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

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2016 Intel Corporation
   3  */
   4
   5 #include <stdio.h>
   6 #include <stdlib.h>
   7 #include <stdint.h>
   8 #include <inttypes.h>
   9 #include <sys/types.h>
  10 #include <string.h>
  11 #include <sys/queue.h>
  12 #include <stdarg.h>
  13 #include <errno.h>
  14 #include <getopt.h>
  15
  16 #include <rte_common.h>
  17 #include <rte_vect.h>
  18 #include <rte_byteorder.h>
  19 #include <rte_log.h>
  20 #include <rte_memory.h>
  21 #include <rte_memcpy.h>
  22 #include <rte_eal.h>
  23 #include <rte_launch.h>
  24 #include <rte_atomic.h>
  25 #include <rte_cycles.h>
  26 #include <rte_prefetch.h>
  27 #include <rte_lcore.h>
  28 #include <rte_per_lcore.h>
  29 #include <rte_branch_prediction.h>
  30 #include <rte_interrupts.h>
  31 #include <rte_random.h>
  32 #include <rte_debug.h>
  33 #include <rte_ether.h>
  34 #include <rte_ethdev.h>
  35 #include <rte_ring.h>
  36 #include <rte_mempool.h>
  37 #include <rte_mbuf.h>
  38 #include <rte_ip.h>
  39 #include <rte_tcp.h>
  40 #include <rte_udp.h>
  41 #include <rte_string_fns.h>
  42 #include <rte_pause.h>
  43
  44 #include <cmdline_parse.h>
  45 #include <cmdline_parse_etheraddr.h>
  46
  47 #include <lthread_api.h>
  48
  49 #define APP_LOOKUP_EXACT_MATCH          0
  50 #define APP_LOOKUP_LPM                  1
  51 #define DO_RFC_1812_CHECKS
  52
  53 /* Enable cpu-load stats 0-off, 1-on */
  54 #define APP_CPU_LOAD                 1
  55
  56 #ifndef APP_LOOKUP_METHOD
  57 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  58 #endif
  59
  60 #ifndef __GLIBC__ /* sched_getcpu() is glibc specific */
  61 #define sched_getcpu() rte_lcore_id()
  62 #endif
  63
  64 static int
  65 check_ptype(int portid)
  66 {
  67         int i, ret;
  68         int ipv4 = 0, ipv6 = 0;
  69
  70         ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK, NULL,
  71                         0);
  72         if (ret <= 0)
  73                 return 0;
  74
  75         uint32_t ptypes[ret];
  76
  77         ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK,
  78                         ptypes, ret);
  79         for (i = 0; i < ret; ++i) {
  80                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
  81                         ipv4 = 1;
  82                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
  83                         ipv6 = 1;
  84         }
  85
  86         if (ipv4 && ipv6)
  87                 return 1;
  88
  89         return 0;
  90 }
  91
  92 static inline void
  93 parse_ptype(struct rte_mbuf *m)
  94 {
  95         struct ether_hdr *eth_hdr;
  96         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
  97         uint16_t ether_type;
  98
  99         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 100         ether_type = eth_hdr->ether_type;
 101         if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
 102                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 103         else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
 104                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 105
 106         m->packet_type = packet_type;
 107 }
 108
 109 static uint16_t
 110 cb_parse_ptype(__rte_unused uint16_t port, __rte_unused uint16_t queue,
 111                 struct rte_mbuf *pkts[], uint16_t nb_pkts,
 112                 __rte_unused uint16_t max_pkts, __rte_unused void *user_param)
 113 {
 114         unsigned int i;
 115
 116         for (i = 0; i < nb_pkts; i++)
 117                 parse_ptype(pkts[i]);
 118
 119         return nb_pkts;
 120 }
 121
 122 /*
 123  *  When set to zero, simple forwaring path is eanbled.
 124  *  When set to one, optimized forwarding path is enabled.
 125  *  Note that LPM optimisation path uses SSE4.1 instructions.
 126  */
 127 #define ENABLE_MULTI_BUFFER_OPTIMIZE    1
 128
 129 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 130 #include <rte_hash.h>
 131 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 132 #include <rte_lpm.h>
 133 #include <rte_lpm6.h>
 134 #else
 135 #error "APP_LOOKUP_METHOD set to incorrect value"
 136 #endif
 137
 138 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
 139
 140 #define MAX_JUMBO_PKT_LEN  9600
 141
 142 #define IPV6_ADDR_LEN 16
 143
 144 #define MEMPOOL_CACHE_SIZE 256
 145
 146 /*
 147  * This expression is used to calculate the number of mbufs needed depending on
 148  * user input, taking into account memory for rx and tx hardware rings, cache
 149  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 150  * NB_MBUF never goes below a minimum value of 8192
 151  */
 152
 153 #define NB_MBUF RTE_MAX(\
 154                 (nb_ports*nb_rx_queue*nb_rxd +      \
 155                 nb_ports*nb_lcores*MAX_PKT_BURST +  \
 156                 nb_ports*n_tx_queue*nb_txd +        \
 157                 nb_lcores*MEMPOOL_CACHE_SIZE),      \
 158                 (unsigned)8192)
 159
 160 #define MAX_PKT_BURST     32
 161 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 162
 163 /*
 164  * Try to avoid TX buffering if we have at least MAX_TX_BURST packets to send.
 165  */
 166 #define MAX_TX_BURST  (MAX_PKT_BURST / 2)
 167 #define BURST_SIZE    MAX_TX_BURST
 168
 169 #define NB_SOCKETS 8
 170
 171 /* Configure how many packets ahead to prefetch, when reading packets */
 172 #define PREFETCH_OFFSET 3
 173
 174 /* Used to mark destination port as 'invalid'. */
 175 #define BAD_PORT        ((uint16_t)-1)
 176
 177 #define FWDSTEP 4
 178
 179 /*
 180  * Configurable number of RX/TX ring descriptors
 181  */
 182 #define RTE_TEST_RX_DESC_DEFAULT 1024
 183 #define RTE_TEST_TX_DESC_DEFAULT 1024
 184 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 185 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 186
 187 /* ethernet addresses of ports */
 188 static uint64_t dest_eth_addr[RTE_MAX_ETHPORTS];
 189 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 190
 191 static xmm_t val_eth[RTE_MAX_ETHPORTS];
 192
 193 /* replace first 12B of the ethernet header. */
 194 #define MASK_ETH 0x3f
 195
 196 /* mask of enabled ports */
 197 static uint32_t enabled_port_mask;
 198 static int promiscuous_on; /**< Set in promiscuous mode off by default. */
 199 static int numa_on = 1;    /**< NUMA is enabled by default. */
 200 static int parse_ptype_on;
 201
 202 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 203 static int ipv6;           /**< ipv6 is false by default. */
 204 #endif
 205
 206 #if (APP_CPU_LOAD == 1)
 207
 208 #define MAX_CPU RTE_MAX_LCORE
 209 #define CPU_LOAD_TIMEOUT_US (5 * 1000 * 1000)  /**< Timeout for collecting 5s */
 210
 211 #define CPU_PROCESS     0
 212 #define CPU_POLL        1
 213 #define MAX_CPU_COUNTER 2
 214
 215 struct cpu_load {
 216         uint16_t       n_cpu;
 217         uint64_t       counter;
 218         uint64_t       hits[MAX_CPU_COUNTER][MAX_CPU];
 219 } __rte_cache_aligned;
 220
 221 static struct cpu_load cpu_load;
 222 static int cpu_load_lcore_id = -1;
 223
 224 #define SET_CPU_BUSY(thread, counter) \
 225                 thread->conf.busy[counter] = 1
 226
 227 #define SET_CPU_IDLE(thread, counter) \
 228                 thread->conf.busy[counter] = 0
 229
 230 #define IS_CPU_BUSY(thread, counter) \
 231                 (thread->conf.busy[counter] > 0)
 232
 233 #else
 234
 235 #define SET_CPU_BUSY(thread, counter)
 236 #define SET_CPU_IDLE(thread, counter)
 237 #define IS_CPU_BUSY(thread, counter) 0
 238
 239 #endif
 240
 241 struct mbuf_table {
 242         uint16_t len;
 243         struct rte_mbuf *m_table[MAX_PKT_BURST];
 244 };
 245
 246 struct lcore_rx_queue {
 247         uint16_t port_id;
 248         uint8_t queue_id;
 249 } __rte_cache_aligned;
 250
 251 #define MAX_RX_QUEUE_PER_LCORE 16
 252 #define MAX_TX_QUEUE_PER_PORT  RTE_MAX_ETHPORTS
 253 #define MAX_RX_QUEUE_PER_PORT  128
 254
 255 #define MAX_LCORE_PARAMS       1024
 256 struct rx_thread_params {
 257         uint16_t port_id;
 258         uint8_t queue_id;
 259         uint8_t lcore_id;
 260         uint8_t thread_id;
 261 } __rte_cache_aligned;
 262
 263 static struct rx_thread_params rx_thread_params_array[MAX_LCORE_PARAMS];
 264 static struct rx_thread_params rx_thread_params_array_default[] = {
 265         {0, 0, 2, 0},
 266         {0, 1, 2, 1},
 267         {0, 2, 2, 2},
 268         {1, 0, 2, 3},
 269         {1, 1, 2, 4},
 270         {1, 2, 2, 5},
 271         {2, 0, 2, 6},
 272         {3, 0, 3, 7},
 273         {3, 1, 3, 8},
 274 };
 275
 276 static struct rx_thread_params *rx_thread_params =
 277                 rx_thread_params_array_default;
 278 static uint16_t nb_rx_thread_params = RTE_DIM(rx_thread_params_array_default);
 279
 280 struct tx_thread_params {
 281         uint8_t lcore_id;
 282         uint8_t thread_id;
 283 } __rte_cache_aligned;
 284
 285 static struct tx_thread_params tx_thread_params_array[MAX_LCORE_PARAMS];
 286 static struct tx_thread_params tx_thread_params_array_default[] = {
 287         {4, 0},
 288         {5, 1},
 289         {6, 2},
 290         {7, 3},
 291         {8, 4},
 292         {9, 5},
 293         {10, 6},
 294         {11, 7},
 295         {12, 8},
 296 };
 297
 298 static struct tx_thread_params *tx_thread_params =
 299                 tx_thread_params_array_default;
 300 static uint16_t nb_tx_thread_params = RTE_DIM(tx_thread_params_array_default);
 301
 302 static struct rte_eth_conf port_conf = {
 303         .rxmode = {
 304                 .mq_mode = ETH_MQ_RX_RSS,
 305                 .max_rx_pkt_len = ETHER_MAX_LEN,
 306                 .split_hdr_size = 0,
 307                 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
 308         },
 309         .rx_adv_conf = {
 310                 .rss_conf = {
 311                         .rss_key = NULL,
 312                         .rss_hf = ETH_RSS_TCP,
 313                 },
 314         },
 315         .txmode = {
 316                 .mq_mode = ETH_MQ_TX_NONE,
 317         },
 318 };
 319
 320 static struct rte_mempool *pktmbuf_pool[NB_SOCKETS];
 321
 322 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 323
 324 #include <rte_hash_crc.h>
 325 #define DEFAULT_HASH_FUNC       rte_hash_crc
 326
 327 struct ipv4_5tuple {
 328         uint32_t ip_dst;
 329         uint32_t ip_src;
 330         uint16_t port_dst;
 331         uint16_t port_src;
 332         uint8_t  proto;
 333 } __attribute__((__packed__));
 334
 335 union ipv4_5tuple_host {
 336         struct {
 337                 uint8_t  pad0;
 338                 uint8_t  proto;
 339                 uint16_t pad1;
 340                 uint32_t ip_src;
 341                 uint32_t ip_dst;
 342                 uint16_t port_src;
 343                 uint16_t port_dst;
 344         };
 345         __m128i xmm;
 346 };
 347
 348 #define XMM_NUM_IN_IPV6_5TUPLE 3
 349
 350 struct ipv6_5tuple {
 351         uint8_t  ip_dst[IPV6_ADDR_LEN];
 352         uint8_t  ip_src[IPV6_ADDR_LEN];
 353         uint16_t port_dst;
 354         uint16_t port_src;
 355         uint8_t  proto;
 356 } __attribute__((__packed__));
 357
 358 union ipv6_5tuple_host {
 359         struct {
 360                 uint16_t pad0;
 361                 uint8_t  proto;
 362                 uint8_t  pad1;
 363                 uint8_t  ip_src[IPV6_ADDR_LEN];
 364                 uint8_t  ip_dst[IPV6_ADDR_LEN];
 365                 uint16_t port_src;
 366                 uint16_t port_dst;
 367                 uint64_t reserve;
 368         };
 369         __m128i xmm[XMM_NUM_IN_IPV6_5TUPLE];
 370 };
 371
 372 struct ipv4_l3fwd_route {
 373         struct ipv4_5tuple key;
 374         uint8_t if_out;
 375 };
 376
 377 struct ipv6_l3fwd_route {
 378         struct ipv6_5tuple key;
 379         uint8_t if_out;
 380 };
 381
 382 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 383         {{IPv4(101, 0, 0, 0), IPv4(100, 10, 0, 1),  101, 11, IPPROTO_TCP}, 0},
 384         {{IPv4(201, 0, 0, 0), IPv4(200, 20, 0, 1),  102, 12, IPPROTO_TCP}, 1},
 385         {{IPv4(111, 0, 0, 0), IPv4(100, 30, 0, 1),  101, 11, IPPROTO_TCP}, 2},
 386         {{IPv4(211, 0, 0, 0), IPv4(200, 40, 0, 1),  102, 12, IPPROTO_TCP}, 3},
 387 };
 388
 389 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 390         {{
 391         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 392         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 393                         0x05},
 394         101, 11, IPPROTO_TCP}, 0},
 395
 396         {{
 397         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 398         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 399                         0x05},
 400         102, 12, IPPROTO_TCP}, 1},
 401
 402         {{
 403         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 404         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 405                         0x05},
 406         101, 11, IPPROTO_TCP}, 2},
 407
 408         {{
 409         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 410         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 411                         0x05},
 412         102, 12, IPPROTO_TCP}, 3},
 413 };
 414
 415 typedef struct rte_hash lookup_struct_t;
 416 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 417 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 418
 419 #ifdef RTE_ARCH_X86_64
 420 /* default to 4 million hash entries (approx) */
 421 #define L3FWD_HASH_ENTRIES (1024*1024*4)
 422 #else
 423 /* 32-bit has less address-space for hugepage memory, limit to 1M entries */
 424 #define L3FWD_HASH_ENTRIES (1024*1024*1)
 425 #endif
 426 #define HASH_ENTRY_NUMBER_DEFAULT 4
 427
 428 static uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
 429
 430 static inline uint32_t
 431 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
 432                 uint32_t init_val)
 433 {
 434         const union ipv4_5tuple_host *k;
 435         uint32_t t;
 436         const uint32_t *p;
 437
 438         k = data;
 439         t = k->proto;
 440         p = (const uint32_t *)&k->port_src;
 441
 442         init_val = rte_hash_crc_4byte(t, init_val);
 443         init_val = rte_hash_crc_4byte(k->ip_src, init_val);
 444         init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
 445         init_val = rte_hash_crc_4byte(*p, init_val);
 446         return init_val;
 447 }
 448
 449 static inline uint32_t
 450 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len,
 451                 uint32_t init_val)
 452 {
 453         const union ipv6_5tuple_host *k;
 454         uint32_t t;
 455         const uint32_t *p;
 456         const uint32_t *ip_src0, *ip_src1, *ip_src2, *ip_src3;
 457         const uint32_t *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
 458
 459         k = data;
 460         t = k->proto;
 461         p = (const uint32_t *)&k->port_src;
 462
 463         ip_src0 = (const uint32_t *) k->ip_src;
 464         ip_src1 = (const uint32_t *)(k->ip_src + 4);
 465         ip_src2 = (const uint32_t *)(k->ip_src + 8);
 466         ip_src3 = (const uint32_t *)(k->ip_src + 12);
 467         ip_dst0 = (const uint32_t *) k->ip_dst;
 468         ip_dst1 = (const uint32_t *)(k->ip_dst + 4);
 469         ip_dst2 = (const uint32_t *)(k->ip_dst + 8);
 470         ip_dst3 = (const uint32_t *)(k->ip_dst + 12);
 471         init_val = rte_hash_crc_4byte(t, init_val);
 472         init_val = rte_hash_crc_4byte(*ip_src0, init_val);
 473         init_val = rte_hash_crc_4byte(*ip_src1, init_val);
 474         init_val = rte_hash_crc_4byte(*ip_src2, init_val);
 475         init_val = rte_hash_crc_4byte(*ip_src3, init_val);
 476         init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
 477         init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
 478         init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
 479         init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
 480         init_val = rte_hash_crc_4byte(*p, init_val);
 481         return init_val;
 482 }
 483
 484 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
 485 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
 486
 487 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 488 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 489
 490 #endif
 491
 492 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 493 struct ipv4_l3fwd_route {
 494         uint32_t ip;
 495         uint8_t  depth;
 496         uint8_t  if_out;
 497 };
 498
 499 struct ipv6_l3fwd_route {
 500         uint8_t ip[16];
 501         uint8_t depth;
 502         uint8_t if_out;
 503 };
 504
 505 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 506         {IPv4(1, 1, 1, 0), 24, 0},
 507         {IPv4(2, 1, 1, 0), 24, 1},
 508         {IPv4(3, 1, 1, 0), 24, 2},
 509         {IPv4(4, 1, 1, 0), 24, 3},
 510         {IPv4(5, 1, 1, 0), 24, 4},
 511         {IPv4(6, 1, 1, 0), 24, 5},
 512         {IPv4(7, 1, 1, 0), 24, 6},
 513         {IPv4(8, 1, 1, 0), 24, 7},
 514 };
 515
 516 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 517         {{1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 0},
 518         {{2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 1},
 519         {{3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 2},
 520         {{4, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 3},
 521         {{5, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 4},
 522         {{6, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 5},
 523         {{7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 6},
 524         {{8, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 7},
 525 };
 526
 527 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
 528 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
 529
 530 #define IPV4_L3FWD_LPM_MAX_RULES         1024
 531 #define IPV6_L3FWD_LPM_MAX_RULES         1024
 532 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
 533
 534 typedef struct rte_lpm lookup_struct_t;
 535 typedef struct rte_lpm6 lookup6_struct_t;
 536 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 537 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 538 #endif
 539
 540 struct lcore_conf {
 541         lookup_struct_t *ipv4_lookup_struct;
 542 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 543         lookup6_struct_t *ipv6_lookup_struct;
 544 #else
 545         lookup_struct_t *ipv6_lookup_struct;
 546 #endif
 547         void *data;
 548 } __rte_cache_aligned;
 549
 550 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 551 RTE_DEFINE_PER_LCORE(struct lcore_conf *, lcore_conf);
 552
 553 #define MAX_RX_QUEUE_PER_THREAD 16
 554 #define MAX_TX_PORT_PER_THREAD  RTE_MAX_ETHPORTS
 555 #define MAX_TX_QUEUE_PER_PORT   RTE_MAX_ETHPORTS
 556 #define MAX_RX_QUEUE_PER_PORT   128
 557
 558 #define MAX_RX_THREAD 1024
 559 #define MAX_TX_THREAD 1024
 560 #define MAX_THREAD    (MAX_RX_THREAD + MAX_TX_THREAD)
 561
 562 /**
 563  * Producers and consumers threads configuration
 564  */
 565 static int lthreads_on = 1; /**< Use lthreads for processing*/
 566
 567 rte_atomic16_t rx_counter;  /**< Number of spawned rx threads */
 568 rte_atomic16_t tx_counter;  /**< Number of spawned tx threads */
 569
 570 struct thread_conf {
 571         uint16_t lcore_id;      /**< Initial lcore for rx thread */
 572         uint16_t cpu_id;        /**< Cpu id for cpu load stats counter */
 573         uint16_t thread_id;     /**< Thread ID */
 574
 575 #if (APP_CPU_LOAD > 0)
 576         int busy[MAX_CPU_COUNTER];
 577 #endif
 578 };
 579
 580 struct thread_rx_conf {
 581         struct thread_conf conf;
 582
 583         uint16_t n_rx_queue;
 584         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 585
 586         uint16_t n_ring;        /**< Number of output rings */
 587         struct rte_ring *ring[RTE_MAX_LCORE];
 588         struct lthread_cond *ready[RTE_MAX_LCORE];
 589
 590 #if (APP_CPU_LOAD > 0)
 591         int busy[MAX_CPU_COUNTER];
 592 #endif
 593 } __rte_cache_aligned;
 594
 595 uint16_t n_rx_thread;
 596 struct thread_rx_conf rx_thread[MAX_RX_THREAD];
 597
 598 struct thread_tx_conf {
 599         struct thread_conf conf;
 600
 601         uint16_t tx_queue_id[RTE_MAX_LCORE];
 602         struct mbuf_table tx_mbufs[RTE_MAX_LCORE];
 603
 604         struct rte_ring *ring;
 605         struct lthread_cond **ready;
 606
 607 } __rte_cache_aligned;
 608
 609 uint16_t n_tx_thread;
 610 struct thread_tx_conf tx_thread[MAX_TX_THREAD];
 611
 612 /* Send burst of packets on an output interface */
 613 static inline int
 614 send_burst(struct thread_tx_conf *qconf, uint16_t n, uint16_t port)
 615 {
 616         struct rte_mbuf **m_table;
 617         int ret;
 618         uint16_t queueid;
 619
 620         queueid = qconf->tx_queue_id[port];
 621         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 622
 623         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 624         if (unlikely(ret < n)) {
 625                 do {
 626                         rte_pktmbuf_free(m_table[ret]);
 627                 } while (++ret < n);
 628         }
 629
 630         return 0;
 631 }
 632
 633 /* Enqueue a single packet, and send burst if queue is filled */
 634 static inline int
 635 send_single_packet(struct rte_mbuf *m, uint16_t port)
 636 {
 637         uint16_t len;
 638         struct thread_tx_conf *qconf;
 639
 640         if (lthreads_on)
 641                 qconf = (struct thread_tx_conf *)lthread_get_data();
 642         else
 643                 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
 644
 645         len = qconf->tx_mbufs[port].len;
 646         qconf->tx_mbufs[port].m_table[len] = m;
 647         len++;
 648
 649         /* enough pkts to be sent */
 650         if (unlikely(len == MAX_PKT_BURST)) {
 651                 send_burst(qconf, MAX_PKT_BURST, port);
 652                 len = 0;
 653         }
 654
 655         qconf->tx_mbufs[port].len = len;
 656         return 0;
 657 }
 658
 659 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
 660         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
 661 static __rte_always_inline void
 662 send_packetsx4(uint16_t port,
 663         struct rte_mbuf *m[], uint32_t num)
 664 {
 665         uint32_t len, j, n;
 666         struct thread_tx_conf *qconf;
 667
 668         if (lthreads_on)
 669                 qconf = (struct thread_tx_conf *)lthread_get_data();
 670         else
 671                 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
 672
 673         len = qconf->tx_mbufs[port].len;
 674
 675         /*
 676          * If TX buffer for that queue is empty, and we have enough packets,
 677          * then send them straightway.
 678          */
 679         if (num >= MAX_TX_BURST && len == 0) {
 680                 n = rte_eth_tx_burst(port, qconf->tx_queue_id[port], m, num);
 681                 if (unlikely(n < num)) {
 682                         do {
 683                                 rte_pktmbuf_free(m[n]);
 684                         } while (++n < num);
 685                 }
 686                 return;
 687         }
 688
 689         /*
 690          * Put packets into TX buffer for that queue.
 691          */
 692
 693         n = len + num;
 694         n = (n > MAX_PKT_BURST) ? MAX_PKT_BURST - len : num;
 695
 696         j = 0;
 697         switch (n % FWDSTEP) {
 698         while (j < n) {
 699         case 0:
 700                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 701                 j++;
 702                 /* fall-through */
 703         case 3:
 704                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 705                 j++;
 706                 /* fall-through */
 707         case 2:
 708                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 709                 j++;
 710                 /* fall-through */
 711         case 1:
 712                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 713                 j++;
 714         }
 715         }
 716
 717         len += n;
 718
 719         /* enough pkts to be sent */
 720         if (unlikely(len == MAX_PKT_BURST)) {
 721
 722                 send_burst(qconf, MAX_PKT_BURST, port);
 723
 724                 /* copy rest of the packets into the TX buffer. */
 725                 len = num - n;
 726                 j = 0;
 727                 switch (len % FWDSTEP) {
 728                 while (j < len) {
 729                 case 0:
 730                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 731                         j++;
 732                         /* fall-through */
 733                 case 3:
 734                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 735                         j++;
 736                         /* fall-through */
 737                 case 2:
 738                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 739                         j++;
 740                         /* fall-through */
 741                 case 1:
 742                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 743                         j++;
 744                 }
 745                 }
 746         }
 747
 748         qconf->tx_mbufs[port].len = len;
 749 }
 750 #endif /* APP_LOOKUP_LPM */
 751
 752 #ifdef DO_RFC_1812_CHECKS
 753 static inline int
 754 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 755 {
 756         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 757         /*
 758          * 1. The packet length reported by the Link Layer must be large
 759          * enough to hold the minimum length legal IP datagram (20 bytes).
 760          */
 761         if (link_len < sizeof(struct ipv4_hdr))
 762                 return -1;
 763
 764         /* 2. The IP checksum must be correct. */
 765         /* this is checked in H/W */
 766
 767         /*
 768          * 3. The IP version number must be 4. If the version number is not 4
 769          * then the packet may be another version of IP, such as IPng or
 770          * ST-II.
 771          */
 772         if (((pkt->version_ihl) >> 4) != 4)
 773                 return -3;
 774         /*
 775          * 4. The IP header length field must be large enough to hold the
 776          * minimum length legal IP datagram (20 bytes = 5 words).
 777          */
 778         if ((pkt->version_ihl & 0xf) < 5)
 779                 return -4;
 780
 781         /*
 782          * 5. The IP total length field must be large enough to hold the IP
 783          * datagram header, whose length is specified in the IP header length
 784          * field.
 785          */
 786         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 787                 return -5;
 788
 789         return 0;
 790 }
 791 #endif
 792
 793 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 794
 795 static __m128i mask0;
 796 static __m128i mask1;
 797 static __m128i mask2;
 798 static inline uint16_t
 799 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid,
 800                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 801 {
 802         int ret = 0;
 803         union ipv4_5tuple_host key;
 804
 805         ipv4_hdr = (uint8_t *)ipv4_hdr + offsetof(struct ipv4_hdr, time_to_live);
 806         __m128i data = _mm_loadu_si128((__m128i *)(ipv4_hdr));
 807         /* Get 5 tuple: dst port, src port, dst IP address, src IP address and
 808            protocol */
 809         key.xmm = _mm_and_si128(data, mask0);
 810         /* Find destination port */
 811         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 812         return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
 813 }
 814
 815 static inline uint16_t
 816 get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid,
 817                 lookup_struct_t *ipv6_l3fwd_lookup_struct)
 818 {
 819         int ret = 0;
 820         union ipv6_5tuple_host key;
 821
 822         ipv6_hdr = (uint8_t *)ipv6_hdr + offsetof(struct ipv6_hdr, payload_len);
 823         __m128i data0 = _mm_loadu_si128((__m128i *)(ipv6_hdr));
 824         __m128i data1 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
 825                         sizeof(__m128i)));
 826         __m128i data2 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
 827                         sizeof(__m128i) + sizeof(__m128i)));
 828         /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
 829         key.xmm[0] = _mm_and_si128(data0, mask1);
 830         /* Get part of 5 tuple: dst IP address lower 96 bits and src IP address
 831            higher 32 bits */
 832         key.xmm[1] = data1;
 833         /* Get part of 5 tuple: dst port and src port and dst IP address higher
 834            32 bits */
 835         key.xmm[2] = _mm_and_si128(data2, mask2);
 836
 837         /* Find destination port */
 838         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 839         return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
 840 }
 841 #endif
 842
 843 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 844
 845 static inline uint16_t
 846 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid,
 847                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 848 {
 849         uint32_t next_hop;
 850
 851         return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 852                 rte_be_to_cpu_32(((struct ipv4_hdr *)ipv4_hdr)->dst_addr),
 853                 &next_hop) == 0) ? next_hop : portid);
 854 }
 855
 856 static inline uint16_t
 857 get_ipv6_dst_port(void *ipv6_hdr,  uint16_t portid,
 858                 lookup6_struct_t *ipv6_l3fwd_lookup_struct)
 859 {
 860         uint32_t next_hop;
 861
 862         return ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
 863                         ((struct ipv6_hdr *)ipv6_hdr)->dst_addr, &next_hop) == 0) ?
 864                         next_hop : portid);
 865 }
 866 #endif
 867
 868 static inline void l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid)
 869                 __attribute__((unused));
 870
 871 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) && \
 872         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
 873
 874 #define MASK_ALL_PKTS   0xff
 875 #define EXCLUDE_1ST_PKT 0xfe
 876 #define EXCLUDE_2ND_PKT 0xfd
 877 #define EXCLUDE_3RD_PKT 0xfb
 878 #define EXCLUDE_4TH_PKT 0xf7
 879 #define EXCLUDE_5TH_PKT 0xef
 880 #define EXCLUDE_6TH_PKT 0xdf
 881 #define EXCLUDE_7TH_PKT 0xbf
 882 #define EXCLUDE_8TH_PKT 0x7f
 883
 884 static inline void
 885 simple_ipv4_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid)
 886 {
 887         struct ether_hdr *eth_hdr[8];
 888         struct ipv4_hdr *ipv4_hdr[8];
 889         uint16_t dst_port[8];
 890         int32_t ret[8];
 891         union ipv4_5tuple_host key[8];
 892         __m128i data[8];
 893
 894         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
 895         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
 896         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
 897         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
 898         eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *);
 899         eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *);
 900         eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *);
 901         eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *);
 902
 903         /* Handle IPv4 headers.*/
 904         ipv4_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv4_hdr *,
 905                         sizeof(struct ether_hdr));
 906         ipv4_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv4_hdr *,
 907                         sizeof(struct ether_hdr));
 908         ipv4_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv4_hdr *,
 909                         sizeof(struct ether_hdr));
 910         ipv4_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv4_hdr *,
 911                         sizeof(struct ether_hdr));
 912         ipv4_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv4_hdr *,
 913                         sizeof(struct ether_hdr));
 914         ipv4_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv4_hdr *,
 915                         sizeof(struct ether_hdr));
 916         ipv4_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv4_hdr *,
 917                         sizeof(struct ether_hdr));
 918         ipv4_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv4_hdr *,
 919                         sizeof(struct ether_hdr));
 920
 921 #ifdef DO_RFC_1812_CHECKS
 922         /* Check to make sure the packet is valid (RFC1812) */
 923         uint8_t valid_mask = MASK_ALL_PKTS;
 924
 925         if (is_valid_ipv4_pkt(ipv4_hdr[0], m[0]->pkt_len) < 0) {
 926                 rte_pktmbuf_free(m[0]);
 927                 valid_mask &= EXCLUDE_1ST_PKT;
 928         }
 929         if (is_valid_ipv4_pkt(ipv4_hdr[1], m[1]->pkt_len) < 0) {
 930                 rte_pktmbuf_free(m[1]);
 931                 valid_mask &= EXCLUDE_2ND_PKT;
 932         }
 933         if (is_valid_ipv4_pkt(ipv4_hdr[2], m[2]->pkt_len) < 0) {
 934                 rte_pktmbuf_free(m[2]);
 935                 valid_mask &= EXCLUDE_3RD_PKT;
 936         }
 937         if (is_valid_ipv4_pkt(ipv4_hdr[3], m[3]->pkt_len) < 0) {
 938                 rte_pktmbuf_free(m[3]);
 939                 valid_mask &= EXCLUDE_4TH_PKT;
 940         }
 941         if (is_valid_ipv4_pkt(ipv4_hdr[4], m[4]->pkt_len) < 0) {
 942                 rte_pktmbuf_free(m[4]);
 943                 valid_mask &= EXCLUDE_5TH_PKT;
 944         }
 945         if (is_valid_ipv4_pkt(ipv4_hdr[5], m[5]->pkt_len) < 0) {
 946                 rte_pktmbuf_free(m[5]);
 947                 valid_mask &= EXCLUDE_6TH_PKT;
 948         }
 949         if (is_valid_ipv4_pkt(ipv4_hdr[6], m[6]->pkt_len) < 0) {
 950                 rte_pktmbuf_free(m[6]);
 951                 valid_mask &= EXCLUDE_7TH_PKT;
 952         }
 953         if (is_valid_ipv4_pkt(ipv4_hdr[7], m[7]->pkt_len) < 0) {
 954                 rte_pktmbuf_free(m[7]);
 955                 valid_mask &= EXCLUDE_8TH_PKT;
 956         }
 957         if (unlikely(valid_mask != MASK_ALL_PKTS)) {
 958                 if (valid_mask == 0)
 959                         return;
 960
 961                 uint8_t i = 0;
 962
 963                 for (i = 0; i < 8; i++)
 964                         if ((0x1 << i) & valid_mask)
 965                                 l3fwd_simple_forward(m[i], portid);
 966         }
 967 #endif /* End of #ifdef DO_RFC_1812_CHECKS */
 968
 969         data[0] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[0], __m128i *,
 970                         sizeof(struct ether_hdr) +
 971                         offsetof(struct ipv4_hdr, time_to_live)));
 972         data[1] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[1], __m128i *,
 973                         sizeof(struct ether_hdr) +
 974                         offsetof(struct ipv4_hdr, time_to_live)));
 975         data[2] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[2], __m128i *,
 976                         sizeof(struct ether_hdr) +
 977                         offsetof(struct ipv4_hdr, time_to_live)));
 978         data[3] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[3], __m128i *,
 979                         sizeof(struct ether_hdr) +
 980                         offsetof(struct ipv4_hdr, time_to_live)));
 981         data[4] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[4], __m128i *,
 982                         sizeof(struct ether_hdr) +
 983                         offsetof(struct ipv4_hdr, time_to_live)));
 984         data[5] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[5], __m128i *,
 985                         sizeof(struct ether_hdr) +
 986                         offsetof(struct ipv4_hdr, time_to_live)));
 987         data[6] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[6], __m128i *,
 988                         sizeof(struct ether_hdr) +
 989                         offsetof(struct ipv4_hdr, time_to_live)));
 990         data[7] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[7], __m128i *,
 991                         sizeof(struct ether_hdr) +
 992                         offsetof(struct ipv4_hdr, time_to_live)));
 993
 994         key[0].xmm = _mm_and_si128(data[0], mask0);
 995         key[1].xmm = _mm_and_si128(data[1], mask0);
 996         key[2].xmm = _mm_and_si128(data[2], mask0);
 997         key[3].xmm = _mm_and_si128(data[3], mask0);
 998         key[4].xmm = _mm_and_si128(data[4], mask0);
 999         key[5].xmm = _mm_and_si128(data[5], mask0);
1000         key[6].xmm = _mm_and_si128(data[6], mask0);
1001         key[7].xmm = _mm_and_si128(data[7], mask0);
1002
1003         const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1004                         &key[4], &key[5], &key[6], &key[7]};
1005
1006         rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct,
1007                         &key_array[0], 8, ret);
1008         dst_port[0] = ((ret[0] < 0) ? portid : ipv4_l3fwd_out_if[ret[0]]);
1009         dst_port[1] = ((ret[1] < 0) ? portid : ipv4_l3fwd_out_if[ret[1]]);
1010         dst_port[2] = ((ret[2] < 0) ? portid : ipv4_l3fwd_out_if[ret[2]]);
1011         dst_port[3] = ((ret[3] < 0) ? portid : ipv4_l3fwd_out_if[ret[3]]);
1012         dst_port[4] = ((ret[4] < 0) ? portid : ipv4_l3fwd_out_if[ret[4]]);
1013         dst_port[5] = ((ret[5] < 0) ? portid : ipv4_l3fwd_out_if[ret[5]]);
1014         dst_port[6] = ((ret[6] < 0) ? portid : ipv4_l3fwd_out_if[ret[6]]);
1015         dst_port[7] = ((ret[7] < 0) ? portid : ipv4_l3fwd_out_if[ret[7]]);
1016
1017         if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1018                         (enabled_port_mask & 1 << dst_port[0]) == 0)
1019                 dst_port[0] = portid;
1020         if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1021                         (enabled_port_mask & 1 << dst_port[1]) == 0)
1022                 dst_port[1] = portid;
1023         if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1024                         (enabled_port_mask & 1 << dst_port[2]) == 0)
1025                 dst_port[2] = portid;
1026         if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1027                         (enabled_port_mask & 1 << dst_port[3]) == 0)
1028                 dst_port[3] = portid;
1029         if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1030                         (enabled_port_mask & 1 << dst_port[4]) == 0)
1031                 dst_port[4] = portid;
1032         if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1033                         (enabled_port_mask & 1 << dst_port[5]) == 0)
1034                 dst_port[5] = portid;
1035         if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1036                         (enabled_port_mask & 1 << dst_port[6]) == 0)
1037                 dst_port[6] = portid;
1038         if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1039                         (enabled_port_mask & 1 << dst_port[7]) == 0)
1040                 dst_port[7] = portid;
1041
1042 #ifdef DO_RFC_1812_CHECKS
1043         /* Update time to live and header checksum */
1044         --(ipv4_hdr[0]->time_to_live);
1045         --(ipv4_hdr[1]->time_to_live);
1046         --(ipv4_hdr[2]->time_to_live);
1047         --(ipv4_hdr[3]->time_to_live);
1048         ++(ipv4_hdr[0]->hdr_checksum);
1049         ++(ipv4_hdr[1]->hdr_checksum);
1050         ++(ipv4_hdr[2]->hdr_checksum);
1051         ++(ipv4_hdr[3]->hdr_checksum);
1052         --(ipv4_hdr[4]->time_to_live);
1053         --(ipv4_hdr[5]->time_to_live);
1054         --(ipv4_hdr[6]->time_to_live);
1055         --(ipv4_hdr[7]->time_to_live);
1056         ++(ipv4_hdr[4]->hdr_checksum);
1057         ++(ipv4_hdr[5]->hdr_checksum);
1058         ++(ipv4_hdr[6]->hdr_checksum);
1059         ++(ipv4_hdr[7]->hdr_checksum);
1060 #endif
1061
1062         /* dst addr */
1063         *(uint64_t *)&eth_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1064         *(uint64_t *)&eth_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1065         *(uint64_t *)&eth_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1066         *(uint64_t *)&eth_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1067         *(uint64_t *)&eth_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1068         *(uint64_t *)&eth_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1069         *(uint64_t *)&eth_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1070         *(uint64_t *)&eth_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1071
1072         /* src addr */
1073         ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
1074         ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
1075         ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
1076         ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
1077         ether_addr_copy(&ports_eth_addr[dst_port[4]], &eth_hdr[4]->s_addr);
1078         ether_addr_copy(&ports_eth_addr[dst_port[5]], &eth_hdr[5]->s_addr);
1079         ether_addr_copy(&ports_eth_addr[dst_port[6]], &eth_hdr[6]->s_addr);
1080         ether_addr_copy(&ports_eth_addr[dst_port[7]], &eth_hdr[7]->s_addr);
1081
1082         send_single_packet(m[0], (uint8_t)dst_port[0]);
1083         send_single_packet(m[1], (uint8_t)dst_port[1]);
1084         send_single_packet(m[2], (uint8_t)dst_port[2]);
1085         send_single_packet(m[3], (uint8_t)dst_port[3]);
1086         send_single_packet(m[4], (uint8_t)dst_port[4]);
1087         send_single_packet(m[5], (uint8_t)dst_port[5]);
1088         send_single_packet(m[6], (uint8_t)dst_port[6]);
1089         send_single_packet(m[7], (uint8_t)dst_port[7]);
1090
1091 }
1092
1093 static inline void get_ipv6_5tuple(struct rte_mbuf *m0, __m128i mask0,
1094                 __m128i mask1, union ipv6_5tuple_host *key)
1095 {
1096         __m128i tmpdata0 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1097                         __m128i *, sizeof(struct ether_hdr) +
1098                         offsetof(struct ipv6_hdr, payload_len)));
1099         __m128i tmpdata1 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1100                         __m128i *, sizeof(struct ether_hdr) +
1101                         offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i)));
1102         __m128i tmpdata2 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1103                         __m128i *, sizeof(struct ether_hdr) +
1104                         offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i) +
1105                         sizeof(__m128i)));
1106         key->xmm[0] = _mm_and_si128(tmpdata0, mask0);
1107         key->xmm[1] = tmpdata1;
1108         key->xmm[2] = _mm_and_si128(tmpdata2, mask1);
1109 }
1110
1111 static inline void
1112 simple_ipv6_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid)
1113 {
1114         int32_t ret[8];
1115         uint16_t dst_port[8];
1116         struct ether_hdr *eth_hdr[8];
1117         union ipv6_5tuple_host key[8];
1118
1119         __attribute__((unused)) struct ipv6_hdr *ipv6_hdr[8];
1120
1121         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
1122         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
1123         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
1124         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
1125         eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *);
1126         eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *);
1127         eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *);
1128         eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *);
1129
1130         /* Handle IPv6 headers.*/
1131         ipv6_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv6_hdr *,
1132                         sizeof(struct ether_hdr));
1133         ipv6_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv6_hdr *,
1134                         sizeof(struct ether_hdr));
1135         ipv6_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv6_hdr *,
1136                         sizeof(struct ether_hdr));
1137         ipv6_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv6_hdr *,
1138                         sizeof(struct ether_hdr));
1139         ipv6_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv6_hdr *,
1140                         sizeof(struct ether_hdr));
1141         ipv6_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv6_hdr *,
1142                         sizeof(struct ether_hdr));
1143         ipv6_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv6_hdr *,
1144                         sizeof(struct ether_hdr));
1145         ipv6_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv6_hdr *,
1146                         sizeof(struct ether_hdr));
1147
1148         get_ipv6_5tuple(m[0], mask1, mask2, &key[0]);
1149         get_ipv6_5tuple(m[1], mask1, mask2, &key[1]);
1150         get_ipv6_5tuple(m[2], mask1, mask2, &key[2]);
1151         get_ipv6_5tuple(m[3], mask1, mask2, &key[3]);
1152         get_ipv6_5tuple(m[4], mask1, mask2, &key[4]);
1153         get_ipv6_5tuple(m[5], mask1, mask2, &key[5]);
1154         get_ipv6_5tuple(m[6], mask1, mask2, &key[6]);
1155         get_ipv6_5tuple(m[7], mask1, mask2, &key[7]);
1156
1157         const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1158                         &key[4], &key[5], &key[6], &key[7]};
1159
1160         rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1161                         &key_array[0], 4, ret);
1162         dst_port[0] = ((ret[0] < 0) ? portid : ipv6_l3fwd_out_if[ret[0]]);
1163         dst_port[1] = ((ret[1] < 0) ? portid : ipv6_l3fwd_out_if[ret[1]]);
1164         dst_port[2] = ((ret[2] < 0) ? portid : ipv6_l3fwd_out_if[ret[2]]);
1165         dst_port[3] = ((ret[3] < 0) ? portid : ipv6_l3fwd_out_if[ret[3]]);
1166         dst_port[4] = ((ret[4] < 0) ? portid : ipv6_l3fwd_out_if[ret[4]]);
1167         dst_port[5] = ((ret[5] < 0) ? portid : ipv6_l3fwd_out_if[ret[5]]);
1168         dst_port[6] = ((ret[6] < 0) ? portid : ipv6_l3fwd_out_if[ret[6]]);
1169         dst_port[7] = ((ret[7] < 0) ? portid : ipv6_l3fwd_out_if[ret[7]]);
1170
1171         if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1172                         (enabled_port_mask & 1 << dst_port[0]) == 0)
1173                 dst_port[0] = portid;
1174         if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1175                         (enabled_port_mask & 1 << dst_port[1]) == 0)
1176                 dst_port[1] = portid;
1177         if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1178                         (enabled_port_mask & 1 << dst_port[2]) == 0)
1179                 dst_port[2] = portid;
1180         if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1181                         (enabled_port_mask & 1 << dst_port[3]) == 0)
1182                 dst_port[3] = portid;
1183         if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1184                         (enabled_port_mask & 1 << dst_port[4]) == 0)
1185                 dst_port[4] = portid;
1186         if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1187                         (enabled_port_mask & 1 << dst_port[5]) == 0)
1188                 dst_port[5] = portid;
1189         if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1190                         (enabled_port_mask & 1 << dst_port[6]) == 0)
1191                 dst_port[6] = portid;
1192         if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1193                         (enabled_port_mask & 1 << dst_port[7]) == 0)
1194                 dst_port[7] = portid;
1195
1196         /* dst addr */
1197         *(uint64_t *)&eth_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1198         *(uint64_t *)&eth_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1199         *(uint64_t *)&eth_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1200         *(uint64_t *)&eth_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1201         *(uint64_t *)&eth_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1202         *(uint64_t *)&eth_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1203         *(uint64_t *)&eth_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1204         *(uint64_t *)&eth_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1205
1206         /* src addr */
1207         ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
1208         ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
1209         ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
1210         ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
1211         ether_addr_copy(&ports_eth_addr[dst_port[4]], &eth_hdr[4]->s_addr);
1212         ether_addr_copy(&ports_eth_addr[dst_port[5]], &eth_hdr[5]->s_addr);
1213         ether_addr_copy(&ports_eth_addr[dst_port[6]], &eth_hdr[6]->s_addr);
1214         ether_addr_copy(&ports_eth_addr[dst_port[7]], &eth_hdr[7]->s_addr);
1215
1216         send_single_packet(m[0], dst_port[0]);
1217         send_single_packet(m[1], dst_port[1]);
1218         send_single_packet(m[2], dst_port[2]);
1219         send_single_packet(m[3], dst_port[3]);
1220         send_single_packet(m[4], dst_port[4]);
1221         send_single_packet(m[5], dst_port[5]);
1222         send_single_packet(m[6], dst_port[6]);
1223         send_single_packet(m[7], dst_port[7]);
1224
1225 }
1226 #endif /* APP_LOOKUP_METHOD */
1227
1228 static __rte_always_inline void
1229 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid)
1230 {
1231         struct ether_hdr *eth_hdr;
1232         struct ipv4_hdr *ipv4_hdr;
1233         uint16_t dst_port;
1234
1235         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
1236
1237         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
1238                 /* Handle IPv4 headers.*/
1239                 ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
1240                                 sizeof(struct ether_hdr));
1241
1242 #ifdef DO_RFC_1812_CHECKS
1243                 /* Check to make sure the packet is valid (RFC1812) */
1244                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
1245                         rte_pktmbuf_free(m);
1246                         return;
1247                 }
1248 #endif
1249
1250                  dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
1251                         RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct);
1252                 if (dst_port >= RTE_MAX_ETHPORTS ||
1253                                 (enabled_port_mask & 1 << dst_port) == 0)
1254                         dst_port = portid;
1255
1256 #ifdef DO_RFC_1812_CHECKS
1257                 /* Update time to live and header checksum */
1258                 --(ipv4_hdr->time_to_live);
1259                 ++(ipv4_hdr->hdr_checksum);
1260 #endif
1261                 /* dst addr */
1262                 *(uint64_t *)&eth_hdr->d_addr = dest_eth_addr[dst_port];
1263
1264                 /* src addr */
1265                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
1266
1267                 send_single_packet(m, dst_port);
1268         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
1269                 /* Handle IPv6 headers.*/
1270                 struct ipv6_hdr *ipv6_hdr;
1271
1272                 ipv6_hdr = rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
1273                                 sizeof(struct ether_hdr));
1274
1275                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
1276                                 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct);
1277
1278                 if (dst_port >= RTE_MAX_ETHPORTS ||
1279                                 (enabled_port_mask & 1 << dst_port) == 0)
1280                         dst_port = portid;
1281
1282                 /* dst addr */
1283                 *(uint64_t *)&eth_hdr->d_addr = dest_eth_addr[dst_port];
1284
1285                 /* src addr */
1286                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
1287
1288                 send_single_packet(m, dst_port);
1289         } else
1290                 /* Free the mbuf that contains non-IPV4/IPV6 packet */
1291                 rte_pktmbuf_free(m);
1292 }
1293
1294 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1295         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1296 #ifdef DO_RFC_1812_CHECKS
1297
1298 #define IPV4_MIN_VER_IHL        0x45
1299 #define IPV4_MAX_VER_IHL        0x4f
1300 #define IPV4_MAX_VER_IHL_DIFF   (IPV4_MAX_VER_IHL - IPV4_MIN_VER_IHL)
1301
1302 /* Minimum value of IPV4 total length (20B) in network byte order. */
1303 #define IPV4_MIN_LEN_BE (sizeof(struct ipv4_hdr) << 8)
1304
1305 /*
1306  * From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2:
1307  * - The IP version number must be 4.
1308  * - The IP header length field must be large enough to hold the
1309  *    minimum length legal IP datagram (20 bytes = 5 words).
1310  * - The IP total length field must be large enough to hold the IP
1311  *   datagram header, whose length is specified in the IP header length
1312  *   field.
1313  * If we encounter invalid IPV4 packet, then set destination port for it
1314  * to BAD_PORT value.
1315  */
1316 static __rte_always_inline void
1317 rfc1812_process(struct ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t ptype)
1318 {
1319         uint8_t ihl;
1320
1321         if (RTE_ETH_IS_IPV4_HDR(ptype)) {
1322                 ihl = ipv4_hdr->version_ihl - IPV4_MIN_VER_IHL;
1323
1324                 ipv4_hdr->time_to_live--;
1325                 ipv4_hdr->hdr_checksum++;
1326
1327                 if (ihl > IPV4_MAX_VER_IHL_DIFF ||
1328                                 ((uint8_t)ipv4_hdr->total_length == 0 &&
1329                                 ipv4_hdr->total_length < IPV4_MIN_LEN_BE)) {
1330                         dp[0] = BAD_PORT;
1331                 }
1332         }
1333 }
1334
1335 #else
1336 #define rfc1812_process(mb, dp, ptype)  do { } while (0)
1337 #endif /* DO_RFC_1812_CHECKS */
1338 #endif /* APP_LOOKUP_LPM && ENABLE_MULTI_BUFFER_OPTIMIZE */
1339
1340
1341 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1342         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1343
1344 static __rte_always_inline uint16_t
1345 get_dst_port(struct rte_mbuf *pkt, uint32_t dst_ipv4, uint16_t portid)
1346 {
1347         uint32_t next_hop;
1348         struct ipv6_hdr *ipv6_hdr;
1349         struct ether_hdr *eth_hdr;
1350
1351         if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
1352                 return (uint16_t) ((rte_lpm_lookup(
1353                                 RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dst_ipv4,
1354                                 &next_hop) == 0) ? next_hop : portid);
1355
1356         } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
1357
1358                 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1359                 ipv6_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
1360
1361                 return (uint16_t) ((rte_lpm6_lookup(
1362                                 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1363                                 ipv6_hdr->dst_addr, &next_hop) == 0) ?
1364                                 next_hop : portid);
1365
1366         }
1367
1368         return portid;
1369 }
1370
1371 static inline void
1372 process_packet(struct rte_mbuf *pkt, uint16_t *dst_port, uint16_t portid)
1373 {
1374         struct ether_hdr *eth_hdr;
1375         struct ipv4_hdr *ipv4_hdr;
1376         uint32_t dst_ipv4;
1377         uint16_t dp;
1378         __m128i te, ve;
1379
1380         eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1381         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1382
1383         dst_ipv4 = ipv4_hdr->dst_addr;
1384         dst_ipv4 = rte_be_to_cpu_32(dst_ipv4);
1385         dp = get_dst_port(pkt, dst_ipv4, portid);
1386
1387         te = _mm_load_si128((__m128i *)eth_hdr);
1388         ve = val_eth[dp];
1389
1390         dst_port[0] = dp;
1391         rfc1812_process(ipv4_hdr, dst_port, pkt->packet_type);
1392
1393         te =  _mm_blend_epi16(te, ve, MASK_ETH);
1394         _mm_store_si128((__m128i *)eth_hdr, te);
1395 }
1396
1397 /*
1398  * Read packet_type and destination IPV4 addresses from 4 mbufs.
1399  */
1400 static inline void
1401 processx4_step1(struct rte_mbuf *pkt[FWDSTEP],
1402                 __m128i *dip,
1403                 uint32_t *ipv4_flag)
1404 {
1405         struct ipv4_hdr *ipv4_hdr;
1406         struct ether_hdr *eth_hdr;
1407         uint32_t x0, x1, x2, x3;
1408
1409         eth_hdr = rte_pktmbuf_mtod(pkt[0], struct ether_hdr *);
1410         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1411         x0 = ipv4_hdr->dst_addr;
1412         ipv4_flag[0] = pkt[0]->packet_type & RTE_PTYPE_L3_IPV4;
1413
1414         eth_hdr = rte_pktmbuf_mtod(pkt[1], struct ether_hdr *);
1415         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1416         x1 = ipv4_hdr->dst_addr;
1417         ipv4_flag[0] &= pkt[1]->packet_type;
1418
1419         eth_hdr = rte_pktmbuf_mtod(pkt[2], struct ether_hdr *);
1420         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1421         x2 = ipv4_hdr->dst_addr;
1422         ipv4_flag[0] &= pkt[2]->packet_type;
1423
1424         eth_hdr = rte_pktmbuf_mtod(pkt[3], struct ether_hdr *);
1425         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1426         x3 = ipv4_hdr->dst_addr;
1427         ipv4_flag[0] &= pkt[3]->packet_type;
1428
1429         dip[0] = _mm_set_epi32(x3, x2, x1, x0);
1430 }
1431
1432 /*
1433  * Lookup into LPM for destination port.
1434  * If lookup fails, use incoming port (portid) as destination port.
1435  */
1436 static inline void
1437 processx4_step2(__m128i dip,
1438                 uint32_t ipv4_flag,
1439                 uint16_t portid,
1440                 struct rte_mbuf *pkt[FWDSTEP],
1441                 uint16_t dprt[FWDSTEP])
1442 {
1443         rte_xmm_t dst;
1444         const __m128i bswap_mask = _mm_set_epi8(12, 13, 14, 15, 8, 9, 10, 11,
1445                         4, 5, 6, 7, 0, 1, 2, 3);
1446
1447         /* Byte swap 4 IPV4 addresses. */
1448         dip = _mm_shuffle_epi8(dip, bswap_mask);
1449
1450         /* if all 4 packets are IPV4. */
1451         if (likely(ipv4_flag)) {
1452                 rte_lpm_lookupx4(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dip,
1453                                 dst.u32, portid);
1454
1455                 /* get rid of unused upper 16 bit for each dport. */
1456                 dst.x = _mm_packs_epi32(dst.x, dst.x);
1457                 *(uint64_t *)dprt = dst.u64[0];
1458         } else {
1459                 dst.x = dip;
1460                 dprt[0] = get_dst_port(pkt[0], dst.u32[0], portid);
1461                 dprt[1] = get_dst_port(pkt[1], dst.u32[1], portid);
1462                 dprt[2] = get_dst_port(pkt[2], dst.u32[2], portid);
1463                 dprt[3] = get_dst_port(pkt[3], dst.u32[3], portid);
1464         }
1465 }
1466
1467 /*
1468  * Update source and destination MAC addresses in the ethernet header.
1469  * Perform RFC1812 checks and updates for IPV4 packets.
1470  */
1471 static inline void
1472 processx4_step3(struct rte_mbuf *pkt[FWDSTEP], uint16_t dst_port[FWDSTEP])
1473 {
1474         __m128i te[FWDSTEP];
1475         __m128i ve[FWDSTEP];
1476         __m128i *p[FWDSTEP];
1477
1478         p[0] = rte_pktmbuf_mtod(pkt[0], __m128i *);
1479         p[1] = rte_pktmbuf_mtod(pkt[1], __m128i *);
1480         p[2] = rte_pktmbuf_mtod(pkt[2], __m128i *);
1481         p[3] = rte_pktmbuf_mtod(pkt[3], __m128i *);
1482
1483         ve[0] = val_eth[dst_port[0]];
1484         te[0] = _mm_load_si128(p[0]);
1485
1486         ve[1] = val_eth[dst_port[1]];
1487         te[1] = _mm_load_si128(p[1]);
1488
1489         ve[2] = val_eth[dst_port[2]];
1490         te[2] = _mm_load_si128(p[2]);
1491
1492         ve[3] = val_eth[dst_port[3]];
1493         te[3] = _mm_load_si128(p[3]);
1494
1495         /* Update first 12 bytes, keep rest bytes intact. */
1496         te[0] =  _mm_blend_epi16(te[0], ve[0], MASK_ETH);
1497         te[1] =  _mm_blend_epi16(te[1], ve[1], MASK_ETH);
1498         te[2] =  _mm_blend_epi16(te[2], ve[2], MASK_ETH);
1499         te[3] =  _mm_blend_epi16(te[3], ve[3], MASK_ETH);
1500
1501         _mm_store_si128(p[0], te[0]);
1502         _mm_store_si128(p[1], te[1]);
1503         _mm_store_si128(p[2], te[2]);
1504         _mm_store_si128(p[3], te[3]);
1505
1506         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[0] + 1),
1507                         &dst_port[0], pkt[0]->packet_type);
1508         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[1] + 1),
1509                         &dst_port[1], pkt[1]->packet_type);
1510         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[2] + 1),
1511                         &dst_port[2], pkt[2]->packet_type);
1512         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[3] + 1),
1513                         &dst_port[3], pkt[3]->packet_type);
1514 }
1515
1516 /*
1517  * We group consecutive packets with the same destionation port into one burst.
1518  * To avoid extra latency this is done together with some other packet
1519  * processing, but after we made a final decision about packet's destination.
1520  * To do this we maintain:
1521  * pnum - array of number of consecutive packets with the same dest port for
1522  * each packet in the input burst.
1523  * lp - pointer to the last updated element in the pnum.
1524  * dlp - dest port value lp corresponds to.
1525  */
1526
1527 #define GRPSZ   (1 << FWDSTEP)
1528 #define GRPMSK  (GRPSZ - 1)
1529
1530 #define GROUP_PORT_STEP(dlp, dcp, lp, pn, idx)  do { \
1531         if (likely((dlp) == (dcp)[(idx)])) {         \
1532                 (lp)[0]++;                           \
1533         } else {                                     \
1534                 (dlp) = (dcp)[idx];                  \
1535                 (lp) = (pn) + (idx);                 \
1536                 (lp)[0] = 1;                         \
1537         }                                            \
1538 } while (0)
1539
1540 /*
1541  * Group consecutive packets with the same destination port in bursts of 4.
1542  * Suppose we have array of destionation ports:
1543  * dst_port[] = {a, b, c, d,, e, ... }
1544  * dp1 should contain: <a, b, c, d>, dp2: <b, c, d, e>.
1545  * We doing 4 comparisons at once and the result is 4 bit mask.
1546  * This mask is used as an index into prebuild array of pnum values.
1547  */
1548 static inline uint16_t *
1549 port_groupx4(uint16_t pn[FWDSTEP + 1], uint16_t *lp, __m128i dp1, __m128i dp2)
1550 {
1551         static const struct {
1552                 uint64_t pnum; /* prebuild 4 values for pnum[]. */
1553                 int32_t  idx;  /* index for new last updated elemnet. */
1554                 uint16_t lpv;  /* add value to the last updated element. */
1555         } gptbl[GRPSZ] = {
1556         {
1557                 /* 0: a != b, b != c, c != d, d != e */
1558                 .pnum = UINT64_C(0x0001000100010001),
1559                 .idx = 4,
1560                 .lpv = 0,
1561         },
1562         {
1563                 /* 1: a == b, b != c, c != d, d != e */
1564                 .pnum = UINT64_C(0x0001000100010002),
1565                 .idx = 4,
1566                 .lpv = 1,
1567         },
1568         {
1569                 /* 2: a != b, b == c, c != d, d != e */
1570                 .pnum = UINT64_C(0x0001000100020001),
1571                 .idx = 4,
1572                 .lpv = 0,
1573         },
1574         {
1575                 /* 3: a == b, b == c, c != d, d != e */
1576                 .pnum = UINT64_C(0x0001000100020003),
1577                 .idx = 4,
1578                 .lpv = 2,
1579         },
1580         {
1581                 /* 4: a != b, b != c, c == d, d != e */
1582                 .pnum = UINT64_C(0x0001000200010001),
1583                 .idx = 4,
1584                 .lpv = 0,
1585         },
1586         {
1587                 /* 5: a == b, b != c, c == d, d != e */
1588                 .pnum = UINT64_C(0x0001000200010002),
1589                 .idx = 4,
1590                 .lpv = 1,
1591         },
1592         {
1593                 /* 6: a != b, b == c, c == d, d != e */
1594                 .pnum = UINT64_C(0x0001000200030001),
1595                 .idx = 4,
1596                 .lpv = 0,
1597         },
1598         {
1599                 /* 7: a == b, b == c, c == d, d != e */
1600                 .pnum = UINT64_C(0x0001000200030004),
1601                 .idx = 4,
1602                 .lpv = 3,
1603         },
1604         {
1605                 /* 8: a != b, b != c, c != d, d == e */
1606                 .pnum = UINT64_C(0x0002000100010001),
1607                 .idx = 3,
1608                 .lpv = 0,
1609         },
1610         {
1611                 /* 9: a == b, b != c, c != d, d == e */
1612                 .pnum = UINT64_C(0x0002000100010002),
1613                 .idx = 3,
1614                 .lpv = 1,
1615         },
1616         {
1617                 /* 0xa: a != b, b == c, c != d, d == e */
1618                 .pnum = UINT64_C(0x0002000100020001),
1619                 .idx = 3,
1620                 .lpv = 0,
1621         },
1622         {
1623                 /* 0xb: a == b, b == c, c != d, d == e */
1624                 .pnum = UINT64_C(0x0002000100020003),
1625                 .idx = 3,
1626                 .lpv = 2,
1627         },
1628         {
1629                 /* 0xc: a != b, b != c, c == d, d == e */
1630                 .pnum = UINT64_C(0x0002000300010001),
1631                 .idx = 2,
1632                 .lpv = 0,
1633         },
1634         {
1635                 /* 0xd: a == b, b != c, c == d, d == e */
1636                 .pnum = UINT64_C(0x0002000300010002),
1637                 .idx = 2,
1638                 .lpv = 1,
1639         },
1640         {
1641                 /* 0xe: a != b, b == c, c == d, d == e */
1642                 .pnum = UINT64_C(0x0002000300040001),
1643                 .idx = 1,
1644                 .lpv = 0,
1645         },
1646         {
1647                 /* 0xf: a == b, b == c, c == d, d == e */
1648                 .pnum = UINT64_C(0x0002000300040005),
1649                 .idx = 0,
1650                 .lpv = 4,
1651         },
1652         };
1653
1654         union {
1655                 uint16_t u16[FWDSTEP + 1];
1656                 uint64_t u64;
1657         } *pnum = (void *)pn;
1658
1659         int32_t v;
1660
1661         dp1 = _mm_cmpeq_epi16(dp1, dp2);
1662         dp1 = _mm_unpacklo_epi16(dp1, dp1);
1663         v = _mm_movemask_ps((__m128)dp1);
1664
1665         /* update last port counter. */
1666         lp[0] += gptbl[v].lpv;
1667
1668         /* if dest port value has changed. */
1669         if (v != GRPMSK) {
1670                 pnum->u64 = gptbl[v].pnum;
1671                 pnum->u16[FWDSTEP] = 1;
1672                 lp = pnum->u16 + gptbl[v].idx;
1673         }
1674
1675         return lp;
1676 }
1677
1678 #endif /* APP_LOOKUP_METHOD */
1679
1680 static void
1681 process_burst(struct rte_mbuf *pkts_burst[MAX_PKT_BURST], int nb_rx,
1682                 uint16_t portid)
1683 {
1684
1685         int j;
1686
1687 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1688         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1689         int32_t k;
1690         uint16_t dlp;
1691         uint16_t *lp;
1692         uint16_t dst_port[MAX_PKT_BURST];
1693         __m128i dip[MAX_PKT_BURST / FWDSTEP];
1694         uint32_t ipv4_flag[MAX_PKT_BURST / FWDSTEP];
1695         uint16_t pnum[MAX_PKT_BURST + 1];
1696 #endif
1697
1698
1699 #if (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
1700 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1701         {
1702                 /*
1703                  * Send nb_rx - nb_rx%8 packets
1704                  * in groups of 8.
1705                  */
1706                 int32_t n = RTE_ALIGN_FLOOR(nb_rx, 8);
1707
1708                 for (j = 0; j < n; j += 8) {
1709                         uint32_t pkt_type =
1710                                 pkts_burst[j]->packet_type &
1711                                 pkts_burst[j+1]->packet_type &
1712                                 pkts_burst[j+2]->packet_type &
1713                                 pkts_burst[j+3]->packet_type &
1714                                 pkts_burst[j+4]->packet_type &
1715                                 pkts_burst[j+5]->packet_type &
1716                                 pkts_burst[j+6]->packet_type &
1717                                 pkts_burst[j+7]->packet_type;
1718                         if (pkt_type & RTE_PTYPE_L3_IPV4) {
1719                                 simple_ipv4_fwd_8pkts(&pkts_burst[j], portid);
1720                         } else if (pkt_type &
1721                                 RTE_PTYPE_L3_IPV6) {
1722                                 simple_ipv6_fwd_8pkts(&pkts_burst[j], portid);
1723                         } else {
1724                                 l3fwd_simple_forward(pkts_burst[j], portid);
1725                                 l3fwd_simple_forward(pkts_burst[j+1], portid);
1726                                 l3fwd_simple_forward(pkts_burst[j+2], portid);
1727                                 l3fwd_simple_forward(pkts_burst[j+3], portid);
1728                                 l3fwd_simple_forward(pkts_burst[j+4], portid);
1729                                 l3fwd_simple_forward(pkts_burst[j+5], portid);
1730                                 l3fwd_simple_forward(pkts_burst[j+6], portid);
1731                                 l3fwd_simple_forward(pkts_burst[j+7], portid);
1732                         }
1733                 }
1734                 for (; j < nb_rx ; j++)
1735                         l3fwd_simple_forward(pkts_burst[j], portid);
1736         }
1737 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1738
1739         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1740         for (j = 0; j != k; j += FWDSTEP)
1741                 processx4_step1(&pkts_burst[j], &dip[j / FWDSTEP],
1742                                 &ipv4_flag[j / FWDSTEP]);
1743
1744         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1745         for (j = 0; j != k; j += FWDSTEP)
1746                 processx4_step2(dip[j / FWDSTEP], ipv4_flag[j / FWDSTEP],
1747                                 portid, &pkts_burst[j], &dst_port[j]);
1748
1749         /*
1750          * Finish packet processing and group consecutive
1751          * packets with the same destination port.
1752          */
1753         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1754         if (k != 0) {
1755                 __m128i dp1, dp2;
1756
1757                 lp = pnum;
1758                 lp[0] = 1;
1759
1760                 processx4_step3(pkts_burst, dst_port);
1761
1762                 /* dp1: <d[0], d[1], d[2], d[3], ... > */
1763                 dp1 = _mm_loadu_si128((__m128i *)dst_port);
1764
1765                 for (j = FWDSTEP; j != k; j += FWDSTEP) {
1766                         processx4_step3(&pkts_burst[j], &dst_port[j]);
1767
1768                         /*
1769                          * dp2:
1770                          * <d[j-3], d[j-2], d[j-1], d[j], ... >
1771                          */
1772                         dp2 = _mm_loadu_si128(
1773                                         (__m128i *)&dst_port[j - FWDSTEP + 1]);
1774                         lp  = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1775
1776                         /*
1777                          * dp1:
1778                          * <d[j], d[j+1], d[j+2], d[j+3], ... >
1779                          */
1780                         dp1 = _mm_srli_si128(dp2, (FWDSTEP - 1) *
1781                                         sizeof(dst_port[0]));
1782                 }
1783
1784                 /*
1785                  * dp2: <d[j-3], d[j-2], d[j-1], d[j-1], ... >
1786                  */
1787                 dp2 = _mm_shufflelo_epi16(dp1, 0xf9);
1788                 lp  = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1789
1790                 /*
1791                  * remove values added by the last repeated
1792                  * dst port.
1793                  */
1794                 lp[0]--;
1795                 dlp = dst_port[j - 1];
1796         } else {
1797                 /* set dlp and lp to the never used values. */
1798                 dlp = BAD_PORT - 1;
1799                 lp = pnum + MAX_PKT_BURST;
1800         }
1801
1802         /* Process up to last 3 packets one by one. */
1803         switch (nb_rx % FWDSTEP) {
1804         case 3:
1805                 process_packet(pkts_burst[j], dst_port + j, portid);
1806                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1807                 j++;
1808                 /* fall-through */
1809         case 2:
1810                 process_packet(pkts_burst[j], dst_port + j, portid);
1811                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1812                 j++;
1813                 /* fall-through */
1814         case 1:
1815                 process_packet(pkts_burst[j], dst_port + j, portid);
1816                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1817                 j++;
1818         }
1819
1820         /*
1821          * Send packets out, through destination port.
1822          * Consecuteve pacekts with the same destination port
1823          * are already grouped together.
1824          * If destination port for the packet equals BAD_PORT,
1825          * then free the packet without sending it out.
1826          */
1827         for (j = 0; j < nb_rx; j += k) {
1828
1829                 int32_t m;
1830                 uint16_t pn;
1831
1832                 pn = dst_port[j];
1833                 k = pnum[j];
1834
1835                 if (likely(pn != BAD_PORT))
1836                         send_packetsx4(pn, pkts_burst + j, k);
1837                 else
1838                         for (m = j; m != j + k; m++)
1839                                 rte_pktmbuf_free(pkts_burst[m]);
1840
1841         }
1842
1843 #endif /* APP_LOOKUP_METHOD */
1844 #else /* ENABLE_MULTI_BUFFER_OPTIMIZE == 0 */
1845
1846         /* Prefetch first packets */
1847         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++)
1848                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[j], void *));
1849
1850         /* Prefetch and forward already prefetched packets */
1851         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1852                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1853                                 j + PREFETCH_OFFSET], void *));
1854                 l3fwd_simple_forward(pkts_burst[j], portid);
1855         }
1856
1857         /* Forward remaining prefetched packets */
1858         for (; j < nb_rx; j++)
1859                 l3fwd_simple_forward(pkts_burst[j], portid);
1860
1861 #endif /* ENABLE_MULTI_BUFFER_OPTIMIZE */
1862
1863 }
1864
1865 #if (APP_CPU_LOAD > 0)
1866
1867 /*
1868  * CPU-load stats collector
1869  */
1870 static int
1871 cpu_load_collector(__rte_unused void *arg) {
1872         unsigned i, j, k;
1873         uint64_t hits;
1874         uint64_t prev_tsc, diff_tsc, cur_tsc;
1875         uint64_t total[MAX_CPU] = { 0 };
1876         unsigned min_cpu = MAX_CPU;
1877         unsigned max_cpu = 0;
1878         unsigned cpu_id;
1879         int busy_total = 0;
1880         int busy_flag = 0;
1881
1882         unsigned int n_thread_per_cpu[MAX_CPU] = { 0 };
1883         struct thread_conf *thread_per_cpu[MAX_CPU][MAX_THREAD];
1884
1885         struct thread_conf *thread_conf;
1886
1887         const uint64_t interval_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1888                 US_PER_S * CPU_LOAD_TIMEOUT_US;
1889
1890         prev_tsc = 0;
1891         /*
1892          * Wait for all threads
1893          */
1894
1895         printf("Waiting for %d rx threads and %d tx threads\n", n_rx_thread,
1896                         n_tx_thread);
1897
1898         while (rte_atomic16_read(&rx_counter) < n_rx_thread)
1899                 rte_pause();
1900
1901         while (rte_atomic16_read(&tx_counter) < n_tx_thread)
1902                 rte_pause();
1903
1904         for (i = 0; i < n_rx_thread; i++) {
1905
1906                 thread_conf = &rx_thread[i].conf;
1907                 cpu_id = thread_conf->cpu_id;
1908                 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1909
1910                 if (cpu_id > max_cpu)
1911                         max_cpu = cpu_id;
1912                 if (cpu_id < min_cpu)
1913                         min_cpu = cpu_id;
1914         }
1915         for (i = 0; i < n_tx_thread; i++) {
1916
1917                 thread_conf = &tx_thread[i].conf;
1918                 cpu_id = thread_conf->cpu_id;
1919                 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1920
1921                 if (thread_conf->cpu_id > max_cpu)
1922                         max_cpu = thread_conf->cpu_id;
1923                 if (thread_conf->cpu_id < min_cpu)
1924                         min_cpu = thread_conf->cpu_id;
1925         }
1926
1927         while (1) {
1928
1929                 cpu_load.counter++;
1930                 for (i = min_cpu; i <= max_cpu; i++) {
1931                         for (j = 0; j < MAX_CPU_COUNTER; j++) {
1932                                 for (k = 0; k < n_thread_per_cpu[i]; k++)
1933                                         if (thread_per_cpu[i][k]->busy[j]) {
1934                                                 busy_flag = 1;
1935                                                 break;
1936                                         }
1937                                 if (busy_flag) {
1938                                         cpu_load.hits[j][i]++;
1939                                         busy_total = 1;
1940                                         busy_flag = 0;
1941                                 }
1942                         }
1943
1944                         if (busy_total) {
1945                                 total[i]++;
1946                                 busy_total = 0;
1947                         }
1948                 }
1949
1950                 cur_tsc = rte_rdtsc();
1951
1952                 diff_tsc = cur_tsc - prev_tsc;
1953                 if (unlikely(diff_tsc > interval_tsc)) {
1954
1955                         printf("\033c");
1956
1957                         printf("Cpu usage for %d rx threads and %d tx threads:\n\n",
1958                                         n_rx_thread, n_tx_thread);
1959
1960                         printf("cpu#     proc%%  poll%%  overhead%%\n\n");
1961
1962                         for (i = min_cpu; i <= max_cpu; i++) {
1963                                 hits = 0;
1964                                 printf("CPU %d:", i);
1965                                 for (j = 0; j < MAX_CPU_COUNTER; j++) {
1966                                         printf("%7" PRIu64 "",
1967                                                         cpu_load.hits[j][i] * 100 / cpu_load.counter);
1968                                         hits += cpu_load.hits[j][i];
1969                                         cpu_load.hits[j][i] = 0;
1970                                 }
1971                                 printf("%7" PRIu64 "\n",
1972                                                 100 - total[i] * 100 / cpu_load.counter);
1973                                 total[i] = 0;
1974                         }
1975                         cpu_load.counter = 0;
1976
1977                         prev_tsc = cur_tsc;
1978                 }
1979
1980         }
1981 }
1982 #endif /* APP_CPU_LOAD */
1983
1984 /*
1985  * Null processing lthread loop
1986  *
1987  * This loop is used to start empty scheduler on lcore.
1988  */
1989 static void *
1990 lthread_null(__rte_unused void *args)
1991 {
1992         int lcore_id = rte_lcore_id();
1993
1994         RTE_LOG(INFO, L3FWD, "Starting scheduler on lcore %d.\n", lcore_id);
1995         lthread_exit(NULL);
1996         return NULL;
1997 }
1998
1999 /* main processing loop */
2000 static void *
2001 lthread_tx_per_ring(void *dummy)
2002 {
2003         int nb_rx;
2004         uint16_t portid;
2005         struct rte_ring *ring;
2006         struct thread_tx_conf *tx_conf;
2007         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2008         struct lthread_cond *ready;
2009
2010         tx_conf = (struct thread_tx_conf *)dummy;
2011         ring = tx_conf->ring;
2012         ready = *tx_conf->ready;
2013
2014         lthread_set_data((void *)tx_conf);
2015
2016         /*
2017          * Move this lthread to lcore
2018          */
2019         lthread_set_affinity(tx_conf->conf.lcore_id);
2020
2021         RTE_LOG(INFO, L3FWD, "entering main tx loop on lcore %u\n", rte_lcore_id());
2022
2023         nb_rx = 0;
2024         rte_atomic16_inc(&tx_counter);
2025         while (1) {
2026
2027                 /*
2028                  * Read packet from ring
2029                  */
2030                 SET_CPU_BUSY(tx_conf, CPU_POLL);
2031                 nb_rx = rte_ring_sc_dequeue_burst(ring, (void **)pkts_burst,
2032                                 MAX_PKT_BURST, NULL);
2033                 SET_CPU_IDLE(tx_conf, CPU_POLL);
2034
2035                 if (nb_rx > 0) {
2036                         SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2037                         portid = pkts_burst[0]->port;
2038                         process_burst(pkts_burst, nb_rx, portid);
2039                         SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2040                         lthread_yield();
2041                 } else
2042                         lthread_cond_wait(ready, 0);
2043
2044         }
2045         return NULL;
2046 }
2047
2048 /*
2049  * Main tx-lthreads spawner lthread.
2050  *
2051  * This lthread is used to spawn one new lthread per ring from producers.
2052  *
2053  */
2054 static void *
2055 lthread_tx(void *args)
2056 {
2057         struct lthread *lt;
2058
2059         unsigned lcore_id;
2060         uint16_t portid;
2061         struct thread_tx_conf *tx_conf;
2062
2063         tx_conf = (struct thread_tx_conf *)args;
2064         lthread_set_data((void *)tx_conf);
2065
2066         /*
2067          * Move this lthread to the selected lcore
2068          */
2069         lthread_set_affinity(tx_conf->conf.lcore_id);
2070
2071         /*
2072          * Spawn tx readers (one per input ring)
2073          */
2074         lthread_create(&lt, tx_conf->conf.lcore_id, lthread_tx_per_ring,
2075                         (void *)tx_conf);
2076
2077         lcore_id = rte_lcore_id();
2078
2079         RTE_LOG(INFO, L3FWD, "Entering Tx main loop on lcore %u\n", lcore_id);
2080
2081         tx_conf->conf.cpu_id = sched_getcpu();
2082         while (1) {
2083
2084                 lthread_sleep(BURST_TX_DRAIN_US * 1000);
2085
2086                 /*
2087                  * TX burst queue drain
2088                  */
2089                 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2090                         if (tx_conf->tx_mbufs[portid].len == 0)
2091                                 continue;
2092                         SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2093                         send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2094                         SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2095                         tx_conf->tx_mbufs[portid].len = 0;
2096                 }
2097
2098         }
2099         return NULL;
2100 }
2101
2102 static void *
2103 lthread_rx(void *dummy)
2104 {
2105         int ret;
2106         uint16_t nb_rx;
2107         int i;
2108         uint16_t portid;
2109         uint8_t queueid;
2110         int worker_id;
2111         int len[RTE_MAX_LCORE] = { 0 };
2112         int old_len, new_len;
2113         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2114         struct thread_rx_conf *rx_conf;
2115
2116         rx_conf = (struct thread_rx_conf *)dummy;
2117         lthread_set_data((void *)rx_conf);
2118
2119         /*
2120          * Move this lthread to lcore
2121          */
2122         lthread_set_affinity(rx_conf->conf.lcore_id);
2123
2124         if (rx_conf->n_rx_queue == 0) {
2125                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", rte_lcore_id());
2126                 return NULL;
2127         }
2128
2129         RTE_LOG(INFO, L3FWD, "Entering main Rx loop on lcore %u\n", rte_lcore_id());
2130
2131         for (i = 0; i < rx_conf->n_rx_queue; i++) {
2132
2133                 portid = rx_conf->rx_queue_list[i].port_id;
2134                 queueid = rx_conf->rx_queue_list[i].queue_id;
2135                 RTE_LOG(INFO, L3FWD,
2136                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2137                                 rte_lcore_id(), portid, queueid);
2138         }
2139
2140         /*
2141          * Init all condition variables (one per rx thread)
2142          */
2143         for (i = 0; i < rx_conf->n_rx_queue; i++)
2144                 lthread_cond_init(NULL, &rx_conf->ready[i], NULL);
2145
2146         worker_id = 0;
2147
2148         rx_conf->conf.cpu_id = sched_getcpu();
2149         rte_atomic16_inc(&rx_counter);
2150         while (1) {
2151
2152                 /*
2153                  * Read packet from RX queues
2154                  */
2155                 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2156                         portid = rx_conf->rx_queue_list[i].port_id;
2157                         queueid = rx_conf->rx_queue_list[i].queue_id;
2158
2159                         SET_CPU_BUSY(rx_conf, CPU_POLL);
2160                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2161                                 MAX_PKT_BURST);
2162                         SET_CPU_IDLE(rx_conf, CPU_POLL);
2163
2164                         if (nb_rx != 0) {
2165                                 worker_id = (worker_id + 1) % rx_conf->n_ring;
2166                                 old_len = len[worker_id];
2167
2168                                 SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2169                                 ret = rte_ring_sp_enqueue_burst(
2170                                                 rx_conf->ring[worker_id],
2171                                                 (void **) pkts_burst,
2172                                                 nb_rx, NULL);
2173
2174                                 new_len = old_len + ret;
2175
2176                                 if (new_len >= BURST_SIZE) {
2177                                         lthread_cond_signal(rx_conf->ready[worker_id]);
2178                                         new_len = 0;
2179                                 }
2180
2181                                 len[worker_id] = new_len;
2182
2183                                 if (unlikely(ret < nb_rx)) {
2184                                         uint32_t k;
2185
2186                                         for (k = ret; k < nb_rx; k++) {
2187                                                 struct rte_mbuf *m = pkts_burst[k];
2188
2189                                                 rte_pktmbuf_free(m);
2190                                         }
2191                                 }
2192                                 SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2193                         }
2194
2195                         lthread_yield();
2196                 }
2197         }
2198         return NULL;
2199 }
2200
2201 /*
2202  * Start scheduler with initial lthread on lcore
2203  *
2204  * This lthread loop spawns all rx and tx lthreads on master lcore
2205  */
2206
2207 static void *
2208 lthread_spawner(__rte_unused void *arg)
2209 {
2210         struct lthread *lt[MAX_THREAD];
2211         int i;
2212         int n_thread = 0;
2213
2214         printf("Entering lthread_spawner\n");
2215
2216         /*
2217          * Create producers (rx threads) on default lcore
2218          */
2219         for (i = 0; i < n_rx_thread; i++) {
2220                 rx_thread[i].conf.thread_id = i;
2221                 lthread_create(&lt[n_thread], -1, lthread_rx,
2222                                 (void *)&rx_thread[i]);
2223                 n_thread++;
2224         }
2225
2226         /*
2227          * Wait for all producers. Until some producers can be started on the same
2228          * scheduler as this lthread, yielding is required to let them to run and
2229          * prevent deadlock here.
2230          */
2231         while (rte_atomic16_read(&rx_counter) < n_rx_thread)
2232                 lthread_sleep(100000);
2233
2234         /*
2235          * Create consumers (tx threads) on default lcore_id
2236          */
2237         for (i = 0; i < n_tx_thread; i++) {
2238                 tx_thread[i].conf.thread_id = i;
2239                 lthread_create(&lt[n_thread], -1, lthread_tx,
2240                                 (void *)&tx_thread[i]);
2241                 n_thread++;
2242         }
2243
2244         /*
2245          * Wait for all threads finished
2246          */
2247         for (i = 0; i < n_thread; i++)
2248                 lthread_join(lt[i], NULL);
2249
2250         return NULL;
2251 }
2252
2253 /*
2254  * Start master scheduler with initial lthread spawning rx and tx lthreads
2255  * (main_lthread_master).
2256  */
2257 static int
2258 lthread_master_spawner(__rte_unused void *arg) {
2259         struct lthread *lt;
2260         int lcore_id = rte_lcore_id();
2261
2262         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2263         lthread_create(&lt, -1, lthread_spawner, NULL);
2264         lthread_run();
2265
2266         return 0;
2267 }
2268
2269 /*
2270  * Start scheduler on lcore.
2271  */
2272 static int
2273 sched_spawner(__rte_unused void *arg) {
2274         struct lthread *lt;
2275         int lcore_id = rte_lcore_id();
2276
2277 #if (APP_CPU_LOAD)
2278         if (lcore_id == cpu_load_lcore_id) {
2279                 cpu_load_collector(arg);
2280                 return 0;
2281         }
2282 #endif /* APP_CPU_LOAD */
2283
2284         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2285         lthread_create(&lt, -1, lthread_null, NULL);
2286         lthread_run();
2287
2288         return 0;
2289 }
2290
2291 /* main processing loop */
2292 static int
2293 pthread_tx(void *dummy)
2294 {
2295         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2296         uint64_t prev_tsc, diff_tsc, cur_tsc;
2297         int nb_rx;
2298         uint16_t portid;
2299         struct thread_tx_conf *tx_conf;
2300
2301         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
2302                 US_PER_S * BURST_TX_DRAIN_US;
2303
2304         prev_tsc = 0;
2305
2306         tx_conf = (struct thread_tx_conf *)dummy;
2307
2308         RTE_LOG(INFO, L3FWD, "Entering main Tx loop on lcore %u\n", rte_lcore_id());
2309
2310         tx_conf->conf.cpu_id = sched_getcpu();
2311         rte_atomic16_inc(&tx_counter);
2312         while (1) {
2313
2314                 cur_tsc = rte_rdtsc();
2315
2316                 /*
2317                  * TX burst queue drain
2318                  */
2319                 diff_tsc = cur_tsc - prev_tsc;
2320                 if (unlikely(diff_tsc > drain_tsc)) {
2321
2322                         /*
2323                          * This could be optimized (use queueid instead of
2324                          * portid), but it is not called so often
2325                          */
2326                         SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2327                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2328                                 if (tx_conf->tx_mbufs[portid].len == 0)
2329                                         continue;
2330                                 send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2331                                 tx_conf->tx_mbufs[portid].len = 0;
2332                         }
2333                         SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2334
2335                         prev_tsc = cur_tsc;
2336                 }
2337
2338                 /*
2339                  * Read packet from ring
2340                  */
2341                 SET_CPU_BUSY(tx_conf, CPU_POLL);
2342                 nb_rx = rte_ring_sc_dequeue_burst(tx_conf->ring,
2343                                 (void **)pkts_burst, MAX_PKT_BURST, NULL);
2344                 SET_CPU_IDLE(tx_conf, CPU_POLL);
2345
2346                 if (unlikely(nb_rx == 0)) {
2347                         sched_yield();
2348                         continue;
2349                 }
2350
2351                 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2352                 portid = pkts_burst[0]->port;
2353                 process_burst(pkts_burst, nb_rx, portid);
2354                 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2355
2356         }
2357 }
2358
2359 static int
2360 pthread_rx(void *dummy)
2361 {
2362         int i;
2363         int worker_id;
2364         uint32_t n;
2365         uint32_t nb_rx;
2366         unsigned lcore_id;
2367         uint8_t queueid;
2368         uint16_t portid;
2369         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2370
2371         struct thread_rx_conf *rx_conf;
2372
2373         lcore_id = rte_lcore_id();
2374         rx_conf = (struct thread_rx_conf *)dummy;
2375
2376         if (rx_conf->n_rx_queue == 0) {
2377                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
2378                 return 0;
2379         }
2380
2381         RTE_LOG(INFO, L3FWD, "entering main rx loop on lcore %u\n", lcore_id);
2382
2383         for (i = 0; i < rx_conf->n_rx_queue; i++) {
2384
2385                 portid = rx_conf->rx_queue_list[i].port_id;
2386                 queueid = rx_conf->rx_queue_list[i].queue_id;
2387                 RTE_LOG(INFO, L3FWD,
2388                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2389                                 lcore_id, portid, queueid);
2390         }
2391
2392         worker_id = 0;
2393         rx_conf->conf.cpu_id = sched_getcpu();
2394         rte_atomic16_inc(&rx_counter);
2395         while (1) {
2396
2397                 /*
2398                  * Read packet from RX queues
2399                  */
2400                 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2401                         portid = rx_conf->rx_queue_list[i].port_id;
2402                         queueid = rx_conf->rx_queue_list[i].queue_id;
2403
2404                         SET_CPU_BUSY(rx_conf, CPU_POLL);
2405                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2406                                 MAX_PKT_BURST);
2407                         SET_CPU_IDLE(rx_conf, CPU_POLL);
2408
2409                         if (nb_rx == 0) {
2410                                 sched_yield();
2411                                 continue;
2412                         }
2413
2414                         SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2415                         worker_id = (worker_id + 1) % rx_conf->n_ring;
2416                         n = rte_ring_sp_enqueue_burst(rx_conf->ring[worker_id],
2417                                         (void **)pkts_burst, nb_rx, NULL);
2418
2419                         if (unlikely(n != nb_rx)) {
2420                                 uint32_t k;
2421
2422                                 for (k = n; k < nb_rx; k++) {
2423                                         struct rte_mbuf *m = pkts_burst[k];
2424
2425                                         rte_pktmbuf_free(m);
2426                                 }
2427                         }
2428
2429                         SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2430
2431                 }
2432         }
2433 }
2434
2435 /*
2436  * P-Thread spawner.
2437  */
2438 static int
2439 pthread_run(__rte_unused void *arg) {
2440         int lcore_id = rte_lcore_id();
2441         int i;
2442
2443         for (i = 0; i < n_rx_thread; i++)
2444                 if (rx_thread[i].conf.lcore_id == lcore_id) {
2445                         printf("Start rx thread on %d...\n", lcore_id);
2446                         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2447                         RTE_PER_LCORE(lcore_conf)->data = (void *)&rx_thread[i];
2448                         pthread_rx((void *)&rx_thread[i]);
2449                         return 0;
2450                 }
2451
2452         for (i = 0; i < n_tx_thread; i++)
2453                 if (tx_thread[i].conf.lcore_id == lcore_id) {
2454                         printf("Start tx thread on %d...\n", lcore_id);
2455                         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2456                         RTE_PER_LCORE(lcore_conf)->data = (void *)&tx_thread[i];
2457                         pthread_tx((void *)&tx_thread[i]);
2458                         return 0;
2459                 }
2460
2461 #if (APP_CPU_LOAD)
2462         if (lcore_id == cpu_load_lcore_id)
2463                 cpu_load_collector(arg);
2464 #endif /* APP_CPU_LOAD */
2465
2466         return 0;
2467 }
2468
2469 static int
2470 check_lcore_params(void)
2471 {
2472         uint8_t queue, lcore;
2473         uint16_t i;
2474         int socketid;
2475
2476         for (i = 0; i < nb_rx_thread_params; ++i) {
2477                 queue = rx_thread_params[i].queue_id;
2478                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
2479                         printf("invalid queue number: %hhu\n", queue);
2480                         return -1;
2481                 }
2482                 lcore = rx_thread_params[i].lcore_id;
2483                 if (!rte_lcore_is_enabled(lcore)) {
2484                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
2485                         return -1;
2486                 }
2487                 socketid = rte_lcore_to_socket_id(lcore);
2488                 if ((socketid != 0) && (numa_on == 0))
2489                         printf("warning: lcore %hhu is on socket %d with numa off\n",
2490                                 lcore, socketid);
2491         }
2492         return 0;
2493 }
2494
2495 static int
2496 check_port_config(void)
2497 {
2498         unsigned portid;
2499         uint16_t i;
2500
2501         for (i = 0; i < nb_rx_thread_params; ++i) {
2502                 portid = rx_thread_params[i].port_id;
2503                 if ((enabled_port_mask & (1 << portid)) == 0) {
2504                         printf("port %u is not enabled in port mask\n", portid);
2505                         return -1;
2506                 }
2507                 if (!rte_eth_dev_is_valid_port(portid)) {
2508                         printf("port %u is not present on the board\n", portid);
2509                         return -1;
2510                 }
2511         }
2512         return 0;
2513 }
2514
2515 static uint8_t
2516 get_port_n_rx_queues(const uint16_t port)
2517 {
2518         int queue = -1;
2519         uint16_t i;
2520
2521         for (i = 0; i < nb_rx_thread_params; ++i)
2522                 if (rx_thread_params[i].port_id == port &&
2523                                 rx_thread_params[i].queue_id > queue)
2524                         queue = rx_thread_params[i].queue_id;
2525
2526         return (uint8_t)(++queue);
2527 }
2528
2529 static int
2530 init_rx_rings(void)
2531 {
2532         unsigned socket_io;
2533         struct thread_rx_conf *rx_conf;
2534         struct thread_tx_conf *tx_conf;
2535         unsigned rx_thread_id, tx_thread_id;
2536         char name[256];
2537         struct rte_ring *ring = NULL;
2538
2539         for (tx_thread_id = 0; tx_thread_id < n_tx_thread; tx_thread_id++) {
2540
2541                 tx_conf = &tx_thread[tx_thread_id];
2542
2543                 printf("Connecting tx-thread %d with rx-thread %d\n", tx_thread_id,
2544                                 tx_conf->conf.thread_id);
2545
2546                 rx_thread_id = tx_conf->conf.thread_id;
2547                 if (rx_thread_id > n_tx_thread) {
2548                         printf("connection from tx-thread %u to rx-thread %u fails "
2549                                         "(rx-thread not defined)\n", tx_thread_id, rx_thread_id);
2550                         return -1;
2551                 }
2552
2553                 rx_conf = &rx_thread[rx_thread_id];
2554                 socket_io = rte_lcore_to_socket_id(rx_conf->conf.lcore_id);
2555
2556                 snprintf(name, sizeof(name), "app_ring_s%u_rx%u_tx%u",
2557                                 socket_io, rx_thread_id, tx_thread_id);
2558
2559                 ring = rte_ring_create(name, 1024 * 4, socket_io,
2560                                 RING_F_SP_ENQ | RING_F_SC_DEQ);
2561
2562                 if (ring == NULL) {
2563                         rte_panic("Cannot create ring to connect rx-thread %u "
2564                                         "with tx-thread %u\n", rx_thread_id, tx_thread_id);
2565                 }
2566
2567                 rx_conf->ring[rx_conf->n_ring] = ring;
2568
2569                 tx_conf->ring = ring;
2570                 tx_conf->ready = &rx_conf->ready[rx_conf->n_ring];
2571
2572                 rx_conf->n_ring++;
2573         }
2574         return 0;
2575 }
2576
2577 static int
2578 init_rx_queues(void)
2579 {
2580         uint16_t i, nb_rx_queue;
2581         uint8_t thread;
2582
2583         n_rx_thread = 0;
2584
2585         for (i = 0; i < nb_rx_thread_params; ++i) {
2586                 thread = rx_thread_params[i].thread_id;
2587                 nb_rx_queue = rx_thread[thread].n_rx_queue;
2588
2589                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
2590                         printf("error: too many queues (%u) for thread: %u\n",
2591                                 (unsigned)nb_rx_queue + 1, (unsigned)thread);
2592                         return -1;
2593                 }
2594
2595                 rx_thread[thread].conf.thread_id = thread;
2596                 rx_thread[thread].conf.lcore_id = rx_thread_params[i].lcore_id;
2597                 rx_thread[thread].rx_queue_list[nb_rx_queue].port_id =
2598                         rx_thread_params[i].port_id;
2599                 rx_thread[thread].rx_queue_list[nb_rx_queue].queue_id =
2600                         rx_thread_params[i].queue_id;
2601                 rx_thread[thread].n_rx_queue++;
2602
2603                 if (thread >= n_rx_thread)
2604                         n_rx_thread = thread + 1;
2605
2606         }
2607         return 0;
2608 }
2609
2610 static int
2611 init_tx_threads(void)
2612 {
2613         int i;
2614
2615         n_tx_thread = 0;
2616         for (i = 0; i < nb_tx_thread_params; ++i) {
2617                 tx_thread[n_tx_thread].conf.thread_id = tx_thread_params[i].thread_id;
2618                 tx_thread[n_tx_thread].conf.lcore_id = tx_thread_params[i].lcore_id;
2619                 n_tx_thread++;
2620         }
2621         return 0;
2622 }
2623
2624 /* display usage */
2625 static void
2626 print_usage(const char *prgname)
2627 {
2628         printf("%s [EAL options] -- -p PORTMASK -P"
2629                 "  [--rx (port,queue,lcore,thread)[,(port,queue,lcore,thread]]"
2630                 "  [--tx (lcore,thread)[,(lcore,thread]]"
2631                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
2632                 "  [--parse-ptype]\n\n"
2633                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
2634                 "  -P : enable promiscuous mode\n"
2635                 "  --rx (port,queue,lcore,thread): rx queues configuration\n"
2636                 "  --tx (lcore,thread): tx threads configuration\n"
2637                 "  --stat-lcore LCORE: use lcore for stat collector\n"
2638                 "  --eth-dest=X,MM:MM:MM:MM:MM:MM: optional, ethernet destination for port X\n"
2639                 "  --no-numa: optional, disable numa awareness\n"
2640                 "  --ipv6: optional, specify it if running ipv6 packets\n"
2641                 "  --enable-jumbo: enable jumbo frame"
2642                 " which max packet len is PKTLEN in decimal (64-9600)\n"
2643                 "  --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n"
2644                 "  --no-lthreads: turn off lthread model\n"
2645                 "  --parse-ptype: set to use software to analyze packet type\n\n",
2646                 prgname);
2647 }
2648
2649 static int parse_max_pkt_len(const char *pktlen)
2650 {
2651         char *end = NULL;
2652         unsigned long len;
2653
2654         /* parse decimal string */
2655         len = strtoul(pktlen, &end, 10);
2656         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
2657                 return -1;
2658
2659         if (len == 0)
2660                 return -1;
2661
2662         return len;
2663 }
2664
2665 static int
2666 parse_portmask(const char *portmask)
2667 {
2668         char *end = NULL;
2669         unsigned long pm;
2670
2671         /* parse hexadecimal string */
2672         pm = strtoul(portmask, &end, 16);
2673         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
2674                 return -1;
2675
2676         if (pm == 0)
2677                 return -1;
2678
2679         return pm;
2680 }
2681
2682 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2683 static int
2684 parse_hash_entry_number(const char *hash_entry_num)
2685 {
2686         char *end = NULL;
2687         unsigned long hash_en;
2688
2689         /* parse hexadecimal string */
2690         hash_en = strtoul(hash_entry_num, &end, 16);
2691         if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
2692                 return -1;
2693
2694         if (hash_en == 0)
2695                 return -1;
2696
2697         return hash_en;
2698 }
2699 #endif
2700
2701 static int
2702 parse_rx_config(const char *q_arg)
2703 {
2704         char s[256];
2705         const char *p, *p0 = q_arg;
2706         char *end;
2707         enum fieldnames {
2708                 FLD_PORT = 0,
2709                 FLD_QUEUE,
2710                 FLD_LCORE,
2711                 FLD_THREAD,
2712                 _NUM_FLD
2713         };
2714         unsigned long int_fld[_NUM_FLD];
2715         char *str_fld[_NUM_FLD];
2716         int i;
2717         unsigned size;
2718
2719         nb_rx_thread_params = 0;
2720
2721         while ((p = strchr(p0, '(')) != NULL) {
2722                 ++p;
2723                 p0 = strchr(p, ')');
2724                 if (p0 == NULL)
2725                         return -1;
2726
2727                 size = p0 - p;
2728                 if (size >= sizeof(s))
2729                         return -1;
2730
2731                 snprintf(s, sizeof(s), "%.*s", size, p);
2732                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2733                         return -1;
2734                 for (i = 0; i < _NUM_FLD; i++) {
2735                         errno = 0;
2736                         int_fld[i] = strtoul(str_fld[i], &end, 0);
2737                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2738                                 return -1;
2739                 }
2740                 if (nb_rx_thread_params >= MAX_LCORE_PARAMS) {
2741                         printf("exceeded max number of rx params: %hu\n",
2742                                         nb_rx_thread_params);
2743                         return -1;
2744                 }
2745                 rx_thread_params_array[nb_rx_thread_params].port_id =
2746                                 int_fld[FLD_PORT];
2747                 rx_thread_params_array[nb_rx_thread_params].queue_id =
2748                                 (uint8_t)int_fld[FLD_QUEUE];
2749                 rx_thread_params_array[nb_rx_thread_params].lcore_id =
2750                                 (uint8_t)int_fld[FLD_LCORE];
2751                 rx_thread_params_array[nb_rx_thread_params].thread_id =
2752                                 (uint8_t)int_fld[FLD_THREAD];
2753                 ++nb_rx_thread_params;
2754         }
2755         rx_thread_params = rx_thread_params_array;
2756         return 0;
2757 }
2758
2759 static int
2760 parse_tx_config(const char *q_arg)
2761 {
2762         char s[256];
2763         const char *p, *p0 = q_arg;
2764         char *end;
2765         enum fieldnames {
2766                 FLD_LCORE = 0,
2767                 FLD_THREAD,
2768                 _NUM_FLD
2769         };
2770         unsigned long int_fld[_NUM_FLD];
2771         char *str_fld[_NUM_FLD];
2772         int i;
2773         unsigned size;
2774
2775         nb_tx_thread_params = 0;
2776
2777         while ((p = strchr(p0, '(')) != NULL) {
2778                 ++p;
2779                 p0 = strchr(p, ')');
2780                 if (p0 == NULL)
2781                         return -1;
2782
2783                 size = p0 - p;
2784                 if (size >= sizeof(s))
2785                         return -1;
2786
2787                 snprintf(s, sizeof(s), "%.*s", size, p);
2788                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2789                         return -1;
2790                 for (i = 0; i < _NUM_FLD; i++) {
2791                         errno = 0;
2792                         int_fld[i] = strtoul(str_fld[i], &end, 0);
2793                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2794                                 return -1;
2795                 }
2796                 if (nb_tx_thread_params >= MAX_LCORE_PARAMS) {
2797                         printf("exceeded max number of tx params: %hu\n",
2798                                 nb_tx_thread_params);
2799                         return -1;
2800                 }
2801                 tx_thread_params_array[nb_tx_thread_params].lcore_id =
2802                                 (uint8_t)int_fld[FLD_LCORE];
2803                 tx_thread_params_array[nb_tx_thread_params].thread_id =
2804                                 (uint8_t)int_fld[FLD_THREAD];
2805                 ++nb_tx_thread_params;
2806         }
2807         tx_thread_params = tx_thread_params_array;
2808
2809         return 0;
2810 }
2811
2812 #if (APP_CPU_LOAD > 0)
2813 static int
2814 parse_stat_lcore(const char *stat_lcore)
2815 {
2816         char *end = NULL;
2817         unsigned long lcore_id;
2818
2819         lcore_id = strtoul(stat_lcore, &end, 10);
2820         if ((stat_lcore[0] == '\0') || (end == NULL) || (*end != '\0'))
2821                 return -1;
2822
2823         return lcore_id;
2824 }
2825 #endif
2826
2827 static void
2828 parse_eth_dest(const char *optarg)
2829 {
2830         uint16_t portid;
2831         char *port_end;
2832         uint8_t c, *dest, peer_addr[6];
2833
2834         errno = 0;
2835         portid = strtoul(optarg, &port_end, 10);
2836         if (errno != 0 || port_end == optarg || *port_end++ != ',')
2837                 rte_exit(EXIT_FAILURE,
2838                 "Invalid eth-dest: %s", optarg);
2839         if (portid >= RTE_MAX_ETHPORTS)
2840                 rte_exit(EXIT_FAILURE,
2841                 "eth-dest: port %d >= RTE_MAX_ETHPORTS(%d)\n",
2842                 portid, RTE_MAX_ETHPORTS);
2843
2844         if (cmdline_parse_etheraddr(NULL, port_end,
2845                 &peer_addr, sizeof(peer_addr)) < 0)
2846                 rte_exit(EXIT_FAILURE,
2847                 "Invalid ethernet address: %s\n",
2848                 port_end);
2849         dest = (uint8_t *)&dest_eth_addr[portid];
2850         for (c = 0; c < 6; c++)
2851                 dest[c] = peer_addr[c];
2852         *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
2853 }
2854
2855 #define CMD_LINE_OPT_RX_CONFIG "rx"
2856 #define CMD_LINE_OPT_TX_CONFIG "tx"
2857 #define CMD_LINE_OPT_STAT_LCORE "stat-lcore"
2858 #define CMD_LINE_OPT_ETH_DEST "eth-dest"
2859 #define CMD_LINE_OPT_NO_NUMA "no-numa"
2860 #define CMD_LINE_OPT_IPV6 "ipv6"
2861 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
2862 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
2863 #define CMD_LINE_OPT_NO_LTHREADS "no-lthreads"
2864 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
2865
2866 /* Parse the argument given in the command line of the application */
2867 static int
2868 parse_args(int argc, char **argv)
2869 {
2870         int opt, ret;
2871         char **argvopt;
2872         int option_index;
2873         char *prgname = argv[0];
2874         static struct option lgopts[] = {
2875                 {CMD_LINE_OPT_RX_CONFIG, 1, 0, 0},
2876                 {CMD_LINE_OPT_TX_CONFIG, 1, 0, 0},
2877                 {CMD_LINE_OPT_STAT_LCORE, 1, 0, 0},
2878                 {CMD_LINE_OPT_ETH_DEST, 1, 0, 0},
2879                 {CMD_LINE_OPT_NO_NUMA, 0, 0, 0},
2880                 {CMD_LINE_OPT_IPV6, 0, 0, 0},
2881                 {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, 0},
2882                 {CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, 0},
2883                 {CMD_LINE_OPT_NO_LTHREADS, 0, 0, 0},
2884                 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
2885                 {NULL, 0, 0, 0}
2886         };
2887
2888         argvopt = argv;
2889
2890         while ((opt = getopt_long(argc, argvopt, "p:P",
2891                                 lgopts, &option_index)) != EOF) {
2892
2893                 switch (opt) {
2894                 /* portmask */
2895                 case 'p':
2896                         enabled_port_mask = parse_portmask(optarg);
2897                         if (enabled_port_mask == 0) {
2898                                 printf("invalid portmask\n");
2899                                 print_usage(prgname);
2900                                 return -1;
2901                         }
2902                         break;
2903                 case 'P':
2904                         printf("Promiscuous mode selected\n");
2905                         promiscuous_on = 1;
2906                         break;
2907
2908                 /* long options */
2909                 case 0:
2910                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_RX_CONFIG,
2911                                 sizeof(CMD_LINE_OPT_RX_CONFIG))) {
2912                                 ret = parse_rx_config(optarg);
2913                                 if (ret) {
2914                                         printf("invalid rx-config\n");
2915                                         print_usage(prgname);
2916                                         return -1;
2917                                 }
2918                         }
2919
2920                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_TX_CONFIG,
2921                                 sizeof(CMD_LINE_OPT_TX_CONFIG))) {
2922                                 ret = parse_tx_config(optarg);
2923                                 if (ret) {
2924                                         printf("invalid tx-config\n");
2925                                         print_usage(prgname);
2926                                         return -1;
2927                                 }
2928                         }
2929
2930 #if (APP_CPU_LOAD > 0)
2931                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_STAT_LCORE,
2932                                         sizeof(CMD_LINE_OPT_STAT_LCORE))) {
2933                                 cpu_load_lcore_id = parse_stat_lcore(optarg);
2934                         }
2935 #endif
2936
2937                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ETH_DEST,
2938                                 sizeof(CMD_LINE_OPT_ETH_DEST)))
2939                                         parse_eth_dest(optarg);
2940
2941                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_NUMA,
2942                                 sizeof(CMD_LINE_OPT_NO_NUMA))) {
2943                                 printf("numa is disabled\n");
2944                                 numa_on = 0;
2945                         }
2946
2947 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2948                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_IPV6,
2949                                 sizeof(CMD_LINE_OPT_IPV6))) {
2950                                 printf("ipv6 is specified\n");
2951                                 ipv6 = 1;
2952                         }
2953 #endif
2954
2955                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_LTHREADS,
2956                                         sizeof(CMD_LINE_OPT_NO_LTHREADS))) {
2957                                 printf("l-threads model is disabled\n");
2958                                 lthreads_on = 0;
2959                         }
2960
2961                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_PARSE_PTYPE,
2962                                         sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
2963                                 printf("software packet type parsing enabled\n");
2964                                 parse_ptype_on = 1;
2965                         }
2966
2967                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ENABLE_JUMBO,
2968                                 sizeof(CMD_LINE_OPT_ENABLE_JUMBO))) {
2969                                 struct option lenopts = {"max-pkt-len", required_argument, 0,
2970                                                 0};
2971
2972                                 printf("jumbo frame is enabled - disabling simple TX path\n");
2973                                 port_conf.rxmode.offloads |=
2974                                                 DEV_RX_OFFLOAD_JUMBO_FRAME;
2975                                 port_conf.txmode.offloads |=
2976                                                 DEV_TX_OFFLOAD_MULTI_SEGS;
2977
2978                                 /* if no max-pkt-len set, use the default value ETHER_MAX_LEN */
2979                                 if (0 == getopt_long(argc, argvopt, "", &lenopts,
2980                                                 &option_index)) {
2981
2982                                         ret = parse_max_pkt_len(optarg);
2983                                         if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)) {
2984                                                 printf("invalid packet length\n");
2985                                                 print_usage(prgname);
2986                                                 return -1;
2987                                         }
2988                                         port_conf.rxmode.max_rx_pkt_len = ret;
2989                                 }
2990                                 printf("set jumbo frame max packet length to %u\n",
2991                                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
2992                         }
2993 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2994                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_HASH_ENTRY_NUM,
2995                                 sizeof(CMD_LINE_OPT_HASH_ENTRY_NUM))) {
2996                                 ret = parse_hash_entry_number(optarg);
2997                                 if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
2998                                         hash_entry_number = ret;
2999                                 } else {
3000                                         printf("invalid hash entry number\n");
3001                                         print_usage(prgname);
3002                                         return -1;
3003                                 }
3004                         }
3005 #endif
3006                         break;
3007
3008                 default:
3009                         print_usage(prgname);
3010                         return -1;
3011                 }
3012         }
3013
3014         if (optind >= 0)
3015                 argv[optind-1] = prgname;
3016
3017         ret = optind-1;
3018         optind = 1; /* reset getopt lib */
3019         return ret;
3020 }
3021
3022 static void
3023 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
3024 {
3025         char buf[ETHER_ADDR_FMT_SIZE];
3026
3027         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
3028         printf("%s%s", name, buf);
3029 }
3030
3031 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
3032
3033 static void convert_ipv4_5tuple(struct ipv4_5tuple *key1,
3034                 union ipv4_5tuple_host *key2)
3035 {
3036         key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
3037         key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
3038         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3039         key2->port_src = rte_cpu_to_be_16(key1->port_src);
3040         key2->proto = key1->proto;
3041         key2->pad0 = 0;
3042         key2->pad1 = 0;
3043 }
3044
3045 static void convert_ipv6_5tuple(struct ipv6_5tuple *key1,
3046                 union ipv6_5tuple_host *key2)
3047 {
3048         uint32_t i;
3049
3050         for (i = 0; i < 16; i++) {
3051                 key2->ip_dst[i] = key1->ip_dst[i];
3052                 key2->ip_src[i] = key1->ip_src[i];
3053         }
3054         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3055         key2->port_src = rte_cpu_to_be_16(key1->port_src);
3056         key2->proto = key1->proto;
3057         key2->pad0 = 0;
3058         key2->pad1 = 0;
3059         key2->reserve = 0;
3060 }
3061
3062 #define BYTE_VALUE_MAX 256
3063 #define ALL_32_BITS 0xffffffff
3064 #define BIT_8_TO_15 0x0000ff00
3065 static inline void
3066 populate_ipv4_few_flow_into_table(const struct rte_hash *h)
3067 {
3068         uint32_t i;
3069         int32_t ret;
3070         uint32_t array_len = RTE_DIM(ipv4_l3fwd_route_array);
3071
3072         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3073         for (i = 0; i < array_len; i++) {
3074                 struct ipv4_l3fwd_route  entry;
3075                 union ipv4_5tuple_host newkey;
3076
3077                 entry = ipv4_l3fwd_route_array[i];
3078                 convert_ipv4_5tuple(&entry.key, &newkey);
3079                 ret = rte_hash_add_key(h, (void *)&newkey);
3080                 if (ret < 0) {
3081                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3082                                 " to the l3fwd hash.\n", i);
3083                 }
3084                 ipv4_l3fwd_out_if[ret] = entry.if_out;
3085         }
3086         printf("Hash: Adding 0x%" PRIx32 " keys\n", array_len);
3087 }
3088
3089 #define BIT_16_TO_23 0x00ff0000
3090 static inline void
3091 populate_ipv6_few_flow_into_table(const struct rte_hash *h)
3092 {
3093         uint32_t i;
3094         int32_t ret;
3095         uint32_t array_len = RTE_DIM(ipv6_l3fwd_route_array);
3096
3097         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3098         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3099         for (i = 0; i < array_len; i++) {
3100                 struct ipv6_l3fwd_route entry;
3101                 union ipv6_5tuple_host newkey;
3102
3103                 entry = ipv6_l3fwd_route_array[i];
3104                 convert_ipv6_5tuple(&entry.key, &newkey);
3105                 ret = rte_hash_add_key(h, (void *)&newkey);
3106                 if (ret < 0) {
3107                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3108                                 " to the l3fwd hash.\n", i);
3109                 }
3110                 ipv6_l3fwd_out_if[ret] = entry.if_out;
3111         }
3112         printf("Hash: Adding 0x%" PRIx32 "keys\n", array_len);
3113 }
3114
3115 #define NUMBER_PORT_USED 4
3116 static inline void
3117 populate_ipv4_many_flow_into_table(const struct rte_hash *h,
3118                 unsigned int nr_flow)
3119 {
3120         unsigned i;
3121
3122         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3123
3124         for (i = 0; i < nr_flow; i++) {
3125                 struct ipv4_l3fwd_route entry;
3126                 union ipv4_5tuple_host newkey;
3127                 uint8_t a = (uint8_t)((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3128                 uint8_t b = (uint8_t)(((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3129                                 BYTE_VALUE_MAX);
3130                 uint8_t c = (uint8_t)((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3131                                 BYTE_VALUE_MAX));
3132                 /* Create the ipv4 exact match flow */
3133                 memset(&entry, 0, sizeof(entry));
3134                 switch (i & (NUMBER_PORT_USED - 1)) {
3135                 case 0:
3136                         entry = ipv4_l3fwd_route_array[0];
3137                         entry.key.ip_dst = IPv4(101, c, b, a);
3138                         break;
3139                 case 1:
3140                         entry = ipv4_l3fwd_route_array[1];
3141                         entry.key.ip_dst = IPv4(201, c, b, a);
3142                         break;
3143                 case 2:
3144                         entry = ipv4_l3fwd_route_array[2];
3145                         entry.key.ip_dst = IPv4(111, c, b, a);
3146                         break;
3147                 case 3:
3148                         entry = ipv4_l3fwd_route_array[3];
3149                         entry.key.ip_dst = IPv4(211, c, b, a);
3150                         break;
3151                 };
3152                 convert_ipv4_5tuple(&entry.key, &newkey);
3153                 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3154
3155                 if (ret < 0)
3156                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3157
3158                 ipv4_l3fwd_out_if[ret] = (uint8_t)entry.if_out;
3159
3160         }
3161         printf("Hash: Adding 0x%x keys\n", nr_flow);
3162 }
3163
3164 static inline void
3165 populate_ipv6_many_flow_into_table(const struct rte_hash *h,
3166                 unsigned int nr_flow)
3167 {
3168         unsigned i;
3169
3170         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3171         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3172         for (i = 0; i < nr_flow; i++) {
3173                 struct ipv6_l3fwd_route entry;
3174                 union ipv6_5tuple_host newkey;
3175
3176                 uint8_t a = (uint8_t) ((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3177                 uint8_t b = (uint8_t) (((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3178                                 BYTE_VALUE_MAX);
3179                 uint8_t c = (uint8_t) ((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3180                                 BYTE_VALUE_MAX));
3181
3182                 /* Create the ipv6 exact match flow */
3183                 memset(&entry, 0, sizeof(entry));
3184                 switch (i & (NUMBER_PORT_USED - 1)) {
3185                 case 0:
3186                         entry = ipv6_l3fwd_route_array[0];
3187                         break;
3188                 case 1:
3189                         entry = ipv6_l3fwd_route_array[1];
3190                         break;
3191                 case 2:
3192                         entry = ipv6_l3fwd_route_array[2];
3193                         break;
3194                 case 3:
3195                         entry = ipv6_l3fwd_route_array[3];
3196                         break;
3197                 };
3198                 entry.key.ip_dst[13] = c;
3199                 entry.key.ip_dst[14] = b;
3200                 entry.key.ip_dst[15] = a;
3201                 convert_ipv6_5tuple(&entry.key, &newkey);
3202                 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3203
3204                 if (ret < 0)
3205                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3206
3207                 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
3208
3209         }
3210         printf("Hash: Adding 0x%x keys\n", nr_flow);
3211 }
3212
3213 static void
3214 setup_hash(int socketid)
3215 {
3216         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
3217                 .name = NULL,
3218                 .entries = L3FWD_HASH_ENTRIES,
3219                 .key_len = sizeof(union ipv4_5tuple_host),
3220                 .hash_func = ipv4_hash_crc,
3221                 .hash_func_init_val = 0,
3222         };
3223
3224         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
3225                 .name = NULL,
3226                 .entries = L3FWD_HASH_ENTRIES,
3227                 .key_len = sizeof(union ipv6_5tuple_host),
3228                 .hash_func = ipv6_hash_crc,
3229                 .hash_func_init_val = 0,
3230         };
3231
3232         char s[64];
3233
3234         /* create ipv4 hash */
3235         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
3236         ipv4_l3fwd_hash_params.name = s;
3237         ipv4_l3fwd_hash_params.socket_id = socketid;
3238         ipv4_l3fwd_lookup_struct[socketid] =
3239                         rte_hash_create(&ipv4_l3fwd_hash_params);
3240         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3241                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3242                                 "socket %d\n", socketid);
3243
3244         /* create ipv6 hash */
3245         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
3246         ipv6_l3fwd_hash_params.name = s;
3247         ipv6_l3fwd_hash_params.socket_id = socketid;
3248         ipv6_l3fwd_lookup_struct[socketid] =
3249                         rte_hash_create(&ipv6_l3fwd_hash_params);
3250         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3251                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3252                                 "socket %d\n", socketid);
3253
3254         if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
3255                 /* For testing hash matching with a large number of flows we
3256                  * generate millions of IP 5-tuples with an incremented dst
3257                  * address to initialize the hash table. */
3258                 if (ipv6 == 0) {
3259                         /* populate the ipv4 hash */
3260                         populate_ipv4_many_flow_into_table(
3261                                 ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
3262                 } else {
3263                         /* populate the ipv6 hash */
3264                         populate_ipv6_many_flow_into_table(
3265                                 ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
3266                 }
3267         } else {
3268                 /* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize
3269                  * the hash table */
3270                 if (ipv6 == 0) {
3271                         /* populate the ipv4 hash */
3272                         populate_ipv4_few_flow_into_table(
3273                                         ipv4_l3fwd_lookup_struct[socketid]);
3274                 } else {
3275                         /* populate the ipv6 hash */
3276                         populate_ipv6_few_flow_into_table(
3277                                         ipv6_l3fwd_lookup_struct[socketid]);
3278                 }
3279         }
3280 }
3281 #endif
3282
3283 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3284 static void
3285 setup_lpm(int socketid)
3286 {
3287         struct rte_lpm6_config config;
3288         struct rte_lpm_config lpm_ipv4_config;
3289         unsigned i;
3290         int ret;
3291         char s[64];
3292
3293         /* create the LPM table */
3294         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
3295         lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
3296         lpm_ipv4_config.number_tbl8s = 256;
3297         lpm_ipv4_config.flags = 0;
3298         ipv4_l3fwd_lookup_struct[socketid] =
3299                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
3300         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3301                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3302                                 " on socket %d\n", socketid);
3303
3304         /* populate the LPM table */
3305         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
3306
3307                 /* skip unused ports */
3308                 if ((1 << ipv4_l3fwd_route_array[i].if_out &
3309                                 enabled_port_mask) == 0)
3310                         continue;
3311
3312                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
3313                         ipv4_l3fwd_route_array[i].ip,
3314                         ipv4_l3fwd_route_array[i].depth,
3315                         ipv4_l3fwd_route_array[i].if_out);
3316
3317                 if (ret < 0) {
3318                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3319                                 "l3fwd LPM table on socket %d\n",
3320                                 i, socketid);
3321                 }
3322
3323                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
3324                         (unsigned)ipv4_l3fwd_route_array[i].ip,
3325                         ipv4_l3fwd_route_array[i].depth,
3326                         ipv4_l3fwd_route_array[i].if_out);
3327         }
3328
3329         /* create the LPM6 table */
3330         snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
3331
3332         config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
3333         config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
3334         config.flags = 0;
3335         ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
3336                                 &config);
3337         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3338                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3339                                 " on socket %d\n", socketid);
3340
3341         /* populate the LPM table */
3342         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
3343
3344                 /* skip unused ports */
3345                 if ((1 << ipv6_l3fwd_route_array[i].if_out &
3346                                 enabled_port_mask) == 0)
3347                         continue;
3348
3349                 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
3350                         ipv6_l3fwd_route_array[i].ip,
3351                         ipv6_l3fwd_route_array[i].depth,
3352                         ipv6_l3fwd_route_array[i].if_out);
3353
3354                 if (ret < 0) {
3355                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3356                                 "l3fwd LPM table on socket %d\n",
3357                                 i, socketid);
3358                 }
3359
3360                 printf("LPM: Adding route %s / %d (%d)\n",
3361                         "IPV6",
3362                         ipv6_l3fwd_route_array[i].depth,
3363                         ipv6_l3fwd_route_array[i].if_out);
3364         }
3365 }
3366 #endif
3367
3368 static int
3369 init_mem(unsigned nb_mbuf)
3370 {
3371         struct lcore_conf *qconf;
3372         int socketid;
3373         unsigned lcore_id;
3374         char s[64];
3375
3376         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3377                 if (rte_lcore_is_enabled(lcore_id) == 0)
3378                         continue;
3379
3380                 if (numa_on)
3381                         socketid = rte_lcore_to_socket_id(lcore_id);
3382                 else
3383                         socketid = 0;
3384
3385                 if (socketid >= NB_SOCKETS) {
3386                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
3387                                 socketid, lcore_id, NB_SOCKETS);
3388                 }
3389                 if (pktmbuf_pool[socketid] == NULL) {
3390                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
3391                         pktmbuf_pool[socketid] =
3392                                 rte_pktmbuf_pool_create(s, nb_mbuf,
3393                                         MEMPOOL_CACHE_SIZE, 0,
3394                                         RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
3395                         if (pktmbuf_pool[socketid] == NULL)
3396                                 rte_exit(EXIT_FAILURE,
3397                                                 "Cannot init mbuf pool on socket %d\n", socketid);
3398                         else
3399                                 printf("Allocated mbuf pool on socket %d\n", socketid);
3400
3401 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3402                         setup_lpm(socketid);
3403 #else
3404                         setup_hash(socketid);
3405 #endif
3406                 }
3407                 qconf = &lcore_conf[lcore_id];
3408                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
3409                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
3410         }
3411         return 0;
3412 }
3413
3414 /* Check the link status of all ports in up to 9s, and print them finally */
3415 static void
3416 check_all_ports_link_status(uint32_t port_mask)
3417 {
3418 #define CHECK_INTERVAL 100 /* 100ms */
3419 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
3420         uint16_t portid;
3421         uint8_t count, all_ports_up, print_flag = 0;
3422         struct rte_eth_link link;
3423
3424         printf("\nChecking link status");
3425         fflush(stdout);
3426         for (count = 0; count <= MAX_CHECK_TIME; count++) {
3427                 all_ports_up = 1;
3428                 RTE_ETH_FOREACH_DEV(portid) {
3429                         if ((port_mask & (1 << portid)) == 0)
3430                                 continue;
3431                         memset(&link, 0, sizeof(link));
3432                         rte_eth_link_get_nowait(portid, &link);
3433                         /* print link status if flag set */
3434                         if (print_flag == 1) {
3435                                 if (link.link_status)
3436                                         printf(
3437                                         "Port%d Link Up. Speed %u Mbps - %s\n",
3438                                                 portid, link.link_speed,
3439                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
3440                                         ("full-duplex") : ("half-duplex\n"));
3441                                 else
3442                                         printf("Port %d Link Down\n", portid);
3443                                 continue;
3444                         }
3445                         /* clear all_ports_up flag if any link down */
3446                         if (link.link_status == ETH_LINK_DOWN) {
3447                                 all_ports_up = 0;
3448                                 break;
3449                         }
3450                 }
3451                 /* after finally printing all link status, get out */
3452                 if (print_flag == 1)
3453                         break;
3454
3455                 if (all_ports_up == 0) {
3456                         printf(".");
3457                         fflush(stdout);
3458                         rte_delay_ms(CHECK_INTERVAL);
3459                 }
3460
3461                 /* set the print_flag if all ports up or timeout */
3462                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
3463                         print_flag = 1;
3464                         printf("done\n");
3465                 }
3466         }
3467 }
3468
3469 int
3470 main(int argc, char **argv)
3471 {
3472         struct rte_eth_dev_info dev_info;
3473         struct rte_eth_txconf *txconf;
3474         int ret;
3475         int i;
3476         unsigned nb_ports;
3477         uint16_t queueid, portid;
3478         unsigned lcore_id;
3479         uint32_t n_tx_queue, nb_lcores;
3480         uint8_t nb_rx_queue, queue, socketid;
3481
3482         /* init EAL */
3483         ret = rte_eal_init(argc, argv);
3484         if (ret < 0)
3485                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
3486         argc -= ret;
3487         argv += ret;
3488
3489         /* pre-init dst MACs for all ports to 02:00:00:00:00:xx */
3490         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
3491                 dest_eth_addr[portid] = ETHER_LOCAL_ADMIN_ADDR +
3492                                 ((uint64_t)portid << 40);
3493                 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
3494         }
3495
3496         /* parse application arguments (after the EAL ones) */
3497         ret = parse_args(argc, argv);
3498         if (ret < 0)
3499                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
3500
3501         if (check_lcore_params() < 0)
3502                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
3503
3504         printf("Initializing rx-queues...\n");
3505         ret = init_rx_queues();
3506         if (ret < 0)
3507                 rte_exit(EXIT_FAILURE, "init_rx_queues failed\n");
3508
3509         printf("Initializing tx-threads...\n");
3510         ret = init_tx_threads();
3511         if (ret < 0)
3512                 rte_exit(EXIT_FAILURE, "init_tx_threads failed\n");
3513
3514         printf("Initializing rings...\n");
3515         ret = init_rx_rings();
3516         if (ret < 0)
3517                 rte_exit(EXIT_FAILURE, "init_rx_rings failed\n");
3518
3519         nb_ports = rte_eth_dev_count_avail();
3520
3521         if (check_port_config() < 0)
3522                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
3523
3524         nb_lcores = rte_lcore_count();
3525
3526         /* initialize all ports */
3527         RTE_ETH_FOREACH_DEV(portid) {
3528                 struct rte_eth_conf local_port_conf = port_conf;
3529
3530                 /* skip ports that are not enabled */
3531                 if ((enabled_port_mask & (1 << portid)) == 0) {
3532                         printf("\nSkipping disabled port %d\n", portid);
3533                         continue;
3534                 }
3535
3536                 /* init port */
3537                 printf("Initializing port %d ... ", portid);
3538                 fflush(stdout);
3539
3540                 nb_rx_queue = get_port_n_rx_queues(portid);
3541                 n_tx_queue = nb_lcores;
3542                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
3543                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
3544                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
3545                         nb_rx_queue, (unsigned)n_tx_queue);
3546                 rte_eth_dev_info_get(portid, &dev_info);
3547                 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
3548                         local_port_conf.txmode.offloads |=
3549                                 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
3550
3551                 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
3552                         dev_info.flow_type_rss_offloads;
3553                 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
3554                                 port_conf.rx_adv_conf.rss_conf.rss_hf) {
3555                         printf("Port %u modified RSS hash function based on hardware support,"
3556                                 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
3557                                 portid,
3558                                 port_conf.rx_adv_conf.rss_conf.rss_hf,
3559                                 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
3560                 }
3561
3562                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
3563                                         (uint16_t)n_tx_queue, &local_port_conf);
3564                 if (ret < 0)
3565                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
3566                                 ret, portid);
3567
3568                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
3569                                                        &nb_txd);
3570                 if (ret < 0)
3571                         rte_exit(EXIT_FAILURE,
3572                                  "rte_eth_dev_adjust_nb_rx_tx_desc: err=%d, port=%d\n",
3573                                  ret, portid);
3574
3575                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
3576                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
3577                 printf(", ");
3578                 print_ethaddr("Destination:",
3579                         (const struct ether_addr *)&dest_eth_addr[portid]);
3580                 printf(", ");
3581
3582                 /*
3583                  * prepare src MACs for each port.
3584                  */
3585                 ether_addr_copy(&ports_eth_addr[portid],
3586                         (struct ether_addr *)(val_eth + portid) + 1);
3587
3588                 /* init memory */
3589                 ret = init_mem(NB_MBUF);
3590                 if (ret < 0)
3591                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
3592
3593                 /* init one TX queue per couple (lcore,port) */
3594                 queueid = 0;
3595                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3596                         if (rte_lcore_is_enabled(lcore_id) == 0)
3597                                 continue;
3598
3599                         if (numa_on)
3600                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3601                         else
3602                                 socketid = 0;
3603
3604                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
3605                         fflush(stdout);
3606
3607                         txconf = &dev_info.default_txconf;
3608                         txconf->offloads = local_port_conf.txmode.offloads;
3609                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
3610                                                      socketid, txconf);
3611                         if (ret < 0)
3612                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
3613                                         "port=%d\n", ret, portid);
3614
3615                         tx_thread[lcore_id].tx_queue_id[portid] = queueid;
3616                         queueid++;
3617                 }
3618                 printf("\n");
3619         }
3620
3621         for (i = 0; i < n_rx_thread; i++) {
3622                 lcore_id = rx_thread[i].conf.lcore_id;
3623
3624                 if (rte_lcore_is_enabled(lcore_id) == 0) {
3625                         rte_exit(EXIT_FAILURE,
3626                                         "Cannot start Rx thread on lcore %u: lcore disabled\n",
3627                                         lcore_id
3628                                 );
3629                 }
3630
3631                 printf("\nInitializing rx queues for Rx thread %d on lcore %u ... ",
3632                                 i, lcore_id);
3633                 fflush(stdout);
3634
3635                 /* init RX queues */
3636                 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3637                         struct rte_eth_dev *dev;
3638                         struct rte_eth_conf *conf;
3639                         struct rte_eth_rxconf rxq_conf;
3640
3641                         portid = rx_thread[i].rx_queue_list[queue].port_id;
3642                         queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3643                         dev = &rte_eth_devices[portid];
3644                         conf = &dev->data->dev_conf;
3645
3646                         if (numa_on)
3647                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3648                         else
3649                                 socketid = 0;
3650
3651                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
3652                         fflush(stdout);
3653
3654                         rte_eth_dev_info_get(portid, &dev_info);
3655                         rxq_conf = dev_info.default_rxconf;
3656                         rxq_conf.offloads = conf->rxmode.offloads;
3657                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
3658                                         socketid,
3659                                         &rxq_conf,
3660                                         pktmbuf_pool[socketid]);
3661                         if (ret < 0)
3662                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, "
3663                                                 "port=%d\n", ret, portid);
3664                 }
3665         }
3666
3667         printf("\n");
3668
3669         /* start ports */
3670         RTE_ETH_FOREACH_DEV(portid) {
3671                 if ((enabled_port_mask & (1 << portid)) == 0)
3672                         continue;
3673
3674                 /* Start device */
3675                 ret = rte_eth_dev_start(portid);
3676                 if (ret < 0)
3677                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
3678                                 ret, portid);
3679
3680                 /*
3681                  * If enabled, put device in promiscuous mode.
3682                  * This allows IO forwarding mode to forward packets
3683                  * to itself through 2 cross-connected  ports of the
3684                  * target machine.
3685                  */
3686                 if (promiscuous_on)
3687                         rte_eth_promiscuous_enable(portid);
3688         }
3689
3690         for (i = 0; i < n_rx_thread; i++) {
3691                 lcore_id = rx_thread[i].conf.lcore_id;
3692                 if (rte_lcore_is_enabled(lcore_id) == 0)
3693                         continue;
3694
3695                 /* check if hw packet type is supported */
3696                 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3697                         portid = rx_thread[i].rx_queue_list[queue].port_id;
3698                         queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3699
3700                         if (parse_ptype_on) {
3701                                 if (!rte_eth_add_rx_callback(portid, queueid,
3702                                                 cb_parse_ptype, NULL))
3703                                         rte_exit(EXIT_FAILURE,
3704                                                 "Failed to add rx callback: "
3705                                                 "port=%d\n", portid);
3706                         } else if (!check_ptype(portid))
3707                                 rte_exit(EXIT_FAILURE,
3708                                         "Port %d cannot parse packet type.\n\n"
3709                                         "Please add --parse-ptype to use sw "
3710                                         "packet type analyzer.\n\n",
3711                                         portid);
3712                 }
3713         }
3714
3715         check_all_ports_link_status(enabled_port_mask);
3716
3717         if (lthreads_on) {
3718                 printf("Starting L-Threading Model\n");
3719
3720 #if (APP_CPU_LOAD > 0)
3721                 if (cpu_load_lcore_id > 0)
3722                         /* Use one lcore for cpu load collector */
3723                         nb_lcores--;
3724 #endif
3725
3726                 lthread_num_schedulers_set(nb_lcores);
3727                 rte_eal_mp_remote_launch(sched_spawner, NULL, SKIP_MASTER);
3728                 lthread_master_spawner(NULL);
3729
3730         } else {
3731                 printf("Starting P-Threading Model\n");
3732                 /* launch per-lcore init on every lcore */
3733                 rte_eal_mp_remote_launch(pthread_run, NULL, CALL_MASTER);
3734                 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
3735                         if (rte_eal_wait_lcore(lcore_id) < 0)
3736                                 return -1;
3737                 }
3738         }
3739
3740         return 0;
3741 }