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 = 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         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 262         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 263         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 264         {{IPv4(100,40,0,4), 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         {IPv4(1,1,1,0), 24, 0},
 304         {IPv4(2,1,1,0), 24, 1},
 305         {IPv4(3,1,1,0), 24, 2},
 306         {IPv4(4,1,1,0), 24, 3},
 307         {IPv4(5,1,1,0), 24, 4},
 308         {IPv4(6,1,1,0), 24, 5},
 309         {IPv4(7,1,1,0), 24, 6},
 310         {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 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 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 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 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 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 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 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 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 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 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(ETHER_TYPE_IPv4))
 628                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 629         else if (ether_type == rte_cpu_to_be_16(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 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 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                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 704
 705                 send_single_packet(m, dst_port);
 706         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 707                 /* Handle IPv6 headers.*/
 708 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 709                 struct ipv6_hdr *ipv6_hdr;
 710
 711                 ipv6_hdr =
 712                         rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
 713                                                 sizeof(struct rte_ether_hdr));
 714
 715                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
 716                                         qconf->ipv6_lookup_struct);
 717
 718                 if (dst_port >= RTE_MAX_ETHPORTS ||
 719                                 (enabled_port_mask & 1 << dst_port) == 0)
 720                         dst_port = portid;
 721
 722                 /* 02:00:00:00:00:xx */
 723                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 724                 *((uint64_t *)d_addr_bytes) =
 725                         0x000000000002 + ((uint64_t)dst_port << 40);
 726
 727                 /* src addr */
 728                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 729
 730                 send_single_packet(m, dst_port);
 731 #else
 732                 /* We don't currently handle IPv6 packets in LPM mode. */
 733                 rte_pktmbuf_free(m);
 734 #endif
 735         } else
 736                 rte_pktmbuf_free(m);
 737
 738 }
 739
 740 #define MINIMUM_SLEEP_TIME         1
 741 #define SUSPEND_THRESHOLD          300
 742
 743 static inline uint32_t
 744 power_idle_heuristic(uint32_t zero_rx_packet_count)
 745 {
 746         /* If zero count is less than 100,  sleep 1us */
 747         if (zero_rx_packet_count < SUSPEND_THRESHOLD)
 748                 return MINIMUM_SLEEP_TIME;
 749         /* If zero count is less than 1000, sleep 100 us which is the
 750                 minimum latency switching from C3/C6 to C0
 751         */
 752         else
 753                 return SUSPEND_THRESHOLD;
 754 }
 755
 756 static inline enum freq_scale_hint_t
 757 power_freq_scaleup_heuristic(unsigned lcore_id,
 758                              uint16_t port_id,
 759                              uint16_t queue_id)
 760 {
 761         uint32_t rxq_count = rte_eth_rx_queue_count(port_id, queue_id);
 762 /**
 763  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
 764  * per iteration
 765  */
 766 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
 767 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
 768 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
 769 #define FREQ_UP_TREND1_ACC   1
 770 #define FREQ_UP_TREND2_ACC   100
 771 #define FREQ_UP_THRESHOLD    10000
 772
 773         if (likely(rxq_count > FREQ_GEAR3_RX_PACKET_THRESHOLD)) {
 774                 stats[lcore_id].trend = 0;
 775                 return FREQ_HIGHEST;
 776         } else if (likely(rxq_count > FREQ_GEAR2_RX_PACKET_THRESHOLD))
 777                 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
 778         else if (likely(rxq_count > FREQ_GEAR1_RX_PACKET_THRESHOLD))
 779                 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
 780
 781         if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
 782                 stats[lcore_id].trend = 0;
 783                 return FREQ_HIGHER;
 784         }
 785
 786         return FREQ_CURRENT;
 787 }
 788
 789 /**
 790  * force polling thread sleep until one-shot rx interrupt triggers
 791  * @param port_id
 792  *  Port id.
 793  * @param queue_id
 794  *  Rx queue id.
 795  * @return
 796  *  0 on success
 797  */
 798 static int
 799 sleep_until_rx_interrupt(int num)
 800 {
 801         struct rte_epoll_event event[num];
 802         int n, i;
 803         uint16_t port_id;
 804         uint8_t queue_id;
 805         void *data;
 806
 807         RTE_LOG(INFO, L3FWD_POWER,
 808                 "lcore %u sleeps until interrupt triggers\n",
 809                 rte_lcore_id());
 810
 811         n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
 812         for (i = 0; i < n; i++) {
 813                 data = event[i].epdata.data;
 814                 port_id = ((uintptr_t)data) >> CHAR_BIT;
 815                 queue_id = ((uintptr_t)data) &
 816                         RTE_LEN2MASK(CHAR_BIT, uint8_t);
 817                 rte_eth_dev_rx_intr_disable(port_id, queue_id);
 818                 RTE_LOG(INFO, L3FWD_POWER,
 819                         "lcore %u is waked up from rx interrupt on"
 820                         " port %d queue %d\n",
 821                         rte_lcore_id(), port_id, queue_id);
 822         }
 823
 824         return 0;
 825 }
 826
 827 static void turn_on_intr(struct lcore_conf *qconf)
 828 {
 829         int i;
 830         struct lcore_rx_queue *rx_queue;
 831         uint8_t queue_id;
 832         uint16_t port_id;
 833
 834         for (i = 0; i < qconf->n_rx_queue; ++i) {
 835                 rx_queue = &(qconf->rx_queue_list[i]);
 836                 port_id = rx_queue->port_id;
 837                 queue_id = rx_queue->queue_id;
 838
 839                 rte_spinlock_lock(&(locks[port_id]));
 840                 rte_eth_dev_rx_intr_enable(port_id, queue_id);
 841                 rte_spinlock_unlock(&(locks[port_id]));
 842         }
 843 }
 844
 845 static int event_register(struct lcore_conf *qconf)
 846 {
 847         struct lcore_rx_queue *rx_queue;
 848         uint8_t queueid;
 849         uint16_t portid;
 850         uint32_t data;
 851         int ret;
 852         int i;
 853
 854         for (i = 0; i < qconf->n_rx_queue; ++i) {
 855                 rx_queue = &(qconf->rx_queue_list[i]);
 856                 portid = rx_queue->port_id;
 857                 queueid = rx_queue->queue_id;
 858                 data = portid << CHAR_BIT | queueid;
 859
 860                 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
 861                                                 RTE_EPOLL_PER_THREAD,
 862                                                 RTE_INTR_EVENT_ADD,
 863                                                 (void *)((uintptr_t)data));
 864                 if (ret)
 865                         return ret;
 866         }
 867
 868         return 0;
 869 }
 870 /* main processing loop */
 871 static int
 872 main_empty_poll_loop(__attribute__((unused)) void *dummy)
 873 {
 874         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 875         unsigned int lcore_id;
 876         uint64_t prev_tsc, diff_tsc, cur_tsc;
 877         int i, j, nb_rx;
 878         uint8_t queueid;
 879         uint16_t portid;
 880         struct lcore_conf *qconf;
 881         struct lcore_rx_queue *rx_queue;
 882
 883         const uint64_t drain_tsc =
 884                 (rte_get_tsc_hz() + US_PER_S - 1) /
 885                 US_PER_S * BURST_TX_DRAIN_US;
 886
 887         prev_tsc = 0;
 888
 889         lcore_id = rte_lcore_id();
 890         qconf = &lcore_conf[lcore_id];
 891
 892         if (qconf->n_rx_queue == 0) {
 893                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
 894                         lcore_id);
 895                 return 0;
 896         }
 897
 898         for (i = 0; i < qconf->n_rx_queue; i++) {
 899                 portid = qconf->rx_queue_list[i].port_id;
 900                 queueid = qconf->rx_queue_list[i].queue_id;
 901                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
 902                                 "rxqueueid=%hhu\n", lcore_id, portid, queueid);
 903         }
 904
 905         while (!is_done()) {
 906                 stats[lcore_id].nb_iteration_looped++;
 907
 908                 cur_tsc = rte_rdtsc();
 909                 /*
 910                  * TX burst queue drain
 911                  */
 912                 diff_tsc = cur_tsc - prev_tsc;
 913                 if (unlikely(diff_tsc > drain_tsc)) {
 914                         for (i = 0; i < qconf->n_tx_port; ++i) {
 915                                 portid = qconf->tx_port_id[i];
 916                                 rte_eth_tx_buffer_flush(portid,
 917                                                 qconf->tx_queue_id[portid],
 918                                                 qconf->tx_buffer[portid]);
 919                         }
 920                         prev_tsc = cur_tsc;
 921                 }
 922
 923                 /*
 924                  * Read packet from RX queues
 925                  */
 926                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 927                         rx_queue = &(qconf->rx_queue_list[i]);
 928                         rx_queue->idle_hint = 0;
 929                         portid = rx_queue->port_id;
 930                         queueid = rx_queue->queue_id;
 931
 932                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 933                                         MAX_PKT_BURST);
 934
 935                         stats[lcore_id].nb_rx_processed += nb_rx;
 936
 937                         if (nb_rx == 0) {
 938
 939                                 rte_power_empty_poll_stat_update(lcore_id);
 940
 941                                 continue;
 942                         } else {
 943                                 rte_power_poll_stat_update(lcore_id, nb_rx);
 944                         }
 945
 946
 947                         /* Prefetch first packets */
 948                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 949                                 rte_prefetch0(rte_pktmbuf_mtod(
 950                                                         pkts_burst[j], void *));
 951                         }
 952
 953                         /* Prefetch and forward already prefetched packets */
 954                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 955                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 956                                                         j + PREFETCH_OFFSET],
 957                                                         void *));
 958                                 l3fwd_simple_forward(pkts_burst[j], portid,
 959                                                 qconf);
 960                         }
 961
 962                         /* Forward remaining prefetched packets */
 963                         for (; j < nb_rx; j++) {
 964                                 l3fwd_simple_forward(pkts_burst[j], portid,
 965                                                 qconf);
 966                         }
 967
 968                 }
 969
 970         }
 971
 972         return 0;
 973 }
 974 /* main processing loop */
 975 static int
 976 main_loop(__attribute__((unused)) void *dummy)
 977 {
 978         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 979         unsigned lcore_id;
 980         uint64_t prev_tsc, diff_tsc, cur_tsc, tim_res_tsc, hz;
 981         uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
 982         int i, j, nb_rx;
 983         uint8_t queueid;
 984         uint16_t portid;
 985         struct lcore_conf *qconf;
 986         struct lcore_rx_queue *rx_queue;
 987         enum freq_scale_hint_t lcore_scaleup_hint;
 988         uint32_t lcore_rx_idle_count = 0;
 989         uint32_t lcore_idle_hint = 0;
 990         int intr_en = 0;
 991
 992         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 993
 994         prev_tsc = 0;
 995         hz = rte_get_timer_hz();
 996         tim_res_tsc = hz/TIMER_NUMBER_PER_SECOND;
 997
 998         lcore_id = rte_lcore_id();
 999         qconf = &lcore_conf[lcore_id];
1000
1001         if (qconf->n_rx_queue == 0) {
1002                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
1003                 return 0;
1004         }
1005
1006         RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
1007
1008         for (i = 0; i < qconf->n_rx_queue; i++) {
1009                 portid = qconf->rx_queue_list[i].port_id;
1010                 queueid = qconf->rx_queue_list[i].queue_id;
1011                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1012                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
1013         }
1014
1015         /* add into event wait list */
1016         if (event_register(qconf) == 0)
1017                 intr_en = 1;
1018         else
1019                 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
1020
1021         while (1) {
1022                 stats[lcore_id].nb_iteration_looped++;
1023
1024                 cur_tsc = rte_rdtsc();
1025                 cur_tsc_power = cur_tsc;
1026
1027                 /*
1028                  * TX burst queue drain
1029                  */
1030                 diff_tsc = cur_tsc - prev_tsc;
1031                 if (unlikely(diff_tsc > drain_tsc)) {
1032                         for (i = 0; i < qconf->n_tx_port; ++i) {
1033                                 portid = qconf->tx_port_id[i];
1034                                 rte_eth_tx_buffer_flush(portid,
1035                                                 qconf->tx_queue_id[portid],
1036                                                 qconf->tx_buffer[portid]);
1037                         }
1038                         prev_tsc = cur_tsc;
1039                 }
1040
1041                 diff_tsc_power = cur_tsc_power - prev_tsc_power;
1042                 if (diff_tsc_power > tim_res_tsc) {
1043                         rte_timer_manage();
1044                         prev_tsc_power = cur_tsc_power;
1045                 }
1046
1047 start_rx:
1048                 /*
1049                  * Read packet from RX queues
1050                  */
1051                 lcore_scaleup_hint = FREQ_CURRENT;
1052                 lcore_rx_idle_count = 0;
1053                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1054                         rx_queue = &(qconf->rx_queue_list[i]);
1055                         rx_queue->idle_hint = 0;
1056                         portid = rx_queue->port_id;
1057                         queueid = rx_queue->queue_id;
1058
1059                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1060                                                                 MAX_PKT_BURST);
1061
1062                         stats[lcore_id].nb_rx_processed += nb_rx;
1063                         if (unlikely(nb_rx == 0)) {
1064                                 /**
1065                                  * no packet received from rx queue, try to
1066                                  * sleep for a while forcing CPU enter deeper
1067                                  * C states.
1068                                  */
1069                                 rx_queue->zero_rx_packet_count++;
1070
1071                                 if (rx_queue->zero_rx_packet_count <=
1072                                                         MIN_ZERO_POLL_COUNT)
1073                                         continue;
1074
1075                                 rx_queue->idle_hint = power_idle_heuristic(\
1076                                         rx_queue->zero_rx_packet_count);
1077                                 lcore_rx_idle_count++;
1078                         } else {
1079                                 rx_queue->zero_rx_packet_count = 0;
1080
1081                                 /**
1082                                  * do not scale up frequency immediately as
1083                                  * user to kernel space communication is costly
1084                                  * which might impact packet I/O for received
1085                                  * packets.
1086                                  */
1087                                 rx_queue->freq_up_hint =
1088                                         power_freq_scaleup_heuristic(lcore_id,
1089                                                         portid, queueid);
1090                         }
1091
1092                         /* Prefetch first packets */
1093                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1094                                 rte_prefetch0(rte_pktmbuf_mtod(
1095                                                 pkts_burst[j], void *));
1096                         }
1097
1098                         /* Prefetch and forward already prefetched packets */
1099                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1100                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1101                                                 j + PREFETCH_OFFSET], void *));
1102                                 l3fwd_simple_forward(pkts_burst[j], portid,
1103                                                                 qconf);
1104                         }
1105
1106                         /* Forward remaining prefetched packets */
1107                         for (; j < nb_rx; j++) {
1108                                 l3fwd_simple_forward(pkts_burst[j], portid,
1109                                                                 qconf);
1110                         }
1111                 }
1112
1113                 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
1114                         for (i = 1, lcore_scaleup_hint =
1115                                 qconf->rx_queue_list[0].freq_up_hint;
1116                                         i < qconf->n_rx_queue; ++i) {
1117                                 rx_queue = &(qconf->rx_queue_list[i]);
1118                                 if (rx_queue->freq_up_hint >
1119                                                 lcore_scaleup_hint)
1120                                         lcore_scaleup_hint =
1121                                                 rx_queue->freq_up_hint;
1122                         }
1123
1124                         if (lcore_scaleup_hint == FREQ_HIGHEST) {
1125                                 if (rte_power_freq_max)
1126                                         rte_power_freq_max(lcore_id);
1127                         } else if (lcore_scaleup_hint == FREQ_HIGHER) {
1128                                 if (rte_power_freq_up)
1129                                         rte_power_freq_up(lcore_id);
1130                         }
1131                 } else {
1132                         /**
1133                          * All Rx queues empty in recent consecutive polls,
1134                          * sleep in a conservative manner, meaning sleep as
1135                          * less as possible.
1136                          */
1137                         for (i = 1, lcore_idle_hint =
1138                                 qconf->rx_queue_list[0].idle_hint;
1139                                         i < qconf->n_rx_queue; ++i) {
1140                                 rx_queue = &(qconf->rx_queue_list[i]);
1141                                 if (rx_queue->idle_hint < lcore_idle_hint)
1142                                         lcore_idle_hint = rx_queue->idle_hint;
1143                         }
1144
1145                         if (lcore_idle_hint < SUSPEND_THRESHOLD)
1146                                 /**
1147                                  * execute "pause" instruction to avoid context
1148                                  * switch which generally take hundred of
1149                                  * microseconds for short sleep.
1150                                  */
1151                                 rte_delay_us(lcore_idle_hint);
1152                         else {
1153                                 /* suspend until rx interrupt trigges */
1154                                 if (intr_en) {
1155                                         turn_on_intr(qconf);
1156                                         sleep_until_rx_interrupt(
1157                                                 qconf->n_rx_queue);
1158                                         /**
1159                                          * start receiving packets immediately
1160                                          */
1161                                         goto start_rx;
1162                                 }
1163                         }
1164                         stats[lcore_id].sleep_time += lcore_idle_hint;
1165                 }
1166         }
1167 }
1168
1169 static int
1170 check_lcore_params(void)
1171 {
1172         uint8_t queue, lcore;
1173         uint16_t i;
1174         int socketid;
1175
1176         for (i = 0; i < nb_lcore_params; ++i) {
1177                 queue = lcore_params[i].queue_id;
1178                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1179                         printf("invalid queue number: %hhu\n", queue);
1180                         return -1;
1181                 }
1182                 lcore = lcore_params[i].lcore_id;
1183                 if (!rte_lcore_is_enabled(lcore)) {
1184                         printf("error: lcore %hhu is not enabled in lcore "
1185                                                         "mask\n", lcore);
1186                         return -1;
1187                 }
1188                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1189                                                         (numa_on == 0)) {
1190                         printf("warning: lcore %hhu is on socket %d with numa "
1191                                                 "off\n", lcore, socketid);
1192                 }
1193         }
1194         return 0;
1195 }
1196
1197 static int
1198 check_port_config(void)
1199 {
1200         unsigned portid;
1201         uint16_t i;
1202
1203         for (i = 0; i < nb_lcore_params; ++i) {
1204                 portid = lcore_params[i].port_id;
1205                 if ((enabled_port_mask & (1 << portid)) == 0) {
1206                         printf("port %u is not enabled in port mask\n",
1207                                                                 portid);
1208                         return -1;
1209                 }
1210                 if (!rte_eth_dev_is_valid_port(portid)) {
1211                         printf("port %u is not present on the board\n",
1212                                                                 portid);
1213                         return -1;
1214                 }
1215         }
1216         return 0;
1217 }
1218
1219 static uint8_t
1220 get_port_n_rx_queues(const uint16_t port)
1221 {
1222         int queue = -1;
1223         uint16_t i;
1224
1225         for (i = 0; i < nb_lcore_params; ++i) {
1226                 if (lcore_params[i].port_id == port &&
1227                                 lcore_params[i].queue_id > queue)
1228                         queue = lcore_params[i].queue_id;
1229         }
1230         return (uint8_t)(++queue);
1231 }
1232
1233 static int
1234 init_lcore_rx_queues(void)
1235 {
1236         uint16_t i, nb_rx_queue;
1237         uint8_t lcore;
1238
1239         for (i = 0; i < nb_lcore_params; ++i) {
1240                 lcore = lcore_params[i].lcore_id;
1241                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1242                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1243                         printf("error: too many queues (%u) for lcore: %u\n",
1244                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1245                         return -1;
1246                 } else {
1247                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1248                                 lcore_params[i].port_id;
1249                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1250                                 lcore_params[i].queue_id;
1251                         lcore_conf[lcore].n_rx_queue++;
1252                 }
1253         }
1254         return 0;
1255 }
1256
1257 /* display usage */
1258 static void
1259 print_usage(const char *prgname)
1260 {
1261         printf ("%s [EAL options] -- -p PORTMASK -P"
1262                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1263                 "  [--high-perf-cores CORELIST"
1264                 "  [--perf-config (port,queue,hi_perf,lcore_index)[,(port,queue,hi_perf,lcore_index]]"
1265                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1266                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1267                 "  -P : enable promiscuous mode\n"
1268                 "  --config (port,queue,lcore): rx queues configuration\n"
1269                 "  --high-perf-cores CORELIST: list of high performance cores\n"
1270                 "  --perf-config: similar as config, cores specified as indices"
1271                 " for bins containing high or regular performance cores\n"
1272                 "  --no-numa: optional, disable numa awareness\n"
1273                 "  --enable-jumbo: enable jumbo frame"
1274                 " which max packet len is PKTLEN in decimal (64-9600)\n"
1275                 "  --parse-ptype: parse packet type by software\n"
1276                 "  --empty-poll: enable empty poll detection"
1277                 " follow (training_flag, high_threshold, med_threshold)\n",
1278                 prgname);
1279 }
1280
1281 static int parse_max_pkt_len(const char *pktlen)
1282 {
1283         char *end = NULL;
1284         unsigned long len;
1285
1286         /* parse decimal string */
1287         len = strtoul(pktlen, &end, 10);
1288         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1289                 return -1;
1290
1291         if (len == 0)
1292                 return -1;
1293
1294         return len;
1295 }
1296
1297 static int
1298 parse_portmask(const char *portmask)
1299 {
1300         char *end = NULL;
1301         unsigned long pm;
1302
1303         /* parse hexadecimal string */
1304         pm = strtoul(portmask, &end, 16);
1305         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1306                 return -1;
1307
1308         if (pm == 0)
1309                 return -1;
1310
1311         return pm;
1312 }
1313
1314 static int
1315 parse_config(const char *q_arg)
1316 {
1317         char s[256];
1318         const char *p, *p0 = q_arg;
1319         char *end;
1320         enum fieldnames {
1321                 FLD_PORT = 0,
1322                 FLD_QUEUE,
1323                 FLD_LCORE,
1324                 _NUM_FLD
1325         };
1326         unsigned long int_fld[_NUM_FLD];
1327         char *str_fld[_NUM_FLD];
1328         int i;
1329         unsigned size;
1330
1331         nb_lcore_params = 0;
1332
1333         while ((p = strchr(p0,'(')) != NULL) {
1334                 ++p;
1335                 if((p0 = strchr(p,')')) == NULL)
1336                         return -1;
1337
1338                 size = p0 - p;
1339                 if(size >= sizeof(s))
1340                         return -1;
1341
1342                 snprintf(s, sizeof(s), "%.*s", size, p);
1343                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1344                                                                 _NUM_FLD)
1345                         return -1;
1346                 for (i = 0; i < _NUM_FLD; i++){
1347                         errno = 0;
1348                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1349                         if (errno != 0 || end == str_fld[i] || int_fld[i] >
1350                                                                         255)
1351                                 return -1;
1352                 }
1353                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1354                         printf("exceeded max number of lcore params: %hu\n",
1355                                 nb_lcore_params);
1356                         return -1;
1357                 }
1358                 lcore_params_array[nb_lcore_params].port_id =
1359                                 (uint8_t)int_fld[FLD_PORT];
1360                 lcore_params_array[nb_lcore_params].queue_id =
1361                                 (uint8_t)int_fld[FLD_QUEUE];
1362                 lcore_params_array[nb_lcore_params].lcore_id =
1363                                 (uint8_t)int_fld[FLD_LCORE];
1364                 ++nb_lcore_params;
1365         }
1366         lcore_params = lcore_params_array;
1367
1368         return 0;
1369 }
1370 static int
1371 parse_ep_config(const char *q_arg)
1372 {
1373         char s[256];
1374         const char *p = q_arg;
1375         char *end;
1376         int  num_arg;
1377
1378         char *str_fld[3];
1379
1380         int training_flag;
1381         int med_edpi;
1382         int hgh_edpi;
1383
1384         ep_med_edpi = EMPTY_POLL_MED_THRESHOLD;
1385         ep_hgh_edpi = EMPTY_POLL_MED_THRESHOLD;
1386
1387         strlcpy(s, p, sizeof(s));
1388
1389         num_arg = rte_strsplit(s, sizeof(s), str_fld, 3, ',');
1390
1391         empty_poll_train = false;
1392
1393         if (num_arg == 0)
1394                 return 0;
1395
1396         if (num_arg == 3) {
1397
1398                 training_flag = strtoul(str_fld[0], &end, 0);
1399                 med_edpi = strtoul(str_fld[1], &end, 0);
1400                 hgh_edpi = strtoul(str_fld[2], &end, 0);
1401
1402                 if (training_flag == 1)
1403                         empty_poll_train = true;
1404
1405                 if (med_edpi > 0)
1406                         ep_med_edpi = med_edpi;
1407
1408                 if (med_edpi > 0)
1409                         ep_hgh_edpi = hgh_edpi;
1410
1411         } else {
1412
1413                 return -1;
1414         }
1415
1416         return 0;
1417
1418 }
1419 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1420
1421 /* Parse the argument given in the command line of the application */
1422 static int
1423 parse_args(int argc, char **argv)
1424 {
1425         int opt, ret;
1426         char **argvopt;
1427         int option_index;
1428         uint32_t limit;
1429         char *prgname = argv[0];
1430         static struct option lgopts[] = {
1431                 {"config", 1, 0, 0},
1432                 {"perf-config", 1, 0, 0},
1433                 {"high-perf-cores", 1, 0, 0},
1434                 {"no-numa", 0, 0, 0},
1435                 {"enable-jumbo", 0, 0, 0},
1436                 {"empty-poll", 1, 0, 0},
1437                 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
1438                 {NULL, 0, 0, 0}
1439         };
1440
1441         argvopt = argv;
1442
1443         while ((opt = getopt_long(argc, argvopt, "p:l:m:h:P",
1444                                 lgopts, &option_index)) != EOF) {
1445
1446                 switch (opt) {
1447                 /* portmask */
1448                 case 'p':
1449                         enabled_port_mask = parse_portmask(optarg);
1450                         if (enabled_port_mask == 0) {
1451                                 printf("invalid portmask\n");
1452                                 print_usage(prgname);
1453                                 return -1;
1454                         }
1455                         break;
1456                 case 'P':
1457                         printf("Promiscuous mode selected\n");
1458                         promiscuous_on = 1;
1459                         break;
1460                 case 'l':
1461                         limit = parse_max_pkt_len(optarg);
1462                         freq_tlb[LOW] = limit;
1463                         break;
1464                 case 'm':
1465                         limit = parse_max_pkt_len(optarg);
1466                         freq_tlb[MED] = limit;
1467                         break;
1468                 case 'h':
1469                         limit = parse_max_pkt_len(optarg);
1470                         freq_tlb[HGH] = limit;
1471                         break;
1472                 /* long options */
1473                 case 0:
1474                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1475                                 ret = parse_config(optarg);
1476                                 if (ret) {
1477                                         printf("invalid config\n");
1478                                         print_usage(prgname);
1479                                         return -1;
1480                                 }
1481                         }
1482
1483                         if (!strncmp(lgopts[option_index].name,
1484                                         "perf-config", 11)) {
1485                                 ret = parse_perf_config(optarg);
1486                                 if (ret) {
1487                                         printf("invalid perf-config\n");
1488                                         print_usage(prgname);
1489                                         return -1;
1490                                 }
1491                         }
1492
1493                         if (!strncmp(lgopts[option_index].name,
1494                                         "high-perf-cores", 15)) {
1495                                 ret = parse_perf_core_list(optarg);
1496                                 if (ret) {
1497                                         printf("invalid high-perf-cores\n");
1498                                         print_usage(prgname);
1499                                         return -1;
1500                                 }
1501                         }
1502
1503                         if (!strncmp(lgopts[option_index].name,
1504                                                 "no-numa", 7)) {
1505                                 printf("numa is disabled \n");
1506                                 numa_on = 0;
1507                         }
1508
1509                         if (!strncmp(lgopts[option_index].name,
1510                                                 "empty-poll", 10)) {
1511                                 printf("empty-poll is enabled\n");
1512                                 empty_poll_on = true;
1513                                 ret = parse_ep_config(optarg);
1514
1515                                 if (ret) {
1516                                         printf("invalid empty poll config\n");
1517                                         print_usage(prgname);
1518                                         return -1;
1519                                 }
1520
1521                         }
1522
1523                         if (!strncmp(lgopts[option_index].name,
1524                                         "enable-jumbo", 12)) {
1525                                 struct option lenopts =
1526                                         {"max-pkt-len", required_argument, \
1527                                                                         0, 0};
1528
1529                                 printf("jumbo frame is enabled \n");
1530                                 port_conf.rxmode.offloads |=
1531                                                 DEV_RX_OFFLOAD_JUMBO_FRAME;
1532                                 port_conf.txmode.offloads |=
1533                                                 DEV_TX_OFFLOAD_MULTI_SEGS;
1534
1535                                 /**
1536                                  * if no max-pkt-len set, use the default value
1537                                  * ETHER_MAX_LEN
1538                                  */
1539                                 if (0 == getopt_long(argc, argvopt, "",
1540                                                 &lenopts, &option_index)) {
1541                                         ret = parse_max_pkt_len(optarg);
1542                                         if ((ret < 64) ||
1543                                                 (ret > MAX_JUMBO_PKT_LEN)){
1544                                                 printf("invalid packet "
1545                                                                 "length\n");
1546                                                 print_usage(prgname);
1547                                                 return -1;
1548                                         }
1549                                         port_conf.rxmode.max_rx_pkt_len = ret;
1550                                 }
1551                                 printf("set jumbo frame "
1552                                         "max packet length to %u\n",
1553                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1554                         }
1555
1556                         if (!strncmp(lgopts[option_index].name,
1557                                      CMD_LINE_OPT_PARSE_PTYPE,
1558                                      sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
1559                                 printf("soft parse-ptype is enabled\n");
1560                                 parse_ptype = 1;
1561                         }
1562
1563                         break;
1564
1565                 default:
1566                         print_usage(prgname);
1567                         return -1;
1568                 }
1569         }
1570
1571         if (optind >= 0)
1572                 argv[optind-1] = prgname;
1573
1574         ret = optind-1;
1575         optind = 1; /* reset getopt lib */
1576         return ret;
1577 }
1578
1579 static void
1580 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1581 {
1582         char buf[ETHER_ADDR_FMT_SIZE];
1583         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1584         printf("%s%s", name, buf);
1585 }
1586
1587 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1588 static void
1589 setup_hash(int socketid)
1590 {
1591         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1592                 .name = NULL,
1593                 .entries = L3FWD_HASH_ENTRIES,
1594                 .key_len = sizeof(struct ipv4_5tuple),
1595                 .hash_func = DEFAULT_HASH_FUNC,
1596                 .hash_func_init_val = 0,
1597         };
1598
1599         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1600                 .name = NULL,
1601                 .entries = L3FWD_HASH_ENTRIES,
1602                 .key_len = sizeof(struct ipv6_5tuple),
1603                 .hash_func = DEFAULT_HASH_FUNC,
1604                 .hash_func_init_val = 0,
1605         };
1606
1607         unsigned i;
1608         int ret;
1609         char s[64];
1610
1611         /* create ipv4 hash */
1612         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1613         ipv4_l3fwd_hash_params.name = s;
1614         ipv4_l3fwd_hash_params.socket_id = socketid;
1615         ipv4_l3fwd_lookup_struct[socketid] =
1616                 rte_hash_create(&ipv4_l3fwd_hash_params);
1617         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1618                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1619                                 "socket %d\n", socketid);
1620
1621         /* create ipv6 hash */
1622         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1623         ipv6_l3fwd_hash_params.name = s;
1624         ipv6_l3fwd_hash_params.socket_id = socketid;
1625         ipv6_l3fwd_lookup_struct[socketid] =
1626                 rte_hash_create(&ipv6_l3fwd_hash_params);
1627         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1628                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1629                                 "socket %d\n", socketid);
1630
1631
1632         /* populate the ipv4 hash */
1633         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1634                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1635                                 (void *) &ipv4_l3fwd_route_array[i].key);
1636                 if (ret < 0) {
1637                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1638                                 "l3fwd hash on socket %d\n", i, socketid);
1639                 }
1640                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1641                 printf("Hash: Adding key\n");
1642                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1643         }
1644
1645         /* populate the ipv6 hash */
1646         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1647                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1648                                 (void *) &ipv6_l3fwd_route_array[i].key);
1649                 if (ret < 0) {
1650                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1651                                 "l3fwd hash on socket %d\n", i, socketid);
1652                 }
1653                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1654                 printf("Hash: Adding key\n");
1655                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1656         }
1657 }
1658 #endif
1659
1660 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1661 static void
1662 setup_lpm(int socketid)
1663 {
1664         unsigned i;
1665         int ret;
1666         char s[64];
1667
1668         /* create the LPM table */
1669         struct rte_lpm_config lpm_ipv4_config;
1670
1671         lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
1672         lpm_ipv4_config.number_tbl8s = 256;
1673         lpm_ipv4_config.flags = 0;
1674
1675         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1676         ipv4_l3fwd_lookup_struct[socketid] =
1677                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
1678         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1679                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1680                                 " on socket %d\n", socketid);
1681
1682         /* populate the LPM table */
1683         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1684                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1685                         ipv4_l3fwd_route_array[i].ip,
1686                         ipv4_l3fwd_route_array[i].depth,
1687                         ipv4_l3fwd_route_array[i].if_out);
1688
1689                 if (ret < 0) {
1690                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1691                                 "l3fwd LPM table on socket %d\n",
1692                                 i, socketid);
1693                 }
1694
1695                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1696                         (unsigned)ipv4_l3fwd_route_array[i].ip,
1697                         ipv4_l3fwd_route_array[i].depth,
1698                         ipv4_l3fwd_route_array[i].if_out);
1699         }
1700 }
1701 #endif
1702
1703 static int
1704 init_mem(unsigned nb_mbuf)
1705 {
1706         struct lcore_conf *qconf;
1707         int socketid;
1708         unsigned lcore_id;
1709         char s[64];
1710
1711         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1712                 if (rte_lcore_is_enabled(lcore_id) == 0)
1713                         continue;
1714
1715                 if (numa_on)
1716                         socketid = rte_lcore_to_socket_id(lcore_id);
1717                 else
1718                         socketid = 0;
1719
1720                 if (socketid >= NB_SOCKETS) {
1721                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1722                                         "out of range %d\n", socketid,
1723                                                 lcore_id, NB_SOCKETS);
1724                 }
1725                 if (pktmbuf_pool[socketid] == NULL) {
1726                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1727                         pktmbuf_pool[socketid] =
1728                                 rte_pktmbuf_pool_create(s, nb_mbuf,
1729                                         MEMPOOL_CACHE_SIZE, 0,
1730                                         RTE_MBUF_DEFAULT_BUF_SIZE,
1731                                         socketid);
1732                         if (pktmbuf_pool[socketid] == NULL)
1733                                 rte_exit(EXIT_FAILURE,
1734                                         "Cannot init mbuf pool on socket %d\n",
1735                                                                 socketid);
1736                         else
1737                                 printf("Allocated mbuf pool on socket %d\n",
1738                                                                 socketid);
1739
1740 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1741                         setup_lpm(socketid);
1742 #else
1743                         setup_hash(socketid);
1744 #endif
1745                 }
1746                 qconf = &lcore_conf[lcore_id];
1747                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1748 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1749                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1750 #endif
1751         }
1752         return 0;
1753 }
1754
1755 /* Check the link status of all ports in up to 9s, and print them finally */
1756 static void
1757 check_all_ports_link_status(uint32_t port_mask)
1758 {
1759 #define CHECK_INTERVAL 100 /* 100ms */
1760 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1761         uint8_t count, all_ports_up, print_flag = 0;
1762         uint16_t portid;
1763         struct rte_eth_link link;
1764
1765         printf("\nChecking link status");
1766         fflush(stdout);
1767         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1768                 all_ports_up = 1;
1769                 RTE_ETH_FOREACH_DEV(portid) {
1770                         if ((port_mask & (1 << portid)) == 0)
1771                                 continue;
1772                         memset(&link, 0, sizeof(link));
1773                         rte_eth_link_get_nowait(portid, &link);
1774                         /* print link status if flag set */
1775                         if (print_flag == 1) {
1776                                 if (link.link_status)
1777                                         printf("Port %d Link Up - speed %u "
1778                                                 "Mbps - %s\n", (uint8_t)portid,
1779                                                 (unsigned)link.link_speed,
1780                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1781                                         ("full-duplex") : ("half-duplex\n"));
1782                                 else
1783                                         printf("Port %d Link Down\n",
1784                                                 (uint8_t)portid);
1785                                 continue;
1786                         }
1787                         /* clear all_ports_up flag if any link down */
1788                         if (link.link_status == ETH_LINK_DOWN) {
1789                                 all_ports_up = 0;
1790                                 break;
1791                         }
1792                 }
1793                 /* after finally printing all link status, get out */
1794                 if (print_flag == 1)
1795                         break;
1796
1797                 if (all_ports_up == 0) {
1798                         printf(".");
1799                         fflush(stdout);
1800                         rte_delay_ms(CHECK_INTERVAL);
1801                 }
1802
1803                 /* set the print_flag if all ports up or timeout */
1804                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1805                         print_flag = 1;
1806                         printf("done\n");
1807                 }
1808         }
1809 }
1810
1811 static int check_ptype(uint16_t portid)
1812 {
1813         int i, ret;
1814         int ptype_l3_ipv4 = 0;
1815 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1816         int ptype_l3_ipv6 = 0;
1817 #endif
1818         uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
1819
1820         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
1821         if (ret <= 0)
1822                 return 0;
1823
1824         uint32_t ptypes[ret];
1825
1826         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
1827         for (i = 0; i < ret; ++i) {
1828                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
1829                         ptype_l3_ipv4 = 1;
1830 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1831                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
1832                         ptype_l3_ipv6 = 1;
1833 #endif
1834         }
1835
1836         if (ptype_l3_ipv4 == 0)
1837                 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
1838
1839 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1840         if (ptype_l3_ipv6 == 0)
1841                 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
1842 #endif
1843
1844 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1845         if (ptype_l3_ipv4)
1846 #else /* APP_LOOKUP_EXACT_MATCH */
1847         if (ptype_l3_ipv4 && ptype_l3_ipv6)
1848 #endif
1849                 return 1;
1850
1851         return 0;
1852
1853 }
1854
1855 static int
1856 init_power_library(void)
1857 {
1858         int ret = 0, lcore_id;
1859         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1860                 if (rte_lcore_is_enabled(lcore_id)) {
1861                         /* init power management library */
1862                         ret = rte_power_init(lcore_id);
1863                         if (ret)
1864                                 RTE_LOG(ERR, POWER,
1865                                 "Library initialization failed on core %u\n",
1866                                 lcore_id);
1867                 }
1868         }
1869         return ret;
1870 }
1871 static void
1872 empty_poll_setup_timer(void)
1873 {
1874         int lcore_id = rte_lcore_id();
1875         uint64_t hz = rte_get_timer_hz();
1876
1877         struct  ep_params *ep_ptr = ep_params;
1878
1879         ep_ptr->interval_ticks = hz / INTERVALS_PER_SECOND;
1880
1881         rte_timer_reset_sync(&ep_ptr->timer0,
1882                         ep_ptr->interval_ticks,
1883                         PERIODICAL,
1884                         lcore_id,
1885                         rte_empty_poll_detection,
1886                         (void *)ep_ptr);
1887
1888 }
1889 static int
1890 launch_timer(unsigned int lcore_id)
1891 {
1892         int64_t prev_tsc = 0, cur_tsc, diff_tsc, cycles_10ms;
1893
1894         RTE_SET_USED(lcore_id);
1895
1896
1897         if (rte_get_master_lcore() != lcore_id) {
1898                 rte_panic("timer on lcore:%d which is not master core:%d\n",
1899                                 lcore_id,
1900                                 rte_get_master_lcore());
1901         }
1902
1903         RTE_LOG(INFO, POWER, "Bring up the Timer\n");
1904
1905         empty_poll_setup_timer();
1906
1907         cycles_10ms = rte_get_timer_hz() / 100;
1908
1909         while (!is_done()) {
1910                 cur_tsc = rte_rdtsc();
1911                 diff_tsc = cur_tsc - prev_tsc;
1912                 if (diff_tsc > cycles_10ms) {
1913                         rte_timer_manage();
1914                         prev_tsc = cur_tsc;
1915                         cycles_10ms = rte_get_timer_hz() / 100;
1916                 }
1917         }
1918
1919         RTE_LOG(INFO, POWER, "Timer_subsystem is done\n");
1920
1921         return 0;
1922 }
1923
1924
1925 int
1926 main(int argc, char **argv)
1927 {
1928         struct lcore_conf *qconf;
1929         struct rte_eth_dev_info dev_info;
1930         struct rte_eth_txconf *txconf;
1931         int ret;
1932         uint16_t nb_ports;
1933         uint16_t queueid;
1934         unsigned lcore_id;
1935         uint64_t hz;
1936         uint32_t n_tx_queue, nb_lcores;
1937         uint32_t dev_rxq_num, dev_txq_num;
1938         uint8_t nb_rx_queue, queue, socketid;
1939         uint16_t portid;
1940
1941         /* catch SIGINT and restore cpufreq governor to ondemand */
1942         signal(SIGINT, signal_exit_now);
1943
1944         /* init EAL */
1945         ret = rte_eal_init(argc, argv);
1946         if (ret < 0)
1947                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1948         argc -= ret;
1949         argv += ret;
1950
1951         /* init RTE timer library to be used late */
1952         rte_timer_subsystem_init();
1953
1954         /* parse application arguments (after the EAL ones) */
1955         ret = parse_args(argc, argv);
1956         if (ret < 0)
1957                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1958
1959         if (init_power_library())
1960                 RTE_LOG(ERR, L3FWD_POWER, "init_power_library failed\n");
1961
1962         if (update_lcore_params() < 0)
1963                 rte_exit(EXIT_FAILURE, "update_lcore_params failed\n");
1964
1965         if (check_lcore_params() < 0)
1966                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1967
1968         ret = init_lcore_rx_queues();
1969         if (ret < 0)
1970                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1971
1972         nb_ports = rte_eth_dev_count_avail();
1973
1974         if (check_port_config() < 0)
1975                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1976
1977         nb_lcores = rte_lcore_count();
1978
1979         /* initialize all ports */
1980         RTE_ETH_FOREACH_DEV(portid) {
1981                 struct rte_eth_conf local_port_conf = port_conf;
1982
1983                 /* skip ports that are not enabled */
1984                 if ((enabled_port_mask & (1 << portid)) == 0) {
1985                         printf("\nSkipping disabled port %d\n", portid);
1986                         continue;
1987                 }
1988
1989                 /* init port */
1990                 printf("Initializing port %d ... ", portid );
1991                 fflush(stdout);
1992
1993                 rte_eth_dev_info_get(portid, &dev_info);
1994                 dev_rxq_num = dev_info.max_rx_queues;
1995                 dev_txq_num = dev_info.max_tx_queues;
1996
1997                 nb_rx_queue = get_port_n_rx_queues(portid);
1998                 if (nb_rx_queue > dev_rxq_num)
1999                         rte_exit(EXIT_FAILURE,
2000                                 "Cannot configure not existed rxq: "
2001                                 "port=%d\n", portid);
2002
2003                 n_tx_queue = nb_lcores;
2004                 if (n_tx_queue > dev_txq_num)
2005                         n_tx_queue = dev_txq_num;
2006                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
2007                         nb_rx_queue, (unsigned)n_tx_queue );
2008                 /* If number of Rx queue is 0, no need to enable Rx interrupt */
2009                 if (nb_rx_queue == 0)
2010                         local_port_conf.intr_conf.rxq = 0;
2011                 rte_eth_dev_info_get(portid, &dev_info);
2012                 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
2013                         local_port_conf.txmode.offloads |=
2014                                 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
2015
2016                 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2017                         dev_info.flow_type_rss_offloads;
2018                 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2019                                 port_conf.rx_adv_conf.rss_conf.rss_hf) {
2020                         printf("Port %u modified RSS hash function based on hardware support,"
2021                                 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2022                                 portid,
2023                                 port_conf.rx_adv_conf.rss_conf.rss_hf,
2024                                 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2025                 }
2026
2027                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
2028                                         (uint16_t)n_tx_queue, &local_port_conf);
2029                 if (ret < 0)
2030                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
2031                                         "err=%d, port=%d\n", ret, portid);
2032
2033                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
2034                                                        &nb_txd);
2035                 if (ret < 0)
2036                         rte_exit(EXIT_FAILURE,
2037                                  "Cannot adjust number of descriptors: err=%d, port=%d\n",
2038                                  ret, portid);
2039
2040                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
2041                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
2042                 printf(", ");
2043
2044                 /* init memory */
2045                 ret = init_mem(NB_MBUF);
2046                 if (ret < 0)
2047                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
2048
2049                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2050                         if (rte_lcore_is_enabled(lcore_id) == 0)
2051                                 continue;
2052
2053                         /* Initialize TX buffers */
2054                         qconf = &lcore_conf[lcore_id];
2055                         qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
2056                                 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
2057                                 rte_eth_dev_socket_id(portid));
2058                         if (qconf->tx_buffer[portid] == NULL)
2059                                 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
2060                                                  portid);
2061
2062                         rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
2063                 }
2064
2065                 /* init one TX queue per couple (lcore,port) */
2066                 queueid = 0;
2067                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2068                         if (rte_lcore_is_enabled(lcore_id) == 0)
2069                                 continue;
2070
2071                         if (queueid >= dev_txq_num)
2072                                 continue;
2073
2074                         if (numa_on)
2075                                 socketid = \
2076                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
2077                         else
2078                                 socketid = 0;
2079
2080                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
2081                         fflush(stdout);
2082
2083                         txconf = &dev_info.default_txconf;
2084                         txconf->offloads = local_port_conf.txmode.offloads;
2085                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
2086                                                      socketid, txconf);
2087                         if (ret < 0)
2088                                 rte_exit(EXIT_FAILURE,
2089                                         "rte_eth_tx_queue_setup: err=%d, "
2090                                                 "port=%d\n", ret, portid);
2091
2092                         qconf = &lcore_conf[lcore_id];
2093                         qconf->tx_queue_id[portid] = queueid;
2094                         queueid++;
2095
2096                         qconf->tx_port_id[qconf->n_tx_port] = portid;
2097                         qconf->n_tx_port++;
2098                 }
2099                 printf("\n");
2100         }
2101
2102         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2103                 if (rte_lcore_is_enabled(lcore_id) == 0)
2104                         continue;
2105
2106                 if (empty_poll_on == false) {
2107                         /* init timer structures for each enabled lcore */
2108                         rte_timer_init(&power_timers[lcore_id]);
2109                         hz = rte_get_timer_hz();
2110                         rte_timer_reset(&power_timers[lcore_id],
2111                                         hz/TIMER_NUMBER_PER_SECOND,
2112                                         SINGLE, lcore_id,
2113                                         power_timer_cb, NULL);
2114                 }
2115                 qconf = &lcore_conf[lcore_id];
2116                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
2117                 fflush(stdout);
2118                 /* init RX queues */
2119                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
2120                         struct rte_eth_rxconf rxq_conf;
2121                         struct rte_eth_dev *dev;
2122                         struct rte_eth_conf *conf;
2123
2124                         portid = qconf->rx_queue_list[queue].port_id;
2125                         queueid = qconf->rx_queue_list[queue].queue_id;
2126                         dev = &rte_eth_devices[portid];
2127                         conf = &dev->data->dev_conf;
2128
2129                         if (numa_on)
2130                                 socketid = \
2131                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
2132                         else
2133                                 socketid = 0;
2134
2135                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
2136                         fflush(stdout);
2137
2138                         rte_eth_dev_info_get(portid, &dev_info);
2139                         rxq_conf = dev_info.default_rxconf;
2140                         rxq_conf.offloads = conf->rxmode.offloads;
2141                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
2142                                 socketid, &rxq_conf,
2143                                 pktmbuf_pool[socketid]);
2144                         if (ret < 0)
2145                                 rte_exit(EXIT_FAILURE,
2146                                         "rte_eth_rx_queue_setup: err=%d, "
2147                                                 "port=%d\n", ret, portid);
2148
2149                         if (parse_ptype) {
2150                                 if (add_cb_parse_ptype(portid, queueid) < 0)
2151                                         rte_exit(EXIT_FAILURE,
2152                                                  "Fail to add ptype cb\n");
2153                         } else if (!check_ptype(portid))
2154                                 rte_exit(EXIT_FAILURE,
2155                                          "PMD can not provide needed ptypes\n");
2156                 }
2157         }
2158
2159         printf("\n");
2160
2161         /* start ports */
2162         RTE_ETH_FOREACH_DEV(portid) {
2163                 if ((enabled_port_mask & (1 << portid)) == 0) {
2164                         continue;
2165                 }
2166                 /* Start device */
2167                 ret = rte_eth_dev_start(portid);
2168                 if (ret < 0)
2169                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
2170                                                 "port=%d\n", ret, portid);
2171                 /*
2172                  * If enabled, put device in promiscuous mode.
2173                  * This allows IO forwarding mode to forward packets
2174                  * to itself through 2 cross-connected  ports of the
2175                  * target machine.
2176                  */
2177                 if (promiscuous_on)
2178                         rte_eth_promiscuous_enable(portid);
2179                 /* initialize spinlock for each port */
2180                 rte_spinlock_init(&(locks[portid]));
2181         }
2182
2183         check_all_ports_link_status(enabled_port_mask);
2184
2185         if (empty_poll_on == true) {
2186
2187                 if (empty_poll_train) {
2188                         policy.state = TRAINING;
2189                 } else {
2190                         policy.state = MED_NORMAL;
2191                         policy.med_base_edpi = ep_med_edpi;
2192                         policy.hgh_base_edpi = ep_hgh_edpi;
2193                 }
2194
2195                 ret = rte_power_empty_poll_stat_init(&ep_params,
2196                                 freq_tlb,
2197                                 &policy);
2198                 if (ret < 0)
2199                         rte_exit(EXIT_FAILURE, "empty poll init failed");
2200         }
2201
2202
2203         /* launch per-lcore init on every lcore */
2204         if (empty_poll_on == false) {
2205                 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2206         } else {
2207                 empty_poll_stop = false;
2208                 rte_eal_mp_remote_launch(main_empty_poll_loop, NULL,
2209                                 SKIP_MASTER);
2210         }
2211
2212         if (empty_poll_on == true)
2213                 launch_timer(rte_lcore_id());
2214
2215         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2216                 if (rte_eal_wait_lcore(lcore_id) < 0)
2217                         return -1;
2218         }
2219
2220         if (empty_poll_on)
2221                 rte_power_empty_poll_stat_free();
2222
2223         return 0;
2224 }