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_cycles.h>
  28 #include <rte_prefetch.h>
  29 #include <rte_lcore.h>
  30 #include <rte_per_lcore.h>
  31 #include <rte_branch_prediction.h>
  32 #include <rte_interrupts.h>
  33 #include <rte_random.h>
  34 #include <rte_debug.h>
  35 #include <rte_ether.h>
  36 #include <rte_ethdev.h>
  37 #include <rte_mempool.h>
  38 #include <rte_mbuf.h>
  39 #include <rte_ip.h>
  40 #include <rte_tcp.h>
  41 #include <rte_udp.h>
  42 #include <rte_string_fns.h>
  43 #include <rte_timer.h>
  44 #include <rte_power.h>
  45 #include <rte_spinlock.h>
  46 #include <rte_power_empty_poll.h>
  47 #include <rte_metrics.h>
  48 #include <rte_telemetry.h>
  49 #include <rte_power_pmd_mgmt.h>
  50
  51 #include "perf_core.h"
  52 #include "main.h"
  53
  54 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
  55
  56 #define MAX_PKT_BURST 32
  57
  58 #define MIN_ZERO_POLL_COUNT 10
  59
  60 /* 100 ms interval */
  61 #define TIMER_NUMBER_PER_SECOND           10
  62 /* (10ms) */
  63 #define INTERVALS_PER_SECOND             100
  64 /* 100000 us */
  65 #define SCALING_PERIOD                    (1000000/TIMER_NUMBER_PER_SECOND)
  66 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
  67
  68 #define APP_LOOKUP_EXACT_MATCH          0
  69 #define APP_LOOKUP_LPM                  1
  70 #define DO_RFC_1812_CHECKS
  71
  72 #ifndef APP_LOOKUP_METHOD
  73 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  74 #endif
  75
  76 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  77 #include <rte_hash.h>
  78 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
  79 #include <rte_lpm.h>
  80 #else
  81 #error "APP_LOOKUP_METHOD set to incorrect value"
  82 #endif
  83
  84 #ifndef IPv6_BYTES
  85 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
  86                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
  87 #define IPv6_BYTES(addr) \
  88         addr[0],  addr[1], addr[2],  addr[3], \
  89         addr[4],  addr[5], addr[6],  addr[7], \
  90         addr[8],  addr[9], addr[10], addr[11],\
  91         addr[12], addr[13],addr[14], addr[15]
  92 #endif
  93
  94 #define MAX_JUMBO_PKT_LEN  9600
  95
  96 #define IPV6_ADDR_LEN 16
  97
  98 #define MEMPOOL_CACHE_SIZE 256
  99
 100 /*
 101  * This expression is used to calculate the number of mbufs needed depending on
 102  * user input, taking into account memory for rx and tx hardware rings, cache
 103  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 104  * NB_MBUF never goes below a minimum value of 8192.
 105  */
 106
 107 #define NB_MBUF RTE_MAX ( \
 108         (nb_ports*nb_rx_queue*nb_rxd + \
 109         nb_ports*nb_lcores*MAX_PKT_BURST + \
 110         nb_ports*n_tx_queue*nb_txd + \
 111         nb_lcores*MEMPOOL_CACHE_SIZE), \
 112         (unsigned)8192)
 113
 114 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 115
 116 #define NB_SOCKETS 8
 117
 118 /* Configure how many packets ahead to prefetch, when reading packets */
 119 #define PREFETCH_OFFSET 3
 120
 121 /*
 122  * Configurable number of RX/TX ring descriptors
 123  */
 124 #define RTE_TEST_RX_DESC_DEFAULT 1024
 125 #define RTE_TEST_TX_DESC_DEFAULT 1024
 126
 127 /*
 128  * These two thresholds were decided on by running the training algorithm on
 129  * a 2.5GHz Xeon. These defaults can be overridden by supplying non-zero values
 130  * for the med_threshold and high_threshold parameters on the command line.
 131  */
 132 #define EMPTY_POLL_MED_THRESHOLD 350000UL
 133 #define EMPTY_POLL_HGH_THRESHOLD 580000UL
 134
 135 #define NUM_TELSTATS RTE_DIM(telstats_strings)
 136
 137 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 138 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 139
 140 /* ethernet addresses of ports */
 141 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 142
 143 /* ethernet addresses of ports */
 144 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
 145
 146 /* mask of enabled ports */
 147 static uint32_t enabled_port_mask = 0;
 148 /* Ports set in promiscuous mode off by default. */
 149 static int promiscuous_on = 0;
 150 /* NUMA is enabled by default. */
 151 static int numa_on = 1;
 152 static bool empty_poll_stop;
 153 static bool empty_poll_train;
 154 volatile bool quit_signal;
 155 static struct  ep_params *ep_params;
 156 static struct  ep_policy policy;
 157 static long  ep_med_edpi, ep_hgh_edpi;
 158 /* timer to update telemetry every 500ms */
 159 static struct rte_timer telemetry_timer;
 160
 161 /* stats index returned by metrics lib */
 162 int telstats_index;
 163
 164 struct telstats_name {
 165         char name[RTE_ETH_XSTATS_NAME_SIZE];
 166 };
 167
 168 /* telemetry stats to be reported */
 169 const struct telstats_name telstats_strings[] = {
 170         {"empty_poll"},
 171         {"full_poll"},
 172         {"busy_percent"}
 173 };
 174
 175 /* core busyness in percentage */
 176 enum busy_rate {
 177         ZERO = 0,
 178         PARTIAL = 50,
 179         FULL = 100
 180 };
 181
 182 /* reference poll count to measure core busyness */
 183 #define DEFAULT_COUNT 10000
 184 /*
 185  * reference CYCLES to be used to
 186  * measure core busyness based on poll count
 187  */
 188 #define MIN_CYCLES  1500000ULL
 189 #define MAX_CYCLES 22000000ULL
 190
 191 /* (500ms) */
 192 #define TELEMETRY_INTERVALS_PER_SEC 2
 193
 194 static int parse_ptype; /**< Parse packet type using rx callback, and */
 195                         /**< disabled by default */
 196
 197 enum appmode {
 198         APP_MODE_DEFAULT = 0,
 199         APP_MODE_LEGACY,
 200         APP_MODE_EMPTY_POLL,
 201         APP_MODE_TELEMETRY,
 202         APP_MODE_INTERRUPT,
 203         APP_MODE_PMD_MGMT
 204 };
 205
 206 enum appmode app_mode;
 207
 208 static enum rte_power_pmd_mgmt_type pmgmt_type;
 209 bool baseline_enabled;
 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        = RTE_ETH_MQ_RX_RSS,
 253                 .split_hdr_size = 0,
 254                 .offloads = RTE_ETH_RX_OFFLOAD_CHECKSUM,
 255         },
 256         .rx_adv_conf = {
 257                 .rss_conf = {
 258                         .rss_key = NULL,
 259                         .rss_hf = RTE_ETH_RSS_UDP,
 260                 },
 261         },
 262         .txmode = {
 263                 .mq_mode = RTE_ETH_MQ_TX_NONE,
 264         }
 265 };
 266
 267 static uint32_t max_pkt_len;
 268
 269 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 270
 271
 272 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 273
 274 #ifdef RTE_ARCH_X86
 275 #include <rte_hash_crc.h>
 276 #define DEFAULT_HASH_FUNC       rte_hash_crc
 277 #else
 278 #include <rte_jhash.h>
 279 #define DEFAULT_HASH_FUNC       rte_jhash
 280 #endif
 281
 282 struct ipv4_5tuple {
 283         uint32_t ip_dst;
 284         uint32_t ip_src;
 285         uint16_t port_dst;
 286         uint16_t port_src;
 287         uint8_t  proto;
 288 } __rte_packed;
 289
 290 struct ipv6_5tuple {
 291         uint8_t  ip_dst[IPV6_ADDR_LEN];
 292         uint8_t  ip_src[IPV6_ADDR_LEN];
 293         uint16_t port_dst;
 294         uint16_t port_src;
 295         uint8_t  proto;
 296 } __rte_packed;
 297
 298 struct ipv4_l3fwd_route {
 299         struct ipv4_5tuple key;
 300         uint8_t if_out;
 301 };
 302
 303 struct ipv6_l3fwd_route {
 304         struct ipv6_5tuple key;
 305         uint8_t if_out;
 306 };
 307
 308 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 309         {{RTE_IPV4(100,10,0,1), RTE_IPV4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 310         {{RTE_IPV4(100,20,0,2), RTE_IPV4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 311         {{RTE_IPV4(100,30,0,3), RTE_IPV4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 312         {{RTE_IPV4(100,40,0,4), RTE_IPV4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 313 };
 314
 315 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 316         {
 317                 {
 318                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 319                          0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 320                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 321                          0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
 322                          1, 10, IPPROTO_UDP
 323                 }, 4
 324         },
 325 };
 326
 327 typedef struct rte_hash lookup_struct_t;
 328 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 329 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 330
 331 #define L3FWD_HASH_ENTRIES      1024
 332
 333 static uint16_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 334 static uint16_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 335 #endif
 336
 337 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 338 struct ipv4_l3fwd_route {
 339         uint32_t ip;
 340         uint8_t  depth;
 341         uint8_t  if_out;
 342 };
 343
 344 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 345         {RTE_IPV4(1,1,1,0), 24, 0},
 346         {RTE_IPV4(2,1,1,0), 24, 1},
 347         {RTE_IPV4(3,1,1,0), 24, 2},
 348         {RTE_IPV4(4,1,1,0), 24, 3},
 349         {RTE_IPV4(5,1,1,0), 24, 4},
 350         {RTE_IPV4(6,1,1,0), 24, 5},
 351         {RTE_IPV4(7,1,1,0), 24, 6},
 352         {RTE_IPV4(8,1,1,0), 24, 7},
 353 };
 354
 355 #define IPV4_L3FWD_LPM_MAX_RULES     1024
 356
 357 typedef struct rte_lpm lookup_struct_t;
 358 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 359 #endif
 360
 361 struct lcore_conf {
 362         uint16_t n_rx_queue;
 363         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 364         uint16_t n_tx_port;
 365         uint16_t tx_port_id[RTE_MAX_ETHPORTS];
 366         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 367         struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
 368         lookup_struct_t * ipv4_lookup_struct;
 369         lookup_struct_t * ipv6_lookup_struct;
 370 } __rte_cache_aligned;
 371
 372 struct lcore_stats {
 373         /* total sleep time in ms since last frequency scaling down */
 374         uint32_t sleep_time;
 375         /* number of long sleep recently */
 376         uint32_t nb_long_sleep;
 377         /* freq. scaling up trend */
 378         uint32_t trend;
 379         /* total packet processed recently */
 380         uint64_t nb_rx_processed;
 381         /* total iterations looped recently */
 382         uint64_t nb_iteration_looped;
 383         /*
 384          * Represents empty and non empty polls
 385          * of rte_eth_rx_burst();
 386          * ep_nep[0] holds non empty polls
 387          * i.e. 0 < nb_rx <= MAX_BURST
 388          * ep_nep[1] holds empty polls.
 389          * i.e. nb_rx == 0
 390          */
 391         uint64_t ep_nep[2];
 392         /*
 393          * Represents full and empty+partial
 394          * polls of rte_eth_rx_burst();
 395          * ep_nep[0] holds empty+partial polls.
 396          * i.e. 0 <= nb_rx < MAX_BURST
 397          * ep_nep[1] holds full polls
 398          * i.e. nb_rx == MAX_BURST
 399          */
 400         uint64_t fp_nfp[2];
 401         enum busy_rate br;
 402         rte_spinlock_t telemetry_lock;
 403 } __rte_cache_aligned;
 404
 405 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
 406 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
 407 static struct rte_timer power_timers[RTE_MAX_LCORE];
 408
 409 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
 410 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
 411                 unsigned int lcore_id, uint16_t port_id, uint16_t queue_id);
 412
 413
 414 /*
 415  * These defaults are using the max frequency index (1), a medium index (9)
 416  * and a typical low frequency index (14). These can be adjusted to use
 417  * different indexes using the relevant command line parameters.
 418  */
 419 static uint8_t  freq_tlb[] = {14, 9, 1};
 420
 421 static int is_done(void)
 422 {
 423         return quit_signal;
 424 }
 425
 426 /* exit signal handler */
 427 static void
 428 signal_exit_now(int sigtype)
 429 {
 430
 431         if (sigtype == SIGINT)
 432                 quit_signal = true;
 433
 434 }
 435
 436 /*  Freqency scale down timer callback */
 437 static void
 438 power_timer_cb(__rte_unused struct rte_timer *tim,
 439                           __rte_unused void *arg)
 440 {
 441         uint64_t hz;
 442         float sleep_time_ratio;
 443         unsigned lcore_id = rte_lcore_id();
 444
 445         /* accumulate total execution time in us when callback is invoked */
 446         sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
 447                                         (float)SCALING_PERIOD;
 448         /**
 449          * check whether need to scale down frequency a step if it sleep a lot.
 450          */
 451         if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
 452                 if (rte_power_freq_down)
 453                         rte_power_freq_down(lcore_id);
 454         }
 455         else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
 456                 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
 457                 /**
 458                  * scale down a step if average packet per iteration less
 459                  * than expectation.
 460                  */
 461                 if (rte_power_freq_down)
 462                         rte_power_freq_down(lcore_id);
 463         }
 464
 465         /**
 466          * initialize another timer according to current frequency to ensure
 467          * timer interval is relatively fixed.
 468          */
 469         hz = rte_get_timer_hz();
 470         rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
 471                                 SINGLE, lcore_id, power_timer_cb, NULL);
 472
 473         stats[lcore_id].nb_rx_processed = 0;
 474         stats[lcore_id].nb_iteration_looped = 0;
 475
 476         stats[lcore_id].sleep_time = 0;
 477 }
 478
 479 /* Enqueue a single packet, and send burst if queue is filled */
 480 static inline int
 481 send_single_packet(struct rte_mbuf *m, uint16_t port)
 482 {
 483         uint32_t lcore_id;
 484         struct lcore_conf *qconf;
 485
 486         lcore_id = rte_lcore_id();
 487         qconf = &lcore_conf[lcore_id];
 488
 489         rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
 490                         qconf->tx_buffer[port], m);
 491
 492         return 0;
 493 }
 494
 495 #ifdef DO_RFC_1812_CHECKS
 496 static inline int
 497 is_valid_ipv4_pkt(struct rte_ipv4_hdr *pkt, uint32_t link_len)
 498 {
 499         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 500         /*
 501          * 1. The packet length reported by the Link Layer must be large
 502          * enough to hold the minimum length legal IP datagram (20 bytes).
 503          */
 504         if (link_len < sizeof(struct rte_ipv4_hdr))
 505                 return -1;
 506
 507         /* 2. The IP checksum must be correct. */
 508         /* if this is not checked in H/W, check it. */
 509         if ((port_conf.rxmode.offloads & RTE_ETH_RX_OFFLOAD_IPV4_CKSUM) == 0) {
 510                 uint16_t actual_cksum, expected_cksum;
 511                 actual_cksum = pkt->hdr_checksum;
 512                 pkt->hdr_checksum = 0;
 513                 expected_cksum = rte_ipv4_cksum(pkt);
 514                 if (actual_cksum != expected_cksum)
 515                         return -2;
 516         }
 517
 518         /*
 519          * 3. The IP version number must be 4. If the version number is not 4
 520          * then the packet may be another version of IP, such as IPng or
 521          * ST-II.
 522          */
 523         if (((pkt->version_ihl) >> 4) != 4)
 524                 return -3;
 525         /*
 526          * 4. The IP header length field must be large enough to hold the
 527          * minimum length legal IP datagram (20 bytes = 5 words).
 528          */
 529         if ((pkt->version_ihl & 0xf) < 5)
 530                 return -4;
 531
 532         /*
 533          * 5. The IP total length field must be large enough to hold the IP
 534          * datagram header, whose length is specified in the IP header length
 535          * field.
 536          */
 537         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct rte_ipv4_hdr))
 538                 return -5;
 539
 540         return 0;
 541 }
 542 #endif
 543
 544 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 545 static void
 546 print_ipv4_key(struct ipv4_5tuple key)
 547 {
 548         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
 549                 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
 550                                 key.port_dst, key.port_src, key.proto);
 551 }
 552 static void
 553 print_ipv6_key(struct ipv6_5tuple key)
 554 {
 555         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 556                 "port dst = %d, port src = %d, proto = %d\n",
 557                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 558                 key.port_dst, key.port_src, key.proto);
 559 }
 560
 561 static inline uint16_t
 562 get_ipv4_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
 563                 lookup_struct_t * ipv4_l3fwd_lookup_struct)
 564 {
 565         struct ipv4_5tuple key;
 566         struct rte_tcp_hdr *tcp;
 567         struct rte_udp_hdr *udp;
 568         int ret = 0;
 569
 570         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 571         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 572         key.proto = ipv4_hdr->next_proto_id;
 573
 574         switch (ipv4_hdr->next_proto_id) {
 575         case IPPROTO_TCP:
 576                 tcp = (struct rte_tcp_hdr *)((unsigned char *)ipv4_hdr +
 577                                         sizeof(struct rte_ipv4_hdr));
 578                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 579                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 580                 break;
 581
 582         case IPPROTO_UDP:
 583                 udp = (struct rte_udp_hdr *)((unsigned char *)ipv4_hdr +
 584                                         sizeof(struct rte_ipv4_hdr));
 585                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 586                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 587                 break;
 588
 589         default:
 590                 key.port_dst = 0;
 591                 key.port_src = 0;
 592                 break;
 593         }
 594
 595         /* Find destination port */
 596         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 597         return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
 598 }
 599
 600 static inline uint16_t
 601 get_ipv6_dst_port(struct rte_ipv6_hdr *ipv6_hdr, uint16_t portid,
 602                         lookup_struct_t *ipv6_l3fwd_lookup_struct)
 603 {
 604         struct ipv6_5tuple key;
 605         struct rte_tcp_hdr *tcp;
 606         struct rte_udp_hdr *udp;
 607         int ret = 0;
 608
 609         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 610         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 611
 612         key.proto = ipv6_hdr->proto;
 613
 614         switch (ipv6_hdr->proto) {
 615         case IPPROTO_TCP:
 616                 tcp = (struct rte_tcp_hdr *)((unsigned char *) ipv6_hdr +
 617                                         sizeof(struct rte_ipv6_hdr));
 618                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 619                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 620                 break;
 621
 622         case IPPROTO_UDP:
 623                 udp = (struct rte_udp_hdr *)((unsigned char *) ipv6_hdr +
 624                                         sizeof(struct rte_ipv6_hdr));
 625                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 626                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 627                 break;
 628
 629         default:
 630                 key.port_dst = 0;
 631                 key.port_src = 0;
 632                 break;
 633         }
 634
 635         /* Find destination port */
 636         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 637         return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
 638 }
 639 #endif
 640
 641 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 642 static inline uint16_t
 643 get_ipv4_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
 644                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 645 {
 646         uint32_t next_hop;
 647
 648         return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 649                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 650                         next_hop : portid);
 651 }
 652 #endif
 653
 654 static inline void
 655 parse_ptype_one(struct rte_mbuf *m)
 656 {
 657         struct rte_ether_hdr *eth_hdr;
 658         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 659         uint16_t ether_type;
 660
 661         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 662         ether_type = eth_hdr->ether_type;
 663         if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
 664                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 665         else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6))
 666                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 667
 668         m->packet_type = packet_type;
 669 }
 670
 671 static uint16_t
 672 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
 673                struct rte_mbuf *pkts[], uint16_t nb_pkts,
 674                uint16_t max_pkts __rte_unused,
 675                void *user_param __rte_unused)
 676 {
 677         unsigned int i;
 678
 679         for (i = 0; i < nb_pkts; ++i)
 680                 parse_ptype_one(pkts[i]);
 681
 682         return nb_pkts;
 683 }
 684
 685 static int
 686 add_cb_parse_ptype(uint16_t portid, uint16_t queueid)
 687 {
 688         printf("Port %d: softly parse packet type info\n", portid);
 689         if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL))
 690                 return 0;
 691
 692         printf("Failed to add rx callback: port=%d\n", portid);
 693         return -1;
 694 }
 695
 696 static inline void
 697 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
 698                                 struct lcore_conf *qconf)
 699 {
 700         struct rte_ether_hdr *eth_hdr;
 701         struct rte_ipv4_hdr *ipv4_hdr;
 702         void *d_addr_bytes;
 703         uint16_t dst_port;
 704
 705         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 706
 707         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 708                 /* Handle IPv4 headers.*/
 709                 ipv4_hdr =
 710                         rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
 711                                                 sizeof(struct rte_ether_hdr));
 712
 713 #ifdef DO_RFC_1812_CHECKS
 714                 /* Check to make sure the packet is valid (RFC1812) */
 715                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 716                         rte_pktmbuf_free(m);
 717                         return;
 718                 }
 719 #endif
 720
 721                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 722                                         qconf->ipv4_lookup_struct);
 723                 if (dst_port >= RTE_MAX_ETHPORTS ||
 724                                 (enabled_port_mask & 1 << dst_port) == 0)
 725                         dst_port = portid;
 726
 727                 /* 02:00:00:00:00:xx */
 728                 d_addr_bytes = &eth_hdr->dst_addr.addr_bytes[0];
 729                 *((uint64_t *)d_addr_bytes) =
 730                         0x000000000002 + ((uint64_t)dst_port << 40);
 731
 732 #ifdef DO_RFC_1812_CHECKS
 733                 /* Update time to live and header checksum */
 734                 --(ipv4_hdr->time_to_live);
 735                 ++(ipv4_hdr->hdr_checksum);
 736 #endif
 737
 738                 /* src addr */
 739                 rte_ether_addr_copy(&ports_eth_addr[dst_port],
 740                                 &eth_hdr->src_addr);
 741
 742                 send_single_packet(m, dst_port);
 743         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 744                 /* Handle IPv6 headers.*/
 745 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 746                 struct rte_ipv6_hdr *ipv6_hdr;
 747
 748                 ipv6_hdr =
 749                         rte_pktmbuf_mtod_offset(m, struct rte_ipv6_hdr *,
 750                                                 sizeof(struct rte_ether_hdr));
 751
 752                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
 753                                         qconf->ipv6_lookup_struct);
 754
 755                 if (dst_port >= RTE_MAX_ETHPORTS ||
 756                                 (enabled_port_mask & 1 << dst_port) == 0)
 757                         dst_port = portid;
 758
 759                 /* 02:00:00:00:00:xx */
 760                 d_addr_bytes = &eth_hdr->dst_addr.addr_bytes[0];
 761                 *((uint64_t *)d_addr_bytes) =
 762                         0x000000000002 + ((uint64_t)dst_port << 40);
 763
 764                 /* src addr */
 765                 rte_ether_addr_copy(&ports_eth_addr[dst_port],
 766                                 &eth_hdr->src_addr);
 767
 768                 send_single_packet(m, dst_port);
 769 #else
 770                 /* We don't currently handle IPv6 packets in LPM mode. */
 771                 rte_pktmbuf_free(m);
 772 #endif
 773         } else
 774                 rte_pktmbuf_free(m);
 775
 776 }
 777
 778 #define MINIMUM_SLEEP_TIME         1
 779 #define SUSPEND_THRESHOLD          300
 780
 781 static inline uint32_t
 782 power_idle_heuristic(uint32_t zero_rx_packet_count)
 783 {
 784         /* If zero count is less than 100,  sleep 1us */
 785         if (zero_rx_packet_count < SUSPEND_THRESHOLD)
 786                 return MINIMUM_SLEEP_TIME;
 787         /* If zero count is less than 1000, sleep 100 us which is the
 788                 minimum latency switching from C3/C6 to C0
 789         */
 790         else
 791                 return SUSPEND_THRESHOLD;
 792 }
 793
 794 static inline enum freq_scale_hint_t
 795 power_freq_scaleup_heuristic(unsigned lcore_id,
 796                              uint16_t port_id,
 797                              uint16_t queue_id)
 798 {
 799         uint32_t rxq_count = rte_eth_rx_queue_count(port_id, queue_id);
 800 /**
 801  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
 802  * per iteration
 803  */
 804 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
 805 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
 806 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
 807 #define FREQ_UP_TREND1_ACC   1
 808 #define FREQ_UP_TREND2_ACC   100
 809 #define FREQ_UP_THRESHOLD    10000
 810
 811         if (likely(rxq_count > FREQ_GEAR3_RX_PACKET_THRESHOLD)) {
 812                 stats[lcore_id].trend = 0;
 813                 return FREQ_HIGHEST;
 814         } else if (likely(rxq_count > FREQ_GEAR2_RX_PACKET_THRESHOLD))
 815                 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
 816         else if (likely(rxq_count > FREQ_GEAR1_RX_PACKET_THRESHOLD))
 817                 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
 818
 819         if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
 820                 stats[lcore_id].trend = 0;
 821                 return FREQ_HIGHER;
 822         }
 823
 824         return FREQ_CURRENT;
 825 }
 826
 827 /**
 828  * force polling thread sleep until one-shot rx interrupt triggers
 829  * @param port_id
 830  *  Port id.
 831  * @param queue_id
 832  *  Rx queue id.
 833  * @return
 834  *  0 on success
 835  */
 836 static int
 837 sleep_until_rx_interrupt(int num, int lcore)
 838 {
 839         /*
 840          * we want to track when we are woken up by traffic so that we can go
 841          * back to sleep again without log spamming. Avoid cache line sharing
 842          * to prevent threads stepping on each others' toes.
 843          */
 844         static struct {
 845                 bool wakeup;
 846         } __rte_cache_aligned status[RTE_MAX_LCORE];
 847         struct rte_epoll_event event[num];
 848         int n, i;
 849         uint16_t port_id;
 850         uint8_t queue_id;
 851         void *data;
 852
 853         if (status[lcore].wakeup) {
 854                 RTE_LOG(INFO, L3FWD_POWER,
 855                                 "lcore %u sleeps until interrupt triggers\n",
 856                                 rte_lcore_id());
 857         }
 858
 859         n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, 10);
 860         for (i = 0; i < n; i++) {
 861                 data = event[i].epdata.data;
 862                 port_id = ((uintptr_t)data) >> CHAR_BIT;
 863                 queue_id = ((uintptr_t)data) &
 864                         RTE_LEN2MASK(CHAR_BIT, uint8_t);
 865                 RTE_LOG(INFO, L3FWD_POWER,
 866                         "lcore %u is waked up from rx interrupt on"
 867                         " port %d queue %d\n",
 868                         rte_lcore_id(), port_id, queue_id);
 869         }
 870         status[lcore].wakeup = n != 0;
 871
 872         return 0;
 873 }
 874
 875 static void turn_on_off_intr(struct lcore_conf *qconf, bool on)
 876 {
 877         int i;
 878         struct lcore_rx_queue *rx_queue;
 879         uint8_t queue_id;
 880         uint16_t port_id;
 881
 882         for (i = 0; i < qconf->n_rx_queue; ++i) {
 883                 rx_queue = &(qconf->rx_queue_list[i]);
 884                 port_id = rx_queue->port_id;
 885                 queue_id = rx_queue->queue_id;
 886
 887                 rte_spinlock_lock(&(locks[port_id]));
 888                 if (on)
 889                         rte_eth_dev_rx_intr_enable(port_id, queue_id);
 890                 else
 891                         rte_eth_dev_rx_intr_disable(port_id, queue_id);
 892                 rte_spinlock_unlock(&(locks[port_id]));
 893         }
 894 }
 895
 896 static int event_register(struct lcore_conf *qconf)
 897 {
 898         struct lcore_rx_queue *rx_queue;
 899         uint8_t queueid;
 900         uint16_t portid;
 901         uint32_t data;
 902         int ret;
 903         int i;
 904
 905         for (i = 0; i < qconf->n_rx_queue; ++i) {
 906                 rx_queue = &(qconf->rx_queue_list[i]);
 907                 portid = rx_queue->port_id;
 908                 queueid = rx_queue->queue_id;
 909                 data = portid << CHAR_BIT | queueid;
 910
 911                 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
 912                                                 RTE_EPOLL_PER_THREAD,
 913                                                 RTE_INTR_EVENT_ADD,
 914                                                 (void *)((uintptr_t)data));
 915                 if (ret)
 916                         return ret;
 917         }
 918
 919         return 0;
 920 }
 921
 922 /* Main processing loop. 8< */
 923 static int main_intr_loop(__rte_unused void *dummy)
 924 {
 925         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 926         unsigned int lcore_id;
 927         uint64_t prev_tsc, diff_tsc, cur_tsc;
 928         int i, j, nb_rx;
 929         uint8_t queueid;
 930         uint16_t portid;
 931         struct lcore_conf *qconf;
 932         struct lcore_rx_queue *rx_queue;
 933         uint32_t lcore_rx_idle_count = 0;
 934         uint32_t lcore_idle_hint = 0;
 935         int intr_en = 0;
 936
 937         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
 938                                    US_PER_S * BURST_TX_DRAIN_US;
 939
 940         prev_tsc = 0;
 941
 942         lcore_id = rte_lcore_id();
 943         qconf = &lcore_conf[lcore_id];
 944
 945         if (qconf->n_rx_queue == 0) {
 946                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
 947                                 lcore_id);
 948                 return 0;
 949         }
 950
 951         RTE_LOG(INFO, L3FWD_POWER, "entering main interrupt loop on lcore %u\n",
 952                         lcore_id);
 953
 954         for (i = 0; i < qconf->n_rx_queue; i++) {
 955                 portid = qconf->rx_queue_list[i].port_id;
 956                 queueid = qconf->rx_queue_list[i].queue_id;
 957                 RTE_LOG(INFO, L3FWD_POWER,
 958                                 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
 959                                 lcore_id, portid, queueid);
 960         }
 961
 962         /* add into event wait list */
 963         if (event_register(qconf) == 0)
 964                 intr_en = 1;
 965         else
 966                 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
 967
 968         while (!is_done()) {
 969                 stats[lcore_id].nb_iteration_looped++;
 970
 971                 cur_tsc = rte_rdtsc();
 972
 973                 /*
 974                  * TX burst queue drain
 975                  */
 976                 diff_tsc = cur_tsc - prev_tsc;
 977                 if (unlikely(diff_tsc > drain_tsc)) {
 978                         for (i = 0; i < qconf->n_tx_port; ++i) {
 979                                 portid = qconf->tx_port_id[i];
 980                                 rte_eth_tx_buffer_flush(portid,
 981                                                 qconf->tx_queue_id[portid],
 982                                                 qconf->tx_buffer[portid]);
 983                         }
 984                         prev_tsc = cur_tsc;
 985                 }
 986
 987 start_rx:
 988                 /*
 989                  * Read packet from RX queues
 990                  */
 991                 lcore_rx_idle_count = 0;
 992                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 993                         rx_queue = &(qconf->rx_queue_list[i]);
 994                         rx_queue->idle_hint = 0;
 995                         portid = rx_queue->port_id;
 996                         queueid = rx_queue->queue_id;
 997
 998                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 999                                         MAX_PKT_BURST);
1000
1001                         stats[lcore_id].nb_rx_processed += nb_rx;
1002                         if (unlikely(nb_rx == 0)) {
1003                                 /**
1004                                  * no packet received from rx queue, try to
1005                                  * sleep for a while forcing CPU enter deeper
1006                                  * C states.
1007                                  */
1008                                 rx_queue->zero_rx_packet_count++;
1009
1010                                 if (rx_queue->zero_rx_packet_count <=
1011                                                 MIN_ZERO_POLL_COUNT)
1012                                         continue;
1013
1014                                 rx_queue->idle_hint = power_idle_heuristic(
1015                                                 rx_queue->zero_rx_packet_count);
1016                                 lcore_rx_idle_count++;
1017                         } else {
1018                                 rx_queue->zero_rx_packet_count = 0;
1019                         }
1020
1021                         /* Prefetch first packets */
1022                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1023                                 rte_prefetch0(rte_pktmbuf_mtod(
1024                                                 pkts_burst[j], void *));
1025                         }
1026
1027                         /* Prefetch and forward already prefetched packets */
1028                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1029                                 rte_prefetch0(rte_pktmbuf_mtod(
1030                                                 pkts_burst[j + PREFETCH_OFFSET],
1031                                                 void *));
1032                                 l3fwd_simple_forward(
1033                                                 pkts_burst[j], portid, qconf);
1034                         }
1035
1036                         /* Forward remaining prefetched packets */
1037                         for (; j < nb_rx; j++) {
1038                                 l3fwd_simple_forward(
1039                                                 pkts_burst[j], portid, qconf);
1040                         }
1041                 }
1042
1043                 if (unlikely(lcore_rx_idle_count == qconf->n_rx_queue)) {
1044                         /**
1045                          * All Rx queues empty in recent consecutive polls,
1046                          * sleep in a conservative manner, meaning sleep as
1047                          * less as possible.
1048                          */
1049                         for (i = 1,
1050                             lcore_idle_hint = qconf->rx_queue_list[0].idle_hint;
1051                                         i < qconf->n_rx_queue; ++i) {
1052                                 rx_queue = &(qconf->rx_queue_list[i]);
1053                                 if (rx_queue->idle_hint < lcore_idle_hint)
1054                                         lcore_idle_hint = rx_queue->idle_hint;
1055                         }
1056
1057                         if (lcore_idle_hint < SUSPEND_THRESHOLD)
1058                                 /**
1059                                  * execute "pause" instruction to avoid context
1060                                  * switch which generally take hundred of
1061                                  * microseconds for short sleep.
1062                                  */
1063                                 rte_delay_us(lcore_idle_hint);
1064                         else {
1065                                 /* suspend until rx interrupt triggers */
1066                                 if (intr_en) {
1067                                         turn_on_off_intr(qconf, 1);
1068                                         sleep_until_rx_interrupt(
1069                                                         qconf->n_rx_queue,
1070                                                         lcore_id);
1071                                         turn_on_off_intr(qconf, 0);
1072                                         /**
1073                                          * start receiving packets immediately
1074                                          */
1075                                         if (likely(!is_done()))
1076                                                 goto start_rx;
1077                                 }
1078                         }
1079                         stats[lcore_id].sleep_time += lcore_idle_hint;
1080                 }
1081         }
1082
1083         return 0;
1084 }
1085 /* >8 End of main processing loop. */
1086
1087 /* main processing loop */
1088 static int
1089 main_telemetry_loop(__rte_unused void *dummy)
1090 {
1091         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1092         unsigned int lcore_id;
1093         uint64_t prev_tsc, diff_tsc, cur_tsc, prev_tel_tsc;
1094         int i, j, nb_rx;
1095         uint8_t queueid;
1096         uint16_t portid;
1097         struct lcore_conf *qconf;
1098         struct lcore_rx_queue *rx_queue;
1099         uint64_t ep_nep[2] = {0}, fp_nfp[2] = {0};
1100         uint64_t poll_count;
1101         enum busy_rate br;
1102
1103         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1104                                         US_PER_S * BURST_TX_DRAIN_US;
1105
1106         poll_count = 0;
1107         prev_tsc = 0;
1108         prev_tel_tsc = 0;
1109
1110         lcore_id = rte_lcore_id();
1111         qconf = &lcore_conf[lcore_id];
1112
1113         if (qconf->n_rx_queue == 0) {
1114                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
1115                         lcore_id);
1116                 return 0;
1117         }
1118
1119         RTE_LOG(INFO, L3FWD_POWER, "entering main telemetry loop on lcore %u\n",
1120                 lcore_id);
1121
1122         for (i = 0; i < qconf->n_rx_queue; i++) {
1123                 portid = qconf->rx_queue_list[i].port_id;
1124                 queueid = qconf->rx_queue_list[i].queue_id;
1125                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1126                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
1127         }
1128
1129         while (!is_done()) {
1130
1131                 cur_tsc = rte_rdtsc();
1132                 /*
1133                  * TX burst queue drain
1134                  */
1135                 diff_tsc = cur_tsc - prev_tsc;
1136                 if (unlikely(diff_tsc > drain_tsc)) {
1137                         for (i = 0; i < qconf->n_tx_port; ++i) {
1138                                 portid = qconf->tx_port_id[i];
1139                                 rte_eth_tx_buffer_flush(portid,
1140                                                 qconf->tx_queue_id[portid],
1141                                                 qconf->tx_buffer[portid]);
1142                         }
1143                         prev_tsc = cur_tsc;
1144                 }
1145
1146                 /*
1147                  * Read packet from RX queues
1148                  */
1149                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1150                         rx_queue = &(qconf->rx_queue_list[i]);
1151                         portid = rx_queue->port_id;
1152                         queueid = rx_queue->queue_id;
1153
1154                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1155                                                                 MAX_PKT_BURST);
1156                         ep_nep[nb_rx == 0]++;
1157                         fp_nfp[nb_rx == MAX_PKT_BURST]++;
1158                         poll_count++;
1159                         if (unlikely(nb_rx == 0))
1160                                 continue;
1161
1162                         /* Prefetch first packets */
1163                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1164                                 rte_prefetch0(rte_pktmbuf_mtod(
1165                                                 pkts_burst[j], void *));
1166                         }
1167
1168                         /* Prefetch and forward already prefetched packets */
1169                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1170                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1171                                                 j + PREFETCH_OFFSET], void *));
1172                                 l3fwd_simple_forward(pkts_burst[j], portid,
1173                                                                 qconf);
1174                         }
1175
1176                         /* Forward remaining prefetched packets */
1177                         for (; j < nb_rx; j++) {
1178                                 l3fwd_simple_forward(pkts_burst[j], portid,
1179                                                                 qconf);
1180                         }
1181                 }
1182                 if (unlikely(poll_count >= DEFAULT_COUNT)) {
1183                         diff_tsc = cur_tsc - prev_tel_tsc;
1184                         if (diff_tsc >= MAX_CYCLES) {
1185                                 br = FULL;
1186                         } else if (diff_tsc > MIN_CYCLES &&
1187                                         diff_tsc < MAX_CYCLES) {
1188                                 br = (diff_tsc * 100) / MAX_CYCLES;
1189                         } else {
1190                                 br = ZERO;
1191                         }
1192                         poll_count = 0;
1193                         prev_tel_tsc = cur_tsc;
1194                         /* update stats for telemetry */
1195                         rte_spinlock_lock(&stats[lcore_id].telemetry_lock);
1196                         stats[lcore_id].ep_nep[0] = ep_nep[0];
1197                         stats[lcore_id].ep_nep[1] = ep_nep[1];
1198                         stats[lcore_id].fp_nfp[0] = fp_nfp[0];
1199                         stats[lcore_id].fp_nfp[1] = fp_nfp[1];
1200                         stats[lcore_id].br = br;
1201                         rte_spinlock_unlock(&stats[lcore_id].telemetry_lock);
1202                 }
1203         }
1204
1205         return 0;
1206 }
1207 /* main processing loop */
1208 static int
1209 main_empty_poll_loop(__rte_unused void *dummy)
1210 {
1211         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1212         unsigned int lcore_id;
1213         uint64_t prev_tsc, diff_tsc, cur_tsc;
1214         int i, j, nb_rx;
1215         uint8_t queueid;
1216         uint16_t portid;
1217         struct lcore_conf *qconf;
1218         struct lcore_rx_queue *rx_queue;
1219
1220         const uint64_t drain_tsc =
1221                 (rte_get_tsc_hz() + US_PER_S - 1) /
1222                 US_PER_S * BURST_TX_DRAIN_US;
1223
1224         prev_tsc = 0;
1225
1226         lcore_id = rte_lcore_id();
1227         qconf = &lcore_conf[lcore_id];
1228
1229         if (qconf->n_rx_queue == 0) {
1230                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
1231                         lcore_id);
1232                 return 0;
1233         }
1234
1235         for (i = 0; i < qconf->n_rx_queue; i++) {
1236                 portid = qconf->rx_queue_list[i].port_id;
1237                 queueid = qconf->rx_queue_list[i].queue_id;
1238                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1239                                 "rxqueueid=%hhu\n", lcore_id, portid, queueid);
1240         }
1241
1242         while (!is_done()) {
1243                 stats[lcore_id].nb_iteration_looped++;
1244
1245                 cur_tsc = rte_rdtsc();
1246                 /*
1247                  * TX burst queue drain
1248                  */
1249                 diff_tsc = cur_tsc - prev_tsc;
1250                 if (unlikely(diff_tsc > drain_tsc)) {
1251                         for (i = 0; i < qconf->n_tx_port; ++i) {
1252                                 portid = qconf->tx_port_id[i];
1253                                 rte_eth_tx_buffer_flush(portid,
1254                                                 qconf->tx_queue_id[portid],
1255                                                 qconf->tx_buffer[portid]);
1256                         }
1257                         prev_tsc = cur_tsc;
1258                 }
1259
1260                 /*
1261                  * Read packet from RX queues
1262                  */
1263                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1264                         rx_queue = &(qconf->rx_queue_list[i]);
1265                         rx_queue->idle_hint = 0;
1266                         portid = rx_queue->port_id;
1267                         queueid = rx_queue->queue_id;
1268
1269                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1270                                         MAX_PKT_BURST);
1271
1272                         stats[lcore_id].nb_rx_processed += nb_rx;
1273
1274                         if (nb_rx == 0) {
1275
1276                                 rte_power_empty_poll_stat_update(lcore_id);
1277
1278                                 continue;
1279                         } else {
1280                                 rte_power_poll_stat_update(lcore_id, nb_rx);
1281                         }
1282
1283
1284                         /* Prefetch first packets */
1285                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1286                                 rte_prefetch0(rte_pktmbuf_mtod(
1287                                                         pkts_burst[j], void *));
1288                         }
1289
1290                         /* Prefetch and forward already prefetched packets */
1291                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1292                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1293                                                         j + PREFETCH_OFFSET],
1294                                                         void *));
1295                                 l3fwd_simple_forward(pkts_burst[j], portid,
1296                                                 qconf);
1297                         }
1298
1299                         /* Forward remaining prefetched packets */
1300                         for (; j < nb_rx; j++) {
1301                                 l3fwd_simple_forward(pkts_burst[j], portid,
1302                                                 qconf);
1303                         }
1304
1305                 }
1306
1307         }
1308
1309         return 0;
1310 }
1311 /* main processing loop */
1312 static int
1313 main_legacy_loop(__rte_unused void *dummy)
1314 {
1315         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1316         unsigned lcore_id;
1317         uint64_t prev_tsc, diff_tsc, cur_tsc, tim_res_tsc, hz;
1318         uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
1319         int i, j, nb_rx;
1320         uint8_t queueid;
1321         uint16_t portid;
1322         struct lcore_conf *qconf;
1323         struct lcore_rx_queue *rx_queue;
1324         enum freq_scale_hint_t lcore_scaleup_hint;
1325         uint32_t lcore_rx_idle_count = 0;
1326         uint32_t lcore_idle_hint = 0;
1327         int intr_en = 0;
1328
1329         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
1330
1331         prev_tsc = 0;
1332         hz = rte_get_timer_hz();
1333         tim_res_tsc = hz/TIMER_NUMBER_PER_SECOND;
1334
1335         lcore_id = rte_lcore_id();
1336         qconf = &lcore_conf[lcore_id];
1337
1338         if (qconf->n_rx_queue == 0) {
1339                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
1340                 return 0;
1341         }
1342
1343         RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
1344
1345         for (i = 0; i < qconf->n_rx_queue; i++) {
1346                 portid = qconf->rx_queue_list[i].port_id;
1347                 queueid = qconf->rx_queue_list[i].queue_id;
1348                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1349                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
1350         }
1351
1352         /* add into event wait list */
1353         if (event_register(qconf) == 0)
1354                 intr_en = 1;
1355         else
1356                 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
1357
1358         while (!is_done()) {
1359                 stats[lcore_id].nb_iteration_looped++;
1360
1361                 cur_tsc = rte_rdtsc();
1362                 cur_tsc_power = cur_tsc;
1363
1364                 /*
1365                  * TX burst queue drain
1366                  */
1367                 diff_tsc = cur_tsc - prev_tsc;
1368                 if (unlikely(diff_tsc > drain_tsc)) {
1369                         for (i = 0; i < qconf->n_tx_port; ++i) {
1370                                 portid = qconf->tx_port_id[i];
1371                                 rte_eth_tx_buffer_flush(portid,
1372                                                 qconf->tx_queue_id[portid],
1373                                                 qconf->tx_buffer[portid]);
1374                         }
1375                         prev_tsc = cur_tsc;
1376                 }
1377
1378                 diff_tsc_power = cur_tsc_power - prev_tsc_power;
1379                 if (diff_tsc_power > tim_res_tsc) {
1380                         rte_timer_manage();
1381                         prev_tsc_power = cur_tsc_power;
1382                 }
1383
1384 start_rx:
1385                 /*
1386                  * Read packet from RX queues
1387                  */
1388                 lcore_scaleup_hint = FREQ_CURRENT;
1389                 lcore_rx_idle_count = 0;
1390                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1391                         rx_queue = &(qconf->rx_queue_list[i]);
1392                         rx_queue->idle_hint = 0;
1393                         portid = rx_queue->port_id;
1394                         queueid = rx_queue->queue_id;
1395
1396                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1397                                                                 MAX_PKT_BURST);
1398
1399                         stats[lcore_id].nb_rx_processed += nb_rx;
1400                         if (unlikely(nb_rx == 0)) {
1401                                 /**
1402                                  * no packet received from rx queue, try to
1403                                  * sleep for a while forcing CPU enter deeper
1404                                  * C states.
1405                                  */
1406                                 rx_queue->zero_rx_packet_count++;
1407
1408                                 if (rx_queue->zero_rx_packet_count <=
1409                                                         MIN_ZERO_POLL_COUNT)
1410                                         continue;
1411
1412                                 rx_queue->idle_hint = power_idle_heuristic(\
1413                                         rx_queue->zero_rx_packet_count);
1414                                 lcore_rx_idle_count++;
1415                         } else {
1416                                 rx_queue->zero_rx_packet_count = 0;
1417
1418                                 /**
1419                                  * do not scale up frequency immediately as
1420                                  * user to kernel space communication is costly
1421                                  * which might impact packet I/O for received
1422                                  * packets.
1423                                  */
1424                                 rx_queue->freq_up_hint =
1425                                         power_freq_scaleup_heuristic(lcore_id,
1426                                                         portid, queueid);
1427                         }
1428
1429                         /* Prefetch first packets */
1430                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1431                                 rte_prefetch0(rte_pktmbuf_mtod(
1432                                                 pkts_burst[j], void *));
1433                         }
1434
1435                         /* Prefetch and forward already prefetched packets */
1436                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1437                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1438                                                 j + PREFETCH_OFFSET], void *));
1439                                 l3fwd_simple_forward(pkts_burst[j], portid,
1440                                                                 qconf);
1441                         }
1442
1443                         /* Forward remaining prefetched packets */
1444                         for (; j < nb_rx; j++) {
1445                                 l3fwd_simple_forward(pkts_burst[j], portid,
1446                                                                 qconf);
1447                         }
1448                 }
1449
1450                 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
1451                         for (i = 1, lcore_scaleup_hint =
1452                                 qconf->rx_queue_list[0].freq_up_hint;
1453                                         i < qconf->n_rx_queue; ++i) {
1454                                 rx_queue = &(qconf->rx_queue_list[i]);
1455                                 if (rx_queue->freq_up_hint >
1456                                                 lcore_scaleup_hint)
1457                                         lcore_scaleup_hint =
1458                                                 rx_queue->freq_up_hint;
1459                         }
1460
1461                         if (lcore_scaleup_hint == FREQ_HIGHEST) {
1462                                 if (rte_power_freq_max)
1463                                         rte_power_freq_max(lcore_id);
1464                         } else if (lcore_scaleup_hint == FREQ_HIGHER) {
1465                                 if (rte_power_freq_up)
1466                                         rte_power_freq_up(lcore_id);
1467                         }
1468                 } else {
1469                         /**
1470                          * All Rx queues empty in recent consecutive polls,
1471                          * sleep in a conservative manner, meaning sleep as
1472                          * less as possible.
1473                          */
1474                         for (i = 1, lcore_idle_hint =
1475                                 qconf->rx_queue_list[0].idle_hint;
1476                                         i < qconf->n_rx_queue; ++i) {
1477                                 rx_queue = &(qconf->rx_queue_list[i]);
1478                                 if (rx_queue->idle_hint < lcore_idle_hint)
1479                                         lcore_idle_hint = rx_queue->idle_hint;
1480                         }
1481
1482                         if (lcore_idle_hint < SUSPEND_THRESHOLD)
1483                                 /**
1484                                  * execute "pause" instruction to avoid context
1485                                  * switch which generally take hundred of
1486                                  * microseconds for short sleep.
1487                                  */
1488                                 rte_delay_us(lcore_idle_hint);
1489                         else {
1490                                 /* suspend until rx interrupt triggers */
1491                                 if (intr_en) {
1492                                         turn_on_off_intr(qconf, 1);
1493                                         sleep_until_rx_interrupt(
1494                                                         qconf->n_rx_queue,
1495                                                         lcore_id);
1496                                         turn_on_off_intr(qconf, 0);
1497                                         /**
1498                                          * start receiving packets immediately
1499                                          */
1500                                         if (likely(!is_done()))
1501                                                 goto start_rx;
1502                                 }
1503                         }
1504                         stats[lcore_id].sleep_time += lcore_idle_hint;
1505                 }
1506         }
1507
1508         return 0;
1509 }
1510
1511 static int
1512 check_lcore_params(void)
1513 {
1514         uint8_t queue, lcore;
1515         uint16_t i;
1516         int socketid;
1517
1518         for (i = 0; i < nb_lcore_params; ++i) {
1519                 queue = lcore_params[i].queue_id;
1520                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1521                         printf("invalid queue number: %hhu\n", queue);
1522                         return -1;
1523                 }
1524                 lcore = lcore_params[i].lcore_id;
1525                 if (!rte_lcore_is_enabled(lcore)) {
1526                         printf("error: lcore %hhu is not enabled in lcore "
1527                                                         "mask\n", lcore);
1528                         return -1;
1529                 }
1530                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1531                                                         (numa_on == 0)) {
1532                         printf("warning: lcore %hhu is on socket %d with numa "
1533                                                 "off\n", lcore, socketid);
1534                 }
1535                 if (app_mode == APP_MODE_TELEMETRY && lcore == rte_lcore_id()) {
1536                         printf("cannot enable main core %d in config for telemetry mode\n",
1537                                 rte_lcore_id());
1538                         return -1;
1539                 }
1540         }
1541         return 0;
1542 }
1543
1544 static int
1545 check_port_config(void)
1546 {
1547         unsigned portid;
1548         uint16_t i;
1549
1550         for (i = 0; i < nb_lcore_params; ++i) {
1551                 portid = lcore_params[i].port_id;
1552                 if ((enabled_port_mask & (1 << portid)) == 0) {
1553                         printf("port %u is not enabled in port mask\n",
1554                                                                 portid);
1555                         return -1;
1556                 }
1557                 if (!rte_eth_dev_is_valid_port(portid)) {
1558                         printf("port %u is not present on the board\n",
1559                                                                 portid);
1560                         return -1;
1561                 }
1562         }
1563         return 0;
1564 }
1565
1566 static uint8_t
1567 get_port_n_rx_queues(const uint16_t port)
1568 {
1569         int queue = -1;
1570         uint16_t i;
1571
1572         for (i = 0; i < nb_lcore_params; ++i) {
1573                 if (lcore_params[i].port_id == port &&
1574                                 lcore_params[i].queue_id > queue)
1575                         queue = lcore_params[i].queue_id;
1576         }
1577         return (uint8_t)(++queue);
1578 }
1579
1580 static int
1581 init_lcore_rx_queues(void)
1582 {
1583         uint16_t i, nb_rx_queue;
1584         uint8_t lcore;
1585
1586         for (i = 0; i < nb_lcore_params; ++i) {
1587                 lcore = lcore_params[i].lcore_id;
1588                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1589                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1590                         printf("error: too many queues (%u) for lcore: %u\n",
1591                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1592                         return -1;
1593                 } else {
1594                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1595                                 lcore_params[i].port_id;
1596                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1597                                 lcore_params[i].queue_id;
1598                         lcore_conf[lcore].n_rx_queue++;
1599                 }
1600         }
1601         return 0;
1602 }
1603
1604 /* display usage */
1605 static void
1606 print_usage(const char *prgname)
1607 {
1608         printf ("%s [EAL options] -- -p PORTMASK -P"
1609                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1610                 "  [--high-perf-cores CORELIST"
1611                 "  [--perf-config (port,queue,hi_perf,lcore_index)[,(port,queue,hi_perf,lcore_index]]"
1612                 "  [--max-pkt-len PKTLEN]\n"
1613                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1614                 "  -P: enable promiscuous mode\n"
1615                 "  --config (port,queue,lcore): rx queues configuration\n"
1616                 "  --high-perf-cores CORELIST: list of high performance cores\n"
1617                 "  --perf-config: similar as config, cores specified as indices"
1618                 " for bins containing high or regular performance cores\n"
1619                 "  --no-numa: optional, disable numa awareness\n"
1620                 "  --max-pkt-len PKTLEN: maximum packet length in decimal (64-9600)\n"
1621                 "  --parse-ptype: parse packet type by software\n"
1622                 "  --legacy: use legacy interrupt-based scaling\n"
1623                 "  --empty-poll: enable empty poll detection"
1624                 " follow (training_flag, high_threshold, med_threshold)\n"
1625                 " --telemetry: enable telemetry mode, to update"
1626                 " empty polls, full polls, and core busyness to telemetry\n"
1627                 " --interrupt-only: enable interrupt-only mode\n"
1628                 " --pmd-mgmt MODE: enable PMD power management mode. "
1629                 "Currently supported modes: baseline, monitor, pause, scale\n",
1630                 prgname);
1631 }
1632
1633 static int parse_max_pkt_len(const char *pktlen)
1634 {
1635         char *end = NULL;
1636         unsigned long len;
1637
1638         /* parse decimal string */
1639         len = strtoul(pktlen, &end, 10);
1640         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1641                 return -1;
1642
1643         if (len == 0)
1644                 return -1;
1645
1646         return len;
1647 }
1648
1649 static int
1650 parse_portmask(const char *portmask)
1651 {
1652         char *end = NULL;
1653         unsigned long pm;
1654
1655         /* parse hexadecimal string */
1656         pm = strtoul(portmask, &end, 16);
1657         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1658                 return 0;
1659
1660         return pm;
1661 }
1662
1663 static int
1664 parse_config(const char *q_arg)
1665 {
1666         char s[256];
1667         const char *p, *p0 = q_arg;
1668         char *end;
1669         enum fieldnames {
1670                 FLD_PORT = 0,
1671                 FLD_QUEUE,
1672                 FLD_LCORE,
1673                 _NUM_FLD
1674         };
1675         unsigned long int_fld[_NUM_FLD];
1676         char *str_fld[_NUM_FLD];
1677         int i;
1678         unsigned size;
1679
1680         nb_lcore_params = 0;
1681
1682         while ((p = strchr(p0,'(')) != NULL) {
1683                 ++p;
1684                 if((p0 = strchr(p,')')) == NULL)
1685                         return -1;
1686
1687                 size = p0 - p;
1688                 if(size >= sizeof(s))
1689                         return -1;
1690
1691                 snprintf(s, sizeof(s), "%.*s", size, p);
1692                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1693                                                                 _NUM_FLD)
1694                         return -1;
1695                 for (i = 0; i < _NUM_FLD; i++){
1696                         errno = 0;
1697                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1698                         if (errno != 0 || end == str_fld[i] || int_fld[i] >
1699                                                                         255)
1700                                 return -1;
1701                 }
1702                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1703                         printf("exceeded max number of lcore params: %hu\n",
1704                                 nb_lcore_params);
1705                         return -1;
1706                 }
1707                 lcore_params_array[nb_lcore_params].port_id =
1708                                 (uint8_t)int_fld[FLD_PORT];
1709                 lcore_params_array[nb_lcore_params].queue_id =
1710                                 (uint8_t)int_fld[FLD_QUEUE];
1711                 lcore_params_array[nb_lcore_params].lcore_id =
1712                                 (uint8_t)int_fld[FLD_LCORE];
1713                 ++nb_lcore_params;
1714         }
1715         lcore_params = lcore_params_array;
1716
1717         return 0;
1718 }
1719
1720 static int
1721 parse_pmd_mgmt_config(const char *name)
1722 {
1723 #define PMD_MGMT_MONITOR "monitor"
1724 #define PMD_MGMT_PAUSE   "pause"
1725 #define PMD_MGMT_SCALE   "scale"
1726 #define PMD_MGMT_BASELINE  "baseline"
1727
1728         if (strncmp(PMD_MGMT_MONITOR, name, sizeof(PMD_MGMT_MONITOR)) == 0) {
1729                 pmgmt_type = RTE_POWER_MGMT_TYPE_MONITOR;
1730                 return 0;
1731         }
1732
1733         if (strncmp(PMD_MGMT_PAUSE, name, sizeof(PMD_MGMT_PAUSE)) == 0) {
1734                 pmgmt_type = RTE_POWER_MGMT_TYPE_PAUSE;
1735                 return 0;
1736         }
1737
1738         if (strncmp(PMD_MGMT_SCALE, name, sizeof(PMD_MGMT_SCALE)) == 0) {
1739                 pmgmt_type = RTE_POWER_MGMT_TYPE_SCALE;
1740                 return 0;
1741         }
1742         if (strncmp(PMD_MGMT_BASELINE, name, sizeof(PMD_MGMT_BASELINE)) == 0) {
1743                 baseline_enabled = true;
1744                 return 0;
1745         }
1746         /* unknown PMD power management mode */
1747         return -1;
1748 }
1749
1750 static int
1751 parse_ep_config(const char *q_arg)
1752 {
1753         char s[256];
1754         const char *p = q_arg;
1755         char *end;
1756         int  num_arg;
1757
1758         char *str_fld[3];
1759
1760         int training_flag;
1761         int med_edpi;
1762         int hgh_edpi;
1763
1764         ep_med_edpi = EMPTY_POLL_MED_THRESHOLD;
1765         ep_hgh_edpi = EMPTY_POLL_HGH_THRESHOLD;
1766
1767         strlcpy(s, p, sizeof(s));
1768
1769         num_arg = rte_strsplit(s, sizeof(s), str_fld, 3, ',');
1770
1771         empty_poll_train = false;
1772
1773         if (num_arg == 0)
1774                 return 0;
1775
1776         if (num_arg == 3) {
1777
1778                 training_flag = strtoul(str_fld[0], &end, 0);
1779                 med_edpi = strtoul(str_fld[1], &end, 0);
1780                 hgh_edpi = strtoul(str_fld[2], &end, 0);
1781
1782                 if (training_flag == 1)
1783                         empty_poll_train = true;
1784
1785                 if (med_edpi > 0)
1786                         ep_med_edpi = med_edpi;
1787
1788                 if (hgh_edpi > 0)
1789                         ep_hgh_edpi = hgh_edpi;
1790
1791         } else {
1792
1793                 return -1;
1794         }
1795
1796         return 0;
1797
1798 }
1799 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1800 #define CMD_LINE_OPT_LEGACY "legacy"
1801 #define CMD_LINE_OPT_EMPTY_POLL "empty-poll"
1802 #define CMD_LINE_OPT_INTERRUPT_ONLY "interrupt-only"
1803 #define CMD_LINE_OPT_TELEMETRY "telemetry"
1804 #define CMD_LINE_OPT_PMD_MGMT "pmd-mgmt"
1805 #define CMD_LINE_OPT_MAX_PKT_LEN "max-pkt-len"
1806
1807 /* Parse the argument given in the command line of the application */
1808 static int
1809 parse_args(int argc, char **argv)
1810 {
1811         int opt, ret;
1812         char **argvopt;
1813         int option_index;
1814         uint32_t limit;
1815         char *prgname = argv[0];
1816         static struct option lgopts[] = {
1817                 {"config", 1, 0, 0},
1818                 {"perf-config", 1, 0, 0},
1819                 {"high-perf-cores", 1, 0, 0},
1820                 {"no-numa", 0, 0, 0},
1821                 {CMD_LINE_OPT_MAX_PKT_LEN, 1, 0, 0},
1822                 {CMD_LINE_OPT_EMPTY_POLL, 1, 0, 0},
1823                 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
1824                 {CMD_LINE_OPT_LEGACY, 0, 0, 0},
1825                 {CMD_LINE_OPT_TELEMETRY, 0, 0, 0},
1826                 {CMD_LINE_OPT_INTERRUPT_ONLY, 0, 0, 0},
1827                 {CMD_LINE_OPT_PMD_MGMT, 1, 0, 0},
1828                 {NULL, 0, 0, 0}
1829         };
1830
1831         argvopt = argv;
1832
1833         while ((opt = getopt_long(argc, argvopt, "p:l:m:h:P",
1834                                 lgopts, &option_index)) != EOF) {
1835
1836                 switch (opt) {
1837                 /* portmask */
1838                 case 'p':
1839                         enabled_port_mask = parse_portmask(optarg);
1840                         if (enabled_port_mask == 0) {
1841                                 printf("invalid portmask\n");
1842                                 print_usage(prgname);
1843                                 return -1;
1844                         }
1845                         break;
1846                 case 'P':
1847                         printf("Promiscuous mode selected\n");
1848                         promiscuous_on = 1;
1849                         break;
1850                 case 'l':
1851                         limit = parse_max_pkt_len(optarg);
1852                         freq_tlb[LOW] = limit;
1853                         break;
1854                 case 'm':
1855                         limit = parse_max_pkt_len(optarg);
1856                         freq_tlb[MED] = limit;
1857                         break;
1858                 case 'h':
1859                         limit = parse_max_pkt_len(optarg);
1860                         freq_tlb[HGH] = limit;
1861                         break;
1862                 /* long options */
1863                 case 0:
1864                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1865                                 ret = parse_config(optarg);
1866                                 if (ret) {
1867                                         printf("invalid config\n");
1868                                         print_usage(prgname);
1869                                         return -1;
1870                                 }
1871                         }
1872
1873                         if (!strncmp(lgopts[option_index].name,
1874                                         "perf-config", 11)) {
1875                                 ret = parse_perf_config(optarg);
1876                                 if (ret) {
1877                                         printf("invalid perf-config\n");
1878                                         print_usage(prgname);
1879                                         return -1;
1880                                 }
1881                         }
1882
1883                         if (!strncmp(lgopts[option_index].name,
1884                                         "high-perf-cores", 15)) {
1885                                 ret = parse_perf_core_list(optarg);
1886                                 if (ret) {
1887                                         printf("invalid high-perf-cores\n");
1888                                         print_usage(prgname);
1889                                         return -1;
1890                                 }
1891                         }
1892
1893                         if (!strncmp(lgopts[option_index].name,
1894                                                 "no-numa", 7)) {
1895                                 printf("numa is disabled \n");
1896                                 numa_on = 0;
1897                         }
1898
1899                         if (!strncmp(lgopts[option_index].name,
1900                                         CMD_LINE_OPT_LEGACY,
1901                                         sizeof(CMD_LINE_OPT_LEGACY))) {
1902                                 if (app_mode != APP_MODE_DEFAULT) {
1903                                         printf(" legacy mode is mutually exclusive with other modes\n");
1904                                         return -1;
1905                                 }
1906                                 app_mode = APP_MODE_LEGACY;
1907                                 printf("legacy mode is enabled\n");
1908                         }
1909
1910                         if (!strncmp(lgopts[option_index].name,
1911                                         CMD_LINE_OPT_EMPTY_POLL, 10)) {
1912                                 if (app_mode != APP_MODE_DEFAULT) {
1913                                         printf(" empty-poll mode is mutually exclusive with other modes\n");
1914                                         return -1;
1915                                 }
1916                                 app_mode = APP_MODE_EMPTY_POLL;
1917                                 ret = parse_ep_config(optarg);
1918
1919                                 if (ret) {
1920                                         printf("invalid empty poll config\n");
1921                                         print_usage(prgname);
1922                                         return -1;
1923                                 }
1924                                 printf("empty-poll is enabled\n");
1925                         }
1926
1927                         if (!strncmp(lgopts[option_index].name,
1928                                         CMD_LINE_OPT_TELEMETRY,
1929                                         sizeof(CMD_LINE_OPT_TELEMETRY))) {
1930                                 if (app_mode != APP_MODE_DEFAULT) {
1931                                         printf(" telemetry mode is mutually exclusive with other modes\n");
1932                                         return -1;
1933                                 }
1934                                 app_mode = APP_MODE_TELEMETRY;
1935                                 printf("telemetry mode is enabled\n");
1936                         }
1937
1938                         if (!strncmp(lgopts[option_index].name,
1939                                         CMD_LINE_OPT_PMD_MGMT,
1940                                         sizeof(CMD_LINE_OPT_PMD_MGMT))) {
1941                                 if (app_mode != APP_MODE_DEFAULT) {
1942                                         printf(" power mgmt mode is mutually exclusive with other modes\n");
1943                                         return -1;
1944                                 }
1945                                 if (parse_pmd_mgmt_config(optarg) < 0) {
1946                                         printf(" Invalid PMD power management mode: %s\n",
1947                                                         optarg);
1948                                         return -1;
1949                                 }
1950                                 app_mode = APP_MODE_PMD_MGMT;
1951                                 printf("PMD power mgmt mode is enabled\n");
1952                         }
1953                         if (!strncmp(lgopts[option_index].name,
1954                                         CMD_LINE_OPT_INTERRUPT_ONLY,
1955                                         sizeof(CMD_LINE_OPT_INTERRUPT_ONLY))) {
1956                                 if (app_mode != APP_MODE_DEFAULT) {
1957                                         printf(" interrupt-only mode is mutually exclusive with other modes\n");
1958                                         return -1;
1959                                 }
1960                                 app_mode = APP_MODE_INTERRUPT;
1961                                 printf("interrupt-only mode is enabled\n");
1962                         }
1963
1964                         if (!strncmp(lgopts[option_index].name,
1965                                         CMD_LINE_OPT_MAX_PKT_LEN,
1966                                         sizeof(CMD_LINE_OPT_MAX_PKT_LEN))) {
1967                                 printf("Custom frame size is configured\n");
1968                                 max_pkt_len = parse_max_pkt_len(optarg);
1969                         }
1970
1971                         if (!strncmp(lgopts[option_index].name,
1972                                      CMD_LINE_OPT_PARSE_PTYPE,
1973                                      sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
1974                                 printf("soft parse-ptype is enabled\n");
1975                                 parse_ptype = 1;
1976                         }
1977
1978                         break;
1979
1980                 default:
1981                         print_usage(prgname);
1982                         return -1;
1983                 }
1984         }
1985
1986         if (optind >= 0)
1987                 argv[optind-1] = prgname;
1988
1989         ret = optind-1;
1990         optind = 1; /* reset getopt lib */
1991         return ret;
1992 }
1993
1994 static void
1995 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1996 {
1997         char buf[RTE_ETHER_ADDR_FMT_SIZE];
1998         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
1999         printf("%s%s", name, buf);
2000 }
2001
2002 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2003 static void
2004 setup_hash(int socketid)
2005 {
2006         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
2007                 .name = NULL,
2008                 .entries = L3FWD_HASH_ENTRIES,
2009                 .key_len = sizeof(struct ipv4_5tuple),
2010                 .hash_func = DEFAULT_HASH_FUNC,
2011                 .hash_func_init_val = 0,
2012         };
2013
2014         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
2015                 .name = NULL,
2016                 .entries = L3FWD_HASH_ENTRIES,
2017                 .key_len = sizeof(struct ipv6_5tuple),
2018                 .hash_func = DEFAULT_HASH_FUNC,
2019                 .hash_func_init_val = 0,
2020         };
2021
2022         unsigned i;
2023         int ret;
2024         char s[64];
2025
2026         /* create ipv4 hash */
2027         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
2028         ipv4_l3fwd_hash_params.name = s;
2029         ipv4_l3fwd_hash_params.socket_id = socketid;
2030         ipv4_l3fwd_lookup_struct[socketid] =
2031                 rte_hash_create(&ipv4_l3fwd_hash_params);
2032         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2033                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2034                                 "socket %d\n", socketid);
2035
2036         /* create ipv6 hash */
2037         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
2038         ipv6_l3fwd_hash_params.name = s;
2039         ipv6_l3fwd_hash_params.socket_id = socketid;
2040         ipv6_l3fwd_lookup_struct[socketid] =
2041                 rte_hash_create(&ipv6_l3fwd_hash_params);
2042         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
2043                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2044                                 "socket %d\n", socketid);
2045
2046
2047         /* populate the ipv4 hash */
2048         for (i = 0; i < RTE_DIM(ipv4_l3fwd_route_array); i++) {
2049                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
2050                                 (void *) &ipv4_l3fwd_route_array[i].key);
2051                 if (ret < 0) {
2052                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
2053                                 "l3fwd hash on socket %d\n", i, socketid);
2054                 }
2055                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
2056                 printf("Hash: Adding key\n");
2057                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
2058         }
2059
2060         /* populate the ipv6 hash */
2061         for (i = 0; i < RTE_DIM(ipv6_l3fwd_route_array); i++) {
2062                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
2063                                 (void *) &ipv6_l3fwd_route_array[i].key);
2064                 if (ret < 0) {
2065                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
2066                                 "l3fwd hash on socket %d\n", i, socketid);
2067                 }
2068                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
2069                 printf("Hash: Adding key\n");
2070                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
2071         }
2072 }
2073 #endif
2074
2075 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2076 static void
2077 setup_lpm(int socketid)
2078 {
2079         unsigned i;
2080         int ret;
2081         char s[64];
2082
2083         /* create the LPM table */
2084         struct rte_lpm_config lpm_ipv4_config;
2085
2086         lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
2087         lpm_ipv4_config.number_tbl8s = 256;
2088         lpm_ipv4_config.flags = 0;
2089
2090         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
2091         ipv4_l3fwd_lookup_struct[socketid] =
2092                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
2093         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2094                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
2095                                 " on socket %d\n", socketid);
2096
2097         /* populate the LPM table */
2098         for (i = 0; i < RTE_DIM(ipv4_l3fwd_route_array); i++) {
2099                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
2100                         ipv4_l3fwd_route_array[i].ip,
2101                         ipv4_l3fwd_route_array[i].depth,
2102                         ipv4_l3fwd_route_array[i].if_out);
2103
2104                 if (ret < 0) {
2105                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
2106                                 "l3fwd LPM table on socket %d\n",
2107                                 i, socketid);
2108                 }
2109
2110                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
2111                         (unsigned)ipv4_l3fwd_route_array[i].ip,
2112                         ipv4_l3fwd_route_array[i].depth,
2113                         ipv4_l3fwd_route_array[i].if_out);
2114         }
2115 }
2116 #endif
2117
2118 static int
2119 init_mem(unsigned nb_mbuf)
2120 {
2121         struct lcore_conf *qconf;
2122         int socketid;
2123         unsigned lcore_id;
2124         char s[64];
2125
2126         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2127                 if (rte_lcore_is_enabled(lcore_id) == 0)
2128                         continue;
2129
2130                 if (numa_on)
2131                         socketid = rte_lcore_to_socket_id(lcore_id);
2132                 else
2133                         socketid = 0;
2134
2135                 if (socketid >= NB_SOCKETS) {
2136                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
2137                                         "out of range %d\n", socketid,
2138                                                 lcore_id, NB_SOCKETS);
2139                 }
2140                 if (pktmbuf_pool[socketid] == NULL) {
2141                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
2142                         pktmbuf_pool[socketid] =
2143                                 rte_pktmbuf_pool_create(s, nb_mbuf,
2144                                         MEMPOOL_CACHE_SIZE, 0,
2145                                         RTE_MBUF_DEFAULT_BUF_SIZE,
2146                                         socketid);
2147                         if (pktmbuf_pool[socketid] == NULL)
2148                                 rte_exit(EXIT_FAILURE,
2149                                         "Cannot init mbuf pool on socket %d\n",
2150                                                                 socketid);
2151                         else
2152                                 printf("Allocated mbuf pool on socket %d\n",
2153                                                                 socketid);
2154
2155 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2156                         setup_lpm(socketid);
2157 #else
2158                         setup_hash(socketid);
2159 #endif
2160                 }
2161                 qconf = &lcore_conf[lcore_id];
2162                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
2163 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2164                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
2165 #endif
2166         }
2167         return 0;
2168 }
2169
2170 /* Check the link status of all ports in up to 9s, and print them finally */
2171 static void
2172 check_all_ports_link_status(uint32_t port_mask)
2173 {
2174 #define CHECK_INTERVAL 100 /* 100ms */
2175 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
2176         uint8_t count, all_ports_up, print_flag = 0;
2177         uint16_t portid;
2178         struct rte_eth_link link;
2179         int ret;
2180         char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
2181
2182         printf("\nChecking link status");
2183         fflush(stdout);
2184         for (count = 0; count <= MAX_CHECK_TIME; count++) {
2185                 all_ports_up = 1;
2186                 RTE_ETH_FOREACH_DEV(portid) {
2187                         if ((port_mask & (1 << portid)) == 0)
2188                                 continue;
2189                         memset(&link, 0, sizeof(link));
2190                         ret = rte_eth_link_get_nowait(portid, &link);
2191                         if (ret < 0) {
2192                                 all_ports_up = 0;
2193                                 if (print_flag == 1)
2194                                         printf("Port %u link get failed: %s\n",
2195                                                 portid, rte_strerror(-ret));
2196                                 continue;
2197                         }
2198                         /* print link status if flag set */
2199                         if (print_flag == 1) {
2200                                 rte_eth_link_to_str(link_status_text,
2201                                         sizeof(link_status_text), &link);
2202                                 printf("Port %d %s\n", portid,
2203                                        link_status_text);
2204                                 continue;
2205                         }
2206                         /* clear all_ports_up flag if any link down */
2207                         if (link.link_status == RTE_ETH_LINK_DOWN) {
2208                                 all_ports_up = 0;
2209                                 break;
2210                         }
2211                 }
2212                 /* after finally printing all link status, get out */
2213                 if (print_flag == 1)
2214                         break;
2215
2216                 if (all_ports_up == 0) {
2217                         printf(".");
2218                         fflush(stdout);
2219                         rte_delay_ms(CHECK_INTERVAL);
2220                 }
2221
2222                 /* set the print_flag if all ports up or timeout */
2223                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
2224                         print_flag = 1;
2225                         printf("done\n");
2226                 }
2227         }
2228 }
2229
2230 static int check_ptype(uint16_t portid)
2231 {
2232         int i, ret;
2233         int ptype_l3_ipv4 = 0;
2234 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2235         int ptype_l3_ipv6 = 0;
2236 #endif
2237         uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
2238
2239         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
2240         if (ret <= 0)
2241                 return 0;
2242
2243         uint32_t ptypes[ret];
2244
2245         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
2246         for (i = 0; i < ret; ++i) {
2247                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
2248                         ptype_l3_ipv4 = 1;
2249 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2250                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
2251                         ptype_l3_ipv6 = 1;
2252 #endif
2253         }
2254
2255         if (ptype_l3_ipv4 == 0)
2256                 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
2257
2258 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2259         if (ptype_l3_ipv6 == 0)
2260                 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
2261 #endif
2262
2263 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2264         if (ptype_l3_ipv4)
2265 #else /* APP_LOOKUP_EXACT_MATCH */
2266         if (ptype_l3_ipv4 && ptype_l3_ipv6)
2267 #endif
2268                 return 1;
2269
2270         return 0;
2271
2272 }
2273
2274 static int
2275 init_power_library(void)
2276 {
2277         enum power_management_env env;
2278         unsigned int lcore_id;
2279         int ret = 0;
2280
2281         RTE_LCORE_FOREACH(lcore_id) {
2282                 /* init power management library */
2283                 ret = rte_power_init(lcore_id);
2284                 if (ret) {
2285                         RTE_LOG(ERR, POWER,
2286                                 "Library initialization failed on core %u\n",
2287                                 lcore_id);
2288                         return ret;
2289                 }
2290                 /* we're not supporting the VM channel mode */
2291                 env = rte_power_get_env();
2292                 if (env != PM_ENV_ACPI_CPUFREQ &&
2293                                 env != PM_ENV_PSTATE_CPUFREQ) {
2294                         RTE_LOG(ERR, POWER,
2295                                 "Only ACPI and PSTATE mode are supported\n");
2296                         return -1;
2297                 }
2298         }
2299         return ret;
2300 }
2301
2302 static int
2303 deinit_power_library(void)
2304 {
2305         unsigned int lcore_id;
2306         int ret = 0;
2307
2308         RTE_LCORE_FOREACH(lcore_id) {
2309                 /* deinit power management library */
2310                 ret = rte_power_exit(lcore_id);
2311                 if (ret) {
2312                         RTE_LOG(ERR, POWER,
2313                                 "Library deinitialization failed on core %u\n",
2314                                 lcore_id);
2315                         return ret;
2316                 }
2317         }
2318         return ret;
2319 }
2320
2321 static void
2322 get_current_stat_values(uint64_t *values)
2323 {
2324         unsigned int lcore_id = rte_lcore_id();
2325         struct lcore_conf *qconf;
2326         uint64_t app_eps = 0, app_fps = 0, app_br = 0;
2327         uint64_t count = 0;
2328
2329         RTE_LCORE_FOREACH_WORKER(lcore_id) {
2330                 qconf = &lcore_conf[lcore_id];
2331                 if (qconf->n_rx_queue == 0)
2332                         continue;
2333                 count++;
2334                 rte_spinlock_lock(&stats[lcore_id].telemetry_lock);
2335                 app_eps += stats[lcore_id].ep_nep[1];
2336                 app_fps += stats[lcore_id].fp_nfp[1];
2337                 app_br += stats[lcore_id].br;
2338                 rte_spinlock_unlock(&stats[lcore_id].telemetry_lock);
2339         }
2340
2341         if (count > 0) {
2342                 values[0] = app_eps/count;
2343                 values[1] = app_fps/count;
2344                 values[2] = app_br/count;
2345         } else
2346                 memset(values, 0, sizeof(uint64_t) * NUM_TELSTATS);
2347
2348 }
2349
2350 static void
2351 update_telemetry(__rte_unused struct rte_timer *tim,
2352                 __rte_unused void *arg)
2353 {
2354         int ret;
2355         uint64_t values[NUM_TELSTATS] = {0};
2356
2357         get_current_stat_values(values);
2358         ret = rte_metrics_update_values(RTE_METRICS_GLOBAL, telstats_index,
2359                                         values, RTE_DIM(values));
2360         if (ret < 0)
2361                 RTE_LOG(WARNING, POWER, "failed to update metrcis\n");
2362 }
2363
2364 static int
2365 handle_app_stats(const char *cmd __rte_unused,
2366                 const char *params __rte_unused,
2367                 struct rte_tel_data *d)
2368 {
2369         uint64_t values[NUM_TELSTATS] = {0};
2370         uint32_t i;
2371
2372         rte_tel_data_start_dict(d);
2373         get_current_stat_values(values);
2374         for (i = 0; i < NUM_TELSTATS; i++)
2375                 rte_tel_data_add_dict_u64(d, telstats_strings[i].name,
2376                                 values[i]);
2377         return 0;
2378 }
2379
2380 static void
2381 telemetry_setup_timer(void)
2382 {
2383         int lcore_id = rte_lcore_id();
2384         uint64_t hz = rte_get_timer_hz();
2385         uint64_t ticks;
2386
2387         ticks = hz / TELEMETRY_INTERVALS_PER_SEC;
2388         rte_timer_reset_sync(&telemetry_timer,
2389                         ticks,
2390                         PERIODICAL,
2391                         lcore_id,
2392                         update_telemetry,
2393                         NULL);
2394 }
2395 static void
2396 empty_poll_setup_timer(void)
2397 {
2398         int lcore_id = rte_lcore_id();
2399         uint64_t hz = rte_get_timer_hz();
2400
2401         struct  ep_params *ep_ptr = ep_params;
2402
2403         ep_ptr->interval_ticks = hz / INTERVALS_PER_SECOND;
2404
2405         rte_timer_reset_sync(&ep_ptr->timer0,
2406                         ep_ptr->interval_ticks,
2407                         PERIODICAL,
2408                         lcore_id,
2409                         rte_empty_poll_detection,
2410                         (void *)ep_ptr);
2411
2412 }
2413 static int
2414 launch_timer(unsigned int lcore_id)
2415 {
2416         int64_t prev_tsc = 0, cur_tsc, diff_tsc, cycles_10ms;
2417
2418         RTE_SET_USED(lcore_id);
2419
2420
2421         if (rte_get_main_lcore() != lcore_id) {
2422                 rte_panic("timer on lcore:%d which is not main core:%d\n",
2423                                 lcore_id,
2424                                 rte_get_main_lcore());
2425         }
2426
2427         RTE_LOG(INFO, POWER, "Bring up the Timer\n");
2428
2429         if (app_mode == APP_MODE_EMPTY_POLL)
2430                 empty_poll_setup_timer();
2431         else
2432                 telemetry_setup_timer();
2433
2434         cycles_10ms = rte_get_timer_hz() / 100;
2435
2436         while (!is_done()) {
2437                 cur_tsc = rte_rdtsc();
2438                 diff_tsc = cur_tsc - prev_tsc;
2439                 if (diff_tsc > cycles_10ms) {
2440                         rte_timer_manage();
2441                         prev_tsc = cur_tsc;
2442                         cycles_10ms = rte_get_timer_hz() / 100;
2443                 }
2444         }
2445
2446         RTE_LOG(INFO, POWER, "Timer_subsystem is done\n");
2447
2448         return 0;
2449 }
2450
2451 static int
2452 autodetect_mode(void)
2453 {
2454         RTE_LOG(NOTICE, L3FWD_POWER, "Operating mode not specified, probing frequency scaling support...\n");
2455
2456         /*
2457          * Empty poll and telemetry modes have to be specifically requested to
2458          * be enabled, but we can auto-detect between interrupt mode with or
2459          * without frequency scaling. Both ACPI and pstate can be used.
2460          */
2461         if (rte_power_check_env_supported(PM_ENV_ACPI_CPUFREQ))
2462                 return APP_MODE_LEGACY;
2463         if (rte_power_check_env_supported(PM_ENV_PSTATE_CPUFREQ))
2464                 return APP_MODE_LEGACY;
2465
2466         RTE_LOG(NOTICE, L3FWD_POWER, "Frequency scaling not supported, selecting interrupt-only mode\n");
2467
2468         return APP_MODE_INTERRUPT;
2469 }
2470
2471 static const char *
2472 mode_to_str(enum appmode mode)
2473 {
2474         switch (mode) {
2475         case APP_MODE_LEGACY:
2476                 return "legacy";
2477         case APP_MODE_EMPTY_POLL:
2478                 return "empty poll";
2479         case APP_MODE_TELEMETRY:
2480                 return "telemetry";
2481         case APP_MODE_INTERRUPT:
2482                 return "interrupt-only";
2483         case APP_MODE_PMD_MGMT:
2484                 return "pmd mgmt";
2485         default:
2486                 return "invalid";
2487         }
2488 }
2489
2490 static uint32_t
2491 eth_dev_get_overhead_len(uint32_t max_rx_pktlen, uint16_t max_mtu)
2492 {
2493         uint32_t overhead_len;
2494
2495         if (max_mtu != UINT16_MAX && max_rx_pktlen > max_mtu)
2496                 overhead_len = max_rx_pktlen - max_mtu;
2497         else
2498                 overhead_len = RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN;
2499
2500         return overhead_len;
2501 }
2502
2503 static int
2504 config_port_max_pkt_len(struct rte_eth_conf *conf,
2505                 struct rte_eth_dev_info *dev_info)
2506 {
2507         uint32_t overhead_len;
2508
2509         if (max_pkt_len == 0)
2510                 return 0;
2511
2512         if (max_pkt_len < RTE_ETHER_MIN_LEN || max_pkt_len > MAX_JUMBO_PKT_LEN)
2513                 return -1;
2514
2515         overhead_len = eth_dev_get_overhead_len(dev_info->max_rx_pktlen,
2516                         dev_info->max_mtu);
2517         conf->rxmode.mtu = max_pkt_len - overhead_len;
2518
2519         if (conf->rxmode.mtu > RTE_ETHER_MTU)
2520                 conf->txmode.offloads |= RTE_ETH_TX_OFFLOAD_MULTI_SEGS;
2521
2522         return 0;
2523 }
2524
2525 /* Power library initialized in the main routine. 8< */
2526 int
2527 main(int argc, char **argv)
2528 {
2529         struct lcore_conf *qconf;
2530         struct rte_eth_dev_info dev_info;
2531         struct rte_eth_txconf *txconf;
2532         int ret;
2533         uint16_t nb_ports;
2534         uint16_t queueid;
2535         unsigned lcore_id;
2536         uint64_t hz;
2537         uint32_t n_tx_queue, nb_lcores;
2538         uint32_t dev_rxq_num, dev_txq_num;
2539         uint8_t nb_rx_queue, queue, socketid;
2540         uint16_t portid;
2541         const char *ptr_strings[NUM_TELSTATS];
2542
2543         /* init EAL */
2544         ret = rte_eal_init(argc, argv);
2545         if (ret < 0)
2546                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2547         argc -= ret;
2548         argv += ret;
2549
2550         /* catch SIGINT and restore cpufreq governor to ondemand */
2551         signal(SIGINT, signal_exit_now);
2552
2553         /* init RTE timer library to be used late */
2554         rte_timer_subsystem_init();
2555
2556         /* if we're running pmd-mgmt mode, don't default to baseline mode */
2557         baseline_enabled = false;
2558
2559         /* parse application arguments (after the EAL ones) */
2560         ret = parse_args(argc, argv);
2561         if (ret < 0)
2562                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
2563
2564         if (app_mode == APP_MODE_DEFAULT)
2565                 app_mode = autodetect_mode();
2566
2567         RTE_LOG(INFO, L3FWD_POWER, "Selected operation mode: %s\n",
2568                         mode_to_str(app_mode));
2569
2570         /* only legacy and empty poll mode rely on power library */
2571         if ((app_mode == APP_MODE_LEGACY || app_mode == APP_MODE_EMPTY_POLL) &&
2572                         init_power_library())
2573                 rte_exit(EXIT_FAILURE, "init_power_library failed\n");
2574
2575         if (update_lcore_params() < 0)
2576                 rte_exit(EXIT_FAILURE, "update_lcore_params failed\n");
2577
2578         if (check_lcore_params() < 0)
2579                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
2580
2581         ret = init_lcore_rx_queues();
2582         if (ret < 0)
2583                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2584
2585         nb_ports = rte_eth_dev_count_avail();
2586
2587         if (check_port_config() < 0)
2588                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
2589
2590         nb_lcores = rte_lcore_count();
2591
2592         /* initialize all ports */
2593         RTE_ETH_FOREACH_DEV(portid) {
2594                 struct rte_eth_conf local_port_conf = port_conf;
2595                 /* not all app modes need interrupts */
2596                 bool need_intr = app_mode == APP_MODE_LEGACY ||
2597                                 app_mode == APP_MODE_INTERRUPT;
2598
2599                 /* skip ports that are not enabled */
2600                 if ((enabled_port_mask & (1 << portid)) == 0) {
2601                         printf("\nSkipping disabled port %d\n", portid);
2602                         continue;
2603                 }
2604
2605                 /* init port */
2606                 printf("Initializing port %d ... ", portid );
2607                 fflush(stdout);
2608
2609                 ret = rte_eth_dev_info_get(portid, &dev_info);
2610                 if (ret != 0)
2611                         rte_exit(EXIT_FAILURE,
2612                                 "Error during getting device (port %u) info: %s\n",
2613                                 portid, strerror(-ret));
2614
2615                 dev_rxq_num = dev_info.max_rx_queues;
2616                 dev_txq_num = dev_info.max_tx_queues;
2617
2618                 nb_rx_queue = get_port_n_rx_queues(portid);
2619                 if (nb_rx_queue > dev_rxq_num)
2620                         rte_exit(EXIT_FAILURE,
2621                                 "Cannot configure not existed rxq: "
2622                                 "port=%d\n", portid);
2623
2624                 n_tx_queue = nb_lcores;
2625                 if (n_tx_queue > dev_txq_num)
2626                         n_tx_queue = dev_txq_num;
2627                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
2628                         nb_rx_queue, (unsigned)n_tx_queue );
2629                 /* If number of Rx queue is 0, no need to enable Rx interrupt */
2630                 if (nb_rx_queue == 0)
2631                         need_intr = false;
2632
2633                 if (need_intr)
2634                         local_port_conf.intr_conf.rxq = 1;
2635
2636                 ret = rte_eth_dev_info_get(portid, &dev_info);
2637                 if (ret != 0)
2638                         rte_exit(EXIT_FAILURE,
2639                                 "Error during getting device (port %u) info: %s\n",
2640                                 portid, strerror(-ret));
2641
2642                 ret = config_port_max_pkt_len(&local_port_conf, &dev_info);
2643                 if (ret != 0)
2644                         rte_exit(EXIT_FAILURE,
2645                                 "Invalid max packet length: %u (port %u)\n",
2646                                 max_pkt_len, portid);
2647
2648                 if (dev_info.tx_offload_capa & RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE)
2649                         local_port_conf.txmode.offloads |=
2650                                 RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE;
2651
2652                 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2653                         dev_info.flow_type_rss_offloads;
2654                 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2655                                 port_conf.rx_adv_conf.rss_conf.rss_hf) {
2656                         printf("Port %u modified RSS hash function based on hardware support,"
2657                                 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2658                                 portid,
2659                                 port_conf.rx_adv_conf.rss_conf.rss_hf,
2660                                 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2661                 }
2662
2663                 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf == 0)
2664                         local_port_conf.rxmode.mq_mode = RTE_ETH_MQ_RX_NONE;
2665                 local_port_conf.rxmode.offloads &= dev_info.rx_offload_capa;
2666                 port_conf.rxmode.offloads = local_port_conf.rxmode.offloads;
2667
2668                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
2669                                         (uint16_t)n_tx_queue, &local_port_conf);
2670                 if (ret < 0)
2671                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
2672                                         "err=%d, port=%d\n", ret, portid);
2673
2674                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
2675                                                        &nb_txd);
2676                 if (ret < 0)
2677                         rte_exit(EXIT_FAILURE,
2678                                  "Cannot adjust number of descriptors: err=%d, port=%d\n",
2679                                  ret, portid);
2680
2681                 ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
2682                 if (ret < 0)
2683                         rte_exit(EXIT_FAILURE,
2684                                  "Cannot get MAC address: err=%d, port=%d\n",
2685                                  ret, portid);
2686
2687                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
2688                 printf(", ");
2689
2690                 /* init memory */
2691                 ret = init_mem(NB_MBUF);
2692                 if (ret < 0)
2693                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
2694
2695                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2696                         if (rte_lcore_is_enabled(lcore_id) == 0)
2697                                 continue;
2698
2699                         /* Initialize TX buffers */
2700                         qconf = &lcore_conf[lcore_id];
2701                         qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
2702                                 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
2703                                 rte_eth_dev_socket_id(portid));
2704                         if (qconf->tx_buffer[portid] == NULL)
2705                                 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
2706                                                  portid);
2707
2708                         rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
2709                 }
2710
2711                 /* init one TX queue per couple (lcore,port) */
2712                 queueid = 0;
2713                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2714                         if (rte_lcore_is_enabled(lcore_id) == 0)
2715                                 continue;
2716
2717                         if (queueid >= dev_txq_num)
2718                                 continue;
2719
2720                         if (numa_on)
2721                                 socketid = \
2722                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
2723                         else
2724                                 socketid = 0;
2725
2726                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
2727                         fflush(stdout);
2728
2729                         txconf = &dev_info.default_txconf;
2730                         txconf->offloads = local_port_conf.txmode.offloads;
2731                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
2732                                                      socketid, txconf);
2733                         if (ret < 0)
2734                                 rte_exit(EXIT_FAILURE,
2735                                         "rte_eth_tx_queue_setup: err=%d, "
2736                                                 "port=%d\n", ret, portid);
2737
2738                         qconf = &lcore_conf[lcore_id];
2739                         qconf->tx_queue_id[portid] = queueid;
2740                         queueid++;
2741
2742                         qconf->tx_port_id[qconf->n_tx_port] = portid;
2743                         qconf->n_tx_port++;
2744                 }
2745                 printf("\n");
2746         }
2747
2748         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2749                 if (rte_lcore_is_enabled(lcore_id) == 0)
2750                         continue;
2751
2752                 if (app_mode == APP_MODE_LEGACY) {
2753                         /* init timer structures for each enabled lcore */
2754                         rte_timer_init(&power_timers[lcore_id]);
2755                         hz = rte_get_timer_hz();
2756                         rte_timer_reset(&power_timers[lcore_id],
2757                                         hz/TIMER_NUMBER_PER_SECOND,
2758                                         SINGLE, lcore_id,
2759                                         power_timer_cb, NULL);
2760                 }
2761                 qconf = &lcore_conf[lcore_id];
2762                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
2763                 fflush(stdout);
2764
2765                 /* init RX queues */
2766                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
2767                         struct rte_eth_rxconf rxq_conf;
2768
2769                         portid = qconf->rx_queue_list[queue].port_id;
2770                         queueid = qconf->rx_queue_list[queue].queue_id;
2771
2772                         if (numa_on)
2773                                 socketid = \
2774                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
2775                         else
2776                                 socketid = 0;
2777
2778                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
2779                         fflush(stdout);
2780
2781                         ret = rte_eth_dev_info_get(portid, &dev_info);
2782                         if (ret != 0)
2783                                 rte_exit(EXIT_FAILURE,
2784                                         "Error during getting device (port %u) info: %s\n",
2785                                         portid, strerror(-ret));
2786
2787                         rxq_conf = dev_info.default_rxconf;
2788                         rxq_conf.offloads = port_conf.rxmode.offloads;
2789                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
2790                                 socketid, &rxq_conf,
2791                                 pktmbuf_pool[socketid]);
2792                         if (ret < 0)
2793                                 rte_exit(EXIT_FAILURE,
2794                                         "rte_eth_rx_queue_setup: err=%d, "
2795                                                 "port=%d\n", ret, portid);
2796
2797                         if (parse_ptype) {
2798                                 if (add_cb_parse_ptype(portid, queueid) < 0)
2799                                         rte_exit(EXIT_FAILURE,
2800                                                  "Fail to add ptype cb\n");
2801                         }
2802
2803                         if (app_mode == APP_MODE_PMD_MGMT && !baseline_enabled) {
2804                                 ret = rte_power_ethdev_pmgmt_queue_enable(
2805                                                 lcore_id, portid, queueid,
2806                                                 pmgmt_type);
2807                                 if (ret < 0)
2808                                         rte_exit(EXIT_FAILURE,
2809                                                 "rte_power_ethdev_pmgmt_queue_enable: err=%d, port=%d\n",
2810                                                         ret, portid);
2811                         }
2812                 }
2813         }
2814         /* >8 End of power library initialization. */
2815
2816         printf("\n");
2817
2818         /* start ports */
2819         RTE_ETH_FOREACH_DEV(portid) {
2820                 if ((enabled_port_mask & (1 << portid)) == 0) {
2821                         continue;
2822                 }
2823                 /* Start device */
2824                 ret = rte_eth_dev_start(portid);
2825                 if (ret < 0)
2826                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
2827                                                 "port=%d\n", ret, portid);
2828                 /*
2829                  * If enabled, put device in promiscuous mode.
2830                  * This allows IO forwarding mode to forward packets
2831                  * to itself through 2 cross-connected  ports of the
2832                  * target machine.
2833                  */
2834                 if (promiscuous_on) {
2835                         ret = rte_eth_promiscuous_enable(portid);
2836                         if (ret != 0)
2837                                 rte_exit(EXIT_FAILURE,
2838                                         "rte_eth_promiscuous_enable: err=%s, port=%u\n",
2839                                         rte_strerror(-ret), portid);
2840                 }
2841                 /* initialize spinlock for each port */
2842                 rte_spinlock_init(&(locks[portid]));
2843
2844                 if (!parse_ptype)
2845                         if (!check_ptype(portid))
2846                                 rte_exit(EXIT_FAILURE,
2847                                         "PMD can not provide needed ptypes\n");
2848         }
2849
2850         check_all_ports_link_status(enabled_port_mask);
2851
2852         if (app_mode == APP_MODE_EMPTY_POLL) {
2853
2854                 if (empty_poll_train) {
2855                         policy.state = TRAINING;
2856                 } else {
2857                         policy.state = MED_NORMAL;
2858                         policy.med_base_edpi = ep_med_edpi;
2859                         policy.hgh_base_edpi = ep_hgh_edpi;
2860                 }
2861
2862                 ret = rte_power_empty_poll_stat_init(&ep_params,
2863                                 freq_tlb,
2864                                 &policy);
2865                 if (ret < 0)
2866                         rte_exit(EXIT_FAILURE, "empty poll init failed");
2867         }
2868
2869
2870         /* launch per-lcore init on every lcore */
2871         if (app_mode == APP_MODE_LEGACY) {
2872                 rte_eal_mp_remote_launch(main_legacy_loop, NULL, CALL_MAIN);
2873         } else if (app_mode == APP_MODE_EMPTY_POLL) {
2874                 empty_poll_stop = false;
2875                 rte_eal_mp_remote_launch(main_empty_poll_loop, NULL,
2876                                 SKIP_MAIN);
2877         } else if (app_mode == APP_MODE_TELEMETRY) {
2878                 unsigned int i;
2879
2880                 /* Init metrics library */
2881                 rte_metrics_init(rte_socket_id());
2882                 /** Register stats with metrics library */
2883                 for (i = 0; i < NUM_TELSTATS; i++)
2884                         ptr_strings[i] = telstats_strings[i].name;
2885
2886                 ret = rte_metrics_reg_names(ptr_strings, NUM_TELSTATS);
2887                 if (ret >= 0)
2888                         telstats_index = ret;
2889                 else
2890                         rte_exit(EXIT_FAILURE, "failed to register metrics names");
2891
2892                 RTE_LCORE_FOREACH_WORKER(lcore_id) {
2893                         rte_spinlock_init(&stats[lcore_id].telemetry_lock);
2894                 }
2895                 rte_timer_init(&telemetry_timer);
2896                 rte_telemetry_register_cmd("/l3fwd-power/stats",
2897                                 handle_app_stats,
2898                                 "Returns global power stats. Parameters: None");
2899                 rte_eal_mp_remote_launch(main_telemetry_loop, NULL,
2900                                                 SKIP_MAIN);
2901         } else if (app_mode == APP_MODE_INTERRUPT) {
2902                 rte_eal_mp_remote_launch(main_intr_loop, NULL, CALL_MAIN);
2903         } else if (app_mode == APP_MODE_PMD_MGMT) {
2904                 /* reuse telemetry loop for PMD power management mode */
2905                 rte_eal_mp_remote_launch(main_telemetry_loop, NULL, CALL_MAIN);
2906         }
2907
2908         if (app_mode == APP_MODE_EMPTY_POLL || app_mode == APP_MODE_TELEMETRY)
2909                 launch_timer(rte_lcore_id());
2910
2911         RTE_LCORE_FOREACH_WORKER(lcore_id) {
2912                 if (rte_eal_wait_lcore(lcore_id) < 0)
2913                         return -1;
2914         }
2915
2916         if (app_mode == APP_MODE_PMD_MGMT) {
2917                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2918                         if (rte_lcore_is_enabled(lcore_id) == 0)
2919                                 continue;
2920                         qconf = &lcore_conf[lcore_id];
2921                         for (queue = 0; queue < qconf->n_rx_queue; ++queue) {
2922                                 portid = qconf->rx_queue_list[queue].port_id;
2923                                 queueid = qconf->rx_queue_list[queue].queue_id;
2924
2925                                 rte_power_ethdev_pmgmt_queue_disable(lcore_id,
2926                                                 portid, queueid);
2927                         }
2928                 }
2929         }
2930
2931         RTE_ETH_FOREACH_DEV(portid)
2932         {
2933                 if ((enabled_port_mask & (1 << portid)) == 0)
2934                         continue;
2935
2936                 ret = rte_eth_dev_stop(portid);
2937                 if (ret != 0)
2938                         RTE_LOG(ERR, L3FWD_POWER, "rte_eth_dev_stop: err=%d, port=%u\n",
2939                                 ret, portid);
2940
2941                 rte_eth_dev_close(portid);
2942         }
2943
2944         if (app_mode == APP_MODE_EMPTY_POLL)
2945                 rte_power_empty_poll_stat_free();
2946
2947         if ((app_mode == APP_MODE_LEGACY || app_mode == APP_MODE_EMPTY_POLL) &&
2948                         deinit_power_library())
2949                 rte_exit(EXIT_FAILURE, "deinit_power_library failed\n");
2950
2951         if (rte_eal_cleanup() < 0)
2952                 RTE_LOG(ERR, L3FWD_POWER, "EAL cleanup failed\n");
2953
2954         return 0;
2955 }