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 #include <rte_power_pmd_mgmt.h>
  51
  52 #include "perf_core.h"
  53 #include "main.h"
  54
  55 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
  56
  57 #define MAX_PKT_BURST 32
  58
  59 #define MIN_ZERO_POLL_COUNT 10
  60
  61 /* 100 ms interval */
  62 #define TIMER_NUMBER_PER_SECOND           10
  63 /* (10ms) */
  64 #define INTERVALS_PER_SECOND             100
  65 /* 100000 us */
  66 #define SCALING_PERIOD                    (1000000/TIMER_NUMBER_PER_SECOND)
  67 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
  68
  69 #define APP_LOOKUP_EXACT_MATCH          0
  70 #define APP_LOOKUP_LPM                  1
  71 #define DO_RFC_1812_CHECKS
  72
  73 #ifndef APP_LOOKUP_METHOD
  74 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  75 #endif
  76
  77 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  78 #include <rte_hash.h>
  79 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
  80 #include <rte_lpm.h>
  81 #else
  82 #error "APP_LOOKUP_METHOD set to incorrect value"
  83 #endif
  84
  85 #ifndef IPv6_BYTES
  86 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
  87                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
  88 #define IPv6_BYTES(addr) \
  89         addr[0],  addr[1], addr[2],  addr[3], \
  90         addr[4],  addr[5], addr[6],  addr[7], \
  91         addr[8],  addr[9], addr[10], addr[11],\
  92         addr[12], addr[13],addr[14], addr[15]
  93 #endif
  94
  95 #define MAX_JUMBO_PKT_LEN  9600
  96
  97 #define IPV6_ADDR_LEN 16
  98
  99 #define MEMPOOL_CACHE_SIZE 256
 100
 101 /*
 102  * This expression is used to calculate the number of mbufs needed depending on
 103  * user input, taking into account memory for rx and tx hardware rings, cache
 104  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 105  * NB_MBUF never goes below a minimum value of 8192.
 106  */
 107
 108 #define NB_MBUF RTE_MAX ( \
 109         (nb_ports*nb_rx_queue*nb_rxd + \
 110         nb_ports*nb_lcores*MAX_PKT_BURST + \
 111         nb_ports*n_tx_queue*nb_txd + \
 112         nb_lcores*MEMPOOL_CACHE_SIZE), \
 113         (unsigned)8192)
 114
 115 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 116
 117 #define NB_SOCKETS 8
 118
 119 /* Configure how many packets ahead to prefetch, when reading packets */
 120 #define PREFETCH_OFFSET 3
 121
 122 /*
 123  * Configurable number of RX/TX ring descriptors
 124  */
 125 #define RTE_TEST_RX_DESC_DEFAULT 1024
 126 #define RTE_TEST_TX_DESC_DEFAULT 1024
 127
 128 /*
 129  * These two thresholds were decided on by running the training algorithm on
 130  * a 2.5GHz Xeon. These defaults can be overridden by supplying non-zero values
 131  * for the med_threshold and high_threshold parameters on the command line.
 132  */
 133 #define EMPTY_POLL_MED_THRESHOLD 350000UL
 134 #define EMPTY_POLL_HGH_THRESHOLD 580000UL
 135
 136 #define NUM_TELSTATS RTE_DIM(telstats_strings)
 137
 138 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 139 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 140
 141 /* ethernet addresses of ports */
 142 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 143
 144 /* ethernet addresses of ports */
 145 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
 146
 147 /* mask of enabled ports */
 148 static uint32_t enabled_port_mask = 0;
 149 /* Ports set in promiscuous mode off by default. */
 150 static int promiscuous_on = 0;
 151 /* NUMA is enabled by default. */
 152 static int numa_on = 1;
 153 static bool empty_poll_stop;
 154 static bool empty_poll_train;
 155 volatile bool quit_signal;
 156 static struct  ep_params *ep_params;
 157 static struct  ep_policy policy;
 158 static long  ep_med_edpi, ep_hgh_edpi;
 159 /* timer to update telemetry every 500ms */
 160 static struct rte_timer telemetry_timer;
 161
 162 /* stats index returned by metrics lib */
 163 int telstats_index;
 164
 165 struct telstats_name {
 166         char name[RTE_ETH_XSTATS_NAME_SIZE];
 167 };
 168
 169 /* telemetry stats to be reported */
 170 const struct telstats_name telstats_strings[] = {
 171         {"empty_poll"},
 172         {"full_poll"},
 173         {"busy_percent"}
 174 };
 175
 176 /* core busyness in percentage */
 177 enum busy_rate {
 178         ZERO = 0,
 179         PARTIAL = 50,
 180         FULL = 100
 181 };
 182
 183 /* reference poll count to measure core busyness */
 184 #define DEFAULT_COUNT 10000
 185 /*
 186  * reference CYCLES to be used to
 187  * measure core busyness based on poll count
 188  */
 189 #define MIN_CYCLES  1500000ULL
 190 #define MAX_CYCLES 22000000ULL
 191
 192 /* (500ms) */
 193 #define TELEMETRY_INTERVALS_PER_SEC 2
 194
 195 static int parse_ptype; /**< Parse packet type using rx callback, and */
 196                         /**< disabled by default */
 197
 198 enum appmode {
 199         APP_MODE_DEFAULT = 0,
 200         APP_MODE_LEGACY,
 201         APP_MODE_EMPTY_POLL,
 202         APP_MODE_TELEMETRY,
 203         APP_MODE_INTERRUPT,
 204         APP_MODE_PMD_MGMT
 205 };
 206
 207 enum appmode app_mode;
 208
 209 static enum rte_power_pmd_mgmt_type pmgmt_type;
 210
 211 enum freq_scale_hint_t
 212 {
 213         FREQ_LOWER    =      -1,
 214         FREQ_CURRENT  =       0,
 215         FREQ_HIGHER   =       1,
 216         FREQ_HIGHEST  =       2
 217 };
 218
 219 struct lcore_rx_queue {
 220         uint16_t port_id;
 221         uint8_t queue_id;
 222         enum freq_scale_hint_t freq_up_hint;
 223         uint32_t zero_rx_packet_count;
 224         uint32_t idle_hint;
 225 } __rte_cache_aligned;
 226
 227 #define MAX_RX_QUEUE_PER_LCORE 16
 228 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 229 #define MAX_RX_QUEUE_PER_PORT 128
 230
 231 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
 232
 233
 234 struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 235 static struct lcore_params lcore_params_array_default[] = {
 236         {0, 0, 2},
 237         {0, 1, 2},
 238         {0, 2, 2},
 239         {1, 0, 2},
 240         {1, 1, 2},
 241         {1, 2, 2},
 242         {2, 0, 2},
 243         {3, 0, 3},
 244         {3, 1, 3},
 245 };
 246
 247 struct lcore_params *lcore_params = lcore_params_array_default;
 248 uint16_t nb_lcore_params = RTE_DIM(lcore_params_array_default);
 249
 250 static struct rte_eth_conf port_conf = {
 251         .rxmode = {
 252                 .mq_mode        = ETH_MQ_RX_RSS,
 253                 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
 254                 .split_hdr_size = 0,
 255                 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
 256         },
 257         .rx_adv_conf = {
 258                 .rss_conf = {
 259                         .rss_key = NULL,
 260                         .rss_hf = ETH_RSS_UDP,
 261                 },
 262         },
 263         .txmode = {
 264                 .mq_mode = ETH_MQ_TX_NONE,
 265         }
 266 };
 267
 268 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 269
 270
 271 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 272
 273 #ifdef RTE_ARCH_X86
 274 #include <rte_hash_crc.h>
 275 #define DEFAULT_HASH_FUNC       rte_hash_crc
 276 #else
 277 #include <rte_jhash.h>
 278 #define DEFAULT_HASH_FUNC       rte_jhash
 279 #endif
 280
 281 struct ipv4_5tuple {
 282         uint32_t ip_dst;
 283         uint32_t ip_src;
 284         uint16_t port_dst;
 285         uint16_t port_src;
 286         uint8_t  proto;
 287 } __rte_packed;
 288
 289 struct ipv6_5tuple {
 290         uint8_t  ip_dst[IPV6_ADDR_LEN];
 291         uint8_t  ip_src[IPV6_ADDR_LEN];
 292         uint16_t port_dst;
 293         uint16_t port_src;
 294         uint8_t  proto;
 295 } __rte_packed;
 296
 297 struct ipv4_l3fwd_route {
 298         struct ipv4_5tuple key;
 299         uint8_t if_out;
 300 };
 301
 302 struct ipv6_l3fwd_route {
 303         struct ipv6_5tuple key;
 304         uint8_t if_out;
 305 };
 306
 307 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 308         {{RTE_IPV4(100,10,0,1), RTE_IPV4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 309         {{RTE_IPV4(100,20,0,2), RTE_IPV4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 310         {{RTE_IPV4(100,30,0,3), RTE_IPV4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 311         {{RTE_IPV4(100,40,0,4), RTE_IPV4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 312 };
 313
 314 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 315         {
 316                 {
 317                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 318                          0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 319                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 320                          0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
 321                          1, 10, IPPROTO_UDP
 322                 }, 4
 323         },
 324 };
 325
 326 typedef struct rte_hash lookup_struct_t;
 327 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 328 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 329
 330 #define L3FWD_HASH_ENTRIES      1024
 331
 332 static uint16_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 333 static uint16_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 334 #endif
 335
 336 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 337 struct ipv4_l3fwd_route {
 338         uint32_t ip;
 339         uint8_t  depth;
 340         uint8_t  if_out;
 341 };
 342
 343 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 344         {RTE_IPV4(1,1,1,0), 24, 0},
 345         {RTE_IPV4(2,1,1,0), 24, 1},
 346         {RTE_IPV4(3,1,1,0), 24, 2},
 347         {RTE_IPV4(4,1,1,0), 24, 3},
 348         {RTE_IPV4(5,1,1,0), 24, 4},
 349         {RTE_IPV4(6,1,1,0), 24, 5},
 350         {RTE_IPV4(7,1,1,0), 24, 6},
 351         {RTE_IPV4(8,1,1,0), 24, 7},
 352 };
 353
 354 #define IPV4_L3FWD_LPM_MAX_RULES     1024
 355
 356 typedef struct rte_lpm lookup_struct_t;
 357 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 358 #endif
 359
 360 struct lcore_conf {
 361         uint16_t n_rx_queue;
 362         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 363         uint16_t n_tx_port;
 364         uint16_t tx_port_id[RTE_MAX_ETHPORTS];
 365         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 366         struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
 367         lookup_struct_t * ipv4_lookup_struct;
 368         lookup_struct_t * ipv6_lookup_struct;
 369 } __rte_cache_aligned;
 370
 371 struct lcore_stats {
 372         /* total sleep time in ms since last frequency scaling down */
 373         uint32_t sleep_time;
 374         /* number of long sleep recently */
 375         uint32_t nb_long_sleep;
 376         /* freq. scaling up trend */
 377         uint32_t trend;
 378         /* total packet processed recently */
 379         uint64_t nb_rx_processed;
 380         /* total iterations looped recently */
 381         uint64_t nb_iteration_looped;
 382         /*
 383          * Represents empty and non empty polls
 384          * of rte_eth_rx_burst();
 385          * ep_nep[0] holds non empty polls
 386          * i.e. 0 < nb_rx <= MAX_BURST
 387          * ep_nep[1] holds empty polls.
 388          * i.e. nb_rx == 0
 389          */
 390         uint64_t ep_nep[2];
 391         /*
 392          * Represents full and empty+partial
 393          * polls of rte_eth_rx_burst();
 394          * ep_nep[0] holds empty+partial polls.
 395          * i.e. 0 <= nb_rx < MAX_BURST
 396          * ep_nep[1] holds full polls
 397          * i.e. nb_rx == MAX_BURST
 398          */
 399         uint64_t fp_nfp[2];
 400         enum busy_rate br;
 401         rte_spinlock_t telemetry_lock;
 402 } __rte_cache_aligned;
 403
 404 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
 405 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
 406 static struct rte_timer power_timers[RTE_MAX_LCORE];
 407
 408 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
 409 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
 410                 unsigned int lcore_id, uint16_t port_id, uint16_t queue_id);
 411
 412
 413 /*
 414  * These defaults are using the max frequency index (1), a medium index (9)
 415  * and a typical low frequency index (14). These can be adjusted to use
 416  * different indexes using the relevant command line parameters.
 417  */
 418 static uint8_t  freq_tlb[] = {14, 9, 1};
 419
 420 static int is_done(void)
 421 {
 422         return quit_signal;
 423 }
 424
 425 /* exit signal handler */
 426 static void
 427 signal_exit_now(int sigtype)
 428 {
 429
 430         if (sigtype == SIGINT)
 431                 quit_signal = true;
 432
 433 }
 434
 435 /*  Freqency scale down timer callback */
 436 static void
 437 power_timer_cb(__rte_unused struct rte_timer *tim,
 438                           __rte_unused void *arg)
 439 {
 440         uint64_t hz;
 441         float sleep_time_ratio;
 442         unsigned lcore_id = rte_lcore_id();
 443
 444         /* accumulate total execution time in us when callback is invoked */
 445         sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
 446                                         (float)SCALING_PERIOD;
 447         /**
 448          * check whether need to scale down frequency a step if it sleep a lot.
 449          */
 450         if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
 451                 if (rte_power_freq_down)
 452                         rte_power_freq_down(lcore_id);
 453         }
 454         else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
 455                 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
 456                 /**
 457                  * scale down a step if average packet per iteration less
 458                  * than expectation.
 459                  */
 460                 if (rte_power_freq_down)
 461                         rte_power_freq_down(lcore_id);
 462         }
 463
 464         /**
 465          * initialize another timer according to current frequency to ensure
 466          * timer interval is relatively fixed.
 467          */
 468         hz = rte_get_timer_hz();
 469         rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
 470                                 SINGLE, lcore_id, power_timer_cb, NULL);
 471
 472         stats[lcore_id].nb_rx_processed = 0;
 473         stats[lcore_id].nb_iteration_looped = 0;
 474
 475         stats[lcore_id].sleep_time = 0;
 476 }
 477
 478 /* Enqueue a single packet, and send burst if queue is filled */
 479 static inline int
 480 send_single_packet(struct rte_mbuf *m, uint16_t port)
 481 {
 482         uint32_t lcore_id;
 483         struct lcore_conf *qconf;
 484
 485         lcore_id = rte_lcore_id();
 486         qconf = &lcore_conf[lcore_id];
 487
 488         rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
 489                         qconf->tx_buffer[port], m);
 490
 491         return 0;
 492 }
 493
 494 #ifdef DO_RFC_1812_CHECKS
 495 static inline int
 496 is_valid_ipv4_pkt(struct rte_ipv4_hdr *pkt, uint32_t link_len)
 497 {
 498         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 499         /*
 500          * 1. The packet length reported by the Link Layer must be large
 501          * enough to hold the minimum length legal IP datagram (20 bytes).
 502          */
 503         if (link_len < sizeof(struct rte_ipv4_hdr))
 504                 return -1;
 505
 506         /* 2. The IP checksum must be correct. */
 507         /* this is checked in H/W */
 508
 509         /*
 510          * 3. The IP version number must be 4. If the version number is not 4
 511          * then the packet may be another version of IP, such as IPng or
 512          * ST-II.
 513          */
 514         if (((pkt->version_ihl) >> 4) != 4)
 515                 return -3;
 516         /*
 517          * 4. The IP header length field must be large enough to hold the
 518          * minimum length legal IP datagram (20 bytes = 5 words).
 519          */
 520         if ((pkt->version_ihl & 0xf) < 5)
 521                 return -4;
 522
 523         /*
 524          * 5. The IP total length field must be large enough to hold the IP
 525          * datagram header, whose length is specified in the IP header length
 526          * field.
 527          */
 528         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct rte_ipv4_hdr))
 529                 return -5;
 530
 531         return 0;
 532 }
 533 #endif
 534
 535 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 536 static void
 537 print_ipv4_key(struct ipv4_5tuple key)
 538 {
 539         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
 540                 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
 541                                 key.port_dst, key.port_src, key.proto);
 542 }
 543 static void
 544 print_ipv6_key(struct ipv6_5tuple key)
 545 {
 546         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 547                 "port dst = %d, port src = %d, proto = %d\n",
 548                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 549                 key.port_dst, key.port_src, key.proto);
 550 }
 551
 552 static inline uint16_t
 553 get_ipv4_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
 554                 lookup_struct_t * ipv4_l3fwd_lookup_struct)
 555 {
 556         struct ipv4_5tuple key;
 557         struct rte_tcp_hdr *tcp;
 558         struct rte_udp_hdr *udp;
 559         int ret = 0;
 560
 561         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 562         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 563         key.proto = ipv4_hdr->next_proto_id;
 564
 565         switch (ipv4_hdr->next_proto_id) {
 566         case IPPROTO_TCP:
 567                 tcp = (struct rte_tcp_hdr *)((unsigned char *)ipv4_hdr +
 568                                         sizeof(struct rte_ipv4_hdr));
 569                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 570                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 571                 break;
 572
 573         case IPPROTO_UDP:
 574                 udp = (struct rte_udp_hdr *)((unsigned char *)ipv4_hdr +
 575                                         sizeof(struct rte_ipv4_hdr));
 576                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 577                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 578                 break;
 579
 580         default:
 581                 key.port_dst = 0;
 582                 key.port_src = 0;
 583                 break;
 584         }
 585
 586         /* Find destination port */
 587         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 588         return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
 589 }
 590
 591 static inline uint16_t
 592 get_ipv6_dst_port(struct rte_ipv6_hdr *ipv6_hdr, uint16_t portid,
 593                         lookup_struct_t *ipv6_l3fwd_lookup_struct)
 594 {
 595         struct ipv6_5tuple key;
 596         struct rte_tcp_hdr *tcp;
 597         struct rte_udp_hdr *udp;
 598         int ret = 0;
 599
 600         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 601         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 602
 603         key.proto = ipv6_hdr->proto;
 604
 605         switch (ipv6_hdr->proto) {
 606         case IPPROTO_TCP:
 607                 tcp = (struct rte_tcp_hdr *)((unsigned char *) ipv6_hdr +
 608                                         sizeof(struct rte_ipv6_hdr));
 609                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 610                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 611                 break;
 612
 613         case IPPROTO_UDP:
 614                 udp = (struct rte_udp_hdr *)((unsigned char *) ipv6_hdr +
 615                                         sizeof(struct rte_ipv6_hdr));
 616                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 617                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 618                 break;
 619
 620         default:
 621                 key.port_dst = 0;
 622                 key.port_src = 0;
 623                 break;
 624         }
 625
 626         /* Find destination port */
 627         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 628         return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
 629 }
 630 #endif
 631
 632 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 633 static inline uint16_t
 634 get_ipv4_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
 635                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 636 {
 637         uint32_t next_hop;
 638
 639         return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 640                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 641                         next_hop : portid);
 642 }
 643 #endif
 644
 645 static inline void
 646 parse_ptype_one(struct rte_mbuf *m)
 647 {
 648         struct rte_ether_hdr *eth_hdr;
 649         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 650         uint16_t ether_type;
 651
 652         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 653         ether_type = eth_hdr->ether_type;
 654         if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
 655                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 656         else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6))
 657                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 658
 659         m->packet_type = packet_type;
 660 }
 661
 662 static uint16_t
 663 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
 664                struct rte_mbuf *pkts[], uint16_t nb_pkts,
 665                uint16_t max_pkts __rte_unused,
 666                void *user_param __rte_unused)
 667 {
 668         unsigned int i;
 669
 670         for (i = 0; i < nb_pkts; ++i)
 671                 parse_ptype_one(pkts[i]);
 672
 673         return nb_pkts;
 674 }
 675
 676 static int
 677 add_cb_parse_ptype(uint16_t portid, uint16_t queueid)
 678 {
 679         printf("Port %d: softly parse packet type info\n", portid);
 680         if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL))
 681                 return 0;
 682
 683         printf("Failed to add rx callback: port=%d\n", portid);
 684         return -1;
 685 }
 686
 687 static inline void
 688 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
 689                                 struct lcore_conf *qconf)
 690 {
 691         struct rte_ether_hdr *eth_hdr;
 692         struct rte_ipv4_hdr *ipv4_hdr;
 693         void *d_addr_bytes;
 694         uint16_t dst_port;
 695
 696         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 697
 698         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 699                 /* Handle IPv4 headers.*/
 700                 ipv4_hdr =
 701                         rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
 702                                                 sizeof(struct rte_ether_hdr));
 703
 704 #ifdef DO_RFC_1812_CHECKS
 705                 /* Check to make sure the packet is valid (RFC1812) */
 706                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 707                         rte_pktmbuf_free(m);
 708                         return;
 709                 }
 710 #endif
 711
 712                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 713                                         qconf->ipv4_lookup_struct);
 714                 if (dst_port >= RTE_MAX_ETHPORTS ||
 715                                 (enabled_port_mask & 1 << dst_port) == 0)
 716                         dst_port = portid;
 717
 718                 /* 02:00:00:00:00:xx */
 719                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 720                 *((uint64_t *)d_addr_bytes) =
 721                         0x000000000002 + ((uint64_t)dst_port << 40);
 722
 723 #ifdef DO_RFC_1812_CHECKS
 724                 /* Update time to live and header checksum */
 725                 --(ipv4_hdr->time_to_live);
 726                 ++(ipv4_hdr->hdr_checksum);
 727 #endif
 728
 729                 /* src addr */
 730                 rte_ether_addr_copy(&ports_eth_addr[dst_port],
 731                                 &eth_hdr->s_addr);
 732
 733                 send_single_packet(m, dst_port);
 734         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 735                 /* Handle IPv6 headers.*/
 736 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 737                 struct rte_ipv6_hdr *ipv6_hdr;
 738
 739                 ipv6_hdr =
 740                         rte_pktmbuf_mtod_offset(m, struct rte_ipv6_hdr *,
 741                                                 sizeof(struct rte_ether_hdr));
 742
 743                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
 744                                         qconf->ipv6_lookup_struct);
 745
 746                 if (dst_port >= RTE_MAX_ETHPORTS ||
 747                                 (enabled_port_mask & 1 << dst_port) == 0)
 748                         dst_port = portid;
 749
 750                 /* 02:00:00:00:00:xx */
 751                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 752                 *((uint64_t *)d_addr_bytes) =
 753                         0x000000000002 + ((uint64_t)dst_port << 40);
 754
 755                 /* src addr */
 756                 rte_ether_addr_copy(&ports_eth_addr[dst_port],
 757                                 &eth_hdr->s_addr);
 758
 759                 send_single_packet(m, dst_port);
 760 #else
 761                 /* We don't currently handle IPv6 packets in LPM mode. */
 762                 rte_pktmbuf_free(m);
 763 #endif
 764         } else
 765                 rte_pktmbuf_free(m);
 766
 767 }
 768
 769 #define MINIMUM_SLEEP_TIME         1
 770 #define SUSPEND_THRESHOLD          300
 771
 772 static inline uint32_t
 773 power_idle_heuristic(uint32_t zero_rx_packet_count)
 774 {
 775         /* If zero count is less than 100,  sleep 1us */
 776         if (zero_rx_packet_count < SUSPEND_THRESHOLD)
 777                 return MINIMUM_SLEEP_TIME;
 778         /* If zero count is less than 1000, sleep 100 us which is the
 779                 minimum latency switching from C3/C6 to C0
 780         */
 781         else
 782                 return SUSPEND_THRESHOLD;
 783 }
 784
 785 static inline enum freq_scale_hint_t
 786 power_freq_scaleup_heuristic(unsigned lcore_id,
 787                              uint16_t port_id,
 788                              uint16_t queue_id)
 789 {
 790         uint32_t rxq_count = rte_eth_rx_queue_count(port_id, queue_id);
 791 /**
 792  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
 793  * per iteration
 794  */
 795 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
 796 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
 797 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
 798 #define FREQ_UP_TREND1_ACC   1
 799 #define FREQ_UP_TREND2_ACC   100
 800 #define FREQ_UP_THRESHOLD    10000
 801
 802         if (likely(rxq_count > FREQ_GEAR3_RX_PACKET_THRESHOLD)) {
 803                 stats[lcore_id].trend = 0;
 804                 return FREQ_HIGHEST;
 805         } else if (likely(rxq_count > FREQ_GEAR2_RX_PACKET_THRESHOLD))
 806                 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
 807         else if (likely(rxq_count > FREQ_GEAR1_RX_PACKET_THRESHOLD))
 808                 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
 809
 810         if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
 811                 stats[lcore_id].trend = 0;
 812                 return FREQ_HIGHER;
 813         }
 814
 815         return FREQ_CURRENT;
 816 }
 817
 818 /**
 819  * force polling thread sleep until one-shot rx interrupt triggers
 820  * @param port_id
 821  *  Port id.
 822  * @param queue_id
 823  *  Rx queue id.
 824  * @return
 825  *  0 on success
 826  */
 827 static int
 828 sleep_until_rx_interrupt(int num, int lcore)
 829 {
 830         /*
 831          * we want to track when we are woken up by traffic so that we can go
 832          * back to sleep again without log spamming. Avoid cache line sharing
 833          * to prevent threads stepping on each others' toes.
 834          */
 835         static struct {
 836                 bool wakeup;
 837         } __rte_cache_aligned status[RTE_MAX_LCORE];
 838         struct rte_epoll_event event[num];
 839         int n, i;
 840         uint16_t port_id;
 841         uint8_t queue_id;
 842         void *data;
 843
 844         if (status[lcore].wakeup) {
 845                 RTE_LOG(INFO, L3FWD_POWER,
 846                                 "lcore %u sleeps until interrupt triggers\n",
 847                                 rte_lcore_id());
 848         }
 849
 850         n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, 10);
 851         for (i = 0; i < n; i++) {
 852                 data = event[i].epdata.data;
 853                 port_id = ((uintptr_t)data) >> CHAR_BIT;
 854                 queue_id = ((uintptr_t)data) &
 855                         RTE_LEN2MASK(CHAR_BIT, uint8_t);
 856                 RTE_LOG(INFO, L3FWD_POWER,
 857                         "lcore %u is waked up from rx interrupt on"
 858                         " port %d queue %d\n",
 859                         rte_lcore_id(), port_id, queue_id);
 860         }
 861         status[lcore].wakeup = n != 0;
 862
 863         return 0;
 864 }
 865
 866 static void turn_on_off_intr(struct lcore_conf *qconf, bool on)
 867 {
 868         int i;
 869         struct lcore_rx_queue *rx_queue;
 870         uint8_t queue_id;
 871         uint16_t port_id;
 872
 873         for (i = 0; i < qconf->n_rx_queue; ++i) {
 874                 rx_queue = &(qconf->rx_queue_list[i]);
 875                 port_id = rx_queue->port_id;
 876                 queue_id = rx_queue->queue_id;
 877
 878                 rte_spinlock_lock(&(locks[port_id]));
 879                 if (on)
 880                         rte_eth_dev_rx_intr_enable(port_id, queue_id);
 881                 else
 882                         rte_eth_dev_rx_intr_disable(port_id, queue_id);
 883                 rte_spinlock_unlock(&(locks[port_id]));
 884         }
 885 }
 886
 887 static int event_register(struct lcore_conf *qconf)
 888 {
 889         struct lcore_rx_queue *rx_queue;
 890         uint8_t queueid;
 891         uint16_t portid;
 892         uint32_t data;
 893         int ret;
 894         int i;
 895
 896         for (i = 0; i < qconf->n_rx_queue; ++i) {
 897                 rx_queue = &(qconf->rx_queue_list[i]);
 898                 portid = rx_queue->port_id;
 899                 queueid = rx_queue->queue_id;
 900                 data = portid << CHAR_BIT | queueid;
 901
 902                 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
 903                                                 RTE_EPOLL_PER_THREAD,
 904                                                 RTE_INTR_EVENT_ADD,
 905                                                 (void *)((uintptr_t)data));
 906                 if (ret)
 907                         return ret;
 908         }
 909
 910         return 0;
 911 }
 912
 913 /* main processing loop */
 914 static int main_intr_loop(__rte_unused void *dummy)
 915 {
 916         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 917         unsigned int lcore_id;
 918         uint64_t prev_tsc, diff_tsc, cur_tsc;
 919         int i, j, nb_rx;
 920         uint8_t queueid;
 921         uint16_t portid;
 922         struct lcore_conf *qconf;
 923         struct lcore_rx_queue *rx_queue;
 924         uint32_t lcore_rx_idle_count = 0;
 925         uint32_t lcore_idle_hint = 0;
 926         int intr_en = 0;
 927
 928         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
 929                                    US_PER_S * BURST_TX_DRAIN_US;
 930
 931         prev_tsc = 0;
 932
 933         lcore_id = rte_lcore_id();
 934         qconf = &lcore_conf[lcore_id];
 935
 936         if (qconf->n_rx_queue == 0) {
 937                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
 938                                 lcore_id);
 939                 return 0;
 940         }
 941
 942         RTE_LOG(INFO, L3FWD_POWER, "entering main interrupt loop on lcore %u\n",
 943                         lcore_id);
 944
 945         for (i = 0; i < qconf->n_rx_queue; i++) {
 946                 portid = qconf->rx_queue_list[i].port_id;
 947                 queueid = qconf->rx_queue_list[i].queue_id;
 948                 RTE_LOG(INFO, L3FWD_POWER,
 949                                 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
 950                                 lcore_id, portid, queueid);
 951         }
 952
 953         /* add into event wait list */
 954         if (event_register(qconf) == 0)
 955                 intr_en = 1;
 956         else
 957                 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
 958
 959         while (!is_done()) {
 960                 stats[lcore_id].nb_iteration_looped++;
 961
 962                 cur_tsc = rte_rdtsc();
 963
 964                 /*
 965                  * TX burst queue drain
 966                  */
 967                 diff_tsc = cur_tsc - prev_tsc;
 968                 if (unlikely(diff_tsc > drain_tsc)) {
 969                         for (i = 0; i < qconf->n_tx_port; ++i) {
 970                                 portid = qconf->tx_port_id[i];
 971                                 rte_eth_tx_buffer_flush(portid,
 972                                                 qconf->tx_queue_id[portid],
 973                                                 qconf->tx_buffer[portid]);
 974                         }
 975                         prev_tsc = cur_tsc;
 976                 }
 977
 978 start_rx:
 979                 /*
 980                  * Read packet from RX queues
 981                  */
 982                 lcore_rx_idle_count = 0;
 983                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 984                         rx_queue = &(qconf->rx_queue_list[i]);
 985                         rx_queue->idle_hint = 0;
 986                         portid = rx_queue->port_id;
 987                         queueid = rx_queue->queue_id;
 988
 989                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 990                                         MAX_PKT_BURST);
 991
 992                         stats[lcore_id].nb_rx_processed += nb_rx;
 993                         if (unlikely(nb_rx == 0)) {
 994                                 /**
 995                                  * no packet received from rx queue, try to
 996                                  * sleep for a while forcing CPU enter deeper
 997                                  * C states.
 998                                  */
 999                                 rx_queue->zero_rx_packet_count++;
1000
1001                                 if (rx_queue->zero_rx_packet_count <=
1002                                                 MIN_ZERO_POLL_COUNT)
1003                                         continue;
1004
1005                                 rx_queue->idle_hint = power_idle_heuristic(
1006                                                 rx_queue->zero_rx_packet_count);
1007                                 lcore_rx_idle_count++;
1008                         } else {
1009                                 rx_queue->zero_rx_packet_count = 0;
1010                         }
1011
1012                         /* Prefetch first packets */
1013                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1014                                 rte_prefetch0(rte_pktmbuf_mtod(
1015                                                 pkts_burst[j], void *));
1016                         }
1017
1018                         /* Prefetch and forward already prefetched packets */
1019                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1020                                 rte_prefetch0(rte_pktmbuf_mtod(
1021                                                 pkts_burst[j + PREFETCH_OFFSET],
1022                                                 void *));
1023                                 l3fwd_simple_forward(
1024                                                 pkts_burst[j], portid, qconf);
1025                         }
1026
1027                         /* Forward remaining prefetched packets */
1028                         for (; j < nb_rx; j++) {
1029                                 l3fwd_simple_forward(
1030                                                 pkts_burst[j], portid, qconf);
1031                         }
1032                 }
1033
1034                 if (unlikely(lcore_rx_idle_count == qconf->n_rx_queue)) {
1035                         /**
1036                          * All Rx queues empty in recent consecutive polls,
1037                          * sleep in a conservative manner, meaning sleep as
1038                          * less as possible.
1039                          */
1040                         for (i = 1,
1041                             lcore_idle_hint = qconf->rx_queue_list[0].idle_hint;
1042                                         i < qconf->n_rx_queue; ++i) {
1043                                 rx_queue = &(qconf->rx_queue_list[i]);
1044                                 if (rx_queue->idle_hint < lcore_idle_hint)
1045                                         lcore_idle_hint = rx_queue->idle_hint;
1046                         }
1047
1048                         if (lcore_idle_hint < SUSPEND_THRESHOLD)
1049                                 /**
1050                                  * execute "pause" instruction to avoid context
1051                                  * switch which generally take hundred of
1052                                  * microseconds for short sleep.
1053                                  */
1054                                 rte_delay_us(lcore_idle_hint);
1055                         else {
1056                                 /* suspend until rx interrupt triggers */
1057                                 if (intr_en) {
1058                                         turn_on_off_intr(qconf, 1);
1059                                         sleep_until_rx_interrupt(
1060                                                         qconf->n_rx_queue,
1061                                                         lcore_id);
1062                                         turn_on_off_intr(qconf, 0);
1063                                         /**
1064                                          * start receiving packets immediately
1065                                          */
1066                                         if (likely(!is_done()))
1067                                                 goto start_rx;
1068                                 }
1069                         }
1070                         stats[lcore_id].sleep_time += lcore_idle_hint;
1071                 }
1072         }
1073
1074         return 0;
1075 }
1076
1077 /* main processing loop */
1078 static int
1079 main_telemetry_loop(__rte_unused void *dummy)
1080 {
1081         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1082         unsigned int lcore_id;
1083         uint64_t prev_tsc, diff_tsc, cur_tsc, prev_tel_tsc;
1084         int i, j, nb_rx;
1085         uint8_t queueid;
1086         uint16_t portid;
1087         struct lcore_conf *qconf;
1088         struct lcore_rx_queue *rx_queue;
1089         uint64_t ep_nep[2] = {0}, fp_nfp[2] = {0};
1090         uint64_t poll_count;
1091         enum busy_rate br;
1092
1093         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1094                                         US_PER_S * BURST_TX_DRAIN_US;
1095
1096         poll_count = 0;
1097         prev_tsc = 0;
1098         prev_tel_tsc = 0;
1099
1100         lcore_id = rte_lcore_id();
1101         qconf = &lcore_conf[lcore_id];
1102
1103         if (qconf->n_rx_queue == 0) {
1104                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
1105                         lcore_id);
1106                 return 0;
1107         }
1108
1109         RTE_LOG(INFO, L3FWD_POWER, "entering main telemetry loop on lcore %u\n",
1110                 lcore_id);
1111
1112         for (i = 0; i < qconf->n_rx_queue; i++) {
1113                 portid = qconf->rx_queue_list[i].port_id;
1114                 queueid = qconf->rx_queue_list[i].queue_id;
1115                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1116                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
1117         }
1118
1119         while (!is_done()) {
1120
1121                 cur_tsc = rte_rdtsc();
1122                 /*
1123                  * TX burst queue drain
1124                  */
1125                 diff_tsc = cur_tsc - prev_tsc;
1126                 if (unlikely(diff_tsc > drain_tsc)) {
1127                         for (i = 0; i < qconf->n_tx_port; ++i) {
1128                                 portid = qconf->tx_port_id[i];
1129                                 rte_eth_tx_buffer_flush(portid,
1130                                                 qconf->tx_queue_id[portid],
1131                                                 qconf->tx_buffer[portid]);
1132                         }
1133                         prev_tsc = cur_tsc;
1134                 }
1135
1136                 /*
1137                  * Read packet from RX queues
1138                  */
1139                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1140                         rx_queue = &(qconf->rx_queue_list[i]);
1141                         portid = rx_queue->port_id;
1142                         queueid = rx_queue->queue_id;
1143
1144                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1145                                                                 MAX_PKT_BURST);
1146                         ep_nep[nb_rx == 0]++;
1147                         fp_nfp[nb_rx == MAX_PKT_BURST]++;
1148                         poll_count++;
1149                         if (unlikely(nb_rx == 0))
1150                                 continue;
1151
1152                         /* Prefetch first packets */
1153                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1154                                 rte_prefetch0(rte_pktmbuf_mtod(
1155                                                 pkts_burst[j], void *));
1156                         }
1157
1158                         /* Prefetch and forward already prefetched packets */
1159                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1160                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1161                                                 j + PREFETCH_OFFSET], void *));
1162                                 l3fwd_simple_forward(pkts_burst[j], portid,
1163                                                                 qconf);
1164                         }
1165
1166                         /* Forward remaining prefetched packets */
1167                         for (; j < nb_rx; j++) {
1168                                 l3fwd_simple_forward(pkts_burst[j], portid,
1169                                                                 qconf);
1170                         }
1171                 }
1172                 if (unlikely(poll_count >= DEFAULT_COUNT)) {
1173                         diff_tsc = cur_tsc - prev_tel_tsc;
1174                         if (diff_tsc >= MAX_CYCLES) {
1175                                 br = FULL;
1176                         } else if (diff_tsc > MIN_CYCLES &&
1177                                         diff_tsc < MAX_CYCLES) {
1178                                 br = (diff_tsc * 100) / MAX_CYCLES;
1179                         } else {
1180                                 br = ZERO;
1181                         }
1182                         poll_count = 0;
1183                         prev_tel_tsc = cur_tsc;
1184                         /* update stats for telemetry */
1185                         rte_spinlock_lock(&stats[lcore_id].telemetry_lock);
1186                         stats[lcore_id].ep_nep[0] = ep_nep[0];
1187                         stats[lcore_id].ep_nep[1] = ep_nep[1];
1188                         stats[lcore_id].fp_nfp[0] = fp_nfp[0];
1189                         stats[lcore_id].fp_nfp[1] = fp_nfp[1];
1190                         stats[lcore_id].br = br;
1191                         rte_spinlock_unlock(&stats[lcore_id].telemetry_lock);
1192                 }
1193         }
1194
1195         return 0;
1196 }
1197 /* main processing loop */
1198 static int
1199 main_empty_poll_loop(__rte_unused void *dummy)
1200 {
1201         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1202         unsigned int lcore_id;
1203         uint64_t prev_tsc, diff_tsc, cur_tsc;
1204         int i, j, nb_rx;
1205         uint8_t queueid;
1206         uint16_t portid;
1207         struct lcore_conf *qconf;
1208         struct lcore_rx_queue *rx_queue;
1209
1210         const uint64_t drain_tsc =
1211                 (rte_get_tsc_hz() + US_PER_S - 1) /
1212                 US_PER_S * BURST_TX_DRAIN_US;
1213
1214         prev_tsc = 0;
1215
1216         lcore_id = rte_lcore_id();
1217         qconf = &lcore_conf[lcore_id];
1218
1219         if (qconf->n_rx_queue == 0) {
1220                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
1221                         lcore_id);
1222                 return 0;
1223         }
1224
1225         for (i = 0; i < qconf->n_rx_queue; i++) {
1226                 portid = qconf->rx_queue_list[i].port_id;
1227                 queueid = qconf->rx_queue_list[i].queue_id;
1228                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1229                                 "rxqueueid=%hhu\n", lcore_id, portid, queueid);
1230         }
1231
1232         while (!is_done()) {
1233                 stats[lcore_id].nb_iteration_looped++;
1234
1235                 cur_tsc = rte_rdtsc();
1236                 /*
1237                  * TX burst queue drain
1238                  */
1239                 diff_tsc = cur_tsc - prev_tsc;
1240                 if (unlikely(diff_tsc > drain_tsc)) {
1241                         for (i = 0; i < qconf->n_tx_port; ++i) {
1242                                 portid = qconf->tx_port_id[i];
1243                                 rte_eth_tx_buffer_flush(portid,
1244                                                 qconf->tx_queue_id[portid],
1245                                                 qconf->tx_buffer[portid]);
1246                         }
1247                         prev_tsc = cur_tsc;
1248                 }
1249
1250                 /*
1251                  * Read packet from RX queues
1252                  */
1253                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1254                         rx_queue = &(qconf->rx_queue_list[i]);
1255                         rx_queue->idle_hint = 0;
1256                         portid = rx_queue->port_id;
1257                         queueid = rx_queue->queue_id;
1258
1259                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1260                                         MAX_PKT_BURST);
1261
1262                         stats[lcore_id].nb_rx_processed += nb_rx;
1263
1264                         if (nb_rx == 0) {
1265
1266                                 rte_power_empty_poll_stat_update(lcore_id);
1267
1268                                 continue;
1269                         } else {
1270                                 rte_power_poll_stat_update(lcore_id, nb_rx);
1271                         }
1272
1273
1274                         /* Prefetch first packets */
1275                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1276                                 rte_prefetch0(rte_pktmbuf_mtod(
1277                                                         pkts_burst[j], void *));
1278                         }
1279
1280                         /* Prefetch and forward already prefetched packets */
1281                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1282                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1283                                                         j + PREFETCH_OFFSET],
1284                                                         void *));
1285                                 l3fwd_simple_forward(pkts_burst[j], portid,
1286                                                 qconf);
1287                         }
1288
1289                         /* Forward remaining prefetched packets */
1290                         for (; j < nb_rx; j++) {
1291                                 l3fwd_simple_forward(pkts_burst[j], portid,
1292                                                 qconf);
1293                         }
1294
1295                 }
1296
1297         }
1298
1299         return 0;
1300 }
1301 /* main processing loop */
1302 static int
1303 main_legacy_loop(__rte_unused void *dummy)
1304 {
1305         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1306         unsigned lcore_id;
1307         uint64_t prev_tsc, diff_tsc, cur_tsc, tim_res_tsc, hz;
1308         uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
1309         int i, j, nb_rx;
1310         uint8_t queueid;
1311         uint16_t portid;
1312         struct lcore_conf *qconf;
1313         struct lcore_rx_queue *rx_queue;
1314         enum freq_scale_hint_t lcore_scaleup_hint;
1315         uint32_t lcore_rx_idle_count = 0;
1316         uint32_t lcore_idle_hint = 0;
1317         int intr_en = 0;
1318
1319         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
1320
1321         prev_tsc = 0;
1322         hz = rte_get_timer_hz();
1323         tim_res_tsc = hz/TIMER_NUMBER_PER_SECOND;
1324
1325         lcore_id = rte_lcore_id();
1326         qconf = &lcore_conf[lcore_id];
1327
1328         if (qconf->n_rx_queue == 0) {
1329                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
1330                 return 0;
1331         }
1332
1333         RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
1334
1335         for (i = 0; i < qconf->n_rx_queue; i++) {
1336                 portid = qconf->rx_queue_list[i].port_id;
1337                 queueid = qconf->rx_queue_list[i].queue_id;
1338                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1339                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
1340         }
1341
1342         /* add into event wait list */
1343         if (event_register(qconf) == 0)
1344                 intr_en = 1;
1345         else
1346                 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
1347
1348         while (!is_done()) {
1349                 stats[lcore_id].nb_iteration_looped++;
1350
1351                 cur_tsc = rte_rdtsc();
1352                 cur_tsc_power = cur_tsc;
1353
1354                 /*
1355                  * TX burst queue drain
1356                  */
1357                 diff_tsc = cur_tsc - prev_tsc;
1358                 if (unlikely(diff_tsc > drain_tsc)) {
1359                         for (i = 0; i < qconf->n_tx_port; ++i) {
1360                                 portid = qconf->tx_port_id[i];
1361                                 rte_eth_tx_buffer_flush(portid,
1362                                                 qconf->tx_queue_id[portid],
1363                                                 qconf->tx_buffer[portid]);
1364                         }
1365                         prev_tsc = cur_tsc;
1366                 }
1367
1368                 diff_tsc_power = cur_tsc_power - prev_tsc_power;
1369                 if (diff_tsc_power > tim_res_tsc) {
1370                         rte_timer_manage();
1371                         prev_tsc_power = cur_tsc_power;
1372                 }
1373
1374 start_rx:
1375                 /*
1376                  * Read packet from RX queues
1377                  */
1378                 lcore_scaleup_hint = FREQ_CURRENT;
1379                 lcore_rx_idle_count = 0;
1380                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1381                         rx_queue = &(qconf->rx_queue_list[i]);
1382                         rx_queue->idle_hint = 0;
1383                         portid = rx_queue->port_id;
1384                         queueid = rx_queue->queue_id;
1385
1386                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1387                                                                 MAX_PKT_BURST);
1388
1389                         stats[lcore_id].nb_rx_processed += nb_rx;
1390                         if (unlikely(nb_rx == 0)) {
1391                                 /**
1392                                  * no packet received from rx queue, try to
1393                                  * sleep for a while forcing CPU enter deeper
1394                                  * C states.
1395                                  */
1396                                 rx_queue->zero_rx_packet_count++;
1397
1398                                 if (rx_queue->zero_rx_packet_count <=
1399                                                         MIN_ZERO_POLL_COUNT)
1400                                         continue;
1401
1402                                 rx_queue->idle_hint = power_idle_heuristic(\
1403                                         rx_queue->zero_rx_packet_count);
1404                                 lcore_rx_idle_count++;
1405                         } else {
1406                                 rx_queue->zero_rx_packet_count = 0;
1407
1408                                 /**
1409                                  * do not scale up frequency immediately as
1410                                  * user to kernel space communication is costly
1411                                  * which might impact packet I/O for received
1412                                  * packets.
1413                                  */
1414                                 rx_queue->freq_up_hint =
1415                                         power_freq_scaleup_heuristic(lcore_id,
1416                                                         portid, queueid);
1417                         }
1418
1419                         /* Prefetch first packets */
1420                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1421                                 rte_prefetch0(rte_pktmbuf_mtod(
1422                                                 pkts_burst[j], void *));
1423                         }
1424
1425                         /* Prefetch and forward already prefetched packets */
1426                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1427                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1428                                                 j + PREFETCH_OFFSET], void *));
1429                                 l3fwd_simple_forward(pkts_burst[j], portid,
1430                                                                 qconf);
1431                         }
1432
1433                         /* Forward remaining prefetched packets */
1434                         for (; j < nb_rx; j++) {
1435                                 l3fwd_simple_forward(pkts_burst[j], portid,
1436                                                                 qconf);
1437                         }
1438                 }
1439
1440                 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
1441                         for (i = 1, lcore_scaleup_hint =
1442                                 qconf->rx_queue_list[0].freq_up_hint;
1443                                         i < qconf->n_rx_queue; ++i) {
1444                                 rx_queue = &(qconf->rx_queue_list[i]);
1445                                 if (rx_queue->freq_up_hint >
1446                                                 lcore_scaleup_hint)
1447                                         lcore_scaleup_hint =
1448                                                 rx_queue->freq_up_hint;
1449                         }
1450
1451                         if (lcore_scaleup_hint == FREQ_HIGHEST) {
1452                                 if (rte_power_freq_max)
1453                                         rte_power_freq_max(lcore_id);
1454                         } else if (lcore_scaleup_hint == FREQ_HIGHER) {
1455                                 if (rte_power_freq_up)
1456                                         rte_power_freq_up(lcore_id);
1457                         }
1458                 } else {
1459                         /**
1460                          * All Rx queues empty in recent consecutive polls,
1461                          * sleep in a conservative manner, meaning sleep as
1462                          * less as possible.
1463                          */
1464                         for (i = 1, lcore_idle_hint =
1465                                 qconf->rx_queue_list[0].idle_hint;
1466                                         i < qconf->n_rx_queue; ++i) {
1467                                 rx_queue = &(qconf->rx_queue_list[i]);
1468                                 if (rx_queue->idle_hint < lcore_idle_hint)
1469                                         lcore_idle_hint = rx_queue->idle_hint;
1470                         }
1471
1472                         if (lcore_idle_hint < SUSPEND_THRESHOLD)
1473                                 /**
1474                                  * execute "pause" instruction to avoid context
1475                                  * switch which generally take hundred of
1476                                  * microseconds for short sleep.
1477                                  */
1478                                 rte_delay_us(lcore_idle_hint);
1479                         else {
1480                                 /* suspend until rx interrupt triggers */
1481                                 if (intr_en) {
1482                                         turn_on_off_intr(qconf, 1);
1483                                         sleep_until_rx_interrupt(
1484                                                         qconf->n_rx_queue,
1485                                                         lcore_id);
1486                                         turn_on_off_intr(qconf, 0);
1487                                         /**
1488                                          * start receiving packets immediately
1489                                          */
1490                                         if (likely(!is_done()))
1491                                                 goto start_rx;
1492                                 }
1493                         }
1494                         stats[lcore_id].sleep_time += lcore_idle_hint;
1495                 }
1496         }
1497
1498         return 0;
1499 }
1500
1501 static int
1502 check_lcore_params(void)
1503 {
1504         uint8_t queue, lcore;
1505         uint16_t i;
1506         int socketid;
1507
1508         for (i = 0; i < nb_lcore_params; ++i) {
1509                 queue = lcore_params[i].queue_id;
1510                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1511                         printf("invalid queue number: %hhu\n", queue);
1512                         return -1;
1513                 }
1514                 lcore = lcore_params[i].lcore_id;
1515                 if (!rte_lcore_is_enabled(lcore)) {
1516                         printf("error: lcore %hhu is not enabled in lcore "
1517                                                         "mask\n", lcore);
1518                         return -1;
1519                 }
1520                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1521                                                         (numa_on == 0)) {
1522                         printf("warning: lcore %hhu is on socket %d with numa "
1523                                                 "off\n", lcore, socketid);
1524                 }
1525                 if (app_mode == APP_MODE_TELEMETRY && lcore == rte_lcore_id()) {
1526                         printf("cannot enable main core %d in config for telemetry mode\n",
1527                                 rte_lcore_id());
1528                         return -1;
1529                 }
1530         }
1531         return 0;
1532 }
1533
1534 static int
1535 check_port_config(void)
1536 {
1537         unsigned portid;
1538         uint16_t i;
1539
1540         for (i = 0; i < nb_lcore_params; ++i) {
1541                 portid = lcore_params[i].port_id;
1542                 if ((enabled_port_mask & (1 << portid)) == 0) {
1543                         printf("port %u is not enabled in port mask\n",
1544                                                                 portid);
1545                         return -1;
1546                 }
1547                 if (!rte_eth_dev_is_valid_port(portid)) {
1548                         printf("port %u is not present on the board\n",
1549                                                                 portid);
1550                         return -1;
1551                 }
1552         }
1553         return 0;
1554 }
1555
1556 static uint8_t
1557 get_port_n_rx_queues(const uint16_t port)
1558 {
1559         int queue = -1;
1560         uint16_t i;
1561
1562         for (i = 0; i < nb_lcore_params; ++i) {
1563                 if (lcore_params[i].port_id == port &&
1564                                 lcore_params[i].queue_id > queue)
1565                         queue = lcore_params[i].queue_id;
1566         }
1567         return (uint8_t)(++queue);
1568 }
1569
1570 static int
1571 init_lcore_rx_queues(void)
1572 {
1573         uint16_t i, nb_rx_queue;
1574         uint8_t lcore;
1575
1576         for (i = 0; i < nb_lcore_params; ++i) {
1577                 lcore = lcore_params[i].lcore_id;
1578                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1579                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1580                         printf("error: too many queues (%u) for lcore: %u\n",
1581                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1582                         return -1;
1583                 } else {
1584                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1585                                 lcore_params[i].port_id;
1586                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1587                                 lcore_params[i].queue_id;
1588                         lcore_conf[lcore].n_rx_queue++;
1589                 }
1590         }
1591         return 0;
1592 }
1593
1594 /* display usage */
1595 static void
1596 print_usage(const char *prgname)
1597 {
1598         printf ("%s [EAL options] -- -p PORTMASK -P"
1599                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1600                 "  [--high-perf-cores CORELIST"
1601                 "  [--perf-config (port,queue,hi_perf,lcore_index)[,(port,queue,hi_perf,lcore_index]]"
1602                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1603                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1604                 "  -P : enable promiscuous mode\n"
1605                 "  --config (port,queue,lcore): rx queues configuration\n"
1606                 "  --high-perf-cores CORELIST: list of high performance cores\n"
1607                 "  --perf-config: similar as config, cores specified as indices"
1608                 " for bins containing high or regular performance cores\n"
1609                 "  --no-numa: optional, disable numa awareness\n"
1610                 "  --enable-jumbo: enable jumbo frame"
1611                 " which max packet len is PKTLEN in decimal (64-9600)\n"
1612                 "  --parse-ptype: parse packet type by software\n"
1613                 "  --legacy: use legacy interrupt-based scaling\n"
1614                 "  --empty-poll: enable empty poll detection"
1615                 " follow (training_flag, high_threshold, med_threshold)\n"
1616                 " --telemetry: enable telemetry mode, to update"
1617                 " empty polls, full polls, and core busyness to telemetry\n"
1618                 " --interrupt-only: enable interrupt-only mode\n"
1619                 " --pmd-mgmt MODE: enable PMD power management mode. "
1620                 "Currently supported modes: monitor, pause, scale\n",
1621                 prgname);
1622 }
1623
1624 static int parse_max_pkt_len(const char *pktlen)
1625 {
1626         char *end = NULL;
1627         unsigned long len;
1628
1629         /* parse decimal string */
1630         len = strtoul(pktlen, &end, 10);
1631         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1632                 return -1;
1633
1634         if (len == 0)
1635                 return -1;
1636
1637         return len;
1638 }
1639
1640 static int
1641 parse_portmask(const char *portmask)
1642 {
1643         char *end = NULL;
1644         unsigned long pm;
1645
1646         /* parse hexadecimal string */
1647         pm = strtoul(portmask, &end, 16);
1648         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1649                 return 0;
1650
1651         return pm;
1652 }
1653
1654 static int
1655 parse_config(const char *q_arg)
1656 {
1657         char s[256];
1658         const char *p, *p0 = q_arg;
1659         char *end;
1660         enum fieldnames {
1661                 FLD_PORT = 0,
1662                 FLD_QUEUE,
1663                 FLD_LCORE,
1664                 _NUM_FLD
1665         };
1666         unsigned long int_fld[_NUM_FLD];
1667         char *str_fld[_NUM_FLD];
1668         int i;
1669         unsigned size;
1670
1671         nb_lcore_params = 0;
1672
1673         while ((p = strchr(p0,'(')) != NULL) {
1674                 ++p;
1675                 if((p0 = strchr(p,')')) == NULL)
1676                         return -1;
1677
1678                 size = p0 - p;
1679                 if(size >= sizeof(s))
1680                         return -1;
1681
1682                 snprintf(s, sizeof(s), "%.*s", size, p);
1683                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1684                                                                 _NUM_FLD)
1685                         return -1;
1686                 for (i = 0; i < _NUM_FLD; i++){
1687                         errno = 0;
1688                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1689                         if (errno != 0 || end == str_fld[i] || int_fld[i] >
1690                                                                         255)
1691                                 return -1;
1692                 }
1693                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1694                         printf("exceeded max number of lcore params: %hu\n",
1695                                 nb_lcore_params);
1696                         return -1;
1697                 }
1698                 lcore_params_array[nb_lcore_params].port_id =
1699                                 (uint8_t)int_fld[FLD_PORT];
1700                 lcore_params_array[nb_lcore_params].queue_id =
1701                                 (uint8_t)int_fld[FLD_QUEUE];
1702                 lcore_params_array[nb_lcore_params].lcore_id =
1703                                 (uint8_t)int_fld[FLD_LCORE];
1704                 ++nb_lcore_params;
1705         }
1706         lcore_params = lcore_params_array;
1707
1708         return 0;
1709 }
1710
1711 static int
1712 parse_pmd_mgmt_config(const char *name)
1713 {
1714 #define PMD_MGMT_MONITOR "monitor"
1715 #define PMD_MGMT_PAUSE   "pause"
1716 #define PMD_MGMT_SCALE   "scale"
1717
1718         if (strncmp(PMD_MGMT_MONITOR, name, sizeof(PMD_MGMT_MONITOR)) == 0) {
1719                 pmgmt_type = RTE_POWER_MGMT_TYPE_MONITOR;
1720                 return 0;
1721         }
1722
1723         if (strncmp(PMD_MGMT_PAUSE, name, sizeof(PMD_MGMT_PAUSE)) == 0) {
1724                 pmgmt_type = RTE_POWER_MGMT_TYPE_PAUSE;
1725                 return 0;
1726         }
1727
1728         if (strncmp(PMD_MGMT_SCALE, name, sizeof(PMD_MGMT_SCALE)) == 0) {
1729                 pmgmt_type = RTE_POWER_MGMT_TYPE_SCALE;
1730                 return 0;
1731         }
1732         /* unknown PMD power management mode */
1733         return -1;
1734 }
1735
1736 static int
1737 parse_ep_config(const char *q_arg)
1738 {
1739         char s[256];
1740         const char *p = q_arg;
1741         char *end;
1742         int  num_arg;
1743
1744         char *str_fld[3];
1745
1746         int training_flag;
1747         int med_edpi;
1748         int hgh_edpi;
1749
1750         ep_med_edpi = EMPTY_POLL_MED_THRESHOLD;
1751         ep_hgh_edpi = EMPTY_POLL_HGH_THRESHOLD;
1752
1753         strlcpy(s, p, sizeof(s));
1754
1755         num_arg = rte_strsplit(s, sizeof(s), str_fld, 3, ',');
1756
1757         empty_poll_train = false;
1758
1759         if (num_arg == 0)
1760                 return 0;
1761
1762         if (num_arg == 3) {
1763
1764                 training_flag = strtoul(str_fld[0], &end, 0);
1765                 med_edpi = strtoul(str_fld[1], &end, 0);
1766                 hgh_edpi = strtoul(str_fld[2], &end, 0);
1767
1768                 if (training_flag == 1)
1769                         empty_poll_train = true;
1770
1771                 if (med_edpi > 0)
1772                         ep_med_edpi = med_edpi;
1773
1774                 if (hgh_edpi > 0)
1775                         ep_hgh_edpi = hgh_edpi;
1776
1777         } else {
1778
1779                 return -1;
1780         }
1781
1782         return 0;
1783
1784 }
1785 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1786 #define CMD_LINE_OPT_LEGACY "legacy"
1787 #define CMD_LINE_OPT_EMPTY_POLL "empty-poll"
1788 #define CMD_LINE_OPT_INTERRUPT_ONLY "interrupt-only"
1789 #define CMD_LINE_OPT_TELEMETRY "telemetry"
1790 #define CMD_LINE_OPT_PMD_MGMT "pmd-mgmt"
1791
1792 /* Parse the argument given in the command line of the application */
1793 static int
1794 parse_args(int argc, char **argv)
1795 {
1796         int opt, ret;
1797         char **argvopt;
1798         int option_index;
1799         uint32_t limit;
1800         char *prgname = argv[0];
1801         static struct option lgopts[] = {
1802                 {"config", 1, 0, 0},
1803                 {"perf-config", 1, 0, 0},
1804                 {"high-perf-cores", 1, 0, 0},
1805                 {"no-numa", 0, 0, 0},
1806                 {"enable-jumbo", 0, 0, 0},
1807                 {CMD_LINE_OPT_EMPTY_POLL, 1, 0, 0},
1808                 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
1809                 {CMD_LINE_OPT_LEGACY, 0, 0, 0},
1810                 {CMD_LINE_OPT_TELEMETRY, 0, 0, 0},
1811                 {CMD_LINE_OPT_INTERRUPT_ONLY, 0, 0, 0},
1812                 {CMD_LINE_OPT_PMD_MGMT, 1, 0, 0},
1813                 {NULL, 0, 0, 0}
1814         };
1815
1816         argvopt = argv;
1817
1818         while ((opt = getopt_long(argc, argvopt, "p:l:m:h:P",
1819                                 lgopts, &option_index)) != EOF) {
1820
1821                 switch (opt) {
1822                 /* portmask */
1823                 case 'p':
1824                         enabled_port_mask = parse_portmask(optarg);
1825                         if (enabled_port_mask == 0) {
1826                                 printf("invalid portmask\n");
1827                                 print_usage(prgname);
1828                                 return -1;
1829                         }
1830                         break;
1831                 case 'P':
1832                         printf("Promiscuous mode selected\n");
1833                         promiscuous_on = 1;
1834                         break;
1835                 case 'l':
1836                         limit = parse_max_pkt_len(optarg);
1837                         freq_tlb[LOW] = limit;
1838                         break;
1839                 case 'm':
1840                         limit = parse_max_pkt_len(optarg);
1841                         freq_tlb[MED] = limit;
1842                         break;
1843                 case 'h':
1844                         limit = parse_max_pkt_len(optarg);
1845                         freq_tlb[HGH] = limit;
1846                         break;
1847                 /* long options */
1848                 case 0:
1849                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1850                                 ret = parse_config(optarg);
1851                                 if (ret) {
1852                                         printf("invalid config\n");
1853                                         print_usage(prgname);
1854                                         return -1;
1855                                 }
1856                         }
1857
1858                         if (!strncmp(lgopts[option_index].name,
1859                                         "perf-config", 11)) {
1860                                 ret = parse_perf_config(optarg);
1861                                 if (ret) {
1862                                         printf("invalid perf-config\n");
1863                                         print_usage(prgname);
1864                                         return -1;
1865                                 }
1866                         }
1867
1868                         if (!strncmp(lgopts[option_index].name,
1869                                         "high-perf-cores", 15)) {
1870                                 ret = parse_perf_core_list(optarg);
1871                                 if (ret) {
1872                                         printf("invalid high-perf-cores\n");
1873                                         print_usage(prgname);
1874                                         return -1;
1875                                 }
1876                         }
1877
1878                         if (!strncmp(lgopts[option_index].name,
1879                                                 "no-numa", 7)) {
1880                                 printf("numa is disabled \n");
1881                                 numa_on = 0;
1882                         }
1883
1884                         if (!strncmp(lgopts[option_index].name,
1885                                         CMD_LINE_OPT_LEGACY,
1886                                         sizeof(CMD_LINE_OPT_LEGACY))) {
1887                                 if (app_mode != APP_MODE_DEFAULT) {
1888                                         printf(" legacy mode is mutually exclusive with other modes\n");
1889                                         return -1;
1890                                 }
1891                                 app_mode = APP_MODE_LEGACY;
1892                                 printf("legacy mode is enabled\n");
1893                         }
1894
1895                         if (!strncmp(lgopts[option_index].name,
1896                                         CMD_LINE_OPT_EMPTY_POLL, 10)) {
1897                                 if (app_mode != APP_MODE_DEFAULT) {
1898                                         printf(" empty-poll mode is mutually exclusive with other modes\n");
1899                                         return -1;
1900                                 }
1901                                 app_mode = APP_MODE_EMPTY_POLL;
1902                                 ret = parse_ep_config(optarg);
1903
1904                                 if (ret) {
1905                                         printf("invalid empty poll config\n");
1906                                         print_usage(prgname);
1907                                         return -1;
1908                                 }
1909                                 printf("empty-poll is enabled\n");
1910                         }
1911
1912                         if (!strncmp(lgopts[option_index].name,
1913                                         CMD_LINE_OPT_TELEMETRY,
1914                                         sizeof(CMD_LINE_OPT_TELEMETRY))) {
1915                                 if (app_mode != APP_MODE_DEFAULT) {
1916                                         printf(" telemetry mode is mutually exclusive with other modes\n");
1917                                         return -1;
1918                                 }
1919                                 app_mode = APP_MODE_TELEMETRY;
1920                                 printf("telemetry mode is enabled\n");
1921                         }
1922
1923                         if (!strncmp(lgopts[option_index].name,
1924                                         CMD_LINE_OPT_PMD_MGMT,
1925                                         sizeof(CMD_LINE_OPT_PMD_MGMT))) {
1926                                 if (app_mode != APP_MODE_DEFAULT) {
1927                                         printf(" power mgmt mode is mutually exclusive with other modes\n");
1928                                         return -1;
1929                                 }
1930                                 if (parse_pmd_mgmt_config(optarg) < 0) {
1931                                         printf(" Invalid PMD power management mode: %s\n",
1932                                                         optarg);
1933                                         return -1;
1934                                 }
1935                                 app_mode = APP_MODE_PMD_MGMT;
1936                                 printf("PMD power mgmt mode is enabled\n");
1937                         }
1938                         if (!strncmp(lgopts[option_index].name,
1939                                         CMD_LINE_OPT_INTERRUPT_ONLY,
1940                                         sizeof(CMD_LINE_OPT_INTERRUPT_ONLY))) {
1941                                 if (app_mode != APP_MODE_DEFAULT) {
1942                                         printf(" interrupt-only mode is mutually exclusive with other modes\n");
1943                                         return -1;
1944                                 }
1945                                 app_mode = APP_MODE_INTERRUPT;
1946                                 printf("interrupt-only mode is enabled\n");
1947                         }
1948
1949                         if (!strncmp(lgopts[option_index].name,
1950                                         "enable-jumbo", 12)) {
1951                                 struct option lenopts =
1952                                         {"max-pkt-len", required_argument, \
1953                                                                         0, 0};
1954
1955                                 printf("jumbo frame is enabled \n");
1956                                 port_conf.rxmode.offloads |=
1957                                                 DEV_RX_OFFLOAD_JUMBO_FRAME;
1958                                 port_conf.txmode.offloads |=
1959                                                 DEV_TX_OFFLOAD_MULTI_SEGS;
1960
1961                                 /**
1962                                  * if no max-pkt-len set, use the default value
1963                                  * RTE_ETHER_MAX_LEN
1964                                  */
1965                                 if (0 == getopt_long(argc, argvopt, "",
1966                                                 &lenopts, &option_index)) {
1967                                         ret = parse_max_pkt_len(optarg);
1968                                         if ((ret < 64) ||
1969                                                 (ret > MAX_JUMBO_PKT_LEN)){
1970                                                 printf("invalid packet "
1971                                                                 "length\n");
1972                                                 print_usage(prgname);
1973                                                 return -1;
1974                                         }
1975                                         port_conf.rxmode.max_rx_pkt_len = ret;
1976                                 }
1977                                 printf("set jumbo frame "
1978                                         "max packet length to %u\n",
1979                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1980                         }
1981
1982                         if (!strncmp(lgopts[option_index].name,
1983                                      CMD_LINE_OPT_PARSE_PTYPE,
1984                                      sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
1985                                 printf("soft parse-ptype is enabled\n");
1986                                 parse_ptype = 1;
1987                         }
1988
1989                         break;
1990
1991                 default:
1992                         print_usage(prgname);
1993                         return -1;
1994                 }
1995         }
1996
1997         if (optind >= 0)
1998                 argv[optind-1] = prgname;
1999
2000         ret = optind-1;
2001         optind = 1; /* reset getopt lib */
2002         return ret;
2003 }
2004
2005 static void
2006 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
2007 {
2008         char buf[RTE_ETHER_ADDR_FMT_SIZE];
2009         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
2010         printf("%s%s", name, buf);
2011 }
2012
2013 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2014 static void
2015 setup_hash(int socketid)
2016 {
2017         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
2018                 .name = NULL,
2019                 .entries = L3FWD_HASH_ENTRIES,
2020                 .key_len = sizeof(struct ipv4_5tuple),
2021                 .hash_func = DEFAULT_HASH_FUNC,
2022                 .hash_func_init_val = 0,
2023         };
2024
2025         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
2026                 .name = NULL,
2027                 .entries = L3FWD_HASH_ENTRIES,
2028                 .key_len = sizeof(struct ipv6_5tuple),
2029                 .hash_func = DEFAULT_HASH_FUNC,
2030                 .hash_func_init_val = 0,
2031         };
2032
2033         unsigned i;
2034         int ret;
2035         char s[64];
2036
2037         /* create ipv4 hash */
2038         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
2039         ipv4_l3fwd_hash_params.name = s;
2040         ipv4_l3fwd_hash_params.socket_id = socketid;
2041         ipv4_l3fwd_lookup_struct[socketid] =
2042                 rte_hash_create(&ipv4_l3fwd_hash_params);
2043         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2044                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2045                                 "socket %d\n", socketid);
2046
2047         /* create ipv6 hash */
2048         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
2049         ipv6_l3fwd_hash_params.name = s;
2050         ipv6_l3fwd_hash_params.socket_id = socketid;
2051         ipv6_l3fwd_lookup_struct[socketid] =
2052                 rte_hash_create(&ipv6_l3fwd_hash_params);
2053         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
2054                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2055                                 "socket %d\n", socketid);
2056
2057
2058         /* populate the ipv4 hash */
2059         for (i = 0; i < RTE_DIM(ipv4_l3fwd_route_array); i++) {
2060                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
2061                                 (void *) &ipv4_l3fwd_route_array[i].key);
2062                 if (ret < 0) {
2063                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
2064                                 "l3fwd hash on socket %d\n", i, socketid);
2065                 }
2066                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
2067                 printf("Hash: Adding key\n");
2068                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
2069         }
2070
2071         /* populate the ipv6 hash */
2072         for (i = 0; i < RTE_DIM(ipv6_l3fwd_route_array); i++) {
2073                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
2074                                 (void *) &ipv6_l3fwd_route_array[i].key);
2075                 if (ret < 0) {
2076                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
2077                                 "l3fwd hash on socket %d\n", i, socketid);
2078                 }
2079                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
2080                 printf("Hash: Adding key\n");
2081                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
2082         }
2083 }
2084 #endif
2085
2086 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2087 static void
2088 setup_lpm(int socketid)
2089 {
2090         unsigned i;
2091         int ret;
2092         char s[64];
2093
2094         /* create the LPM table */
2095         struct rte_lpm_config lpm_ipv4_config;
2096
2097         lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
2098         lpm_ipv4_config.number_tbl8s = 256;
2099         lpm_ipv4_config.flags = 0;
2100
2101         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
2102         ipv4_l3fwd_lookup_struct[socketid] =
2103                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
2104         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2105                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
2106                                 " on socket %d\n", socketid);
2107
2108         /* populate the LPM table */
2109         for (i = 0; i < RTE_DIM(ipv4_l3fwd_route_array); i++) {
2110                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
2111                         ipv4_l3fwd_route_array[i].ip,
2112                         ipv4_l3fwd_route_array[i].depth,
2113                         ipv4_l3fwd_route_array[i].if_out);
2114
2115                 if (ret < 0) {
2116                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
2117                                 "l3fwd LPM table on socket %d\n",
2118                                 i, socketid);
2119                 }
2120
2121                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
2122                         (unsigned)ipv4_l3fwd_route_array[i].ip,
2123                         ipv4_l3fwd_route_array[i].depth,
2124                         ipv4_l3fwd_route_array[i].if_out);
2125         }
2126 }
2127 #endif
2128
2129 static int
2130 init_mem(unsigned nb_mbuf)
2131 {
2132         struct lcore_conf *qconf;
2133         int socketid;
2134         unsigned lcore_id;
2135         char s[64];
2136
2137         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2138                 if (rte_lcore_is_enabled(lcore_id) == 0)
2139                         continue;
2140
2141                 if (numa_on)
2142                         socketid = rte_lcore_to_socket_id(lcore_id);
2143                 else
2144                         socketid = 0;
2145
2146                 if (socketid >= NB_SOCKETS) {
2147                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
2148                                         "out of range %d\n", socketid,
2149                                                 lcore_id, NB_SOCKETS);
2150                 }
2151                 if (pktmbuf_pool[socketid] == NULL) {
2152                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
2153                         pktmbuf_pool[socketid] =
2154                                 rte_pktmbuf_pool_create(s, nb_mbuf,
2155                                         MEMPOOL_CACHE_SIZE, 0,
2156                                         RTE_MBUF_DEFAULT_BUF_SIZE,
2157                                         socketid);
2158                         if (pktmbuf_pool[socketid] == NULL)
2159                                 rte_exit(EXIT_FAILURE,
2160                                         "Cannot init mbuf pool on socket %d\n",
2161                                                                 socketid);
2162                         else
2163                                 printf("Allocated mbuf pool on socket %d\n",
2164                                                                 socketid);
2165
2166 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2167                         setup_lpm(socketid);
2168 #else
2169                         setup_hash(socketid);
2170 #endif
2171                 }
2172                 qconf = &lcore_conf[lcore_id];
2173                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
2174 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2175                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
2176 #endif
2177         }
2178         return 0;
2179 }
2180
2181 /* Check the link status of all ports in up to 9s, and print them finally */
2182 static void
2183 check_all_ports_link_status(uint32_t port_mask)
2184 {
2185 #define CHECK_INTERVAL 100 /* 100ms */
2186 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
2187         uint8_t count, all_ports_up, print_flag = 0;
2188         uint16_t portid;
2189         struct rte_eth_link link;
2190         int ret;
2191         char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
2192
2193         printf("\nChecking link status");
2194         fflush(stdout);
2195         for (count = 0; count <= MAX_CHECK_TIME; count++) {
2196                 all_ports_up = 1;
2197                 RTE_ETH_FOREACH_DEV(portid) {
2198                         if ((port_mask & (1 << portid)) == 0)
2199                                 continue;
2200                         memset(&link, 0, sizeof(link));
2201                         ret = rte_eth_link_get_nowait(portid, &link);
2202                         if (ret < 0) {
2203                                 all_ports_up = 0;
2204                                 if (print_flag == 1)
2205                                         printf("Port %u link get failed: %s\n",
2206                                                 portid, rte_strerror(-ret));
2207                                 continue;
2208                         }
2209                         /* print link status if flag set */
2210                         if (print_flag == 1) {
2211                                 rte_eth_link_to_str(link_status_text,
2212                                         sizeof(link_status_text), &link);
2213                                 printf("Port %d %s\n", portid,
2214                                        link_status_text);
2215                                 continue;
2216                         }
2217                         /* clear all_ports_up flag if any link down */
2218                         if (link.link_status == ETH_LINK_DOWN) {
2219                                 all_ports_up = 0;
2220                                 break;
2221                         }
2222                 }
2223                 /* after finally printing all link status, get out */
2224                 if (print_flag == 1)
2225                         break;
2226
2227                 if (all_ports_up == 0) {
2228                         printf(".");
2229                         fflush(stdout);
2230                         rte_delay_ms(CHECK_INTERVAL);
2231                 }
2232
2233                 /* set the print_flag if all ports up or timeout */
2234                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
2235                         print_flag = 1;
2236                         printf("done\n");
2237                 }
2238         }
2239 }
2240
2241 static int check_ptype(uint16_t portid)
2242 {
2243         int i, ret;
2244         int ptype_l3_ipv4 = 0;
2245 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2246         int ptype_l3_ipv6 = 0;
2247 #endif
2248         uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
2249
2250         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
2251         if (ret <= 0)
2252                 return 0;
2253
2254         uint32_t ptypes[ret];
2255
2256         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
2257         for (i = 0; i < ret; ++i) {
2258                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
2259                         ptype_l3_ipv4 = 1;
2260 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2261                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
2262                         ptype_l3_ipv6 = 1;
2263 #endif
2264         }
2265
2266         if (ptype_l3_ipv4 == 0)
2267                 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
2268
2269 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2270         if (ptype_l3_ipv6 == 0)
2271                 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
2272 #endif
2273
2274 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2275         if (ptype_l3_ipv4)
2276 #else /* APP_LOOKUP_EXACT_MATCH */
2277         if (ptype_l3_ipv4 && ptype_l3_ipv6)
2278 #endif
2279                 return 1;
2280
2281         return 0;
2282
2283 }
2284
2285 static int
2286 init_power_library(void)
2287 {
2288         enum power_management_env env;
2289         unsigned int lcore_id;
2290         int ret = 0;
2291
2292         RTE_LCORE_FOREACH(lcore_id) {
2293                 /* init power management library */
2294                 ret = rte_power_init(lcore_id);
2295                 if (ret) {
2296                         RTE_LOG(ERR, POWER,
2297                                 "Library initialization failed on core %u\n",
2298                                 lcore_id);
2299                         return ret;
2300                 }
2301                 /* we're not supporting the VM channel mode */
2302                 env = rte_power_get_env();
2303                 if (env != PM_ENV_ACPI_CPUFREQ &&
2304                                 env != PM_ENV_PSTATE_CPUFREQ) {
2305                         RTE_LOG(ERR, POWER,
2306                                 "Only ACPI and PSTATE mode are supported\n");
2307                         return -1;
2308                 }
2309         }
2310         return ret;
2311 }
2312
2313 static int
2314 deinit_power_library(void)
2315 {
2316         unsigned int lcore_id;
2317         int ret = 0;
2318
2319         RTE_LCORE_FOREACH(lcore_id) {
2320                 /* deinit power management library */
2321                 ret = rte_power_exit(lcore_id);
2322                 if (ret) {
2323                         RTE_LOG(ERR, POWER,
2324                                 "Library deinitialization failed on core %u\n",
2325                                 lcore_id);
2326                         return ret;
2327                 }
2328         }
2329         return ret;
2330 }
2331
2332 static void
2333 get_current_stat_values(uint64_t *values)
2334 {
2335         unsigned int lcore_id = rte_lcore_id();
2336         struct lcore_conf *qconf;
2337         uint64_t app_eps = 0, app_fps = 0, app_br = 0;
2338         uint64_t count = 0;
2339
2340         RTE_LCORE_FOREACH_WORKER(lcore_id) {
2341                 qconf = &lcore_conf[lcore_id];
2342                 if (qconf->n_rx_queue == 0)
2343                         continue;
2344                 count++;
2345                 rte_spinlock_lock(&stats[lcore_id].telemetry_lock);
2346                 app_eps += stats[lcore_id].ep_nep[1];
2347                 app_fps += stats[lcore_id].fp_nfp[1];
2348                 app_br += stats[lcore_id].br;
2349                 rte_spinlock_unlock(&stats[lcore_id].telemetry_lock);
2350         }
2351
2352         if (count > 0) {
2353                 values[0] = app_eps/count;
2354                 values[1] = app_fps/count;
2355                 values[2] = app_br/count;
2356         } else
2357                 memset(values, 0, sizeof(uint64_t) * NUM_TELSTATS);
2358
2359 }
2360
2361 static void
2362 update_telemetry(__rte_unused struct rte_timer *tim,
2363                 __rte_unused void *arg)
2364 {
2365         int ret;
2366         uint64_t values[NUM_TELSTATS] = {0};
2367
2368         get_current_stat_values(values);
2369         ret = rte_metrics_update_values(RTE_METRICS_GLOBAL, telstats_index,
2370                                         values, RTE_DIM(values));
2371         if (ret < 0)
2372                 RTE_LOG(WARNING, POWER, "failed to update metrcis\n");
2373 }
2374
2375 static int
2376 handle_app_stats(const char *cmd __rte_unused,
2377                 const char *params __rte_unused,
2378                 struct rte_tel_data *d)
2379 {
2380         uint64_t values[NUM_TELSTATS] = {0};
2381         uint32_t i;
2382
2383         rte_tel_data_start_dict(d);
2384         get_current_stat_values(values);
2385         for (i = 0; i < NUM_TELSTATS; i++)
2386                 rte_tel_data_add_dict_u64(d, telstats_strings[i].name,
2387                                 values[i]);
2388         return 0;
2389 }
2390
2391 static void
2392 telemetry_setup_timer(void)
2393 {
2394         int lcore_id = rte_lcore_id();
2395         uint64_t hz = rte_get_timer_hz();
2396         uint64_t ticks;
2397
2398         ticks = hz / TELEMETRY_INTERVALS_PER_SEC;
2399         rte_timer_reset_sync(&telemetry_timer,
2400                         ticks,
2401                         PERIODICAL,
2402                         lcore_id,
2403                         update_telemetry,
2404                         NULL);
2405 }
2406 static void
2407 empty_poll_setup_timer(void)
2408 {
2409         int lcore_id = rte_lcore_id();
2410         uint64_t hz = rte_get_timer_hz();
2411
2412         struct  ep_params *ep_ptr = ep_params;
2413
2414         ep_ptr->interval_ticks = hz / INTERVALS_PER_SECOND;
2415
2416         rte_timer_reset_sync(&ep_ptr->timer0,
2417                         ep_ptr->interval_ticks,
2418                         PERIODICAL,
2419                         lcore_id,
2420                         rte_empty_poll_detection,
2421                         (void *)ep_ptr);
2422
2423 }
2424 static int
2425 launch_timer(unsigned int lcore_id)
2426 {
2427         int64_t prev_tsc = 0, cur_tsc, diff_tsc, cycles_10ms;
2428
2429         RTE_SET_USED(lcore_id);
2430
2431
2432         if (rte_get_main_lcore() != lcore_id) {
2433                 rte_panic("timer on lcore:%d which is not main core:%d\n",
2434                                 lcore_id,
2435                                 rte_get_main_lcore());
2436         }
2437
2438         RTE_LOG(INFO, POWER, "Bring up the Timer\n");
2439
2440         if (app_mode == APP_MODE_EMPTY_POLL)
2441                 empty_poll_setup_timer();
2442         else
2443                 telemetry_setup_timer();
2444
2445         cycles_10ms = rte_get_timer_hz() / 100;
2446
2447         while (!is_done()) {
2448                 cur_tsc = rte_rdtsc();
2449                 diff_tsc = cur_tsc - prev_tsc;
2450                 if (diff_tsc > cycles_10ms) {
2451                         rte_timer_manage();
2452                         prev_tsc = cur_tsc;
2453                         cycles_10ms = rte_get_timer_hz() / 100;
2454                 }
2455         }
2456
2457         RTE_LOG(INFO, POWER, "Timer_subsystem is done\n");
2458
2459         return 0;
2460 }
2461
2462 static int
2463 autodetect_mode(void)
2464 {
2465         RTE_LOG(NOTICE, L3FWD_POWER, "Operating mode not specified, probing frequency scaling support...\n");
2466
2467         /*
2468          * Empty poll and telemetry modes have to be specifically requested to
2469          * be enabled, but we can auto-detect between interrupt mode with or
2470          * without frequency scaling. Both ACPI and pstate can be used.
2471          */
2472         if (rte_power_check_env_supported(PM_ENV_ACPI_CPUFREQ))
2473                 return APP_MODE_LEGACY;
2474         if (rte_power_check_env_supported(PM_ENV_PSTATE_CPUFREQ))
2475                 return APP_MODE_LEGACY;
2476
2477         RTE_LOG(NOTICE, L3FWD_POWER, "Frequency scaling not supported, selecting interrupt-only mode\n");
2478
2479         return APP_MODE_INTERRUPT;
2480 }
2481
2482 static const char *
2483 mode_to_str(enum appmode mode)
2484 {
2485         switch (mode) {
2486         case APP_MODE_LEGACY:
2487                 return "legacy";
2488         case APP_MODE_EMPTY_POLL:
2489                 return "empty poll";
2490         case APP_MODE_TELEMETRY:
2491                 return "telemetry";
2492         case APP_MODE_INTERRUPT:
2493                 return "interrupt-only";
2494         case APP_MODE_PMD_MGMT:
2495                 return "pmd mgmt";
2496         default:
2497                 return "invalid";
2498         }
2499 }
2500
2501 int
2502 main(int argc, char **argv)
2503 {
2504         struct lcore_conf *qconf;
2505         struct rte_eth_dev_info dev_info;
2506         struct rte_eth_txconf *txconf;
2507         int ret;
2508         uint16_t nb_ports;
2509         uint16_t queueid;
2510         unsigned lcore_id;
2511         uint64_t hz;
2512         uint32_t n_tx_queue, nb_lcores;
2513         uint32_t dev_rxq_num, dev_txq_num;
2514         uint8_t nb_rx_queue, queue, socketid;
2515         uint16_t portid;
2516         const char *ptr_strings[NUM_TELSTATS];
2517
2518         /* catch SIGINT and restore cpufreq governor to ondemand */
2519         signal(SIGINT, signal_exit_now);
2520
2521         /* init EAL */
2522         ret = rte_eal_init(argc, argv);
2523         if (ret < 0)
2524                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2525         argc -= ret;
2526         argv += ret;
2527
2528         /* init RTE timer library to be used late */
2529         rte_timer_subsystem_init();
2530
2531         /* parse application arguments (after the EAL ones) */
2532         ret = parse_args(argc, argv);
2533         if (ret < 0)
2534                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
2535
2536         if (app_mode == APP_MODE_DEFAULT)
2537                 app_mode = autodetect_mode();
2538
2539         RTE_LOG(INFO, L3FWD_POWER, "Selected operation mode: %s\n",
2540                         mode_to_str(app_mode));
2541
2542         /* only legacy and empty poll mode rely on power library */
2543         if ((app_mode == APP_MODE_LEGACY || app_mode == APP_MODE_EMPTY_POLL) &&
2544                         init_power_library())
2545                 rte_exit(EXIT_FAILURE, "init_power_library failed\n");
2546
2547         if (update_lcore_params() < 0)
2548                 rte_exit(EXIT_FAILURE, "update_lcore_params failed\n");
2549
2550         if (check_lcore_params() < 0)
2551                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
2552
2553         ret = init_lcore_rx_queues();
2554         if (ret < 0)
2555                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2556
2557         nb_ports = rte_eth_dev_count_avail();
2558
2559         if (check_port_config() < 0)
2560                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
2561
2562         nb_lcores = rte_lcore_count();
2563
2564         /* initialize all ports */
2565         RTE_ETH_FOREACH_DEV(portid) {
2566                 struct rte_eth_conf local_port_conf = port_conf;
2567                 /* not all app modes need interrupts */
2568                 bool need_intr = app_mode == APP_MODE_LEGACY ||
2569                                 app_mode == APP_MODE_INTERRUPT;
2570
2571                 /* skip ports that are not enabled */
2572                 if ((enabled_port_mask & (1 << portid)) == 0) {
2573                         printf("\nSkipping disabled port %d\n", portid);
2574                         continue;
2575                 }
2576
2577                 /* init port */
2578                 printf("Initializing port %d ... ", portid );
2579                 fflush(stdout);
2580
2581                 ret = rte_eth_dev_info_get(portid, &dev_info);
2582                 if (ret != 0)
2583                         rte_exit(EXIT_FAILURE,
2584                                 "Error during getting device (port %u) info: %s\n",
2585                                 portid, strerror(-ret));
2586
2587                 dev_rxq_num = dev_info.max_rx_queues;
2588                 dev_txq_num = dev_info.max_tx_queues;
2589
2590                 nb_rx_queue = get_port_n_rx_queues(portid);
2591                 if (nb_rx_queue > dev_rxq_num)
2592                         rte_exit(EXIT_FAILURE,
2593                                 "Cannot configure not existed rxq: "
2594                                 "port=%d\n", portid);
2595
2596                 n_tx_queue = nb_lcores;
2597                 if (n_tx_queue > dev_txq_num)
2598                         n_tx_queue = dev_txq_num;
2599                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
2600                         nb_rx_queue, (unsigned)n_tx_queue );
2601                 /* If number of Rx queue is 0, no need to enable Rx interrupt */
2602                 if (nb_rx_queue == 0)
2603                         need_intr = false;
2604
2605                 if (need_intr)
2606                         local_port_conf.intr_conf.rxq = 1;
2607
2608                 ret = rte_eth_dev_info_get(portid, &dev_info);
2609                 if (ret != 0)
2610                         rte_exit(EXIT_FAILURE,
2611                                 "Error during getting device (port %u) info: %s\n",
2612                                 portid, strerror(-ret));
2613
2614                 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
2615                         local_port_conf.txmode.offloads |=
2616                                 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
2617
2618                 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2619                         dev_info.flow_type_rss_offloads;
2620                 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2621                                 port_conf.rx_adv_conf.rss_conf.rss_hf) {
2622                         printf("Port %u modified RSS hash function based on hardware support,"
2623                                 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2624                                 portid,
2625                                 port_conf.rx_adv_conf.rss_conf.rss_hf,
2626                                 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2627                 }
2628
2629                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
2630                                         (uint16_t)n_tx_queue, &local_port_conf);
2631                 if (ret < 0)
2632                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
2633                                         "err=%d, port=%d\n", ret, portid);
2634
2635                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
2636                                                        &nb_txd);
2637                 if (ret < 0)
2638                         rte_exit(EXIT_FAILURE,
2639                                  "Cannot adjust number of descriptors: err=%d, port=%d\n",
2640                                  ret, portid);
2641
2642                 ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
2643                 if (ret < 0)
2644                         rte_exit(EXIT_FAILURE,
2645                                  "Cannot get MAC address: err=%d, port=%d\n",
2646                                  ret, portid);
2647
2648                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
2649                 printf(", ");
2650
2651                 /* init memory */
2652                 ret = init_mem(NB_MBUF);
2653                 if (ret < 0)
2654                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
2655
2656                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2657                         if (rte_lcore_is_enabled(lcore_id) == 0)
2658                                 continue;
2659
2660                         /* Initialize TX buffers */
2661                         qconf = &lcore_conf[lcore_id];
2662                         qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
2663                                 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
2664                                 rte_eth_dev_socket_id(portid));
2665                         if (qconf->tx_buffer[portid] == NULL)
2666                                 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
2667                                                  portid);
2668
2669                         rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
2670                 }
2671
2672                 /* init one TX queue per couple (lcore,port) */
2673                 queueid = 0;
2674                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2675                         if (rte_lcore_is_enabled(lcore_id) == 0)
2676                                 continue;
2677
2678                         if (queueid >= dev_txq_num)
2679                                 continue;
2680
2681                         if (numa_on)
2682                                 socketid = \
2683                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
2684                         else
2685                                 socketid = 0;
2686
2687                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
2688                         fflush(stdout);
2689
2690                         txconf = &dev_info.default_txconf;
2691                         txconf->offloads = local_port_conf.txmode.offloads;
2692                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
2693                                                      socketid, txconf);
2694                         if (ret < 0)
2695                                 rte_exit(EXIT_FAILURE,
2696                                         "rte_eth_tx_queue_setup: err=%d, "
2697                                                 "port=%d\n", ret, portid);
2698
2699                         qconf = &lcore_conf[lcore_id];
2700                         qconf->tx_queue_id[portid] = queueid;
2701                         queueid++;
2702
2703                         qconf->tx_port_id[qconf->n_tx_port] = portid;
2704                         qconf->n_tx_port++;
2705                 }
2706                 printf("\n");
2707         }
2708
2709         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2710                 if (rte_lcore_is_enabled(lcore_id) == 0)
2711                         continue;
2712
2713                 if (app_mode == APP_MODE_LEGACY) {
2714                         /* init timer structures for each enabled lcore */
2715                         rte_timer_init(&power_timers[lcore_id]);
2716                         hz = rte_get_timer_hz();
2717                         rte_timer_reset(&power_timers[lcore_id],
2718                                         hz/TIMER_NUMBER_PER_SECOND,
2719                                         SINGLE, lcore_id,
2720                                         power_timer_cb, NULL);
2721                 }
2722                 qconf = &lcore_conf[lcore_id];
2723                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
2724                 fflush(stdout);
2725
2726                 /* init RX queues */
2727                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
2728                         struct rte_eth_rxconf rxq_conf;
2729
2730                         portid = qconf->rx_queue_list[queue].port_id;
2731                         queueid = qconf->rx_queue_list[queue].queue_id;
2732
2733                         if (numa_on)
2734                                 socketid = \
2735                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
2736                         else
2737                                 socketid = 0;
2738
2739                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
2740                         fflush(stdout);
2741
2742                         ret = rte_eth_dev_info_get(portid, &dev_info);
2743                         if (ret != 0)
2744                                 rte_exit(EXIT_FAILURE,
2745                                         "Error during getting device (port %u) info: %s\n",
2746                                         portid, strerror(-ret));
2747
2748                         rxq_conf = dev_info.default_rxconf;
2749                         rxq_conf.offloads = port_conf.rxmode.offloads;
2750                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
2751                                 socketid, &rxq_conf,
2752                                 pktmbuf_pool[socketid]);
2753                         if (ret < 0)
2754                                 rte_exit(EXIT_FAILURE,
2755                                         "rte_eth_rx_queue_setup: err=%d, "
2756                                                 "port=%d\n", ret, portid);
2757
2758                         if (parse_ptype) {
2759                                 if (add_cb_parse_ptype(portid, queueid) < 0)
2760                                         rte_exit(EXIT_FAILURE,
2761                                                  "Fail to add ptype cb\n");
2762                         }
2763
2764                         if (app_mode == APP_MODE_PMD_MGMT) {
2765                                 ret = rte_power_ethdev_pmgmt_queue_enable(
2766                                                 lcore_id, portid, queueid,
2767                                                 pmgmt_type);
2768                                 if (ret < 0)
2769                                         rte_exit(EXIT_FAILURE,
2770                                                 "rte_power_ethdev_pmgmt_queue_enable: err=%d, port=%d\n",
2771                                                         ret, portid);
2772                         }
2773                 }
2774         }
2775
2776         printf("\n");
2777
2778         /* start ports */
2779         RTE_ETH_FOREACH_DEV(portid) {
2780                 if ((enabled_port_mask & (1 << portid)) == 0) {
2781                         continue;
2782                 }
2783                 /* Start device */
2784                 ret = rte_eth_dev_start(portid);
2785                 if (ret < 0)
2786                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
2787                                                 "port=%d\n", ret, portid);
2788                 /*
2789                  * If enabled, put device in promiscuous mode.
2790                  * This allows IO forwarding mode to forward packets
2791                  * to itself through 2 cross-connected  ports of the
2792                  * target machine.
2793                  */
2794                 if (promiscuous_on) {
2795                         ret = rte_eth_promiscuous_enable(portid);
2796                         if (ret != 0)
2797                                 rte_exit(EXIT_FAILURE,
2798                                         "rte_eth_promiscuous_enable: err=%s, port=%u\n",
2799                                         rte_strerror(-ret), portid);
2800                 }
2801                 /* initialize spinlock for each port */
2802                 rte_spinlock_init(&(locks[portid]));
2803
2804                 if (!parse_ptype)
2805                         if (!check_ptype(portid))
2806                                 rte_exit(EXIT_FAILURE,
2807                                         "PMD can not provide needed ptypes\n");
2808         }
2809
2810         check_all_ports_link_status(enabled_port_mask);
2811
2812         if (app_mode == APP_MODE_EMPTY_POLL) {
2813
2814                 if (empty_poll_train) {
2815                         policy.state = TRAINING;
2816                 } else {
2817                         policy.state = MED_NORMAL;
2818                         policy.med_base_edpi = ep_med_edpi;
2819                         policy.hgh_base_edpi = ep_hgh_edpi;
2820                 }
2821
2822                 ret = rte_power_empty_poll_stat_init(&ep_params,
2823                                 freq_tlb,
2824                                 &policy);
2825                 if (ret < 0)
2826                         rte_exit(EXIT_FAILURE, "empty poll init failed");
2827         }
2828
2829
2830         /* launch per-lcore init on every lcore */
2831         if (app_mode == APP_MODE_LEGACY) {
2832                 rte_eal_mp_remote_launch(main_legacy_loop, NULL, CALL_MAIN);
2833         } else if (app_mode == APP_MODE_EMPTY_POLL) {
2834                 empty_poll_stop = false;
2835                 rte_eal_mp_remote_launch(main_empty_poll_loop, NULL,
2836                                 SKIP_MAIN);
2837         } else if (app_mode == APP_MODE_TELEMETRY) {
2838                 unsigned int i;
2839
2840                 /* Init metrics library */
2841                 rte_metrics_init(rte_socket_id());
2842                 /** Register stats with metrics library */
2843                 for (i = 0; i < NUM_TELSTATS; i++)
2844                         ptr_strings[i] = telstats_strings[i].name;
2845
2846                 ret = rte_metrics_reg_names(ptr_strings, NUM_TELSTATS);
2847                 if (ret >= 0)
2848                         telstats_index = ret;
2849                 else
2850                         rte_exit(EXIT_FAILURE, "failed to register metrics names");
2851
2852                 RTE_LCORE_FOREACH_WORKER(lcore_id) {
2853                         rte_spinlock_init(&stats[lcore_id].telemetry_lock);
2854                 }
2855                 rte_timer_init(&telemetry_timer);
2856                 rte_telemetry_register_cmd("/l3fwd-power/stats",
2857                                 handle_app_stats,
2858                                 "Returns global power stats. Parameters: None");
2859                 rte_eal_mp_remote_launch(main_telemetry_loop, NULL,
2860                                                 SKIP_MAIN);
2861         } else if (app_mode == APP_MODE_INTERRUPT) {
2862                 rte_eal_mp_remote_launch(main_intr_loop, NULL, CALL_MAIN);
2863         } else if (app_mode == APP_MODE_PMD_MGMT) {
2864                 /* reuse telemetry loop for PMD power management mode */
2865                 rte_eal_mp_remote_launch(main_telemetry_loop, NULL, CALL_MAIN);
2866         }
2867
2868         if (app_mode == APP_MODE_EMPTY_POLL || app_mode == APP_MODE_TELEMETRY)
2869                 launch_timer(rte_lcore_id());
2870
2871         RTE_LCORE_FOREACH_WORKER(lcore_id) {
2872                 if (rte_eal_wait_lcore(lcore_id) < 0)
2873                         return -1;
2874         }
2875
2876         if (app_mode == APP_MODE_PMD_MGMT) {
2877                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2878                         if (rte_lcore_is_enabled(lcore_id) == 0)
2879                                 continue;
2880                         qconf = &lcore_conf[lcore_id];
2881                         for (queue = 0; queue < qconf->n_rx_queue; ++queue) {
2882                                 portid = qconf->rx_queue_list[queue].port_id;
2883                                 queueid = qconf->rx_queue_list[queue].queue_id;
2884
2885                                 rte_power_ethdev_pmgmt_queue_disable(lcore_id,
2886                                                 portid, queueid);
2887                         }
2888                 }
2889         }
2890
2891         RTE_ETH_FOREACH_DEV(portid)
2892         {
2893                 if ((enabled_port_mask & (1 << portid)) == 0)
2894                         continue;
2895
2896                 ret = rte_eth_dev_stop(portid);
2897                 if (ret != 0)
2898                         RTE_LOG(ERR, L3FWD_POWER, "rte_eth_dev_stop: err=%d, port=%u\n",
2899                                 ret, portid);
2900
2901                 rte_eth_dev_close(portid);
2902         }
2903
2904         if (app_mode == APP_MODE_EMPTY_POLL)
2905                 rte_power_empty_poll_stat_free();
2906
2907         if ((app_mode == APP_MODE_LEGACY || app_mode == APP_MODE_EMPTY_POLL) &&
2908                         deinit_power_library())
2909                 rte_exit(EXIT_FAILURE, "deinit_power_library failed\n");
2910
2911         if (rte_eal_cleanup() < 0)
2912                 RTE_LOG(ERR, L3FWD_POWER, "EAL cleanup failed\n");
2913
2914         return 0;
2915 }