examples/l3fwd-power/main.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2018 Intel Corporation
   3  */
   4
   5 #include <stdio.h>
   6 #include <stdlib.h>
   7 #include <stdint.h>
   8 #include <inttypes.h>
   9 #include <sys/types.h>
  10 #include <string.h>
  11 #include <sys/queue.h>
  12 #include <stdarg.h>
  13 #include <errno.h>
  14 #include <getopt.h>
  15 #include <unistd.h>
  16 #include <signal.h>
  17 #include <math.h>
  18
  19 #include <rte_common.h>
  20 #include <rte_byteorder.h>
  21 #include <rte_log.h>
  22 #include <rte_malloc.h>
  23 #include <rte_memory.h>
  24 #include <rte_memcpy.h>
  25 #include <rte_eal.h>
  26 #include <rte_launch.h>
  27 #include <rte_atomic.h>
  28 #include <rte_cycles.h>
  29 #include <rte_prefetch.h>
  30 #include <rte_lcore.h>
  31 #include <rte_per_lcore.h>
  32 #include <rte_branch_prediction.h>
  33 #include <rte_interrupts.h>
  34 #include <rte_random.h>
  35 #include <rte_debug.h>
  36 #include <rte_ether.h>
  37 #include <rte_ethdev.h>
  38 #include <rte_mempool.h>
  39 #include <rte_mbuf.h>
  40 #include <rte_ip.h>
  41 #include <rte_tcp.h>
  42 #include <rte_udp.h>
  43 #include <rte_string_fns.h>
  44 #include <rte_timer.h>
  45 #include <rte_power.h>
  46 #include <rte_spinlock.h>
  47 #include <rte_power_empty_poll.h>
  48 #include <rte_metrics.h>
  49 #include <rte_telemetry.h>
  50
  51 #include "perf_core.h"
  52 #include "main.h"
  53
  54 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
  55
  56 #define MAX_PKT_BURST 32
  57
  58 #define MIN_ZERO_POLL_COUNT 10
  59
  60 /* 100 ms interval */
  61 #define TIMER_NUMBER_PER_SECOND           10
  62 /* (10ms) */
  63 #define INTERVALS_PER_SECOND             100
  64 /* 100000 us */
  65 #define SCALING_PERIOD                    (1000000/TIMER_NUMBER_PER_SECOND)
  66 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
  67
  68 #define APP_LOOKUP_EXACT_MATCH          0
  69 #define APP_LOOKUP_LPM                  1
  70 #define DO_RFC_1812_CHECKS
  71
  72 #ifndef APP_LOOKUP_METHOD
  73 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  74 #endif
  75
  76 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  77 #include <rte_hash.h>
  78 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
  79 #include <rte_lpm.h>
  80 #else
  81 #error "APP_LOOKUP_METHOD set to incorrect value"
  82 #endif
  83
  84 #ifndef IPv6_BYTES
  85 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
  86                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
  87 #define IPv6_BYTES(addr) \
  88         addr[0],  addr[1], addr[2],  addr[3], \
  89         addr[4],  addr[5], addr[6],  addr[7], \
  90         addr[8],  addr[9], addr[10], addr[11],\
  91         addr[12], addr[13],addr[14], addr[15]
  92 #endif
  93
  94 #define MAX_JUMBO_PKT_LEN  9600
  95
  96 #define IPV6_ADDR_LEN 16
  97
  98 #define MEMPOOL_CACHE_SIZE 256
  99
 100 /*
 101  * This expression is used to calculate the number of mbufs needed depending on
 102  * user input, taking into account memory for rx and tx hardware rings, cache
 103  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 104  * NB_MBUF never goes below a minimum value of 8192.
 105  */
 106
 107 #define NB_MBUF RTE_MAX ( \
 108         (nb_ports*nb_rx_queue*nb_rxd + \
 109         nb_ports*nb_lcores*MAX_PKT_BURST + \
 110         nb_ports*n_tx_queue*nb_txd + \
 111         nb_lcores*MEMPOOL_CACHE_SIZE), \
 112         (unsigned)8192)
 113
 114 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 115
 116 #define NB_SOCKETS 8
 117
 118 /* Configure how many packets ahead to prefetch, when reading packets */
 119 #define PREFETCH_OFFSET 3
 120
 121 /*
 122  * Configurable number of RX/TX ring descriptors
 123  */
 124 #define RTE_TEST_RX_DESC_DEFAULT 1024
 125 #define RTE_TEST_TX_DESC_DEFAULT 1024
 126
 127 /*
 128  * These two thresholds were decided on by running the training algorithm on
 129  * a 2.5GHz Xeon. These defaults can be overridden by supplying non-zero values
 130  * for the med_threshold and high_threshold parameters on the command line.
 131  */
 132 #define EMPTY_POLL_MED_THRESHOLD 350000UL
 133 #define EMPTY_POLL_HGH_THRESHOLD 580000UL
 134
 135 #define NUM_TELSTATS RTE_DIM(telstats_strings)
 136
 137 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 138 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 139
 140 /* ethernet addresses of ports */
 141 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 142
 143 /* ethernet addresses of ports */
 144 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
 145
 146 /* mask of enabled ports */
 147 static uint32_t enabled_port_mask = 0;
 148 /* Ports set in promiscuous mode off by default. */
 149 static int promiscuous_on = 0;
 150 /* NUMA is enabled by default. */
 151 static int numa_on = 1;
 152 static bool empty_poll_stop;
 153 static bool empty_poll_train;
 154 volatile bool quit_signal;
 155 static struct  ep_params *ep_params;
 156 static struct  ep_policy policy;
 157 static long  ep_med_edpi, ep_hgh_edpi;
 158 /* timer to update telemetry every 500ms */
 159 static struct rte_timer telemetry_timer;
 160
 161 /* stats index returned by metrics lib */
 162 int telstats_index;
 163
 164 struct telstats_name {
 165         char name[RTE_ETH_XSTATS_NAME_SIZE];
 166 };
 167
 168 /* telemetry stats to be reported */
 169 const struct telstats_name telstats_strings[] = {
 170         {"empty_poll"},
 171         {"full_poll"},
 172         {"busy_percent"}
 173 };
 174
 175 /* core busyness in percentage */
 176 enum busy_rate {
 177         ZERO = 0,
 178         PARTIAL = 50,
 179         FULL = 100
 180 };
 181
 182 /* reference poll count to measure core busyness */
 183 #define DEFAULT_COUNT 10000
 184 /*
 185  * reference CYCLES to be used to
 186  * measure core busyness based on poll count
 187  */
 188 #define MIN_CYCLES  1500000ULL
 189 #define MAX_CYCLES 22000000ULL
 190
 191 /* (500ms) */
 192 #define TELEMETRY_INTERVALS_PER_SEC 2
 193
 194 static int parse_ptype; /**< Parse packet type using rx callback, and */
 195                         /**< disabled by default */
 196
 197 enum appmode {
 198         APP_MODE_DEFAULT = 0,
 199         APP_MODE_LEGACY,
 200         APP_MODE_EMPTY_POLL,
 201         APP_MODE_TELEMETRY,
 202         APP_MODE_INTERRUPT
 203 };
 204
 205 enum appmode app_mode;
 206
 207 enum freq_scale_hint_t
 208 {
 209         FREQ_LOWER    =      -1,
 210         FREQ_CURRENT  =       0,
 211         FREQ_HIGHER   =       1,
 212         FREQ_HIGHEST  =       2
 213 };
 214
 215 struct lcore_rx_queue {
 216         uint16_t port_id;
 217         uint8_t queue_id;
 218         enum freq_scale_hint_t freq_up_hint;
 219         uint32_t zero_rx_packet_count;
 220         uint32_t idle_hint;
 221 } __rte_cache_aligned;
 222
 223 #define MAX_RX_QUEUE_PER_LCORE 16
 224 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 225 #define MAX_RX_QUEUE_PER_PORT 128
 226
 227 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
 228
 229
 230 struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 231 static struct lcore_params lcore_params_array_default[] = {
 232         {0, 0, 2},
 233         {0, 1, 2},
 234         {0, 2, 2},
 235         {1, 0, 2},
 236         {1, 1, 2},
 237         {1, 2, 2},
 238         {2, 0, 2},
 239         {3, 0, 3},
 240         {3, 1, 3},
 241 };
 242
 243 struct lcore_params *lcore_params = lcore_params_array_default;
 244 uint16_t nb_lcore_params = RTE_DIM(lcore_params_array_default);
 245
 246 static struct rte_eth_conf port_conf = {
 247         .rxmode = {
 248                 .mq_mode        = ETH_MQ_RX_RSS,
 249                 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
 250                 .split_hdr_size = 0,
 251                 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
 252         },
 253         .rx_adv_conf = {
 254                 .rss_conf = {
 255                         .rss_key = NULL,
 256                         .rss_hf = ETH_RSS_UDP,
 257                 },
 258         },
 259         .txmode = {
 260                 .mq_mode = ETH_MQ_TX_NONE,
 261         }
 262 };
 263
 264 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 265
 266
 267 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 268
 269 #ifdef RTE_ARCH_X86
 270 #include <rte_hash_crc.h>
 271 #define DEFAULT_HASH_FUNC       rte_hash_crc
 272 #else
 273 #include <rte_jhash.h>
 274 #define DEFAULT_HASH_FUNC       rte_jhash
 275 #endif
 276
 277 struct ipv4_5tuple {
 278         uint32_t ip_dst;
 279         uint32_t ip_src;
 280         uint16_t port_dst;
 281         uint16_t port_src;
 282         uint8_t  proto;
 283 } __rte_packed;
 284
 285 struct ipv6_5tuple {
 286         uint8_t  ip_dst[IPV6_ADDR_LEN];
 287         uint8_t  ip_src[IPV6_ADDR_LEN];
 288         uint16_t port_dst;
 289         uint16_t port_src;
 290         uint8_t  proto;
 291 } __rte_packed;
 292
 293 struct ipv4_l3fwd_route {
 294         struct ipv4_5tuple key;
 295         uint8_t if_out;
 296 };
 297
 298 struct ipv6_l3fwd_route {
 299         struct ipv6_5tuple key;
 300         uint8_t if_out;
 301 };
 302
 303 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 304         {{RTE_IPV4(100,10,0,1), RTE_IPV4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 305         {{RTE_IPV4(100,20,0,2), RTE_IPV4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 306         {{RTE_IPV4(100,30,0,3), RTE_IPV4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 307         {{RTE_IPV4(100,40,0,4), RTE_IPV4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 308 };
 309
 310 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 311         {
 312                 {
 313                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 314                          0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 315                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 316                          0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
 317                          1, 10, IPPROTO_UDP
 318                 }, 4
 319         },
 320 };
 321
 322 typedef struct rte_hash lookup_struct_t;
 323 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 324 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 325
 326 #define L3FWD_HASH_ENTRIES      1024
 327
 328 static uint16_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 329 static uint16_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 330 #endif
 331
 332 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 333 struct ipv4_l3fwd_route {
 334         uint32_t ip;
 335         uint8_t  depth;
 336         uint8_t  if_out;
 337 };
 338
 339 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 340         {RTE_IPV4(1,1,1,0), 24, 0},
 341         {RTE_IPV4(2,1,1,0), 24, 1},
 342         {RTE_IPV4(3,1,1,0), 24, 2},
 343         {RTE_IPV4(4,1,1,0), 24, 3},
 344         {RTE_IPV4(5,1,1,0), 24, 4},
 345         {RTE_IPV4(6,1,1,0), 24, 5},
 346         {RTE_IPV4(7,1,1,0), 24, 6},
 347         {RTE_IPV4(8,1,1,0), 24, 7},
 348 };
 349
 350 #define IPV4_L3FWD_LPM_MAX_RULES     1024
 351
 352 typedef struct rte_lpm lookup_struct_t;
 353 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 354 #endif
 355
 356 struct lcore_conf {
 357         uint16_t n_rx_queue;
 358         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 359         uint16_t n_tx_port;
 360         uint16_t tx_port_id[RTE_MAX_ETHPORTS];
 361         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 362         struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
 363         lookup_struct_t * ipv4_lookup_struct;
 364         lookup_struct_t * ipv6_lookup_struct;
 365 } __rte_cache_aligned;
 366
 367 struct lcore_stats {
 368         /* total sleep time in ms since last frequency scaling down */
 369         uint32_t sleep_time;
 370         /* number of long sleep recently */
 371         uint32_t nb_long_sleep;
 372         /* freq. scaling up trend */
 373         uint32_t trend;
 374         /* total packet processed recently */
 375         uint64_t nb_rx_processed;
 376         /* total iterations looped recently */
 377         uint64_t nb_iteration_looped;
 378         /*
 379          * Represents empty and non empty polls
 380          * of rte_eth_rx_burst();
 381          * ep_nep[0] holds non empty polls
 382          * i.e. 0 < nb_rx <= MAX_BURST
 383          * ep_nep[1] holds empty polls.
 384          * i.e. nb_rx == 0
 385          */
 386         uint64_t ep_nep[2];
 387         /*
 388          * Represents full and empty+partial
 389          * polls of rte_eth_rx_burst();
 390          * ep_nep[0] holds empty+partial polls.
 391          * i.e. 0 <= nb_rx < MAX_BURST
 392          * ep_nep[1] holds full polls
 393          * i.e. nb_rx == MAX_BURST
 394          */
 395         uint64_t fp_nfp[2];
 396         enum busy_rate br;
 397         rte_spinlock_t telemetry_lock;
 398 } __rte_cache_aligned;
 399
 400 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
 401 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
 402 static struct rte_timer power_timers[RTE_MAX_LCORE];
 403
 404 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
 405 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
 406                 unsigned int lcore_id, uint16_t port_id, uint16_t queue_id);
 407
 408
 409 /*
 410  * These defaults are using the max frequency index (1), a medium index (9)
 411  * and a typical low frequency index (14). These can be adjusted to use
 412  * different indexes using the relevant command line parameters.
 413  */
 414 static uint8_t  freq_tlb[] = {14, 9, 1};
 415
 416 static int is_done(void)
 417 {
 418         return quit_signal;
 419 }
 420
 421 /* exit signal handler */
 422 static void
 423 signal_exit_now(int sigtype)
 424 {
 425
 426         if (sigtype == SIGINT)
 427                 quit_signal = true;
 428
 429 }
 430
 431 /*  Freqency scale down timer callback */
 432 static void
 433 power_timer_cb(__rte_unused struct rte_timer *tim,
 434                           __rte_unused void *arg)
 435 {
 436         uint64_t hz;
 437         float sleep_time_ratio;
 438         unsigned lcore_id = rte_lcore_id();
 439
 440         /* accumulate total execution time in us when callback is invoked */
 441         sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
 442                                         (float)SCALING_PERIOD;
 443         /**
 444          * check whether need to scale down frequency a step if it sleep a lot.
 445          */
 446         if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
 447                 if (rte_power_freq_down)
 448                         rte_power_freq_down(lcore_id);
 449         }
 450         else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
 451                 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
 452                 /**
 453                  * scale down a step if average packet per iteration less
 454                  * than expectation.
 455                  */
 456                 if (rte_power_freq_down)
 457                         rte_power_freq_down(lcore_id);
 458         }
 459
 460         /**
 461          * initialize another timer according to current frequency to ensure
 462          * timer interval is relatively fixed.
 463          */
 464         hz = rte_get_timer_hz();
 465         rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
 466                                 SINGLE, lcore_id, power_timer_cb, NULL);
 467
 468         stats[lcore_id].nb_rx_processed = 0;
 469         stats[lcore_id].nb_iteration_looped = 0;
 470
 471         stats[lcore_id].sleep_time = 0;
 472 }
 473
 474 /* Enqueue a single packet, and send burst if queue is filled */
 475 static inline int
 476 send_single_packet(struct rte_mbuf *m, uint16_t port)
 477 {
 478         uint32_t lcore_id;
 479         struct lcore_conf *qconf;
 480
 481         lcore_id = rte_lcore_id();
 482         qconf = &lcore_conf[lcore_id];
 483
 484         rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
 485                         qconf->tx_buffer[port], m);
 486
 487         return 0;
 488 }
 489
 490 #ifdef DO_RFC_1812_CHECKS
 491 static inline int
 492 is_valid_ipv4_pkt(struct rte_ipv4_hdr *pkt, uint32_t link_len)
 493 {
 494         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 495         /*
 496          * 1. The packet length reported by the Link Layer must be large
 497          * enough to hold the minimum length legal IP datagram (20 bytes).
 498          */
 499         if (link_len < sizeof(struct rte_ipv4_hdr))
 500                 return -1;
 501
 502         /* 2. The IP checksum must be correct. */
 503         /* this is checked in H/W */
 504
 505         /*
 506          * 3. The IP version number must be 4. If the version number is not 4
 507          * then the packet may be another version of IP, such as IPng or
 508          * ST-II.
 509          */
 510         if (((pkt->version_ihl) >> 4) != 4)
 511                 return -3;
 512         /*
 513          * 4. The IP header length field must be large enough to hold the
 514          * minimum length legal IP datagram (20 bytes = 5 words).
 515          */
 516         if ((pkt->version_ihl & 0xf) < 5)
 517                 return -4;
 518
 519         /*
 520          * 5. The IP total length field must be large enough to hold the IP
 521          * datagram header, whose length is specified in the IP header length
 522          * field.
 523          */
 524         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct rte_ipv4_hdr))
 525                 return -5;
 526
 527         return 0;
 528 }
 529 #endif
 530
 531 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 532 static void
 533 print_ipv4_key(struct ipv4_5tuple key)
 534 {
 535         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
 536                 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
 537                                 key.port_dst, key.port_src, key.proto);
 538 }
 539 static void
 540 print_ipv6_key(struct ipv6_5tuple key)
 541 {
 542         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 543                 "port dst = %d, port src = %d, proto = %d\n",
 544                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 545                 key.port_dst, key.port_src, key.proto);
 546 }
 547
 548 static inline uint16_t
 549 get_ipv4_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
 550                 lookup_struct_t * ipv4_l3fwd_lookup_struct)
 551 {
 552         struct ipv4_5tuple key;
 553         struct rte_tcp_hdr *tcp;
 554         struct rte_udp_hdr *udp;
 555         int ret = 0;
 556
 557         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 558         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 559         key.proto = ipv4_hdr->next_proto_id;
 560
 561         switch (ipv4_hdr->next_proto_id) {
 562         case IPPROTO_TCP:
 563                 tcp = (struct rte_tcp_hdr *)((unsigned char *)ipv4_hdr +
 564                                         sizeof(struct rte_ipv4_hdr));
 565                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 566                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 567                 break;
 568
 569         case IPPROTO_UDP:
 570                 udp = (struct rte_udp_hdr *)((unsigned char *)ipv4_hdr +
 571                                         sizeof(struct rte_ipv4_hdr));
 572                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 573                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 574                 break;
 575
 576         default:
 577                 key.port_dst = 0;
 578                 key.port_src = 0;
 579                 break;
 580         }
 581
 582         /* Find destination port */
 583         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 584         return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
 585 }
 586
 587 static inline uint16_t
 588 get_ipv6_dst_port(struct rte_ipv6_hdr *ipv6_hdr, uint16_t portid,
 589                         lookup_struct_t *ipv6_l3fwd_lookup_struct)
 590 {
 591         struct ipv6_5tuple key;
 592         struct rte_tcp_hdr *tcp;
 593         struct rte_udp_hdr *udp;
 594         int ret = 0;
 595
 596         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 597         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 598
 599         key.proto = ipv6_hdr->proto;
 600
 601         switch (ipv6_hdr->proto) {
 602         case IPPROTO_TCP:
 603                 tcp = (struct rte_tcp_hdr *)((unsigned char *) ipv6_hdr +
 604                                         sizeof(struct rte_ipv6_hdr));
 605                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 606                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 607                 break;
 608
 609         case IPPROTO_UDP:
 610                 udp = (struct rte_udp_hdr *)((unsigned char *) ipv6_hdr +
 611                                         sizeof(struct rte_ipv6_hdr));
 612                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 613                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 614                 break;
 615
 616         default:
 617                 key.port_dst = 0;
 618                 key.port_src = 0;
 619                 break;
 620         }
 621
 622         /* Find destination port */
 623         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 624         return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
 625 }
 626 #endif
 627
 628 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 629 static inline uint16_t
 630 get_ipv4_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
 631                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 632 {
 633         uint32_t next_hop;
 634
 635         return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 636                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 637                         next_hop : portid);
 638 }
 639 #endif
 640
 641 static inline void
 642 parse_ptype_one(struct rte_mbuf *m)
 643 {
 644         struct rte_ether_hdr *eth_hdr;
 645         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 646         uint16_t ether_type;
 647
 648         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 649         ether_type = eth_hdr->ether_type;
 650         if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
 651                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 652         else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6))
 653                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 654
 655         m->packet_type = packet_type;
 656 }
 657
 658 static uint16_t
 659 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
 660                struct rte_mbuf *pkts[], uint16_t nb_pkts,
 661                uint16_t max_pkts __rte_unused,
 662                void *user_param __rte_unused)
 663 {
 664         unsigned int i;
 665
 666         for (i = 0; i < nb_pkts; ++i)
 667                 parse_ptype_one(pkts[i]);
 668
 669         return nb_pkts;
 670 }
 671
 672 static int
 673 add_cb_parse_ptype(uint16_t portid, uint16_t queueid)
 674 {
 675         printf("Port %d: softly parse packet type info\n", portid);
 676         if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL))
 677                 return 0;
 678
 679         printf("Failed to add rx callback: port=%d\n", portid);
 680         return -1;
 681 }
 682
 683 static inline void
 684 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
 685                                 struct lcore_conf *qconf)
 686 {
 687         struct rte_ether_hdr *eth_hdr;
 688         struct rte_ipv4_hdr *ipv4_hdr;
 689         void *d_addr_bytes;
 690         uint16_t dst_port;
 691
 692         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 693
 694         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 695                 /* Handle IPv4 headers.*/
 696                 ipv4_hdr =
 697                         rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
 698                                                 sizeof(struct rte_ether_hdr));
 699
 700 #ifdef DO_RFC_1812_CHECKS
 701                 /* Check to make sure the packet is valid (RFC1812) */
 702                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 703                         rte_pktmbuf_free(m);
 704                         return;
 705                 }
 706 #endif
 707
 708                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 709                                         qconf->ipv4_lookup_struct);
 710                 if (dst_port >= RTE_MAX_ETHPORTS ||
 711                                 (enabled_port_mask & 1 << dst_port) == 0)
 712                         dst_port = portid;
 713
 714                 /* 02:00:00:00:00:xx */
 715                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 716                 *((uint64_t *)d_addr_bytes) =
 717                         0x000000000002 + ((uint64_t)dst_port << 40);
 718
 719 #ifdef DO_RFC_1812_CHECKS
 720                 /* Update time to live and header checksum */
 721                 --(ipv4_hdr->time_to_live);
 722                 ++(ipv4_hdr->hdr_checksum);
 723 #endif
 724
 725                 /* src addr */
 726                 rte_ether_addr_copy(&ports_eth_addr[dst_port],
 727                                 &eth_hdr->s_addr);
 728
 729                 send_single_packet(m, dst_port);
 730         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 731                 /* Handle IPv6 headers.*/
 732 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 733                 struct rte_ipv6_hdr *ipv6_hdr;
 734
 735                 ipv6_hdr =
 736                         rte_pktmbuf_mtod_offset(m, struct rte_ipv6_hdr *,
 737                                                 sizeof(struct rte_ether_hdr));
 738
 739                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
 740                                         qconf->ipv6_lookup_struct);
 741
 742                 if (dst_port >= RTE_MAX_ETHPORTS ||
 743                                 (enabled_port_mask & 1 << dst_port) == 0)
 744                         dst_port = portid;
 745
 746                 /* 02:00:00:00:00:xx */
 747                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 748                 *((uint64_t *)d_addr_bytes) =
 749                         0x000000000002 + ((uint64_t)dst_port << 40);
 750
 751                 /* src addr */
 752                 rte_ether_addr_copy(&ports_eth_addr[dst_port],
 753                                 &eth_hdr->s_addr);
 754
 755                 send_single_packet(m, dst_port);
 756 #else
 757                 /* We don't currently handle IPv6 packets in LPM mode. */
 758                 rte_pktmbuf_free(m);
 759 #endif
 760         } else
 761                 rte_pktmbuf_free(m);
 762
 763 }
 764
 765 #define MINIMUM_SLEEP_TIME         1
 766 #define SUSPEND_THRESHOLD          300
 767
 768 static inline uint32_t
 769 power_idle_heuristic(uint32_t zero_rx_packet_count)
 770 {
 771         /* If zero count is less than 100,  sleep 1us */
 772         if (zero_rx_packet_count < SUSPEND_THRESHOLD)
 773                 return MINIMUM_SLEEP_TIME;
 774         /* If zero count is less than 1000, sleep 100 us which is the
 775                 minimum latency switching from C3/C6 to C0
 776         */
 777         else
 778                 return SUSPEND_THRESHOLD;
 779 }
 780
 781 static inline enum freq_scale_hint_t
 782 power_freq_scaleup_heuristic(unsigned lcore_id,
 783                              uint16_t port_id,
 784                              uint16_t queue_id)
 785 {
 786         uint32_t rxq_count = rte_eth_rx_queue_count(port_id, queue_id);
 787 /**
 788  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
 789  * per iteration
 790  */
 791 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
 792 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
 793 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
 794 #define FREQ_UP_TREND1_ACC   1
 795 #define FREQ_UP_TREND2_ACC   100
 796 #define FREQ_UP_THRESHOLD    10000
 797
 798         if (likely(rxq_count > FREQ_GEAR3_RX_PACKET_THRESHOLD)) {
 799                 stats[lcore_id].trend = 0;
 800                 return FREQ_HIGHEST;
 801         } else if (likely(rxq_count > FREQ_GEAR2_RX_PACKET_THRESHOLD))
 802                 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
 803         else if (likely(rxq_count > FREQ_GEAR1_RX_PACKET_THRESHOLD))
 804                 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
 805
 806         if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
 807                 stats[lcore_id].trend = 0;
 808                 return FREQ_HIGHER;
 809         }
 810
 811         return FREQ_CURRENT;
 812 }
 813
 814 /**
 815  * force polling thread sleep until one-shot rx interrupt triggers
 816  * @param port_id
 817  *  Port id.
 818  * @param queue_id
 819  *  Rx queue id.
 820  * @return
 821  *  0 on success
 822  */
 823 static int
 824 sleep_until_rx_interrupt(int num)
 825 {
 826         /*
 827          * we want to track when we are woken up by traffic so that we can go
 828          * back to sleep again without log spamming.
 829          */
 830         static bool timeout;
 831         struct rte_epoll_event event[num];
 832         int n, i;
 833         uint16_t port_id;
 834         uint8_t queue_id;
 835         void *data;
 836
 837         if (!timeout) {
 838                 RTE_LOG(INFO, L3FWD_POWER,
 839                                 "lcore %u sleeps until interrupt triggers\n",
 840                                 rte_lcore_id());
 841         }
 842
 843         n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, 10);
 844         for (i = 0; i < n; i++) {
 845                 data = event[i].epdata.data;
 846                 port_id = ((uintptr_t)data) >> CHAR_BIT;
 847                 queue_id = ((uintptr_t)data) &
 848                         RTE_LEN2MASK(CHAR_BIT, uint8_t);
 849                 RTE_LOG(INFO, L3FWD_POWER,
 850                         "lcore %u is waked up from rx interrupt on"
 851                         " port %d queue %d\n",
 852                         rte_lcore_id(), port_id, queue_id);
 853         }
 854         timeout = n == 0;
 855
 856         return 0;
 857 }
 858
 859 static void turn_on_off_intr(struct lcore_conf *qconf, bool on)
 860 {
 861         int i;
 862         struct lcore_rx_queue *rx_queue;
 863         uint8_t queue_id;
 864         uint16_t port_id;
 865
 866         for (i = 0; i < qconf->n_rx_queue; ++i) {
 867                 rx_queue = &(qconf->rx_queue_list[i]);
 868                 port_id = rx_queue->port_id;
 869                 queue_id = rx_queue->queue_id;
 870
 871                 rte_spinlock_lock(&(locks[port_id]));
 872                 if (on)
 873                         rte_eth_dev_rx_intr_enable(port_id, queue_id);
 874                 else
 875                         rte_eth_dev_rx_intr_disable(port_id, queue_id);
 876                 rte_spinlock_unlock(&(locks[port_id]));
 877         }
 878 }
 879
 880 static int event_register(struct lcore_conf *qconf)
 881 {
 882         struct lcore_rx_queue *rx_queue;
 883         uint8_t queueid;
 884         uint16_t portid;
 885         uint32_t data;
 886         int ret;
 887         int i;
 888
 889         for (i = 0; i < qconf->n_rx_queue; ++i) {
 890                 rx_queue = &(qconf->rx_queue_list[i]);
 891                 portid = rx_queue->port_id;
 892                 queueid = rx_queue->queue_id;
 893                 data = portid << CHAR_BIT | queueid;
 894
 895                 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
 896                                                 RTE_EPOLL_PER_THREAD,
 897                                                 RTE_INTR_EVENT_ADD,
 898                                                 (void *)((uintptr_t)data));
 899                 if (ret)
 900                         return ret;
 901         }
 902
 903         return 0;
 904 }
 905
 906 /* main processing loop */
 907 static int main_intr_loop(__rte_unused void *dummy)
 908 {
 909         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 910         unsigned int lcore_id;
 911         uint64_t prev_tsc, diff_tsc, cur_tsc;
 912         int i, j, nb_rx;
 913         uint8_t queueid;
 914         uint16_t portid;
 915         struct lcore_conf *qconf;
 916         struct lcore_rx_queue *rx_queue;
 917         uint32_t lcore_rx_idle_count = 0;
 918         uint32_t lcore_idle_hint = 0;
 919         int intr_en = 0;
 920
 921         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
 922                                    US_PER_S * BURST_TX_DRAIN_US;
 923
 924         prev_tsc = 0;
 925
 926         lcore_id = rte_lcore_id();
 927         qconf = &lcore_conf[lcore_id];
 928
 929         if (qconf->n_rx_queue == 0) {
 930                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
 931                                 lcore_id);
 932                 return 0;
 933         }
 934
 935         RTE_LOG(INFO, L3FWD_POWER, "entering main interrupt loop on lcore %u\n",
 936                         lcore_id);
 937
 938         for (i = 0; i < qconf->n_rx_queue; i++) {
 939                 portid = qconf->rx_queue_list[i].port_id;
 940                 queueid = qconf->rx_queue_list[i].queue_id;
 941                 RTE_LOG(INFO, L3FWD_POWER,
 942                                 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
 943                                 lcore_id, portid, queueid);
 944         }
 945
 946         /* add into event wait list */
 947         if (event_register(qconf) == 0)
 948                 intr_en = 1;
 949         else
 950                 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
 951
 952         while (!is_done()) {
 953                 stats[lcore_id].nb_iteration_looped++;
 954
 955                 cur_tsc = rte_rdtsc();
 956
 957                 /*
 958                  * TX burst queue drain
 959                  */
 960                 diff_tsc = cur_tsc - prev_tsc;
 961                 if (unlikely(diff_tsc > drain_tsc)) {
 962                         for (i = 0; i < qconf->n_tx_port; ++i) {
 963                                 portid = qconf->tx_port_id[i];
 964                                 rte_eth_tx_buffer_flush(portid,
 965                                                 qconf->tx_queue_id[portid],
 966                                                 qconf->tx_buffer[portid]);
 967                         }
 968                         prev_tsc = cur_tsc;
 969                 }
 970
 971 start_rx:
 972                 /*
 973                  * Read packet from RX queues
 974                  */
 975                 lcore_rx_idle_count = 0;
 976                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 977                         rx_queue = &(qconf->rx_queue_list[i]);
 978                         rx_queue->idle_hint = 0;
 979                         portid = rx_queue->port_id;
 980                         queueid = rx_queue->queue_id;
 981
 982                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 983                                         MAX_PKT_BURST);
 984
 985                         stats[lcore_id].nb_rx_processed += nb_rx;
 986                         if (unlikely(nb_rx == 0)) {
 987                                 /**
 988                                  * no packet received from rx queue, try to
 989                                  * sleep for a while forcing CPU enter deeper
 990                                  * C states.
 991                                  */
 992                                 rx_queue->zero_rx_packet_count++;
 993
 994                                 if (rx_queue->zero_rx_packet_count <=
 995                                                 MIN_ZERO_POLL_COUNT)
 996                                         continue;
 997
 998                                 rx_queue->idle_hint = power_idle_heuristic(
 999                                                 rx_queue->zero_rx_packet_count);
1000                                 lcore_rx_idle_count++;
1001                         } else {
1002                                 rx_queue->zero_rx_packet_count = 0;
1003                         }
1004
1005                         /* Prefetch first packets */
1006                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1007                                 rte_prefetch0(rte_pktmbuf_mtod(
1008                                                 pkts_burst[j], void *));
1009                         }
1010
1011                         /* Prefetch and forward already prefetched packets */
1012                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1013                                 rte_prefetch0(rte_pktmbuf_mtod(
1014                                                 pkts_burst[j + PREFETCH_OFFSET],
1015                                                 void *));
1016                                 l3fwd_simple_forward(
1017                                                 pkts_burst[j], portid, qconf);
1018                         }
1019
1020                         /* Forward remaining prefetched packets */
1021                         for (; j < nb_rx; j++) {
1022                                 l3fwd_simple_forward(
1023                                                 pkts_burst[j], portid, qconf);
1024                         }
1025                 }
1026
1027                 if (unlikely(lcore_rx_idle_count == qconf->n_rx_queue)) {
1028                         /**
1029                          * All Rx queues empty in recent consecutive polls,
1030                          * sleep in a conservative manner, meaning sleep as
1031                          * less as possible.
1032                          */
1033                         for (i = 1,
1034                             lcore_idle_hint = qconf->rx_queue_list[0].idle_hint;
1035                                         i < qconf->n_rx_queue; ++i) {
1036                                 rx_queue = &(qconf->rx_queue_list[i]);
1037                                 if (rx_queue->idle_hint < lcore_idle_hint)
1038                                         lcore_idle_hint = rx_queue->idle_hint;
1039                         }
1040
1041                         if (lcore_idle_hint < SUSPEND_THRESHOLD)
1042                                 /**
1043                                  * execute "pause" instruction to avoid context
1044                                  * switch which generally take hundred of
1045                                  * microseconds for short sleep.
1046                                  */
1047                                 rte_delay_us(lcore_idle_hint);
1048                         else {
1049                                 /* suspend until rx interrupt triggers */
1050                                 if (intr_en) {
1051                                         turn_on_off_intr(qconf, 1);
1052                                         sleep_until_rx_interrupt(
1053                                                         qconf->n_rx_queue);
1054                                         turn_on_off_intr(qconf, 0);
1055                                         /**
1056                                          * start receiving packets immediately
1057                                          */
1058                                         if (likely(!is_done()))
1059                                                 goto start_rx;
1060                                 }
1061                         }
1062                         stats[lcore_id].sleep_time += lcore_idle_hint;
1063                 }
1064         }
1065
1066         return 0;
1067 }
1068
1069 /* main processing loop */
1070 static int
1071 main_telemetry_loop(__rte_unused void *dummy)
1072 {
1073         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1074         unsigned int lcore_id;
1075         uint64_t prev_tsc, diff_tsc, cur_tsc, prev_tel_tsc;
1076         int i, j, nb_rx;
1077         uint8_t queueid;
1078         uint16_t portid;
1079         struct lcore_conf *qconf;
1080         struct lcore_rx_queue *rx_queue;
1081         uint64_t ep_nep[2] = {0}, fp_nfp[2] = {0};
1082         uint64_t poll_count;
1083         enum busy_rate br;
1084
1085         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1086                                         US_PER_S * BURST_TX_DRAIN_US;
1087
1088         poll_count = 0;
1089         prev_tsc = 0;
1090         prev_tel_tsc = 0;
1091
1092         lcore_id = rte_lcore_id();
1093         qconf = &lcore_conf[lcore_id];
1094
1095         if (qconf->n_rx_queue == 0) {
1096                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
1097                         lcore_id);
1098                 return 0;
1099         }
1100
1101         RTE_LOG(INFO, L3FWD_POWER, "entering main telemetry loop on lcore %u\n",
1102                 lcore_id);
1103
1104         for (i = 0; i < qconf->n_rx_queue; i++) {
1105                 portid = qconf->rx_queue_list[i].port_id;
1106                 queueid = qconf->rx_queue_list[i].queue_id;
1107                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1108                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
1109         }
1110
1111         while (!is_done()) {
1112
1113                 cur_tsc = rte_rdtsc();
1114                 /*
1115                  * TX burst queue drain
1116                  */
1117                 diff_tsc = cur_tsc - prev_tsc;
1118                 if (unlikely(diff_tsc > drain_tsc)) {
1119                         for (i = 0; i < qconf->n_tx_port; ++i) {
1120                                 portid = qconf->tx_port_id[i];
1121                                 rte_eth_tx_buffer_flush(portid,
1122                                                 qconf->tx_queue_id[portid],
1123                                                 qconf->tx_buffer[portid]);
1124                         }
1125                         prev_tsc = cur_tsc;
1126                 }
1127
1128                 /*
1129                  * Read packet from RX queues
1130                  */
1131                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1132                         rx_queue = &(qconf->rx_queue_list[i]);
1133                         portid = rx_queue->port_id;
1134                         queueid = rx_queue->queue_id;
1135
1136                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1137                                                                 MAX_PKT_BURST);
1138                         ep_nep[nb_rx == 0]++;
1139                         fp_nfp[nb_rx == MAX_PKT_BURST]++;
1140                         poll_count++;
1141                         if (unlikely(nb_rx == 0))
1142                                 continue;
1143
1144                         /* Prefetch first packets */
1145                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1146                                 rte_prefetch0(rte_pktmbuf_mtod(
1147                                                 pkts_burst[j], void *));
1148                         }
1149
1150                         /* Prefetch and forward already prefetched packets */
1151                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1152                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1153                                                 j + PREFETCH_OFFSET], void *));
1154                                 l3fwd_simple_forward(pkts_burst[j], portid,
1155                                                                 qconf);
1156                         }
1157
1158                         /* Forward remaining prefetched packets */
1159                         for (; j < nb_rx; j++) {
1160                                 l3fwd_simple_forward(pkts_burst[j], portid,
1161                                                                 qconf);
1162                         }
1163                 }
1164                 if (unlikely(poll_count >= DEFAULT_COUNT)) {
1165                         diff_tsc = cur_tsc - prev_tel_tsc;
1166                         if (diff_tsc >= MAX_CYCLES) {
1167                                 br = FULL;
1168                         } else if (diff_tsc > MIN_CYCLES &&
1169                                         diff_tsc < MAX_CYCLES) {
1170                                 br = (diff_tsc * 100) / MAX_CYCLES;
1171                         } else {
1172                                 br = ZERO;
1173                         }
1174                         poll_count = 0;
1175                         prev_tel_tsc = cur_tsc;
1176                         /* update stats for telemetry */
1177                         rte_spinlock_lock(&stats[lcore_id].telemetry_lock);
1178                         stats[lcore_id].ep_nep[0] = ep_nep[0];
1179                         stats[lcore_id].ep_nep[1] = ep_nep[1];
1180                         stats[lcore_id].fp_nfp[0] = fp_nfp[0];
1181                         stats[lcore_id].fp_nfp[1] = fp_nfp[1];
1182                         stats[lcore_id].br = br;
1183                         rte_spinlock_unlock(&stats[lcore_id].telemetry_lock);
1184                 }
1185         }
1186
1187         return 0;
1188 }
1189 /* main processing loop */
1190 static int
1191 main_empty_poll_loop(__rte_unused void *dummy)
1192 {
1193         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1194         unsigned int lcore_id;
1195         uint64_t prev_tsc, diff_tsc, cur_tsc;
1196         int i, j, nb_rx;
1197         uint8_t queueid;
1198         uint16_t portid;
1199         struct lcore_conf *qconf;
1200         struct lcore_rx_queue *rx_queue;
1201
1202         const uint64_t drain_tsc =
1203                 (rte_get_tsc_hz() + US_PER_S - 1) /
1204                 US_PER_S * BURST_TX_DRAIN_US;
1205
1206         prev_tsc = 0;
1207
1208         lcore_id = rte_lcore_id();
1209         qconf = &lcore_conf[lcore_id];
1210
1211         if (qconf->n_rx_queue == 0) {
1212                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
1213                         lcore_id);
1214                 return 0;
1215         }
1216
1217         for (i = 0; i < qconf->n_rx_queue; i++) {
1218                 portid = qconf->rx_queue_list[i].port_id;
1219                 queueid = qconf->rx_queue_list[i].queue_id;
1220                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1221                                 "rxqueueid=%hhu\n", lcore_id, portid, queueid);
1222         }
1223
1224         while (!is_done()) {
1225                 stats[lcore_id].nb_iteration_looped++;
1226
1227                 cur_tsc = rte_rdtsc();
1228                 /*
1229                  * TX burst queue drain
1230                  */
1231                 diff_tsc = cur_tsc - prev_tsc;
1232                 if (unlikely(diff_tsc > drain_tsc)) {
1233                         for (i = 0; i < qconf->n_tx_port; ++i) {
1234                                 portid = qconf->tx_port_id[i];
1235                                 rte_eth_tx_buffer_flush(portid,
1236                                                 qconf->tx_queue_id[portid],
1237                                                 qconf->tx_buffer[portid]);
1238                         }
1239                         prev_tsc = cur_tsc;
1240                 }
1241
1242                 /*
1243                  * Read packet from RX queues
1244                  */
1245                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1246                         rx_queue = &(qconf->rx_queue_list[i]);
1247                         rx_queue->idle_hint = 0;
1248                         portid = rx_queue->port_id;
1249                         queueid = rx_queue->queue_id;
1250
1251                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1252                                         MAX_PKT_BURST);
1253
1254                         stats[lcore_id].nb_rx_processed += nb_rx;
1255
1256                         if (nb_rx == 0) {
1257
1258                                 rte_power_empty_poll_stat_update(lcore_id);
1259
1260                                 continue;
1261                         } else {
1262                                 rte_power_poll_stat_update(lcore_id, nb_rx);
1263                         }
1264
1265
1266                         /* Prefetch first packets */
1267                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1268                                 rte_prefetch0(rte_pktmbuf_mtod(
1269                                                         pkts_burst[j], void *));
1270                         }
1271
1272                         /* Prefetch and forward already prefetched packets */
1273                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1274                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1275                                                         j + PREFETCH_OFFSET],
1276                                                         void *));
1277                                 l3fwd_simple_forward(pkts_burst[j], portid,
1278                                                 qconf);
1279                         }
1280
1281                         /* Forward remaining prefetched packets */
1282                         for (; j < nb_rx; j++) {
1283                                 l3fwd_simple_forward(pkts_burst[j], portid,
1284                                                 qconf);
1285                         }
1286
1287                 }
1288
1289         }
1290
1291         return 0;
1292 }
1293 /* main processing loop */
1294 static int
1295 main_legacy_loop(__rte_unused void *dummy)
1296 {
1297         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1298         unsigned lcore_id;
1299         uint64_t prev_tsc, diff_tsc, cur_tsc, tim_res_tsc, hz;
1300         uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
1301         int i, j, nb_rx;
1302         uint8_t queueid;
1303         uint16_t portid;
1304         struct lcore_conf *qconf;
1305         struct lcore_rx_queue *rx_queue;
1306         enum freq_scale_hint_t lcore_scaleup_hint;
1307         uint32_t lcore_rx_idle_count = 0;
1308         uint32_t lcore_idle_hint = 0;
1309         int intr_en = 0;
1310
1311         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
1312
1313         prev_tsc = 0;
1314         hz = rte_get_timer_hz();
1315         tim_res_tsc = hz/TIMER_NUMBER_PER_SECOND;
1316
1317         lcore_id = rte_lcore_id();
1318         qconf = &lcore_conf[lcore_id];
1319
1320         if (qconf->n_rx_queue == 0) {
1321                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
1322                 return 0;
1323         }
1324
1325         RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
1326
1327         for (i = 0; i < qconf->n_rx_queue; i++) {
1328                 portid = qconf->rx_queue_list[i].port_id;
1329                 queueid = qconf->rx_queue_list[i].queue_id;
1330                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1331                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
1332         }
1333
1334         /* add into event wait list */
1335         if (event_register(qconf) == 0)
1336                 intr_en = 1;
1337         else
1338                 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
1339
1340         while (!is_done()) {
1341                 stats[lcore_id].nb_iteration_looped++;
1342
1343                 cur_tsc = rte_rdtsc();
1344                 cur_tsc_power = cur_tsc;
1345
1346                 /*
1347                  * TX burst queue drain
1348                  */
1349                 diff_tsc = cur_tsc - prev_tsc;
1350                 if (unlikely(diff_tsc > drain_tsc)) {
1351                         for (i = 0; i < qconf->n_tx_port; ++i) {
1352                                 portid = qconf->tx_port_id[i];
1353                                 rte_eth_tx_buffer_flush(portid,
1354                                                 qconf->tx_queue_id[portid],
1355                                                 qconf->tx_buffer[portid]);
1356                         }
1357                         prev_tsc = cur_tsc;
1358                 }
1359
1360                 diff_tsc_power = cur_tsc_power - prev_tsc_power;
1361                 if (diff_tsc_power > tim_res_tsc) {
1362                         rte_timer_manage();
1363                         prev_tsc_power = cur_tsc_power;
1364                 }
1365
1366 start_rx:
1367                 /*
1368                  * Read packet from RX queues
1369                  */
1370                 lcore_scaleup_hint = FREQ_CURRENT;
1371                 lcore_rx_idle_count = 0;
1372                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1373                         rx_queue = &(qconf->rx_queue_list[i]);
1374                         rx_queue->idle_hint = 0;
1375                         portid = rx_queue->port_id;
1376                         queueid = rx_queue->queue_id;
1377
1378                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1379                                                                 MAX_PKT_BURST);
1380
1381                         stats[lcore_id].nb_rx_processed += nb_rx;
1382                         if (unlikely(nb_rx == 0)) {
1383                                 /**
1384                                  * no packet received from rx queue, try to
1385                                  * sleep for a while forcing CPU enter deeper
1386                                  * C states.
1387                                  */
1388                                 rx_queue->zero_rx_packet_count++;
1389
1390                                 if (rx_queue->zero_rx_packet_count <=
1391                                                         MIN_ZERO_POLL_COUNT)
1392                                         continue;
1393
1394                                 rx_queue->idle_hint = power_idle_heuristic(\
1395                                         rx_queue->zero_rx_packet_count);
1396                                 lcore_rx_idle_count++;
1397                         } else {
1398                                 rx_queue->zero_rx_packet_count = 0;
1399
1400                                 /**
1401                                  * do not scale up frequency immediately as
1402                                  * user to kernel space communication is costly
1403                                  * which might impact packet I/O for received
1404                                  * packets.
1405                                  */
1406                                 rx_queue->freq_up_hint =
1407                                         power_freq_scaleup_heuristic(lcore_id,
1408                                                         portid, queueid);
1409                         }
1410
1411                         /* Prefetch first packets */
1412                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1413                                 rte_prefetch0(rte_pktmbuf_mtod(
1414                                                 pkts_burst[j], void *));
1415                         }
1416
1417                         /* Prefetch and forward already prefetched packets */
1418                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1419                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1420                                                 j + PREFETCH_OFFSET], void *));
1421                                 l3fwd_simple_forward(pkts_burst[j], portid,
1422                                                                 qconf);
1423                         }
1424
1425                         /* Forward remaining prefetched packets */
1426                         for (; j < nb_rx; j++) {
1427                                 l3fwd_simple_forward(pkts_burst[j], portid,
1428                                                                 qconf);
1429                         }
1430                 }
1431
1432                 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
1433                         for (i = 1, lcore_scaleup_hint =
1434                                 qconf->rx_queue_list[0].freq_up_hint;
1435                                         i < qconf->n_rx_queue; ++i) {
1436                                 rx_queue = &(qconf->rx_queue_list[i]);
1437                                 if (rx_queue->freq_up_hint >
1438                                                 lcore_scaleup_hint)
1439                                         lcore_scaleup_hint =
1440                                                 rx_queue->freq_up_hint;
1441                         }
1442
1443                         if (lcore_scaleup_hint == FREQ_HIGHEST) {
1444                                 if (rte_power_freq_max)
1445                                         rte_power_freq_max(lcore_id);
1446                         } else if (lcore_scaleup_hint == FREQ_HIGHER) {
1447                                 if (rte_power_freq_up)
1448                                         rte_power_freq_up(lcore_id);
1449                         }
1450                 } else {
1451                         /**
1452                          * All Rx queues empty in recent consecutive polls,
1453                          * sleep in a conservative manner, meaning sleep as
1454                          * less as possible.
1455                          */
1456                         for (i = 1, lcore_idle_hint =
1457                                 qconf->rx_queue_list[0].idle_hint;
1458                                         i < qconf->n_rx_queue; ++i) {
1459                                 rx_queue = &(qconf->rx_queue_list[i]);
1460                                 if (rx_queue->idle_hint < lcore_idle_hint)
1461                                         lcore_idle_hint = rx_queue->idle_hint;
1462                         }
1463
1464                         if (lcore_idle_hint < SUSPEND_THRESHOLD)
1465                                 /**
1466                                  * execute "pause" instruction to avoid context
1467                                  * switch which generally take hundred of
1468                                  * microseconds for short sleep.
1469                                  */
1470                                 rte_delay_us(lcore_idle_hint);
1471                         else {
1472                                 /* suspend until rx interrupt triggers */
1473                                 if (intr_en) {
1474                                         turn_on_off_intr(qconf, 1);
1475                                         sleep_until_rx_interrupt(
1476                                                 qconf->n_rx_queue);
1477                                         turn_on_off_intr(qconf, 0);
1478                                         /**
1479                                          * start receiving packets immediately
1480                                          */
1481                                         if (likely(!is_done()))
1482                                                 goto start_rx;
1483                                 }
1484                         }
1485                         stats[lcore_id].sleep_time += lcore_idle_hint;
1486                 }
1487         }
1488
1489         return 0;
1490 }
1491
1492 static int
1493 check_lcore_params(void)
1494 {
1495         uint8_t queue, lcore;
1496         uint16_t i;
1497         int socketid;
1498
1499         for (i = 0; i < nb_lcore_params; ++i) {
1500                 queue = lcore_params[i].queue_id;
1501                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1502                         printf("invalid queue number: %hhu\n", queue);
1503                         return -1;
1504                 }
1505                 lcore = lcore_params[i].lcore_id;
1506                 if (!rte_lcore_is_enabled(lcore)) {
1507                         printf("error: lcore %hhu is not enabled in lcore "
1508                                                         "mask\n", lcore);
1509                         return -1;
1510                 }
1511                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1512                                                         (numa_on == 0)) {
1513                         printf("warning: lcore %hhu is on socket %d with numa "
1514                                                 "off\n", lcore, socketid);
1515                 }
1516                 if (app_mode == APP_MODE_TELEMETRY && lcore == rte_lcore_id()) {
1517                         printf("cannot enable master core %d in config for telemetry mode\n",
1518                                 rte_lcore_id());
1519                         return -1;
1520                 }
1521         }
1522         return 0;
1523 }
1524
1525 static int
1526 check_port_config(void)
1527 {
1528         unsigned portid;
1529         uint16_t i;
1530
1531         for (i = 0; i < nb_lcore_params; ++i) {
1532                 portid = lcore_params[i].port_id;
1533                 if ((enabled_port_mask & (1 << portid)) == 0) {
1534                         printf("port %u is not enabled in port mask\n",
1535                                                                 portid);
1536                         return -1;
1537                 }
1538                 if (!rte_eth_dev_is_valid_port(portid)) {
1539                         printf("port %u is not present on the board\n",
1540                                                                 portid);
1541                         return -1;
1542                 }
1543         }
1544         return 0;
1545 }
1546
1547 static uint8_t
1548 get_port_n_rx_queues(const uint16_t port)
1549 {
1550         int queue = -1;
1551         uint16_t i;
1552
1553         for (i = 0; i < nb_lcore_params; ++i) {
1554                 if (lcore_params[i].port_id == port &&
1555                                 lcore_params[i].queue_id > queue)
1556                         queue = lcore_params[i].queue_id;
1557         }
1558         return (uint8_t)(++queue);
1559 }
1560
1561 static int
1562 init_lcore_rx_queues(void)
1563 {
1564         uint16_t i, nb_rx_queue;
1565         uint8_t lcore;
1566
1567         for (i = 0; i < nb_lcore_params; ++i) {
1568                 lcore = lcore_params[i].lcore_id;
1569                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1570                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1571                         printf("error: too many queues (%u) for lcore: %u\n",
1572                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1573                         return -1;
1574                 } else {
1575                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1576                                 lcore_params[i].port_id;
1577                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1578                                 lcore_params[i].queue_id;
1579                         lcore_conf[lcore].n_rx_queue++;
1580                 }
1581         }
1582         return 0;
1583 }
1584
1585 /* display usage */
1586 static void
1587 print_usage(const char *prgname)
1588 {
1589         printf ("%s [EAL options] -- -p PORTMASK -P"
1590                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1591                 "  [--high-perf-cores CORELIST"
1592                 "  [--perf-config (port,queue,hi_perf,lcore_index)[,(port,queue,hi_perf,lcore_index]]"
1593                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1594                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1595                 "  -P : enable promiscuous mode\n"
1596                 "  --config (port,queue,lcore): rx queues configuration\n"
1597                 "  --high-perf-cores CORELIST: list of high performance cores\n"
1598                 "  --perf-config: similar as config, cores specified as indices"
1599                 " for bins containing high or regular performance cores\n"
1600                 "  --no-numa: optional, disable numa awareness\n"
1601                 "  --enable-jumbo: enable jumbo frame"
1602                 " which max packet len is PKTLEN in decimal (64-9600)\n"
1603                 "  --parse-ptype: parse packet type by software\n"
1604                 "  --legacy: use legacy interrupt-based scaling\n"
1605                 "  --empty-poll: enable empty poll detection"
1606                 " follow (training_flag, high_threshold, med_threshold)\n"
1607                 " --telemetry: enable telemetry mode, to update"
1608                 " empty polls, full polls, and core busyness to telemetry\n"
1609                 " --interrupt-only: enable interrupt-only mode\n",
1610                 prgname);
1611 }
1612
1613 static int parse_max_pkt_len(const char *pktlen)
1614 {
1615         char *end = NULL;
1616         unsigned long len;
1617
1618         /* parse decimal string */
1619         len = strtoul(pktlen, &end, 10);
1620         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1621                 return -1;
1622
1623         if (len == 0)
1624                 return -1;
1625
1626         return len;
1627 }
1628
1629 static int
1630 parse_portmask(const char *portmask)
1631 {
1632         char *end = NULL;
1633         unsigned long pm;
1634
1635         /* parse hexadecimal string */
1636         pm = strtoul(portmask, &end, 16);
1637         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1638                 return 0;
1639
1640         return pm;
1641 }
1642
1643 static int
1644 parse_config(const char *q_arg)
1645 {
1646         char s[256];
1647         const char *p, *p0 = q_arg;
1648         char *end;
1649         enum fieldnames {
1650                 FLD_PORT = 0,
1651                 FLD_QUEUE,
1652                 FLD_LCORE,
1653                 _NUM_FLD
1654         };
1655         unsigned long int_fld[_NUM_FLD];
1656         char *str_fld[_NUM_FLD];
1657         int i;
1658         unsigned size;
1659
1660         nb_lcore_params = 0;
1661
1662         while ((p = strchr(p0,'(')) != NULL) {
1663                 ++p;
1664                 if((p0 = strchr(p,')')) == NULL)
1665                         return -1;
1666
1667                 size = p0 - p;
1668                 if(size >= sizeof(s))
1669                         return -1;
1670
1671                 snprintf(s, sizeof(s), "%.*s", size, p);
1672                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1673                                                                 _NUM_FLD)
1674                         return -1;
1675                 for (i = 0; i < _NUM_FLD; i++){
1676                         errno = 0;
1677                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1678                         if (errno != 0 || end == str_fld[i] || int_fld[i] >
1679                                                                         255)
1680                                 return -1;
1681                 }
1682                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1683                         printf("exceeded max number of lcore params: %hu\n",
1684                                 nb_lcore_params);
1685                         return -1;
1686                 }
1687                 lcore_params_array[nb_lcore_params].port_id =
1688                                 (uint8_t)int_fld[FLD_PORT];
1689                 lcore_params_array[nb_lcore_params].queue_id =
1690                                 (uint8_t)int_fld[FLD_QUEUE];
1691                 lcore_params_array[nb_lcore_params].lcore_id =
1692                                 (uint8_t)int_fld[FLD_LCORE];
1693                 ++nb_lcore_params;
1694         }
1695         lcore_params = lcore_params_array;
1696
1697         return 0;
1698 }
1699 static int
1700 parse_ep_config(const char *q_arg)
1701 {
1702         char s[256];
1703         const char *p = q_arg;
1704         char *end;
1705         int  num_arg;
1706
1707         char *str_fld[3];
1708
1709         int training_flag;
1710         int med_edpi;
1711         int hgh_edpi;
1712
1713         ep_med_edpi = EMPTY_POLL_MED_THRESHOLD;
1714         ep_hgh_edpi = EMPTY_POLL_MED_THRESHOLD;
1715
1716         strlcpy(s, p, sizeof(s));
1717
1718         num_arg = rte_strsplit(s, sizeof(s), str_fld, 3, ',');
1719
1720         empty_poll_train = false;
1721
1722         if (num_arg == 0)
1723                 return 0;
1724
1725         if (num_arg == 3) {
1726
1727                 training_flag = strtoul(str_fld[0], &end, 0);
1728                 med_edpi = strtoul(str_fld[1], &end, 0);
1729                 hgh_edpi = strtoul(str_fld[2], &end, 0);
1730
1731                 if (training_flag == 1)
1732                         empty_poll_train = true;
1733
1734                 if (med_edpi > 0)
1735                         ep_med_edpi = med_edpi;
1736
1737                 if (med_edpi > 0)
1738                         ep_hgh_edpi = hgh_edpi;
1739
1740         } else {
1741
1742                 return -1;
1743         }
1744
1745         return 0;
1746
1747 }
1748 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1749 #define CMD_LINE_OPT_LEGACY "legacy"
1750 #define CMD_LINE_OPT_EMPTY_POLL "empty-poll"
1751 #define CMD_LINE_OPT_INTERRUPT_ONLY "interrupt-only"
1752 #define CMD_LINE_OPT_TELEMETRY "telemetry"
1753
1754 /* Parse the argument given in the command line of the application */
1755 static int
1756 parse_args(int argc, char **argv)
1757 {
1758         int opt, ret;
1759         char **argvopt;
1760         int option_index;
1761         uint32_t limit;
1762         char *prgname = argv[0];
1763         static struct option lgopts[] = {
1764                 {"config", 1, 0, 0},
1765                 {"perf-config", 1, 0, 0},
1766                 {"high-perf-cores", 1, 0, 0},
1767                 {"no-numa", 0, 0, 0},
1768                 {"enable-jumbo", 0, 0, 0},
1769                 {CMD_LINE_OPT_EMPTY_POLL, 1, 0, 0},
1770                 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
1771                 {CMD_LINE_OPT_LEGACY, 0, 0, 0},
1772                 {CMD_LINE_OPT_TELEMETRY, 0, 0, 0},
1773                 {CMD_LINE_OPT_INTERRUPT_ONLY, 0, 0, 0},
1774                 {NULL, 0, 0, 0}
1775         };
1776
1777         argvopt = argv;
1778
1779         while ((opt = getopt_long(argc, argvopt, "p:l:m:h:P",
1780                                 lgopts, &option_index)) != EOF) {
1781
1782                 switch (opt) {
1783                 /* portmask */
1784                 case 'p':
1785                         enabled_port_mask = parse_portmask(optarg);
1786                         if (enabled_port_mask == 0) {
1787                                 printf("invalid portmask\n");
1788                                 print_usage(prgname);
1789                                 return -1;
1790                         }
1791                         break;
1792                 case 'P':
1793                         printf("Promiscuous mode selected\n");
1794                         promiscuous_on = 1;
1795                         break;
1796                 case 'l':
1797                         limit = parse_max_pkt_len(optarg);
1798                         freq_tlb[LOW] = limit;
1799                         break;
1800                 case 'm':
1801                         limit = parse_max_pkt_len(optarg);
1802                         freq_tlb[MED] = limit;
1803                         break;
1804                 case 'h':
1805                         limit = parse_max_pkt_len(optarg);
1806                         freq_tlb[HGH] = limit;
1807                         break;
1808                 /* long options */
1809                 case 0:
1810                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1811                                 ret = parse_config(optarg);
1812                                 if (ret) {
1813                                         printf("invalid config\n");
1814                                         print_usage(prgname);
1815                                         return -1;
1816                                 }
1817                         }
1818
1819                         if (!strncmp(lgopts[option_index].name,
1820                                         "perf-config", 11)) {
1821                                 ret = parse_perf_config(optarg);
1822                                 if (ret) {
1823                                         printf("invalid perf-config\n");
1824                                         print_usage(prgname);
1825                                         return -1;
1826                                 }
1827                         }
1828
1829                         if (!strncmp(lgopts[option_index].name,
1830                                         "high-perf-cores", 15)) {
1831                                 ret = parse_perf_core_list(optarg);
1832                                 if (ret) {
1833                                         printf("invalid high-perf-cores\n");
1834                                         print_usage(prgname);
1835                                         return -1;
1836                                 }
1837                         }
1838
1839                         if (!strncmp(lgopts[option_index].name,
1840                                                 "no-numa", 7)) {
1841                                 printf("numa is disabled \n");
1842                                 numa_on = 0;
1843                         }
1844
1845                         if (!strncmp(lgopts[option_index].name,
1846                                         CMD_LINE_OPT_LEGACY,
1847                                         sizeof(CMD_LINE_OPT_LEGACY))) {
1848                                 if (app_mode != APP_MODE_DEFAULT) {
1849                                         printf(" legacy mode is mutually exclusive with other modes\n");
1850                                         return -1;
1851                                 }
1852                                 app_mode = APP_MODE_LEGACY;
1853                                 printf("legacy mode is enabled\n");
1854                         }
1855
1856                         if (!strncmp(lgopts[option_index].name,
1857                                         CMD_LINE_OPT_EMPTY_POLL, 10)) {
1858                                 if (app_mode != APP_MODE_DEFAULT) {
1859                                         printf(" empty-poll mode is mutually exclusive with other modes\n");
1860                                         return -1;
1861                                 }
1862                                 app_mode = APP_MODE_EMPTY_POLL;
1863                                 ret = parse_ep_config(optarg);
1864
1865                                 if (ret) {
1866                                         printf("invalid empty poll config\n");
1867                                         print_usage(prgname);
1868                                         return -1;
1869                                 }
1870                                 printf("empty-poll is enabled\n");
1871                         }
1872
1873                         if (!strncmp(lgopts[option_index].name,
1874                                         CMD_LINE_OPT_TELEMETRY,
1875                                         sizeof(CMD_LINE_OPT_TELEMETRY))) {
1876                                 if (app_mode != APP_MODE_DEFAULT) {
1877                                         printf(" telemetry mode is mutually exclusive with other modes\n");
1878                                         return -1;
1879                                 }
1880                                 app_mode = APP_MODE_TELEMETRY;
1881                                 printf("telemetry mode is enabled\n");
1882                         }
1883
1884                         if (!strncmp(lgopts[option_index].name,
1885                                         CMD_LINE_OPT_INTERRUPT_ONLY,
1886                                         sizeof(CMD_LINE_OPT_INTERRUPT_ONLY))) {
1887                                 if (app_mode != APP_MODE_DEFAULT) {
1888                                         printf(" interrupt-only mode is mutually exclusive with other modes\n");
1889                                         return -1;
1890                                 }
1891                                 app_mode = APP_MODE_INTERRUPT;
1892                                 printf("interrupt-only mode is enabled\n");
1893                         }
1894
1895                         if (!strncmp(lgopts[option_index].name,
1896                                         "enable-jumbo", 12)) {
1897                                 struct option lenopts =
1898                                         {"max-pkt-len", required_argument, \
1899                                                                         0, 0};
1900
1901                                 printf("jumbo frame is enabled \n");
1902                                 port_conf.rxmode.offloads |=
1903                                                 DEV_RX_OFFLOAD_JUMBO_FRAME;
1904                                 port_conf.txmode.offloads |=
1905                                                 DEV_TX_OFFLOAD_MULTI_SEGS;
1906
1907                                 /**
1908                                  * if no max-pkt-len set, use the default value
1909                                  * RTE_ETHER_MAX_LEN
1910                                  */
1911                                 if (0 == getopt_long(argc, argvopt, "",
1912                                                 &lenopts, &option_index)) {
1913                                         ret = parse_max_pkt_len(optarg);
1914                                         if ((ret < 64) ||
1915                                                 (ret > MAX_JUMBO_PKT_LEN)){
1916                                                 printf("invalid packet "
1917                                                                 "length\n");
1918                                                 print_usage(prgname);
1919                                                 return -1;
1920                                         }
1921                                         port_conf.rxmode.max_rx_pkt_len = ret;
1922                                 }
1923                                 printf("set jumbo frame "
1924                                         "max packet length to %u\n",
1925                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1926                         }
1927
1928                         if (!strncmp(lgopts[option_index].name,
1929                                      CMD_LINE_OPT_PARSE_PTYPE,
1930                                      sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
1931                                 printf("soft parse-ptype is enabled\n");
1932                                 parse_ptype = 1;
1933                         }
1934
1935                         break;
1936
1937                 default:
1938                         print_usage(prgname);
1939                         return -1;
1940                 }
1941         }
1942
1943         if (optind >= 0)
1944                 argv[optind-1] = prgname;
1945
1946         ret = optind-1;
1947         optind = 1; /* reset getopt lib */
1948         return ret;
1949 }
1950
1951 static void
1952 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1953 {
1954         char buf[RTE_ETHER_ADDR_FMT_SIZE];
1955         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
1956         printf("%s%s", name, buf);
1957 }
1958
1959 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1960 static void
1961 setup_hash(int socketid)
1962 {
1963         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1964                 .name = NULL,
1965                 .entries = L3FWD_HASH_ENTRIES,
1966                 .key_len = sizeof(struct ipv4_5tuple),
1967                 .hash_func = DEFAULT_HASH_FUNC,
1968                 .hash_func_init_val = 0,
1969         };
1970
1971         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1972                 .name = NULL,
1973                 .entries = L3FWD_HASH_ENTRIES,
1974                 .key_len = sizeof(struct ipv6_5tuple),
1975                 .hash_func = DEFAULT_HASH_FUNC,
1976                 .hash_func_init_val = 0,
1977         };
1978
1979         unsigned i;
1980         int ret;
1981         char s[64];
1982
1983         /* create ipv4 hash */
1984         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1985         ipv4_l3fwd_hash_params.name = s;
1986         ipv4_l3fwd_hash_params.socket_id = socketid;
1987         ipv4_l3fwd_lookup_struct[socketid] =
1988                 rte_hash_create(&ipv4_l3fwd_hash_params);
1989         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1990                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1991                                 "socket %d\n", socketid);
1992
1993         /* create ipv6 hash */
1994         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1995         ipv6_l3fwd_hash_params.name = s;
1996         ipv6_l3fwd_hash_params.socket_id = socketid;
1997         ipv6_l3fwd_lookup_struct[socketid] =
1998                 rte_hash_create(&ipv6_l3fwd_hash_params);
1999         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
2000                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2001                                 "socket %d\n", socketid);
2002
2003
2004         /* populate the ipv4 hash */
2005         for (i = 0; i < RTE_DIM(ipv4_l3fwd_route_array); i++) {
2006                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
2007                                 (void *) &ipv4_l3fwd_route_array[i].key);
2008                 if (ret < 0) {
2009                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
2010                                 "l3fwd hash on socket %d\n", i, socketid);
2011                 }
2012                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
2013                 printf("Hash: Adding key\n");
2014                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
2015         }
2016
2017         /* populate the ipv6 hash */
2018         for (i = 0; i < RTE_DIM(ipv6_l3fwd_route_array); i++) {
2019                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
2020                                 (void *) &ipv6_l3fwd_route_array[i].key);
2021                 if (ret < 0) {
2022                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
2023                                 "l3fwd hash on socket %d\n", i, socketid);
2024                 }
2025                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
2026                 printf("Hash: Adding key\n");
2027                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
2028         }
2029 }
2030 #endif
2031
2032 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2033 static void
2034 setup_lpm(int socketid)
2035 {
2036         unsigned i;
2037         int ret;
2038         char s[64];
2039
2040         /* create the LPM table */
2041         struct rte_lpm_config lpm_ipv4_config;
2042
2043         lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
2044         lpm_ipv4_config.number_tbl8s = 256;
2045         lpm_ipv4_config.flags = 0;
2046
2047         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
2048         ipv4_l3fwd_lookup_struct[socketid] =
2049                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
2050         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2051                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
2052                                 " on socket %d\n", socketid);
2053
2054         /* populate the LPM table */
2055         for (i = 0; i < RTE_DIM(ipv4_l3fwd_route_array); i++) {
2056                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
2057                         ipv4_l3fwd_route_array[i].ip,
2058                         ipv4_l3fwd_route_array[i].depth,
2059                         ipv4_l3fwd_route_array[i].if_out);
2060
2061                 if (ret < 0) {
2062                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
2063                                 "l3fwd LPM table on socket %d\n",
2064                                 i, socketid);
2065                 }
2066
2067                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
2068                         (unsigned)ipv4_l3fwd_route_array[i].ip,
2069                         ipv4_l3fwd_route_array[i].depth,
2070                         ipv4_l3fwd_route_array[i].if_out);
2071         }
2072 }
2073 #endif
2074
2075 static int
2076 init_mem(unsigned nb_mbuf)
2077 {
2078         struct lcore_conf *qconf;
2079         int socketid;
2080         unsigned lcore_id;
2081         char s[64];
2082
2083         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2084                 if (rte_lcore_is_enabled(lcore_id) == 0)
2085                         continue;
2086
2087                 if (numa_on)
2088                         socketid = rte_lcore_to_socket_id(lcore_id);
2089                 else
2090                         socketid = 0;
2091
2092                 if (socketid >= NB_SOCKETS) {
2093                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
2094                                         "out of range %d\n", socketid,
2095                                                 lcore_id, NB_SOCKETS);
2096                 }
2097                 if (pktmbuf_pool[socketid] == NULL) {
2098                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
2099                         pktmbuf_pool[socketid] =
2100                                 rte_pktmbuf_pool_create(s, nb_mbuf,
2101                                         MEMPOOL_CACHE_SIZE, 0,
2102                                         RTE_MBUF_DEFAULT_BUF_SIZE,
2103                                         socketid);
2104                         if (pktmbuf_pool[socketid] == NULL)
2105                                 rte_exit(EXIT_FAILURE,
2106                                         "Cannot init mbuf pool on socket %d\n",
2107                                                                 socketid);
2108                         else
2109                                 printf("Allocated mbuf pool on socket %d\n",
2110                                                                 socketid);
2111
2112 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2113                         setup_lpm(socketid);
2114 #else
2115                         setup_hash(socketid);
2116 #endif
2117                 }
2118                 qconf = &lcore_conf[lcore_id];
2119                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
2120 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2121                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
2122 #endif
2123         }
2124         return 0;
2125 }
2126
2127 /* Check the link status of all ports in up to 9s, and print them finally */
2128 static void
2129 check_all_ports_link_status(uint32_t port_mask)
2130 {
2131 #define CHECK_INTERVAL 100 /* 100ms */
2132 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
2133         uint8_t count, all_ports_up, print_flag = 0;
2134         uint16_t portid;
2135         struct rte_eth_link link;
2136         int ret;
2137
2138         printf("\nChecking link status");
2139         fflush(stdout);
2140         for (count = 0; count <= MAX_CHECK_TIME; count++) {
2141                 all_ports_up = 1;
2142                 RTE_ETH_FOREACH_DEV(portid) {
2143                         if ((port_mask & (1 << portid)) == 0)
2144                                 continue;
2145                         memset(&link, 0, sizeof(link));
2146                         ret = rte_eth_link_get_nowait(portid, &link);
2147                         if (ret < 0) {
2148                                 all_ports_up = 0;
2149                                 if (print_flag == 1)
2150                                         printf("Port %u link get failed: %s\n",
2151                                                 portid, rte_strerror(-ret));
2152                                 continue;
2153                         }
2154                         /* print link status if flag set */
2155                         if (print_flag == 1) {
2156                                 if (link.link_status)
2157                                         printf("Port %d Link Up - speed %u "
2158                                                 "Mbps - %s\n", (uint8_t)portid,
2159                                                 (unsigned)link.link_speed,
2160                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
2161                                         ("full-duplex") : ("half-duplex"));
2162                                 else
2163                                         printf("Port %d Link Down\n",
2164                                                 (uint8_t)portid);
2165                                 continue;
2166                         }
2167                         /* clear all_ports_up flag if any link down */
2168                         if (link.link_status == ETH_LINK_DOWN) {
2169                                 all_ports_up = 0;
2170                                 break;
2171                         }
2172                 }
2173                 /* after finally printing all link status, get out */
2174                 if (print_flag == 1)
2175                         break;
2176
2177                 if (all_ports_up == 0) {
2178                         printf(".");
2179                         fflush(stdout);
2180                         rte_delay_ms(CHECK_INTERVAL);
2181                 }
2182
2183                 /* set the print_flag if all ports up or timeout */
2184                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
2185                         print_flag = 1;
2186                         printf("done\n");
2187                 }
2188         }
2189 }
2190
2191 static int check_ptype(uint16_t portid)
2192 {
2193         int i, ret;
2194         int ptype_l3_ipv4 = 0;
2195 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2196         int ptype_l3_ipv6 = 0;
2197 #endif
2198         uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
2199
2200         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
2201         if (ret <= 0)
2202                 return 0;
2203
2204         uint32_t ptypes[ret];
2205
2206         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
2207         for (i = 0; i < ret; ++i) {
2208                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
2209                         ptype_l3_ipv4 = 1;
2210 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2211                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
2212                         ptype_l3_ipv6 = 1;
2213 #endif
2214         }
2215
2216         if (ptype_l3_ipv4 == 0)
2217                 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
2218
2219 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2220         if (ptype_l3_ipv6 == 0)
2221                 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
2222 #endif
2223
2224 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2225         if (ptype_l3_ipv4)
2226 #else /* APP_LOOKUP_EXACT_MATCH */
2227         if (ptype_l3_ipv4 && ptype_l3_ipv6)
2228 #endif
2229                 return 1;
2230
2231         return 0;
2232
2233 }
2234
2235 static int
2236 init_power_library(void)
2237 {
2238         enum power_management_env env;
2239         unsigned int lcore_id;
2240         int ret = 0;
2241
2242         RTE_LCORE_FOREACH(lcore_id) {
2243                 /* init power management library */
2244                 ret = rte_power_init(lcore_id);
2245                 if (ret) {
2246                         RTE_LOG(ERR, POWER,
2247                                 "Library initialization failed on core %u\n",
2248                                 lcore_id);
2249                         return ret;
2250                 }
2251                 /* we're not supporting the VM channel mode */
2252                 env = rte_power_get_env();
2253                 if (env != PM_ENV_ACPI_CPUFREQ &&
2254                                 env != PM_ENV_PSTATE_CPUFREQ) {
2255                         RTE_LOG(ERR, POWER,
2256                                 "Only ACPI and PSTATE mode are supported\n");
2257                         return -1;
2258                 }
2259         }
2260         return ret;
2261 }
2262
2263 static int
2264 deinit_power_library(void)
2265 {
2266         unsigned int lcore_id;
2267         int ret = 0;
2268
2269         RTE_LCORE_FOREACH(lcore_id) {
2270                 /* deinit power management library */
2271                 ret = rte_power_exit(lcore_id);
2272                 if (ret) {
2273                         RTE_LOG(ERR, POWER,
2274                                 "Library deinitialization failed on core %u\n",
2275                                 lcore_id);
2276                         return ret;
2277                 }
2278         }
2279         return ret;
2280 }
2281
2282 static void
2283 get_current_stat_values(uint64_t *values)
2284 {
2285         unsigned int lcore_id = rte_lcore_id();
2286         struct lcore_conf *qconf;
2287         uint64_t app_eps = 0, app_fps = 0, app_br = 0;
2288         uint64_t count = 0;
2289
2290         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2291                 qconf = &lcore_conf[lcore_id];
2292                 if (qconf->n_rx_queue == 0)
2293                         continue;
2294                 count++;
2295                 rte_spinlock_lock(&stats[lcore_id].telemetry_lock);
2296                 app_eps += stats[lcore_id].ep_nep[1];
2297                 app_fps += stats[lcore_id].fp_nfp[1];
2298                 app_br += stats[lcore_id].br;
2299                 rte_spinlock_unlock(&stats[lcore_id].telemetry_lock);
2300         }
2301
2302         if (count > 0) {
2303                 values[0] = app_eps/count;
2304                 values[1] = app_fps/count;
2305                 values[2] = app_br/count;
2306         } else
2307                 memset(values, 0, sizeof(uint64_t) * NUM_TELSTATS);
2308
2309 }
2310
2311 static void
2312 update_telemetry(__rte_unused struct rte_timer *tim,
2313                 __rte_unused void *arg)
2314 {
2315         int ret;
2316         uint64_t values[NUM_TELSTATS] = {0};
2317
2318         get_current_stat_values(values);
2319         ret = rte_metrics_update_values(RTE_METRICS_GLOBAL, telstats_index,
2320                                         values, RTE_DIM(values));
2321         if (ret < 0)
2322                 RTE_LOG(WARNING, POWER, "failed to update metrcis\n");
2323 }
2324
2325 static int
2326 handle_app_stats(const char *cmd __rte_unused,
2327                 const char *params __rte_unused,
2328                 struct rte_tel_data *d)
2329 {
2330         uint64_t values[NUM_TELSTATS] = {0};
2331         uint32_t i;
2332
2333         rte_tel_data_start_dict(d);
2334         get_current_stat_values(values);
2335         for (i = 0; i < NUM_TELSTATS; i++)
2336                 rte_tel_data_add_dict_u64(d, telstats_strings[i].name,
2337                                 values[i]);
2338         return 0;
2339 }
2340
2341 static void
2342 telemetry_setup_timer(void)
2343 {
2344         int lcore_id = rte_lcore_id();
2345         uint64_t hz = rte_get_timer_hz();
2346         uint64_t ticks;
2347
2348         ticks = hz / TELEMETRY_INTERVALS_PER_SEC;
2349         rte_timer_reset_sync(&telemetry_timer,
2350                         ticks,
2351                         PERIODICAL,
2352                         lcore_id,
2353                         update_telemetry,
2354                         NULL);
2355 }
2356 static void
2357 empty_poll_setup_timer(void)
2358 {
2359         int lcore_id = rte_lcore_id();
2360         uint64_t hz = rte_get_timer_hz();
2361
2362         struct  ep_params *ep_ptr = ep_params;
2363
2364         ep_ptr->interval_ticks = hz / INTERVALS_PER_SECOND;
2365
2366         rte_timer_reset_sync(&ep_ptr->timer0,
2367                         ep_ptr->interval_ticks,
2368                         PERIODICAL,
2369                         lcore_id,
2370                         rte_empty_poll_detection,
2371                         (void *)ep_ptr);
2372
2373 }
2374 static int
2375 launch_timer(unsigned int lcore_id)
2376 {
2377         int64_t prev_tsc = 0, cur_tsc, diff_tsc, cycles_10ms;
2378
2379         RTE_SET_USED(lcore_id);
2380
2381
2382         if (rte_get_master_lcore() != lcore_id) {
2383                 rte_panic("timer on lcore:%d which is not master core:%d\n",
2384                                 lcore_id,
2385                                 rte_get_master_lcore());
2386         }
2387
2388         RTE_LOG(INFO, POWER, "Bring up the Timer\n");
2389
2390         if (app_mode == APP_MODE_EMPTY_POLL)
2391                 empty_poll_setup_timer();
2392         else
2393                 telemetry_setup_timer();
2394
2395         cycles_10ms = rte_get_timer_hz() / 100;
2396
2397         while (!is_done()) {
2398                 cur_tsc = rte_rdtsc();
2399                 diff_tsc = cur_tsc - prev_tsc;
2400                 if (diff_tsc > cycles_10ms) {
2401                         rte_timer_manage();
2402                         prev_tsc = cur_tsc;
2403                         cycles_10ms = rte_get_timer_hz() / 100;
2404                 }
2405         }
2406
2407         RTE_LOG(INFO, POWER, "Timer_subsystem is done\n");
2408
2409         return 0;
2410 }
2411
2412 static int
2413 autodetect_mode(void)
2414 {
2415         RTE_LOG(NOTICE, L3FWD_POWER, "Operating mode not specified, probing frequency scaling support...\n");
2416
2417         /*
2418          * Empty poll and telemetry modes have to be specifically requested to
2419          * be enabled, but we can auto-detect between interrupt mode with or
2420          * without frequency scaling. Both ACPI and pstate can be used.
2421          */
2422         if (rte_power_check_env_supported(PM_ENV_ACPI_CPUFREQ))
2423                 return APP_MODE_LEGACY;
2424         if (rte_power_check_env_supported(PM_ENV_PSTATE_CPUFREQ))
2425                 return APP_MODE_LEGACY;
2426
2427         RTE_LOG(NOTICE, L3FWD_POWER, "Frequency scaling not supported, selecting interrupt-only mode\n");
2428
2429         return APP_MODE_INTERRUPT;
2430 }
2431
2432 static const char *
2433 mode_to_str(enum appmode mode)
2434 {
2435         switch (mode) {
2436         case APP_MODE_LEGACY:
2437                 return "legacy";
2438         case APP_MODE_EMPTY_POLL:
2439                 return "empty poll";
2440         case APP_MODE_TELEMETRY:
2441                 return "telemetry";
2442         case APP_MODE_INTERRUPT:
2443                 return "interrupt-only";
2444         default:
2445                 return "invalid";
2446         }
2447 }
2448
2449 int
2450 main(int argc, char **argv)
2451 {
2452         struct lcore_conf *qconf;
2453         struct rte_eth_dev_info dev_info;
2454         struct rte_eth_txconf *txconf;
2455         int ret;
2456         uint16_t nb_ports;
2457         uint16_t queueid;
2458         unsigned lcore_id;
2459         uint64_t hz;
2460         uint32_t n_tx_queue, nb_lcores;
2461         uint32_t dev_rxq_num, dev_txq_num;
2462         uint8_t nb_rx_queue, queue, socketid;
2463         uint16_t portid;
2464         const char *ptr_strings[NUM_TELSTATS];
2465
2466         /* catch SIGINT and restore cpufreq governor to ondemand */
2467         signal(SIGINT, signal_exit_now);
2468
2469         /* init EAL */
2470         ret = rte_eal_init(argc, argv);
2471         if (ret < 0)
2472                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2473         argc -= ret;
2474         argv += ret;
2475
2476         /* init RTE timer library to be used late */
2477         rte_timer_subsystem_init();
2478
2479         /* parse application arguments (after the EAL ones) */
2480         ret = parse_args(argc, argv);
2481         if (ret < 0)
2482                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
2483
2484         if (app_mode == APP_MODE_DEFAULT)
2485                 app_mode = autodetect_mode();
2486
2487         RTE_LOG(INFO, L3FWD_POWER, "Selected operation mode: %s\n",
2488                         mode_to_str(app_mode));
2489
2490         /* only legacy and empty poll mode rely on power library */
2491         if ((app_mode == APP_MODE_LEGACY || app_mode == APP_MODE_EMPTY_POLL) &&
2492                         init_power_library())
2493                 rte_exit(EXIT_FAILURE, "init_power_library failed\n");
2494
2495         if (update_lcore_params() < 0)
2496                 rte_exit(EXIT_FAILURE, "update_lcore_params failed\n");
2497
2498         if (check_lcore_params() < 0)
2499                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
2500
2501         ret = init_lcore_rx_queues();
2502         if (ret < 0)
2503                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2504
2505         nb_ports = rte_eth_dev_count_avail();
2506
2507         if (check_port_config() < 0)
2508                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
2509
2510         nb_lcores = rte_lcore_count();
2511
2512         /* initialize all ports */
2513         RTE_ETH_FOREACH_DEV(portid) {
2514                 struct rte_eth_conf local_port_conf = port_conf;
2515                 /* not all app modes need interrupts */
2516                 bool need_intr = app_mode == APP_MODE_LEGACY ||
2517                                 app_mode == APP_MODE_INTERRUPT;
2518
2519                 /* skip ports that are not enabled */
2520                 if ((enabled_port_mask & (1 << portid)) == 0) {
2521                         printf("\nSkipping disabled port %d\n", portid);
2522                         continue;
2523                 }
2524
2525                 /* init port */
2526                 printf("Initializing port %d ... ", portid );
2527                 fflush(stdout);
2528
2529                 ret = rte_eth_dev_info_get(portid, &dev_info);
2530                 if (ret != 0)
2531                         rte_exit(EXIT_FAILURE,
2532                                 "Error during getting device (port %u) info: %s\n",
2533                                 portid, strerror(-ret));
2534
2535                 dev_rxq_num = dev_info.max_rx_queues;
2536                 dev_txq_num = dev_info.max_tx_queues;
2537
2538                 nb_rx_queue = get_port_n_rx_queues(portid);
2539                 if (nb_rx_queue > dev_rxq_num)
2540                         rte_exit(EXIT_FAILURE,
2541                                 "Cannot configure not existed rxq: "
2542                                 "port=%d\n", portid);
2543
2544                 n_tx_queue = nb_lcores;
2545                 if (n_tx_queue > dev_txq_num)
2546                         n_tx_queue = dev_txq_num;
2547                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
2548                         nb_rx_queue, (unsigned)n_tx_queue );
2549                 /* If number of Rx queue is 0, no need to enable Rx interrupt */
2550                 if (nb_rx_queue == 0)
2551                         need_intr = false;
2552
2553                 if (need_intr)
2554                         local_port_conf.intr_conf.rxq = 1;
2555
2556                 ret = rte_eth_dev_info_get(portid, &dev_info);
2557                 if (ret != 0)
2558                         rte_exit(EXIT_FAILURE,
2559                                 "Error during getting device (port %u) info: %s\n",
2560                                 portid, strerror(-ret));
2561
2562                 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
2563                         local_port_conf.txmode.offloads |=
2564                                 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
2565
2566                 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2567                         dev_info.flow_type_rss_offloads;
2568                 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2569                                 port_conf.rx_adv_conf.rss_conf.rss_hf) {
2570                         printf("Port %u modified RSS hash function based on hardware support,"
2571                                 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2572                                 portid,
2573                                 port_conf.rx_adv_conf.rss_conf.rss_hf,
2574                                 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2575                 }
2576
2577                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
2578                                         (uint16_t)n_tx_queue, &local_port_conf);
2579                 if (ret < 0)
2580                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
2581                                         "err=%d, port=%d\n", ret, portid);
2582
2583                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
2584                                                        &nb_txd);
2585                 if (ret < 0)
2586                         rte_exit(EXIT_FAILURE,
2587                                  "Cannot adjust number of descriptors: err=%d, port=%d\n",
2588                                  ret, portid);
2589
2590                 ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
2591                 if (ret < 0)
2592                         rte_exit(EXIT_FAILURE,
2593                                  "Cannot get MAC address: err=%d, port=%d\n",
2594                                  ret, portid);
2595
2596                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
2597                 printf(", ");
2598
2599                 /* init memory */
2600                 ret = init_mem(NB_MBUF);
2601                 if (ret < 0)
2602                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
2603
2604                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2605                         if (rte_lcore_is_enabled(lcore_id) == 0)
2606                                 continue;
2607
2608                         /* Initialize TX buffers */
2609                         qconf = &lcore_conf[lcore_id];
2610                         qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
2611                                 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
2612                                 rte_eth_dev_socket_id(portid));
2613                         if (qconf->tx_buffer[portid] == NULL)
2614                                 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
2615                                                  portid);
2616
2617                         rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
2618                 }
2619
2620                 /* init one TX queue per couple (lcore,port) */
2621                 queueid = 0;
2622                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2623                         if (rte_lcore_is_enabled(lcore_id) == 0)
2624                                 continue;
2625
2626                         if (queueid >= dev_txq_num)
2627                                 continue;
2628
2629                         if (numa_on)
2630                                 socketid = \
2631                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
2632                         else
2633                                 socketid = 0;
2634
2635                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
2636                         fflush(stdout);
2637
2638                         txconf = &dev_info.default_txconf;
2639                         txconf->offloads = local_port_conf.txmode.offloads;
2640                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
2641                                                      socketid, txconf);
2642                         if (ret < 0)
2643                                 rte_exit(EXIT_FAILURE,
2644                                         "rte_eth_tx_queue_setup: err=%d, "
2645                                                 "port=%d\n", ret, portid);
2646
2647                         qconf = &lcore_conf[lcore_id];
2648                         qconf->tx_queue_id[portid] = queueid;
2649                         queueid++;
2650
2651                         qconf->tx_port_id[qconf->n_tx_port] = portid;
2652                         qconf->n_tx_port++;
2653                 }
2654                 printf("\n");
2655         }
2656
2657         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2658                 if (rte_lcore_is_enabled(lcore_id) == 0)
2659                         continue;
2660
2661                 if (app_mode == APP_MODE_LEGACY) {
2662                         /* init timer structures for each enabled lcore */
2663                         rte_timer_init(&power_timers[lcore_id]);
2664                         hz = rte_get_timer_hz();
2665                         rte_timer_reset(&power_timers[lcore_id],
2666                                         hz/TIMER_NUMBER_PER_SECOND,
2667                                         SINGLE, lcore_id,
2668                                         power_timer_cb, NULL);
2669                 }
2670                 qconf = &lcore_conf[lcore_id];
2671                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
2672                 fflush(stdout);
2673                 /* init RX queues */
2674                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
2675                         struct rte_eth_rxconf rxq_conf;
2676
2677                         portid = qconf->rx_queue_list[queue].port_id;
2678                         queueid = qconf->rx_queue_list[queue].queue_id;
2679
2680                         if (numa_on)
2681                                 socketid = \
2682                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
2683                         else
2684                                 socketid = 0;
2685
2686                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
2687                         fflush(stdout);
2688
2689                         ret = rte_eth_dev_info_get(portid, &dev_info);
2690                         if (ret != 0)
2691                                 rte_exit(EXIT_FAILURE,
2692                                         "Error during getting device (port %u) info: %s\n",
2693                                         portid, strerror(-ret));
2694
2695                         rxq_conf = dev_info.default_rxconf;
2696                         rxq_conf.offloads = port_conf.rxmode.offloads;
2697                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
2698                                 socketid, &rxq_conf,
2699                                 pktmbuf_pool[socketid]);
2700                         if (ret < 0)
2701                                 rte_exit(EXIT_FAILURE,
2702                                         "rte_eth_rx_queue_setup: err=%d, "
2703                                                 "port=%d\n", ret, portid);
2704
2705                         if (parse_ptype) {
2706                                 if (add_cb_parse_ptype(portid, queueid) < 0)
2707                                         rte_exit(EXIT_FAILURE,
2708                                                  "Fail to add ptype cb\n");
2709                         } else if (!check_ptype(portid))
2710                                 rte_exit(EXIT_FAILURE,
2711                                          "PMD can not provide needed ptypes\n");
2712                 }
2713         }
2714
2715         printf("\n");
2716
2717         /* start ports */
2718         RTE_ETH_FOREACH_DEV(portid) {
2719                 if ((enabled_port_mask & (1 << portid)) == 0) {
2720                         continue;
2721                 }
2722                 /* Start device */
2723                 ret = rte_eth_dev_start(portid);
2724                 if (ret < 0)
2725                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
2726                                                 "port=%d\n", ret, portid);
2727                 /*
2728                  * If enabled, put device in promiscuous mode.
2729                  * This allows IO forwarding mode to forward packets
2730                  * to itself through 2 cross-connected  ports of the
2731                  * target machine.
2732                  */
2733                 if (promiscuous_on) {
2734                         ret = rte_eth_promiscuous_enable(portid);
2735                         if (ret != 0)
2736                                 rte_exit(EXIT_FAILURE,
2737                                         "rte_eth_promiscuous_enable: err=%s, port=%u\n",
2738                                         rte_strerror(-ret), portid);
2739                 }
2740                 /* initialize spinlock for each port */
2741                 rte_spinlock_init(&(locks[portid]));
2742         }
2743
2744         check_all_ports_link_status(enabled_port_mask);
2745
2746         if (app_mode == APP_MODE_EMPTY_POLL) {
2747
2748                 if (empty_poll_train) {
2749                         policy.state = TRAINING;
2750                 } else {
2751                         policy.state = MED_NORMAL;
2752                         policy.med_base_edpi = ep_med_edpi;
2753                         policy.hgh_base_edpi = ep_hgh_edpi;
2754                 }
2755
2756                 ret = rte_power_empty_poll_stat_init(&ep_params,
2757                                 freq_tlb,
2758                                 &policy);
2759                 if (ret < 0)
2760                         rte_exit(EXIT_FAILURE, "empty poll init failed");
2761         }
2762
2763
2764         /* launch per-lcore init on every lcore */
2765         if (app_mode == APP_MODE_LEGACY) {
2766                 rte_eal_mp_remote_launch(main_legacy_loop, NULL, CALL_MASTER);
2767         } else if (app_mode == APP_MODE_EMPTY_POLL) {
2768                 empty_poll_stop = false;
2769                 rte_eal_mp_remote_launch(main_empty_poll_loop, NULL,
2770                                 SKIP_MASTER);
2771         } else if (app_mode == APP_MODE_TELEMETRY) {
2772                 unsigned int i;
2773
2774                 /* Init metrics library */
2775                 rte_metrics_init(rte_socket_id());
2776                 /** Register stats with metrics library */
2777                 for (i = 0; i < NUM_TELSTATS; i++)
2778                         ptr_strings[i] = telstats_strings[i].name;
2779
2780                 ret = rte_metrics_reg_names(ptr_strings, NUM_TELSTATS);
2781                 if (ret >= 0)
2782                         telstats_index = ret;
2783                 else
2784                         rte_exit(EXIT_FAILURE, "failed to register metrics names");
2785
2786                 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2787                         rte_spinlock_init(&stats[lcore_id].telemetry_lock);
2788                 }
2789                 rte_timer_init(&telemetry_timer);
2790                 rte_telemetry_register_cmd("/l3fwd-power/stats",
2791                                 handle_app_stats,
2792                                 "Returns global power stats. Parameters: None");
2793                 rte_eal_mp_remote_launch(main_telemetry_loop, NULL,
2794                                                 SKIP_MASTER);
2795         } else if (app_mode == APP_MODE_INTERRUPT) {
2796                 rte_eal_mp_remote_launch(main_intr_loop, NULL, CALL_MASTER);
2797         }
2798
2799         if (app_mode == APP_MODE_EMPTY_POLL || app_mode == APP_MODE_TELEMETRY)
2800                 launch_timer(rte_lcore_id());
2801
2802         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2803                 if (rte_eal_wait_lcore(lcore_id) < 0)
2804                         return -1;
2805         }
2806
2807         RTE_ETH_FOREACH_DEV(portid)
2808         {
2809                 if ((enabled_port_mask & (1 << portid)) == 0)
2810                         continue;
2811
2812                 rte_eth_dev_stop(portid);
2813                 rte_eth_dev_close(portid);
2814         }
2815
2816         if (app_mode == APP_MODE_EMPTY_POLL)
2817                 rte_power_empty_poll_stat_free();
2818
2819         if ((app_mode == APP_MODE_LEGACY || app_mode == APP_MODE_EMPTY_POLL) &&
2820                         deinit_power_library())
2821                 rte_exit(EXIT_FAILURE, "deinit_power_library failed\n");
2822
2823         if (rte_eal_cleanup() < 0)
2824                 RTE_LOG(ERR, L3FWD_POWER, "EAL cleanup failed\n");
2825
2826         return 0;
2827 }