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