examples/l3fwd-power/main.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2018 Intel Corporation
   3  */
   4
   5 #include <stdio.h>
   6 #include <stdlib.h>
   7 #include <stdint.h>
   8 #include <inttypes.h>
   9 #include <sys/types.h>
  10 #include <string.h>
  11 #include <sys/queue.h>
  12 #include <stdarg.h>
  13 #include <errno.h>
  14 #include <getopt.h>
  15 #include <unistd.h>
  16 #include <signal.h>
  17 #include <math.h>
  18
  19 #include <rte_common.h>
  20 #include <rte_byteorder.h>
  21 #include <rte_log.h>
  22 #include <rte_malloc.h>
  23 #include <rte_memory.h>
  24 #include <rte_memcpy.h>
  25 #include <rte_eal.h>
  26 #include <rte_launch.h>
  27 #include <rte_atomic.h>
  28 #include <rte_cycles.h>
  29 #include <rte_prefetch.h>
  30 #include <rte_lcore.h>
  31 #include <rte_per_lcore.h>
  32 #include <rte_branch_prediction.h>
  33 #include <rte_interrupts.h>
  34 #include <rte_random.h>
  35 #include <rte_debug.h>
  36 #include <rte_ether.h>
  37 #include <rte_ethdev.h>
  38 #include <rte_mempool.h>
  39 #include <rte_mbuf.h>
  40 #include <rte_ip.h>
  41 #include <rte_tcp.h>
  42 #include <rte_udp.h>
  43 #include <rte_string_fns.h>
  44 #include <rte_timer.h>
  45 #include <rte_power.h>
  46 #include <rte_spinlock.h>
  47 #include <rte_power_empty_poll.h>
  48 #include <rte_metrics.h>
  49 #include <rte_telemetry.h>
  50 #include <rte_power_pmd_mgmt.h>
  51
  52 #include "perf_core.h"
  53 #include "main.h"
  54
  55 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
  56
  57 #define MAX_PKT_BURST 32
  58
  59 #define MIN_ZERO_POLL_COUNT 10
  60
  61 /* 100 ms interval */
  62 #define TIMER_NUMBER_PER_SECOND           10
  63 /* (10ms) */
  64 #define INTERVALS_PER_SECOND             100
  65 /* 100000 us */
  66 #define SCALING_PERIOD                    (1000000/TIMER_NUMBER_PER_SECOND)
  67 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
  68
  69 #define APP_LOOKUP_EXACT_MATCH          0
  70 #define APP_LOOKUP_LPM                  1
  71 #define DO_RFC_1812_CHECKS
  72
  73 #ifndef APP_LOOKUP_METHOD
  74 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  75 #endif
  76
  77 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  78 #include <rte_hash.h>
  79 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
  80 #include <rte_lpm.h>
  81 #else
  82 #error "APP_LOOKUP_METHOD set to incorrect value"
  83 #endif
  84
  85 #ifndef IPv6_BYTES
  86 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
  87                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
  88 #define IPv6_BYTES(addr) \
  89         addr[0],  addr[1], addr[2],  addr[3], \
  90         addr[4],  addr[5], addr[6],  addr[7], \
  91         addr[8],  addr[9], addr[10], addr[11],\
  92         addr[12], addr[13],addr[14], addr[15]
  93 #endif
  94
  95 #define MAX_JUMBO_PKT_LEN  9600
  96
  97 #define IPV6_ADDR_LEN 16
  98
  99 #define MEMPOOL_CACHE_SIZE 256
 100
 101 /*
 102  * This expression is used to calculate the number of mbufs needed depending on
 103  * user input, taking into account memory for rx and tx hardware rings, cache
 104  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 105  * NB_MBUF never goes below a minimum value of 8192.
 106  */
 107
 108 #define NB_MBUF RTE_MAX ( \
 109         (nb_ports*nb_rx_queue*nb_rxd + \
 110         nb_ports*nb_lcores*MAX_PKT_BURST + \
 111         nb_ports*n_tx_queue*nb_txd + \
 112         nb_lcores*MEMPOOL_CACHE_SIZE), \
 113         (unsigned)8192)
 114
 115 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 116
 117 #define NB_SOCKETS 8
 118
 119 /* Configure how many packets ahead to prefetch, when reading packets */
 120 #define PREFETCH_OFFSET 3
 121
 122 /*
 123  * Configurable number of RX/TX ring descriptors
 124  */
 125 #define RTE_TEST_RX_DESC_DEFAULT 1024
 126 #define RTE_TEST_TX_DESC_DEFAULT 1024
 127
 128 /*
 129  * These two thresholds were decided on by running the training algorithm on
 130  * a 2.5GHz Xeon. These defaults can be overridden by supplying non-zero values
 131  * for the med_threshold and high_threshold parameters on the command line.
 132  */
 133 #define EMPTY_POLL_MED_THRESHOLD 350000UL
 134 #define EMPTY_POLL_HGH_THRESHOLD 580000UL
 135
 136 #define NUM_TELSTATS RTE_DIM(telstats_strings)
 137
 138 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 139 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 140
 141 /* ethernet addresses of ports */
 142 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 143
 144 /* ethernet addresses of ports */
 145 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
 146
 147 /* mask of enabled ports */
 148 static uint32_t enabled_port_mask = 0;
 149 /* Ports set in promiscuous mode off by default. */
 150 static int promiscuous_on = 0;
 151 /* NUMA is enabled by default. */
 152 static int numa_on = 1;
 153 static bool empty_poll_stop;
 154 static bool empty_poll_train;
 155 volatile bool quit_signal;
 156 static struct  ep_params *ep_params;
 157 static struct  ep_policy policy;
 158 static long  ep_med_edpi, ep_hgh_edpi;
 159 /* timer to update telemetry every 500ms */
 160 static struct rte_timer telemetry_timer;
 161
 162 /* stats index returned by metrics lib */
 163 int telstats_index;
 164
 165 struct telstats_name {
 166         char name[RTE_ETH_XSTATS_NAME_SIZE];
 167 };
 168
 169 /* telemetry stats to be reported */
 170 const struct telstats_name telstats_strings[] = {
 171         {"empty_poll"},
 172         {"full_poll"},
 173         {"busy_percent"}
 174 };
 175
 176 /* core busyness in percentage */
 177 enum busy_rate {
 178         ZERO = 0,
 179         PARTIAL = 50,
 180         FULL = 100
 181 };
 182
 183 /* reference poll count to measure core busyness */
 184 #define DEFAULT_COUNT 10000
 185 /*
 186  * reference CYCLES to be used to
 187  * measure core busyness based on poll count
 188  */
 189 #define MIN_CYCLES  1500000ULL
 190 #define MAX_CYCLES 22000000ULL
 191
 192 /* (500ms) */
 193 #define TELEMETRY_INTERVALS_PER_SEC 2
 194
 195 static int parse_ptype; /**< Parse packet type using rx callback, and */
 196                         /**< disabled by default */
 197
 198 enum appmode {
 199         APP_MODE_DEFAULT = 0,
 200         APP_MODE_LEGACY,
 201         APP_MODE_EMPTY_POLL,
 202         APP_MODE_TELEMETRY,
 203         APP_MODE_INTERRUPT,
 204         APP_MODE_PMD_MGMT
 205 };
 206
 207 enum appmode app_mode;
 208
 209 static enum rte_power_pmd_mgmt_type pmgmt_type;
 210 bool baseline_enabled;
 211
 212 enum freq_scale_hint_t
 213 {
 214         FREQ_LOWER    =      -1,
 215         FREQ_CURRENT  =       0,
 216         FREQ_HIGHER   =       1,
 217         FREQ_HIGHEST  =       2
 218 };
 219
 220 struct lcore_rx_queue {
 221         uint16_t port_id;
 222         uint8_t queue_id;
 223         enum freq_scale_hint_t freq_up_hint;
 224         uint32_t zero_rx_packet_count;
 225         uint32_t idle_hint;
 226 } __rte_cache_aligned;
 227
 228 #define MAX_RX_QUEUE_PER_LCORE 16
 229 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 230 #define MAX_RX_QUEUE_PER_PORT 128
 231
 232 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
 233
 234
 235 struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 236 static struct lcore_params lcore_params_array_default[] = {
 237         {0, 0, 2},
 238         {0, 1, 2},
 239         {0, 2, 2},
 240         {1, 0, 2},
 241         {1, 1, 2},
 242         {1, 2, 2},
 243         {2, 0, 2},
 244         {3, 0, 3},
 245         {3, 1, 3},
 246 };
 247
 248 struct lcore_params *lcore_params = lcore_params_array_default;
 249 uint16_t nb_lcore_params = RTE_DIM(lcore_params_array_default);
 250
 251 static struct rte_eth_conf port_conf = {
 252         .rxmode = {
 253                 .mq_mode        = ETH_MQ_RX_RSS,
 254                 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
 255                 .split_hdr_size = 0,
 256                 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
 257         },
 258         .rx_adv_conf = {
 259                 .rss_conf = {
 260                         .rss_key = NULL,
 261                         .rss_hf = ETH_RSS_UDP,
 262                 },
 263         },
 264         .txmode = {
 265                 .mq_mode = ETH_MQ_TX_NONE,
 266         }
 267 };
 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         /* this is checked in H/W */
 509
 510         /*
 511          * 3. The IP version number must be 4. If the version number is not 4
 512          * then the packet may be another version of IP, such as IPng or
 513          * ST-II.
 514          */
 515         if (((pkt->version_ihl) >> 4) != 4)
 516                 return -3;
 517         /*
 518          * 4. The IP header length field must be large enough to hold the
 519          * minimum length legal IP datagram (20 bytes = 5 words).
 520          */
 521         if ((pkt->version_ihl & 0xf) < 5)
 522                 return -4;
 523
 524         /*
 525          * 5. The IP total length field must be large enough to hold the IP
 526          * datagram header, whose length is specified in the IP header length
 527          * field.
 528          */
 529         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct rte_ipv4_hdr))
 530                 return -5;
 531
 532         return 0;
 533 }
 534 #endif
 535
 536 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 537 static void
 538 print_ipv4_key(struct ipv4_5tuple key)
 539 {
 540         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
 541                 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
 542                                 key.port_dst, key.port_src, key.proto);
 543 }
 544 static void
 545 print_ipv6_key(struct ipv6_5tuple key)
 546 {
 547         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 548                 "port dst = %d, port src = %d, proto = %d\n",
 549                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 550                 key.port_dst, key.port_src, key.proto);
 551 }
 552
 553 static inline uint16_t
 554 get_ipv4_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
 555                 lookup_struct_t * ipv4_l3fwd_lookup_struct)
 556 {
 557         struct ipv4_5tuple key;
 558         struct rte_tcp_hdr *tcp;
 559         struct rte_udp_hdr *udp;
 560         int ret = 0;
 561
 562         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 563         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 564         key.proto = ipv4_hdr->next_proto_id;
 565
 566         switch (ipv4_hdr->next_proto_id) {
 567         case IPPROTO_TCP:
 568                 tcp = (struct rte_tcp_hdr *)((unsigned char *)ipv4_hdr +
 569                                         sizeof(struct rte_ipv4_hdr));
 570                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 571                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 572                 break;
 573
 574         case IPPROTO_UDP:
 575                 udp = (struct rte_udp_hdr *)((unsigned char *)ipv4_hdr +
 576                                         sizeof(struct rte_ipv4_hdr));
 577                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 578                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 579                 break;
 580
 581         default:
 582                 key.port_dst = 0;
 583                 key.port_src = 0;
 584                 break;
 585         }
 586
 587         /* Find destination port */
 588         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 589         return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
 590 }
 591
 592 static inline uint16_t
 593 get_ipv6_dst_port(struct rte_ipv6_hdr *ipv6_hdr, uint16_t portid,
 594                         lookup_struct_t *ipv6_l3fwd_lookup_struct)
 595 {
 596         struct ipv6_5tuple key;
 597         struct rte_tcp_hdr *tcp;
 598         struct rte_udp_hdr *udp;
 599         int ret = 0;
 600
 601         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 602         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 603
 604         key.proto = ipv6_hdr->proto;
 605
 606         switch (ipv6_hdr->proto) {
 607         case IPPROTO_TCP:
 608                 tcp = (struct rte_tcp_hdr *)((unsigned char *) ipv6_hdr +
 609                                         sizeof(struct rte_ipv6_hdr));
 610                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 611                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 612                 break;
 613
 614         case IPPROTO_UDP:
 615                 udp = (struct rte_udp_hdr *)((unsigned char *) ipv6_hdr +
 616                                         sizeof(struct rte_ipv6_hdr));
 617                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 618                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 619                 break;
 620
 621         default:
 622                 key.port_dst = 0;
 623                 key.port_src = 0;
 624                 break;
 625         }
 626
 627         /* Find destination port */
 628         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 629         return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
 630 }
 631 #endif
 632
 633 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 634 static inline uint16_t
 635 get_ipv4_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
 636                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 637 {
 638         uint32_t next_hop;
 639
 640         return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 641                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 642                         next_hop : portid);
 643 }
 644 #endif
 645
 646 static inline void
 647 parse_ptype_one(struct rte_mbuf *m)
 648 {
 649         struct rte_ether_hdr *eth_hdr;
 650         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 651         uint16_t ether_type;
 652
 653         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 654         ether_type = eth_hdr->ether_type;
 655         if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
 656                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 657         else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6))
 658                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 659
 660         m->packet_type = packet_type;
 661 }
 662
 663 static uint16_t
 664 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
 665                struct rte_mbuf *pkts[], uint16_t nb_pkts,
 666                uint16_t max_pkts __rte_unused,
 667                void *user_param __rte_unused)
 668 {
 669         unsigned int i;
 670
 671         for (i = 0; i < nb_pkts; ++i)
 672                 parse_ptype_one(pkts[i]);
 673
 674         return nb_pkts;
 675 }
 676
 677 static int
 678 add_cb_parse_ptype(uint16_t portid, uint16_t queueid)
 679 {
 680         printf("Port %d: softly parse packet type info\n", portid);
 681         if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL))
 682                 return 0;
 683
 684         printf("Failed to add rx callback: port=%d\n", portid);
 685         return -1;
 686 }
 687
 688 static inline void
 689 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
 690                                 struct lcore_conf *qconf)
 691 {
 692         struct rte_ether_hdr *eth_hdr;
 693         struct rte_ipv4_hdr *ipv4_hdr;
 694         void *d_addr_bytes;
 695         uint16_t dst_port;
 696
 697         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 698
 699         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 700                 /* Handle IPv4 headers.*/
 701                 ipv4_hdr =
 702                         rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
 703                                                 sizeof(struct rte_ether_hdr));
 704
 705 #ifdef DO_RFC_1812_CHECKS
 706                 /* Check to make sure the packet is valid (RFC1812) */
 707                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 708                         rte_pktmbuf_free(m);
 709                         return;
 710                 }
 711 #endif
 712
 713                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 714                                         qconf->ipv4_lookup_struct);
 715                 if (dst_port >= RTE_MAX_ETHPORTS ||
 716                                 (enabled_port_mask & 1 << dst_port) == 0)
 717                         dst_port = portid;
 718
 719                 /* 02:00:00:00:00:xx */
 720                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 721                 *((uint64_t *)d_addr_bytes) =
 722                         0x000000000002 + ((uint64_t)dst_port << 40);
 723
 724 #ifdef DO_RFC_1812_CHECKS
 725                 /* Update time to live and header checksum */
 726                 --(ipv4_hdr->time_to_live);
 727                 ++(ipv4_hdr->hdr_checksum);
 728 #endif
 729
 730                 /* src addr */
 731                 rte_ether_addr_copy(&ports_eth_addr[dst_port],
 732                                 &eth_hdr->s_addr);
 733
 734                 send_single_packet(m, dst_port);
 735         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 736                 /* Handle IPv6 headers.*/
 737 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 738                 struct rte_ipv6_hdr *ipv6_hdr;
 739
 740                 ipv6_hdr =
 741                         rte_pktmbuf_mtod_offset(m, struct rte_ipv6_hdr *,
 742                                                 sizeof(struct rte_ether_hdr));
 743
 744                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
 745                                         qconf->ipv6_lookup_struct);
 746
 747                 if (dst_port >= RTE_MAX_ETHPORTS ||
 748                                 (enabled_port_mask & 1 << dst_port) == 0)
 749                         dst_port = portid;
 750
 751                 /* 02:00:00:00:00:xx */
 752                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 753                 *((uint64_t *)d_addr_bytes) =
 754                         0x000000000002 + ((uint64_t)dst_port << 40);
 755
 756                 /* src addr */
 757                 rte_ether_addr_copy(&ports_eth_addr[dst_port],
 758                                 &eth_hdr->s_addr);
 759
 760                 send_single_packet(m, dst_port);
 761 #else
 762                 /* We don't currently handle IPv6 packets in LPM mode. */
 763                 rte_pktmbuf_free(m);
 764 #endif
 765         } else
 766                 rte_pktmbuf_free(m);
 767
 768 }
 769
 770 #define MINIMUM_SLEEP_TIME         1
 771 #define SUSPEND_THRESHOLD          300
 772
 773 static inline uint32_t
 774 power_idle_heuristic(uint32_t zero_rx_packet_count)
 775 {
 776         /* If zero count is less than 100,  sleep 1us */
 777         if (zero_rx_packet_count < SUSPEND_THRESHOLD)
 778                 return MINIMUM_SLEEP_TIME;
 779         /* If zero count is less than 1000, sleep 100 us which is the
 780                 minimum latency switching from C3/C6 to C0
 781         */
 782         else
 783                 return SUSPEND_THRESHOLD;
 784 }
 785
 786 static inline enum freq_scale_hint_t
 787 power_freq_scaleup_heuristic(unsigned lcore_id,
 788                              uint16_t port_id,
 789                              uint16_t queue_id)
 790 {
 791         uint32_t rxq_count = rte_eth_rx_queue_count(port_id, queue_id);
 792 /**
 793  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
 794  * per iteration
 795  */
 796 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
 797 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
 798 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
 799 #define FREQ_UP_TREND1_ACC   1
 800 #define FREQ_UP_TREND2_ACC   100
 801 #define FREQ_UP_THRESHOLD    10000
 802
 803         if (likely(rxq_count > FREQ_GEAR3_RX_PACKET_THRESHOLD)) {
 804                 stats[lcore_id].trend = 0;
 805                 return FREQ_HIGHEST;
 806         } else if (likely(rxq_count > FREQ_GEAR2_RX_PACKET_THRESHOLD))
 807                 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
 808         else if (likely(rxq_count > FREQ_GEAR1_RX_PACKET_THRESHOLD))
 809                 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
 810
 811         if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
 812                 stats[lcore_id].trend = 0;
 813                 return FREQ_HIGHER;
 814         }
 815
 816         return FREQ_CURRENT;
 817 }
 818
 819 /**
 820  * force polling thread sleep until one-shot rx interrupt triggers
 821  * @param port_id
 822  *  Port id.
 823  * @param queue_id
 824  *  Rx queue id.
 825  * @return
 826  *  0 on success
 827  */
 828 static int
 829 sleep_until_rx_interrupt(int num, int lcore)
 830 {
 831         /*
 832          * we want to track when we are woken up by traffic so that we can go
 833          * back to sleep again without log spamming. Avoid cache line sharing
 834          * to prevent threads stepping on each others' toes.
 835          */
 836         static struct {
 837                 bool wakeup;
 838         } __rte_cache_aligned status[RTE_MAX_LCORE];
 839         struct rte_epoll_event event[num];
 840         int n, i;
 841         uint16_t port_id;
 842         uint8_t queue_id;
 843         void *data;
 844
 845         if (status[lcore].wakeup) {
 846                 RTE_LOG(INFO, L3FWD_POWER,
 847                                 "lcore %u sleeps until interrupt triggers\n",
 848                                 rte_lcore_id());
 849         }
 850
 851         n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, 10);
 852         for (i = 0; i < n; i++) {
 853                 data = event[i].epdata.data;
 854                 port_id = ((uintptr_t)data) >> CHAR_BIT;
 855                 queue_id = ((uintptr_t)data) &
 856                         RTE_LEN2MASK(CHAR_BIT, uint8_t);
 857                 RTE_LOG(INFO, L3FWD_POWER,
 858                         "lcore %u is waked up from rx interrupt on"
 859                         " port %d queue %d\n",
 860                         rte_lcore_id(), port_id, queue_id);
 861         }
 862         status[lcore].wakeup = n != 0;
 863
 864         return 0;
 865 }
 866
 867 static void turn_on_off_intr(struct lcore_conf *qconf, bool on)
 868 {
 869         int i;
 870         struct lcore_rx_queue *rx_queue;
 871         uint8_t queue_id;
 872         uint16_t port_id;
 873
 874         for (i = 0; i < qconf->n_rx_queue; ++i) {
 875                 rx_queue = &(qconf->rx_queue_list[i]);
 876                 port_id = rx_queue->port_id;
 877                 queue_id = rx_queue->queue_id;
 878
 879                 rte_spinlock_lock(&(locks[port_id]));
 880                 if (on)
 881                         rte_eth_dev_rx_intr_enable(port_id, queue_id);
 882                 else
 883                         rte_eth_dev_rx_intr_disable(port_id, queue_id);
 884                 rte_spinlock_unlock(&(locks[port_id]));
 885         }
 886 }
 887
 888 static int event_register(struct lcore_conf *qconf)
 889 {
 890         struct lcore_rx_queue *rx_queue;
 891         uint8_t queueid;
 892         uint16_t portid;
 893         uint32_t data;
 894         int ret;
 895         int i;
 896
 897         for (i = 0; i < qconf->n_rx_queue; ++i) {
 898                 rx_queue = &(qconf->rx_queue_list[i]);
 899                 portid = rx_queue->port_id;
 900                 queueid = rx_queue->queue_id;
 901                 data = portid << CHAR_BIT | queueid;
 902
 903                 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
 904                                                 RTE_EPOLL_PER_THREAD,
 905                                                 RTE_INTR_EVENT_ADD,
 906                                                 (void *)((uintptr_t)data));
 907                 if (ret)
 908                         return ret;
 909         }
 910
 911         return 0;
 912 }
 913
 914 /* main processing loop */
 915 static int main_intr_loop(__rte_unused void *dummy)
 916 {
 917         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 918         unsigned int lcore_id;
 919         uint64_t prev_tsc, diff_tsc, cur_tsc;
 920         int i, j, nb_rx;
 921         uint8_t queueid;
 922         uint16_t portid;
 923         struct lcore_conf *qconf;
 924         struct lcore_rx_queue *rx_queue;
 925         uint32_t lcore_rx_idle_count = 0;
 926         uint32_t lcore_idle_hint = 0;
 927         int intr_en = 0;
 928
 929         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
 930                                    US_PER_S * BURST_TX_DRAIN_US;
 931
 932         prev_tsc = 0;
 933
 934         lcore_id = rte_lcore_id();
 935         qconf = &lcore_conf[lcore_id];
 936
 937         if (qconf->n_rx_queue == 0) {
 938                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
 939                                 lcore_id);
 940                 return 0;
 941         }
 942
 943         RTE_LOG(INFO, L3FWD_POWER, "entering main interrupt loop on lcore %u\n",
 944                         lcore_id);
 945
 946         for (i = 0; i < qconf->n_rx_queue; i++) {
 947                 portid = qconf->rx_queue_list[i].port_id;
 948                 queueid = qconf->rx_queue_list[i].queue_id;
 949                 RTE_LOG(INFO, L3FWD_POWER,
 950                                 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
 951                                 lcore_id, portid, queueid);
 952         }
 953
 954         /* add into event wait list */
 955         if (event_register(qconf) == 0)
 956                 intr_en = 1;
 957         else
 958                 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
 959
 960         while (!is_done()) {
 961                 stats[lcore_id].nb_iteration_looped++;
 962
 963                 cur_tsc = rte_rdtsc();
 964
 965                 /*
 966                  * TX burst queue drain
 967                  */
 968                 diff_tsc = cur_tsc - prev_tsc;
 969                 if (unlikely(diff_tsc > drain_tsc)) {
 970                         for (i = 0; i < qconf->n_tx_port; ++i) {
 971                                 portid = qconf->tx_port_id[i];
 972                                 rte_eth_tx_buffer_flush(portid,
 973                                                 qconf->tx_queue_id[portid],
 974                                                 qconf->tx_buffer[portid]);
 975                         }
 976                         prev_tsc = cur_tsc;
 977                 }
 978
 979 start_rx:
 980                 /*
 981                  * Read packet from RX queues
 982                  */
 983                 lcore_rx_idle_count = 0;
 984                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 985                         rx_queue = &(qconf->rx_queue_list[i]);
 986                         rx_queue->idle_hint = 0;
 987                         portid = rx_queue->port_id;
 988                         queueid = rx_queue->queue_id;
 989
 990                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 991                                         MAX_PKT_BURST);
 992
 993                         stats[lcore_id].nb_rx_processed += nb_rx;
 994                         if (unlikely(nb_rx == 0)) {
 995                                 /**
 996                                  * no packet received from rx queue, try to
 997                                  * sleep for a while forcing CPU enter deeper
 998                                  * C states.
 999                                  */
1000                                 rx_queue->zero_rx_packet_count++;
1001
1002                                 if (rx_queue->zero_rx_packet_count <=
1003                                                 MIN_ZERO_POLL_COUNT)
1004                                         continue;
1005
1006                                 rx_queue->idle_hint = power_idle_heuristic(
1007                                                 rx_queue->zero_rx_packet_count);
1008                                 lcore_rx_idle_count++;
1009                         } else {
1010                                 rx_queue->zero_rx_packet_count = 0;
1011                         }
1012
1013                         /* Prefetch first packets */
1014                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1015                                 rte_prefetch0(rte_pktmbuf_mtod(
1016                                                 pkts_burst[j], void *));
1017                         }
1018
1019                         /* Prefetch and forward already prefetched packets */
1020                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1021                                 rte_prefetch0(rte_pktmbuf_mtod(
1022                                                 pkts_burst[j + PREFETCH_OFFSET],
1023                                                 void *));
1024                                 l3fwd_simple_forward(
1025                                                 pkts_burst[j], portid, qconf);
1026                         }
1027
1028                         /* Forward remaining prefetched packets */
1029                         for (; j < nb_rx; j++) {
1030                                 l3fwd_simple_forward(
1031                                                 pkts_burst[j], portid, qconf);
1032                         }
1033                 }
1034
1035                 if (unlikely(lcore_rx_idle_count == qconf->n_rx_queue)) {
1036                         /**
1037                          * All Rx queues empty in recent consecutive polls,
1038                          * sleep in a conservative manner, meaning sleep as
1039                          * less as possible.
1040                          */
1041                         for (i = 1,
1042                             lcore_idle_hint = qconf->rx_queue_list[0].idle_hint;
1043                                         i < qconf->n_rx_queue; ++i) {
1044                                 rx_queue = &(qconf->rx_queue_list[i]);
1045                                 if (rx_queue->idle_hint < lcore_idle_hint)
1046                                         lcore_idle_hint = rx_queue->idle_hint;
1047                         }
1048
1049                         if (lcore_idle_hint < SUSPEND_THRESHOLD)
1050                                 /**
1051                                  * execute "pause" instruction to avoid context
1052                                  * switch which generally take hundred of
1053                                  * microseconds for short sleep.
1054                                  */
1055                                 rte_delay_us(lcore_idle_hint);
1056                         else {
1057                                 /* suspend until rx interrupt triggers */
1058                                 if (intr_en) {
1059                                         turn_on_off_intr(qconf, 1);
1060                                         sleep_until_rx_interrupt(
1061                                                         qconf->n_rx_queue,
1062                                                         lcore_id);
1063                                         turn_on_off_intr(qconf, 0);
1064                                         /**
1065                                          * start receiving packets immediately
1066                                          */
1067                                         if (likely(!is_done()))
1068                                                 goto start_rx;
1069                                 }
1070                         }
1071                         stats[lcore_id].sleep_time += lcore_idle_hint;
1072                 }
1073         }
1074
1075         return 0;
1076 }
1077
1078 /* main processing loop */
1079 static int
1080 main_telemetry_loop(__rte_unused void *dummy)
1081 {
1082         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1083         unsigned int lcore_id;
1084         uint64_t prev_tsc, diff_tsc, cur_tsc, prev_tel_tsc;
1085         int i, j, nb_rx;
1086         uint8_t queueid;
1087         uint16_t portid;
1088         struct lcore_conf *qconf;
1089         struct lcore_rx_queue *rx_queue;
1090         uint64_t ep_nep[2] = {0}, fp_nfp[2] = {0};
1091         uint64_t poll_count;
1092         enum busy_rate br;
1093
1094         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1095                                         US_PER_S * BURST_TX_DRAIN_US;
1096
1097         poll_count = 0;
1098         prev_tsc = 0;
1099         prev_tel_tsc = 0;
1100
1101         lcore_id = rte_lcore_id();
1102         qconf = &lcore_conf[lcore_id];
1103
1104         if (qconf->n_rx_queue == 0) {
1105                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
1106                         lcore_id);
1107                 return 0;
1108         }
1109
1110         RTE_LOG(INFO, L3FWD_POWER, "entering main telemetry loop on lcore %u\n",
1111                 lcore_id);
1112
1113         for (i = 0; i < qconf->n_rx_queue; i++) {
1114                 portid = qconf->rx_queue_list[i].port_id;
1115                 queueid = qconf->rx_queue_list[i].queue_id;
1116                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1117                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
1118         }
1119
1120         while (!is_done()) {
1121
1122                 cur_tsc = rte_rdtsc();
1123                 /*
1124                  * TX burst queue drain
1125                  */
1126                 diff_tsc = cur_tsc - prev_tsc;
1127                 if (unlikely(diff_tsc > drain_tsc)) {
1128                         for (i = 0; i < qconf->n_tx_port; ++i) {
1129                                 portid = qconf->tx_port_id[i];
1130                                 rte_eth_tx_buffer_flush(portid,
1131                                                 qconf->tx_queue_id[portid],
1132                                                 qconf->tx_buffer[portid]);
1133                         }
1134                         prev_tsc = cur_tsc;
1135                 }
1136
1137                 /*
1138                  * Read packet from RX queues
1139                  */
1140                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1141                         rx_queue = &(qconf->rx_queue_list[i]);
1142                         portid = rx_queue->port_id;
1143                         queueid = rx_queue->queue_id;
1144
1145                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1146                                                                 MAX_PKT_BURST);
1147                         ep_nep[nb_rx == 0]++;
1148                         fp_nfp[nb_rx == MAX_PKT_BURST]++;
1149                         poll_count++;
1150                         if (unlikely(nb_rx == 0))
1151                                 continue;
1152
1153                         /* Prefetch first packets */
1154                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1155                                 rte_prefetch0(rte_pktmbuf_mtod(
1156                                                 pkts_burst[j], void *));
1157                         }
1158
1159                         /* Prefetch and forward already prefetched packets */
1160                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1161                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1162                                                 j + PREFETCH_OFFSET], void *));
1163                                 l3fwd_simple_forward(pkts_burst[j], portid,
1164                                                                 qconf);
1165                         }
1166
1167                         /* Forward remaining prefetched packets */
1168                         for (; j < nb_rx; j++) {
1169                                 l3fwd_simple_forward(pkts_burst[j], portid,
1170                                                                 qconf);
1171                         }
1172                 }
1173                 if (unlikely(poll_count >= DEFAULT_COUNT)) {
1174                         diff_tsc = cur_tsc - prev_tel_tsc;
1175                         if (diff_tsc >= MAX_CYCLES) {
1176                                 br = FULL;
1177                         } else if (diff_tsc > MIN_CYCLES &&
1178                                         diff_tsc < MAX_CYCLES) {
1179                                 br = (diff_tsc * 100) / MAX_CYCLES;
1180                         } else {
1181                                 br = ZERO;
1182                         }
1183                         poll_count = 0;
1184                         prev_tel_tsc = cur_tsc;
1185                         /* update stats for telemetry */
1186                         rte_spinlock_lock(&stats[lcore_id].telemetry_lock);
1187                         stats[lcore_id].ep_nep[0] = ep_nep[0];
1188                         stats[lcore_id].ep_nep[1] = ep_nep[1];
1189                         stats[lcore_id].fp_nfp[0] = fp_nfp[0];
1190                         stats[lcore_id].fp_nfp[1] = fp_nfp[1];
1191                         stats[lcore_id].br = br;
1192                         rte_spinlock_unlock(&stats[lcore_id].telemetry_lock);
1193                 }
1194         }
1195
1196         return 0;
1197 }
1198 /* main processing loop */
1199 static int
1200 main_empty_poll_loop(__rte_unused void *dummy)
1201 {
1202         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1203         unsigned int lcore_id;
1204         uint64_t prev_tsc, diff_tsc, cur_tsc;
1205         int i, j, nb_rx;
1206         uint8_t queueid;
1207         uint16_t portid;
1208         struct lcore_conf *qconf;
1209         struct lcore_rx_queue *rx_queue;
1210
1211         const uint64_t drain_tsc =
1212                 (rte_get_tsc_hz() + US_PER_S - 1) /
1213                 US_PER_S * BURST_TX_DRAIN_US;
1214
1215         prev_tsc = 0;
1216
1217         lcore_id = rte_lcore_id();
1218         qconf = &lcore_conf[lcore_id];
1219
1220         if (qconf->n_rx_queue == 0) {
1221                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
1222                         lcore_id);
1223                 return 0;
1224         }
1225
1226         for (i = 0; i < qconf->n_rx_queue; i++) {
1227                 portid = qconf->rx_queue_list[i].port_id;
1228                 queueid = qconf->rx_queue_list[i].queue_id;
1229                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1230                                 "rxqueueid=%hhu\n", lcore_id, portid, queueid);
1231         }
1232
1233         while (!is_done()) {
1234                 stats[lcore_id].nb_iteration_looped++;
1235
1236                 cur_tsc = rte_rdtsc();
1237                 /*
1238                  * TX burst queue drain
1239                  */
1240                 diff_tsc = cur_tsc - prev_tsc;
1241                 if (unlikely(diff_tsc > drain_tsc)) {
1242                         for (i = 0; i < qconf->n_tx_port; ++i) {
1243                                 portid = qconf->tx_port_id[i];
1244                                 rte_eth_tx_buffer_flush(portid,
1245                                                 qconf->tx_queue_id[portid],
1246                                                 qconf->tx_buffer[portid]);
1247                         }
1248                         prev_tsc = cur_tsc;
1249                 }
1250
1251                 /*
1252                  * Read packet from RX queues
1253                  */
1254                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1255                         rx_queue = &(qconf->rx_queue_list[i]);
1256                         rx_queue->idle_hint = 0;
1257                         portid = rx_queue->port_id;
1258                         queueid = rx_queue->queue_id;
1259
1260                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1261                                         MAX_PKT_BURST);
1262
1263                         stats[lcore_id].nb_rx_processed += nb_rx;
1264
1265                         if (nb_rx == 0) {
1266
1267                                 rte_power_empty_poll_stat_update(lcore_id);
1268
1269                                 continue;
1270                         } else {
1271                                 rte_power_poll_stat_update(lcore_id, nb_rx);
1272                         }
1273
1274
1275                         /* Prefetch first packets */
1276                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1277                                 rte_prefetch0(rte_pktmbuf_mtod(
1278                                                         pkts_burst[j], void *));
1279                         }
1280
1281                         /* Prefetch and forward already prefetched packets */
1282                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1283                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1284                                                         j + PREFETCH_OFFSET],
1285                                                         void *));
1286                                 l3fwd_simple_forward(pkts_burst[j], portid,
1287                                                 qconf);
1288                         }
1289
1290                         /* Forward remaining prefetched packets */
1291                         for (; j < nb_rx; j++) {
1292                                 l3fwd_simple_forward(pkts_burst[j], portid,
1293                                                 qconf);
1294                         }
1295
1296                 }
1297
1298         }
1299
1300         return 0;
1301 }
1302 /* main processing loop */
1303 static int
1304 main_legacy_loop(__rte_unused void *dummy)
1305 {
1306         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1307         unsigned lcore_id;
1308         uint64_t prev_tsc, diff_tsc, cur_tsc, tim_res_tsc, hz;
1309         uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
1310         int i, j, nb_rx;
1311         uint8_t queueid;
1312         uint16_t portid;
1313         struct lcore_conf *qconf;
1314         struct lcore_rx_queue *rx_queue;
1315         enum freq_scale_hint_t lcore_scaleup_hint;
1316         uint32_t lcore_rx_idle_count = 0;
1317         uint32_t lcore_idle_hint = 0;
1318         int intr_en = 0;
1319
1320         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
1321
1322         prev_tsc = 0;
1323         hz = rte_get_timer_hz();
1324         tim_res_tsc = hz/TIMER_NUMBER_PER_SECOND;
1325
1326         lcore_id = rte_lcore_id();
1327         qconf = &lcore_conf[lcore_id];
1328
1329         if (qconf->n_rx_queue == 0) {
1330                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
1331                 return 0;
1332         }
1333
1334         RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
1335
1336         for (i = 0; i < qconf->n_rx_queue; i++) {
1337                 portid = qconf->rx_queue_list[i].port_id;
1338                 queueid = qconf->rx_queue_list[i].queue_id;
1339                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1340                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
1341         }
1342
1343         /* add into event wait list */
1344         if (event_register(qconf) == 0)
1345                 intr_en = 1;
1346         else
1347                 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
1348
1349         while (!is_done()) {
1350                 stats[lcore_id].nb_iteration_looped++;
1351
1352                 cur_tsc = rte_rdtsc();
1353                 cur_tsc_power = cur_tsc;
1354
1355                 /*
1356                  * TX burst queue drain
1357                  */
1358                 diff_tsc = cur_tsc - prev_tsc;
1359                 if (unlikely(diff_tsc > drain_tsc)) {
1360                         for (i = 0; i < qconf->n_tx_port; ++i) {
1361                                 portid = qconf->tx_port_id[i];
1362                                 rte_eth_tx_buffer_flush(portid,
1363                                                 qconf->tx_queue_id[portid],
1364                                                 qconf->tx_buffer[portid]);
1365                         }
1366                         prev_tsc = cur_tsc;
1367                 }
1368
1369                 diff_tsc_power = cur_tsc_power - prev_tsc_power;
1370                 if (diff_tsc_power > tim_res_tsc) {
1371                         rte_timer_manage();
1372                         prev_tsc_power = cur_tsc_power;
1373                 }
1374
1375 start_rx:
1376                 /*
1377                  * Read packet from RX queues
1378                  */
1379                 lcore_scaleup_hint = FREQ_CURRENT;
1380                 lcore_rx_idle_count = 0;
1381                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1382                         rx_queue = &(qconf->rx_queue_list[i]);
1383                         rx_queue->idle_hint = 0;
1384                         portid = rx_queue->port_id;
1385                         queueid = rx_queue->queue_id;
1386
1387                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1388                                                                 MAX_PKT_BURST);
1389
1390                         stats[lcore_id].nb_rx_processed += nb_rx;
1391                         if (unlikely(nb_rx == 0)) {
1392                                 /**
1393                                  * no packet received from rx queue, try to
1394                                  * sleep for a while forcing CPU enter deeper
1395                                  * C states.
1396                                  */
1397                                 rx_queue->zero_rx_packet_count++;
1398
1399                                 if (rx_queue->zero_rx_packet_count <=
1400                                                         MIN_ZERO_POLL_COUNT)
1401                                         continue;
1402
1403                                 rx_queue->idle_hint = power_idle_heuristic(\
1404                                         rx_queue->zero_rx_packet_count);
1405                                 lcore_rx_idle_count++;
1406                         } else {
1407                                 rx_queue->zero_rx_packet_count = 0;
1408
1409                                 /**
1410                                  * do not scale up frequency immediately as
1411                                  * user to kernel space communication is costly
1412                                  * which might impact packet I/O for received
1413                                  * packets.
1414                                  */
1415                                 rx_queue->freq_up_hint =
1416                                         power_freq_scaleup_heuristic(lcore_id,
1417                                                         portid, queueid);
1418                         }
1419
1420                         /* Prefetch first packets */
1421                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1422                                 rte_prefetch0(rte_pktmbuf_mtod(
1423                                                 pkts_burst[j], void *));
1424                         }
1425
1426                         /* Prefetch and forward already prefetched packets */
1427                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1428                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1429                                                 j + PREFETCH_OFFSET], void *));
1430                                 l3fwd_simple_forward(pkts_burst[j], portid,
1431                                                                 qconf);
1432                         }
1433
1434                         /* Forward remaining prefetched packets */
1435                         for (; j < nb_rx; j++) {
1436                                 l3fwd_simple_forward(pkts_burst[j], portid,
1437                                                                 qconf);
1438                         }
1439                 }
1440
1441                 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
1442                         for (i = 1, lcore_scaleup_hint =
1443                                 qconf->rx_queue_list[0].freq_up_hint;
1444                                         i < qconf->n_rx_queue; ++i) {
1445                                 rx_queue = &(qconf->rx_queue_list[i]);
1446                                 if (rx_queue->freq_up_hint >
1447                                                 lcore_scaleup_hint)
1448                                         lcore_scaleup_hint =
1449                                                 rx_queue->freq_up_hint;
1450                         }
1451
1452                         if (lcore_scaleup_hint == FREQ_HIGHEST) {
1453                                 if (rte_power_freq_max)
1454                                         rte_power_freq_max(lcore_id);
1455                         } else if (lcore_scaleup_hint == FREQ_HIGHER) {
1456                                 if (rte_power_freq_up)
1457                                         rte_power_freq_up(lcore_id);
1458                         }
1459                 } else {
1460                         /**
1461                          * All Rx queues empty in recent consecutive polls,
1462                          * sleep in a conservative manner, meaning sleep as
1463                          * less as possible.
1464                          */
1465                         for (i = 1, lcore_idle_hint =
1466                                 qconf->rx_queue_list[0].idle_hint;
1467                                         i < qconf->n_rx_queue; ++i) {
1468                                 rx_queue = &(qconf->rx_queue_list[i]);
1469                                 if (rx_queue->idle_hint < lcore_idle_hint)
1470                                         lcore_idle_hint = rx_queue->idle_hint;
1471                         }
1472
1473                         if (lcore_idle_hint < SUSPEND_THRESHOLD)
1474                                 /**
1475                                  * execute "pause" instruction to avoid context
1476                                  * switch which generally take hundred of
1477                                  * microseconds for short sleep.
1478                                  */
1479                                 rte_delay_us(lcore_idle_hint);
1480                         else {
1481                                 /* suspend until rx interrupt triggers */
1482                                 if (intr_en) {
1483                                         turn_on_off_intr(qconf, 1);
1484                                         sleep_until_rx_interrupt(
1485                                                         qconf->n_rx_queue,
1486                                                         lcore_id);
1487                                         turn_on_off_intr(qconf, 0);
1488                                         /**
1489                                          * start receiving packets immediately
1490                                          */
1491                                         if (likely(!is_done()))
1492                                                 goto start_rx;
1493                                 }
1494                         }
1495                         stats[lcore_id].sleep_time += lcore_idle_hint;
1496                 }
1497         }
1498
1499         return 0;
1500 }
1501
1502 static int
1503 check_lcore_params(void)
1504 {
1505         uint8_t queue, lcore;
1506         uint16_t i;
1507         int socketid;
1508
1509         for (i = 0; i < nb_lcore_params; ++i) {
1510                 queue = lcore_params[i].queue_id;
1511                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1512                         printf("invalid queue number: %hhu\n", queue);
1513                         return -1;
1514                 }
1515                 lcore = lcore_params[i].lcore_id;
1516                 if (!rte_lcore_is_enabled(lcore)) {
1517                         printf("error: lcore %hhu is not enabled in lcore "
1518                                                         "mask\n", lcore);
1519                         return -1;
1520                 }
1521                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1522                                                         (numa_on == 0)) {
1523                         printf("warning: lcore %hhu is on socket %d with numa "
1524                                                 "off\n", lcore, socketid);
1525                 }
1526                 if (app_mode == APP_MODE_TELEMETRY && lcore == rte_lcore_id()) {
1527                         printf("cannot enable main core %d in config for telemetry mode\n",
1528                                 rte_lcore_id());
1529                         return -1;
1530                 }
1531         }
1532         return 0;
1533 }
1534
1535 static int
1536 check_port_config(void)
1537 {
1538         unsigned portid;
1539         uint16_t i;
1540
1541         for (i = 0; i < nb_lcore_params; ++i) {
1542                 portid = lcore_params[i].port_id;
1543                 if ((enabled_port_mask & (1 << portid)) == 0) {
1544                         printf("port %u is not enabled in port mask\n",
1545                                                                 portid);
1546                         return -1;
1547                 }
1548                 if (!rte_eth_dev_is_valid_port(portid)) {
1549                         printf("port %u is not present on the board\n",
1550                                                                 portid);
1551                         return -1;
1552                 }
1553         }
1554         return 0;
1555 }
1556
1557 static uint8_t
1558 get_port_n_rx_queues(const uint16_t port)
1559 {
1560         int queue = -1;
1561         uint16_t i;
1562
1563         for (i = 0; i < nb_lcore_params; ++i) {
1564                 if (lcore_params[i].port_id == port &&
1565                                 lcore_params[i].queue_id > queue)
1566                         queue = lcore_params[i].queue_id;
1567         }
1568         return (uint8_t)(++queue);
1569 }
1570
1571 static int
1572 init_lcore_rx_queues(void)
1573 {
1574         uint16_t i, nb_rx_queue;
1575         uint8_t lcore;
1576
1577         for (i = 0; i < nb_lcore_params; ++i) {
1578                 lcore = lcore_params[i].lcore_id;
1579                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1580                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1581                         printf("error: too many queues (%u) for lcore: %u\n",
1582                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1583                         return -1;
1584                 } else {
1585                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1586                                 lcore_params[i].port_id;
1587                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1588                                 lcore_params[i].queue_id;
1589                         lcore_conf[lcore].n_rx_queue++;
1590                 }
1591         }
1592         return 0;
1593 }
1594
1595 /* display usage */
1596 static void
1597 print_usage(const char *prgname)
1598 {
1599         printf ("%s [EAL options] -- -p PORTMASK -P"
1600                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1601                 "  [--high-perf-cores CORELIST"
1602                 "  [--perf-config (port,queue,hi_perf,lcore_index)[,(port,queue,hi_perf,lcore_index]]"
1603                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1604                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1605                 "  -P : enable promiscuous mode\n"
1606                 "  --config (port,queue,lcore): rx queues configuration\n"
1607                 "  --high-perf-cores CORELIST: list of high performance cores\n"
1608                 "  --perf-config: similar as config, cores specified as indices"
1609                 " for bins containing high or regular performance cores\n"
1610                 "  --no-numa: optional, disable numa awareness\n"
1611                 "  --enable-jumbo: enable jumbo frame"
1612                 " which max packet len is PKTLEN in decimal (64-9600)\n"
1613                 "  --parse-ptype: parse packet type by software\n"
1614                 "  --legacy: use legacy interrupt-based scaling\n"
1615                 "  --empty-poll: enable empty poll detection"
1616                 " follow (training_flag, high_threshold, med_threshold)\n"
1617                 " --telemetry: enable telemetry mode, to update"
1618                 " empty polls, full polls, and core busyness to telemetry\n"
1619                 " --interrupt-only: enable interrupt-only mode\n"
1620                 " --pmd-mgmt MODE: enable PMD power management mode. "
1621                 "Currently supported modes: baseline, monitor, pause, scale\n",
1622                 prgname);
1623 }
1624
1625 static int parse_max_pkt_len(const char *pktlen)
1626 {
1627         char *end = NULL;
1628         unsigned long len;
1629
1630         /* parse decimal string */
1631         len = strtoul(pktlen, &end, 10);
1632         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1633                 return -1;
1634
1635         if (len == 0)
1636                 return -1;
1637
1638         return len;
1639 }
1640
1641 static int
1642 parse_portmask(const char *portmask)
1643 {
1644         char *end = NULL;
1645         unsigned long pm;
1646
1647         /* parse hexadecimal string */
1648         pm = strtoul(portmask, &end, 16);
1649         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1650                 return 0;
1651
1652         return pm;
1653 }
1654
1655 static int
1656 parse_config(const char *q_arg)
1657 {
1658         char s[256];
1659         const char *p, *p0 = q_arg;
1660         char *end;
1661         enum fieldnames {
1662                 FLD_PORT = 0,
1663                 FLD_QUEUE,
1664                 FLD_LCORE,
1665                 _NUM_FLD
1666         };
1667         unsigned long int_fld[_NUM_FLD];
1668         char *str_fld[_NUM_FLD];
1669         int i;
1670         unsigned size;
1671
1672         nb_lcore_params = 0;
1673
1674         while ((p = strchr(p0,'(')) != NULL) {
1675                 ++p;
1676                 if((p0 = strchr(p,')')) == NULL)
1677                         return -1;
1678
1679                 size = p0 - p;
1680                 if(size >= sizeof(s))
1681                         return -1;
1682
1683                 snprintf(s, sizeof(s), "%.*s", size, p);
1684                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1685                                                                 _NUM_FLD)
1686                         return -1;
1687                 for (i = 0; i < _NUM_FLD; i++){
1688                         errno = 0;
1689                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1690                         if (errno != 0 || end == str_fld[i] || int_fld[i] >
1691                                                                         255)
1692                                 return -1;
1693                 }
1694                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1695                         printf("exceeded max number of lcore params: %hu\n",
1696                                 nb_lcore_params);
1697                         return -1;
1698                 }
1699                 lcore_params_array[nb_lcore_params].port_id =
1700                                 (uint8_t)int_fld[FLD_PORT];
1701                 lcore_params_array[nb_lcore_params].queue_id =
1702                                 (uint8_t)int_fld[FLD_QUEUE];
1703                 lcore_params_array[nb_lcore_params].lcore_id =
1704                                 (uint8_t)int_fld[FLD_LCORE];
1705                 ++nb_lcore_params;
1706         }
1707         lcore_params = lcore_params_array;
1708
1709         return 0;
1710 }
1711
1712 static int
1713 parse_pmd_mgmt_config(const char *name)
1714 {
1715 #define PMD_MGMT_MONITOR "monitor"
1716 #define PMD_MGMT_PAUSE   "pause"
1717 #define PMD_MGMT_SCALE   "scale"
1718 #define PMD_MGMT_BASELINE  "baseline"
1719
1720         if (strncmp(PMD_MGMT_MONITOR, name, sizeof(PMD_MGMT_MONITOR)) == 0) {
1721                 pmgmt_type = RTE_POWER_MGMT_TYPE_MONITOR;
1722                 return 0;
1723         }
1724
1725         if (strncmp(PMD_MGMT_PAUSE, name, sizeof(PMD_MGMT_PAUSE)) == 0) {
1726                 pmgmt_type = RTE_POWER_MGMT_TYPE_PAUSE;
1727                 return 0;
1728         }
1729
1730         if (strncmp(PMD_MGMT_SCALE, name, sizeof(PMD_MGMT_SCALE)) == 0) {
1731                 pmgmt_type = RTE_POWER_MGMT_TYPE_SCALE;
1732                 return 0;
1733         }
1734         if (strncmp(PMD_MGMT_BASELINE, name, sizeof(PMD_MGMT_BASELINE)) == 0) {
1735                 baseline_enabled = true;
1736                 return 0;
1737         }
1738         /* unknown PMD power management mode */
1739         return -1;
1740 }
1741
1742 static int
1743 parse_ep_config(const char *q_arg)
1744 {
1745         char s[256];
1746         const char *p = q_arg;
1747         char *end;
1748         int  num_arg;
1749
1750         char *str_fld[3];
1751
1752         int training_flag;
1753         int med_edpi;
1754         int hgh_edpi;
1755
1756         ep_med_edpi = EMPTY_POLL_MED_THRESHOLD;
1757         ep_hgh_edpi = EMPTY_POLL_HGH_THRESHOLD;
1758
1759         strlcpy(s, p, sizeof(s));
1760
1761         num_arg = rte_strsplit(s, sizeof(s), str_fld, 3, ',');
1762
1763         empty_poll_train = false;
1764
1765         if (num_arg == 0)
1766                 return 0;
1767
1768         if (num_arg == 3) {
1769
1770                 training_flag = strtoul(str_fld[0], &end, 0);
1771                 med_edpi = strtoul(str_fld[1], &end, 0);
1772                 hgh_edpi = strtoul(str_fld[2], &end, 0);
1773
1774                 if (training_flag == 1)
1775                         empty_poll_train = true;
1776
1777                 if (med_edpi > 0)
1778                         ep_med_edpi = med_edpi;
1779
1780                 if (hgh_edpi > 0)
1781                         ep_hgh_edpi = hgh_edpi;
1782
1783         } else {
1784
1785                 return -1;
1786         }
1787
1788         return 0;
1789
1790 }
1791 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1792 #define CMD_LINE_OPT_LEGACY "legacy"
1793 #define CMD_LINE_OPT_EMPTY_POLL "empty-poll"
1794 #define CMD_LINE_OPT_INTERRUPT_ONLY "interrupt-only"
1795 #define CMD_LINE_OPT_TELEMETRY "telemetry"
1796 #define CMD_LINE_OPT_PMD_MGMT "pmd-mgmt"
1797
1798 /* Parse the argument given in the command line of the application */
1799 static int
1800 parse_args(int argc, char **argv)
1801 {
1802         int opt, ret;
1803         char **argvopt;
1804         int option_index;
1805         uint32_t limit;
1806         char *prgname = argv[0];
1807         static struct option lgopts[] = {
1808                 {"config", 1, 0, 0},
1809                 {"perf-config", 1, 0, 0},
1810                 {"high-perf-cores", 1, 0, 0},
1811                 {"no-numa", 0, 0, 0},
1812                 {"enable-jumbo", 0, 0, 0},
1813                 {CMD_LINE_OPT_EMPTY_POLL, 1, 0, 0},
1814                 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
1815                 {CMD_LINE_OPT_LEGACY, 0, 0, 0},
1816                 {CMD_LINE_OPT_TELEMETRY, 0, 0, 0},
1817                 {CMD_LINE_OPT_INTERRUPT_ONLY, 0, 0, 0},
1818                 {CMD_LINE_OPT_PMD_MGMT, 1, 0, 0},
1819                 {NULL, 0, 0, 0}
1820         };
1821
1822         argvopt = argv;
1823
1824         while ((opt = getopt_long(argc, argvopt, "p:l:m:h:P",
1825                                 lgopts, &option_index)) != EOF) {
1826
1827                 switch (opt) {
1828                 /* portmask */
1829                 case 'p':
1830                         enabled_port_mask = parse_portmask(optarg);
1831                         if (enabled_port_mask == 0) {
1832                                 printf("invalid portmask\n");
1833                                 print_usage(prgname);
1834                                 return -1;
1835                         }
1836                         break;
1837                 case 'P':
1838                         printf("Promiscuous mode selected\n");
1839                         promiscuous_on = 1;
1840                         break;
1841                 case 'l':
1842                         limit = parse_max_pkt_len(optarg);
1843                         freq_tlb[LOW] = limit;
1844                         break;
1845                 case 'm':
1846                         limit = parse_max_pkt_len(optarg);
1847                         freq_tlb[MED] = limit;
1848                         break;
1849                 case 'h':
1850                         limit = parse_max_pkt_len(optarg);
1851                         freq_tlb[HGH] = limit;
1852                         break;
1853                 /* long options */
1854                 case 0:
1855                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1856                                 ret = parse_config(optarg);
1857                                 if (ret) {
1858                                         printf("invalid config\n");
1859                                         print_usage(prgname);
1860                                         return -1;
1861                                 }
1862                         }
1863
1864                         if (!strncmp(lgopts[option_index].name,
1865                                         "perf-config", 11)) {
1866                                 ret = parse_perf_config(optarg);
1867                                 if (ret) {
1868                                         printf("invalid perf-config\n");
1869                                         print_usage(prgname);
1870                                         return -1;
1871                                 }
1872                         }
1873
1874                         if (!strncmp(lgopts[option_index].name,
1875                                         "high-perf-cores", 15)) {
1876                                 ret = parse_perf_core_list(optarg);
1877                                 if (ret) {
1878                                         printf("invalid high-perf-cores\n");
1879                                         print_usage(prgname);
1880                                         return -1;
1881                                 }
1882                         }
1883
1884                         if (!strncmp(lgopts[option_index].name,
1885                                                 "no-numa", 7)) {
1886                                 printf("numa is disabled \n");
1887                                 numa_on = 0;
1888                         }
1889
1890                         if (!strncmp(lgopts[option_index].name,
1891                                         CMD_LINE_OPT_LEGACY,
1892                                         sizeof(CMD_LINE_OPT_LEGACY))) {
1893                                 if (app_mode != APP_MODE_DEFAULT) {
1894                                         printf(" legacy mode is mutually exclusive with other modes\n");
1895                                         return -1;
1896                                 }
1897                                 app_mode = APP_MODE_LEGACY;
1898                                 printf("legacy mode is enabled\n");
1899                         }
1900
1901                         if (!strncmp(lgopts[option_index].name,
1902                                         CMD_LINE_OPT_EMPTY_POLL, 10)) {
1903                                 if (app_mode != APP_MODE_DEFAULT) {
1904                                         printf(" empty-poll mode is mutually exclusive with other modes\n");
1905                                         return -1;
1906                                 }
1907                                 app_mode = APP_MODE_EMPTY_POLL;
1908                                 ret = parse_ep_config(optarg);
1909
1910                                 if (ret) {
1911                                         printf("invalid empty poll config\n");
1912                                         print_usage(prgname);
1913                                         return -1;
1914                                 }
1915                                 printf("empty-poll is enabled\n");
1916                         }
1917
1918                         if (!strncmp(lgopts[option_index].name,
1919                                         CMD_LINE_OPT_TELEMETRY,
1920                                         sizeof(CMD_LINE_OPT_TELEMETRY))) {
1921                                 if (app_mode != APP_MODE_DEFAULT) {
1922                                         printf(" telemetry mode is mutually exclusive with other modes\n");
1923                                         return -1;
1924                                 }
1925                                 app_mode = APP_MODE_TELEMETRY;
1926                                 printf("telemetry mode is enabled\n");
1927                         }
1928
1929                         if (!strncmp(lgopts[option_index].name,
1930                                         CMD_LINE_OPT_PMD_MGMT,
1931                                         sizeof(CMD_LINE_OPT_PMD_MGMT))) {
1932                                 if (app_mode != APP_MODE_DEFAULT) {
1933                                         printf(" power mgmt mode is mutually exclusive with other modes\n");
1934                                         return -1;
1935                                 }
1936                                 if (parse_pmd_mgmt_config(optarg) < 0) {
1937                                         printf(" Invalid PMD power management mode: %s\n",
1938                                                         optarg);
1939                                         return -1;
1940                                 }
1941                                 app_mode = APP_MODE_PMD_MGMT;
1942                                 printf("PMD power mgmt mode is enabled\n");
1943                         }
1944                         if (!strncmp(lgopts[option_index].name,
1945                                         CMD_LINE_OPT_INTERRUPT_ONLY,
1946                                         sizeof(CMD_LINE_OPT_INTERRUPT_ONLY))) {
1947                                 if (app_mode != APP_MODE_DEFAULT) {
1948                                         printf(" interrupt-only mode is mutually exclusive with other modes\n");
1949                                         return -1;
1950                                 }
1951                                 app_mode = APP_MODE_INTERRUPT;
1952                                 printf("interrupt-only mode is enabled\n");
1953                         }
1954
1955                         if (!strncmp(lgopts[option_index].name,
1956                                         "enable-jumbo", 12)) {
1957                                 struct option lenopts =
1958                                         {"max-pkt-len", required_argument, \
1959                                                                         0, 0};
1960
1961                                 printf("jumbo frame is enabled \n");
1962                                 port_conf.rxmode.offloads |=
1963                                                 DEV_RX_OFFLOAD_JUMBO_FRAME;
1964                                 port_conf.txmode.offloads |=
1965                                                 DEV_TX_OFFLOAD_MULTI_SEGS;
1966
1967                                 /**
1968                                  * if no max-pkt-len set, use the default value
1969                                  * RTE_ETHER_MAX_LEN
1970                                  */
1971                                 if (0 == getopt_long(argc, argvopt, "",
1972                                                 &lenopts, &option_index)) {
1973                                         ret = parse_max_pkt_len(optarg);
1974                                         if ((ret < 64) ||
1975                                                 (ret > MAX_JUMBO_PKT_LEN)){
1976                                                 printf("invalid packet "
1977                                                                 "length\n");
1978                                                 print_usage(prgname);
1979                                                 return -1;
1980                                         }
1981                                         port_conf.rxmode.max_rx_pkt_len = ret;
1982                                 }
1983                                 printf("set jumbo frame "
1984                                         "max packet length to %u\n",
1985                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1986                         }
1987
1988                         if (!strncmp(lgopts[option_index].name,
1989                                      CMD_LINE_OPT_PARSE_PTYPE,
1990                                      sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
1991                                 printf("soft parse-ptype is enabled\n");
1992                                 parse_ptype = 1;
1993                         }
1994
1995                         break;
1996
1997                 default:
1998                         print_usage(prgname);
1999                         return -1;
2000                 }
2001         }
2002
2003         if (optind >= 0)
2004                 argv[optind-1] = prgname;
2005
2006         ret = optind-1;
2007         optind = 1; /* reset getopt lib */
2008         return ret;
2009 }
2010
2011 static void
2012 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
2013 {
2014         char buf[RTE_ETHER_ADDR_FMT_SIZE];
2015         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
2016         printf("%s%s", name, buf);
2017 }
2018
2019 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2020 static void
2021 setup_hash(int socketid)
2022 {
2023         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
2024                 .name = NULL,
2025                 .entries = L3FWD_HASH_ENTRIES,
2026                 .key_len = sizeof(struct ipv4_5tuple),
2027                 .hash_func = DEFAULT_HASH_FUNC,
2028                 .hash_func_init_val = 0,
2029         };
2030
2031         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
2032                 .name = NULL,
2033                 .entries = L3FWD_HASH_ENTRIES,
2034                 .key_len = sizeof(struct ipv6_5tuple),
2035                 .hash_func = DEFAULT_HASH_FUNC,
2036                 .hash_func_init_val = 0,
2037         };
2038
2039         unsigned i;
2040         int ret;
2041         char s[64];
2042
2043         /* create ipv4 hash */
2044         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
2045         ipv4_l3fwd_hash_params.name = s;
2046         ipv4_l3fwd_hash_params.socket_id = socketid;
2047         ipv4_l3fwd_lookup_struct[socketid] =
2048                 rte_hash_create(&ipv4_l3fwd_hash_params);
2049         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2050                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2051                                 "socket %d\n", socketid);
2052
2053         /* create ipv6 hash */
2054         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
2055         ipv6_l3fwd_hash_params.name = s;
2056         ipv6_l3fwd_hash_params.socket_id = socketid;
2057         ipv6_l3fwd_lookup_struct[socketid] =
2058                 rte_hash_create(&ipv6_l3fwd_hash_params);
2059         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
2060                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2061                                 "socket %d\n", socketid);
2062
2063
2064         /* populate the ipv4 hash */
2065         for (i = 0; i < RTE_DIM(ipv4_l3fwd_route_array); i++) {
2066                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
2067                                 (void *) &ipv4_l3fwd_route_array[i].key);
2068                 if (ret < 0) {
2069                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
2070                                 "l3fwd hash on socket %d\n", i, socketid);
2071                 }
2072                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
2073                 printf("Hash: Adding key\n");
2074                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
2075         }
2076
2077         /* populate the ipv6 hash */
2078         for (i = 0; i < RTE_DIM(ipv6_l3fwd_route_array); i++) {
2079                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
2080                                 (void *) &ipv6_l3fwd_route_array[i].key);
2081                 if (ret < 0) {
2082                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
2083                                 "l3fwd hash on socket %d\n", i, socketid);
2084                 }
2085                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
2086                 printf("Hash: Adding key\n");
2087                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
2088         }
2089 }
2090 #endif
2091
2092 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2093 static void
2094 setup_lpm(int socketid)
2095 {
2096         unsigned i;
2097         int ret;
2098         char s[64];
2099
2100         /* create the LPM table */
2101         struct rte_lpm_config lpm_ipv4_config;
2102
2103         lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
2104         lpm_ipv4_config.number_tbl8s = 256;
2105         lpm_ipv4_config.flags = 0;
2106
2107         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
2108         ipv4_l3fwd_lookup_struct[socketid] =
2109                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
2110         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2111                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
2112                                 " on socket %d\n", socketid);
2113
2114         /* populate the LPM table */
2115         for (i = 0; i < RTE_DIM(ipv4_l3fwd_route_array); i++) {
2116                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
2117                         ipv4_l3fwd_route_array[i].ip,
2118                         ipv4_l3fwd_route_array[i].depth,
2119                         ipv4_l3fwd_route_array[i].if_out);
2120
2121                 if (ret < 0) {
2122                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
2123                                 "l3fwd LPM table on socket %d\n",
2124                                 i, socketid);
2125                 }
2126
2127                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
2128                         (unsigned)ipv4_l3fwd_route_array[i].ip,
2129                         ipv4_l3fwd_route_array[i].depth,
2130                         ipv4_l3fwd_route_array[i].if_out);
2131         }
2132 }
2133 #endif
2134
2135 static int
2136 init_mem(unsigned nb_mbuf)
2137 {
2138         struct lcore_conf *qconf;
2139         int socketid;
2140         unsigned lcore_id;
2141         char s[64];
2142
2143         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2144                 if (rte_lcore_is_enabled(lcore_id) == 0)
2145                         continue;
2146
2147                 if (numa_on)
2148                         socketid = rte_lcore_to_socket_id(lcore_id);
2149                 else
2150                         socketid = 0;
2151
2152                 if (socketid >= NB_SOCKETS) {
2153                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
2154                                         "out of range %d\n", socketid,
2155                                                 lcore_id, NB_SOCKETS);
2156                 }
2157                 if (pktmbuf_pool[socketid] == NULL) {
2158                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
2159                         pktmbuf_pool[socketid] =
2160                                 rte_pktmbuf_pool_create(s, nb_mbuf,
2161                                         MEMPOOL_CACHE_SIZE, 0,
2162                                         RTE_MBUF_DEFAULT_BUF_SIZE,
2163                                         socketid);
2164                         if (pktmbuf_pool[socketid] == NULL)
2165                                 rte_exit(EXIT_FAILURE,
2166                                         "Cannot init mbuf pool on socket %d\n",
2167                                                                 socketid);
2168                         else
2169                                 printf("Allocated mbuf pool on socket %d\n",
2170                                                                 socketid);
2171
2172 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2173                         setup_lpm(socketid);
2174 #else
2175                         setup_hash(socketid);
2176 #endif
2177                 }
2178                 qconf = &lcore_conf[lcore_id];
2179                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
2180 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2181                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
2182 #endif
2183         }
2184         return 0;
2185 }
2186
2187 /* Check the link status of all ports in up to 9s, and print them finally */
2188 static void
2189 check_all_ports_link_status(uint32_t port_mask)
2190 {
2191 #define CHECK_INTERVAL 100 /* 100ms */
2192 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
2193         uint8_t count, all_ports_up, print_flag = 0;
2194         uint16_t portid;
2195         struct rte_eth_link link;
2196         int ret;
2197         char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
2198
2199         printf("\nChecking link status");
2200         fflush(stdout);
2201         for (count = 0; count <= MAX_CHECK_TIME; count++) {
2202                 all_ports_up = 1;
2203                 RTE_ETH_FOREACH_DEV(portid) {
2204                         if ((port_mask & (1 << portid)) == 0)
2205                                 continue;
2206                         memset(&link, 0, sizeof(link));
2207                         ret = rte_eth_link_get_nowait(portid, &link);
2208                         if (ret < 0) {
2209                                 all_ports_up = 0;
2210                                 if (print_flag == 1)
2211                                         printf("Port %u link get failed: %s\n",
2212                                                 portid, rte_strerror(-ret));
2213                                 continue;
2214                         }
2215                         /* print link status if flag set */
2216                         if (print_flag == 1) {
2217                                 rte_eth_link_to_str(link_status_text,
2218                                         sizeof(link_status_text), &link);
2219                                 printf("Port %d %s\n", portid,
2220                                        link_status_text);
2221                                 continue;
2222                         }
2223                         /* clear all_ports_up flag if any link down */
2224                         if (link.link_status == ETH_LINK_DOWN) {
2225                                 all_ports_up = 0;
2226                                 break;
2227                         }
2228                 }
2229                 /* after finally printing all link status, get out */
2230                 if (print_flag == 1)
2231                         break;
2232
2233                 if (all_ports_up == 0) {
2234                         printf(".");
2235                         fflush(stdout);
2236                         rte_delay_ms(CHECK_INTERVAL);
2237                 }
2238
2239                 /* set the print_flag if all ports up or timeout */
2240                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
2241                         print_flag = 1;
2242                         printf("done\n");
2243                 }
2244         }
2245 }
2246
2247 static int check_ptype(uint16_t portid)
2248 {
2249         int i, ret;
2250         int ptype_l3_ipv4 = 0;
2251 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2252         int ptype_l3_ipv6 = 0;
2253 #endif
2254         uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
2255
2256         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
2257         if (ret <= 0)
2258                 return 0;
2259
2260         uint32_t ptypes[ret];
2261
2262         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
2263         for (i = 0; i < ret; ++i) {
2264                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
2265                         ptype_l3_ipv4 = 1;
2266 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2267                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
2268                         ptype_l3_ipv6 = 1;
2269 #endif
2270         }
2271
2272         if (ptype_l3_ipv4 == 0)
2273                 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
2274
2275 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2276         if (ptype_l3_ipv6 == 0)
2277                 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
2278 #endif
2279
2280 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2281         if (ptype_l3_ipv4)
2282 #else /* APP_LOOKUP_EXACT_MATCH */
2283         if (ptype_l3_ipv4 && ptype_l3_ipv6)
2284 #endif
2285                 return 1;
2286
2287         return 0;
2288
2289 }
2290
2291 static int
2292 init_power_library(void)
2293 {
2294         enum power_management_env env;
2295         unsigned int lcore_id;
2296         int ret = 0;
2297
2298         RTE_LCORE_FOREACH(lcore_id) {
2299                 /* init power management library */
2300                 ret = rte_power_init(lcore_id);
2301                 if (ret) {
2302                         RTE_LOG(ERR, POWER,
2303                                 "Library initialization failed on core %u\n",
2304                                 lcore_id);
2305                         return ret;
2306                 }
2307                 /* we're not supporting the VM channel mode */
2308                 env = rte_power_get_env();
2309                 if (env != PM_ENV_ACPI_CPUFREQ &&
2310                                 env != PM_ENV_PSTATE_CPUFREQ) {
2311                         RTE_LOG(ERR, POWER,
2312                                 "Only ACPI and PSTATE mode are supported\n");
2313                         return -1;
2314                 }
2315         }
2316         return ret;
2317 }
2318
2319 static int
2320 deinit_power_library(void)
2321 {
2322         unsigned int lcore_id;
2323         int ret = 0;
2324
2325         RTE_LCORE_FOREACH(lcore_id) {
2326                 /* deinit power management library */
2327                 ret = rte_power_exit(lcore_id);
2328                 if (ret) {
2329                         RTE_LOG(ERR, POWER,
2330                                 "Library deinitialization failed on core %u\n",
2331                                 lcore_id);
2332                         return ret;
2333                 }
2334         }
2335         return ret;
2336 }
2337
2338 static void
2339 get_current_stat_values(uint64_t *values)
2340 {
2341         unsigned int lcore_id = rte_lcore_id();
2342         struct lcore_conf *qconf;
2343         uint64_t app_eps = 0, app_fps = 0, app_br = 0;
2344         uint64_t count = 0;
2345
2346         RTE_LCORE_FOREACH_WORKER(lcore_id) {
2347                 qconf = &lcore_conf[lcore_id];
2348                 if (qconf->n_rx_queue == 0)
2349                         continue;
2350                 count++;
2351                 rte_spinlock_lock(&stats[lcore_id].telemetry_lock);
2352                 app_eps += stats[lcore_id].ep_nep[1];
2353                 app_fps += stats[lcore_id].fp_nfp[1];
2354                 app_br += stats[lcore_id].br;
2355                 rte_spinlock_unlock(&stats[lcore_id].telemetry_lock);
2356         }
2357
2358         if (count > 0) {
2359                 values[0] = app_eps/count;
2360                 values[1] = app_fps/count;
2361                 values[2] = app_br/count;
2362         } else
2363                 memset(values, 0, sizeof(uint64_t) * NUM_TELSTATS);
2364
2365 }
2366
2367 static void
2368 update_telemetry(__rte_unused struct rte_timer *tim,
2369                 __rte_unused void *arg)
2370 {
2371         int ret;
2372         uint64_t values[NUM_TELSTATS] = {0};
2373
2374         get_current_stat_values(values);
2375         ret = rte_metrics_update_values(RTE_METRICS_GLOBAL, telstats_index,
2376                                         values, RTE_DIM(values));
2377         if (ret < 0)
2378                 RTE_LOG(WARNING, POWER, "failed to update metrcis\n");
2379 }
2380
2381 static int
2382 handle_app_stats(const char *cmd __rte_unused,
2383                 const char *params __rte_unused,
2384                 struct rte_tel_data *d)
2385 {
2386         uint64_t values[NUM_TELSTATS] = {0};
2387         uint32_t i;
2388
2389         rte_tel_data_start_dict(d);
2390         get_current_stat_values(values);
2391         for (i = 0; i < NUM_TELSTATS; i++)
2392                 rte_tel_data_add_dict_u64(d, telstats_strings[i].name,
2393                                 values[i]);
2394         return 0;
2395 }
2396
2397 static void
2398 telemetry_setup_timer(void)
2399 {
2400         int lcore_id = rte_lcore_id();
2401         uint64_t hz = rte_get_timer_hz();
2402         uint64_t ticks;
2403
2404         ticks = hz / TELEMETRY_INTERVALS_PER_SEC;
2405         rte_timer_reset_sync(&telemetry_timer,
2406                         ticks,
2407                         PERIODICAL,
2408                         lcore_id,
2409                         update_telemetry,
2410                         NULL);
2411 }
2412 static void
2413 empty_poll_setup_timer(void)
2414 {
2415         int lcore_id = rte_lcore_id();
2416         uint64_t hz = rte_get_timer_hz();
2417
2418         struct  ep_params *ep_ptr = ep_params;
2419
2420         ep_ptr->interval_ticks = hz / INTERVALS_PER_SECOND;
2421
2422         rte_timer_reset_sync(&ep_ptr->timer0,
2423                         ep_ptr->interval_ticks,
2424                         PERIODICAL,
2425                         lcore_id,
2426                         rte_empty_poll_detection,
2427                         (void *)ep_ptr);
2428
2429 }
2430 static int
2431 launch_timer(unsigned int lcore_id)
2432 {
2433         int64_t prev_tsc = 0, cur_tsc, diff_tsc, cycles_10ms;
2434
2435         RTE_SET_USED(lcore_id);
2436
2437
2438         if (rte_get_main_lcore() != lcore_id) {
2439                 rte_panic("timer on lcore:%d which is not main core:%d\n",
2440                                 lcore_id,
2441                                 rte_get_main_lcore());
2442         }
2443
2444         RTE_LOG(INFO, POWER, "Bring up the Timer\n");
2445
2446         if (app_mode == APP_MODE_EMPTY_POLL)
2447                 empty_poll_setup_timer();
2448         else
2449                 telemetry_setup_timer();
2450
2451         cycles_10ms = rte_get_timer_hz() / 100;
2452
2453         while (!is_done()) {
2454                 cur_tsc = rte_rdtsc();
2455                 diff_tsc = cur_tsc - prev_tsc;
2456                 if (diff_tsc > cycles_10ms) {
2457                         rte_timer_manage();
2458                         prev_tsc = cur_tsc;
2459                         cycles_10ms = rte_get_timer_hz() / 100;
2460                 }
2461         }
2462
2463         RTE_LOG(INFO, POWER, "Timer_subsystem is done\n");
2464
2465         return 0;
2466 }
2467
2468 static int
2469 autodetect_mode(void)
2470 {
2471         RTE_LOG(NOTICE, L3FWD_POWER, "Operating mode not specified, probing frequency scaling support...\n");
2472
2473         /*
2474          * Empty poll and telemetry modes have to be specifically requested to
2475          * be enabled, but we can auto-detect between interrupt mode with or
2476          * without frequency scaling. Both ACPI and pstate can be used.
2477          */
2478         if (rte_power_check_env_supported(PM_ENV_ACPI_CPUFREQ))
2479                 return APP_MODE_LEGACY;
2480         if (rte_power_check_env_supported(PM_ENV_PSTATE_CPUFREQ))
2481                 return APP_MODE_LEGACY;
2482
2483         RTE_LOG(NOTICE, L3FWD_POWER, "Frequency scaling not supported, selecting interrupt-only mode\n");
2484
2485         return APP_MODE_INTERRUPT;
2486 }
2487
2488 static const char *
2489 mode_to_str(enum appmode mode)
2490 {
2491         switch (mode) {
2492         case APP_MODE_LEGACY:
2493                 return "legacy";
2494         case APP_MODE_EMPTY_POLL:
2495                 return "empty poll";
2496         case APP_MODE_TELEMETRY:
2497                 return "telemetry";
2498         case APP_MODE_INTERRUPT:
2499                 return "interrupt-only";
2500         case APP_MODE_PMD_MGMT:
2501                 return "pmd mgmt";
2502         default:
2503                 return "invalid";
2504         }
2505 }
2506
2507 int
2508 main(int argc, char **argv)
2509 {
2510         struct lcore_conf *qconf;
2511         struct rte_eth_dev_info dev_info;
2512         struct rte_eth_txconf *txconf;
2513         int ret;
2514         uint16_t nb_ports;
2515         uint16_t queueid;
2516         unsigned lcore_id;
2517         uint64_t hz;
2518         uint32_t n_tx_queue, nb_lcores;
2519         uint32_t dev_rxq_num, dev_txq_num;
2520         uint8_t nb_rx_queue, queue, socketid;
2521         uint16_t portid;
2522         const char *ptr_strings[NUM_TELSTATS];
2523
2524         /* catch SIGINT and restore cpufreq governor to ondemand */
2525         signal(SIGINT, signal_exit_now);
2526
2527         /* init EAL */
2528         ret = rte_eal_init(argc, argv);
2529         if (ret < 0)
2530                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2531         argc -= ret;
2532         argv += ret;
2533
2534         /* init RTE timer library to be used late */
2535         rte_timer_subsystem_init();
2536
2537         /* if we're running pmd-mgmt mode, don't default to baseline mode */
2538         baseline_enabled = false;
2539
2540         /* parse application arguments (after the EAL ones) */
2541         ret = parse_args(argc, argv);
2542         if (ret < 0)
2543                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
2544
2545         if (app_mode == APP_MODE_DEFAULT)
2546                 app_mode = autodetect_mode();
2547
2548         RTE_LOG(INFO, L3FWD_POWER, "Selected operation mode: %s\n",
2549                         mode_to_str(app_mode));
2550
2551         /* only legacy and empty poll mode rely on power library */
2552         if ((app_mode == APP_MODE_LEGACY || app_mode == APP_MODE_EMPTY_POLL) &&
2553                         init_power_library())
2554                 rte_exit(EXIT_FAILURE, "init_power_library failed\n");
2555
2556         if (update_lcore_params() < 0)
2557                 rte_exit(EXIT_FAILURE, "update_lcore_params failed\n");
2558
2559         if (check_lcore_params() < 0)
2560                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
2561
2562         ret = init_lcore_rx_queues();
2563         if (ret < 0)
2564                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2565
2566         nb_ports = rte_eth_dev_count_avail();
2567
2568         if (check_port_config() < 0)
2569                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
2570
2571         nb_lcores = rte_lcore_count();
2572
2573         /* initialize all ports */
2574         RTE_ETH_FOREACH_DEV(portid) {
2575                 struct rte_eth_conf local_port_conf = port_conf;
2576                 /* not all app modes need interrupts */
2577                 bool need_intr = app_mode == APP_MODE_LEGACY ||
2578                                 app_mode == APP_MODE_INTERRUPT;
2579
2580                 /* skip ports that are not enabled */
2581                 if ((enabled_port_mask & (1 << portid)) == 0) {
2582                         printf("\nSkipping disabled port %d\n", portid);
2583                         continue;
2584                 }
2585
2586                 /* init port */
2587                 printf("Initializing port %d ... ", portid );
2588                 fflush(stdout);
2589
2590                 ret = rte_eth_dev_info_get(portid, &dev_info);
2591                 if (ret != 0)
2592                         rte_exit(EXIT_FAILURE,
2593                                 "Error during getting device (port %u) info: %s\n",
2594                                 portid, strerror(-ret));
2595
2596                 dev_rxq_num = dev_info.max_rx_queues;
2597                 dev_txq_num = dev_info.max_tx_queues;
2598
2599                 nb_rx_queue = get_port_n_rx_queues(portid);
2600                 if (nb_rx_queue > dev_rxq_num)
2601                         rte_exit(EXIT_FAILURE,
2602                                 "Cannot configure not existed rxq: "
2603                                 "port=%d\n", portid);
2604
2605                 n_tx_queue = nb_lcores;
2606                 if (n_tx_queue > dev_txq_num)
2607                         n_tx_queue = dev_txq_num;
2608                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
2609                         nb_rx_queue, (unsigned)n_tx_queue );
2610                 /* If number of Rx queue is 0, no need to enable Rx interrupt */
2611                 if (nb_rx_queue == 0)
2612                         need_intr = false;
2613
2614                 if (need_intr)
2615                         local_port_conf.intr_conf.rxq = 1;
2616
2617                 ret = rte_eth_dev_info_get(portid, &dev_info);
2618                 if (ret != 0)
2619                         rte_exit(EXIT_FAILURE,
2620                                 "Error during getting device (port %u) info: %s\n",
2621                                 portid, strerror(-ret));
2622
2623                 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
2624                         local_port_conf.txmode.offloads |=
2625                                 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
2626
2627                 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2628                         dev_info.flow_type_rss_offloads;
2629                 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2630                                 port_conf.rx_adv_conf.rss_conf.rss_hf) {
2631                         printf("Port %u modified RSS hash function based on hardware support,"
2632                                 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2633                                 portid,
2634                                 port_conf.rx_adv_conf.rss_conf.rss_hf,
2635                                 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2636                 }
2637
2638                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
2639                                         (uint16_t)n_tx_queue, &local_port_conf);
2640                 if (ret < 0)
2641                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
2642                                         "err=%d, port=%d\n", ret, portid);
2643
2644                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
2645                                                        &nb_txd);
2646                 if (ret < 0)
2647                         rte_exit(EXIT_FAILURE,
2648                                  "Cannot adjust number of descriptors: err=%d, port=%d\n",
2649                                  ret, portid);
2650
2651                 ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
2652                 if (ret < 0)
2653                         rte_exit(EXIT_FAILURE,
2654                                  "Cannot get MAC address: err=%d, port=%d\n",
2655                                  ret, portid);
2656
2657                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
2658                 printf(", ");
2659
2660                 /* init memory */
2661                 ret = init_mem(NB_MBUF);
2662                 if (ret < 0)
2663                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
2664
2665                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2666                         if (rte_lcore_is_enabled(lcore_id) == 0)
2667                                 continue;
2668
2669                         /* Initialize TX buffers */
2670                         qconf = &lcore_conf[lcore_id];
2671                         qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
2672                                 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
2673                                 rte_eth_dev_socket_id(portid));
2674                         if (qconf->tx_buffer[portid] == NULL)
2675                                 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
2676                                                  portid);
2677
2678                         rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
2679                 }
2680
2681                 /* init one TX queue per couple (lcore,port) */
2682                 queueid = 0;
2683                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2684                         if (rte_lcore_is_enabled(lcore_id) == 0)
2685                                 continue;
2686
2687                         if (queueid >= dev_txq_num)
2688                                 continue;
2689
2690                         if (numa_on)
2691                                 socketid = \
2692                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
2693                         else
2694                                 socketid = 0;
2695
2696                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
2697                         fflush(stdout);
2698
2699                         txconf = &dev_info.default_txconf;
2700                         txconf->offloads = local_port_conf.txmode.offloads;
2701                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
2702                                                      socketid, txconf);
2703                         if (ret < 0)
2704                                 rte_exit(EXIT_FAILURE,
2705                                         "rte_eth_tx_queue_setup: err=%d, "
2706                                                 "port=%d\n", ret, portid);
2707
2708                         qconf = &lcore_conf[lcore_id];
2709                         qconf->tx_queue_id[portid] = queueid;
2710                         queueid++;
2711
2712                         qconf->tx_port_id[qconf->n_tx_port] = portid;
2713                         qconf->n_tx_port++;
2714                 }
2715                 printf("\n");
2716         }
2717
2718         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2719                 if (rte_lcore_is_enabled(lcore_id) == 0)
2720                         continue;
2721
2722                 if (app_mode == APP_MODE_LEGACY) {
2723                         /* init timer structures for each enabled lcore */
2724                         rte_timer_init(&power_timers[lcore_id]);
2725                         hz = rte_get_timer_hz();
2726                         rte_timer_reset(&power_timers[lcore_id],
2727                                         hz/TIMER_NUMBER_PER_SECOND,
2728                                         SINGLE, lcore_id,
2729                                         power_timer_cb, NULL);
2730                 }
2731                 qconf = &lcore_conf[lcore_id];
2732                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
2733                 fflush(stdout);
2734
2735                 /* init RX queues */
2736                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
2737                         struct rte_eth_rxconf rxq_conf;
2738
2739                         portid = qconf->rx_queue_list[queue].port_id;
2740                         queueid = qconf->rx_queue_list[queue].queue_id;
2741
2742                         if (numa_on)
2743                                 socketid = \
2744                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
2745                         else
2746                                 socketid = 0;
2747
2748                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
2749                         fflush(stdout);
2750
2751                         ret = rte_eth_dev_info_get(portid, &dev_info);
2752                         if (ret != 0)
2753                                 rte_exit(EXIT_FAILURE,
2754                                         "Error during getting device (port %u) info: %s\n",
2755                                         portid, strerror(-ret));
2756
2757                         rxq_conf = dev_info.default_rxconf;
2758                         rxq_conf.offloads = port_conf.rxmode.offloads;
2759                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
2760                                 socketid, &rxq_conf,
2761                                 pktmbuf_pool[socketid]);
2762                         if (ret < 0)
2763                                 rte_exit(EXIT_FAILURE,
2764                                         "rte_eth_rx_queue_setup: err=%d, "
2765                                                 "port=%d\n", ret, portid);
2766
2767                         if (parse_ptype) {
2768                                 if (add_cb_parse_ptype(portid, queueid) < 0)
2769                                         rte_exit(EXIT_FAILURE,
2770                                                  "Fail to add ptype cb\n");
2771                         }
2772
2773                         if (app_mode == APP_MODE_PMD_MGMT && !baseline_enabled) {
2774                                 ret = rte_power_ethdev_pmgmt_queue_enable(
2775                                                 lcore_id, portid, queueid,
2776                                                 pmgmt_type);
2777                                 if (ret < 0)
2778                                         rte_exit(EXIT_FAILURE,
2779                                                 "rte_power_ethdev_pmgmt_queue_enable: err=%d, port=%d\n",
2780                                                         ret, portid);
2781                         }
2782                 }
2783         }
2784
2785         printf("\n");
2786
2787         /* start ports */
2788         RTE_ETH_FOREACH_DEV(portid) {
2789                 if ((enabled_port_mask & (1 << portid)) == 0) {
2790                         continue;
2791                 }
2792                 /* Start device */
2793                 ret = rte_eth_dev_start(portid);
2794                 if (ret < 0)
2795                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
2796                                                 "port=%d\n", ret, portid);
2797                 /*
2798                  * If enabled, put device in promiscuous mode.
2799                  * This allows IO forwarding mode to forward packets
2800                  * to itself through 2 cross-connected  ports of the
2801                  * target machine.
2802                  */
2803                 if (promiscuous_on) {
2804                         ret = rte_eth_promiscuous_enable(portid);
2805                         if (ret != 0)
2806                                 rte_exit(EXIT_FAILURE,
2807                                         "rte_eth_promiscuous_enable: err=%s, port=%u\n",
2808                                         rte_strerror(-ret), portid);
2809                 }
2810                 /* initialize spinlock for each port */
2811                 rte_spinlock_init(&(locks[portid]));
2812
2813                 if (!parse_ptype)
2814                         if (!check_ptype(portid))
2815                                 rte_exit(EXIT_FAILURE,
2816                                         "PMD can not provide needed ptypes\n");
2817         }
2818
2819         check_all_ports_link_status(enabled_port_mask);
2820
2821         if (app_mode == APP_MODE_EMPTY_POLL) {
2822
2823                 if (empty_poll_train) {
2824                         policy.state = TRAINING;
2825                 } else {
2826                         policy.state = MED_NORMAL;
2827                         policy.med_base_edpi = ep_med_edpi;
2828                         policy.hgh_base_edpi = ep_hgh_edpi;
2829                 }
2830
2831                 ret = rte_power_empty_poll_stat_init(&ep_params,
2832                                 freq_tlb,
2833                                 &policy);
2834                 if (ret < 0)
2835                         rte_exit(EXIT_FAILURE, "empty poll init failed");
2836         }
2837
2838
2839         /* launch per-lcore init on every lcore */
2840         if (app_mode == APP_MODE_LEGACY) {
2841                 rte_eal_mp_remote_launch(main_legacy_loop, NULL, CALL_MAIN);
2842         } else if (app_mode == APP_MODE_EMPTY_POLL) {
2843                 empty_poll_stop = false;
2844                 rte_eal_mp_remote_launch(main_empty_poll_loop, NULL,
2845                                 SKIP_MAIN);
2846         } else if (app_mode == APP_MODE_TELEMETRY) {
2847                 unsigned int i;
2848
2849                 /* Init metrics library */
2850                 rte_metrics_init(rte_socket_id());
2851                 /** Register stats with metrics library */
2852                 for (i = 0; i < NUM_TELSTATS; i++)
2853                         ptr_strings[i] = telstats_strings[i].name;
2854
2855                 ret = rte_metrics_reg_names(ptr_strings, NUM_TELSTATS);
2856                 if (ret >= 0)
2857                         telstats_index = ret;
2858                 else
2859                         rte_exit(EXIT_FAILURE, "failed to register metrics names");
2860
2861                 RTE_LCORE_FOREACH_WORKER(lcore_id) {
2862                         rte_spinlock_init(&stats[lcore_id].telemetry_lock);
2863                 }
2864                 rte_timer_init(&telemetry_timer);
2865                 rte_telemetry_register_cmd("/l3fwd-power/stats",
2866                                 handle_app_stats,
2867                                 "Returns global power stats. Parameters: None");
2868                 rte_eal_mp_remote_launch(main_telemetry_loop, NULL,
2869                                                 SKIP_MAIN);
2870         } else if (app_mode == APP_MODE_INTERRUPT) {
2871                 rte_eal_mp_remote_launch(main_intr_loop, NULL, CALL_MAIN);
2872         } else if (app_mode == APP_MODE_PMD_MGMT) {
2873                 /* reuse telemetry loop for PMD power management mode */
2874                 rte_eal_mp_remote_launch(main_telemetry_loop, NULL, CALL_MAIN);
2875         }
2876
2877         if (app_mode == APP_MODE_EMPTY_POLL || app_mode == APP_MODE_TELEMETRY)
2878                 launch_timer(rte_lcore_id());
2879
2880         RTE_LCORE_FOREACH_WORKER(lcore_id) {
2881                 if (rte_eal_wait_lcore(lcore_id) < 0)
2882                         return -1;
2883         }
2884
2885         if (app_mode == APP_MODE_PMD_MGMT) {
2886                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2887                         if (rte_lcore_is_enabled(lcore_id) == 0)
2888                                 continue;
2889                         qconf = &lcore_conf[lcore_id];
2890                         for (queue = 0; queue < qconf->n_rx_queue; ++queue) {
2891                                 portid = qconf->rx_queue_list[queue].port_id;
2892                                 queueid = qconf->rx_queue_list[queue].queue_id;
2893
2894                                 rte_power_ethdev_pmgmt_queue_disable(lcore_id,
2895                                                 portid, queueid);
2896                         }
2897                 }
2898         }
2899
2900         RTE_ETH_FOREACH_DEV(portid)
2901         {
2902                 if ((enabled_port_mask & (1 << portid)) == 0)
2903                         continue;
2904
2905                 ret = rte_eth_dev_stop(portid);
2906                 if (ret != 0)
2907                         RTE_LOG(ERR, L3FWD_POWER, "rte_eth_dev_stop: err=%d, port=%u\n",
2908                                 ret, portid);
2909
2910                 rte_eth_dev_close(portid);
2911         }
2912
2913         if (app_mode == APP_MODE_EMPTY_POLL)
2914                 rte_power_empty_poll_stat_free();
2915
2916         if ((app_mode == APP_MODE_LEGACY || app_mode == APP_MODE_EMPTY_POLL) &&
2917                         deinit_power_library())
2918                 rte_exit(EXIT_FAILURE, "deinit_power_library failed\n");
2919
2920         if (rte_eal_cleanup() < 0)
2921                 RTE_LOG(ERR, L3FWD_POWER, "EAL cleanup failed\n");
2922
2923         return 0;
2924 }