examples/performance-thread/l3fwd-thread/main.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2016 Intel Corporation
   3  */
   4
   5 #ifndef _GNU_SOURCE
   6 #define _GNU_SOURCE
   7 #endif
   8
   9 #include <stdio.h>
  10 #include <stdlib.h>
  11 #include <stdint.h>
  12 #include <inttypes.h>
  13 #include <sys/types.h>
  14 #include <string.h>
  15 #include <sys/queue.h>
  16 #include <stdarg.h>
  17 #include <errno.h>
  18 #include <getopt.h>
  19 #include <sched.h>
  20
  21 #include <rte_common.h>
  22 #include <rte_vect.h>
  23 #include <rte_byteorder.h>
  24 #include <rte_log.h>
  25 #include <rte_memory.h>
  26 #include <rte_memcpy.h>
  27 #include <rte_eal.h>
  28 #include <rte_launch.h>
  29 #include <rte_atomic.h>
  30 #include <rte_cycles.h>
  31 #include <rte_prefetch.h>
  32 #include <rte_lcore.h>
  33 #include <rte_per_lcore.h>
  34 #include <rte_branch_prediction.h>
  35 #include <rte_interrupts.h>
  36 #include <rte_random.h>
  37 #include <rte_debug.h>
  38 #include <rte_ether.h>
  39 #include <rte_ethdev.h>
  40 #include <rte_ring.h>
  41 #include <rte_mempool.h>
  42 #include <rte_mbuf.h>
  43 #include <rte_ip.h>
  44 #include <rte_tcp.h>
  45 #include <rte_udp.h>
  46 #include <rte_string_fns.h>
  47 #include <rte_pause.h>
  48 #include <rte_timer.h>
  49
  50 #include <cmdline_parse.h>
  51 #include <cmdline_parse_etheraddr.h>
  52
  53 #include <lthread_api.h>
  54
  55 #define APP_LOOKUP_EXACT_MATCH          0
  56 #define APP_LOOKUP_LPM                  1
  57 #define DO_RFC_1812_CHECKS
  58
  59 /* Enable cpu-load stats 0-off, 1-on */
  60 #define APP_CPU_LOAD                 1
  61
  62 #ifndef APP_LOOKUP_METHOD
  63 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  64 #endif
  65
  66 #ifndef __GLIBC__ /* sched_getcpu() is glibc specific */
  67 #define sched_getcpu() rte_lcore_id()
  68 #endif
  69
  70 static int
  71 check_ptype(int portid)
  72 {
  73         int i, ret;
  74         int ipv4 = 0, ipv6 = 0;
  75
  76         ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK, NULL,
  77                         0);
  78         if (ret <= 0)
  79                 return 0;
  80
  81         uint32_t ptypes[ret];
  82
  83         ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK,
  84                         ptypes, ret);
  85         for (i = 0; i < ret; ++i) {
  86                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
  87                         ipv4 = 1;
  88                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
  89                         ipv6 = 1;
  90         }
  91
  92         if (ipv4 && ipv6)
  93                 return 1;
  94
  95         return 0;
  96 }
  97
  98 static inline void
  99 parse_ptype(struct rte_mbuf *m)
 100 {
 101         struct rte_ether_hdr *eth_hdr;
 102         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 103         uint16_t ether_type;
 104
 105         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 106         ether_type = eth_hdr->ether_type;
 107         if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
 108                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 109         else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6))
 110                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 111
 112         m->packet_type = packet_type;
 113 }
 114
 115 static uint16_t
 116 cb_parse_ptype(__rte_unused uint16_t port, __rte_unused uint16_t queue,
 117                 struct rte_mbuf *pkts[], uint16_t nb_pkts,
 118                 __rte_unused uint16_t max_pkts, __rte_unused void *user_param)
 119 {
 120         unsigned int i;
 121
 122         for (i = 0; i < nb_pkts; i++)
 123                 parse_ptype(pkts[i]);
 124
 125         return nb_pkts;
 126 }
 127
 128 /*
 129  *  When set to zero, simple forwaring path is eanbled.
 130  *  When set to one, optimized forwarding path is enabled.
 131  *  Note that LPM optimisation path uses SSE4.1 instructions.
 132  */
 133 #define ENABLE_MULTI_BUFFER_OPTIMIZE    1
 134
 135 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 136 #include <rte_hash.h>
 137 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 138 #include <rte_lpm.h>
 139 #include <rte_lpm6.h>
 140 #else
 141 #error "APP_LOOKUP_METHOD set to incorrect value"
 142 #endif
 143
 144 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
 145
 146 #define MAX_JUMBO_PKT_LEN  9600
 147
 148 #define IPV6_ADDR_LEN 16
 149
 150 #define MEMPOOL_CACHE_SIZE 256
 151
 152 /*
 153  * This expression is used to calculate the number of mbufs needed depending on
 154  * user input, taking into account memory for rx and tx hardware rings, cache
 155  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 156  * NB_MBUF never goes below a minimum value of 8192
 157  */
 158
 159 #define NB_MBUF RTE_MAX(\
 160                 (nb_ports*nb_rx_queue*nb_rxd +      \
 161                 nb_ports*nb_lcores*MAX_PKT_BURST +  \
 162                 nb_ports*n_tx_queue*nb_txd +        \
 163                 nb_lcores*MEMPOOL_CACHE_SIZE),      \
 164                 (unsigned)8192)
 165
 166 #define MAX_PKT_BURST     32
 167 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 168
 169 /*
 170  * Try to avoid TX buffering if we have at least MAX_TX_BURST packets to send.
 171  */
 172 #define MAX_TX_BURST  (MAX_PKT_BURST / 2)
 173 #define BURST_SIZE    MAX_TX_BURST
 174
 175 #define NB_SOCKETS 8
 176
 177 /* Configure how many packets ahead to prefetch, when reading packets */
 178 #define PREFETCH_OFFSET 3
 179
 180 /* Used to mark destination port as 'invalid'. */
 181 #define BAD_PORT        ((uint16_t)-1)
 182
 183 #define FWDSTEP 4
 184
 185 /*
 186  * Configurable number of RX/TX ring descriptors
 187  */
 188 #define RTE_TEST_RX_DESC_DEFAULT 1024
 189 #define RTE_TEST_TX_DESC_DEFAULT 1024
 190 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 191 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 192
 193 /* ethernet addresses of ports */
 194 static uint64_t dest_eth_addr[RTE_MAX_ETHPORTS];
 195 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 196
 197 static xmm_t val_eth[RTE_MAX_ETHPORTS];
 198
 199 /* replace first 12B of the ethernet header. */
 200 #define MASK_ETH 0x3f
 201
 202 /* mask of enabled ports */
 203 static uint32_t enabled_port_mask;
 204 static int promiscuous_on; /**< Set in promiscuous mode off by default. */
 205 static int numa_on = 1;    /**< NUMA is enabled by default. */
 206 static int parse_ptype_on;
 207
 208 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 209 static int ipv6;           /**< ipv6 is false by default. */
 210 #endif
 211
 212 #if (APP_CPU_LOAD == 1)
 213
 214 #define MAX_CPU RTE_MAX_LCORE
 215 #define CPU_LOAD_TIMEOUT_US (5 * 1000 * 1000)  /**< Timeout for collecting 5s */
 216
 217 #define CPU_PROCESS     0
 218 #define CPU_POLL        1
 219 #define MAX_CPU_COUNTER 2
 220
 221 struct cpu_load {
 222         uint16_t       n_cpu;
 223         uint64_t       counter;
 224         uint64_t       hits[MAX_CPU_COUNTER][MAX_CPU];
 225 } __rte_cache_aligned;
 226
 227 static struct cpu_load cpu_load;
 228 static int cpu_load_lcore_id = -1;
 229
 230 #define SET_CPU_BUSY(thread, counter) \
 231                 thread->conf.busy[counter] = 1
 232
 233 #define SET_CPU_IDLE(thread, counter) \
 234                 thread->conf.busy[counter] = 0
 235
 236 #define IS_CPU_BUSY(thread, counter) \
 237                 (thread->conf.busy[counter] > 0)
 238
 239 #else
 240
 241 #define SET_CPU_BUSY(thread, counter)
 242 #define SET_CPU_IDLE(thread, counter)
 243 #define IS_CPU_BUSY(thread, counter) 0
 244
 245 #endif
 246
 247 struct mbuf_table {
 248         uint16_t len;
 249         struct rte_mbuf *m_table[MAX_PKT_BURST];
 250 };
 251
 252 struct lcore_rx_queue {
 253         uint16_t port_id;
 254         uint8_t queue_id;
 255 } __rte_cache_aligned;
 256
 257 #define MAX_RX_QUEUE_PER_LCORE 16
 258 #define MAX_TX_QUEUE_PER_PORT  RTE_MAX_ETHPORTS
 259 #define MAX_RX_QUEUE_PER_PORT  128
 260
 261 #define MAX_LCORE_PARAMS       1024
 262 struct rx_thread_params {
 263         uint16_t port_id;
 264         uint8_t queue_id;
 265         uint8_t lcore_id;
 266         uint8_t thread_id;
 267 } __rte_cache_aligned;
 268
 269 static struct rx_thread_params rx_thread_params_array[MAX_LCORE_PARAMS];
 270 static struct rx_thread_params rx_thread_params_array_default[] = {
 271         {0, 0, 2, 0},
 272         {0, 1, 2, 1},
 273         {0, 2, 2, 2},
 274         {1, 0, 2, 3},
 275         {1, 1, 2, 4},
 276         {1, 2, 2, 5},
 277         {2, 0, 2, 6},
 278         {3, 0, 3, 7},
 279         {3, 1, 3, 8},
 280 };
 281
 282 static struct rx_thread_params *rx_thread_params =
 283                 rx_thread_params_array_default;
 284 static uint16_t nb_rx_thread_params = RTE_DIM(rx_thread_params_array_default);
 285
 286 struct tx_thread_params {
 287         uint8_t lcore_id;
 288         uint8_t thread_id;
 289 } __rte_cache_aligned;
 290
 291 static struct tx_thread_params tx_thread_params_array[MAX_LCORE_PARAMS];
 292 static struct tx_thread_params tx_thread_params_array_default[] = {
 293         {4, 0},
 294         {5, 1},
 295         {6, 2},
 296         {7, 3},
 297         {8, 4},
 298         {9, 5},
 299         {10, 6},
 300         {11, 7},
 301         {12, 8},
 302 };
 303
 304 static struct tx_thread_params *tx_thread_params =
 305                 tx_thread_params_array_default;
 306 static uint16_t nb_tx_thread_params = RTE_DIM(tx_thread_params_array_default);
 307
 308 static struct rte_eth_conf port_conf = {
 309         .rxmode = {
 310                 .mq_mode = ETH_MQ_RX_RSS,
 311                 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
 312                 .split_hdr_size = 0,
 313                 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
 314         },
 315         .rx_adv_conf = {
 316                 .rss_conf = {
 317                         .rss_key = NULL,
 318                         .rss_hf = ETH_RSS_TCP,
 319                 },
 320         },
 321         .txmode = {
 322                 .mq_mode = ETH_MQ_TX_NONE,
 323         },
 324 };
 325
 326 static struct rte_mempool *pktmbuf_pool[NB_SOCKETS];
 327
 328 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 329
 330 #include <rte_hash_crc.h>
 331 #define DEFAULT_HASH_FUNC       rte_hash_crc
 332
 333 struct ipv4_5tuple {
 334         uint32_t ip_dst;
 335         uint32_t ip_src;
 336         uint16_t port_dst;
 337         uint16_t port_src;
 338         uint8_t  proto;
 339 } __rte_packed;
 340
 341 union ipv4_5tuple_host {
 342         struct {
 343                 uint8_t  pad0;
 344                 uint8_t  proto;
 345                 uint16_t pad1;
 346                 uint32_t ip_src;
 347                 uint32_t ip_dst;
 348                 uint16_t port_src;
 349                 uint16_t port_dst;
 350         };
 351         __m128i xmm;
 352 };
 353
 354 #define XMM_NUM_IN_IPV6_5TUPLE 3
 355
 356 struct ipv6_5tuple {
 357         uint8_t  ip_dst[IPV6_ADDR_LEN];
 358         uint8_t  ip_src[IPV6_ADDR_LEN];
 359         uint16_t port_dst;
 360         uint16_t port_src;
 361         uint8_t  proto;
 362 } __rte_packed;
 363
 364 union ipv6_5tuple_host {
 365         struct {
 366                 uint16_t pad0;
 367                 uint8_t  proto;
 368                 uint8_t  pad1;
 369                 uint8_t  ip_src[IPV6_ADDR_LEN];
 370                 uint8_t  ip_dst[IPV6_ADDR_LEN];
 371                 uint16_t port_src;
 372                 uint16_t port_dst;
 373                 uint64_t reserve;
 374         };
 375         __m128i xmm[XMM_NUM_IN_IPV6_5TUPLE];
 376 };
 377
 378 struct ipv4_l3fwd_route {
 379         struct ipv4_5tuple key;
 380         uint8_t if_out;
 381 };
 382
 383 struct ipv6_l3fwd_route {
 384         struct ipv6_5tuple key;
 385         uint8_t if_out;
 386 };
 387
 388 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 389         {{RTE_IPV4(101, 0, 0, 0), RTE_IPV4(100, 10, 0, 1),  101, 11, IPPROTO_TCP}, 0},
 390         {{RTE_IPV4(201, 0, 0, 0), RTE_IPV4(200, 20, 0, 1),  102, 12, IPPROTO_TCP}, 1},
 391         {{RTE_IPV4(111, 0, 0, 0), RTE_IPV4(100, 30, 0, 1),  101, 11, IPPROTO_TCP}, 2},
 392         {{RTE_IPV4(211, 0, 0, 0), RTE_IPV4(200, 40, 0, 1),  102, 12, IPPROTO_TCP}, 3},
 393 };
 394
 395 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 396         {{
 397         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 398         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 399                         0x05},
 400         101, 11, IPPROTO_TCP}, 0},
 401
 402         {{
 403         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 404         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 405                         0x05},
 406         102, 12, IPPROTO_TCP}, 1},
 407
 408         {{
 409         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 410         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 411                         0x05},
 412         101, 11, IPPROTO_TCP}, 2},
 413
 414         {{
 415         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 416         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 417                         0x05},
 418         102, 12, IPPROTO_TCP}, 3},
 419 };
 420
 421 typedef struct rte_hash lookup_struct_t;
 422 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 423 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 424
 425 #ifdef RTE_ARCH_X86_64
 426 /* default to 4 million hash entries (approx) */
 427 #define L3FWD_HASH_ENTRIES (1024*1024*4)
 428 #else
 429 /* 32-bit has less address-space for hugepage memory, limit to 1M entries */
 430 #define L3FWD_HASH_ENTRIES (1024*1024*1)
 431 #endif
 432 #define HASH_ENTRY_NUMBER_DEFAULT 4
 433
 434 static uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
 435
 436 static inline uint32_t
 437 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
 438                 uint32_t init_val)
 439 {
 440         const union ipv4_5tuple_host *k;
 441         uint32_t t;
 442         const uint32_t *p;
 443
 444         k = data;
 445         t = k->proto;
 446         p = (const uint32_t *)&k->port_src;
 447
 448         init_val = rte_hash_crc_4byte(t, init_val);
 449         init_val = rte_hash_crc_4byte(k->ip_src, init_val);
 450         init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
 451         init_val = rte_hash_crc_4byte(*p, init_val);
 452         return init_val;
 453 }
 454
 455 static inline uint32_t
 456 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len,
 457                 uint32_t init_val)
 458 {
 459         const union ipv6_5tuple_host *k;
 460         uint32_t t;
 461         const uint32_t *p;
 462         const uint32_t *ip_src0, *ip_src1, *ip_src2, *ip_src3;
 463         const uint32_t *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
 464
 465         k = data;
 466         t = k->proto;
 467         p = (const uint32_t *)&k->port_src;
 468
 469         ip_src0 = (const uint32_t *) k->ip_src;
 470         ip_src1 = (const uint32_t *)(k->ip_src + 4);
 471         ip_src2 = (const uint32_t *)(k->ip_src + 8);
 472         ip_src3 = (const uint32_t *)(k->ip_src + 12);
 473         ip_dst0 = (const uint32_t *) k->ip_dst;
 474         ip_dst1 = (const uint32_t *)(k->ip_dst + 4);
 475         ip_dst2 = (const uint32_t *)(k->ip_dst + 8);
 476         ip_dst3 = (const uint32_t *)(k->ip_dst + 12);
 477         init_val = rte_hash_crc_4byte(t, init_val);
 478         init_val = rte_hash_crc_4byte(*ip_src0, init_val);
 479         init_val = rte_hash_crc_4byte(*ip_src1, init_val);
 480         init_val = rte_hash_crc_4byte(*ip_src2, init_val);
 481         init_val = rte_hash_crc_4byte(*ip_src3, init_val);
 482         init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
 483         init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
 484         init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
 485         init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
 486         init_val = rte_hash_crc_4byte(*p, init_val);
 487         return init_val;
 488 }
 489
 490 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
 491 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
 492
 493 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 494 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 495
 496 #endif
 497
 498 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 499 struct ipv4_l3fwd_route {
 500         uint32_t ip;
 501         uint8_t  depth;
 502         uint8_t  if_out;
 503 };
 504
 505 struct ipv6_l3fwd_route {
 506         uint8_t ip[16];
 507         uint8_t depth;
 508         uint8_t if_out;
 509 };
 510
 511 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 512         {RTE_IPV4(1, 1, 1, 0), 24, 0},
 513         {RTE_IPV4(2, 1, 1, 0), 24, 1},
 514         {RTE_IPV4(3, 1, 1, 0), 24, 2},
 515         {RTE_IPV4(4, 1, 1, 0), 24, 3},
 516         {RTE_IPV4(5, 1, 1, 0), 24, 4},
 517         {RTE_IPV4(6, 1, 1, 0), 24, 5},
 518         {RTE_IPV4(7, 1, 1, 0), 24, 6},
 519         {RTE_IPV4(8, 1, 1, 0), 24, 7},
 520 };
 521
 522 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 523         {{1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 0},
 524         {{2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 1},
 525         {{3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 2},
 526         {{4, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 3},
 527         {{5, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 4},
 528         {{6, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 5},
 529         {{7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 6},
 530         {{8, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 7},
 531 };
 532
 533 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
 534 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
 535
 536 #define IPV4_L3FWD_LPM_MAX_RULES         1024
 537 #define IPV6_L3FWD_LPM_MAX_RULES         1024
 538 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
 539
 540 typedef struct rte_lpm lookup_struct_t;
 541 typedef struct rte_lpm6 lookup6_struct_t;
 542 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 543 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 544 #endif
 545
 546 struct lcore_conf {
 547         lookup_struct_t *ipv4_lookup_struct;
 548 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 549         lookup6_struct_t *ipv6_lookup_struct;
 550 #else
 551         lookup_struct_t *ipv6_lookup_struct;
 552 #endif
 553         void *data;
 554 } __rte_cache_aligned;
 555
 556 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 557 RTE_DEFINE_PER_LCORE(struct lcore_conf *, lcore_conf);
 558
 559 #define MAX_RX_QUEUE_PER_THREAD 16
 560 #define MAX_TX_PORT_PER_THREAD  RTE_MAX_ETHPORTS
 561 #define MAX_TX_QUEUE_PER_PORT   RTE_MAX_ETHPORTS
 562 #define MAX_RX_QUEUE_PER_PORT   128
 563
 564 #define MAX_RX_THREAD 1024
 565 #define MAX_TX_THREAD 1024
 566 #define MAX_THREAD    (MAX_RX_THREAD + MAX_TX_THREAD)
 567
 568 /**
 569  * Producers and consumers threads configuration
 570  */
 571 static int lthreads_on = 1; /**< Use lthreads for processing*/
 572
 573 rte_atomic16_t rx_counter;  /**< Number of spawned rx threads */
 574 rte_atomic16_t tx_counter;  /**< Number of spawned tx threads */
 575
 576 struct thread_conf {
 577         uint16_t lcore_id;      /**< Initial lcore for rx thread */
 578         uint16_t cpu_id;        /**< Cpu id for cpu load stats counter */
 579         uint16_t thread_id;     /**< Thread ID */
 580
 581 #if (APP_CPU_LOAD > 0)
 582         int busy[MAX_CPU_COUNTER];
 583 #endif
 584 };
 585
 586 struct thread_rx_conf {
 587         struct thread_conf conf;
 588
 589         uint16_t n_rx_queue;
 590         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 591
 592         uint16_t n_ring;        /**< Number of output rings */
 593         struct rte_ring *ring[RTE_MAX_LCORE];
 594         struct lthread_cond *ready[RTE_MAX_LCORE];
 595
 596 #if (APP_CPU_LOAD > 0)
 597         int busy[MAX_CPU_COUNTER];
 598 #endif
 599 } __rte_cache_aligned;
 600
 601 uint16_t n_rx_thread;
 602 struct thread_rx_conf rx_thread[MAX_RX_THREAD];
 603
 604 struct thread_tx_conf {
 605         struct thread_conf conf;
 606
 607         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 608         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 609
 610         struct rte_ring *ring;
 611         struct lthread_cond **ready;
 612
 613 } __rte_cache_aligned;
 614
 615 uint16_t n_tx_thread;
 616 struct thread_tx_conf tx_thread[MAX_TX_THREAD];
 617
 618 /* Send burst of packets on an output interface */
 619 static inline int
 620 send_burst(struct thread_tx_conf *qconf, uint16_t n, uint16_t port)
 621 {
 622         struct rte_mbuf **m_table;
 623         int ret;
 624         uint16_t queueid;
 625
 626         queueid = qconf->tx_queue_id[port];
 627         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 628
 629         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 630         if (unlikely(ret < n)) {
 631                 do {
 632                         rte_pktmbuf_free(m_table[ret]);
 633                 } while (++ret < n);
 634         }
 635
 636         return 0;
 637 }
 638
 639 /* Enqueue a single packet, and send burst if queue is filled */
 640 static inline int
 641 send_single_packet(struct rte_mbuf *m, uint16_t port)
 642 {
 643         uint16_t len;
 644         struct thread_tx_conf *qconf;
 645
 646         if (lthreads_on)
 647                 qconf = (struct thread_tx_conf *)lthread_get_data();
 648         else
 649                 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
 650
 651         len = qconf->tx_mbufs[port].len;
 652         qconf->tx_mbufs[port].m_table[len] = m;
 653         len++;
 654
 655         /* enough pkts to be sent */
 656         if (unlikely(len == MAX_PKT_BURST)) {
 657                 send_burst(qconf, MAX_PKT_BURST, port);
 658                 len = 0;
 659         }
 660
 661         qconf->tx_mbufs[port].len = len;
 662         return 0;
 663 }
 664
 665 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
 666         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
 667 static __rte_always_inline void
 668 send_packetsx4(uint16_t port,
 669         struct rte_mbuf *m[], uint32_t num)
 670 {
 671         uint32_t len, j, n;
 672         struct thread_tx_conf *qconf;
 673
 674         if (lthreads_on)
 675                 qconf = (struct thread_tx_conf *)lthread_get_data();
 676         else
 677                 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
 678
 679         len = qconf->tx_mbufs[port].len;
 680
 681         /*
 682          * If TX buffer for that queue is empty, and we have enough packets,
 683          * then send them straightway.
 684          */
 685         if (num >= MAX_TX_BURST && len == 0) {
 686                 n = rte_eth_tx_burst(port, qconf->tx_queue_id[port], m, num);
 687                 if (unlikely(n < num)) {
 688                         do {
 689                                 rte_pktmbuf_free(m[n]);
 690                         } while (++n < num);
 691                 }
 692                 return;
 693         }
 694
 695         /*
 696          * Put packets into TX buffer for that queue.
 697          */
 698
 699         n = len + num;
 700         n = (n > MAX_PKT_BURST) ? MAX_PKT_BURST - len : num;
 701
 702         j = 0;
 703         switch (n % FWDSTEP) {
 704         while (j < n) {
 705         case 0:
 706                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 707                 j++;
 708                 /* fall-through */
 709         case 3:
 710                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 711                 j++;
 712                 /* fall-through */
 713         case 2:
 714                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 715                 j++;
 716                 /* fall-through */
 717         case 1:
 718                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 719                 j++;
 720         }
 721         }
 722
 723         len += n;
 724
 725         /* enough pkts to be sent */
 726         if (unlikely(len == MAX_PKT_BURST)) {
 727
 728                 send_burst(qconf, MAX_PKT_BURST, port);
 729
 730                 /* copy rest of the packets into the TX buffer. */
 731                 len = num - n;
 732                 j = 0;
 733                 switch (len % FWDSTEP) {
 734                 while (j < len) {
 735                 case 0:
 736                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 737                         j++;
 738                         /* fall-through */
 739                 case 3:
 740                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 741                         j++;
 742                         /* fall-through */
 743                 case 2:
 744                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 745                         j++;
 746                         /* fall-through */
 747                 case 1:
 748                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 749                         j++;
 750                 }
 751                 }
 752         }
 753
 754         qconf->tx_mbufs[port].len = len;
 755 }
 756 #endif /* APP_LOOKUP_LPM */
 757
 758 #ifdef DO_RFC_1812_CHECKS
 759 static inline int
 760 is_valid_ipv4_pkt(struct rte_ipv4_hdr *pkt, uint32_t link_len)
 761 {
 762         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 763         /*
 764          * 1. The packet length reported by the Link Layer must be large
 765          * enough to hold the minimum length legal IP datagram (20 bytes).
 766          */
 767         if (link_len < sizeof(struct rte_ipv4_hdr))
 768                 return -1;
 769
 770         /* 2. The IP checksum must be correct. */
 771         /* this is checked in H/W */
 772
 773         /*
 774          * 3. The IP version number must be 4. If the version number is not 4
 775          * then the packet may be another version of IP, such as IPng or
 776          * ST-II.
 777          */
 778         if (((pkt->version_ihl) >> 4) != 4)
 779                 return -3;
 780         /*
 781          * 4. The IP header length field must be large enough to hold the
 782          * minimum length legal IP datagram (20 bytes = 5 words).
 783          */
 784         if ((pkt->version_ihl & 0xf) < 5)
 785                 return -4;
 786
 787         /*
 788          * 5. The IP total length field must be large enough to hold the IP
 789          * datagram header, whose length is specified in the IP header length
 790          * field.
 791          */
 792         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct rte_ipv4_hdr))
 793                 return -5;
 794
 795         return 0;
 796 }
 797 #endif
 798
 799 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 800
 801 static __m128i mask0;
 802 static __m128i mask1;
 803 static __m128i mask2;
 804 static inline uint16_t
 805 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid,
 806                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 807 {
 808         int ret = 0;
 809         union ipv4_5tuple_host key;
 810
 811         ipv4_hdr = (uint8_t *)ipv4_hdr +
 812                 offsetof(struct rte_ipv4_hdr, time_to_live);
 813         __m128i data = _mm_loadu_si128((__m128i *)(ipv4_hdr));
 814         /* Get 5 tuple: dst port, src port, dst IP address, src IP address and
 815            protocol */
 816         key.xmm = _mm_and_si128(data, mask0);
 817         /* Find destination port */
 818         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 819         return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
 820 }
 821
 822 static inline uint16_t
 823 get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid,
 824                 lookup_struct_t *ipv6_l3fwd_lookup_struct)
 825 {
 826         int ret = 0;
 827         union ipv6_5tuple_host key;
 828
 829         ipv6_hdr = (uint8_t *)ipv6_hdr +
 830                 offsetof(struct rte_ipv6_hdr, payload_len);
 831         __m128i data0 = _mm_loadu_si128((__m128i *)(ipv6_hdr));
 832         __m128i data1 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
 833                         sizeof(__m128i)));
 834         __m128i data2 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
 835                         sizeof(__m128i) + sizeof(__m128i)));
 836         /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
 837         key.xmm[0] = _mm_and_si128(data0, mask1);
 838         /* Get part of 5 tuple: dst IP address lower 96 bits and src IP address
 839            higher 32 bits */
 840         key.xmm[1] = data1;
 841         /* Get part of 5 tuple: dst port and src port and dst IP address higher
 842            32 bits */
 843         key.xmm[2] = _mm_and_si128(data2, mask2);
 844
 845         /* Find destination port */
 846         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 847         return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
 848 }
 849 #endif
 850
 851 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 852
 853 static inline uint16_t
 854 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid,
 855                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 856 {
 857         uint32_t next_hop;
 858
 859         return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 860                 rte_be_to_cpu_32(((struct rte_ipv4_hdr *)ipv4_hdr)->dst_addr),
 861                 &next_hop) == 0) ? next_hop : portid);
 862 }
 863
 864 static inline uint16_t
 865 get_ipv6_dst_port(void *ipv6_hdr,  uint16_t portid,
 866                 lookup6_struct_t *ipv6_l3fwd_lookup_struct)
 867 {
 868         uint32_t next_hop;
 869
 870         return ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
 871                 ((struct rte_ipv6_hdr *)ipv6_hdr)->dst_addr, &next_hop) == 0) ?
 872                 next_hop : portid);
 873 }
 874 #endif
 875
 876 static inline void l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid)
 877                 __rte_unused;
 878
 879 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) && \
 880         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
 881
 882 #define MASK_ALL_PKTS   0xff
 883 #define EXCLUDE_1ST_PKT 0xfe
 884 #define EXCLUDE_2ND_PKT 0xfd
 885 #define EXCLUDE_3RD_PKT 0xfb
 886 #define EXCLUDE_4TH_PKT 0xf7
 887 #define EXCLUDE_5TH_PKT 0xef
 888 #define EXCLUDE_6TH_PKT 0xdf
 889 #define EXCLUDE_7TH_PKT 0xbf
 890 #define EXCLUDE_8TH_PKT 0x7f
 891
 892 static inline void
 893 simple_ipv4_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid)
 894 {
 895         struct rte_ether_hdr *eth_hdr[8];
 896         struct rte_ipv4_hdr *ipv4_hdr[8];
 897         uint16_t dst_port[8];
 898         int32_t ret[8];
 899         union ipv4_5tuple_host key[8];
 900         __m128i data[8];
 901
 902         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct rte_ether_hdr *);
 903         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct rte_ether_hdr *);
 904         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct rte_ether_hdr *);
 905         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct rte_ether_hdr *);
 906         eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct rte_ether_hdr *);
 907         eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct rte_ether_hdr *);
 908         eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct rte_ether_hdr *);
 909         eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct rte_ether_hdr *);
 910
 911         /* Handle IPv4 headers.*/
 912         ipv4_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct rte_ipv4_hdr *,
 913                         sizeof(struct rte_ether_hdr));
 914         ipv4_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct rte_ipv4_hdr *,
 915                         sizeof(struct rte_ether_hdr));
 916         ipv4_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct rte_ipv4_hdr *,
 917                         sizeof(struct rte_ether_hdr));
 918         ipv4_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct rte_ipv4_hdr *,
 919                         sizeof(struct rte_ether_hdr));
 920         ipv4_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct rte_ipv4_hdr *,
 921                         sizeof(struct rte_ether_hdr));
 922         ipv4_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct rte_ipv4_hdr *,
 923                         sizeof(struct rte_ether_hdr));
 924         ipv4_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct rte_ipv4_hdr *,
 925                         sizeof(struct rte_ether_hdr));
 926         ipv4_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct rte_ipv4_hdr *,
 927                         sizeof(struct rte_ether_hdr));
 928
 929 #ifdef DO_RFC_1812_CHECKS
 930         /* Check to make sure the packet is valid (RFC1812) */
 931         uint8_t valid_mask = MASK_ALL_PKTS;
 932
 933         if (is_valid_ipv4_pkt(ipv4_hdr[0], m[0]->pkt_len) < 0) {
 934                 rte_pktmbuf_free(m[0]);
 935                 valid_mask &= EXCLUDE_1ST_PKT;
 936         }
 937         if (is_valid_ipv4_pkt(ipv4_hdr[1], m[1]->pkt_len) < 0) {
 938                 rte_pktmbuf_free(m[1]);
 939                 valid_mask &= EXCLUDE_2ND_PKT;
 940         }
 941         if (is_valid_ipv4_pkt(ipv4_hdr[2], m[2]->pkt_len) < 0) {
 942                 rte_pktmbuf_free(m[2]);
 943                 valid_mask &= EXCLUDE_3RD_PKT;
 944         }
 945         if (is_valid_ipv4_pkt(ipv4_hdr[3], m[3]->pkt_len) < 0) {
 946                 rte_pktmbuf_free(m[3]);
 947                 valid_mask &= EXCLUDE_4TH_PKT;
 948         }
 949         if (is_valid_ipv4_pkt(ipv4_hdr[4], m[4]->pkt_len) < 0) {
 950                 rte_pktmbuf_free(m[4]);
 951                 valid_mask &= EXCLUDE_5TH_PKT;
 952         }
 953         if (is_valid_ipv4_pkt(ipv4_hdr[5], m[5]->pkt_len) < 0) {
 954                 rte_pktmbuf_free(m[5]);
 955                 valid_mask &= EXCLUDE_6TH_PKT;
 956         }
 957         if (is_valid_ipv4_pkt(ipv4_hdr[6], m[6]->pkt_len) < 0) {
 958                 rte_pktmbuf_free(m[6]);
 959                 valid_mask &= EXCLUDE_7TH_PKT;
 960         }
 961         if (is_valid_ipv4_pkt(ipv4_hdr[7], m[7]->pkt_len) < 0) {
 962                 rte_pktmbuf_free(m[7]);
 963                 valid_mask &= EXCLUDE_8TH_PKT;
 964         }
 965         if (unlikely(valid_mask != MASK_ALL_PKTS)) {
 966                 if (valid_mask == 0)
 967                         return;
 968
 969                 uint8_t i = 0;
 970
 971                 for (i = 0; i < 8; i++)
 972                         if ((0x1 << i) & valid_mask)
 973                                 l3fwd_simple_forward(m[i], portid);
 974         }
 975 #endif /* End of #ifdef DO_RFC_1812_CHECKS */
 976
 977         data[0] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[0], __m128i *,
 978                         sizeof(struct rte_ether_hdr) +
 979                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 980         data[1] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[1], __m128i *,
 981                         sizeof(struct rte_ether_hdr) +
 982                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 983         data[2] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[2], __m128i *,
 984                         sizeof(struct rte_ether_hdr) +
 985                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 986         data[3] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[3], __m128i *,
 987                         sizeof(struct rte_ether_hdr) +
 988                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 989         data[4] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[4], __m128i *,
 990                         sizeof(struct rte_ether_hdr) +
 991                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 992         data[5] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[5], __m128i *,
 993                         sizeof(struct rte_ether_hdr) +
 994                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 995         data[6] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[6], __m128i *,
 996                         sizeof(struct rte_ether_hdr) +
 997                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 998         data[7] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[7], __m128i *,
 999                         sizeof(struct rte_ether_hdr) +
1000                         offsetof(struct rte_ipv4_hdr, time_to_live)));
1001
1002         key[0].xmm = _mm_and_si128(data[0], mask0);
1003         key[1].xmm = _mm_and_si128(data[1], mask0);
1004         key[2].xmm = _mm_and_si128(data[2], mask0);
1005         key[3].xmm = _mm_and_si128(data[3], mask0);
1006         key[4].xmm = _mm_and_si128(data[4], mask0);
1007         key[5].xmm = _mm_and_si128(data[5], mask0);
1008         key[6].xmm = _mm_and_si128(data[6], mask0);
1009         key[7].xmm = _mm_and_si128(data[7], mask0);
1010
1011         const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1012                         &key[4], &key[5], &key[6], &key[7]};
1013
1014         rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct,
1015                         &key_array[0], 8, ret);
1016         dst_port[0] = ((ret[0] < 0) ? portid : ipv4_l3fwd_out_if[ret[0]]);
1017         dst_port[1] = ((ret[1] < 0) ? portid : ipv4_l3fwd_out_if[ret[1]]);
1018         dst_port[2] = ((ret[2] < 0) ? portid : ipv4_l3fwd_out_if[ret[2]]);
1019         dst_port[3] = ((ret[3] < 0) ? portid : ipv4_l3fwd_out_if[ret[3]]);
1020         dst_port[4] = ((ret[4] < 0) ? portid : ipv4_l3fwd_out_if[ret[4]]);
1021         dst_port[5] = ((ret[5] < 0) ? portid : ipv4_l3fwd_out_if[ret[5]]);
1022         dst_port[6] = ((ret[6] < 0) ? portid : ipv4_l3fwd_out_if[ret[6]]);
1023         dst_port[7] = ((ret[7] < 0) ? portid : ipv4_l3fwd_out_if[ret[7]]);
1024
1025         if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1026                         (enabled_port_mask & 1 << dst_port[0]) == 0)
1027                 dst_port[0] = portid;
1028         if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1029                         (enabled_port_mask & 1 << dst_port[1]) == 0)
1030                 dst_port[1] = portid;
1031         if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1032                         (enabled_port_mask & 1 << dst_port[2]) == 0)
1033                 dst_port[2] = portid;
1034         if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1035                         (enabled_port_mask & 1 << dst_port[3]) == 0)
1036                 dst_port[3] = portid;
1037         if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1038                         (enabled_port_mask & 1 << dst_port[4]) == 0)
1039                 dst_port[4] = portid;
1040         if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1041                         (enabled_port_mask & 1 << dst_port[5]) == 0)
1042                 dst_port[5] = portid;
1043         if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1044                         (enabled_port_mask & 1 << dst_port[6]) == 0)
1045                 dst_port[6] = portid;
1046         if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1047                         (enabled_port_mask & 1 << dst_port[7]) == 0)
1048                 dst_port[7] = portid;
1049
1050 #ifdef DO_RFC_1812_CHECKS
1051         /* Update time to live and header checksum */
1052         --(ipv4_hdr[0]->time_to_live);
1053         --(ipv4_hdr[1]->time_to_live);
1054         --(ipv4_hdr[2]->time_to_live);
1055         --(ipv4_hdr[3]->time_to_live);
1056         ++(ipv4_hdr[0]->hdr_checksum);
1057         ++(ipv4_hdr[1]->hdr_checksum);
1058         ++(ipv4_hdr[2]->hdr_checksum);
1059         ++(ipv4_hdr[3]->hdr_checksum);
1060         --(ipv4_hdr[4]->time_to_live);
1061         --(ipv4_hdr[5]->time_to_live);
1062         --(ipv4_hdr[6]->time_to_live);
1063         --(ipv4_hdr[7]->time_to_live);
1064         ++(ipv4_hdr[4]->hdr_checksum);
1065         ++(ipv4_hdr[5]->hdr_checksum);
1066         ++(ipv4_hdr[6]->hdr_checksum);
1067         ++(ipv4_hdr[7]->hdr_checksum);
1068 #endif
1069
1070         /* dst addr */
1071         *(uint64_t *)&eth_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1072         *(uint64_t *)&eth_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1073         *(uint64_t *)&eth_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1074         *(uint64_t *)&eth_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1075         *(uint64_t *)&eth_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1076         *(uint64_t *)&eth_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1077         *(uint64_t *)&eth_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1078         *(uint64_t *)&eth_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1079
1080         /* src addr */
1081         rte_ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
1082         rte_ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
1083         rte_ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
1084         rte_ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
1085         rte_ether_addr_copy(&ports_eth_addr[dst_port[4]], &eth_hdr[4]->s_addr);
1086         rte_ether_addr_copy(&ports_eth_addr[dst_port[5]], &eth_hdr[5]->s_addr);
1087         rte_ether_addr_copy(&ports_eth_addr[dst_port[6]], &eth_hdr[6]->s_addr);
1088         rte_ether_addr_copy(&ports_eth_addr[dst_port[7]], &eth_hdr[7]->s_addr);
1089
1090         send_single_packet(m[0], (uint8_t)dst_port[0]);
1091         send_single_packet(m[1], (uint8_t)dst_port[1]);
1092         send_single_packet(m[2], (uint8_t)dst_port[2]);
1093         send_single_packet(m[3], (uint8_t)dst_port[3]);
1094         send_single_packet(m[4], (uint8_t)dst_port[4]);
1095         send_single_packet(m[5], (uint8_t)dst_port[5]);
1096         send_single_packet(m[6], (uint8_t)dst_port[6]);
1097         send_single_packet(m[7], (uint8_t)dst_port[7]);
1098
1099 }
1100
1101 static inline void get_ipv6_5tuple(struct rte_mbuf *m0, __m128i mask0,
1102                 __m128i mask1, union ipv6_5tuple_host *key)
1103 {
1104         __m128i tmpdata0 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1105                         __m128i *, sizeof(struct rte_ether_hdr) +
1106                         offsetof(struct rte_ipv6_hdr, payload_len)));
1107         __m128i tmpdata1 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1108                         __m128i *, sizeof(struct rte_ether_hdr) +
1109                         offsetof(struct rte_ipv6_hdr, payload_len) +
1110                         sizeof(__m128i)));
1111         __m128i tmpdata2 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1112                         __m128i *, sizeof(struct rte_ether_hdr) +
1113                         offsetof(struct rte_ipv6_hdr, payload_len) +
1114                         sizeof(__m128i) + sizeof(__m128i)));
1115         key->xmm[0] = _mm_and_si128(tmpdata0, mask0);
1116         key->xmm[1] = tmpdata1;
1117         key->xmm[2] = _mm_and_si128(tmpdata2, mask1);
1118 }
1119
1120 static inline void
1121 simple_ipv6_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid)
1122 {
1123         int32_t ret[8];
1124         uint16_t dst_port[8];
1125         struct rte_ether_hdr *eth_hdr[8];
1126         union ipv6_5tuple_host key[8];
1127
1128         __rte_unused struct rte_ipv6_hdr *ipv6_hdr[8];
1129
1130         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct rte_ether_hdr *);
1131         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct rte_ether_hdr *);
1132         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct rte_ether_hdr *);
1133         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct rte_ether_hdr *);
1134         eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct rte_ether_hdr *);
1135         eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct rte_ether_hdr *);
1136         eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct rte_ether_hdr *);
1137         eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct rte_ether_hdr *);
1138
1139         /* Handle IPv6 headers.*/
1140         ipv6_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct rte_ipv6_hdr *,
1141                         sizeof(struct rte_ether_hdr));
1142         ipv6_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct rte_ipv6_hdr *,
1143                         sizeof(struct rte_ether_hdr));
1144         ipv6_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct rte_ipv6_hdr *,
1145                         sizeof(struct rte_ether_hdr));
1146         ipv6_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct rte_ipv6_hdr *,
1147                         sizeof(struct rte_ether_hdr));
1148         ipv6_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct rte_ipv6_hdr *,
1149                         sizeof(struct rte_ether_hdr));
1150         ipv6_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct rte_ipv6_hdr *,
1151                         sizeof(struct rte_ether_hdr));
1152         ipv6_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct rte_ipv6_hdr *,
1153                         sizeof(struct rte_ether_hdr));
1154         ipv6_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct rte_ipv6_hdr *,
1155                         sizeof(struct rte_ether_hdr));
1156
1157         get_ipv6_5tuple(m[0], mask1, mask2, &key[0]);
1158         get_ipv6_5tuple(m[1], mask1, mask2, &key[1]);
1159         get_ipv6_5tuple(m[2], mask1, mask2, &key[2]);
1160         get_ipv6_5tuple(m[3], mask1, mask2, &key[3]);
1161         get_ipv6_5tuple(m[4], mask1, mask2, &key[4]);
1162         get_ipv6_5tuple(m[5], mask1, mask2, &key[5]);
1163         get_ipv6_5tuple(m[6], mask1, mask2, &key[6]);
1164         get_ipv6_5tuple(m[7], mask1, mask2, &key[7]);
1165
1166         const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1167                         &key[4], &key[5], &key[6], &key[7]};
1168
1169         rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1170                         &key_array[0], 4, ret);
1171         dst_port[0] = ((ret[0] < 0) ? portid : ipv6_l3fwd_out_if[ret[0]]);
1172         dst_port[1] = ((ret[1] < 0) ? portid : ipv6_l3fwd_out_if[ret[1]]);
1173         dst_port[2] = ((ret[2] < 0) ? portid : ipv6_l3fwd_out_if[ret[2]]);
1174         dst_port[3] = ((ret[3] < 0) ? portid : ipv6_l3fwd_out_if[ret[3]]);
1175         dst_port[4] = ((ret[4] < 0) ? portid : ipv6_l3fwd_out_if[ret[4]]);
1176         dst_port[5] = ((ret[5] < 0) ? portid : ipv6_l3fwd_out_if[ret[5]]);
1177         dst_port[6] = ((ret[6] < 0) ? portid : ipv6_l3fwd_out_if[ret[6]]);
1178         dst_port[7] = ((ret[7] < 0) ? portid : ipv6_l3fwd_out_if[ret[7]]);
1179
1180         if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1181                         (enabled_port_mask & 1 << dst_port[0]) == 0)
1182                 dst_port[0] = portid;
1183         if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1184                         (enabled_port_mask & 1 << dst_port[1]) == 0)
1185                 dst_port[1] = portid;
1186         if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1187                         (enabled_port_mask & 1 << dst_port[2]) == 0)
1188                 dst_port[2] = portid;
1189         if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1190                         (enabled_port_mask & 1 << dst_port[3]) == 0)
1191                 dst_port[3] = portid;
1192         if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1193                         (enabled_port_mask & 1 << dst_port[4]) == 0)
1194                 dst_port[4] = portid;
1195         if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1196                         (enabled_port_mask & 1 << dst_port[5]) == 0)
1197                 dst_port[5] = portid;
1198         if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1199                         (enabled_port_mask & 1 << dst_port[6]) == 0)
1200                 dst_port[6] = portid;
1201         if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1202                         (enabled_port_mask & 1 << dst_port[7]) == 0)
1203                 dst_port[7] = portid;
1204
1205         /* dst addr */
1206         *(uint64_t *)&eth_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1207         *(uint64_t *)&eth_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1208         *(uint64_t *)&eth_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1209         *(uint64_t *)&eth_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1210         *(uint64_t *)&eth_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1211         *(uint64_t *)&eth_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1212         *(uint64_t *)&eth_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1213         *(uint64_t *)&eth_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1214
1215         /* src addr */
1216         rte_ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
1217         rte_ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
1218         rte_ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
1219         rte_ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
1220         rte_ether_addr_copy(&ports_eth_addr[dst_port[4]], &eth_hdr[4]->s_addr);
1221         rte_ether_addr_copy(&ports_eth_addr[dst_port[5]], &eth_hdr[5]->s_addr);
1222         rte_ether_addr_copy(&ports_eth_addr[dst_port[6]], &eth_hdr[6]->s_addr);
1223         rte_ether_addr_copy(&ports_eth_addr[dst_port[7]], &eth_hdr[7]->s_addr);
1224
1225         send_single_packet(m[0], dst_port[0]);
1226         send_single_packet(m[1], dst_port[1]);
1227         send_single_packet(m[2], dst_port[2]);
1228         send_single_packet(m[3], dst_port[3]);
1229         send_single_packet(m[4], dst_port[4]);
1230         send_single_packet(m[5], dst_port[5]);
1231         send_single_packet(m[6], dst_port[6]);
1232         send_single_packet(m[7], dst_port[7]);
1233
1234 }
1235 #endif /* APP_LOOKUP_METHOD */
1236
1237 static __rte_always_inline void
1238 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid)
1239 {
1240         struct rte_ether_hdr *eth_hdr;
1241         struct rte_ipv4_hdr *ipv4_hdr;
1242         uint16_t dst_port;
1243
1244         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
1245
1246         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
1247                 /* Handle IPv4 headers.*/
1248                 ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
1249                                 sizeof(struct rte_ether_hdr));
1250
1251 #ifdef DO_RFC_1812_CHECKS
1252                 /* Check to make sure the packet is valid (RFC1812) */
1253                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
1254                         rte_pktmbuf_free(m);
1255                         return;
1256                 }
1257 #endif
1258
1259                  dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
1260                         RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct);
1261                 if (dst_port >= RTE_MAX_ETHPORTS ||
1262                                 (enabled_port_mask & 1 << dst_port) == 0)
1263                         dst_port = portid;
1264
1265 #ifdef DO_RFC_1812_CHECKS
1266                 /* Update time to live and header checksum */
1267                 --(ipv4_hdr->time_to_live);
1268                 ++(ipv4_hdr->hdr_checksum);
1269 #endif
1270                 /* dst addr */
1271                 *(uint64_t *)&eth_hdr->d_addr = dest_eth_addr[dst_port];
1272
1273                 /* src addr */
1274                 rte_ether_addr_copy(&ports_eth_addr[dst_port],
1275                                 &eth_hdr->s_addr);
1276
1277                 send_single_packet(m, dst_port);
1278         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
1279                 /* Handle IPv6 headers.*/
1280                 struct rte_ipv6_hdr *ipv6_hdr;
1281
1282                 ipv6_hdr = rte_pktmbuf_mtod_offset(m, struct rte_ipv6_hdr *,
1283                                 sizeof(struct rte_ether_hdr));
1284
1285                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
1286                                 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct);
1287
1288                 if (dst_port >= RTE_MAX_ETHPORTS ||
1289                                 (enabled_port_mask & 1 << dst_port) == 0)
1290                         dst_port = portid;
1291
1292                 /* dst addr */
1293                 *(uint64_t *)&eth_hdr->d_addr = dest_eth_addr[dst_port];
1294
1295                 /* src addr */
1296                 rte_ether_addr_copy(&ports_eth_addr[dst_port],
1297                                 &eth_hdr->s_addr);
1298
1299                 send_single_packet(m, dst_port);
1300         } else
1301                 /* Free the mbuf that contains non-IPV4/IPV6 packet */
1302                 rte_pktmbuf_free(m);
1303 }
1304
1305 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1306         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1307 #ifdef DO_RFC_1812_CHECKS
1308
1309 #define IPV4_MIN_VER_IHL        0x45
1310 #define IPV4_MAX_VER_IHL        0x4f
1311 #define IPV4_MAX_VER_IHL_DIFF   (IPV4_MAX_VER_IHL - IPV4_MIN_VER_IHL)
1312
1313 /* Minimum value of IPV4 total length (20B) in network byte order. */
1314 #define IPV4_MIN_LEN_BE (sizeof(struct rte_ipv4_hdr) << 8)
1315
1316 /*
1317  * From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2:
1318  * - The IP version number must be 4.
1319  * - The IP header length field must be large enough to hold the
1320  *    minimum length legal IP datagram (20 bytes = 5 words).
1321  * - The IP total length field must be large enough to hold the IP
1322  *   datagram header, whose length is specified in the IP header length
1323  *   field.
1324  * If we encounter invalid IPV4 packet, then set destination port for it
1325  * to BAD_PORT value.
1326  */
1327 static __rte_always_inline void
1328 rfc1812_process(struct rte_ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t ptype)
1329 {
1330         uint8_t ihl;
1331
1332         if (RTE_ETH_IS_IPV4_HDR(ptype)) {
1333                 ihl = ipv4_hdr->version_ihl - IPV4_MIN_VER_IHL;
1334
1335                 ipv4_hdr->time_to_live--;
1336                 ipv4_hdr->hdr_checksum++;
1337
1338                 if (ihl > IPV4_MAX_VER_IHL_DIFF ||
1339                                 ((uint8_t)ipv4_hdr->total_length == 0 &&
1340                                 ipv4_hdr->total_length < IPV4_MIN_LEN_BE)) {
1341                         dp[0] = BAD_PORT;
1342                 }
1343         }
1344 }
1345
1346 #else
1347 #define rfc1812_process(mb, dp, ptype)  do { } while (0)
1348 #endif /* DO_RFC_1812_CHECKS */
1349 #endif /* APP_LOOKUP_LPM && ENABLE_MULTI_BUFFER_OPTIMIZE */
1350
1351
1352 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1353         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1354
1355 static __rte_always_inline uint16_t
1356 get_dst_port(struct rte_mbuf *pkt, uint32_t dst_ipv4, uint16_t portid)
1357 {
1358         uint32_t next_hop;
1359         struct rte_ipv6_hdr *ipv6_hdr;
1360         struct rte_ether_hdr *eth_hdr;
1361
1362         if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
1363                 return (uint16_t) ((rte_lpm_lookup(
1364                                 RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dst_ipv4,
1365                                 &next_hop) == 0) ? next_hop : portid);
1366
1367         } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
1368
1369                 eth_hdr = rte_pktmbuf_mtod(pkt, struct rte_ether_hdr *);
1370                 ipv6_hdr = (struct rte_ipv6_hdr *)(eth_hdr + 1);
1371
1372                 return (uint16_t) ((rte_lpm6_lookup(
1373                                 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1374                                 ipv6_hdr->dst_addr, &next_hop) == 0) ?
1375                                 next_hop : portid);
1376
1377         }
1378
1379         return portid;
1380 }
1381
1382 static inline void
1383 process_packet(struct rte_mbuf *pkt, uint16_t *dst_port, uint16_t portid)
1384 {
1385         struct rte_ether_hdr *eth_hdr;
1386         struct rte_ipv4_hdr *ipv4_hdr;
1387         uint32_t dst_ipv4;
1388         uint16_t dp;
1389         __m128i te, ve;
1390
1391         eth_hdr = rte_pktmbuf_mtod(pkt, struct rte_ether_hdr *);
1392         ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
1393
1394         dst_ipv4 = ipv4_hdr->dst_addr;
1395         dst_ipv4 = rte_be_to_cpu_32(dst_ipv4);
1396         dp = get_dst_port(pkt, dst_ipv4, portid);
1397
1398         te = _mm_load_si128((__m128i *)eth_hdr);
1399         ve = val_eth[dp];
1400
1401         dst_port[0] = dp;
1402         rfc1812_process(ipv4_hdr, dst_port, pkt->packet_type);
1403
1404         te =  _mm_blend_epi16(te, ve, MASK_ETH);
1405         _mm_store_si128((__m128i *)eth_hdr, te);
1406 }
1407
1408 /*
1409  * Read packet_type and destination IPV4 addresses from 4 mbufs.
1410  */
1411 static inline void
1412 processx4_step1(struct rte_mbuf *pkt[FWDSTEP],
1413                 __m128i *dip,
1414                 uint32_t *ipv4_flag)
1415 {
1416         struct rte_ipv4_hdr *ipv4_hdr;
1417         struct rte_ether_hdr *eth_hdr;
1418         uint32_t x0, x1, x2, x3;
1419
1420         eth_hdr = rte_pktmbuf_mtod(pkt[0], struct rte_ether_hdr *);
1421         ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
1422         x0 = ipv4_hdr->dst_addr;
1423         ipv4_flag[0] = pkt[0]->packet_type & RTE_PTYPE_L3_IPV4;
1424
1425         eth_hdr = rte_pktmbuf_mtod(pkt[1], struct rte_ether_hdr *);
1426         ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
1427         x1 = ipv4_hdr->dst_addr;
1428         ipv4_flag[0] &= pkt[1]->packet_type;
1429
1430         eth_hdr = rte_pktmbuf_mtod(pkt[2], struct rte_ether_hdr *);
1431         ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
1432         x2 = ipv4_hdr->dst_addr;
1433         ipv4_flag[0] &= pkt[2]->packet_type;
1434
1435         eth_hdr = rte_pktmbuf_mtod(pkt[3], struct rte_ether_hdr *);
1436         ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
1437         x3 = ipv4_hdr->dst_addr;
1438         ipv4_flag[0] &= pkt[3]->packet_type;
1439
1440         dip[0] = _mm_set_epi32(x3, x2, x1, x0);
1441 }
1442
1443 /*
1444  * Lookup into LPM for destination port.
1445  * If lookup fails, use incoming port (portid) as destination port.
1446  */
1447 static inline void
1448 processx4_step2(__m128i dip,
1449                 uint32_t ipv4_flag,
1450                 uint16_t portid,
1451                 struct rte_mbuf *pkt[FWDSTEP],
1452                 uint16_t dprt[FWDSTEP])
1453 {
1454         rte_xmm_t dst;
1455         const __m128i bswap_mask = _mm_set_epi8(12, 13, 14, 15, 8, 9, 10, 11,
1456                         4, 5, 6, 7, 0, 1, 2, 3);
1457
1458         /* Byte swap 4 IPV4 addresses. */
1459         dip = _mm_shuffle_epi8(dip, bswap_mask);
1460
1461         /* if all 4 packets are IPV4. */
1462         if (likely(ipv4_flag)) {
1463                 rte_lpm_lookupx4(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dip,
1464                                 dst.u32, portid);
1465
1466                 /* get rid of unused upper 16 bit for each dport. */
1467                 dst.x = _mm_packs_epi32(dst.x, dst.x);
1468                 *(uint64_t *)dprt = dst.u64[0];
1469         } else {
1470                 dst.x = dip;
1471                 dprt[0] = get_dst_port(pkt[0], dst.u32[0], portid);
1472                 dprt[1] = get_dst_port(pkt[1], dst.u32[1], portid);
1473                 dprt[2] = get_dst_port(pkt[2], dst.u32[2], portid);
1474                 dprt[3] = get_dst_port(pkt[3], dst.u32[3], portid);
1475         }
1476 }
1477
1478 /*
1479  * Update source and destination MAC addresses in the ethernet header.
1480  * Perform RFC1812 checks and updates for IPV4 packets.
1481  */
1482 static inline void
1483 processx4_step3(struct rte_mbuf *pkt[FWDSTEP], uint16_t dst_port[FWDSTEP])
1484 {
1485         __m128i te[FWDSTEP];
1486         __m128i ve[FWDSTEP];
1487         __m128i *p[FWDSTEP];
1488
1489         p[0] = rte_pktmbuf_mtod(pkt[0], __m128i *);
1490         p[1] = rte_pktmbuf_mtod(pkt[1], __m128i *);
1491         p[2] = rte_pktmbuf_mtod(pkt[2], __m128i *);
1492         p[3] = rte_pktmbuf_mtod(pkt[3], __m128i *);
1493
1494         ve[0] = val_eth[dst_port[0]];
1495         te[0] = _mm_load_si128(p[0]);
1496
1497         ve[1] = val_eth[dst_port[1]];
1498         te[1] = _mm_load_si128(p[1]);
1499
1500         ve[2] = val_eth[dst_port[2]];
1501         te[2] = _mm_load_si128(p[2]);
1502
1503         ve[3] = val_eth[dst_port[3]];
1504         te[3] = _mm_load_si128(p[3]);
1505
1506         /* Update first 12 bytes, keep rest bytes intact. */
1507         te[0] =  _mm_blend_epi16(te[0], ve[0], MASK_ETH);
1508         te[1] =  _mm_blend_epi16(te[1], ve[1], MASK_ETH);
1509         te[2] =  _mm_blend_epi16(te[2], ve[2], MASK_ETH);
1510         te[3] =  _mm_blend_epi16(te[3], ve[3], MASK_ETH);
1511
1512         _mm_store_si128(p[0], te[0]);
1513         _mm_store_si128(p[1], te[1]);
1514         _mm_store_si128(p[2], te[2]);
1515         _mm_store_si128(p[3], te[3]);
1516
1517         rfc1812_process((struct rte_ipv4_hdr *)
1518                         ((struct rte_ether_hdr *)p[0] + 1),
1519                         &dst_port[0], pkt[0]->packet_type);
1520         rfc1812_process((struct rte_ipv4_hdr *)
1521                         ((struct rte_ether_hdr *)p[1] + 1),
1522                         &dst_port[1], pkt[1]->packet_type);
1523         rfc1812_process((struct rte_ipv4_hdr *)
1524                         ((struct rte_ether_hdr *)p[2] + 1),
1525                         &dst_port[2], pkt[2]->packet_type);
1526         rfc1812_process((struct rte_ipv4_hdr *)
1527                         ((struct rte_ether_hdr *)p[3] + 1),
1528                         &dst_port[3], pkt[3]->packet_type);
1529 }
1530
1531 /*
1532  * We group consecutive packets with the same destionation port into one burst.
1533  * To avoid extra latency this is done together with some other packet
1534  * processing, but after we made a final decision about packet's destination.
1535  * To do this we maintain:
1536  * pnum - array of number of consecutive packets with the same dest port for
1537  * each packet in the input burst.
1538  * lp - pointer to the last updated element in the pnum.
1539  * dlp - dest port value lp corresponds to.
1540  */
1541
1542 #define GRPSZ   (1 << FWDSTEP)
1543 #define GRPMSK  (GRPSZ - 1)
1544
1545 #define GROUP_PORT_STEP(dlp, dcp, lp, pn, idx)  do { \
1546         if (likely((dlp) == (dcp)[(idx)])) {         \
1547                 (lp)[0]++;                           \
1548         } else {                                     \
1549                 (dlp) = (dcp)[idx];                  \
1550                 (lp) = (pn) + (idx);                 \
1551                 (lp)[0] = 1;                         \
1552         }                                            \
1553 } while (0)
1554
1555 /*
1556  * Group consecutive packets with the same destination port in bursts of 4.
1557  * Suppose we have array of destionation ports:
1558  * dst_port[] = {a, b, c, d,, e, ... }
1559  * dp1 should contain: <a, b, c, d>, dp2: <b, c, d, e>.
1560  * We doing 4 comparisons at once and the result is 4 bit mask.
1561  * This mask is used as an index into prebuild array of pnum values.
1562  */
1563 static inline uint16_t *
1564 port_groupx4(uint16_t pn[FWDSTEP + 1], uint16_t *lp, __m128i dp1, __m128i dp2)
1565 {
1566         static const struct {
1567                 uint64_t pnum; /* prebuild 4 values for pnum[]. */
1568                 int32_t  idx;  /* index for new last updated elemnet. */
1569                 uint16_t lpv;  /* add value to the last updated element. */
1570         } gptbl[GRPSZ] = {
1571         {
1572                 /* 0: a != b, b != c, c != d, d != e */
1573                 .pnum = UINT64_C(0x0001000100010001),
1574                 .idx = 4,
1575                 .lpv = 0,
1576         },
1577         {
1578                 /* 1: a == b, b != c, c != d, d != e */
1579                 .pnum = UINT64_C(0x0001000100010002),
1580                 .idx = 4,
1581                 .lpv = 1,
1582         },
1583         {
1584                 /* 2: a != b, b == c, c != d, d != e */
1585                 .pnum = UINT64_C(0x0001000100020001),
1586                 .idx = 4,
1587                 .lpv = 0,
1588         },
1589         {
1590                 /* 3: a == b, b == c, c != d, d != e */
1591                 .pnum = UINT64_C(0x0001000100020003),
1592                 .idx = 4,
1593                 .lpv = 2,
1594         },
1595         {
1596                 /* 4: a != b, b != c, c == d, d != e */
1597                 .pnum = UINT64_C(0x0001000200010001),
1598                 .idx = 4,
1599                 .lpv = 0,
1600         },
1601         {
1602                 /* 5: a == b, b != c, c == d, d != e */
1603                 .pnum = UINT64_C(0x0001000200010002),
1604                 .idx = 4,
1605                 .lpv = 1,
1606         },
1607         {
1608                 /* 6: a != b, b == c, c == d, d != e */
1609                 .pnum = UINT64_C(0x0001000200030001),
1610                 .idx = 4,
1611                 .lpv = 0,
1612         },
1613         {
1614                 /* 7: a == b, b == c, c == d, d != e */
1615                 .pnum = UINT64_C(0x0001000200030004),
1616                 .idx = 4,
1617                 .lpv = 3,
1618         },
1619         {
1620                 /* 8: a != b, b != c, c != d, d == e */
1621                 .pnum = UINT64_C(0x0002000100010001),
1622                 .idx = 3,
1623                 .lpv = 0,
1624         },
1625         {
1626                 /* 9: a == b, b != c, c != d, d == e */
1627                 .pnum = UINT64_C(0x0002000100010002),
1628                 .idx = 3,
1629                 .lpv = 1,
1630         },
1631         {
1632                 /* 0xa: a != b, b == c, c != d, d == e */
1633                 .pnum = UINT64_C(0x0002000100020001),
1634                 .idx = 3,
1635                 .lpv = 0,
1636         },
1637         {
1638                 /* 0xb: a == b, b == c, c != d, d == e */
1639                 .pnum = UINT64_C(0x0002000100020003),
1640                 .idx = 3,
1641                 .lpv = 2,
1642         },
1643         {
1644                 /* 0xc: a != b, b != c, c == d, d == e */
1645                 .pnum = UINT64_C(0x0002000300010001),
1646                 .idx = 2,
1647                 .lpv = 0,
1648         },
1649         {
1650                 /* 0xd: a == b, b != c, c == d, d == e */
1651                 .pnum = UINT64_C(0x0002000300010002),
1652                 .idx = 2,
1653                 .lpv = 1,
1654         },
1655         {
1656                 /* 0xe: a != b, b == c, c == d, d == e */
1657                 .pnum = UINT64_C(0x0002000300040001),
1658                 .idx = 1,
1659                 .lpv = 0,
1660         },
1661         {
1662                 /* 0xf: a == b, b == c, c == d, d == e */
1663                 .pnum = UINT64_C(0x0002000300040005),
1664                 .idx = 0,
1665                 .lpv = 4,
1666         },
1667         };
1668
1669         union {
1670                 uint16_t u16[FWDSTEP + 1];
1671                 uint64_t u64;
1672         } *pnum = (void *)pn;
1673
1674         int32_t v;
1675
1676         dp1 = _mm_cmpeq_epi16(dp1, dp2);
1677         dp1 = _mm_unpacklo_epi16(dp1, dp1);
1678         v = _mm_movemask_ps((__m128)dp1);
1679
1680         /* update last port counter. */
1681         lp[0] += gptbl[v].lpv;
1682
1683         /* if dest port value has changed. */
1684         if (v != GRPMSK) {
1685                 pnum->u64 = gptbl[v].pnum;
1686                 pnum->u16[FWDSTEP] = 1;
1687                 lp = pnum->u16 + gptbl[v].idx;
1688         }
1689
1690         return lp;
1691 }
1692
1693 #endif /* APP_LOOKUP_METHOD */
1694
1695 static void
1696 process_burst(struct rte_mbuf *pkts_burst[MAX_PKT_BURST], int nb_rx,
1697                 uint16_t portid)
1698 {
1699
1700         int j;
1701
1702 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1703         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1704         int32_t k;
1705         uint16_t dlp;
1706         uint16_t *lp;
1707         uint16_t dst_port[MAX_PKT_BURST];
1708         __m128i dip[MAX_PKT_BURST / FWDSTEP];
1709         uint32_t ipv4_flag[MAX_PKT_BURST / FWDSTEP];
1710         uint16_t pnum[MAX_PKT_BURST + 1];
1711 #endif
1712
1713
1714 #if (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
1715 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1716         {
1717                 /*
1718                  * Send nb_rx - nb_rx%8 packets
1719                  * in groups of 8.
1720                  */
1721                 int32_t n = RTE_ALIGN_FLOOR(nb_rx, 8);
1722
1723                 for (j = 0; j < n; j += 8) {
1724                         uint32_t pkt_type =
1725                                 pkts_burst[j]->packet_type &
1726                                 pkts_burst[j+1]->packet_type &
1727                                 pkts_burst[j+2]->packet_type &
1728                                 pkts_burst[j+3]->packet_type &
1729                                 pkts_burst[j+4]->packet_type &
1730                                 pkts_burst[j+5]->packet_type &
1731                                 pkts_burst[j+6]->packet_type &
1732                                 pkts_burst[j+7]->packet_type;
1733                         if (pkt_type & RTE_PTYPE_L3_IPV4) {
1734                                 simple_ipv4_fwd_8pkts(&pkts_burst[j], portid);
1735                         } else if (pkt_type &
1736                                 RTE_PTYPE_L3_IPV6) {
1737                                 simple_ipv6_fwd_8pkts(&pkts_burst[j], portid);
1738                         } else {
1739                                 l3fwd_simple_forward(pkts_burst[j], portid);
1740                                 l3fwd_simple_forward(pkts_burst[j+1], portid);
1741                                 l3fwd_simple_forward(pkts_burst[j+2], portid);
1742                                 l3fwd_simple_forward(pkts_burst[j+3], portid);
1743                                 l3fwd_simple_forward(pkts_burst[j+4], portid);
1744                                 l3fwd_simple_forward(pkts_burst[j+5], portid);
1745                                 l3fwd_simple_forward(pkts_burst[j+6], portid);
1746                                 l3fwd_simple_forward(pkts_burst[j+7], portid);
1747                         }
1748                 }
1749                 for (; j < nb_rx ; j++)
1750                         l3fwd_simple_forward(pkts_burst[j], portid);
1751         }
1752 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1753
1754         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1755         for (j = 0; j != k; j += FWDSTEP)
1756                 processx4_step1(&pkts_burst[j], &dip[j / FWDSTEP],
1757                                 &ipv4_flag[j / FWDSTEP]);
1758
1759         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1760         for (j = 0; j != k; j += FWDSTEP)
1761                 processx4_step2(dip[j / FWDSTEP], ipv4_flag[j / FWDSTEP],
1762                                 portid, &pkts_burst[j], &dst_port[j]);
1763
1764         /*
1765          * Finish packet processing and group consecutive
1766          * packets with the same destination port.
1767          */
1768         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1769         if (k != 0) {
1770                 __m128i dp1, dp2;
1771
1772                 lp = pnum;
1773                 lp[0] = 1;
1774
1775                 processx4_step3(pkts_burst, dst_port);
1776
1777                 /* dp1: <d[0], d[1], d[2], d[3], ... > */
1778                 dp1 = _mm_loadu_si128((__m128i *)dst_port);
1779
1780                 for (j = FWDSTEP; j != k; j += FWDSTEP) {
1781                         processx4_step3(&pkts_burst[j], &dst_port[j]);
1782
1783                         /*
1784                          * dp2:
1785                          * <d[j-3], d[j-2], d[j-1], d[j], ... >
1786                          */
1787                         dp2 = _mm_loadu_si128(
1788                                         (__m128i *)&dst_port[j - FWDSTEP + 1]);
1789                         lp  = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1790
1791                         /*
1792                          * dp1:
1793                          * <d[j], d[j+1], d[j+2], d[j+3], ... >
1794                          */
1795                         dp1 = _mm_srli_si128(dp2, (FWDSTEP - 1) *
1796                                         sizeof(dst_port[0]));
1797                 }
1798
1799                 /*
1800                  * dp2: <d[j-3], d[j-2], d[j-1], d[j-1], ... >
1801                  */
1802                 dp2 = _mm_shufflelo_epi16(dp1, 0xf9);
1803                 lp  = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1804
1805                 /*
1806                  * remove values added by the last repeated
1807                  * dst port.
1808                  */
1809                 lp[0]--;
1810                 dlp = dst_port[j - 1];
1811         } else {
1812                 /* set dlp and lp to the never used values. */
1813                 dlp = BAD_PORT - 1;
1814                 lp = pnum + MAX_PKT_BURST;
1815         }
1816
1817         /* Process up to last 3 packets one by one. */
1818         switch (nb_rx % FWDSTEP) {
1819         case 3:
1820                 process_packet(pkts_burst[j], dst_port + j, portid);
1821                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1822                 j++;
1823                 /* fall-through */
1824         case 2:
1825                 process_packet(pkts_burst[j], dst_port + j, portid);
1826                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1827                 j++;
1828                 /* fall-through */
1829         case 1:
1830                 process_packet(pkts_burst[j], dst_port + j, portid);
1831                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1832                 j++;
1833         }
1834
1835         /*
1836          * Send packets out, through destination port.
1837          * Consecuteve pacekts with the same destination port
1838          * are already grouped together.
1839          * If destination port for the packet equals BAD_PORT,
1840          * then free the packet without sending it out.
1841          */
1842         for (j = 0; j < nb_rx; j += k) {
1843
1844                 int32_t m;
1845                 uint16_t pn;
1846
1847                 pn = dst_port[j];
1848                 k = pnum[j];
1849
1850                 if (likely(pn != BAD_PORT))
1851                         send_packetsx4(pn, pkts_burst + j, k);
1852                 else
1853                         for (m = j; m != j + k; m++)
1854                                 rte_pktmbuf_free(pkts_burst[m]);
1855
1856         }
1857
1858 #endif /* APP_LOOKUP_METHOD */
1859 #else /* ENABLE_MULTI_BUFFER_OPTIMIZE == 0 */
1860
1861         /* Prefetch first packets */
1862         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++)
1863                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[j], void *));
1864
1865         /* Prefetch and forward already prefetched packets */
1866         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1867                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1868                                 j + PREFETCH_OFFSET], void *));
1869                 l3fwd_simple_forward(pkts_burst[j], portid);
1870         }
1871
1872         /* Forward remaining prefetched packets */
1873         for (; j < nb_rx; j++)
1874                 l3fwd_simple_forward(pkts_burst[j], portid);
1875
1876 #endif /* ENABLE_MULTI_BUFFER_OPTIMIZE */
1877
1878 }
1879
1880 #if (APP_CPU_LOAD > 0)
1881
1882 /*
1883  * CPU-load stats collector
1884  */
1885 static int __rte_noreturn
1886 cpu_load_collector(__rte_unused void *arg) {
1887         unsigned i, j, k;
1888         uint64_t hits;
1889         uint64_t prev_tsc, diff_tsc, cur_tsc;
1890         uint64_t total[MAX_CPU] = { 0 };
1891         unsigned min_cpu = MAX_CPU;
1892         unsigned max_cpu = 0;
1893         unsigned cpu_id;
1894         int busy_total = 0;
1895         int busy_flag = 0;
1896
1897         unsigned int n_thread_per_cpu[MAX_CPU] = { 0 };
1898         struct thread_conf *thread_per_cpu[MAX_CPU][MAX_THREAD];
1899
1900         struct thread_conf *thread_conf;
1901
1902         const uint64_t interval_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1903                 US_PER_S * CPU_LOAD_TIMEOUT_US;
1904
1905         prev_tsc = 0;
1906         /*
1907          * Wait for all threads
1908          */
1909
1910         printf("Waiting for %d rx threads and %d tx threads\n", n_rx_thread,
1911                         n_tx_thread);
1912
1913         while (rte_atomic16_read(&rx_counter) < n_rx_thread)
1914                 rte_pause();
1915
1916         while (rte_atomic16_read(&tx_counter) < n_tx_thread)
1917                 rte_pause();
1918
1919         for (i = 0; i < n_rx_thread; i++) {
1920
1921                 thread_conf = &rx_thread[i].conf;
1922                 cpu_id = thread_conf->cpu_id;
1923                 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1924
1925                 if (cpu_id > max_cpu)
1926                         max_cpu = cpu_id;
1927                 if (cpu_id < min_cpu)
1928                         min_cpu = cpu_id;
1929         }
1930         for (i = 0; i < n_tx_thread; i++) {
1931
1932                 thread_conf = &tx_thread[i].conf;
1933                 cpu_id = thread_conf->cpu_id;
1934                 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1935
1936                 if (thread_conf->cpu_id > max_cpu)
1937                         max_cpu = thread_conf->cpu_id;
1938                 if (thread_conf->cpu_id < min_cpu)
1939                         min_cpu = thread_conf->cpu_id;
1940         }
1941
1942         while (1) {
1943
1944                 cpu_load.counter++;
1945                 for (i = min_cpu; i <= max_cpu; i++) {
1946                         for (j = 0; j < MAX_CPU_COUNTER; j++) {
1947                                 for (k = 0; k < n_thread_per_cpu[i]; k++)
1948                                         if (thread_per_cpu[i][k]->busy[j]) {
1949                                                 busy_flag = 1;
1950                                                 break;
1951                                         }
1952                                 if (busy_flag) {
1953                                         cpu_load.hits[j][i]++;
1954                                         busy_total = 1;
1955                                         busy_flag = 0;
1956                                 }
1957                         }
1958
1959                         if (busy_total) {
1960                                 total[i]++;
1961                                 busy_total = 0;
1962                         }
1963                 }
1964
1965                 cur_tsc = rte_rdtsc();
1966
1967                 diff_tsc = cur_tsc - prev_tsc;
1968                 if (unlikely(diff_tsc > interval_tsc)) {
1969
1970                         printf("\033c");
1971
1972                         printf("Cpu usage for %d rx threads and %d tx threads:\n\n",
1973                                         n_rx_thread, n_tx_thread);
1974
1975                         printf("cpu#     proc%%  poll%%  overhead%%\n\n");
1976
1977                         for (i = min_cpu; i <= max_cpu; i++) {
1978                                 hits = 0;
1979                                 printf("CPU %d:", i);
1980                                 for (j = 0; j < MAX_CPU_COUNTER; j++) {
1981                                         printf("%7" PRIu64 "",
1982                                                         cpu_load.hits[j][i] * 100 / cpu_load.counter);
1983                                         hits += cpu_load.hits[j][i];
1984                                         cpu_load.hits[j][i] = 0;
1985                                 }
1986                                 printf("%7" PRIu64 "\n",
1987                                                 100 - total[i] * 100 / cpu_load.counter);
1988                                 total[i] = 0;
1989                         }
1990                         cpu_load.counter = 0;
1991
1992                         prev_tsc = cur_tsc;
1993                 }
1994
1995         }
1996 }
1997 #endif /* APP_CPU_LOAD */
1998
1999 /*
2000  * Null processing lthread loop
2001  *
2002  * This loop is used to start empty scheduler on lcore.
2003  */
2004 static void *
2005 lthread_null(__rte_unused void *args)
2006 {
2007         int lcore_id = rte_lcore_id();
2008
2009         RTE_LOG(INFO, L3FWD, "Starting scheduler on lcore %d.\n", lcore_id);
2010         lthread_exit(NULL);
2011         return NULL;
2012 }
2013
2014 /* main processing loop */
2015 static void *
2016 lthread_tx_per_ring(void *dummy)
2017 {
2018         int nb_rx;
2019         uint16_t portid;
2020         struct rte_ring *ring;
2021         struct thread_tx_conf *tx_conf;
2022         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2023         struct lthread_cond *ready;
2024
2025         tx_conf = (struct thread_tx_conf *)dummy;
2026         ring = tx_conf->ring;
2027         ready = *tx_conf->ready;
2028
2029         lthread_set_data((void *)tx_conf);
2030
2031         /*
2032          * Move this lthread to lcore
2033          */
2034         lthread_set_affinity(tx_conf->conf.lcore_id);
2035
2036         RTE_LOG(INFO, L3FWD, "entering main tx loop on lcore %u\n", rte_lcore_id());
2037
2038         nb_rx = 0;
2039         rte_atomic16_inc(&tx_counter);
2040         while (1) {
2041
2042                 /*
2043                  * Read packet from ring
2044                  */
2045                 SET_CPU_BUSY(tx_conf, CPU_POLL);
2046                 nb_rx = rte_ring_sc_dequeue_burst(ring, (void **)pkts_burst,
2047                                 MAX_PKT_BURST, NULL);
2048                 SET_CPU_IDLE(tx_conf, CPU_POLL);
2049
2050                 if (nb_rx > 0) {
2051                         SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2052                         portid = pkts_burst[0]->port;
2053                         process_burst(pkts_burst, nb_rx, portid);
2054                         SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2055                         lthread_yield();
2056                 } else
2057                         lthread_cond_wait(ready, 0);
2058
2059         }
2060         return NULL;
2061 }
2062
2063 /*
2064  * Main tx-lthreads spawner lthread.
2065  *
2066  * This lthread is used to spawn one new lthread per ring from producers.
2067  *
2068  */
2069 static void *
2070 lthread_tx(void *args)
2071 {
2072         struct lthread *lt;
2073
2074         unsigned lcore_id;
2075         uint16_t portid;
2076         struct thread_tx_conf *tx_conf;
2077
2078         tx_conf = (struct thread_tx_conf *)args;
2079         lthread_set_data((void *)tx_conf);
2080
2081         /*
2082          * Move this lthread to the selected lcore
2083          */
2084         lthread_set_affinity(tx_conf->conf.lcore_id);
2085
2086         /*
2087          * Spawn tx readers (one per input ring)
2088          */
2089         lthread_create(&lt, tx_conf->conf.lcore_id, lthread_tx_per_ring,
2090                         (void *)tx_conf);
2091
2092         lcore_id = rte_lcore_id();
2093
2094         RTE_LOG(INFO, L3FWD, "Entering Tx main loop on lcore %u\n", lcore_id);
2095
2096         tx_conf->conf.cpu_id = sched_getcpu();
2097         while (1) {
2098
2099                 lthread_sleep(BURST_TX_DRAIN_US * 1000);
2100
2101                 /*
2102                  * TX burst queue drain
2103                  */
2104                 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2105                         if (tx_conf->tx_mbufs[portid].len == 0)
2106                                 continue;
2107                         SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2108                         send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2109                         SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2110                         tx_conf->tx_mbufs[portid].len = 0;
2111                 }
2112
2113         }
2114         return NULL;
2115 }
2116
2117 static void *
2118 lthread_rx(void *dummy)
2119 {
2120         int ret;
2121         uint16_t nb_rx;
2122         int i;
2123         uint16_t portid;
2124         uint8_t queueid;
2125         int worker_id;
2126         int len[RTE_MAX_LCORE] = { 0 };
2127         int old_len, new_len;
2128         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2129         struct thread_rx_conf *rx_conf;
2130
2131         rx_conf = (struct thread_rx_conf *)dummy;
2132         lthread_set_data((void *)rx_conf);
2133
2134         /*
2135          * Move this lthread to lcore
2136          */
2137         lthread_set_affinity(rx_conf->conf.lcore_id);
2138
2139         if (rx_conf->n_rx_queue == 0) {
2140                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", rte_lcore_id());
2141                 return NULL;
2142         }
2143
2144         RTE_LOG(INFO, L3FWD, "Entering main Rx loop on lcore %u\n", rte_lcore_id());
2145
2146         for (i = 0; i < rx_conf->n_rx_queue; i++) {
2147
2148                 portid = rx_conf->rx_queue_list[i].port_id;
2149                 queueid = rx_conf->rx_queue_list[i].queue_id;
2150                 RTE_LOG(INFO, L3FWD,
2151                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2152                                 rte_lcore_id(), portid, queueid);
2153         }
2154
2155         /*
2156          * Init all condition variables (one per rx thread)
2157          */
2158         for (i = 0; i < rx_conf->n_rx_queue; i++)
2159                 lthread_cond_init(NULL, &rx_conf->ready[i], NULL);
2160
2161         worker_id = 0;
2162
2163         rx_conf->conf.cpu_id = sched_getcpu();
2164         rte_atomic16_inc(&rx_counter);
2165         while (1) {
2166
2167                 /*
2168                  * Read packet from RX queues
2169                  */
2170                 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2171                         portid = rx_conf->rx_queue_list[i].port_id;
2172                         queueid = rx_conf->rx_queue_list[i].queue_id;
2173
2174                         SET_CPU_BUSY(rx_conf, CPU_POLL);
2175                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2176                                 MAX_PKT_BURST);
2177                         SET_CPU_IDLE(rx_conf, CPU_POLL);
2178
2179                         if (nb_rx != 0) {
2180                                 worker_id = (worker_id + 1) % rx_conf->n_ring;
2181                                 old_len = len[worker_id];
2182
2183                                 SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2184                                 ret = rte_ring_sp_enqueue_burst(
2185                                                 rx_conf->ring[worker_id],
2186                                                 (void **) pkts_burst,
2187                                                 nb_rx, NULL);
2188
2189                                 new_len = old_len + ret;
2190
2191                                 if (new_len >= BURST_SIZE) {
2192                                         lthread_cond_signal(rx_conf->ready[worker_id]);
2193                                         new_len = 0;
2194                                 }
2195
2196                                 len[worker_id] = new_len;
2197
2198                                 if (unlikely(ret < nb_rx)) {
2199                                         uint32_t k;
2200
2201                                         for (k = ret; k < nb_rx; k++) {
2202                                                 struct rte_mbuf *m = pkts_burst[k];
2203
2204                                                 rte_pktmbuf_free(m);
2205                                         }
2206                                 }
2207                                 SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2208                         }
2209
2210                         lthread_yield();
2211                 }
2212         }
2213         return NULL;
2214 }
2215
2216 /*
2217  * Start scheduler with initial lthread on lcore
2218  *
2219  * This lthread loop spawns all rx and tx lthreads on main lcore
2220  */
2221
2222 static void *
2223 lthread_spawner(__rte_unused void *arg)
2224 {
2225         struct lthread *lt[MAX_THREAD];
2226         int i;
2227         int n_thread = 0;
2228
2229         printf("Entering lthread_spawner\n");
2230
2231         /*
2232          * Create producers (rx threads) on default lcore
2233          */
2234         for (i = 0; i < n_rx_thread; i++) {
2235                 rx_thread[i].conf.thread_id = i;
2236                 lthread_create(&lt[n_thread], -1, lthread_rx,
2237                                 (void *)&rx_thread[i]);
2238                 n_thread++;
2239         }
2240
2241         /*
2242          * Wait for all producers. Until some producers can be started on the same
2243          * scheduler as this lthread, yielding is required to let them to run and
2244          * prevent deadlock here.
2245          */
2246         while (rte_atomic16_read(&rx_counter) < n_rx_thread)
2247                 lthread_sleep(100000);
2248
2249         /*
2250          * Create consumers (tx threads) on default lcore_id
2251          */
2252         for (i = 0; i < n_tx_thread; i++) {
2253                 tx_thread[i].conf.thread_id = i;
2254                 lthread_create(&lt[n_thread], -1, lthread_tx,
2255                                 (void *)&tx_thread[i]);
2256                 n_thread++;
2257         }
2258
2259         /*
2260          * Wait for all threads finished
2261          */
2262         for (i = 0; i < n_thread; i++)
2263                 lthread_join(lt[i], NULL);
2264
2265         return NULL;
2266 }
2267
2268 /*
2269  * Start main scheduler with initial lthread spawning rx and tx lthreads
2270  * (main_lthread_main).
2271  */
2272 static int
2273 lthread_main_spawner(__rte_unused void *arg) {
2274         struct lthread *lt;
2275         int lcore_id = rte_lcore_id();
2276
2277         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2278         lthread_create(&lt, -1, lthread_spawner, NULL);
2279         lthread_run();
2280
2281         return 0;
2282 }
2283
2284 /*
2285  * Start scheduler on lcore.
2286  */
2287 static int
2288 sched_spawner(__rte_unused void *arg) {
2289         struct lthread *lt;
2290         int lcore_id = rte_lcore_id();
2291
2292 #if (APP_CPU_LOAD)
2293         if (lcore_id == cpu_load_lcore_id) {
2294                 cpu_load_collector(arg);
2295                 return 0;
2296         }
2297 #endif /* APP_CPU_LOAD */
2298
2299         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2300         lthread_create(&lt, -1, lthread_null, NULL);
2301         lthread_run();
2302
2303         return 0;
2304 }
2305
2306 /* main processing loop */
2307 static int __rte_noreturn
2308 pthread_tx(void *dummy)
2309 {
2310         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2311         uint64_t prev_tsc, diff_tsc, cur_tsc;
2312         int nb_rx;
2313         uint16_t portid;
2314         struct thread_tx_conf *tx_conf;
2315
2316         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
2317                 US_PER_S * BURST_TX_DRAIN_US;
2318
2319         prev_tsc = 0;
2320
2321         tx_conf = (struct thread_tx_conf *)dummy;
2322
2323         RTE_LOG(INFO, L3FWD, "Entering main Tx loop on lcore %u\n", rte_lcore_id());
2324
2325         tx_conf->conf.cpu_id = sched_getcpu();
2326         rte_atomic16_inc(&tx_counter);
2327         while (1) {
2328
2329                 cur_tsc = rte_rdtsc();
2330
2331                 /*
2332                  * TX burst queue drain
2333                  */
2334                 diff_tsc = cur_tsc - prev_tsc;
2335                 if (unlikely(diff_tsc > drain_tsc)) {
2336
2337                         /*
2338                          * This could be optimized (use queueid instead of
2339                          * portid), but it is not called so often
2340                          */
2341                         SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2342                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2343                                 if (tx_conf->tx_mbufs[portid].len == 0)
2344                                         continue;
2345                                 send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2346                                 tx_conf->tx_mbufs[portid].len = 0;
2347                         }
2348                         SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2349
2350                         prev_tsc = cur_tsc;
2351                 }
2352
2353                 /*
2354                  * Read packet from ring
2355                  */
2356                 SET_CPU_BUSY(tx_conf, CPU_POLL);
2357                 nb_rx = rte_ring_sc_dequeue_burst(tx_conf->ring,
2358                                 (void **)pkts_burst, MAX_PKT_BURST, NULL);
2359                 SET_CPU_IDLE(tx_conf, CPU_POLL);
2360
2361                 if (unlikely(nb_rx == 0)) {
2362                         sched_yield();
2363                         continue;
2364                 }
2365
2366                 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2367                 portid = pkts_burst[0]->port;
2368                 process_burst(pkts_burst, nb_rx, portid);
2369                 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2370
2371         }
2372 }
2373
2374 static int
2375 pthread_rx(void *dummy)
2376 {
2377         int i;
2378         int worker_id;
2379         uint32_t n;
2380         uint32_t nb_rx;
2381         unsigned lcore_id;
2382         uint8_t queueid;
2383         uint16_t portid;
2384         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2385
2386         struct thread_rx_conf *rx_conf;
2387
2388         lcore_id = rte_lcore_id();
2389         rx_conf = (struct thread_rx_conf *)dummy;
2390
2391         if (rx_conf->n_rx_queue == 0) {
2392                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
2393                 return 0;
2394         }
2395
2396         RTE_LOG(INFO, L3FWD, "entering main rx loop on lcore %u\n", lcore_id);
2397
2398         for (i = 0; i < rx_conf->n_rx_queue; i++) {
2399
2400                 portid = rx_conf->rx_queue_list[i].port_id;
2401                 queueid = rx_conf->rx_queue_list[i].queue_id;
2402                 RTE_LOG(INFO, L3FWD,
2403                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2404                                 lcore_id, portid, queueid);
2405         }
2406
2407         worker_id = 0;
2408         rx_conf->conf.cpu_id = sched_getcpu();
2409         rte_atomic16_inc(&rx_counter);
2410         while (1) {
2411
2412                 /*
2413                  * Read packet from RX queues
2414                  */
2415                 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2416                         portid = rx_conf->rx_queue_list[i].port_id;
2417                         queueid = rx_conf->rx_queue_list[i].queue_id;
2418
2419                         SET_CPU_BUSY(rx_conf, CPU_POLL);
2420                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2421                                 MAX_PKT_BURST);
2422                         SET_CPU_IDLE(rx_conf, CPU_POLL);
2423
2424                         if (nb_rx == 0) {
2425                                 sched_yield();
2426                                 continue;
2427                         }
2428
2429                         SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2430                         worker_id = (worker_id + 1) % rx_conf->n_ring;
2431                         n = rte_ring_sp_enqueue_burst(rx_conf->ring[worker_id],
2432                                         (void **)pkts_burst, nb_rx, NULL);
2433
2434                         if (unlikely(n != nb_rx)) {
2435                                 uint32_t k;
2436
2437                                 for (k = n; k < nb_rx; k++) {
2438                                         struct rte_mbuf *m = pkts_burst[k];
2439
2440                                         rte_pktmbuf_free(m);
2441                                 }
2442                         }
2443
2444                         SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2445
2446                 }
2447         }
2448 }
2449
2450 /*
2451  * P-Thread spawner.
2452  */
2453 static int
2454 pthread_run(__rte_unused void *arg) {
2455         int lcore_id = rte_lcore_id();
2456         int i;
2457
2458         for (i = 0; i < n_rx_thread; i++)
2459                 if (rx_thread[i].conf.lcore_id == lcore_id) {
2460                         printf("Start rx thread on %d...\n", lcore_id);
2461                         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2462                         RTE_PER_LCORE(lcore_conf)->data = (void *)&rx_thread[i];
2463                         pthread_rx((void *)&rx_thread[i]);
2464                         return 0;
2465                 }
2466
2467         for (i = 0; i < n_tx_thread; i++)
2468                 if (tx_thread[i].conf.lcore_id == lcore_id) {
2469                         printf("Start tx thread on %d...\n", lcore_id);
2470                         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2471                         RTE_PER_LCORE(lcore_conf)->data = (void *)&tx_thread[i];
2472                         pthread_tx((void *)&tx_thread[i]);
2473                         return 0;
2474                 }
2475
2476 #if (APP_CPU_LOAD)
2477         if (lcore_id == cpu_load_lcore_id)
2478                 cpu_load_collector(arg);
2479 #endif /* APP_CPU_LOAD */
2480
2481         return 0;
2482 }
2483
2484 static int
2485 check_lcore_params(void)
2486 {
2487         uint8_t queue, lcore;
2488         uint16_t i;
2489         int socketid;
2490
2491         for (i = 0; i < nb_rx_thread_params; ++i) {
2492                 queue = rx_thread_params[i].queue_id;
2493                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
2494                         printf("invalid queue number: %hhu\n", queue);
2495                         return -1;
2496                 }
2497                 lcore = rx_thread_params[i].lcore_id;
2498                 if (!rte_lcore_is_enabled(lcore)) {
2499                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
2500                         return -1;
2501                 }
2502                 socketid = rte_lcore_to_socket_id(lcore);
2503                 if ((socketid != 0) && (numa_on == 0))
2504                         printf("warning: lcore %hhu is on socket %d with numa off\n",
2505                                 lcore, socketid);
2506         }
2507         return 0;
2508 }
2509
2510 static int
2511 check_port_config(void)
2512 {
2513         unsigned portid;
2514         uint16_t i;
2515
2516         for (i = 0; i < nb_rx_thread_params; ++i) {
2517                 portid = rx_thread_params[i].port_id;
2518                 if ((enabled_port_mask & (1 << portid)) == 0) {
2519                         printf("port %u is not enabled in port mask\n", portid);
2520                         return -1;
2521                 }
2522                 if (!rte_eth_dev_is_valid_port(portid)) {
2523                         printf("port %u is not present on the board\n", portid);
2524                         return -1;
2525                 }
2526         }
2527         return 0;
2528 }
2529
2530 static uint8_t
2531 get_port_n_rx_queues(const uint16_t port)
2532 {
2533         int queue = -1;
2534         uint16_t i;
2535
2536         for (i = 0; i < nb_rx_thread_params; ++i)
2537                 if (rx_thread_params[i].port_id == port &&
2538                                 rx_thread_params[i].queue_id > queue)
2539                         queue = rx_thread_params[i].queue_id;
2540
2541         return (uint8_t)(++queue);
2542 }
2543
2544 static int
2545 init_rx_rings(void)
2546 {
2547         unsigned socket_io;
2548         struct thread_rx_conf *rx_conf;
2549         struct thread_tx_conf *tx_conf;
2550         unsigned rx_thread_id, tx_thread_id;
2551         char name[256];
2552         struct rte_ring *ring = NULL;
2553
2554         for (tx_thread_id = 0; tx_thread_id < n_tx_thread; tx_thread_id++) {
2555
2556                 tx_conf = &tx_thread[tx_thread_id];
2557
2558                 printf("Connecting tx-thread %d with rx-thread %d\n", tx_thread_id,
2559                                 tx_conf->conf.thread_id);
2560
2561                 rx_thread_id = tx_conf->conf.thread_id;
2562                 if (rx_thread_id > n_tx_thread) {
2563                         printf("connection from tx-thread %u to rx-thread %u fails "
2564                                         "(rx-thread not defined)\n", tx_thread_id, rx_thread_id);
2565                         return -1;
2566                 }
2567
2568                 rx_conf = &rx_thread[rx_thread_id];
2569                 socket_io = rte_lcore_to_socket_id(rx_conf->conf.lcore_id);
2570
2571                 snprintf(name, sizeof(name), "app_ring_s%u_rx%u_tx%u",
2572                                 socket_io, rx_thread_id, tx_thread_id);
2573
2574                 ring = rte_ring_create(name, 1024 * 4, socket_io,
2575                                 RING_F_SP_ENQ | RING_F_SC_DEQ);
2576
2577                 if (ring == NULL) {
2578                         rte_panic("Cannot create ring to connect rx-thread %u "
2579                                         "with tx-thread %u\n", rx_thread_id, tx_thread_id);
2580                 }
2581
2582                 rx_conf->ring[rx_conf->n_ring] = ring;
2583
2584                 tx_conf->ring = ring;
2585                 tx_conf->ready = &rx_conf->ready[rx_conf->n_ring];
2586
2587                 rx_conf->n_ring++;
2588         }
2589         return 0;
2590 }
2591
2592 static int
2593 init_rx_queues(void)
2594 {
2595         uint16_t i, nb_rx_queue;
2596         uint8_t thread;
2597
2598         n_rx_thread = 0;
2599
2600         for (i = 0; i < nb_rx_thread_params; ++i) {
2601                 thread = rx_thread_params[i].thread_id;
2602                 nb_rx_queue = rx_thread[thread].n_rx_queue;
2603
2604                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
2605                         printf("error: too many queues (%u) for thread: %u\n",
2606                                 (unsigned)nb_rx_queue + 1, (unsigned)thread);
2607                         return -1;
2608                 }
2609
2610                 rx_thread[thread].conf.thread_id = thread;
2611                 rx_thread[thread].conf.lcore_id = rx_thread_params[i].lcore_id;
2612                 rx_thread[thread].rx_queue_list[nb_rx_queue].port_id =
2613                         rx_thread_params[i].port_id;
2614                 rx_thread[thread].rx_queue_list[nb_rx_queue].queue_id =
2615                         rx_thread_params[i].queue_id;
2616                 rx_thread[thread].n_rx_queue++;
2617
2618                 if (thread >= n_rx_thread)
2619                         n_rx_thread = thread + 1;
2620
2621         }
2622         return 0;
2623 }
2624
2625 static int
2626 init_tx_threads(void)
2627 {
2628         int i;
2629
2630         n_tx_thread = 0;
2631         for (i = 0; i < nb_tx_thread_params; ++i) {
2632                 tx_thread[n_tx_thread].conf.thread_id = tx_thread_params[i].thread_id;
2633                 tx_thread[n_tx_thread].conf.lcore_id = tx_thread_params[i].lcore_id;
2634                 n_tx_thread++;
2635         }
2636         return 0;
2637 }
2638
2639 /* display usage */
2640 static void
2641 print_usage(const char *prgname)
2642 {
2643         printf("%s [EAL options] -- -p PORTMASK -P"
2644                 "  [--rx (port,queue,lcore,thread)[,(port,queue,lcore,thread]]"
2645                 "  [--tx (lcore,thread)[,(lcore,thread]]"
2646                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
2647                 "  [--parse-ptype]\n\n"
2648                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
2649                 "  -P : enable promiscuous mode\n"
2650                 "  --rx (port,queue,lcore,thread): rx queues configuration\n"
2651                 "  --tx (lcore,thread): tx threads configuration\n"
2652                 "  --stat-lcore LCORE: use lcore for stat collector\n"
2653                 "  --eth-dest=X,MM:MM:MM:MM:MM:MM: optional, ethernet destination for port X\n"
2654                 "  --no-numa: optional, disable numa awareness\n"
2655                 "  --ipv6: optional, specify it if running ipv6 packets\n"
2656                 "  --enable-jumbo: enable jumbo frame"
2657                 " which max packet len is PKTLEN in decimal (64-9600)\n"
2658                 "  --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n"
2659                 "  --no-lthreads: turn off lthread model\n"
2660                 "  --parse-ptype: set to use software to analyze packet type\n\n",
2661                 prgname);
2662 }
2663
2664 static int parse_max_pkt_len(const char *pktlen)
2665 {
2666         char *end = NULL;
2667         unsigned long len;
2668
2669         /* parse decimal string */
2670         len = strtoul(pktlen, &end, 10);
2671         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
2672                 return -1;
2673
2674         if (len == 0)
2675                 return -1;
2676
2677         return len;
2678 }
2679
2680 static int
2681 parse_portmask(const char *portmask)
2682 {
2683         char *end = NULL;
2684         unsigned long pm;
2685
2686         /* parse hexadecimal string */
2687         pm = strtoul(portmask, &end, 16);
2688         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
2689                 return 0;
2690
2691         return pm;
2692 }
2693
2694 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2695 static int
2696 parse_hash_entry_number(const char *hash_entry_num)
2697 {
2698         char *end = NULL;
2699         unsigned long hash_en;
2700
2701         /* parse hexadecimal string */
2702         hash_en = strtoul(hash_entry_num, &end, 16);
2703         if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
2704                 return -1;
2705
2706         if (hash_en == 0)
2707                 return -1;
2708
2709         return hash_en;
2710 }
2711 #endif
2712
2713 static int
2714 parse_rx_config(const char *q_arg)
2715 {
2716         char s[256];
2717         const char *p, *p0 = q_arg;
2718         char *end;
2719         enum fieldnames {
2720                 FLD_PORT = 0,
2721                 FLD_QUEUE,
2722                 FLD_LCORE,
2723                 FLD_THREAD,
2724                 _NUM_FLD
2725         };
2726         unsigned long int_fld[_NUM_FLD];
2727         char *str_fld[_NUM_FLD];
2728         int i;
2729         unsigned size;
2730
2731         nb_rx_thread_params = 0;
2732
2733         while ((p = strchr(p0, '(')) != NULL) {
2734                 ++p;
2735                 p0 = strchr(p, ')');
2736                 if (p0 == NULL)
2737                         return -1;
2738
2739                 size = p0 - p;
2740                 if (size >= sizeof(s))
2741                         return -1;
2742
2743                 snprintf(s, sizeof(s), "%.*s", size, p);
2744                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2745                         return -1;
2746                 for (i = 0; i < _NUM_FLD; i++) {
2747                         errno = 0;
2748                         int_fld[i] = strtoul(str_fld[i], &end, 0);
2749                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2750                                 return -1;
2751                 }
2752                 if (nb_rx_thread_params >= MAX_LCORE_PARAMS) {
2753                         printf("exceeded max number of rx params: %hu\n",
2754                                         nb_rx_thread_params);
2755                         return -1;
2756                 }
2757                 rx_thread_params_array[nb_rx_thread_params].port_id =
2758                                 int_fld[FLD_PORT];
2759                 rx_thread_params_array[nb_rx_thread_params].queue_id =
2760                                 (uint8_t)int_fld[FLD_QUEUE];
2761                 rx_thread_params_array[nb_rx_thread_params].lcore_id =
2762                                 (uint8_t)int_fld[FLD_LCORE];
2763                 rx_thread_params_array[nb_rx_thread_params].thread_id =
2764                                 (uint8_t)int_fld[FLD_THREAD];
2765                 ++nb_rx_thread_params;
2766         }
2767         rx_thread_params = rx_thread_params_array;
2768         return 0;
2769 }
2770
2771 static int
2772 parse_tx_config(const char *q_arg)
2773 {
2774         char s[256];
2775         const char *p, *p0 = q_arg;
2776         char *end;
2777         enum fieldnames {
2778                 FLD_LCORE = 0,
2779                 FLD_THREAD,
2780                 _NUM_FLD
2781         };
2782         unsigned long int_fld[_NUM_FLD];
2783         char *str_fld[_NUM_FLD];
2784         int i;
2785         unsigned size;
2786
2787         nb_tx_thread_params = 0;
2788
2789         while ((p = strchr(p0, '(')) != NULL) {
2790                 ++p;
2791                 p0 = strchr(p, ')');
2792                 if (p0 == NULL)
2793                         return -1;
2794
2795                 size = p0 - p;
2796                 if (size >= sizeof(s))
2797                         return -1;
2798
2799                 snprintf(s, sizeof(s), "%.*s", size, p);
2800                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2801                         return -1;
2802                 for (i = 0; i < _NUM_FLD; i++) {
2803                         errno = 0;
2804                         int_fld[i] = strtoul(str_fld[i], &end, 0);
2805                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2806                                 return -1;
2807                 }
2808                 if (nb_tx_thread_params >= MAX_LCORE_PARAMS) {
2809                         printf("exceeded max number of tx params: %hu\n",
2810                                 nb_tx_thread_params);
2811                         return -1;
2812                 }
2813                 tx_thread_params_array[nb_tx_thread_params].lcore_id =
2814                                 (uint8_t)int_fld[FLD_LCORE];
2815                 tx_thread_params_array[nb_tx_thread_params].thread_id =
2816                                 (uint8_t)int_fld[FLD_THREAD];
2817                 ++nb_tx_thread_params;
2818         }
2819         tx_thread_params = tx_thread_params_array;
2820
2821         return 0;
2822 }
2823
2824 #if (APP_CPU_LOAD > 0)
2825 static int
2826 parse_stat_lcore(const char *stat_lcore)
2827 {
2828         char *end = NULL;
2829         unsigned long lcore_id;
2830
2831         lcore_id = strtoul(stat_lcore, &end, 10);
2832         if ((stat_lcore[0] == '\0') || (end == NULL) || (*end != '\0'))
2833                 return -1;
2834
2835         return lcore_id;
2836 }
2837 #endif
2838
2839 static void
2840 parse_eth_dest(const char *optarg)
2841 {
2842         uint16_t portid;
2843         char *port_end;
2844         uint8_t c, *dest, peer_addr[6];
2845
2846         errno = 0;
2847         portid = strtoul(optarg, &port_end, 10);
2848         if (errno != 0 || port_end == optarg || *port_end++ != ',')
2849                 rte_exit(EXIT_FAILURE,
2850                 "Invalid eth-dest: %s", optarg);
2851         if (portid >= RTE_MAX_ETHPORTS)
2852                 rte_exit(EXIT_FAILURE,
2853                 "eth-dest: port %d >= RTE_MAX_ETHPORTS(%d)\n",
2854                 portid, RTE_MAX_ETHPORTS);
2855
2856         if (cmdline_parse_etheraddr(NULL, port_end,
2857                 &peer_addr, sizeof(peer_addr)) < 0)
2858                 rte_exit(EXIT_FAILURE,
2859                 "Invalid ethernet address: %s\n",
2860                 port_end);
2861         dest = (uint8_t *)&dest_eth_addr[portid];
2862         for (c = 0; c < 6; c++)
2863                 dest[c] = peer_addr[c];
2864         *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
2865 }
2866
2867 enum {
2868 #define OPT_RX_CONFIG       "rx"
2869         OPT_RX_CONFIG_NUM = 256,
2870 #define OPT_TX_CONFIG       "tx"
2871         OPT_TX_CONFIG_NUM,
2872 #define OPT_STAT_LCORE      "stat-lcore"
2873         OPT_STAT_LCORE_NUM,
2874 #define OPT_ETH_DEST        "eth-dest"
2875         OPT_ETH_DEST_NUM,
2876 #define OPT_NO_NUMA         "no-numa"
2877         OPT_NO_NUMA_NUM,
2878 #define OPT_IPV6            "ipv6"
2879         OPT_IPV6_NUM,
2880 #define OPT_ENABLE_JUMBO    "enable-jumbo"
2881         OPT_ENABLE_JUMBO_NUM,
2882 #define OPT_HASH_ENTRY_NUM  "hash-entry-num"
2883         OPT_HASH_ENTRY_NUM_NUM,
2884 #define OPT_NO_LTHREADS     "no-lthreads"
2885         OPT_NO_LTHREADS_NUM,
2886 #define OPT_PARSE_PTYPE     "parse-ptype"
2887         OPT_PARSE_PTYPE_NUM,
2888 };
2889
2890 /* Parse the argument given in the command line of the application */
2891 static int
2892 parse_args(int argc, char **argv)
2893 {
2894         int opt, ret;
2895         char **argvopt;
2896         int option_index;
2897         char *prgname = argv[0];
2898         static struct option lgopts[] = {
2899                 {OPT_RX_CONFIG,      1, NULL, OPT_RX_CONFIG_NUM      },
2900                 {OPT_TX_CONFIG,      1, NULL, OPT_TX_CONFIG_NUM      },
2901                 {OPT_STAT_LCORE,     1, NULL, OPT_STAT_LCORE_NUM     },
2902                 {OPT_ETH_DEST,       1, NULL, OPT_ETH_DEST_NUM       },
2903                 {OPT_NO_NUMA,        0, NULL, OPT_NO_NUMA_NUM        },
2904                 {OPT_IPV6,           0, NULL, OPT_IPV6_NUM           },
2905                 {OPT_ENABLE_JUMBO,   0, NULL, OPT_ENABLE_JUMBO_NUM   },
2906                 {OPT_HASH_ENTRY_NUM, 1, NULL, OPT_HASH_ENTRY_NUM_NUM },
2907                 {OPT_NO_LTHREADS,    0, NULL, OPT_NO_LTHREADS_NUM    },
2908                 {OPT_PARSE_PTYPE,    0, NULL, OPT_PARSE_PTYPE_NUM    },
2909                 {NULL,               0, 0,    0                      }
2910         };
2911
2912         argvopt = argv;
2913
2914         while ((opt = getopt_long(argc, argvopt, "p:P",
2915                                 lgopts, &option_index)) != EOF) {
2916
2917                 switch (opt) {
2918                 /* portmask */
2919                 case 'p':
2920                         enabled_port_mask = parse_portmask(optarg);
2921                         if (enabled_port_mask == 0) {
2922                                 printf("invalid portmask\n");
2923                                 print_usage(prgname);
2924                                 return -1;
2925                         }
2926                         break;
2927
2928                 case 'P':
2929                         printf("Promiscuous mode selected\n");
2930                         promiscuous_on = 1;
2931                         break;
2932
2933                 /* long options */
2934                 case OPT_RX_CONFIG_NUM:
2935                         ret = parse_rx_config(optarg);
2936                         if (ret) {
2937                                 printf("invalid rx-config\n");
2938                                 print_usage(prgname);
2939                                 return -1;
2940                         }
2941                         break;
2942
2943                 case OPT_TX_CONFIG_NUM:
2944                         ret = parse_tx_config(optarg);
2945                         if (ret) {
2946                                 printf("invalid tx-config\n");
2947                                 print_usage(prgname);
2948                                 return -1;
2949                         }
2950                         break;
2951
2952 #if (APP_CPU_LOAD > 0)
2953                 case OPT_STAT_LCORE_NUM:
2954                         cpu_load_lcore_id = parse_stat_lcore(optarg);
2955                         break;
2956 #endif
2957
2958                 case OPT_ETH_DEST_NUM:
2959                         parse_eth_dest(optarg);
2960                         break;
2961
2962                 case OPT_NO_NUMA_NUM:
2963                         printf("numa is disabled\n");
2964                         numa_on = 0;
2965                         break;
2966
2967 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2968                 case OPT_IPV6_NUM:
2969                         printf("ipv6 is specified\n");
2970                         ipv6 = 1;
2971                         break;
2972 #endif
2973
2974                 case OPT_NO_LTHREADS_NUM:
2975                         printf("l-threads model is disabled\n");
2976                         lthreads_on = 0;
2977                         break;
2978
2979                 case OPT_PARSE_PTYPE_NUM:
2980                         printf("software packet type parsing enabled\n");
2981                         parse_ptype_on = 1;
2982                         break;
2983
2984                 case OPT_ENABLE_JUMBO_NUM:
2985                 {
2986                         struct option lenopts = {"max-pkt-len",
2987                                         required_argument, 0, 0};
2988
2989                         printf("jumbo frame is enabled - disabling simple TX path\n");
2990                         port_conf.rxmode.offloads |=
2991                                         DEV_RX_OFFLOAD_JUMBO_FRAME;
2992                         port_conf.txmode.offloads |=
2993                                         DEV_TX_OFFLOAD_MULTI_SEGS;
2994
2995                         /* if no max-pkt-len set, use the default value
2996                          * RTE_ETHER_MAX_LEN
2997                          */
2998                         if (getopt_long(argc, argvopt, "", &lenopts,
2999                                         &option_index) == 0) {
3000
3001                                 ret = parse_max_pkt_len(optarg);
3002                                 if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)) {
3003                                         printf("invalid packet length\n");
3004                                         print_usage(prgname);
3005                                         return -1;
3006                                 }
3007                                 port_conf.rxmode.max_rx_pkt_len = ret;
3008                         }
3009                         printf("set jumbo frame max packet length to %u\n",
3010                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
3011                         break;
3012                 }
3013 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
3014                 case OPT_HASH_ENTRY_NUM_NUM:
3015                         ret = parse_hash_entry_number(optarg);
3016                         if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
3017                                 hash_entry_number = ret;
3018                         } else {
3019                                 printf("invalid hash entry number\n");
3020                                 print_usage(prgname);
3021                                 return -1;
3022                         }
3023                         break;
3024 #endif
3025
3026                 default:
3027                         print_usage(prgname);
3028                         return -1;
3029                 }
3030         }
3031
3032         if (optind >= 0)
3033                 argv[optind-1] = prgname;
3034
3035         ret = optind-1;
3036         optind = 1; /* reset getopt lib */
3037         return ret;
3038 }
3039
3040 static void
3041 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
3042 {
3043         char buf[RTE_ETHER_ADDR_FMT_SIZE];
3044
3045         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
3046         printf("%s%s", name, buf);
3047 }
3048
3049 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
3050
3051 static void convert_ipv4_5tuple(struct ipv4_5tuple *key1,
3052                 union ipv4_5tuple_host *key2)
3053 {
3054         key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
3055         key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
3056         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3057         key2->port_src = rte_cpu_to_be_16(key1->port_src);
3058         key2->proto = key1->proto;
3059         key2->pad0 = 0;
3060         key2->pad1 = 0;
3061 }
3062
3063 static void convert_ipv6_5tuple(struct ipv6_5tuple *key1,
3064                 union ipv6_5tuple_host *key2)
3065 {
3066         uint32_t i;
3067
3068         for (i = 0; i < 16; i++) {
3069                 key2->ip_dst[i] = key1->ip_dst[i];
3070                 key2->ip_src[i] = key1->ip_src[i];
3071         }
3072         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3073         key2->port_src = rte_cpu_to_be_16(key1->port_src);
3074         key2->proto = key1->proto;
3075         key2->pad0 = 0;
3076         key2->pad1 = 0;
3077         key2->reserve = 0;
3078 }
3079
3080 #define BYTE_VALUE_MAX 256
3081 #define ALL_32_BITS 0xffffffff
3082 #define BIT_8_TO_15 0x0000ff00
3083 static inline void
3084 populate_ipv4_few_flow_into_table(const struct rte_hash *h)
3085 {
3086         uint32_t i;
3087         int32_t ret;
3088         uint32_t array_len = RTE_DIM(ipv4_l3fwd_route_array);
3089
3090         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3091         for (i = 0; i < array_len; i++) {
3092                 struct ipv4_l3fwd_route  entry;
3093                 union ipv4_5tuple_host newkey;
3094
3095                 entry = ipv4_l3fwd_route_array[i];
3096                 convert_ipv4_5tuple(&entry.key, &newkey);
3097                 ret = rte_hash_add_key(h, (void *)&newkey);
3098                 if (ret < 0) {
3099                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3100                                 " to the l3fwd hash.\n", i);
3101                 }
3102                 ipv4_l3fwd_out_if[ret] = entry.if_out;
3103         }
3104         printf("Hash: Adding 0x%" PRIx32 " keys\n", array_len);
3105 }
3106
3107 #define BIT_16_TO_23 0x00ff0000
3108 static inline void
3109 populate_ipv6_few_flow_into_table(const struct rte_hash *h)
3110 {
3111         uint32_t i;
3112         int32_t ret;
3113         uint32_t array_len = RTE_DIM(ipv6_l3fwd_route_array);
3114
3115         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3116         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3117         for (i = 0; i < array_len; i++) {
3118                 struct ipv6_l3fwd_route entry;
3119                 union ipv6_5tuple_host newkey;
3120
3121                 entry = ipv6_l3fwd_route_array[i];
3122                 convert_ipv6_5tuple(&entry.key, &newkey);
3123                 ret = rte_hash_add_key(h, (void *)&newkey);
3124                 if (ret < 0) {
3125                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3126                                 " to the l3fwd hash.\n", i);
3127                 }
3128                 ipv6_l3fwd_out_if[ret] = entry.if_out;
3129         }
3130         printf("Hash: Adding 0x%" PRIx32 "keys\n", array_len);
3131 }
3132
3133 #define NUMBER_PORT_USED 4
3134 static inline void
3135 populate_ipv4_many_flow_into_table(const struct rte_hash *h,
3136                 unsigned int nr_flow)
3137 {
3138         unsigned i;
3139
3140         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3141
3142         for (i = 0; i < nr_flow; i++) {
3143                 struct ipv4_l3fwd_route entry;
3144                 union ipv4_5tuple_host newkey;
3145                 uint8_t a = (uint8_t)((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3146                 uint8_t b = (uint8_t)(((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3147                                 BYTE_VALUE_MAX);
3148                 uint8_t c = (uint8_t)((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3149                                 BYTE_VALUE_MAX));
3150                 /* Create the ipv4 exact match flow */
3151                 memset(&entry, 0, sizeof(entry));
3152                 switch (i & (NUMBER_PORT_USED - 1)) {
3153                 case 0:
3154                         entry = ipv4_l3fwd_route_array[0];
3155                         entry.key.ip_dst = RTE_IPV4(101, c, b, a);
3156                         break;
3157                 case 1:
3158                         entry = ipv4_l3fwd_route_array[1];
3159                         entry.key.ip_dst = RTE_IPV4(201, c, b, a);
3160                         break;
3161                 case 2:
3162                         entry = ipv4_l3fwd_route_array[2];
3163                         entry.key.ip_dst = RTE_IPV4(111, c, b, a);
3164                         break;
3165                 case 3:
3166                         entry = ipv4_l3fwd_route_array[3];
3167                         entry.key.ip_dst = RTE_IPV4(211, c, b, a);
3168                         break;
3169                 };
3170                 convert_ipv4_5tuple(&entry.key, &newkey);
3171                 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3172
3173                 if (ret < 0)
3174                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3175
3176                 ipv4_l3fwd_out_if[ret] = (uint8_t)entry.if_out;
3177
3178         }
3179         printf("Hash: Adding 0x%x keys\n", nr_flow);
3180 }
3181
3182 static inline void
3183 populate_ipv6_many_flow_into_table(const struct rte_hash *h,
3184                 unsigned int nr_flow)
3185 {
3186         unsigned i;
3187
3188         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3189         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3190         for (i = 0; i < nr_flow; i++) {
3191                 struct ipv6_l3fwd_route entry;
3192                 union ipv6_5tuple_host newkey;
3193
3194                 uint8_t a = (uint8_t) ((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3195                 uint8_t b = (uint8_t) (((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3196                                 BYTE_VALUE_MAX);
3197                 uint8_t c = (uint8_t) ((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3198                                 BYTE_VALUE_MAX));
3199
3200                 /* Create the ipv6 exact match flow */
3201                 memset(&entry, 0, sizeof(entry));
3202                 switch (i & (NUMBER_PORT_USED - 1)) {
3203                 case 0:
3204                         entry = ipv6_l3fwd_route_array[0];
3205                         break;
3206                 case 1:
3207                         entry = ipv6_l3fwd_route_array[1];
3208                         break;
3209                 case 2:
3210                         entry = ipv6_l3fwd_route_array[2];
3211                         break;
3212                 case 3:
3213                         entry = ipv6_l3fwd_route_array[3];
3214                         break;
3215                 };
3216                 entry.key.ip_dst[13] = c;
3217                 entry.key.ip_dst[14] = b;
3218                 entry.key.ip_dst[15] = a;
3219                 convert_ipv6_5tuple(&entry.key, &newkey);
3220                 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3221
3222                 if (ret < 0)
3223                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3224
3225                 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
3226
3227         }
3228         printf("Hash: Adding 0x%x keys\n", nr_flow);
3229 }
3230
3231 static void
3232 setup_hash(int socketid)
3233 {
3234         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
3235                 .name = NULL,
3236                 .entries = L3FWD_HASH_ENTRIES,
3237                 .key_len = sizeof(union ipv4_5tuple_host),
3238                 .hash_func = ipv4_hash_crc,
3239                 .hash_func_init_val = 0,
3240         };
3241
3242         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
3243                 .name = NULL,
3244                 .entries = L3FWD_HASH_ENTRIES,
3245                 .key_len = sizeof(union ipv6_5tuple_host),
3246                 .hash_func = ipv6_hash_crc,
3247                 .hash_func_init_val = 0,
3248         };
3249
3250         char s[64];
3251
3252         /* create ipv4 hash */
3253         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
3254         ipv4_l3fwd_hash_params.name = s;
3255         ipv4_l3fwd_hash_params.socket_id = socketid;
3256         ipv4_l3fwd_lookup_struct[socketid] =
3257                         rte_hash_create(&ipv4_l3fwd_hash_params);
3258         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3259                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3260                                 "socket %d\n", socketid);
3261
3262         /* create ipv6 hash */
3263         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
3264         ipv6_l3fwd_hash_params.name = s;
3265         ipv6_l3fwd_hash_params.socket_id = socketid;
3266         ipv6_l3fwd_lookup_struct[socketid] =
3267                         rte_hash_create(&ipv6_l3fwd_hash_params);
3268         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3269                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3270                                 "socket %d\n", socketid);
3271
3272         if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
3273                 /* For testing hash matching with a large number of flows we
3274                  * generate millions of IP 5-tuples with an incremented dst
3275                  * address to initialize the hash table. */
3276                 if (ipv6 == 0) {
3277                         /* populate the ipv4 hash */
3278                         populate_ipv4_many_flow_into_table(
3279                                 ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
3280                 } else {
3281                         /* populate the ipv6 hash */
3282                         populate_ipv6_many_flow_into_table(
3283                                 ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
3284                 }
3285         } else {
3286                 /* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize
3287                  * the hash table */
3288                 if (ipv6 == 0) {
3289                         /* populate the ipv4 hash */
3290                         populate_ipv4_few_flow_into_table(
3291                                         ipv4_l3fwd_lookup_struct[socketid]);
3292                 } else {
3293                         /* populate the ipv6 hash */
3294                         populate_ipv6_few_flow_into_table(
3295                                         ipv6_l3fwd_lookup_struct[socketid]);
3296                 }
3297         }
3298 }
3299 #endif
3300
3301 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3302 static void
3303 setup_lpm(int socketid)
3304 {
3305         struct rte_lpm6_config config;
3306         struct rte_lpm_config lpm_ipv4_config;
3307         unsigned i;
3308         int ret;
3309         char s[64];
3310
3311         /* create the LPM table */
3312         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
3313         lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
3314         lpm_ipv4_config.number_tbl8s = 256;
3315         lpm_ipv4_config.flags = 0;
3316         ipv4_l3fwd_lookup_struct[socketid] =
3317                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
3318         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3319                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3320                                 " on socket %d\n", socketid);
3321
3322         /* populate the LPM table */
3323         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
3324
3325                 /* skip unused ports */
3326                 if ((1 << ipv4_l3fwd_route_array[i].if_out &
3327                                 enabled_port_mask) == 0)
3328                         continue;
3329
3330                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
3331                         ipv4_l3fwd_route_array[i].ip,
3332                         ipv4_l3fwd_route_array[i].depth,
3333                         ipv4_l3fwd_route_array[i].if_out);
3334
3335                 if (ret < 0) {
3336                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3337                                 "l3fwd LPM table on socket %d\n",
3338                                 i, socketid);
3339                 }
3340
3341                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
3342                         (unsigned)ipv4_l3fwd_route_array[i].ip,
3343                         ipv4_l3fwd_route_array[i].depth,
3344                         ipv4_l3fwd_route_array[i].if_out);
3345         }
3346
3347         /* create the LPM6 table */
3348         snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
3349
3350         config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
3351         config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
3352         config.flags = 0;
3353         ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
3354                                 &config);
3355         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3356                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3357                                 " on socket %d\n", socketid);
3358
3359         /* populate the LPM table */
3360         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
3361
3362                 /* skip unused ports */
3363                 if ((1 << ipv6_l3fwd_route_array[i].if_out &
3364                                 enabled_port_mask) == 0)
3365                         continue;
3366
3367                 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
3368                         ipv6_l3fwd_route_array[i].ip,
3369                         ipv6_l3fwd_route_array[i].depth,
3370                         ipv6_l3fwd_route_array[i].if_out);
3371
3372                 if (ret < 0) {
3373                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3374                                 "l3fwd LPM table on socket %d\n",
3375                                 i, socketid);
3376                 }
3377
3378                 printf("LPM: Adding route %s / %d (%d)\n",
3379                         "IPV6",
3380                         ipv6_l3fwd_route_array[i].depth,
3381                         ipv6_l3fwd_route_array[i].if_out);
3382         }
3383 }
3384 #endif
3385
3386 static int
3387 init_mem(unsigned nb_mbuf)
3388 {
3389         struct lcore_conf *qconf;
3390         int socketid;
3391         unsigned lcore_id;
3392         char s[64];
3393
3394         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3395                 if (rte_lcore_is_enabled(lcore_id) == 0)
3396                         continue;
3397
3398                 if (numa_on)
3399                         socketid = rte_lcore_to_socket_id(lcore_id);
3400                 else
3401                         socketid = 0;
3402
3403                 if (socketid >= NB_SOCKETS) {
3404                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
3405                                 socketid, lcore_id, NB_SOCKETS);
3406                 }
3407                 if (pktmbuf_pool[socketid] == NULL) {
3408                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
3409                         pktmbuf_pool[socketid] =
3410                                 rte_pktmbuf_pool_create(s, nb_mbuf,
3411                                         MEMPOOL_CACHE_SIZE, 0,
3412                                         RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
3413                         if (pktmbuf_pool[socketid] == NULL)
3414                                 rte_exit(EXIT_FAILURE,
3415                                                 "Cannot init mbuf pool on socket %d\n", socketid);
3416                         else
3417                                 printf("Allocated mbuf pool on socket %d\n", socketid);
3418
3419 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3420                         setup_lpm(socketid);
3421 #else
3422                         setup_hash(socketid);
3423 #endif
3424                 }
3425                 qconf = &lcore_conf[lcore_id];
3426                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
3427                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
3428         }
3429         return 0;
3430 }
3431
3432 /* Check the link status of all ports in up to 9s, and print them finally */
3433 static void
3434 check_all_ports_link_status(uint32_t port_mask)
3435 {
3436 #define CHECK_INTERVAL 100 /* 100ms */
3437 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
3438         uint16_t portid;
3439         uint8_t count, all_ports_up, print_flag = 0;
3440         struct rte_eth_link link;
3441         int ret;
3442         char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
3443
3444         printf("\nChecking link status");
3445         fflush(stdout);
3446         for (count = 0; count <= MAX_CHECK_TIME; count++) {
3447                 all_ports_up = 1;
3448                 RTE_ETH_FOREACH_DEV(portid) {
3449                         if ((port_mask & (1 << portid)) == 0)
3450                                 continue;
3451                         memset(&link, 0, sizeof(link));
3452                         ret = rte_eth_link_get_nowait(portid, &link);
3453                         if (ret < 0) {
3454                                 all_ports_up = 0;
3455                                 if (print_flag == 1)
3456                                         printf("Port %u link get failed: %s\n",
3457                                                 portid, rte_strerror(-ret));
3458                                 continue;
3459                         }
3460                         /* print link status if flag set */
3461                         if (print_flag == 1) {
3462                                 rte_eth_link_to_str(link_status_text,
3463                                         sizeof(link_status_text), &link);
3464                                 printf("Port %d %s\n", portid,
3465                                         link_status_text);
3466                                 continue;
3467                         }
3468                         /* clear all_ports_up flag if any link down */
3469                         if (link.link_status == ETH_LINK_DOWN) {
3470                                 all_ports_up = 0;
3471                                 break;
3472                         }
3473                 }
3474                 /* after finally printing all link status, get out */
3475                 if (print_flag == 1)
3476                         break;
3477
3478                 if (all_ports_up == 0) {
3479                         printf(".");
3480                         fflush(stdout);
3481                         rte_delay_ms(CHECK_INTERVAL);
3482                 }
3483
3484                 /* set the print_flag if all ports up or timeout */
3485                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
3486                         print_flag = 1;
3487                         printf("done\n");
3488                 }
3489         }
3490 }
3491
3492 int
3493 main(int argc, char **argv)
3494 {
3495         struct rte_eth_dev_info dev_info;
3496         struct rte_eth_txconf *txconf;
3497         int ret;
3498         int i;
3499         unsigned nb_ports;
3500         uint16_t queueid, portid;
3501         unsigned lcore_id;
3502         uint32_t n_tx_queue, nb_lcores;
3503         uint8_t nb_rx_queue, queue, socketid;
3504
3505         /* init EAL */
3506         ret = rte_eal_init(argc, argv);
3507         if (ret < 0)
3508                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
3509         argc -= ret;
3510         argv += ret;
3511
3512         ret = rte_timer_subsystem_init();
3513         if (ret < 0)
3514                 rte_exit(EXIT_FAILURE, "Failed to initialize timer subystem\n");
3515
3516         /* pre-init dst MACs for all ports to 02:00:00:00:00:xx */
3517         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
3518                 dest_eth_addr[portid] = RTE_ETHER_LOCAL_ADMIN_ADDR +
3519                                 ((uint64_t)portid << 40);
3520                 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
3521         }
3522
3523         /* parse application arguments (after the EAL ones) */
3524         ret = parse_args(argc, argv);
3525         if (ret < 0)
3526                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
3527
3528         if (check_lcore_params() < 0)
3529                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
3530
3531         printf("Initializing rx-queues...\n");
3532         ret = init_rx_queues();
3533         if (ret < 0)
3534                 rte_exit(EXIT_FAILURE, "init_rx_queues failed\n");
3535
3536         printf("Initializing tx-threads...\n");
3537         ret = init_tx_threads();
3538         if (ret < 0)
3539                 rte_exit(EXIT_FAILURE, "init_tx_threads failed\n");
3540
3541         printf("Initializing rings...\n");
3542         ret = init_rx_rings();
3543         if (ret < 0)
3544                 rte_exit(EXIT_FAILURE, "init_rx_rings failed\n");
3545
3546         nb_ports = rte_eth_dev_count_avail();
3547
3548         if (check_port_config() < 0)
3549                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
3550
3551         nb_lcores = rte_lcore_count();
3552
3553         /* initialize all ports */
3554         RTE_ETH_FOREACH_DEV(portid) {
3555                 struct rte_eth_conf local_port_conf = port_conf;
3556
3557                 /* skip ports that are not enabled */
3558                 if ((enabled_port_mask & (1 << portid)) == 0) {
3559                         printf("\nSkipping disabled port %d\n", portid);
3560                         continue;
3561                 }
3562
3563                 /* init port */
3564                 printf("Initializing port %d ... ", portid);
3565                 fflush(stdout);
3566
3567                 nb_rx_queue = get_port_n_rx_queues(portid);
3568                 n_tx_queue = nb_lcores;
3569                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
3570                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
3571                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
3572                         nb_rx_queue, (unsigned)n_tx_queue);
3573
3574                 ret = rte_eth_dev_info_get(portid, &dev_info);
3575                 if (ret != 0)
3576                         rte_exit(EXIT_FAILURE,
3577                                 "Error during getting device (port %u) info: %s\n",
3578                                 portid, strerror(-ret));
3579
3580                 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
3581                         local_port_conf.txmode.offloads |=
3582                                 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
3583
3584                 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
3585                         dev_info.flow_type_rss_offloads;
3586                 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
3587                                 port_conf.rx_adv_conf.rss_conf.rss_hf) {
3588                         printf("Port %u modified RSS hash function based on hardware support,"
3589                                 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
3590                                 portid,
3591                                 port_conf.rx_adv_conf.rss_conf.rss_hf,
3592                                 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
3593                 }
3594
3595                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
3596                                         (uint16_t)n_tx_queue, &local_port_conf);
3597                 if (ret < 0)
3598                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
3599                                 ret, portid);
3600
3601                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
3602                                                        &nb_txd);
3603                 if (ret < 0)
3604                         rte_exit(EXIT_FAILURE,
3605                                  "rte_eth_dev_adjust_nb_rx_tx_desc: err=%d, port=%d\n",
3606                                  ret, portid);
3607
3608                 ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
3609                 if (ret < 0)
3610                         rte_exit(EXIT_FAILURE,
3611                                  "rte_eth_macaddr_get: err=%d, port=%d\n",
3612                                  ret, portid);
3613
3614                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
3615                 printf(", ");
3616                 print_ethaddr("Destination:",
3617                         (const struct rte_ether_addr *)&dest_eth_addr[portid]);
3618                 printf(", ");
3619
3620                 /*
3621                  * prepare src MACs for each port.
3622                  */
3623                 rte_ether_addr_copy(&ports_eth_addr[portid],
3624                         (struct rte_ether_addr *)(val_eth + portid) + 1);
3625
3626                 /* init memory */
3627                 ret = init_mem(NB_MBUF);
3628                 if (ret < 0)
3629                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
3630
3631                 /* init one TX queue per couple (lcore,port) */
3632                 queueid = 0;
3633                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3634                         if (rte_lcore_is_enabled(lcore_id) == 0)
3635                                 continue;
3636
3637                         if (numa_on)
3638                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3639                         else
3640                                 socketid = 0;
3641
3642                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
3643                         fflush(stdout);
3644
3645                         txconf = &dev_info.default_txconf;
3646                         txconf->offloads = local_port_conf.txmode.offloads;
3647                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
3648                                                      socketid, txconf);
3649                         if (ret < 0)
3650                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
3651                                         "port=%d\n", ret, portid);
3652
3653                         tx_thread[lcore_id].tx_queue_id[portid] = queueid;
3654                         queueid++;
3655                 }
3656                 printf("\n");
3657         }
3658
3659         for (i = 0; i < n_rx_thread; i++) {
3660                 lcore_id = rx_thread[i].conf.lcore_id;
3661
3662                 if (rte_lcore_is_enabled(lcore_id) == 0) {
3663                         rte_exit(EXIT_FAILURE,
3664                                         "Cannot start Rx thread on lcore %u: lcore disabled\n",
3665                                         lcore_id
3666                                 );
3667                 }
3668
3669                 printf("\nInitializing rx queues for Rx thread %d on lcore %u ... ",
3670                                 i, lcore_id);
3671                 fflush(stdout);
3672
3673                 /* init RX queues */
3674                 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3675                         struct rte_eth_rxconf rxq_conf;
3676
3677                         portid = rx_thread[i].rx_queue_list[queue].port_id;
3678                         queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3679
3680                         if (numa_on)
3681                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3682                         else
3683                                 socketid = 0;
3684
3685                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
3686                         fflush(stdout);
3687
3688                         ret = rte_eth_dev_info_get(portid, &dev_info);
3689                         if (ret != 0)
3690                                 rte_exit(EXIT_FAILURE,
3691                                         "Error during getting device (port %u) info: %s\n",
3692                                         portid, strerror(-ret));
3693
3694                         rxq_conf = dev_info.default_rxconf;
3695                         rxq_conf.offloads = port_conf.rxmode.offloads;
3696                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
3697                                         socketid,
3698                                         &rxq_conf,
3699                                         pktmbuf_pool[socketid]);
3700                         if (ret < 0)
3701                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, "
3702                                                 "port=%d\n", ret, portid);
3703                 }
3704         }
3705
3706         printf("\n");
3707
3708         /* start ports */
3709         RTE_ETH_FOREACH_DEV(portid) {
3710                 if ((enabled_port_mask & (1 << portid)) == 0)
3711                         continue;
3712
3713                 /* Start device */
3714                 ret = rte_eth_dev_start(portid);
3715                 if (ret < 0)
3716                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
3717                                 ret, portid);
3718
3719                 /*
3720                  * If enabled, put device in promiscuous mode.
3721                  * This allows IO forwarding mode to forward packets
3722                  * to itself through 2 cross-connected  ports of the
3723                  * target machine.
3724                  */
3725                 if (promiscuous_on) {
3726                         ret = rte_eth_promiscuous_enable(portid);
3727                         if (ret != 0)
3728                                 rte_exit(EXIT_FAILURE,
3729                                         "rte_eth_promiscuous_enable: err=%s, port=%u\n",
3730                                         rte_strerror(-ret), portid);
3731                 }
3732         }
3733
3734         for (i = 0; i < n_rx_thread; i++) {
3735                 lcore_id = rx_thread[i].conf.lcore_id;
3736                 if (rte_lcore_is_enabled(lcore_id) == 0)
3737                         continue;
3738
3739                 /* check if hw packet type is supported */
3740                 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3741                         portid = rx_thread[i].rx_queue_list[queue].port_id;
3742                         queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3743
3744                         if (parse_ptype_on) {
3745                                 if (!rte_eth_add_rx_callback(portid, queueid,
3746                                                 cb_parse_ptype, NULL))
3747                                         rte_exit(EXIT_FAILURE,
3748                                                 "Failed to add rx callback: "
3749                                                 "port=%d\n", portid);
3750                         } else if (!check_ptype(portid))
3751                                 rte_exit(EXIT_FAILURE,
3752                                         "Port %d cannot parse packet type.\n\n"
3753                                         "Please add --parse-ptype to use sw "
3754                                         "packet type analyzer.\n\n",
3755                                         portid);
3756                 }
3757         }
3758
3759         check_all_ports_link_status(enabled_port_mask);
3760
3761         if (lthreads_on) {
3762                 printf("Starting L-Threading Model\n");
3763
3764 #if (APP_CPU_LOAD > 0)
3765                 if (cpu_load_lcore_id > 0)
3766                         /* Use one lcore for cpu load collector */
3767                         nb_lcores--;
3768 #endif
3769
3770                 lthread_num_schedulers_set(nb_lcores);
3771                 rte_eal_mp_remote_launch(sched_spawner, NULL, SKIP_MAIN);
3772                 lthread_main_spawner(NULL);
3773
3774         } else {
3775                 printf("Starting P-Threading Model\n");
3776                 /* launch per-lcore init on every lcore */
3777                 rte_eal_mp_remote_launch(pthread_run, NULL, CALL_MAIN);
3778                 RTE_LCORE_FOREACH_WORKER(lcore_id) {
3779                         if (rte_eal_wait_lcore(lcore_id) < 0)
3780                                 return -1;
3781                 }
3782         }
3783
3784         /* clean up the EAL */
3785         rte_eal_cleanup();
3786
3787         return 0;
3788 }