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