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