");
2710 			} else if (!check_ptype(portid))
2711 				rte_exit(EXIT_FAILURE,
2712 					 "PMD can not provide needed ptypes\n");
2713 		}
2714 	}
2715 
2716 	printf("\n");
2717 
2718 	/* start ports */
2719 	RTE_ETH_FOREACH_DEV(portid) {
2720 		if ((enabled_port_mask & (1 << portid)) == 0) {
2721 			continue;
2722 		}
2723 		/* Start device */
2724 		ret = rte_eth_dev_start(portid);
2725 		if (ret < 0)
2726 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
2727 						"port=%d\n", ret, portid);
2728 		/*
2729 		 * If enabled, put device in promiscuous mode.
2730 		 * This allows IO forwarding mode to forward packets
2731 		 * to itself through 2 cross-connected  ports of the
2732 		 * target machine.
2733 		 */
2734 		if (promiscuous_on) {
2735 			ret = rte_eth_promiscuous_enable(portid);
2736 			if (ret != 0)
2737 				rte_exit(EXIT_FAILURE,
2738 					"rte_eth_promiscuous_enable: err=%s, port=%u\n",
2739 					rte_strerror(-ret), portid);
2740 		}
2741 		/* initialize spinlock for each port */
2742 		rte_spinlock_init(&(locks[portid]));
2743 	}
2744 
2745 	check_all_ports_link_status(enabled_port_mask);
2746 
2747 	if (app_mode == APP_MODE_EMPTY_POLL) {
2748 
2749 		if (empty_poll_train) {
2750 			policy.state = TRAINING;
2751 		} else {
2752 			policy.state = MED_NORMAL;
2753 			policy.med_base_edpi = ep_med_edpi;
2754 			policy.hgh_base_edpi = ep_hgh_edpi;
2755 		}
2756 
2757 		ret = rte_power_empty_poll_stat_init(&ep_params,
2758 				freq_tlb,
2759 				&policy);
2760 		if (ret < 0)
2761 			rte_exit(EXIT_FAILURE, "empty poll init failed");
2762 	}
2763 
2764 
2765 	/* launch per-lcore init on every lcore */
2766 	if (app_mode == APP_MODE_LEGACY) {
2767 		rte_eal_mp_remote_launch(main_legacy_loop, NULL, CALL_MAIN);
2768 	} else if (app_mode == APP_MODE_EMPTY_POLL) {
2769 		empty_poll_stop = false;
2770 		rte_eal_mp_remote_launch(main_empty_poll_loop, NULL,
2771 				SKIP_MAIN);
2772 	} else if (app_mode == APP_MODE_TELEMETRY) {
2773 		unsigned int i;
2774 
2775 		/* Init metrics library */
2776 		rte_metrics_init(rte_socket_id());
2777 		/** Register stats with metrics library */
2778 		for (i = 0; i < NUM_TELSTATS; i++)
2779 			ptr_strings[i] = telstats_strings[i].name;
2780 
2781 		ret = rte_metrics_reg_names(ptr_strings, NUM_TELSTATS);
2782 		if (ret >= 0)
2783 			telstats_index = ret;
2784 		else
2785 			rte_exit(EXIT_FAILURE, "failed to register metrics names");
2786 
2787 		RTE_LCORE_FOREACH_WORKER(lcore_id) {
2788 			rte_spinlock_init(&stats[lcore_id].telemetry_lock);
2789 		}
2790 		rte_timer_init(&telemetry_timer);
2791 		rte_telemetry_register_cmd("/l3fwd-power/stats",
2792 				handle_app_stats,
2793 				"Returns global power stats. Parameters: None");
2794 		rte_eal_mp_remote_launch(main_telemetry_loop, NULL,
2795 						SKIP_MAIN);
2796 	} else if (app_mode == APP_MODE_INTERRUPT) {
2797 		rte_eal_mp_remote_launch(main_intr_loop, NULL, CALL_MAIN);
2798 	}
2799 
2800 	if (app_mode == APP_MODE_EMPTY_POLL || app_mode == APP_MODE_TELEMETRY)
2801 		launch_timer(rte_lcore_id());
2802 
2803 	RTE_LCORE_FOREACH_WORKER(lcore_id) {
2804 		if (rte_eal_wait_lcore(lcore_id) < 0)
2805 			return -1;
2806 	}
2807 
2808 	RTE_ETH_FOREACH_DEV(portid)
2809 	{
2810 		if ((enabled_port_mask & (1 << portid)) == 0)
2811 			continue;
2812 
2813 		ret = rte_eth_dev_stop(portid);
2814 		if (ret != 0)
2815 			RTE_LOG(ERR, L3FWD_POWER, "rte_eth_dev_stop: err=%d, port=%u\n",
2816 				ret, portid);
2817 
2818 		rte_eth_dev_close(portid);
2819 	}
2820 
2821 	if (app_mode == APP_MODE_EMPTY_POLL)
2822 		rte_power_empty_poll_stat_free();
2823 
2824 	if ((app_mode == APP_MODE_LEGACY || app_mode == APP_MODE_EMPTY_POLL) &&
2825 			deinit_power_library())
2826 		rte_exit(EXIT_FAILURE, "deinit_power_library failed\n");
2827 
2828 	if (rte_eal_cleanup() < 0)
2829 		RTE_LOG(ERR, L3FWD_POWER, "EAL cleanup failed\n");
2830 
2831 	return 0;
2832 }
2833