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
  18 #include <rte_common.h>
  19 #include <rte_byteorder.h>
  20 #include <rte_log.h>
  21 #include <rte_malloc.h>
  22 #include <rte_memory.h>
  23 #include <rte_memcpy.h>
  24 #include <rte_eal.h>
  25 #include <rte_launch.h>
  26 #include <rte_atomic.h>
  27 #include <rte_cycles.h>
  28 #include <rte_prefetch.h>
  29 #include <rte_lcore.h>
  30 #include <rte_per_lcore.h>
  31 #include <rte_branch_prediction.h>
  32 #include <rte_interrupts.h>
  33 #include <rte_random.h>
  34 #include <rte_debug.h>
  35 #include <rte_ether.h>
  36 #include <rte_ethdev.h>
  37 #include <rte_mempool.h>
  38 #include <rte_mbuf.h>
  39 #include <rte_ip.h>
  40 #include <rte_tcp.h>
  41 #include <rte_udp.h>
  42 #include <rte_string_fns.h>
  43 #include <rte_timer.h>
  44 #include <rte_power.h>
  45 #include <rte_spinlock.h>
  46 #include <rte_power_empty_poll.h>
  47
  48 #include "perf_core.h"
  49 #include "main.h"
  50
  51 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
  52
  53 #define MAX_PKT_BURST 32
  54
  55 #define MIN_ZERO_POLL_COUNT 10
  56
  57 /* 100 ms interval */
  58 #define TIMER_NUMBER_PER_SECOND           10
  59 /* (10ms) */
  60 #define INTERVALS_PER_SECOND             100
  61 /* 100000 us */
  62 #define SCALING_PERIOD                    (1000000/TIMER_NUMBER_PER_SECOND)
  63 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
  64
  65 #define APP_LOOKUP_EXACT_MATCH          0
  66 #define APP_LOOKUP_LPM                  1
  67 #define DO_RFC_1812_CHECKS
  68
  69 #ifndef APP_LOOKUP_METHOD
  70 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  71 #endif
  72
  73 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  74 #include <rte_hash.h>
  75 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
  76 #include <rte_lpm.h>
  77 #else
  78 #error "APP_LOOKUP_METHOD set to incorrect value"
  79 #endif
  80
  81 #ifndef IPv6_BYTES
  82 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
  83                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
  84 #define IPv6_BYTES(addr) \
  85         addr[0],  addr[1], addr[2],  addr[3], \
  86         addr[4],  addr[5], addr[6],  addr[7], \
  87         addr[8],  addr[9], addr[10], addr[11],\
  88         addr[12], addr[13],addr[14], addr[15]
  89 #endif
  90
  91 #define MAX_JUMBO_PKT_LEN  9600
  92
  93 #define IPV6_ADDR_LEN 16
  94
  95 #define MEMPOOL_CACHE_SIZE 256
  96
  97 /*
  98  * This expression is used to calculate the number of mbufs needed depending on
  99  * user input, taking into account memory for rx and tx hardware rings, cache
 100  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 101  * NB_MBUF never goes below a minimum value of 8192.
 102  */
 103
 104 #define NB_MBUF RTE_MAX ( \
 105         (nb_ports*nb_rx_queue*nb_rxd + \
 106         nb_ports*nb_lcores*MAX_PKT_BURST + \
 107         nb_ports*n_tx_queue*nb_txd + \
 108         nb_lcores*MEMPOOL_CACHE_SIZE), \
 109         (unsigned)8192)
 110
 111 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 112
 113 #define NB_SOCKETS 8
 114
 115 /* Configure how many packets ahead to prefetch, when reading packets */
 116 #define PREFETCH_OFFSET 3
 117
 118 /*
 119  * Configurable number of RX/TX ring descriptors
 120  */
 121 #define RTE_TEST_RX_DESC_DEFAULT 1024
 122 #define RTE_TEST_TX_DESC_DEFAULT 1024
 123
 124 /*
 125  * These two thresholds were decided on by running the training algorithm on
 126  * a 2.5GHz Xeon. These defaults can be overridden by supplying non-zero values
 127  * for the med_threshold and high_threshold parameters on the command line.
 128  */
 129 #define EMPTY_POLL_MED_THRESHOLD 350000UL
 130 #define EMPTY_POLL_HGH_THRESHOLD 580000UL
 131
 132
 133
 134 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 135 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 136
 137 /* ethernet addresses of ports */
 138 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 139
 140 /* ethernet addresses of ports */
 141 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
 142
 143 /* mask of enabled ports */
 144 static uint32_t enabled_port_mask = 0;
 145 /* Ports set in promiscuous mode off by default. */
 146 static int promiscuous_on = 0;
 147 /* NUMA is enabled by default. */
 148 static int numa_on = 1;
 149 /* emptypoll is disabled by default. */
 150 static bool empty_poll_on;
 151 static bool empty_poll_train;
 152 volatile bool empty_poll_stop;
 153 static struct  ep_params *ep_params;
 154 static struct  ep_policy policy;
 155 static long  ep_med_edpi, ep_hgh_edpi;
 156
 157 static int parse_ptype; /**< Parse packet type using rx callback, and */
 158                         /**< disabled by default */
 159
 160 enum freq_scale_hint_t
 161 {
 162         FREQ_LOWER    =      -1,
 163         FREQ_CURRENT  =       0,
 164         FREQ_HIGHER   =       1,
 165         FREQ_HIGHEST  =       2
 166 };
 167
 168 struct lcore_rx_queue {
 169         uint16_t port_id;
 170         uint8_t queue_id;
 171         enum freq_scale_hint_t freq_up_hint;
 172         uint32_t zero_rx_packet_count;
 173         uint32_t idle_hint;
 174 } __rte_cache_aligned;
 175
 176 #define MAX_RX_QUEUE_PER_LCORE 16
 177 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 178 #define MAX_RX_QUEUE_PER_PORT 128
 179
 180 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
 181
 182
 183 struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 184 static struct lcore_params lcore_params_array_default[] = {
 185         {0, 0, 2},
 186         {0, 1, 2},
 187         {0, 2, 2},
 188         {1, 0, 2},
 189         {1, 1, 2},
 190         {1, 2, 2},
 191         {2, 0, 2},
 192         {3, 0, 3},
 193         {3, 1, 3},
 194 };
 195
 196 struct lcore_params *lcore_params = lcore_params_array_default;
 197 uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 198                                 sizeof(lcore_params_array_default[0]);
 199
 200 static struct rte_eth_conf port_conf = {
 201         .rxmode = {
 202                 .mq_mode        = ETH_MQ_RX_RSS,
 203                 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
 204                 .split_hdr_size = 0,
 205                 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
 206         },
 207         .rx_adv_conf = {
 208                 .rss_conf = {
 209                         .rss_key = NULL,
 210                         .rss_hf = ETH_RSS_UDP,
 211                 },
 212         },
 213         .txmode = {
 214                 .mq_mode = ETH_MQ_TX_NONE,
 215         },
 216         .intr_conf = {
 217                 .rxq = 1,
 218         },
 219 };
 220
 221 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 222
 223
 224 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 225
 226 #ifdef RTE_ARCH_X86
 227 #include <rte_hash_crc.h>
 228 #define DEFAULT_HASH_FUNC       rte_hash_crc
 229 #else
 230 #include <rte_jhash.h>
 231 #define DEFAULT_HASH_FUNC       rte_jhash
 232 #endif
 233
 234 struct ipv4_5tuple {
 235         uint32_t ip_dst;
 236         uint32_t ip_src;
 237         uint16_t port_dst;
 238         uint16_t port_src;
 239         uint8_t  proto;
 240 } __attribute__((__packed__));
 241
 242 struct ipv6_5tuple {
 243         uint8_t  ip_dst[IPV6_ADDR_LEN];
 244         uint8_t  ip_src[IPV6_ADDR_LEN];
 245         uint16_t port_dst;
 246         uint16_t port_src;
 247         uint8_t  proto;
 248 } __attribute__((__packed__));
 249
 250 struct ipv4_l3fwd_route {
 251         struct ipv4_5tuple key;
 252         uint8_t if_out;
 253 };
 254
 255 struct ipv6_l3fwd_route {
 256         struct ipv6_5tuple key;
 257         uint8_t if_out;
 258 };
 259
 260 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 261         {{RTE_IPv4(100,10,0,1), RTE_IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 262         {{RTE_IPv4(100,20,0,2), RTE_IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 263         {{RTE_IPv4(100,30,0,3), RTE_IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 264         {{RTE_IPv4(100,40,0,4), RTE_IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 265 };
 266
 267 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 268         {
 269                 {
 270                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 271                          0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 272                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 273                          0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
 274                          1, 10, IPPROTO_UDP
 275                 }, 4
 276         },
 277 };
 278
 279 typedef struct rte_hash lookup_struct_t;
 280 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 281 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 282
 283 #define L3FWD_HASH_ENTRIES      1024
 284
 285 #define IPV4_L3FWD_NUM_ROUTES \
 286         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 287
 288 #define IPV6_L3FWD_NUM_ROUTES \
 289         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 290
 291 static uint16_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 292 static uint16_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 293 #endif
 294
 295 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 296 struct ipv4_l3fwd_route {
 297         uint32_t ip;
 298         uint8_t  depth;
 299         uint8_t  if_out;
 300 };
 301
 302 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 303         {RTE_IPv4(1,1,1,0), 24, 0},
 304         {RTE_IPv4(2,1,1,0), 24, 1},
 305         {RTE_IPv4(3,1,1,0), 24, 2},
 306         {RTE_IPv4(4,1,1,0), 24, 3},
 307         {RTE_IPv4(5,1,1,0), 24, 4},
 308         {RTE_IPv4(6,1,1,0), 24, 5},
 309         {RTE_IPv4(7,1,1,0), 24, 6},
 310         {RTE_IPv4(8,1,1,0), 24, 7},
 311 };
 312
 313 #define IPV4_L3FWD_NUM_ROUTES \
 314         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 315
 316 #define IPV4_L3FWD_LPM_MAX_RULES     1024
 317
 318 typedef struct rte_lpm lookup_struct_t;
 319 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 320 #endif
 321
 322 struct lcore_conf {
 323         uint16_t n_rx_queue;
 324         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 325         uint16_t n_tx_port;
 326         uint16_t tx_port_id[RTE_MAX_ETHPORTS];
 327         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 328         struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
 329         lookup_struct_t * ipv4_lookup_struct;
 330         lookup_struct_t * ipv6_lookup_struct;
 331 } __rte_cache_aligned;
 332
 333 struct lcore_stats {
 334         /* total sleep time in ms since last frequency scaling down */
 335         uint32_t sleep_time;
 336         /* number of long sleep recently */
 337         uint32_t nb_long_sleep;
 338         /* freq. scaling up trend */
 339         uint32_t trend;
 340         /* total packet processed recently */
 341         uint64_t nb_rx_processed;
 342         /* total iterations looped recently */
 343         uint64_t nb_iteration_looped;
 344         uint32_t padding[9];
 345 } __rte_cache_aligned;
 346
 347 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
 348 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
 349 static struct rte_timer power_timers[RTE_MAX_LCORE];
 350
 351 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
 352 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
 353                 unsigned int lcore_id, uint16_t port_id, uint16_t queue_id);
 354
 355
 356 /*
 357  * These defaults are using the max frequency index (1), a medium index (9)
 358  * and a typical low frequency index (14). These can be adjusted to use
 359  * different indexes using the relevant command line parameters.
 360  */
 361 static uint8_t  freq_tlb[] = {14, 9, 1};
 362
 363 static int is_done(void)
 364 {
 365         return empty_poll_stop;
 366 }
 367
 368 /* exit signal handler */
 369 static void
 370 signal_exit_now(int sigtype)
 371 {
 372         unsigned lcore_id;
 373         unsigned int portid;
 374         int ret;
 375
 376         if (sigtype == SIGINT) {
 377                 if (empty_poll_on)
 378                         empty_poll_stop = true;
 379
 380
 381                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 382                         if (rte_lcore_is_enabled(lcore_id) == 0)
 383                                 continue;
 384
 385                         /* init power management library */
 386                         ret = rte_power_exit(lcore_id);
 387                         if (ret)
 388                                 rte_exit(EXIT_FAILURE, "Power management "
 389                                         "library de-initialization failed on "
 390                                                         "core%u\n", lcore_id);
 391                 }
 392
 393                 if (!empty_poll_on) {
 394                         RTE_ETH_FOREACH_DEV(portid) {
 395                                 if ((enabled_port_mask & (1 << portid)) == 0)
 396                                         continue;
 397
 398                                 rte_eth_dev_stop(portid);
 399                                 rte_eth_dev_close(portid);
 400                         }
 401                 }
 402         }
 403
 404         if (!empty_poll_on)
 405                 rte_exit(EXIT_SUCCESS, "User forced exit\n");
 406 }
 407
 408 /*  Freqency scale down timer callback */
 409 static void
 410 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
 411                           __attribute__((unused)) void *arg)
 412 {
 413         uint64_t hz;
 414         float sleep_time_ratio;
 415         unsigned lcore_id = rte_lcore_id();
 416
 417         /* accumulate total execution time in us when callback is invoked */
 418         sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
 419                                         (float)SCALING_PERIOD;
 420         /**
 421          * check whether need to scale down frequency a step if it sleep a lot.
 422          */
 423         if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
 424                 if (rte_power_freq_down)
 425                         rte_power_freq_down(lcore_id);
 426         }
 427         else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
 428                 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
 429                 /**
 430                  * scale down a step if average packet per iteration less
 431                  * than expectation.
 432                  */
 433                 if (rte_power_freq_down)
 434                         rte_power_freq_down(lcore_id);
 435         }
 436
 437         /**
 438          * initialize another timer according to current frequency to ensure
 439          * timer interval is relatively fixed.
 440          */
 441         hz = rte_get_timer_hz();
 442         rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
 443                                 SINGLE, lcore_id, power_timer_cb, NULL);
 444
 445         stats[lcore_id].nb_rx_processed = 0;
 446         stats[lcore_id].nb_iteration_looped = 0;
 447
 448         stats[lcore_id].sleep_time = 0;
 449 }
 450
 451 /* Enqueue a single packet, and send burst if queue is filled */
 452 static inline int
 453 send_single_packet(struct rte_mbuf *m, uint16_t port)
 454 {
 455         uint32_t lcore_id;
 456         struct lcore_conf *qconf;
 457
 458         lcore_id = rte_lcore_id();
 459         qconf = &lcore_conf[lcore_id];
 460
 461         rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
 462                         qconf->tx_buffer[port], m);
 463
 464         return 0;
 465 }
 466
 467 #ifdef DO_RFC_1812_CHECKS
 468 static inline int
 469 is_valid_ipv4_pkt(struct rte_ipv4_hdr *pkt, uint32_t link_len)
 470 {
 471         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 472         /*
 473          * 1. The packet length reported by the Link Layer must be large
 474          * enough to hold the minimum length legal IP datagram (20 bytes).
 475          */
 476         if (link_len < sizeof(struct rte_ipv4_hdr))
 477                 return -1;
 478
 479         /* 2. The IP checksum must be correct. */
 480         /* this is checked in H/W */
 481
 482         /*
 483          * 3. The IP version number must be 4. If the version number is not 4
 484          * then the packet may be another version of IP, such as IPng or
 485          * ST-II.
 486          */
 487         if (((pkt->version_ihl) >> 4) != 4)
 488                 return -3;
 489         /*
 490          * 4. The IP header length field must be large enough to hold the
 491          * minimum length legal IP datagram (20 bytes = 5 words).
 492          */
 493         if ((pkt->version_ihl & 0xf) < 5)
 494                 return -4;
 495
 496         /*
 497          * 5. The IP total length field must be large enough to hold the IP
 498          * datagram header, whose length is specified in the IP header length
 499          * field.
 500          */
 501         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct rte_ipv4_hdr))
 502                 return -5;
 503
 504         return 0;
 505 }
 506 #endif
 507
 508 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 509 static void
 510 print_ipv4_key(struct ipv4_5tuple key)
 511 {
 512         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
 513                 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
 514                                 key.port_dst, key.port_src, key.proto);
 515 }
 516 static void
 517 print_ipv6_key(struct ipv6_5tuple key)
 518 {
 519         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 520                 "port dst = %d, port src = %d, proto = %d\n",
 521                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 522                 key.port_dst, key.port_src, key.proto);
 523 }
 524
 525 static inline uint16_t
 526 get_ipv4_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
 527                 lookup_struct_t * ipv4_l3fwd_lookup_struct)
 528 {
 529         struct ipv4_5tuple key;
 530         struct tcp_hdr *tcp;
 531         struct udp_hdr *udp;
 532         int ret = 0;
 533
 534         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 535         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 536         key.proto = ipv4_hdr->next_proto_id;
 537
 538         switch (ipv4_hdr->next_proto_id) {
 539         case IPPROTO_TCP:
 540                 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
 541                                         sizeof(struct rte_ipv4_hdr));
 542                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 543                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 544                 break;
 545
 546         case IPPROTO_UDP:
 547                 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
 548                                         sizeof(struct rte_ipv4_hdr));
 549                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 550                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 551                 break;
 552
 553         default:
 554                 key.port_dst = 0;
 555                 key.port_src = 0;
 556                 break;
 557         }
 558
 559         /* Find destination port */
 560         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 561         return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
 562 }
 563
 564 static inline uint16_t
 565 get_ipv6_dst_port(struct rte_ipv6_hdr *ipv6_hdr, uint16_t portid,
 566                         lookup_struct_t *ipv6_l3fwd_lookup_struct)
 567 {
 568         struct ipv6_5tuple key;
 569         struct tcp_hdr *tcp;
 570         struct udp_hdr *udp;
 571         int ret = 0;
 572
 573         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 574         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 575
 576         key.proto = ipv6_hdr->proto;
 577
 578         switch (ipv6_hdr->proto) {
 579         case IPPROTO_TCP:
 580                 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
 581                                         sizeof(struct rte_ipv6_hdr));
 582                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 583                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 584                 break;
 585
 586         case IPPROTO_UDP:
 587                 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
 588                                         sizeof(struct rte_ipv6_hdr));
 589                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 590                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 591                 break;
 592
 593         default:
 594                 key.port_dst = 0;
 595                 key.port_src = 0;
 596                 break;
 597         }
 598
 599         /* Find destination port */
 600         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 601         return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
 602 }
 603 #endif
 604
 605 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 606 static inline uint16_t
 607 get_ipv4_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
 608                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 609 {
 610         uint32_t next_hop;
 611
 612         return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 613                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 614                         next_hop : portid);
 615 }
 616 #endif
 617
 618 static inline void
 619 parse_ptype_one(struct rte_mbuf *m)
 620 {
 621         struct rte_ether_hdr *eth_hdr;
 622         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 623         uint16_t ether_type;
 624
 625         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 626         ether_type = eth_hdr->ether_type;
 627         if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPv4))
 628                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 629         else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPv6))
 630                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 631
 632         m->packet_type = packet_type;
 633 }
 634
 635 static uint16_t
 636 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
 637                struct rte_mbuf *pkts[], uint16_t nb_pkts,
 638                uint16_t max_pkts __rte_unused,
 639                void *user_param __rte_unused)
 640 {
 641         unsigned int i;
 642
 643         for (i = 0; i < nb_pkts; ++i)
 644                 parse_ptype_one(pkts[i]);
 645
 646         return nb_pkts;
 647 }
 648
 649 static int
 650 add_cb_parse_ptype(uint16_t portid, uint16_t queueid)
 651 {
 652         printf("Port %d: softly parse packet type info\n", portid);
 653         if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL))
 654                 return 0;
 655
 656         printf("Failed to add rx callback: port=%d\n", portid);
 657         return -1;
 658 }
 659
 660 static inline void
 661 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
 662                                 struct lcore_conf *qconf)
 663 {
 664         struct rte_ether_hdr *eth_hdr;
 665         struct rte_ipv4_hdr *ipv4_hdr;
 666         void *d_addr_bytes;
 667         uint16_t dst_port;
 668
 669         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 670
 671         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 672                 /* Handle IPv4 headers.*/
 673                 ipv4_hdr =
 674                         rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
 675                                                 sizeof(struct rte_ether_hdr));
 676
 677 #ifdef DO_RFC_1812_CHECKS
 678                 /* Check to make sure the packet is valid (RFC1812) */
 679                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 680                         rte_pktmbuf_free(m);
 681                         return;
 682                 }
 683 #endif
 684
 685                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 686                                         qconf->ipv4_lookup_struct);
 687                 if (dst_port >= RTE_MAX_ETHPORTS ||
 688                                 (enabled_port_mask & 1 << dst_port) == 0)
 689                         dst_port = portid;
 690
 691                 /* 02:00:00:00:00:xx */
 692                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 693                 *((uint64_t *)d_addr_bytes) =
 694                         0x000000000002 + ((uint64_t)dst_port << 40);
 695
 696 #ifdef DO_RFC_1812_CHECKS
 697                 /* Update time to live and header checksum */
 698                 --(ipv4_hdr->time_to_live);
 699                 ++(ipv4_hdr->hdr_checksum);
 700 #endif
 701
 702                 /* src addr */
 703                 rte_ether_addr_copy(&ports_eth_addr[dst_port],
 704                                 &eth_hdr->s_addr);
 705
 706                 send_single_packet(m, dst_port);
 707         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 708                 /* Handle IPv6 headers.*/
 709 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 710                 struct rte_ipv6_hdr *ipv6_hdr;
 711
 712                 ipv6_hdr =
 713                         rte_pktmbuf_mtod_offset(m, struct rte_ipv6_hdr *,
 714                                                 sizeof(struct rte_ether_hdr));
 715
 716                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
 717                                         qconf->ipv6_lookup_struct);
 718
 719                 if (dst_port >= RTE_MAX_ETHPORTS ||
 720                                 (enabled_port_mask & 1 << dst_port) == 0)
 721                         dst_port = portid;
 722
 723                 /* 02:00:00:00:00:xx */
 724                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 725                 *((uint64_t *)d_addr_bytes) =
 726                         0x000000000002 + ((uint64_t)dst_port << 40);
 727
 728                 /* src addr */
 729                 rte_ether_addr_copy(&ports_eth_addr[dst_port],
 730                                 &eth_hdr->s_addr);
 731
 732                 send_single_packet(m, dst_port);
 733 #else
 734                 /* We don't currently handle IPv6 packets in LPM mode. */
 735                 rte_pktmbuf_free(m);
 736 #endif
 737         } else
 738                 rte_pktmbuf_free(m);
 739
 740 }
 741
 742 #define MINIMUM_SLEEP_TIME         1
 743 #define SUSPEND_THRESHOLD          300
 744
 745 static inline uint32_t
 746 power_idle_heuristic(uint32_t zero_rx_packet_count)
 747 {
 748         /* If zero count is less than 100,  sleep 1us */
 749         if (zero_rx_packet_count < SUSPEND_THRESHOLD)
 750                 return MINIMUM_SLEEP_TIME;
 751         /* If zero count is less than 1000, sleep 100 us which is the
 752                 minimum latency switching from C3/C6 to C0
 753         */
 754         else
 755                 return SUSPEND_THRESHOLD;
 756 }
 757
 758 static inline enum freq_scale_hint_t
 759 power_freq_scaleup_heuristic(unsigned lcore_id,
 760                              uint16_t port_id,
 761                              uint16_t queue_id)
 762 {
 763         uint32_t rxq_count = rte_eth_rx_queue_count(port_id, queue_id);
 764 /**
 765  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
 766  * per iteration
 767  */
 768 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
 769 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
 770 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
 771 #define FREQ_UP_TREND1_ACC   1
 772 #define FREQ_UP_TREND2_ACC   100
 773 #define FREQ_UP_THRESHOLD    10000
 774
 775         if (likely(rxq_count > FREQ_GEAR3_RX_PACKET_THRESHOLD)) {
 776                 stats[lcore_id].trend = 0;
 777                 return FREQ_HIGHEST;
 778         } else if (likely(rxq_count > FREQ_GEAR2_RX_PACKET_THRESHOLD))
 779                 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
 780         else if (likely(rxq_count > FREQ_GEAR1_RX_PACKET_THRESHOLD))
 781                 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
 782
 783         if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
 784                 stats[lcore_id].trend = 0;
 785                 return FREQ_HIGHER;
 786         }
 787
 788         return FREQ_CURRENT;
 789 }
 790
 791 /**
 792  * force polling thread sleep until one-shot rx interrupt triggers
 793  * @param port_id
 794  *  Port id.
 795  * @param queue_id
 796  *  Rx queue id.
 797  * @return
 798  *  0 on success
 799  */
 800 static int
 801 sleep_until_rx_interrupt(int num)
 802 {
 803         struct rte_epoll_event event[num];
 804         int n, i;
 805         uint16_t port_id;
 806         uint8_t queue_id;
 807         void *data;
 808
 809         RTE_LOG(INFO, L3FWD_POWER,
 810                 "lcore %u sleeps until interrupt triggers\n",
 811                 rte_lcore_id());
 812
 813         n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
 814         for (i = 0; i < n; i++) {
 815                 data = event[i].epdata.data;
 816                 port_id = ((uintptr_t)data) >> CHAR_BIT;
 817                 queue_id = ((uintptr_t)data) &
 818                         RTE_LEN2MASK(CHAR_BIT, uint8_t);
 819                 rte_eth_dev_rx_intr_disable(port_id, queue_id);
 820                 RTE_LOG(INFO, L3FWD_POWER,
 821                         "lcore %u is waked up from rx interrupt on"
 822                         " port %d queue %d\n",
 823                         rte_lcore_id(), port_id, queue_id);
 824         }
 825
 826         return 0;
 827 }
 828
 829 static void turn_on_intr(struct lcore_conf *qconf)
 830 {
 831         int i;
 832         struct lcore_rx_queue *rx_queue;
 833         uint8_t queue_id;
 834         uint16_t port_id;
 835
 836         for (i = 0; i < qconf->n_rx_queue; ++i) {
 837                 rx_queue = &(qconf->rx_queue_list[i]);
 838                 port_id = rx_queue->port_id;
 839                 queue_id = rx_queue->queue_id;
 840
 841                 rte_spinlock_lock(&(locks[port_id]));
 842                 rte_eth_dev_rx_intr_enable(port_id, queue_id);
 843                 rte_spinlock_unlock(&(locks[port_id]));
 844         }
 845 }
 846
 847 static int event_register(struct lcore_conf *qconf)
 848 {
 849         struct lcore_rx_queue *rx_queue;
 850         uint8_t queueid;
 851         uint16_t portid;
 852         uint32_t data;
 853         int ret;
 854         int i;
 855
 856         for (i = 0; i < qconf->n_rx_queue; ++i) {
 857                 rx_queue = &(qconf->rx_queue_list[i]);
 858                 portid = rx_queue->port_id;
 859                 queueid = rx_queue->queue_id;
 860                 data = portid << CHAR_BIT | queueid;
 861
 862                 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
 863                                                 RTE_EPOLL_PER_THREAD,
 864                                                 RTE_INTR_EVENT_ADD,
 865                                                 (void *)((uintptr_t)data));
 866                 if (ret)
 867                         return ret;
 868         }
 869
 870         return 0;
 871 }
 872 /* main processing loop */
 873 static int
 874 main_empty_poll_loop(__attribute__((unused)) void *dummy)
 875 {
 876         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 877         unsigned int lcore_id;
 878         uint64_t prev_tsc, diff_tsc, cur_tsc;
 879         int i, j, nb_rx;
 880         uint8_t queueid;
 881         uint16_t portid;
 882         struct lcore_conf *qconf;
 883         struct lcore_rx_queue *rx_queue;
 884
 885         const uint64_t drain_tsc =
 886                 (rte_get_tsc_hz() + US_PER_S - 1) /
 887                 US_PER_S * BURST_TX_DRAIN_US;
 888
 889         prev_tsc = 0;
 890
 891         lcore_id = rte_lcore_id();
 892         qconf = &lcore_conf[lcore_id];
 893
 894         if (qconf->n_rx_queue == 0) {
 895                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
 896                         lcore_id);
 897                 return 0;
 898         }
 899
 900         for (i = 0; i < qconf->n_rx_queue; i++) {
 901                 portid = qconf->rx_queue_list[i].port_id;
 902                 queueid = qconf->rx_queue_list[i].queue_id;
 903                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
 904                                 "rxqueueid=%hhu\n", lcore_id, portid, queueid);
 905         }
 906
 907         while (!is_done()) {
 908                 stats[lcore_id].nb_iteration_looped++;
 909
 910                 cur_tsc = rte_rdtsc();
 911                 /*
 912                  * TX burst queue drain
 913                  */
 914                 diff_tsc = cur_tsc - prev_tsc;
 915                 if (unlikely(diff_tsc > drain_tsc)) {
 916                         for (i = 0; i < qconf->n_tx_port; ++i) {
 917                                 portid = qconf->tx_port_id[i];
 918                                 rte_eth_tx_buffer_flush(portid,
 919                                                 qconf->tx_queue_id[portid],
 920                                                 qconf->tx_buffer[portid]);
 921                         }
 922                         prev_tsc = cur_tsc;
 923                 }
 924
 925                 /*
 926                  * Read packet from RX queues
 927                  */
 928                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 929                         rx_queue = &(qconf->rx_queue_list[i]);
 930                         rx_queue->idle_hint = 0;
 931                         portid = rx_queue->port_id;
 932                         queueid = rx_queue->queue_id;
 933
 934                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 935                                         MAX_PKT_BURST);
 936
 937                         stats[lcore_id].nb_rx_processed += nb_rx;
 938
 939                         if (nb_rx == 0) {
 940
 941                                 rte_power_empty_poll_stat_update(lcore_id);
 942
 943                                 continue;
 944                         } else {
 945                                 rte_power_poll_stat_update(lcore_id, nb_rx);
 946                         }
 947
 948
 949                         /* Prefetch first packets */
 950                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 951                                 rte_prefetch0(rte_pktmbuf_mtod(
 952                                                         pkts_burst[j], void *));
 953                         }
 954
 955                         /* Prefetch and forward already prefetched packets */
 956                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 957                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 958                                                         j + PREFETCH_OFFSET],
 959                                                         void *));
 960                                 l3fwd_simple_forward(pkts_burst[j], portid,
 961                                                 qconf);
 962                         }
 963
 964                         /* Forward remaining prefetched packets */
 965                         for (; j < nb_rx; j++) {
 966                                 l3fwd_simple_forward(pkts_burst[j], portid,
 967                                                 qconf);
 968                         }
 969
 970                 }
 971
 972         }
 973
 974         return 0;
 975 }
 976 /* main processing loop */
 977 static int
 978 main_loop(__attribute__((unused)) void *dummy)
 979 {
 980         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 981         unsigned lcore_id;
 982         uint64_t prev_tsc, diff_tsc, cur_tsc, tim_res_tsc, hz;
 983         uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
 984         int i, j, nb_rx;
 985         uint8_t queueid;
 986         uint16_t portid;
 987         struct lcore_conf *qconf;
 988         struct lcore_rx_queue *rx_queue;
 989         enum freq_scale_hint_t lcore_scaleup_hint;
 990         uint32_t lcore_rx_idle_count = 0;
 991         uint32_t lcore_idle_hint = 0;
 992         int intr_en = 0;
 993
 994         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 995
 996         prev_tsc = 0;
 997         hz = rte_get_timer_hz();
 998         tim_res_tsc = hz/TIMER_NUMBER_PER_SECOND;
 999
1000         lcore_id = rte_lcore_id();
1001         qconf = &lcore_conf[lcore_id];
1002
1003         if (qconf->n_rx_queue == 0) {
1004                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
1005                 return 0;
1006         }
1007
1008         RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
1009
1010         for (i = 0; i < qconf->n_rx_queue; i++) {
1011                 portid = qconf->rx_queue_list[i].port_id;
1012                 queueid = qconf->rx_queue_list[i].queue_id;
1013                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1014                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
1015         }
1016
1017         /* add into event wait list */
1018         if (event_register(qconf) == 0)
1019                 intr_en = 1;
1020         else
1021                 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
1022
1023         while (1) {
1024                 stats[lcore_id].nb_iteration_looped++;
1025
1026                 cur_tsc = rte_rdtsc();
1027                 cur_tsc_power = cur_tsc;
1028
1029                 /*
1030                  * TX burst queue drain
1031                  */
1032                 diff_tsc = cur_tsc - prev_tsc;
1033                 if (unlikely(diff_tsc > drain_tsc)) {
1034                         for (i = 0; i < qconf->n_tx_port; ++i) {
1035                                 portid = qconf->tx_port_id[i];
1036                                 rte_eth_tx_buffer_flush(portid,
1037                                                 qconf->tx_queue_id[portid],
1038                                                 qconf->tx_buffer[portid]);
1039                         }
1040                         prev_tsc = cur_tsc;
1041                 }
1042
1043                 diff_tsc_power = cur_tsc_power - prev_tsc_power;
1044                 if (diff_tsc_power > tim_res_tsc) {
1045                         rte_timer_manage();
1046                         prev_tsc_power = cur_tsc_power;
1047                 }
1048
1049 start_rx:
1050                 /*
1051                  * Read packet from RX queues
1052                  */
1053                 lcore_scaleup_hint = FREQ_CURRENT;
1054                 lcore_rx_idle_count = 0;
1055                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1056                         rx_queue = &(qconf->rx_queue_list[i]);
1057                         rx_queue->idle_hint = 0;
1058                         portid = rx_queue->port_id;
1059                         queueid = rx_queue->queue_id;
1060
1061                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1062                                                                 MAX_PKT_BURST);
1063
1064                         stats[lcore_id].nb_rx_processed += nb_rx;
1065                         if (unlikely(nb_rx == 0)) {
1066                                 /**
1067                                  * no packet received from rx queue, try to
1068                                  * sleep for a while forcing CPU enter deeper
1069                                  * C states.
1070                                  */
1071                                 rx_queue->zero_rx_packet_count++;
1072
1073                                 if (rx_queue->zero_rx_packet_count <=
1074                                                         MIN_ZERO_POLL_COUNT)
1075                                         continue;
1076
1077                                 rx_queue->idle_hint = power_idle_heuristic(\
1078                                         rx_queue->zero_rx_packet_count);
1079                                 lcore_rx_idle_count++;
1080                         } else {
1081                                 rx_queue->zero_rx_packet_count = 0;
1082
1083                                 /**
1084                                  * do not scale up frequency immediately as
1085                                  * user to kernel space communication is costly
1086                                  * which might impact packet I/O for received
1087                                  * packets.
1088                                  */
1089                                 rx_queue->freq_up_hint =
1090                                         power_freq_scaleup_heuristic(lcore_id,
1091                                                         portid, queueid);
1092                         }
1093
1094                         /* Prefetch first packets */
1095                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1096                                 rte_prefetch0(rte_pktmbuf_mtod(
1097                                                 pkts_burst[j], void *));
1098                         }
1099
1100                         /* Prefetch and forward already prefetched packets */
1101                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1102                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1103                                                 j + PREFETCH_OFFSET], void *));
1104                                 l3fwd_simple_forward(pkts_burst[j], portid,
1105                                                                 qconf);
1106                         }
1107
1108                         /* Forward remaining prefetched packets */
1109                         for (; j < nb_rx; j++) {
1110                                 l3fwd_simple_forward(pkts_burst[j], portid,
1111                                                                 qconf);
1112                         }
1113                 }
1114
1115                 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
1116                         for (i = 1, lcore_scaleup_hint =
1117                                 qconf->rx_queue_list[0].freq_up_hint;
1118                                         i < qconf->n_rx_queue; ++i) {
1119                                 rx_queue = &(qconf->rx_queue_list[i]);
1120                                 if (rx_queue->freq_up_hint >
1121                                                 lcore_scaleup_hint)
1122                                         lcore_scaleup_hint =
1123                                                 rx_queue->freq_up_hint;
1124                         }
1125
1126                         if (lcore_scaleup_hint == FREQ_HIGHEST) {
1127                                 if (rte_power_freq_max)
1128                                         rte_power_freq_max(lcore_id);
1129                         } else if (lcore_scaleup_hint == FREQ_HIGHER) {
1130                                 if (rte_power_freq_up)
1131                                         rte_power_freq_up(lcore_id);
1132                         }
1133                 } else {
1134                         /**
1135                          * All Rx queues empty in recent consecutive polls,
1136                          * sleep in a conservative manner, meaning sleep as
1137                          * less as possible.
1138                          */
1139                         for (i = 1, lcore_idle_hint =
1140                                 qconf->rx_queue_list[0].idle_hint;
1141                                         i < qconf->n_rx_queue; ++i) {
1142                                 rx_queue = &(qconf->rx_queue_list[i]);
1143                                 if (rx_queue->idle_hint < lcore_idle_hint)
1144                                         lcore_idle_hint = rx_queue->idle_hint;
1145                         }
1146
1147                         if (lcore_idle_hint < SUSPEND_THRESHOLD)
1148                                 /**
1149                                  * execute "pause" instruction to avoid context
1150                                  * switch which generally take hundred of
1151                                  * microseconds for short sleep.
1152                                  */
1153                                 rte_delay_us(lcore_idle_hint);
1154                         else {
1155                                 /* suspend until rx interrupt trigges */
1156                                 if (intr_en) {
1157                                         turn_on_intr(qconf);
1158                                         sleep_until_rx_interrupt(
1159                                                 qconf->n_rx_queue);
1160                                         /**
1161                                          * start receiving packets immediately
1162                                          */
1163                                         goto start_rx;
1164                                 }
1165                         }
1166                         stats[lcore_id].sleep_time += lcore_idle_hint;
1167                 }
1168         }
1169 }
1170
1171 static int
1172 check_lcore_params(void)
1173 {
1174         uint8_t queue, lcore;
1175         uint16_t i;
1176         int socketid;
1177
1178         for (i = 0; i < nb_lcore_params; ++i) {
1179                 queue = lcore_params[i].queue_id;
1180                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1181                         printf("invalid queue number: %hhu\n", queue);
1182                         return -1;
1183                 }
1184                 lcore = lcore_params[i].lcore_id;
1185                 if (!rte_lcore_is_enabled(lcore)) {
1186                         printf("error: lcore %hhu is not enabled in lcore "
1187                                                         "mask\n", lcore);
1188                         return -1;
1189                 }
1190                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1191                                                         (numa_on == 0)) {
1192                         printf("warning: lcore %hhu is on socket %d with numa "
1193                                                 "off\n", lcore, socketid);
1194                 }
1195         }
1196         return 0;
1197 }
1198
1199 static int
1200 check_port_config(void)
1201 {
1202         unsigned portid;
1203         uint16_t i;
1204
1205         for (i = 0; i < nb_lcore_params; ++i) {
1206                 portid = lcore_params[i].port_id;
1207                 if ((enabled_port_mask & (1 << portid)) == 0) {
1208                         printf("port %u is not enabled in port mask\n",
1209                                                                 portid);
1210                         return -1;
1211                 }
1212                 if (!rte_eth_dev_is_valid_port(portid)) {
1213                         printf("port %u is not present on the board\n",
1214                                                                 portid);
1215                         return -1;
1216                 }
1217         }
1218         return 0;
1219 }
1220
1221 static uint8_t
1222 get_port_n_rx_queues(const uint16_t port)
1223 {
1224         int queue = -1;
1225         uint16_t i;
1226
1227         for (i = 0; i < nb_lcore_params; ++i) {
1228                 if (lcore_params[i].port_id == port &&
1229                                 lcore_params[i].queue_id > queue)
1230                         queue = lcore_params[i].queue_id;
1231         }
1232         return (uint8_t)(++queue);
1233 }
1234
1235 static int
1236 init_lcore_rx_queues(void)
1237 {
1238         uint16_t i, nb_rx_queue;
1239         uint8_t lcore;
1240
1241         for (i = 0; i < nb_lcore_params; ++i) {
1242                 lcore = lcore_params[i].lcore_id;
1243                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1244                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1245                         printf("error: too many queues (%u) for lcore: %u\n",
1246                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1247                         return -1;
1248                 } else {
1249                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1250                                 lcore_params[i].port_id;
1251                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1252                                 lcore_params[i].queue_id;
1253                         lcore_conf[lcore].n_rx_queue++;
1254                 }
1255         }
1256         return 0;
1257 }
1258
1259 /* display usage */
1260 static void
1261 print_usage(const char *prgname)
1262 {
1263         printf ("%s [EAL options] -- -p PORTMASK -P"
1264                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1265                 "  [--high-perf-cores CORELIST"
1266                 "  [--perf-config (port,queue,hi_perf,lcore_index)[,(port,queue,hi_perf,lcore_index]]"
1267                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1268                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1269                 "  -P : enable promiscuous mode\n"
1270                 "  --config (port,queue,lcore): rx queues configuration\n"
1271                 "  --high-perf-cores CORELIST: list of high performance cores\n"
1272                 "  --perf-config: similar as config, cores specified as indices"
1273                 " for bins containing high or regular performance cores\n"
1274                 "  --no-numa: optional, disable numa awareness\n"
1275                 "  --enable-jumbo: enable jumbo frame"
1276                 " which max packet len is PKTLEN in decimal (64-9600)\n"
1277                 "  --parse-ptype: parse packet type by software\n"
1278                 "  --empty-poll: enable empty poll detection"
1279                 " follow (training_flag, high_threshold, med_threshold)\n",
1280                 prgname);
1281 }
1282
1283 static int parse_max_pkt_len(const char *pktlen)
1284 {
1285         char *end = NULL;
1286         unsigned long len;
1287
1288         /* parse decimal string */
1289         len = strtoul(pktlen, &end, 10);
1290         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1291                 return -1;
1292
1293         if (len == 0)
1294                 return -1;
1295
1296         return len;
1297 }
1298
1299 static int
1300 parse_portmask(const char *portmask)
1301 {
1302         char *end = NULL;
1303         unsigned long pm;
1304
1305         /* parse hexadecimal string */
1306         pm = strtoul(portmask, &end, 16);
1307         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1308                 return -1;
1309
1310         if (pm == 0)
1311                 return -1;
1312
1313         return pm;
1314 }
1315
1316 static int
1317 parse_config(const char *q_arg)
1318 {
1319         char s[256];
1320         const char *p, *p0 = q_arg;
1321         char *end;
1322         enum fieldnames {
1323                 FLD_PORT = 0,
1324                 FLD_QUEUE,
1325                 FLD_LCORE,
1326                 _NUM_FLD
1327         };
1328         unsigned long int_fld[_NUM_FLD];
1329         char *str_fld[_NUM_FLD];
1330         int i;
1331         unsigned size;
1332
1333         nb_lcore_params = 0;
1334
1335         while ((p = strchr(p0,'(')) != NULL) {
1336                 ++p;
1337                 if((p0 = strchr(p,')')) == NULL)
1338                         return -1;
1339
1340                 size = p0 - p;
1341                 if(size >= sizeof(s))
1342                         return -1;
1343
1344                 snprintf(s, sizeof(s), "%.*s", size, p);
1345                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1346                                                                 _NUM_FLD)
1347                         return -1;
1348                 for (i = 0; i < _NUM_FLD; i++){
1349                         errno = 0;
1350                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1351                         if (errno != 0 || end == str_fld[i] || int_fld[i] >
1352                                                                         255)
1353                                 return -1;
1354                 }
1355                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1356                         printf("exceeded max number of lcore params: %hu\n",
1357                                 nb_lcore_params);
1358                         return -1;
1359                 }
1360                 lcore_params_array[nb_lcore_params].port_id =
1361                                 (uint8_t)int_fld[FLD_PORT];
1362                 lcore_params_array[nb_lcore_params].queue_id =
1363                                 (uint8_t)int_fld[FLD_QUEUE];
1364                 lcore_params_array[nb_lcore_params].lcore_id =
1365                                 (uint8_t)int_fld[FLD_LCORE];
1366                 ++nb_lcore_params;
1367         }
1368         lcore_params = lcore_params_array;
1369
1370         return 0;
1371 }
1372 static int
1373 parse_ep_config(const char *q_arg)
1374 {
1375         char s[256];
1376         const char *p = q_arg;
1377         char *end;
1378         int  num_arg;
1379
1380         char *str_fld[3];
1381
1382         int training_flag;
1383         int med_edpi;
1384         int hgh_edpi;
1385
1386         ep_med_edpi = EMPTY_POLL_MED_THRESHOLD;
1387         ep_hgh_edpi = EMPTY_POLL_MED_THRESHOLD;
1388
1389         strlcpy(s, p, sizeof(s));
1390
1391         num_arg = rte_strsplit(s, sizeof(s), str_fld, 3, ',');
1392
1393         empty_poll_train = false;
1394
1395         if (num_arg == 0)
1396                 return 0;
1397
1398         if (num_arg == 3) {
1399
1400                 training_flag = strtoul(str_fld[0], &end, 0);
1401                 med_edpi = strtoul(str_fld[1], &end, 0);
1402                 hgh_edpi = strtoul(str_fld[2], &end, 0);
1403
1404                 if (training_flag == 1)
1405                         empty_poll_train = true;
1406
1407                 if (med_edpi > 0)
1408                         ep_med_edpi = med_edpi;
1409
1410                 if (med_edpi > 0)
1411                         ep_hgh_edpi = hgh_edpi;
1412
1413         } else {
1414
1415                 return -1;
1416         }
1417
1418         return 0;
1419
1420 }
1421 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1422
1423 /* Parse the argument given in the command line of the application */
1424 static int
1425 parse_args(int argc, char **argv)
1426 {
1427         int opt, ret;
1428         char **argvopt;
1429         int option_index;
1430         uint32_t limit;
1431         char *prgname = argv[0];
1432         static struct option lgopts[] = {
1433                 {"config", 1, 0, 0},
1434                 {"perf-config", 1, 0, 0},
1435                 {"high-perf-cores", 1, 0, 0},
1436                 {"no-numa", 0, 0, 0},
1437                 {"enable-jumbo", 0, 0, 0},
1438                 {"empty-poll", 1, 0, 0},
1439                 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
1440                 {NULL, 0, 0, 0}
1441         };
1442
1443         argvopt = argv;
1444
1445         while ((opt = getopt_long(argc, argvopt, "p:l:m:h:P",
1446                                 lgopts, &option_index)) != EOF) {
1447
1448                 switch (opt) {
1449                 /* portmask */
1450                 case 'p':
1451                         enabled_port_mask = parse_portmask(optarg);
1452                         if (enabled_port_mask == 0) {
1453                                 printf("invalid portmask\n");
1454                                 print_usage(prgname);
1455                                 return -1;
1456                         }
1457                         break;
1458                 case 'P':
1459                         printf("Promiscuous mode selected\n");
1460                         promiscuous_on = 1;
1461                         break;
1462                 case 'l':
1463                         limit = parse_max_pkt_len(optarg);
1464                         freq_tlb[LOW] = limit;
1465                         break;
1466                 case 'm':
1467                         limit = parse_max_pkt_len(optarg);
1468                         freq_tlb[MED] = limit;
1469                         break;
1470                 case 'h':
1471                         limit = parse_max_pkt_len(optarg);
1472                         freq_tlb[HGH] = limit;
1473                         break;
1474                 /* long options */
1475                 case 0:
1476                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1477                                 ret = parse_config(optarg);
1478                                 if (ret) {
1479                                         printf("invalid config\n");
1480                                         print_usage(prgname);
1481                                         return -1;
1482                                 }
1483                         }
1484
1485                         if (!strncmp(lgopts[option_index].name,
1486                                         "perf-config", 11)) {
1487                                 ret = parse_perf_config(optarg);
1488                                 if (ret) {
1489                                         printf("invalid perf-config\n");
1490                                         print_usage(prgname);
1491                                         return -1;
1492                                 }
1493                         }
1494
1495                         if (!strncmp(lgopts[option_index].name,
1496                                         "high-perf-cores", 15)) {
1497                                 ret = parse_perf_core_list(optarg);
1498                                 if (ret) {
1499                                         printf("invalid high-perf-cores\n");
1500                                         print_usage(prgname);
1501                                         return -1;
1502                                 }
1503                         }
1504
1505                         if (!strncmp(lgopts[option_index].name,
1506                                                 "no-numa", 7)) {
1507                                 printf("numa is disabled \n");
1508                                 numa_on = 0;
1509                         }
1510
1511                         if (!strncmp(lgopts[option_index].name,
1512                                                 "empty-poll", 10)) {
1513                                 printf("empty-poll is enabled\n");
1514                                 empty_poll_on = true;
1515                                 ret = parse_ep_config(optarg);
1516
1517                                 if (ret) {
1518                                         printf("invalid empty poll config\n");
1519                                         print_usage(prgname);
1520                                         return -1;
1521                                 }
1522
1523                         }
1524
1525                         if (!strncmp(lgopts[option_index].name,
1526                                         "enable-jumbo", 12)) {
1527                                 struct option lenopts =
1528                                         {"max-pkt-len", required_argument, \
1529                                                                         0, 0};
1530
1531                                 printf("jumbo frame is enabled \n");
1532                                 port_conf.rxmode.offloads |=
1533                                                 DEV_RX_OFFLOAD_JUMBO_FRAME;
1534                                 port_conf.txmode.offloads |=
1535                                                 DEV_TX_OFFLOAD_MULTI_SEGS;
1536
1537                                 /**
1538                                  * if no max-pkt-len set, use the default value
1539                                  * RTE_ETHER_MAX_LEN
1540                                  */
1541                                 if (0 == getopt_long(argc, argvopt, "",
1542                                                 &lenopts, &option_index)) {
1543                                         ret = parse_max_pkt_len(optarg);
1544                                         if ((ret < 64) ||
1545                                                 (ret > MAX_JUMBO_PKT_LEN)){
1546                                                 printf("invalid packet "
1547                                                                 "length\n");
1548                                                 print_usage(prgname);
1549                                                 return -1;
1550                                         }
1551                                         port_conf.rxmode.max_rx_pkt_len = ret;
1552                                 }
1553                                 printf("set jumbo frame "
1554                                         "max packet length to %u\n",
1555                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1556                         }
1557
1558                         if (!strncmp(lgopts[option_index].name,
1559                                      CMD_LINE_OPT_PARSE_PTYPE,
1560                                      sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
1561                                 printf("soft parse-ptype is enabled\n");
1562                                 parse_ptype = 1;
1563                         }
1564
1565                         break;
1566
1567                 default:
1568                         print_usage(prgname);
1569                         return -1;
1570                 }
1571         }
1572
1573         if (optind >= 0)
1574                 argv[optind-1] = prgname;
1575
1576         ret = optind-1;
1577         optind = 1; /* reset getopt lib */
1578         return ret;
1579 }
1580
1581 static void
1582 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1583 {
1584         char buf[RTE_ETHER_ADDR_FMT_SIZE];
1585         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
1586         printf("%s%s", name, buf);
1587 }
1588
1589 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1590 static void
1591 setup_hash(int socketid)
1592 {
1593         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1594                 .name = NULL,
1595                 .entries = L3FWD_HASH_ENTRIES,
1596                 .key_len = sizeof(struct ipv4_5tuple),
1597                 .hash_func = DEFAULT_HASH_FUNC,
1598                 .hash_func_init_val = 0,
1599         };
1600
1601         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1602                 .name = NULL,
1603                 .entries = L3FWD_HASH_ENTRIES,
1604                 .key_len = sizeof(struct ipv6_5tuple),
1605                 .hash_func = DEFAULT_HASH_FUNC,
1606                 .hash_func_init_val = 0,
1607         };
1608
1609         unsigned i;
1610         int ret;
1611         char s[64];
1612
1613         /* create ipv4 hash */
1614         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1615         ipv4_l3fwd_hash_params.name = s;
1616         ipv4_l3fwd_hash_params.socket_id = socketid;
1617         ipv4_l3fwd_lookup_struct[socketid] =
1618                 rte_hash_create(&ipv4_l3fwd_hash_params);
1619         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1620                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1621                                 "socket %d\n", socketid);
1622
1623         /* create ipv6 hash */
1624         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1625         ipv6_l3fwd_hash_params.name = s;
1626         ipv6_l3fwd_hash_params.socket_id = socketid;
1627         ipv6_l3fwd_lookup_struct[socketid] =
1628                 rte_hash_create(&ipv6_l3fwd_hash_params);
1629         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1630                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1631                                 "socket %d\n", socketid);
1632
1633
1634         /* populate the ipv4 hash */
1635         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1636                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1637                                 (void *) &ipv4_l3fwd_route_array[i].key);
1638                 if (ret < 0) {
1639                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1640                                 "l3fwd hash on socket %d\n", i, socketid);
1641                 }
1642                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1643                 printf("Hash: Adding key\n");
1644                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1645         }
1646
1647         /* populate the ipv6 hash */
1648         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1649                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1650                                 (void *) &ipv6_l3fwd_route_array[i].key);
1651                 if (ret < 0) {
1652                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1653                                 "l3fwd hash on socket %d\n", i, socketid);
1654                 }
1655                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1656                 printf("Hash: Adding key\n");
1657                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1658         }
1659 }
1660 #endif
1661
1662 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1663 static void
1664 setup_lpm(int socketid)
1665 {
1666         unsigned i;
1667         int ret;
1668         char s[64];
1669
1670         /* create the LPM table */
1671         struct rte_lpm_config lpm_ipv4_config;
1672
1673         lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
1674         lpm_ipv4_config.number_tbl8s = 256;
1675         lpm_ipv4_config.flags = 0;
1676
1677         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1678         ipv4_l3fwd_lookup_struct[socketid] =
1679                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
1680         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1681                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1682                                 " on socket %d\n", socketid);
1683
1684         /* populate the LPM table */
1685         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1686                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1687                         ipv4_l3fwd_route_array[i].ip,
1688                         ipv4_l3fwd_route_array[i].depth,
1689                         ipv4_l3fwd_route_array[i].if_out);
1690
1691                 if (ret < 0) {
1692                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1693                                 "l3fwd LPM table on socket %d\n",
1694                                 i, socketid);
1695                 }
1696
1697                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1698                         (unsigned)ipv4_l3fwd_route_array[i].ip,
1699                         ipv4_l3fwd_route_array[i].depth,
1700                         ipv4_l3fwd_route_array[i].if_out);
1701         }
1702 }
1703 #endif
1704
1705 static int
1706 init_mem(unsigned nb_mbuf)
1707 {
1708         struct lcore_conf *qconf;
1709         int socketid;
1710         unsigned lcore_id;
1711         char s[64];
1712
1713         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1714                 if (rte_lcore_is_enabled(lcore_id) == 0)
1715                         continue;
1716
1717                 if (numa_on)
1718                         socketid = rte_lcore_to_socket_id(lcore_id);
1719                 else
1720                         socketid = 0;
1721
1722                 if (socketid >= NB_SOCKETS) {
1723                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1724                                         "out of range %d\n", socketid,
1725                                                 lcore_id, NB_SOCKETS);
1726                 }
1727                 if (pktmbuf_pool[socketid] == NULL) {
1728                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1729                         pktmbuf_pool[socketid] =
1730                                 rte_pktmbuf_pool_create(s, nb_mbuf,
1731                                         MEMPOOL_CACHE_SIZE, 0,
1732                                         RTE_MBUF_DEFAULT_BUF_SIZE,
1733                                         socketid);
1734                         if (pktmbuf_pool[socketid] == NULL)
1735                                 rte_exit(EXIT_FAILURE,
1736                                         "Cannot init mbuf pool on socket %d\n",
1737                                                                 socketid);
1738                         else
1739                                 printf("Allocated mbuf pool on socket %d\n",
1740                                                                 socketid);
1741
1742 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1743                         setup_lpm(socketid);
1744 #else
1745                         setup_hash(socketid);
1746 #endif
1747                 }
1748                 qconf = &lcore_conf[lcore_id];
1749                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1750 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1751                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1752 #endif
1753         }
1754         return 0;
1755 }
1756
1757 /* Check the link status of all ports in up to 9s, and print them finally */
1758 static void
1759 check_all_ports_link_status(uint32_t port_mask)
1760 {
1761 #define CHECK_INTERVAL 100 /* 100ms */
1762 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1763         uint8_t count, all_ports_up, print_flag = 0;
1764         uint16_t portid;
1765         struct rte_eth_link link;
1766
1767         printf("\nChecking link status");
1768         fflush(stdout);
1769         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1770                 all_ports_up = 1;
1771                 RTE_ETH_FOREACH_DEV(portid) {
1772                         if ((port_mask & (1 << portid)) == 0)
1773                                 continue;
1774                         memset(&link, 0, sizeof(link));
1775                         rte_eth_link_get_nowait(portid, &link);
1776                         /* print link status if flag set */
1777                         if (print_flag == 1) {
1778                                 if (link.link_status)
1779                                         printf("Port %d Link Up - speed %u "
1780                                                 "Mbps - %s\n", (uint8_t)portid,
1781                                                 (unsigned)link.link_speed,
1782                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1783                                         ("full-duplex") : ("half-duplex\n"));
1784                                 else
1785                                         printf("Port %d Link Down\n",
1786                                                 (uint8_t)portid);
1787                                 continue;
1788                         }
1789                         /* clear all_ports_up flag if any link down */
1790                         if (link.link_status == ETH_LINK_DOWN) {
1791                                 all_ports_up = 0;
1792                                 break;
1793                         }
1794                 }
1795                 /* after finally printing all link status, get out */
1796                 if (print_flag == 1)
1797                         break;
1798
1799                 if (all_ports_up == 0) {
1800                         printf(".");
1801                         fflush(stdout);
1802                         rte_delay_ms(CHECK_INTERVAL);
1803                 }
1804
1805                 /* set the print_flag if all ports up or timeout */
1806                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1807                         print_flag = 1;
1808                         printf("done\n");
1809                 }
1810         }
1811 }
1812
1813 static int check_ptype(uint16_t portid)
1814 {
1815         int i, ret;
1816         int ptype_l3_ipv4 = 0;
1817 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1818         int ptype_l3_ipv6 = 0;
1819 #endif
1820         uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
1821
1822         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
1823         if (ret <= 0)
1824                 return 0;
1825
1826         uint32_t ptypes[ret];
1827
1828         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
1829         for (i = 0; i < ret; ++i) {
1830                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
1831                         ptype_l3_ipv4 = 1;
1832 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1833                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
1834                         ptype_l3_ipv6 = 1;
1835 #endif
1836         }
1837
1838         if (ptype_l3_ipv4 == 0)
1839                 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
1840
1841 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1842         if (ptype_l3_ipv6 == 0)
1843                 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
1844 #endif
1845
1846 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1847         if (ptype_l3_ipv4)
1848 #else /* APP_LOOKUP_EXACT_MATCH */
1849         if (ptype_l3_ipv4 && ptype_l3_ipv6)
1850 #endif
1851                 return 1;
1852
1853         return 0;
1854
1855 }
1856
1857 static int
1858 init_power_library(void)
1859 {
1860         int ret = 0, lcore_id;
1861         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1862                 if (rte_lcore_is_enabled(lcore_id)) {
1863                         /* init power management library */
1864                         ret = rte_power_init(lcore_id);
1865                         if (ret)
1866                                 RTE_LOG(ERR, POWER,
1867                                 "Library initialization failed on core %u\n",
1868                                 lcore_id);
1869                 }
1870         }
1871         return ret;
1872 }
1873 static void
1874 empty_poll_setup_timer(void)
1875 {
1876         int lcore_id = rte_lcore_id();
1877         uint64_t hz = rte_get_timer_hz();
1878
1879         struct  ep_params *ep_ptr = ep_params;
1880
1881         ep_ptr->interval_ticks = hz / INTERVALS_PER_SECOND;
1882
1883         rte_timer_reset_sync(&ep_ptr->timer0,
1884                         ep_ptr->interval_ticks,
1885                         PERIODICAL,
1886                         lcore_id,
1887                         rte_empty_poll_detection,
1888                         (void *)ep_ptr);
1889
1890 }
1891 static int
1892 launch_timer(unsigned int lcore_id)
1893 {
1894         int64_t prev_tsc = 0, cur_tsc, diff_tsc, cycles_10ms;
1895
1896         RTE_SET_USED(lcore_id);
1897
1898
1899         if (rte_get_master_lcore() != lcore_id) {
1900                 rte_panic("timer on lcore:%d which is not master core:%d\n",
1901                                 lcore_id,
1902                                 rte_get_master_lcore());
1903         }
1904
1905         RTE_LOG(INFO, POWER, "Bring up the Timer\n");
1906
1907         empty_poll_setup_timer();
1908
1909         cycles_10ms = rte_get_timer_hz() / 100;
1910
1911         while (!is_done()) {
1912                 cur_tsc = rte_rdtsc();
1913                 diff_tsc = cur_tsc - prev_tsc;
1914                 if (diff_tsc > cycles_10ms) {
1915                         rte_timer_manage();
1916                         prev_tsc = cur_tsc;
1917                         cycles_10ms = rte_get_timer_hz() / 100;
1918                 }
1919         }
1920
1921         RTE_LOG(INFO, POWER, "Timer_subsystem is done\n");
1922
1923         return 0;
1924 }
1925
1926
1927 int
1928 main(int argc, char **argv)
1929 {
1930         struct lcore_conf *qconf;
1931         struct rte_eth_dev_info dev_info;
1932         struct rte_eth_txconf *txconf;
1933         int ret;
1934         uint16_t nb_ports;
1935         uint16_t queueid;
1936         unsigned lcore_id;
1937         uint64_t hz;
1938         uint32_t n_tx_queue, nb_lcores;
1939         uint32_t dev_rxq_num, dev_txq_num;
1940         uint8_t nb_rx_queue, queue, socketid;
1941         uint16_t portid;
1942
1943         /* catch SIGINT and restore cpufreq governor to ondemand */
1944         signal(SIGINT, signal_exit_now);
1945
1946         /* init EAL */
1947         ret = rte_eal_init(argc, argv);
1948         if (ret < 0)
1949                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1950         argc -= ret;
1951         argv += ret;
1952
1953         /* init RTE timer library to be used late */
1954         rte_timer_subsystem_init();
1955
1956         /* parse application arguments (after the EAL ones) */
1957         ret = parse_args(argc, argv);
1958         if (ret < 0)
1959                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1960
1961         if (init_power_library())
1962                 RTE_LOG(ERR, L3FWD_POWER, "init_power_library failed\n");
1963
1964         if (update_lcore_params() < 0)
1965                 rte_exit(EXIT_FAILURE, "update_lcore_params failed\n");
1966
1967         if (check_lcore_params() < 0)
1968                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1969
1970         ret = init_lcore_rx_queues();
1971         if (ret < 0)
1972                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1973
1974         nb_ports = rte_eth_dev_count_avail();
1975
1976         if (check_port_config() < 0)
1977                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1978
1979         nb_lcores = rte_lcore_count();
1980
1981         /* initialize all ports */
1982         RTE_ETH_FOREACH_DEV(portid) {
1983                 struct rte_eth_conf local_port_conf = port_conf;
1984
1985                 /* skip ports that are not enabled */
1986                 if ((enabled_port_mask & (1 << portid)) == 0) {
1987                         printf("\nSkipping disabled port %d\n", portid);
1988                         continue;
1989                 }
1990
1991                 /* init port */
1992                 printf("Initializing port %d ... ", portid );
1993                 fflush(stdout);
1994
1995                 rte_eth_dev_info_get(portid, &dev_info);
1996                 dev_rxq_num = dev_info.max_rx_queues;
1997                 dev_txq_num = dev_info.max_tx_queues;
1998
1999                 nb_rx_queue = get_port_n_rx_queues(portid);
2000                 if (nb_rx_queue > dev_rxq_num)
2001                         rte_exit(EXIT_FAILURE,
2002                                 "Cannot configure not existed rxq: "
2003                                 "port=%d\n", portid);
2004
2005                 n_tx_queue = nb_lcores;
2006                 if (n_tx_queue > dev_txq_num)
2007                         n_tx_queue = dev_txq_num;
2008                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
2009                         nb_rx_queue, (unsigned)n_tx_queue );
2010                 /* If number of Rx queue is 0, no need to enable Rx interrupt */
2011                 if (nb_rx_queue == 0)
2012                         local_port_conf.intr_conf.rxq = 0;
2013                 rte_eth_dev_info_get(portid, &dev_info);
2014                 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
2015                         local_port_conf.txmode.offloads |=
2016                                 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
2017
2018                 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2019                         dev_info.flow_type_rss_offloads;
2020                 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2021                                 port_conf.rx_adv_conf.rss_conf.rss_hf) {
2022                         printf("Port %u modified RSS hash function based on hardware support,"
2023                                 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2024                                 portid,
2025                                 port_conf.rx_adv_conf.rss_conf.rss_hf,
2026                                 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2027                 }
2028
2029                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
2030                                         (uint16_t)n_tx_queue, &local_port_conf);
2031                 if (ret < 0)
2032                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
2033                                         "err=%d, port=%d\n", ret, portid);
2034
2035                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
2036                                                        &nb_txd);
2037                 if (ret < 0)
2038                         rte_exit(EXIT_FAILURE,
2039                                  "Cannot adjust number of descriptors: err=%d, port=%d\n",
2040                                  ret, portid);
2041
2042                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
2043                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
2044                 printf(", ");
2045
2046                 /* init memory */
2047                 ret = init_mem(NB_MBUF);
2048                 if (ret < 0)
2049                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
2050
2051                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2052                         if (rte_lcore_is_enabled(lcore_id) == 0)
2053                                 continue;
2054
2055                         /* Initialize TX buffers */
2056                         qconf = &lcore_conf[lcore_id];
2057                         qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
2058                                 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
2059                                 rte_eth_dev_socket_id(portid));
2060                         if (qconf->tx_buffer[portid] == NULL)
2061                                 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
2062                                                  portid);
2063
2064                         rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
2065                 }
2066
2067                 /* init one TX queue per couple (lcore,port) */
2068                 queueid = 0;
2069                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2070                         if (rte_lcore_is_enabled(lcore_id) == 0)
2071                                 continue;
2072
2073                         if (queueid >= dev_txq_num)
2074                                 continue;
2075
2076                         if (numa_on)
2077                                 socketid = \
2078                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
2079                         else
2080                                 socketid = 0;
2081
2082                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
2083                         fflush(stdout);
2084
2085                         txconf = &dev_info.default_txconf;
2086                         txconf->offloads = local_port_conf.txmode.offloads;
2087                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
2088                                                      socketid, txconf);
2089                         if (ret < 0)
2090                                 rte_exit(EXIT_FAILURE,
2091                                         "rte_eth_tx_queue_setup: err=%d, "
2092                                                 "port=%d\n", ret, portid);
2093
2094                         qconf = &lcore_conf[lcore_id];
2095                         qconf->tx_queue_id[portid] = queueid;
2096                         queueid++;
2097
2098                         qconf->tx_port_id[qconf->n_tx_port] = portid;
2099                         qconf->n_tx_port++;
2100                 }
2101                 printf("\n");
2102         }
2103
2104         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2105                 if (rte_lcore_is_enabled(lcore_id) == 0)
2106                         continue;
2107
2108                 if (empty_poll_on == false) {
2109                         /* init timer structures for each enabled lcore */
2110                         rte_timer_init(&power_timers[lcore_id]);
2111                         hz = rte_get_timer_hz();
2112                         rte_timer_reset(&power_timers[lcore_id],
2113                                         hz/TIMER_NUMBER_PER_SECOND,
2114                                         SINGLE, lcore_id,
2115                                         power_timer_cb, NULL);
2116                 }
2117                 qconf = &lcore_conf[lcore_id];
2118                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
2119                 fflush(stdout);
2120                 /* init RX queues */
2121                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
2122                         struct rte_eth_rxconf rxq_conf;
2123                         struct rte_eth_dev *dev;
2124                         struct rte_eth_conf *conf;
2125
2126                         portid = qconf->rx_queue_list[queue].port_id;
2127                         queueid = qconf->rx_queue_list[queue].queue_id;
2128                         dev = &rte_eth_devices[portid];
2129                         conf = &dev->data->dev_conf;
2130
2131                         if (numa_on)
2132                                 socketid = \
2133                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
2134                         else
2135                                 socketid = 0;
2136
2137                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
2138                         fflush(stdout);
2139
2140                         rte_eth_dev_info_get(portid, &dev_info);
2141                         rxq_conf = dev_info.default_rxconf;
2142                         rxq_conf.offloads = conf->rxmode.offloads;
2143                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
2144                                 socketid, &rxq_conf,
2145                                 pktmbuf_pool[socketid]);
2146                         if (ret < 0)
2147                                 rte_exit(EXIT_FAILURE,
2148                                         "rte_eth_rx_queue_setup: err=%d, "
2149                                                 "port=%d\n", ret, portid);
2150
2151                         if (parse_ptype) {
2152                                 if (add_cb_parse_ptype(portid, queueid) < 0)
2153                                         rte_exit(EXIT_FAILURE,
2154                                                  "Fail to add ptype cb\n");
2155                         } else if (!check_ptype(portid))
2156                                 rte_exit(EXIT_FAILURE,
2157                                          "PMD can not provide needed ptypes\n");
2158                 }
2159         }
2160
2161         printf("\n");
2162
2163         /* start ports */
2164         RTE_ETH_FOREACH_DEV(portid) {
2165                 if ((enabled_port_mask & (1 << portid)) == 0) {
2166                         continue;
2167                 }
2168                 /* Start device */
2169                 ret = rte_eth_dev_start(portid);
2170                 if (ret < 0)
2171                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
2172                                                 "port=%d\n", ret, portid);
2173                 /*
2174                  * If enabled, put device in promiscuous mode.
2175                  * This allows IO forwarding mode to forward packets
2176                  * to itself through 2 cross-connected  ports of the
2177                  * target machine.
2178                  */
2179                 if (promiscuous_on)
2180                         rte_eth_promiscuous_enable(portid);
2181                 /* initialize spinlock for each port */
2182                 rte_spinlock_init(&(locks[portid]));
2183         }
2184
2185         check_all_ports_link_status(enabled_port_mask);
2186
2187         if (empty_poll_on == true) {
2188
2189                 if (empty_poll_train) {
2190                         policy.state = TRAINING;
2191                 } else {
2192                         policy.state = MED_NORMAL;
2193                         policy.med_base_edpi = ep_med_edpi;
2194                         policy.hgh_base_edpi = ep_hgh_edpi;
2195                 }
2196
2197                 ret = rte_power_empty_poll_stat_init(&ep_params,
2198                                 freq_tlb,
2199                                 &policy);
2200                 if (ret < 0)
2201                         rte_exit(EXIT_FAILURE, "empty poll init failed");
2202         }
2203
2204
2205         /* launch per-lcore init on every lcore */
2206         if (empty_poll_on == false) {
2207                 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2208         } else {
2209                 empty_poll_stop = false;
2210                 rte_eal_mp_remote_launch(main_empty_poll_loop, NULL,
2211                                 SKIP_MASTER);
2212         }
2213
2214         if (empty_poll_on == true)
2215                 launch_timer(rte_lcore_id());
2216
2217         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2218                 if (rte_eal_wait_lcore(lcore_id) < 0)
2219                         return -1;
2220         }
2221
2222         if (empty_poll_on)
2223                 rte_power_empty_poll_stat_free();
2224
2225         return 0;
2226 }