examples/l3fwd/main.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <stdio.h>
  35 #include <stdlib.h>
  36 #include <stdint.h>
  37 #include <inttypes.h>
  38 #include <sys/types.h>
  39 #include <string.h>
  40 #include <sys/queue.h>
  41 #include <stdarg.h>
  42 #include <errno.h>
  43 #include <getopt.h>
  44
  45 #include <rte_common.h>
  46 #include <rte_common_vect.h>
  47 #include <rte_byteorder.h>
  48 #include <rte_log.h>
  49 #include <rte_memory.h>
  50 #include <rte_memcpy.h>
  51 #include <rte_memzone.h>
  52 #include <rte_tailq.h>
  53 #include <rte_eal.h>
  54 #include <rte_per_lcore.h>
  55 #include <rte_launch.h>
  56 #include <rte_atomic.h>
  57 #include <rte_cycles.h>
  58 #include <rte_prefetch.h>
  59 #include <rte_lcore.h>
  60 #include <rte_per_lcore.h>
  61 #include <rte_branch_prediction.h>
  62 #include <rte_interrupts.h>
  63 #include <rte_pci.h>
  64 #include <rte_random.h>
  65 #include <rte_debug.h>
  66 #include <rte_ether.h>
  67 #include <rte_ethdev.h>
  68 #include <rte_ring.h>
  69 #include <rte_mempool.h>
  70 #include <rte_mbuf.h>
  71 #include <rte_ip.h>
  72 #include <rte_tcp.h>
  73 #include <rte_udp.h>
  74 #include <rte_string_fns.h>
  75
  76 #define APP_LOOKUP_EXACT_MATCH          0
  77 #define APP_LOOKUP_LPM                  1
  78 #define DO_RFC_1812_CHECKS
  79
  80 #ifndef APP_LOOKUP_METHOD
  81 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  82 #endif
  83
  84 /*
  85  *  When set to zero, simple forwaring path is eanbled.
  86  *  When set to one, optimized forwarding path is enabled.
  87  *  Note that LPM optimisation path uses SSE4.1 instructions.
  88  */
  89 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && !defined(__SSE4_1__))
  90 #define ENABLE_MULTI_BUFFER_OPTIMIZE    0
  91 #else
  92 #define ENABLE_MULTI_BUFFER_OPTIMIZE    1
  93 #endif
  94
  95 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  96 #include <rte_hash.h>
  97 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
  98 #include <rte_lpm.h>
  99 #include <rte_lpm6.h>
 100 #else
 101 #error "APP_LOOKUP_METHOD set to incorrect value"
 102 #endif
 103
 104 #ifndef IPv6_BYTES
 105 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 106                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 107 #define IPv6_BYTES(addr) \
 108         addr[0],  addr[1], addr[2],  addr[3], \
 109         addr[4],  addr[5], addr[6],  addr[7], \
 110         addr[8],  addr[9], addr[10], addr[11],\
 111         addr[12], addr[13],addr[14], addr[15]
 112 #endif
 113
 114
 115 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
 116
 117 #define MAX_JUMBO_PKT_LEN  9600
 118
 119 #define IPV6_ADDR_LEN 16
 120
 121 #define MEMPOOL_CACHE_SIZE 256
 122
 123 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
 124
 125 /*
 126  * This expression is used to calculate the number of mbufs needed depending on user input, taking
 127  *  into account memory for rx and tx hardware rings, cache per lcore and mtable per port per lcore.
 128  *  RTE_MAX is used to ensure that NB_MBUF never goes below a minimum value of 8192
 129  */
 130
 131 #define NB_MBUF RTE_MAX (                                                                                                                                       \
 132                                 (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT +                                                        \
 133                                 nb_ports*nb_lcores*MAX_PKT_BURST +                                                                                      \
 134                                 nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT +                                                          \
 135                                 nb_lcores*MEMPOOL_CACHE_SIZE),                                                                                          \
 136                                 (unsigned)8192)
 137
 138 #define MAX_PKT_BURST     32
 139 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 140
 141 /*
 142  * Try to avoid TX buffering if we have at least MAX_TX_BURST packets to send.
 143  */
 144 #define MAX_TX_BURST    (MAX_PKT_BURST / 2)
 145
 146 #define NB_SOCKETS 8
 147
 148 /* Configure how many packets ahead to prefetch, when reading packets */
 149 #define PREFETCH_OFFSET 3
 150
 151 /* Used to mark destination port as 'invalid'. */
 152 #define BAD_PORT        ((uint16_t)-1)
 153
 154 #define FWDSTEP 4
 155
 156 /*
 157  * Configurable number of RX/TX ring descriptors
 158  */
 159 #define RTE_TEST_RX_DESC_DEFAULT 128
 160 #define RTE_TEST_TX_DESC_DEFAULT 512
 161 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 162 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 163
 164 /* ethernet addresses of ports */
 165 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 166
 167 static __m128i val_eth[RTE_MAX_ETHPORTS];
 168
 169 /* replace first 12B of the ethernet header. */
 170 #define MASK_ETH        0x3f
 171
 172 /* mask of enabled ports */
 173 static uint32_t enabled_port_mask = 0;
 174 static int promiscuous_on = 0; /**< Ports set in promiscuous mode off by default. */
 175 static int numa_on = 1; /**< NUMA is enabled by default. */
 176
 177 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 178 static int ipv6 = 0; /**< ipv6 is false by default. */
 179 #endif
 180
 181 struct mbuf_table {
 182         uint16_t len;
 183         struct rte_mbuf *m_table[MAX_PKT_BURST];
 184 };
 185
 186 struct lcore_rx_queue {
 187         uint8_t port_id;
 188         uint8_t queue_id;
 189 } __rte_cache_aligned;
 190
 191 #define MAX_RX_QUEUE_PER_LCORE 16
 192 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 193 #define MAX_RX_QUEUE_PER_PORT 128
 194
 195 #define MAX_LCORE_PARAMS 1024
 196 struct lcore_params {
 197         uint8_t port_id;
 198         uint8_t queue_id;
 199         uint8_t lcore_id;
 200 } __rte_cache_aligned;
 201
 202 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 203 static struct lcore_params lcore_params_array_default[] = {
 204         {0, 0, 2},
 205         {0, 1, 2},
 206         {0, 2, 2},
 207         {1, 0, 2},
 208         {1, 1, 2},
 209         {1, 2, 2},
 210         {2, 0, 2},
 211         {3, 0, 3},
 212         {3, 1, 3},
 213 };
 214
 215 static struct lcore_params * lcore_params = lcore_params_array_default;
 216 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 217                                 sizeof(lcore_params_array_default[0]);
 218
 219 static struct rte_eth_conf port_conf = {
 220         .rxmode = {
 221                 .mq_mode = ETH_MQ_RX_RSS,
 222                 .max_rx_pkt_len = ETHER_MAX_LEN,
 223                 .split_hdr_size = 0,
 224                 .header_split   = 0, /**< Header Split disabled */
 225                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 226                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 227                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 228                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 229         },
 230         .rx_adv_conf = {
 231                 .rss_conf = {
 232                         .rss_key = NULL,
 233                         .rss_hf = ETH_RSS_IP,
 234                 },
 235         },
 236         .txmode = {
 237                 .mq_mode = ETH_MQ_TX_NONE,
 238         },
 239 };
 240
 241 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 242
 243 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 244
 245 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 246 #include <rte_hash_crc.h>
 247 #define DEFAULT_HASH_FUNC       rte_hash_crc
 248 #else
 249 #include <rte_jhash.h>
 250 #define DEFAULT_HASH_FUNC       rte_jhash
 251 #endif
 252
 253 struct ipv4_5tuple {
 254         uint32_t ip_dst;
 255         uint32_t ip_src;
 256         uint16_t port_dst;
 257         uint16_t port_src;
 258         uint8_t  proto;
 259 } __attribute__((__packed__));
 260
 261 union ipv4_5tuple_host {
 262         struct {
 263                 uint8_t  pad0;
 264                 uint8_t  proto;
 265                 uint16_t pad1;
 266                 uint32_t ip_src;
 267                 uint32_t ip_dst;
 268                 uint16_t port_src;
 269                 uint16_t port_dst;
 270         };
 271         __m128i xmm;
 272 };
 273
 274 #define XMM_NUM_IN_IPV6_5TUPLE 3
 275
 276 struct ipv6_5tuple {
 277         uint8_t  ip_dst[IPV6_ADDR_LEN];
 278         uint8_t  ip_src[IPV6_ADDR_LEN];
 279         uint16_t port_dst;
 280         uint16_t port_src;
 281         uint8_t  proto;
 282 } __attribute__((__packed__));
 283
 284 union ipv6_5tuple_host {
 285         struct {
 286                 uint16_t pad0;
 287                 uint8_t  proto;
 288                 uint8_t  pad1;
 289                 uint8_t  ip_src[IPV6_ADDR_LEN];
 290                 uint8_t  ip_dst[IPV6_ADDR_LEN];
 291                 uint16_t port_src;
 292                 uint16_t port_dst;
 293                 uint64_t reserve;
 294         };
 295         __m128i xmm[XMM_NUM_IN_IPV6_5TUPLE];
 296 };
 297
 298 struct ipv4_l3fwd_route {
 299         struct ipv4_5tuple key;
 300         uint8_t if_out;
 301 };
 302
 303 struct ipv6_l3fwd_route {
 304         struct ipv6_5tuple key;
 305         uint8_t if_out;
 306 };
 307
 308 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 309         {{IPv4(101,0,0,0), IPv4(100,10,0,1),  101, 11, IPPROTO_TCP}, 0},
 310         {{IPv4(201,0,0,0), IPv4(200,20,0,1),  102, 12, IPPROTO_TCP}, 1},
 311         {{IPv4(111,0,0,0), IPv4(100,30,0,1),  101, 11, IPPROTO_TCP}, 2},
 312         {{IPv4(211,0,0,0), IPv4(200,40,0,1),  102, 12, IPPROTO_TCP}, 3},
 313 };
 314
 315 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 316         {{
 317         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 318         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 319         101, 11, IPPROTO_TCP}, 0},
 320
 321         {{
 322         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 323         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 324         102, 12, IPPROTO_TCP}, 1},
 325
 326         {{
 327         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 328         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 329         101, 11, IPPROTO_TCP}, 2},
 330
 331         {{
 332         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 333         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 334         102, 12, IPPROTO_TCP}, 3},
 335 };
 336
 337 typedef struct rte_hash lookup_struct_t;
 338 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 339 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 340
 341 #ifdef RTE_ARCH_X86_64
 342 /* default to 4 million hash entries (approx) */
 343 #define L3FWD_HASH_ENTRIES              1024*1024*4
 344 #else
 345 /* 32-bit has less address-space for hugepage memory, limit to 1M entries */
 346 #define L3FWD_HASH_ENTRIES              1024*1024*1
 347 #endif
 348 #define HASH_ENTRY_NUMBER_DEFAULT       4
 349
 350 static uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
 351
 352 static inline uint32_t
 353 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
 354         uint32_t init_val)
 355 {
 356         const union ipv4_5tuple_host *k;
 357         uint32_t t;
 358         const uint32_t *p;
 359
 360         k = data;
 361         t = k->proto;
 362         p = (const uint32_t *)&k->port_src;
 363
 364 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 365         init_val = rte_hash_crc_4byte(t, init_val);
 366         init_val = rte_hash_crc_4byte(k->ip_src, init_val);
 367         init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
 368         init_val = rte_hash_crc_4byte(*p, init_val);
 369 #else /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 370         init_val = rte_jhash_1word(t, init_val);
 371         init_val = rte_jhash_1word(k->ip_src, init_val);
 372         init_val = rte_jhash_1word(k->ip_dst, init_val);
 373         init_val = rte_jhash_1word(*p, init_val);
 374 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 375         return (init_val);
 376 }
 377
 378 static inline uint32_t
 379 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len, uint32_t init_val)
 380 {
 381         const union ipv6_5tuple_host *k;
 382         uint32_t t;
 383         const uint32_t *p;
 384 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 385         const uint32_t  *ip_src0, *ip_src1, *ip_src2, *ip_src3;
 386         const uint32_t  *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
 387 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 388
 389         k = data;
 390         t = k->proto;
 391         p = (const uint32_t *)&k->port_src;
 392
 393 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 394         ip_src0 = (const uint32_t *) k->ip_src;
 395         ip_src1 = (const uint32_t *)(k->ip_src+4);
 396         ip_src2 = (const uint32_t *)(k->ip_src+8);
 397         ip_src3 = (const uint32_t *)(k->ip_src+12);
 398         ip_dst0 = (const uint32_t *) k->ip_dst;
 399         ip_dst1 = (const uint32_t *)(k->ip_dst+4);
 400         ip_dst2 = (const uint32_t *)(k->ip_dst+8);
 401         ip_dst3 = (const uint32_t *)(k->ip_dst+12);
 402         init_val = rte_hash_crc_4byte(t, init_val);
 403         init_val = rte_hash_crc_4byte(*ip_src0, init_val);
 404         init_val = rte_hash_crc_4byte(*ip_src1, init_val);
 405         init_val = rte_hash_crc_4byte(*ip_src2, init_val);
 406         init_val = rte_hash_crc_4byte(*ip_src3, init_val);
 407         init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
 408         init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
 409         init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
 410         init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
 411         init_val = rte_hash_crc_4byte(*p, init_val);
 412 #else /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 413         init_val = rte_jhash_1word(t, init_val);
 414         init_val = rte_jhash(k->ip_src, sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
 415         init_val = rte_jhash(k->ip_dst, sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
 416         init_val = rte_jhash_1word(*p, init_val);
 417 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 418         return (init_val);
 419 }
 420
 421 #define IPV4_L3FWD_NUM_ROUTES \
 422         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 423
 424 #define IPV6_L3FWD_NUM_ROUTES \
 425         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 426
 427 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 428 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 429
 430 #endif
 431
 432 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 433 struct ipv4_l3fwd_route {
 434         uint32_t ip;
 435         uint8_t  depth;
 436         uint8_t  if_out;
 437 };
 438
 439 struct ipv6_l3fwd_route {
 440         uint8_t ip[16];
 441         uint8_t  depth;
 442         uint8_t  if_out;
 443 };
 444
 445 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 446         {IPv4(1,1,1,0), 24, 0},
 447         {IPv4(2,1,1,0), 24, 1},
 448         {IPv4(3,1,1,0), 24, 2},
 449         {IPv4(4,1,1,0), 24, 3},
 450         {IPv4(5,1,1,0), 24, 4},
 451         {IPv4(6,1,1,0), 24, 5},
 452         {IPv4(7,1,1,0), 24, 6},
 453         {IPv4(8,1,1,0), 24, 7},
 454 };
 455
 456 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 457         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
 458         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
 459         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
 460         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
 461         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
 462         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
 463         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
 464         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
 465 };
 466
 467 #define IPV4_L3FWD_NUM_ROUTES \
 468         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 469 #define IPV6_L3FWD_NUM_ROUTES \
 470         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 471
 472 #define IPV4_L3FWD_LPM_MAX_RULES         1024
 473 #define IPV6_L3FWD_LPM_MAX_RULES         1024
 474 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
 475
 476 typedef struct rte_lpm lookup_struct_t;
 477 typedef struct rte_lpm6 lookup6_struct_t;
 478 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 479 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 480 #endif
 481
 482 struct lcore_conf {
 483         uint16_t n_rx_queue;
 484         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 485         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 486         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 487         lookup_struct_t * ipv4_lookup_struct;
 488 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 489         lookup6_struct_t * ipv6_lookup_struct;
 490 #else
 491         lookup_struct_t * ipv6_lookup_struct;
 492 #endif
 493 } __rte_cache_aligned;
 494
 495 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 496
 497 /* Send burst of packets on an output interface */
 498 static inline int
 499 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 500 {
 501         struct rte_mbuf **m_table;
 502         int ret;
 503         uint16_t queueid;
 504
 505         queueid = qconf->tx_queue_id[port];
 506         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 507
 508         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 509         if (unlikely(ret < n)) {
 510                 do {
 511                         rte_pktmbuf_free(m_table[ret]);
 512                 } while (++ret < n);
 513         }
 514
 515         return 0;
 516 }
 517
 518 /* Enqueue a single packet, and send burst if queue is filled */
 519 static inline int
 520 send_single_packet(struct rte_mbuf *m, uint8_t port)
 521 {
 522         uint32_t lcore_id;
 523         uint16_t len;
 524         struct lcore_conf *qconf;
 525
 526         lcore_id = rte_lcore_id();
 527
 528         qconf = &lcore_conf[lcore_id];
 529         len = qconf->tx_mbufs[port].len;
 530         qconf->tx_mbufs[port].m_table[len] = m;
 531         len++;
 532
 533         /* enough pkts to be sent */
 534         if (unlikely(len == MAX_PKT_BURST)) {
 535                 send_burst(qconf, MAX_PKT_BURST, port);
 536                 len = 0;
 537         }
 538
 539         qconf->tx_mbufs[port].len = len;
 540         return 0;
 541 }
 542
 543 static inline __attribute__((always_inline)) void
 544 send_packetsx4(struct lcore_conf *qconf, uint8_t port,
 545         struct rte_mbuf *m[], uint32_t num)
 546 {
 547         uint32_t len, j, n;
 548
 549         len = qconf->tx_mbufs[port].len;
 550
 551         /*
 552          * If TX buffer for that queue is empty, and we have enough packets,
 553          * then send them straightway.
 554          */
 555         if (num >= MAX_TX_BURST && len == 0) {
 556                 n = rte_eth_tx_burst(port, qconf->tx_queue_id[port], m, num);
 557                 if (unlikely(n < num)) {
 558                         do {
 559                                 rte_pktmbuf_free(m[n]);
 560                         } while (++n < num);
 561                 }
 562                 return;
 563         }
 564
 565         /*
 566          * Put packets into TX buffer for that queue.
 567          */
 568
 569         n = len + num;
 570         n = (n > MAX_PKT_BURST) ? MAX_PKT_BURST - len : num;
 571
 572         j = 0;
 573         switch (n % FWDSTEP) {
 574         while (j < n) {
 575         case 0:
 576                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 577                 j++;
 578         case 3:
 579                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 580                 j++;
 581         case 2:
 582                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 583                 j++;
 584         case 1:
 585                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 586                 j++;
 587         }
 588         }
 589
 590         len += n;
 591
 592         /* enough pkts to be sent */
 593         if (unlikely(len == MAX_PKT_BURST)) {
 594
 595                 send_burst(qconf, MAX_PKT_BURST, port);
 596
 597                 /* copy rest of the packets into the TX buffer. */
 598                 len = num - n;
 599                 j = 0;
 600                 switch (len % FWDSTEP) {
 601                 while (j < len) {
 602                 case 0:
 603                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 604                         j++;
 605                 case 3:
 606                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 607                         j++;
 608                 case 2:
 609                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 610                         j++;
 611                 case 1:
 612                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 613                         j++;
 614                 }
 615                 }
 616         }
 617
 618         qconf->tx_mbufs[port].len = len;
 619 }
 620
 621 #ifdef DO_RFC_1812_CHECKS
 622 static inline int
 623 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 624 {
 625         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 626         /*
 627          * 1. The packet length reported by the Link Layer must be large
 628          * enough to hold the minimum length legal IP datagram (20 bytes).
 629          */
 630         if (link_len < sizeof(struct ipv4_hdr))
 631                 return -1;
 632
 633         /* 2. The IP checksum must be correct. */
 634         /* this is checked in H/W */
 635
 636         /*
 637          * 3. The IP version number must be 4. If the version number is not 4
 638          * then the packet may be another version of IP, such as IPng or
 639          * ST-II.
 640          */
 641         if (((pkt->version_ihl) >> 4) != 4)
 642                 return -3;
 643         /*
 644          * 4. The IP header length field must be large enough to hold the
 645          * minimum length legal IP datagram (20 bytes = 5 words).
 646          */
 647         if ((pkt->version_ihl & 0xf) < 5)
 648                 return -4;
 649
 650         /*
 651          * 5. The IP total length field must be large enough to hold the IP
 652          * datagram header, whose length is specified in the IP header length
 653          * field.
 654          */
 655         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 656                 return -5;
 657
 658         return 0;
 659 }
 660 #endif
 661
 662 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 663
 664 static __m128i mask0;
 665 static __m128i mask1;
 666 static __m128i mask2;
 667 static inline uint8_t
 668 get_ipv4_dst_port(void *ipv4_hdr, uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
 669 {
 670         int ret = 0;
 671         union ipv4_5tuple_host key;
 672
 673         ipv4_hdr = (uint8_t *)ipv4_hdr + offsetof(struct ipv4_hdr, time_to_live);
 674         __m128i data = _mm_loadu_si128((__m128i*)(ipv4_hdr));
 675         /* Get 5 tuple: dst port, src port, dst IP address, src IP address and protocol */
 676         key.xmm = _mm_and_si128(data, mask0);
 677         /* Find destination port */
 678         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 679         return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
 680 }
 681
 682 static inline uint8_t
 683 get_ipv6_dst_port(void *ipv6_hdr,  uint8_t portid, lookup_struct_t * ipv6_l3fwd_lookup_struct)
 684 {
 685         int ret = 0;
 686         union ipv6_5tuple_host key;
 687
 688         ipv6_hdr = (uint8_t *)ipv6_hdr + offsetof(struct ipv6_hdr, payload_len);
 689         __m128i data0 = _mm_loadu_si128((__m128i*)(ipv6_hdr));
 690         __m128i data1 = _mm_loadu_si128((__m128i*)(((uint8_t*)ipv6_hdr)+sizeof(__m128i)));
 691         __m128i data2 = _mm_loadu_si128((__m128i*)(((uint8_t*)ipv6_hdr)+sizeof(__m128i)+sizeof(__m128i)));
 692         /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
 693         key.xmm[0] = _mm_and_si128(data0, mask1);
 694         /* Get part of 5 tuple: dst IP address lower 96 bits and src IP address higher 32 bits */
 695         key.xmm[1] = data1;
 696         /* Get part of 5 tuple: dst port and src port and dst IP address higher 32 bits */
 697         key.xmm[2] = _mm_and_si128(data2, mask2);
 698
 699         /* Find destination port */
 700         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 701         return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
 702 }
 703 #endif
 704
 705 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 706
 707 static inline uint8_t
 708 get_ipv4_dst_port(void *ipv4_hdr,  uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
 709 {
 710         uint8_t next_hop;
 711
 712         return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 713                 rte_be_to_cpu_32(((struct ipv4_hdr *)ipv4_hdr)->dst_addr),
 714                 &next_hop) == 0) ? next_hop : portid);
 715 }
 716
 717 static inline uint8_t
 718 get_ipv6_dst_port(void *ipv6_hdr,  uint8_t portid, lookup6_struct_t * ipv6_l3fwd_lookup_struct)
 719 {
 720         uint8_t next_hop;
 721         return (uint8_t) ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
 722                         ((struct ipv6_hdr*)ipv6_hdr)->dst_addr, &next_hop) == 0)?
 723                         next_hop : portid);
 724 }
 725 #endif
 726
 727 static inline void l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
 728         struct lcore_conf *qconf)  __attribute__((unused));
 729
 730 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) && \
 731         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
 732
 733 #define MASK_ALL_PKTS    0xf
 734 #define EXECLUDE_1ST_PKT 0xe
 735 #define EXECLUDE_2ND_PKT 0xd
 736 #define EXECLUDE_3RD_PKT 0xb
 737 #define EXECLUDE_4TH_PKT 0x7
 738
 739 static inline void
 740 simple_ipv4_fwd_4pkts(struct rte_mbuf* m[4], uint8_t portid, struct lcore_conf *qconf)
 741 {
 742         struct ether_hdr *eth_hdr[4];
 743         struct ipv4_hdr *ipv4_hdr[4];
 744         void *d_addr_bytes[4];
 745         uint8_t dst_port[4];
 746         int32_t ret[4];
 747         union ipv4_5tuple_host key[4];
 748         __m128i data[4];
 749
 750         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
 751         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
 752         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
 753         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
 754
 755         /* Handle IPv4 headers.*/
 756         ipv4_hdr[0] = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m[0], unsigned char *) +
 757                         sizeof(struct ether_hdr));
 758         ipv4_hdr[1] = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m[1], unsigned char *) +
 759                         sizeof(struct ether_hdr));
 760         ipv4_hdr[2] = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m[2], unsigned char *) +
 761                         sizeof(struct ether_hdr));
 762         ipv4_hdr[3] = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m[3], unsigned char *) +
 763                         sizeof(struct ether_hdr));
 764
 765 #ifdef DO_RFC_1812_CHECKS
 766         /* Check to make sure the packet is valid (RFC1812) */
 767         uint8_t valid_mask = MASK_ALL_PKTS;
 768         if (is_valid_ipv4_pkt(ipv4_hdr[0], m[0]->pkt_len) < 0) {
 769                 rte_pktmbuf_free(m[0]);
 770                 valid_mask &= EXECLUDE_1ST_PKT;
 771         }
 772         if (is_valid_ipv4_pkt(ipv4_hdr[1], m[1]->pkt_len) < 0) {
 773                 rte_pktmbuf_free(m[1]);
 774                 valid_mask &= EXECLUDE_2ND_PKT;
 775         }
 776         if (is_valid_ipv4_pkt(ipv4_hdr[2], m[2]->pkt_len) < 0) {
 777                 rte_pktmbuf_free(m[2]);
 778                 valid_mask &= EXECLUDE_3RD_PKT;
 779         }
 780         if (is_valid_ipv4_pkt(ipv4_hdr[3], m[3]->pkt_len) < 0) {
 781                 rte_pktmbuf_free(m[3]);
 782                 valid_mask &= EXECLUDE_4TH_PKT;
 783         }
 784         if (unlikely(valid_mask != MASK_ALL_PKTS)) {
 785                 if (valid_mask == 0){
 786                         return;
 787                 } else {
 788                         uint8_t i = 0;
 789                         for (i = 0; i < 4; i++) {
 790                                 if ((0x1 << i) & valid_mask) {
 791                                         l3fwd_simple_forward(m[i], portid, qconf);
 792                                 }
 793                         }
 794                         return;
 795                 }
 796         }
 797 #endif // End of #ifdef DO_RFC_1812_CHECKS
 798
 799         data[0] = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m[0], unsigned char *) +
 800                 sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
 801         data[1] = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m[1], unsigned char *) +
 802                 sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
 803         data[2] = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m[2], unsigned char *) +
 804                 sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
 805         data[3] = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m[3], unsigned char *) +
 806                 sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
 807
 808         key[0].xmm = _mm_and_si128(data[0], mask0);
 809         key[1].xmm = _mm_and_si128(data[1], mask0);
 810         key[2].xmm = _mm_and_si128(data[2], mask0);
 811         key[3].xmm = _mm_and_si128(data[3], mask0);
 812
 813         const void *key_array[4] = {&key[0], &key[1], &key[2],&key[3]};
 814         rte_hash_lookup_multi(qconf->ipv4_lookup_struct, &key_array[0], 4, ret);
 815         dst_port[0] = (uint8_t) ((ret[0] < 0) ? portid : ipv4_l3fwd_out_if[ret[0]]);
 816         dst_port[1] = (uint8_t) ((ret[1] < 0) ? portid : ipv4_l3fwd_out_if[ret[1]]);
 817         dst_port[2] = (uint8_t) ((ret[2] < 0) ? portid : ipv4_l3fwd_out_if[ret[2]]);
 818         dst_port[3] = (uint8_t) ((ret[3] < 0) ? portid : ipv4_l3fwd_out_if[ret[3]]);
 819
 820         if (dst_port[0] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[0]) == 0)
 821                 dst_port[0] = portid;
 822         if (dst_port[1] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[1]) == 0)
 823                 dst_port[1] = portid;
 824         if (dst_port[2] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[2]) == 0)
 825                 dst_port[2] = portid;
 826         if (dst_port[3] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[3]) == 0)
 827                 dst_port[3] = portid;
 828
 829         /* 02:00:00:00:00:xx */
 830         d_addr_bytes[0] = &eth_hdr[0]->d_addr.addr_bytes[0];
 831         d_addr_bytes[1] = &eth_hdr[1]->d_addr.addr_bytes[0];
 832         d_addr_bytes[2] = &eth_hdr[2]->d_addr.addr_bytes[0];
 833         d_addr_bytes[3] = &eth_hdr[3]->d_addr.addr_bytes[0];
 834         *((uint64_t *)d_addr_bytes[0]) = 0x000000000002 + ((uint64_t)dst_port[0] << 40);
 835         *((uint64_t *)d_addr_bytes[1]) = 0x000000000002 + ((uint64_t)dst_port[1] << 40);
 836         *((uint64_t *)d_addr_bytes[2]) = 0x000000000002 + ((uint64_t)dst_port[2] << 40);
 837         *((uint64_t *)d_addr_bytes[3]) = 0x000000000002 + ((uint64_t)dst_port[3] << 40);
 838
 839 #ifdef DO_RFC_1812_CHECKS
 840         /* Update time to live and header checksum */
 841         --(ipv4_hdr[0]->time_to_live);
 842         --(ipv4_hdr[1]->time_to_live);
 843         --(ipv4_hdr[2]->time_to_live);
 844         --(ipv4_hdr[3]->time_to_live);
 845         ++(ipv4_hdr[0]->hdr_checksum);
 846         ++(ipv4_hdr[1]->hdr_checksum);
 847         ++(ipv4_hdr[2]->hdr_checksum);
 848         ++(ipv4_hdr[3]->hdr_checksum);
 849 #endif
 850
 851         /* src addr */
 852         ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
 853         ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
 854         ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
 855         ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
 856
 857         send_single_packet(m[0], (uint8_t)dst_port[0]);
 858         send_single_packet(m[1], (uint8_t)dst_port[1]);
 859         send_single_packet(m[2], (uint8_t)dst_port[2]);
 860         send_single_packet(m[3], (uint8_t)dst_port[3]);
 861
 862 }
 863
 864 static inline void get_ipv6_5tuple(struct rte_mbuf* m0, __m128i mask0, __m128i mask1,
 865                                  union ipv6_5tuple_host * key)
 866 {
 867         __m128i tmpdata0 = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m0, unsigned char *)
 868                         + sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len)));
 869         __m128i tmpdata1 = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m0, unsigned char *)
 870                         + sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len)
 871                         +  sizeof(__m128i)));
 872         __m128i tmpdata2 = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m0, unsigned char *)
 873                         + sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len)
 874                         + sizeof(__m128i) + sizeof(__m128i)));
 875         key->xmm[0] = _mm_and_si128(tmpdata0, mask0);
 876         key->xmm[1] = tmpdata1;
 877         key->xmm[2] = _mm_and_si128(tmpdata2, mask1);
 878         return;
 879 }
 880
 881 static inline void
 882 simple_ipv6_fwd_4pkts(struct rte_mbuf* m[4], uint8_t portid, struct lcore_conf *qconf)
 883 {
 884         struct ether_hdr *eth_hdr[4];
 885         __attribute__((unused)) struct ipv6_hdr *ipv6_hdr[4];
 886         void *d_addr_bytes[4];
 887         uint8_t dst_port[4];
 888         int32_t ret[4];
 889         union ipv6_5tuple_host key[4];
 890
 891         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
 892         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
 893         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
 894         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
 895
 896         /* Handle IPv6 headers.*/
 897         ipv6_hdr[0] = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m[0], unsigned char *) +
 898                         sizeof(struct ether_hdr));
 899         ipv6_hdr[1] = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m[1], unsigned char *) +
 900                         sizeof(struct ether_hdr));
 901         ipv6_hdr[2] = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m[2], unsigned char *) +
 902                         sizeof(struct ether_hdr));
 903         ipv6_hdr[3] = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m[3], unsigned char *) +
 904                         sizeof(struct ether_hdr));
 905
 906         get_ipv6_5tuple(m[0], mask1, mask2, &key[0]);
 907         get_ipv6_5tuple(m[1], mask1, mask2, &key[1]);
 908         get_ipv6_5tuple(m[2], mask1, mask2, &key[2]);
 909         get_ipv6_5tuple(m[3], mask1, mask2, &key[3]);
 910
 911         const void *key_array[4] = {&key[0], &key[1], &key[2],&key[3]};
 912         rte_hash_lookup_multi(qconf->ipv6_lookup_struct, &key_array[0], 4, ret);
 913         dst_port[0] = (uint8_t) ((ret[0] < 0)? portid:ipv6_l3fwd_out_if[ret[0]]);
 914         dst_port[1] = (uint8_t) ((ret[1] < 0)? portid:ipv6_l3fwd_out_if[ret[1]]);
 915         dst_port[2] = (uint8_t) ((ret[2] < 0)? portid:ipv6_l3fwd_out_if[ret[2]]);
 916         dst_port[3] = (uint8_t) ((ret[3] < 0)? portid:ipv6_l3fwd_out_if[ret[3]]);
 917
 918         if (dst_port[0] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[0]) == 0)
 919                 dst_port[0] = portid;
 920         if (dst_port[1] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[1]) == 0)
 921                 dst_port[1] = portid;
 922         if (dst_port[2] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[2]) == 0)
 923                 dst_port[2] = portid;
 924         if (dst_port[3] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[3]) == 0)
 925                 dst_port[3] = portid;
 926
 927         /* 02:00:00:00:00:xx */
 928         d_addr_bytes[0] = &eth_hdr[0]->d_addr.addr_bytes[0];
 929         d_addr_bytes[1] = &eth_hdr[1]->d_addr.addr_bytes[0];
 930         d_addr_bytes[2] = &eth_hdr[2]->d_addr.addr_bytes[0];
 931         d_addr_bytes[3] = &eth_hdr[3]->d_addr.addr_bytes[0];
 932         *((uint64_t *)d_addr_bytes[0]) = 0x000000000002 + ((uint64_t)dst_port[0] << 40);
 933         *((uint64_t *)d_addr_bytes[1]) = 0x000000000002 + ((uint64_t)dst_port[1] << 40);
 934         *((uint64_t *)d_addr_bytes[2]) = 0x000000000002 + ((uint64_t)dst_port[2] << 40);
 935         *((uint64_t *)d_addr_bytes[3]) = 0x000000000002 + ((uint64_t)dst_port[3] << 40);
 936
 937         /* src addr */
 938         ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
 939         ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
 940         ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
 941         ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
 942
 943         send_single_packet(m[0], (uint8_t)dst_port[0]);
 944         send_single_packet(m[1], (uint8_t)dst_port[1]);
 945         send_single_packet(m[2], (uint8_t)dst_port[2]);
 946         send_single_packet(m[3], (uint8_t)dst_port[3]);
 947
 948 }
 949 #endif /* APP_LOOKUP_METHOD */
 950
 951 static inline __attribute__((always_inline)) void
 952 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, struct lcore_conf *qconf)
 953 {
 954         struct ether_hdr *eth_hdr;
 955         struct ipv4_hdr *ipv4_hdr;
 956         void *d_addr_bytes;
 957         uint8_t dst_port;
 958
 959         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 960
 961         if (m->ol_flags & PKT_RX_IPV4_HDR) {
 962                 /* Handle IPv4 headers.*/
 963                 ipv4_hdr = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
 964                                 sizeof(struct ether_hdr));
 965
 966 #ifdef DO_RFC_1812_CHECKS
 967                 /* Check to make sure the packet is valid (RFC1812) */
 968                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 969                         rte_pktmbuf_free(m);
 970                         return;
 971                 }
 972 #endif
 973
 974                  dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 975                         qconf->ipv4_lookup_struct);
 976                 if (dst_port >= RTE_MAX_ETHPORTS ||
 977                                 (enabled_port_mask & 1 << dst_port) == 0)
 978                         dst_port = portid;
 979
 980                 /* 02:00:00:00:00:xx */
 981                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 982                 *((uint64_t *)d_addr_bytes) = ETHER_LOCAL_ADMIN_ADDR +
 983                         ((uint64_t)dst_port << 40);
 984
 985 #ifdef DO_RFC_1812_CHECKS
 986                 /* Update time to live and header checksum */
 987                 --(ipv4_hdr->time_to_live);
 988                 ++(ipv4_hdr->hdr_checksum);
 989 #endif
 990
 991                 /* src addr */
 992                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 993
 994                 send_single_packet(m, dst_port);
 995
 996         } else {
 997                 /* Handle IPv6 headers.*/
 998                 struct ipv6_hdr *ipv6_hdr;
 999
1000                 ipv6_hdr = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
1001                                 sizeof(struct ether_hdr));
1002
1003                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid, qconf->ipv6_lookup_struct);
1004
1005                 if (dst_port >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port) == 0)
1006                         dst_port = portid;
1007
1008                 /* 02:00:00:00:00:xx */
1009                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
1010                 *((uint64_t *)d_addr_bytes) = ETHER_LOCAL_ADMIN_ADDR +
1011                         ((uint64_t)dst_port << 40);
1012
1013                 /* src addr */
1014                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
1015
1016                 send_single_packet(m, dst_port);
1017         }
1018
1019 }
1020
1021 #ifdef DO_RFC_1812_CHECKS
1022
1023 #define IPV4_MIN_VER_IHL        0x45
1024 #define IPV4_MAX_VER_IHL        0x4f
1025 #define IPV4_MAX_VER_IHL_DIFF   (IPV4_MAX_VER_IHL - IPV4_MIN_VER_IHL)
1026
1027 /* Minimum value of IPV4 total length (20B) in network byte order. */
1028 #define IPV4_MIN_LEN_BE (sizeof(struct ipv4_hdr) << 8)
1029
1030 /*
1031  * From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2:
1032  * - The IP version number must be 4.
1033  * - The IP header length field must be large enough to hold the
1034  *    minimum length legal IP datagram (20 bytes = 5 words).
1035  * - The IP total length field must be large enough to hold the IP
1036  *   datagram header, whose length is specified in the IP header length
1037  *   field.
1038  * If we encounter invalid IPV4 packet, then set destination port for it
1039  * to BAD_PORT value.
1040  */
1041 static inline __attribute__((always_inline)) void
1042 rfc1812_process(struct ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t flags)
1043 {
1044         uint8_t ihl;
1045
1046         if ((flags & PKT_RX_IPV4_HDR) != 0) {
1047
1048                 ihl = ipv4_hdr->version_ihl - IPV4_MIN_VER_IHL;
1049
1050                 ipv4_hdr->time_to_live--;
1051                 ipv4_hdr->hdr_checksum++;
1052
1053                 if (ihl > IPV4_MAX_VER_IHL_DIFF ||
1054                                 ((uint8_t)ipv4_hdr->total_length == 0 &&
1055                                 ipv4_hdr->total_length < IPV4_MIN_LEN_BE)) {
1056                         dp[0] = BAD_PORT;
1057                 }
1058         }
1059 }
1060
1061 #else
1062 #define rfc1812_process(mb, dp) do { } while (0)
1063 #endif /* DO_RFC_1812_CHECKS */
1064
1065
1066 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1067         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1068
1069 static inline __attribute__((always_inline)) uint16_t
1070 get_dst_port(const struct lcore_conf *qconf, struct rte_mbuf *pkt,
1071         uint32_t dst_ipv4, uint8_t portid)
1072 {
1073         uint8_t next_hop;
1074         struct ipv6_hdr *ipv6_hdr;
1075         struct ether_hdr *eth_hdr;
1076
1077         if (pkt->ol_flags & PKT_RX_IPV4_HDR) {
1078                 if (rte_lpm_lookup(qconf->ipv4_lookup_struct, dst_ipv4,
1079                                 &next_hop) != 0)
1080                         next_hop = portid;
1081         } else if (pkt->ol_flags & PKT_RX_IPV6_HDR) {
1082                 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1083                 ipv6_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
1084                 if (rte_lpm6_lookup(qconf->ipv6_lookup_struct,
1085                                 ipv6_hdr->dst_addr, &next_hop) != 0)
1086                         next_hop = portid;
1087         } else {
1088                 next_hop = portid;
1089         }
1090
1091         return next_hop;
1092 }
1093
1094 static inline void
1095 process_packet(struct lcore_conf *qconf, struct rte_mbuf *pkt,
1096         uint16_t *dst_port, uint8_t portid)
1097 {
1098         struct ether_hdr *eth_hdr;
1099         struct ipv4_hdr *ipv4_hdr;
1100         uint32_t dst_ipv4;
1101         uint16_t dp;
1102         __m128i te, ve;
1103
1104         eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1105         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1106
1107         dst_ipv4 = ipv4_hdr->dst_addr;
1108         dst_ipv4 = rte_be_to_cpu_32(dst_ipv4);
1109         dp = get_dst_port(qconf, pkt, dst_ipv4, portid);
1110
1111         te = _mm_load_si128((__m128i *)eth_hdr);
1112         ve = val_eth[dp];
1113
1114         dst_port[0] = dp;
1115         rfc1812_process(ipv4_hdr, dst_port, pkt->ol_flags);
1116
1117         te =  _mm_blend_epi16(te, ve, MASK_ETH);
1118         _mm_store_si128((__m128i *)eth_hdr, te);
1119 }
1120
1121 /*
1122  * Read ol_flags and destination IPV4 addresses from 4 mbufs.
1123  */
1124 static inline void
1125 processx4_step1(struct rte_mbuf *pkt[FWDSTEP], __m128i *dip, uint32_t *flag)
1126 {
1127         struct ipv4_hdr *ipv4_hdr;
1128         struct ether_hdr *eth_hdr;
1129         uint32_t x0, x1, x2, x3;
1130
1131         eth_hdr = rte_pktmbuf_mtod(pkt[0], struct ether_hdr *);
1132         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1133         x0 = ipv4_hdr->dst_addr;
1134         flag[0] = pkt[0]->ol_flags & PKT_RX_IPV4_HDR;
1135
1136         eth_hdr = rte_pktmbuf_mtod(pkt[1], struct ether_hdr *);
1137         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1138         x1 = ipv4_hdr->dst_addr;
1139         flag[0] &= pkt[1]->ol_flags;
1140
1141         eth_hdr = rte_pktmbuf_mtod(pkt[2], struct ether_hdr *);
1142         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1143         x2 = ipv4_hdr->dst_addr;
1144         flag[0] &= pkt[2]->ol_flags;
1145
1146         eth_hdr = rte_pktmbuf_mtod(pkt[3], struct ether_hdr *);
1147         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1148         x3 = ipv4_hdr->dst_addr;
1149         flag[0] &= pkt[3]->ol_flags;
1150
1151         dip[0] = _mm_set_epi32(x3, x2, x1, x0);
1152 }
1153
1154 /*
1155  * Lookup into LPM for destination port.
1156  * If lookup fails, use incoming port (portid) as destination port.
1157  */
1158 static inline void
1159 processx4_step2(const struct lcore_conf *qconf, __m128i dip, uint32_t flag,
1160         uint8_t portid, struct rte_mbuf *pkt[FWDSTEP], uint16_t dprt[FWDSTEP])
1161 {
1162         rte_xmm_t dst;
1163         const  __m128i bswap_mask = _mm_set_epi8(12, 13, 14, 15, 8, 9, 10, 11,
1164                                                 4, 5, 6, 7, 0, 1, 2, 3);
1165
1166         /* Byte swap 4 IPV4 addresses. */
1167         dip = _mm_shuffle_epi8(dip, bswap_mask);
1168
1169         /* if all 4 packets are IPV4. */
1170         if (likely(flag != 0)) {
1171                 rte_lpm_lookupx4(qconf->ipv4_lookup_struct, dip, dprt, portid);
1172         } else {
1173                 dst.x = dip;
1174                 dprt[0] = get_dst_port(qconf, pkt[0], dst.u32[0], portid);
1175                 dprt[1] = get_dst_port(qconf, pkt[1], dst.u32[1], portid);
1176                 dprt[2] = get_dst_port(qconf, pkt[2], dst.u32[2], portid);
1177                 dprt[3] = get_dst_port(qconf, pkt[3], dst.u32[3], portid);
1178         }
1179 }
1180
1181 /*
1182  * Update source and destination MAC addresses in the ethernet header.
1183  * Perform RFC1812 checks and updates for IPV4 packets.
1184  */
1185 static inline void
1186 processx4_step3(struct rte_mbuf *pkt[FWDSTEP], uint16_t dst_port[FWDSTEP])
1187 {
1188         __m128i te[FWDSTEP];
1189         __m128i ve[FWDSTEP];
1190         __m128i *p[FWDSTEP];
1191
1192         p[0] = (rte_pktmbuf_mtod(pkt[0], __m128i *));
1193         p[1] = (rte_pktmbuf_mtod(pkt[1], __m128i *));
1194         p[2] = (rte_pktmbuf_mtod(pkt[2], __m128i *));
1195         p[3] = (rte_pktmbuf_mtod(pkt[3], __m128i *));
1196
1197         ve[0] = val_eth[dst_port[0]];
1198         te[0] = _mm_load_si128(p[0]);
1199
1200         ve[1] = val_eth[dst_port[1]];
1201         te[1] = _mm_load_si128(p[1]);
1202
1203         ve[2] = val_eth[dst_port[2]];
1204         te[2] = _mm_load_si128(p[2]);
1205
1206         ve[3] = val_eth[dst_port[3]];
1207         te[3] = _mm_load_si128(p[3]);
1208
1209         /* Update first 12 bytes, keep rest bytes intact. */
1210         te[0] =  _mm_blend_epi16(te[0], ve[0], MASK_ETH);
1211         te[1] =  _mm_blend_epi16(te[1], ve[1], MASK_ETH);
1212         te[2] =  _mm_blend_epi16(te[2], ve[2], MASK_ETH);
1213         te[3] =  _mm_blend_epi16(te[3], ve[3], MASK_ETH);
1214
1215         _mm_store_si128(p[0], te[0]);
1216         _mm_store_si128(p[1], te[1]);
1217         _mm_store_si128(p[2], te[2]);
1218         _mm_store_si128(p[3], te[3]);
1219
1220         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[0] + 1),
1221                 &dst_port[0], pkt[0]->ol_flags);
1222         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[1] + 1),
1223                 &dst_port[1], pkt[1]->ol_flags);
1224         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[2] + 1),
1225                 &dst_port[2], pkt[2]->ol_flags);
1226         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[3] + 1),
1227                 &dst_port[3], pkt[3]->ol_flags);
1228 }
1229
1230 /*
1231  * We group consecutive packets with the same destionation port into one burst.
1232  * To avoid extra latency this is done together with some other packet
1233  * processing, but after we made a final decision about packet's destination.
1234  * To do this we maintain:
1235  * pnum - array of number of consecutive packets with the same dest port for
1236  * each packet in the input burst.
1237  * lp - pointer to the last updated element in the pnum.
1238  * dlp - dest port value lp corresponds to.
1239  */
1240
1241 #define GRPSZ   (1 << FWDSTEP)
1242 #define GRPMSK  (GRPSZ - 1)
1243
1244 #define GROUP_PORT_STEP(dlp, dcp, lp, pn, idx)  do { \
1245         if (likely((dlp) == (dcp)[(idx)])) {         \
1246                 (lp)[0]++;                           \
1247         } else {                                     \
1248                 (dlp) = (dcp)[idx];                  \
1249                 (lp) = (pn) + (idx);                 \
1250                 (lp)[0] = 1;                         \
1251         }                                            \
1252 } while (0)
1253
1254 /*
1255  * Group consecutive packets with the same destination port in bursts of 4.
1256  * Suppose we have array of destionation ports:
1257  * dst_port[] = {a, b, c, d,, e, ... }
1258  * dp1 should contain: <a, b, c, d>, dp2: <b, c, d, e>.
1259  * We doing 4 comparisions at once and the result is 4 bit mask.
1260  * This mask is used as an index into prebuild array of pnum values.
1261  */
1262 static inline uint16_t *
1263 port_groupx4(uint16_t pn[FWDSTEP + 1], uint16_t *lp, __m128i dp1, __m128i dp2)
1264 {
1265         static const struct {
1266                 uint64_t pnum; /* prebuild 4 values for pnum[]. */
1267                 int32_t  idx;  /* index for new last updated elemnet. */
1268                 uint16_t lpv;  /* add value to the last updated element. */
1269         } gptbl[GRPSZ] = {
1270         {
1271                 /* 0: a != b, b != c, c != d, d != e */
1272                 .pnum = UINT64_C(0x0001000100010001),
1273                 .idx = 4,
1274                 .lpv = 0,
1275         },
1276         {
1277                 /* 1: a == b, b != c, c != d, d != e */
1278                 .pnum = UINT64_C(0x0001000100010002),
1279                 .idx = 4,
1280                 .lpv = 1,
1281         },
1282         {
1283                 /* 2: a != b, b == c, c != d, d != e */
1284                 .pnum = UINT64_C(0x0001000100020001),
1285                 .idx = 4,
1286                 .lpv = 0,
1287         },
1288         {
1289                 /* 3: a == b, b == c, c != d, d != e */
1290                 .pnum = UINT64_C(0x0001000100020003),
1291                 .idx = 4,
1292                 .lpv = 2,
1293         },
1294         {
1295                 /* 4: a != b, b != c, c == d, d != e */
1296                 .pnum = UINT64_C(0x0001000200010001),
1297                 .idx = 4,
1298                 .lpv = 0,
1299         },
1300         {
1301                 /* 5: a == b, b != c, c == d, d != e */
1302                 .pnum = UINT64_C(0x0001000200010002),
1303                 .idx = 4,
1304                 .lpv = 1,
1305         },
1306         {
1307                 /* 6: a != b, b == c, c == d, d != e */
1308                 .pnum = UINT64_C(0x0001000200030001),
1309                 .idx = 4,
1310                 .lpv = 0,
1311         },
1312         {
1313                 /* 7: a == b, b == c, c == d, d != e */
1314                 .pnum = UINT64_C(0x0001000200030004),
1315                 .idx = 4,
1316                 .lpv = 3,
1317         },
1318         {
1319                 /* 8: a != b, b != c, c != d, d == e */
1320                 .pnum = UINT64_C(0x0002000100010001),
1321                 .idx = 3,
1322                 .lpv = 0,
1323         },
1324         {
1325                 /* 9: a == b, b != c, c != d, d == e */
1326                 .pnum = UINT64_C(0x0002000100010002),
1327                 .idx = 3,
1328                 .lpv = 1,
1329         },
1330         {
1331                 /* 0xa: a != b, b == c, c != d, d == e */
1332                 .pnum = UINT64_C(0x0002000100020001),
1333                 .idx = 3,
1334                 .lpv = 0,
1335         },
1336         {
1337                 /* 0xb: a == b, b == c, c != d, d == e */
1338                 .pnum = UINT64_C(0x0002000100020003),
1339                 .idx = 3,
1340                 .lpv = 2,
1341         },
1342         {
1343                 /* 0xc: a != b, b != c, c == d, d == e */
1344                 .pnum = UINT64_C(0x0002000300010001),
1345                 .idx = 2,
1346                 .lpv = 0,
1347         },
1348         {
1349                 /* 0xd: a == b, b != c, c == d, d == e */
1350                 .pnum = UINT64_C(0x0002000300010002),
1351                 .idx = 2,
1352                 .lpv = 1,
1353         },
1354         {
1355                 /* 0xe: a != b, b == c, c == d, d == e */
1356                 .pnum = UINT64_C(0x0002000300040001),
1357                 .idx = 1,
1358                 .lpv = 0,
1359         },
1360         {
1361                 /* 0xf: a == b, b == c, c == d, d == e */
1362                 .pnum = UINT64_C(0x0002000300040005),
1363                 .idx = 0,
1364                 .lpv = 4,
1365         },
1366         };
1367
1368         union {
1369                 uint16_t u16[FWDSTEP + 1];
1370                 uint64_t u64;
1371         } *pnum = (void *)pn;
1372
1373         int32_t v;
1374
1375         dp1 = _mm_cmpeq_epi16(dp1, dp2);
1376         dp1 = _mm_unpacklo_epi16(dp1, dp1);
1377         v = _mm_movemask_ps((__m128)dp1);
1378
1379         /* update last port counter. */
1380         lp[0] += gptbl[v].lpv;
1381
1382         /* if dest port value has changed. */
1383         if (v != GRPMSK) {
1384                 lp = pnum->u16 + gptbl[v].idx;
1385                 lp[0] = 1;
1386                 pnum->u64 = gptbl[v].pnum;
1387         }
1388
1389         return lp;
1390 }
1391
1392 #endif /* APP_LOOKUP_METHOD */
1393
1394 /* main processing loop */
1395 static int
1396 main_loop(__attribute__((unused)) void *dummy)
1397 {
1398         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1399         unsigned lcore_id;
1400         uint64_t prev_tsc, diff_tsc, cur_tsc;
1401         int i, j, nb_rx;
1402         uint8_t portid, queueid;
1403         struct lcore_conf *qconf;
1404         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1405                 US_PER_S * BURST_TX_DRAIN_US;
1406
1407 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1408         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1409         int32_t k;
1410         uint16_t dlp;
1411         uint16_t *lp;
1412         uint16_t dst_port[MAX_PKT_BURST];
1413         __m128i dip[MAX_PKT_BURST / FWDSTEP];
1414         uint32_t flag[MAX_PKT_BURST / FWDSTEP];
1415         uint16_t pnum[MAX_PKT_BURST + 1];
1416 #endif
1417
1418         prev_tsc = 0;
1419
1420         lcore_id = rte_lcore_id();
1421         qconf = &lcore_conf[lcore_id];
1422
1423         if (qconf->n_rx_queue == 0) {
1424                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
1425                 return 0;
1426         }
1427
1428         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
1429
1430         for (i = 0; i < qconf->n_rx_queue; i++) {
1431
1432                 portid = qconf->rx_queue_list[i].port_id;
1433                 queueid = qconf->rx_queue_list[i].queue_id;
1434                 RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n", lcore_id,
1435                         portid, queueid);
1436         }
1437
1438         while (1) {
1439
1440                 cur_tsc = rte_rdtsc();
1441
1442                 /*
1443                  * TX burst queue drain
1444                  */
1445                 diff_tsc = cur_tsc - prev_tsc;
1446                 if (unlikely(diff_tsc > drain_tsc)) {
1447
1448                         /*
1449                          * This could be optimized (use queueid instead of
1450                          * portid), but it is not called so often
1451                          */
1452                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
1453                                 if (qconf->tx_mbufs[portid].len == 0)
1454                                         continue;
1455                                 send_burst(qconf,
1456                                         qconf->tx_mbufs[portid].len,
1457                                         portid);
1458                                 qconf->tx_mbufs[portid].len = 0;
1459                         }
1460
1461                         prev_tsc = cur_tsc;
1462                 }
1463
1464                 /*
1465                  * Read packet from RX queues
1466                  */
1467                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1468                         portid = qconf->rx_queue_list[i].port_id;
1469                         queueid = qconf->rx_queue_list[i].queue_id;
1470                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1471                                 MAX_PKT_BURST);
1472                         if (nb_rx == 0)
1473                                 continue;
1474
1475 #if (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
1476 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1477                         {
1478                                 /*
1479                                  * Send nb_rx - nb_rx%4 packets
1480                                  * in groups of 4.
1481                                  */
1482                                 int32_t n = RTE_ALIGN_FLOOR(nb_rx, 4);
1483                                 for (j = 0; j < n ; j+=4) {
1484                                         uint32_t ol_flag = pkts_burst[j]->ol_flags
1485                                                         & pkts_burst[j+1]->ol_flags
1486                                                         & pkts_burst[j+2]->ol_flags
1487                                                         & pkts_burst[j+3]->ol_flags;
1488                                         if (ol_flag & PKT_RX_IPV4_HDR ) {
1489                                                 simple_ipv4_fwd_4pkts(&pkts_burst[j],
1490                                                                         portid, qconf);
1491                                         } else if (ol_flag & PKT_RX_IPV6_HDR) {
1492                                                 simple_ipv6_fwd_4pkts(&pkts_burst[j],
1493                                                                         portid, qconf);
1494                                         } else {
1495                                                 l3fwd_simple_forward(pkts_burst[j],
1496                                                                         portid, qconf);
1497                                                 l3fwd_simple_forward(pkts_burst[j+1],
1498                                                                         portid, qconf);
1499                                                 l3fwd_simple_forward(pkts_burst[j+2],
1500                                                                         portid, qconf);
1501                                                 l3fwd_simple_forward(pkts_burst[j+3],
1502                                                                         portid, qconf);
1503                                         }
1504                                 }
1505                                 for (; j < nb_rx ; j++) {
1506                                         l3fwd_simple_forward(pkts_burst[j],
1507                                                                 portid, qconf);
1508                                 }
1509                         }
1510 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1511
1512                         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1513                         for (j = 0; j != k; j += FWDSTEP) {
1514                                 processx4_step1(&pkts_burst[j],
1515                                         &dip[j / FWDSTEP],
1516                                         &flag[j / FWDSTEP]);
1517                         }
1518
1519                         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1520                         for (j = 0; j != k; j += FWDSTEP) {
1521                                 processx4_step2(qconf, dip[j / FWDSTEP],
1522                                         flag[j / FWDSTEP], portid,
1523                                         &pkts_burst[j], &dst_port[j]);
1524                         }
1525
1526                         /*
1527                          * Finish packet processing and group consecutive
1528                          * packets with the same destination port.
1529                          */
1530                         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1531                         if (k != 0) {
1532                                 __m128i dp1, dp2;
1533
1534                                 lp = pnum;
1535                                 lp[0] = 1;
1536
1537                                 processx4_step3(pkts_burst, dst_port);
1538
1539                                 /* dp1: <d[0], d[1], d[2], d[3], ... > */
1540                                 dp1 = _mm_loadu_si128((__m128i *)dst_port);
1541
1542                                 for (j = FWDSTEP; j != k; j += FWDSTEP) {
1543                                         processx4_step3(&pkts_burst[j],
1544                                                 &dst_port[j]);
1545
1546                                         /*
1547                                          * dp2:
1548                                          * <d[j-3], d[j-2], d[j-1], d[j], ... >
1549                                          */
1550                                         dp2 = _mm_loadu_si128((__m128i *)
1551                                                 &dst_port[j - FWDSTEP + 1]);
1552                                         lp  = port_groupx4(&pnum[j - FWDSTEP],
1553                                                 lp, dp1, dp2);
1554
1555                                         /*
1556                                          * dp1:
1557                                          * <d[j], d[j+1], d[j+2], d[j+3], ... >
1558                                          */
1559                                         dp1 = _mm_srli_si128(dp2,
1560                                                 (FWDSTEP - 1) *
1561                                                 sizeof(dst_port[0]));
1562                                 }
1563
1564                                 /*
1565                                  * dp2: <d[j-3], d[j-2], d[j-1], d[j-1], ... >
1566                                  */
1567                                 dp2 = _mm_shufflelo_epi16(dp1, 0xf9);
1568                                 lp  = port_groupx4(&pnum[j - FWDSTEP], lp,
1569                                         dp1, dp2);
1570
1571                                 /*
1572                                  * remove values added by the last repeated
1573                                  * dst port.
1574                                  */
1575                                 lp[0]--;
1576                                 dlp = dst_port[j - 1];
1577                         } else {
1578                                 /* set dlp and lp to the never used values. */
1579                                 dlp = BAD_PORT - 1;
1580                                 lp = pnum + MAX_PKT_BURST;
1581                         }
1582
1583                         /* Process up to last 3 packets one by one. */
1584                         switch (nb_rx % FWDSTEP) {
1585                         case 3:
1586                                 process_packet(qconf, pkts_burst[j],
1587                                         dst_port + j, portid);
1588                                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1589                                 j++;
1590                         case 2:
1591                                 process_packet(qconf, pkts_burst[j],
1592                                         dst_port + j, portid);
1593                                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1594                                 j++;
1595                         case 1:
1596                                 process_packet(qconf, pkts_burst[j],
1597                                         dst_port + j, portid);
1598                                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1599                                 j++;
1600                         }
1601
1602                         /*
1603                          * Send packets out, through destination port.
1604                          * Consecuteve pacekts with the same destination port
1605                          * are already grouped together.
1606                          * If destination port for the packet equals BAD_PORT,
1607                          * then free the packet without sending it out.
1608                          */
1609                         for (j = 0; j < nb_rx; j += k) {
1610
1611                                 int32_t m;
1612                                 uint16_t pn;
1613
1614                                 pn = dst_port[j];
1615                                 k = pnum[j];
1616
1617                                 if (likely(pn != BAD_PORT)) {
1618                                         send_packetsx4(qconf, pn,
1619                                                 pkts_burst + j, k);
1620                                 } else {
1621                                         for (m = j; m != j + k; m++)
1622                                                 rte_pktmbuf_free(pkts_burst[m]);
1623                                 }
1624                         }
1625
1626 #endif /* APP_LOOKUP_METHOD */
1627 #else /* ENABLE_MULTI_BUFFER_OPTIMIZE == 0 */
1628
1629                         /* Prefetch first packets */
1630                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1631                                 rte_prefetch0(rte_pktmbuf_mtod(
1632                                                 pkts_burst[j], void *));
1633                         }
1634
1635                         /* Prefetch and forward already prefetched packets */
1636                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1637                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1638                                                 j + PREFETCH_OFFSET], void *));
1639                                 l3fwd_simple_forward(pkts_burst[j], portid,
1640                                         qconf);
1641                         }
1642
1643                         /* Forward remaining prefetched packets */
1644                         for (; j < nb_rx; j++) {
1645                                 l3fwd_simple_forward(pkts_burst[j], portid,
1646                                         qconf);
1647                         }
1648 #endif /* ENABLE_MULTI_BUFFER_OPTIMIZE */
1649
1650                 }
1651         }
1652 }
1653
1654 static int
1655 check_lcore_params(void)
1656 {
1657         uint8_t queue, lcore;
1658         uint16_t i;
1659         int socketid;
1660
1661         for (i = 0; i < nb_lcore_params; ++i) {
1662                 queue = lcore_params[i].queue_id;
1663                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1664                         printf("invalid queue number: %hhu\n", queue);
1665                         return -1;
1666                 }
1667                 lcore = lcore_params[i].lcore_id;
1668                 if (!rte_lcore_is_enabled(lcore)) {
1669                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
1670                         return -1;
1671                 }
1672                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1673                         (numa_on == 0)) {
1674                         printf("warning: lcore %hhu is on socket %d with numa off \n",
1675                                 lcore, socketid);
1676                 }
1677         }
1678         return 0;
1679 }
1680
1681 static int
1682 check_port_config(const unsigned nb_ports)
1683 {
1684         unsigned portid;
1685         uint16_t i;
1686
1687         for (i = 0; i < nb_lcore_params; ++i) {
1688                 portid = lcore_params[i].port_id;
1689                 if ((enabled_port_mask & (1 << portid)) == 0) {
1690                         printf("port %u is not enabled in port mask\n", portid);
1691                         return -1;
1692                 }
1693                 if (portid >= nb_ports) {
1694                         printf("port %u is not present on the board\n", portid);
1695                         return -1;
1696                 }
1697         }
1698         return 0;
1699 }
1700
1701 static uint8_t
1702 get_port_n_rx_queues(const uint8_t port)
1703 {
1704         int queue = -1;
1705         uint16_t i;
1706
1707         for (i = 0; i < nb_lcore_params; ++i) {
1708                 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
1709                         queue = lcore_params[i].queue_id;
1710         }
1711         return (uint8_t)(++queue);
1712 }
1713
1714 static int
1715 init_lcore_rx_queues(void)
1716 {
1717         uint16_t i, nb_rx_queue;
1718         uint8_t lcore;
1719
1720         for (i = 0; i < nb_lcore_params; ++i) {
1721                 lcore = lcore_params[i].lcore_id;
1722                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1723                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1724                         printf("error: too many queues (%u) for lcore: %u\n",
1725                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1726                         return -1;
1727                 } else {
1728                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1729                                 lcore_params[i].port_id;
1730                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1731                                 lcore_params[i].queue_id;
1732                         lcore_conf[lcore].n_rx_queue++;
1733                 }
1734         }
1735         return 0;
1736 }
1737
1738 /* display usage */
1739 static void
1740 print_usage(const char *prgname)
1741 {
1742         printf ("%s [EAL options] -- -p PORTMASK -P"
1743                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1744                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1745                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1746                 "  -P : enable promiscuous mode\n"
1747                 "  --config (port,queue,lcore): rx queues configuration\n"
1748                 "  --no-numa: optional, disable numa awareness\n"
1749                 "  --ipv6: optional, specify it if running ipv6 packets\n"
1750                 "  --enable-jumbo: enable jumbo frame"
1751                 " which max packet len is PKTLEN in decimal (64-9600)\n"
1752                 "  --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n",
1753                 prgname);
1754 }
1755
1756 static int parse_max_pkt_len(const char *pktlen)
1757 {
1758         char *end = NULL;
1759         unsigned long len;
1760
1761         /* parse decimal string */
1762         len = strtoul(pktlen, &end, 10);
1763         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1764                 return -1;
1765
1766         if (len == 0)
1767                 return -1;
1768
1769         return len;
1770 }
1771
1772 static int
1773 parse_portmask(const char *portmask)
1774 {
1775         char *end = NULL;
1776         unsigned long pm;
1777
1778         /* parse hexadecimal string */
1779         pm = strtoul(portmask, &end, 16);
1780         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1781                 return -1;
1782
1783         if (pm == 0)
1784                 return -1;
1785
1786         return pm;
1787 }
1788
1789 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1790 static int
1791 parse_hash_entry_number(const char *hash_entry_num)
1792 {
1793         char *end = NULL;
1794         unsigned long hash_en;
1795         /* parse hexadecimal string */
1796         hash_en = strtoul(hash_entry_num, &end, 16);
1797         if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
1798                 return -1;
1799
1800         if (hash_en == 0)
1801                 return -1;
1802
1803         return hash_en;
1804 }
1805 #endif
1806
1807 static int
1808 parse_config(const char *q_arg)
1809 {
1810         char s[256];
1811         const char *p, *p0 = q_arg;
1812         char *end;
1813         enum fieldnames {
1814                 FLD_PORT = 0,
1815                 FLD_QUEUE,
1816                 FLD_LCORE,
1817                 _NUM_FLD
1818         };
1819         unsigned long int_fld[_NUM_FLD];
1820         char *str_fld[_NUM_FLD];
1821         int i;
1822         unsigned size;
1823
1824         nb_lcore_params = 0;
1825
1826         while ((p = strchr(p0,'(')) != NULL) {
1827                 ++p;
1828                 if((p0 = strchr(p,')')) == NULL)
1829                         return -1;
1830
1831                 size = p0 - p;
1832                 if(size >= sizeof(s))
1833                         return -1;
1834
1835                 snprintf(s, sizeof(s), "%.*s", size, p);
1836                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
1837                         return -1;
1838                 for (i = 0; i < _NUM_FLD; i++){
1839                         errno = 0;
1840                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1841                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1842                                 return -1;
1843                 }
1844                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1845                         printf("exceeded max number of lcore params: %hu\n",
1846                                 nb_lcore_params);
1847                         return -1;
1848                 }
1849                 lcore_params_array[nb_lcore_params].port_id = (uint8_t)int_fld[FLD_PORT];
1850                 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
1851                 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
1852                 ++nb_lcore_params;
1853         }
1854         lcore_params = lcore_params_array;
1855         return 0;
1856 }
1857
1858 #define CMD_LINE_OPT_CONFIG "config"
1859 #define CMD_LINE_OPT_NO_NUMA "no-numa"
1860 #define CMD_LINE_OPT_IPV6 "ipv6"
1861 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
1862 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
1863
1864 /* Parse the argument given in the command line of the application */
1865 static int
1866 parse_args(int argc, char **argv)
1867 {
1868         int opt, ret;
1869         char **argvopt;
1870         int option_index;
1871         char *prgname = argv[0];
1872         static struct option lgopts[] = {
1873                 {CMD_LINE_OPT_CONFIG, 1, 0, 0},
1874                 {CMD_LINE_OPT_NO_NUMA, 0, 0, 0},
1875                 {CMD_LINE_OPT_IPV6, 0, 0, 0},
1876                 {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, 0},
1877                 {CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, 0},
1878                 {NULL, 0, 0, 0}
1879         };
1880
1881         argvopt = argv;
1882
1883         while ((opt = getopt_long(argc, argvopt, "p:P",
1884                                 lgopts, &option_index)) != EOF) {
1885
1886                 switch (opt) {
1887                 /* portmask */
1888                 case 'p':
1889                         enabled_port_mask = parse_portmask(optarg);
1890                         if (enabled_port_mask == 0) {
1891                                 printf("invalid portmask\n");
1892                                 print_usage(prgname);
1893                                 return -1;
1894                         }
1895                         break;
1896                 case 'P':
1897                         printf("Promiscuous mode selected\n");
1898                         promiscuous_on = 1;
1899                         break;
1900
1901                 /* long options */
1902                 case 0:
1903                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_CONFIG,
1904                                 sizeof (CMD_LINE_OPT_CONFIG))) {
1905                                 ret = parse_config(optarg);
1906                                 if (ret) {
1907                                         printf("invalid config\n");
1908                                         print_usage(prgname);
1909                                         return -1;
1910                                 }
1911                         }
1912
1913                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_NUMA,
1914                                 sizeof(CMD_LINE_OPT_NO_NUMA))) {
1915                                 printf("numa is disabled \n");
1916                                 numa_on = 0;
1917                         }
1918
1919 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1920                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_IPV6,
1921                                 sizeof(CMD_LINE_OPT_IPV6))) {
1922                                 printf("ipv6 is specified \n");
1923                                 ipv6 = 1;
1924                         }
1925 #endif
1926
1927                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ENABLE_JUMBO,
1928                                 sizeof (CMD_LINE_OPT_ENABLE_JUMBO))) {
1929                                 struct option lenopts = {"max-pkt-len", required_argument, 0, 0};
1930
1931                                 printf("jumbo frame is enabled - disabling simple TX path\n");
1932                                 port_conf.rxmode.jumbo_frame = 1;
1933
1934                                 /* if no max-pkt-len set, use the default value ETHER_MAX_LEN */
1935                                 if (0 == getopt_long(argc, argvopt, "", &lenopts, &option_index)) {
1936                                         ret = parse_max_pkt_len(optarg);
1937                                         if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)){
1938                                                 printf("invalid packet length\n");
1939                                                 print_usage(prgname);
1940                                                 return -1;
1941                                         }
1942                                         port_conf.rxmode.max_rx_pkt_len = ret;
1943                                 }
1944                                 printf("set jumbo frame max packet length to %u\n",
1945                                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1946                         }
1947 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1948                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_HASH_ENTRY_NUM,
1949                                 sizeof(CMD_LINE_OPT_HASH_ENTRY_NUM))) {
1950                                 ret = parse_hash_entry_number(optarg);
1951                                 if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
1952                                         hash_entry_number = ret;
1953                                 } else {
1954                                         printf("invalid hash entry number\n");
1955                                         print_usage(prgname);
1956                                         return -1;
1957                                 }
1958                         }
1959 #endif
1960                         break;
1961
1962                 default:
1963                         print_usage(prgname);
1964                         return -1;
1965                 }
1966         }
1967
1968         if (optind >= 0)
1969                 argv[optind-1] = prgname;
1970
1971         ret = optind-1;
1972         optind = 0; /* reset getopt lib */
1973         return ret;
1974 }
1975
1976 static void
1977 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1978 {
1979         char buf[ETHER_ADDR_FMT_SIZE];
1980         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1981         printf("%s%s", name, buf);
1982 }
1983
1984 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1985
1986 static void convert_ipv4_5tuple(struct ipv4_5tuple* key1,
1987                 union ipv4_5tuple_host* key2)
1988 {
1989         key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
1990         key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
1991         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
1992         key2->port_src = rte_cpu_to_be_16(key1->port_src);
1993         key2->proto = key1->proto;
1994         key2->pad0 = 0;
1995         key2->pad1 = 0;
1996         return;
1997 }
1998
1999 static void convert_ipv6_5tuple(struct ipv6_5tuple* key1,
2000                 union ipv6_5tuple_host* key2)
2001 {
2002         uint32_t i;
2003         for (i = 0; i < 16; i++)
2004         {
2005                 key2->ip_dst[i] = key1->ip_dst[i];
2006                 key2->ip_src[i] = key1->ip_src[i];
2007         }
2008         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
2009         key2->port_src = rte_cpu_to_be_16(key1->port_src);
2010         key2->proto = key1->proto;
2011         key2->pad0 = 0;
2012         key2->pad1 = 0;
2013         key2->reserve = 0;
2014         return;
2015 }
2016
2017 #define BYTE_VALUE_MAX 256
2018 #define ALL_32_BITS 0xffffffff
2019 #define BIT_8_TO_15 0x0000ff00
2020 static inline void
2021 populate_ipv4_few_flow_into_table(const struct rte_hash* h)
2022 {
2023         uint32_t i;
2024         int32_t ret;
2025         uint32_t array_len = sizeof(ipv4_l3fwd_route_array)/sizeof(ipv4_l3fwd_route_array[0]);
2026
2027         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
2028         for (i = 0; i < array_len; i++) {
2029                 struct ipv4_l3fwd_route  entry;
2030                 union ipv4_5tuple_host newkey;
2031                 entry = ipv4_l3fwd_route_array[i];
2032                 convert_ipv4_5tuple(&entry.key, &newkey);
2033                 ret = rte_hash_add_key (h,(void *) &newkey);
2034                 if (ret < 0) {
2035                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
2036                                 " to the l3fwd hash.\n", i);
2037                 }
2038                 ipv4_l3fwd_out_if[ret] = entry.if_out;
2039         }
2040         printf("Hash: Adding 0x%" PRIx32 " keys\n", array_len);
2041 }
2042
2043 #define BIT_16_TO_23 0x00ff0000
2044 static inline void
2045 populate_ipv6_few_flow_into_table(const struct rte_hash* h)
2046 {
2047         uint32_t i;
2048         int32_t ret;
2049         uint32_t array_len = sizeof(ipv6_l3fwd_route_array)/sizeof(ipv6_l3fwd_route_array[0]);
2050
2051         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
2052         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
2053         for (i = 0; i < array_len; i++) {
2054                 struct ipv6_l3fwd_route entry;
2055                 union ipv6_5tuple_host newkey;
2056                 entry = ipv6_l3fwd_route_array[i];
2057                 convert_ipv6_5tuple(&entry.key, &newkey);
2058                 ret = rte_hash_add_key (h, (void *) &newkey);
2059                 if (ret < 0) {
2060                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
2061                                 " to the l3fwd hash.\n", i);
2062                 }
2063                 ipv6_l3fwd_out_if[ret] = entry.if_out;
2064         }
2065         printf("Hash: Adding 0x%" PRIx32 "keys\n", array_len);
2066 }
2067
2068 #define NUMBER_PORT_USED 4
2069 static inline void
2070 populate_ipv4_many_flow_into_table(const struct rte_hash* h,
2071                 unsigned int nr_flow)
2072 {
2073         unsigned i;
2074         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
2075         for (i = 0; i < nr_flow; i++) {
2076                 struct ipv4_l3fwd_route entry;
2077                 union ipv4_5tuple_host newkey;
2078                 uint8_t a = (uint8_t) ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
2079                 uint8_t b = (uint8_t) (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
2080                 uint8_t c = (uint8_t) ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
2081                 /* Create the ipv4 exact match flow */
2082                 memset(&entry, 0, sizeof(entry));
2083                 switch (i & (NUMBER_PORT_USED -1)) {
2084                 case 0:
2085                         entry = ipv4_l3fwd_route_array[0];
2086                         entry.key.ip_dst = IPv4(101,c,b,a);
2087                         break;
2088                 case 1:
2089                         entry = ipv4_l3fwd_route_array[1];
2090                         entry.key.ip_dst = IPv4(201,c,b,a);
2091                         break;
2092                 case 2:
2093                         entry = ipv4_l3fwd_route_array[2];
2094                         entry.key.ip_dst = IPv4(111,c,b,a);
2095                         break;
2096                 case 3:
2097                         entry = ipv4_l3fwd_route_array[3];
2098                         entry.key.ip_dst = IPv4(211,c,b,a);
2099                         break;
2100                 };
2101                 convert_ipv4_5tuple(&entry.key, &newkey);
2102                 int32_t ret = rte_hash_add_key(h,(void *) &newkey);
2103                 if (ret < 0) {
2104                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
2105                 }
2106                 ipv4_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
2107
2108         }
2109         printf("Hash: Adding 0x%x keys\n", nr_flow);
2110 }
2111
2112 static inline void
2113 populate_ipv6_many_flow_into_table(const struct rte_hash* h,
2114                 unsigned int nr_flow)
2115 {
2116         unsigned i;
2117         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
2118         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
2119         for (i = 0; i < nr_flow; i++) {
2120                 struct ipv6_l3fwd_route entry;
2121                 union ipv6_5tuple_host newkey;
2122                 uint8_t a = (uint8_t) ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
2123                 uint8_t b = (uint8_t) (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
2124                 uint8_t c = (uint8_t) ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
2125                 /* Create the ipv6 exact match flow */
2126                 memset(&entry, 0, sizeof(entry));
2127                 switch (i & (NUMBER_PORT_USED - 1)) {
2128                 case 0: entry = ipv6_l3fwd_route_array[0]; break;
2129                 case 1: entry = ipv6_l3fwd_route_array[1]; break;
2130                 case 2: entry = ipv6_l3fwd_route_array[2]; break;
2131                 case 3: entry = ipv6_l3fwd_route_array[3]; break;
2132                 };
2133                 entry.key.ip_dst[13] = c;
2134                 entry.key.ip_dst[14] = b;
2135                 entry.key.ip_dst[15] = a;
2136                 convert_ipv6_5tuple(&entry.key, &newkey);
2137                 int32_t ret = rte_hash_add_key(h,(void *) &newkey);
2138                 if (ret < 0) {
2139                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
2140                 }
2141                 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
2142
2143         }
2144         printf("Hash: Adding 0x%x keys\n", nr_flow);
2145 }
2146
2147 static void
2148 setup_hash(int socketid)
2149 {
2150     struct rte_hash_parameters ipv4_l3fwd_hash_params = {
2151         .name = NULL,
2152         .entries = L3FWD_HASH_ENTRIES,
2153         .bucket_entries = 4,
2154         .key_len = sizeof(union ipv4_5tuple_host),
2155         .hash_func = ipv4_hash_crc,
2156         .hash_func_init_val = 0,
2157     };
2158
2159     struct rte_hash_parameters ipv6_l3fwd_hash_params = {
2160         .name = NULL,
2161         .entries = L3FWD_HASH_ENTRIES,
2162         .bucket_entries = 4,
2163         .key_len = sizeof(union ipv6_5tuple_host),
2164         .hash_func = ipv6_hash_crc,
2165         .hash_func_init_val = 0,
2166     };
2167
2168     char s[64];
2169
2170         /* create ipv4 hash */
2171         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
2172         ipv4_l3fwd_hash_params.name = s;
2173         ipv4_l3fwd_hash_params.socket_id = socketid;
2174         ipv4_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv4_l3fwd_hash_params);
2175         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2176                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2177                                 "socket %d\n", socketid);
2178
2179         /* create ipv6 hash */
2180         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
2181         ipv6_l3fwd_hash_params.name = s;
2182         ipv6_l3fwd_hash_params.socket_id = socketid;
2183         ipv6_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv6_l3fwd_hash_params);
2184         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
2185                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2186                                 "socket %d\n", socketid);
2187
2188         if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
2189                 /* For testing hash matching with a large number of flows we
2190                  * generate millions of IP 5-tuples with an incremented dst
2191                  * address to initialize the hash table. */
2192                 if (ipv6 == 0) {
2193                         /* populate the ipv4 hash */
2194                         populate_ipv4_many_flow_into_table(
2195                                 ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
2196                 } else {
2197                         /* populate the ipv6 hash */
2198                         populate_ipv6_many_flow_into_table(
2199                                 ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
2200                 }
2201         } else {
2202                 /* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize the hash table */
2203                 if (ipv6 == 0) {
2204                         /* populate the ipv4 hash */
2205                         populate_ipv4_few_flow_into_table(ipv4_l3fwd_lookup_struct[socketid]);
2206                 } else {
2207                         /* populate the ipv6 hash */
2208                         populate_ipv6_few_flow_into_table(ipv6_l3fwd_lookup_struct[socketid]);
2209                 }
2210         }
2211 }
2212 #endif
2213
2214 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2215 static void
2216 setup_lpm(int socketid)
2217 {
2218         struct rte_lpm6_config config;
2219         unsigned i;
2220         int ret;
2221         char s[64];
2222
2223         /* create the LPM table */
2224         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
2225         ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
2226                                 IPV4_L3FWD_LPM_MAX_RULES, 0);
2227         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2228                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
2229                                 " on socket %d\n", socketid);
2230
2231         /* populate the LPM table */
2232         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
2233
2234                 /* skip unused ports */
2235                 if ((1 << ipv4_l3fwd_route_array[i].if_out &
2236                                 enabled_port_mask) == 0)
2237                         continue;
2238
2239                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
2240                         ipv4_l3fwd_route_array[i].ip,
2241                         ipv4_l3fwd_route_array[i].depth,
2242                         ipv4_l3fwd_route_array[i].if_out);
2243
2244                 if (ret < 0) {
2245                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
2246                                 "l3fwd LPM table on socket %d\n",
2247                                 i, socketid);
2248                 }
2249
2250                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
2251                         (unsigned)ipv4_l3fwd_route_array[i].ip,
2252                         ipv4_l3fwd_route_array[i].depth,
2253                         ipv4_l3fwd_route_array[i].if_out);
2254         }
2255
2256         /* create the LPM6 table */
2257         snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
2258
2259         config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
2260         config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
2261         config.flags = 0;
2262         ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
2263                                 &config);
2264         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
2265                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
2266                                 " on socket %d\n", socketid);
2267
2268         /* populate the LPM table */
2269         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
2270
2271                 /* skip unused ports */
2272                 if ((1 << ipv6_l3fwd_route_array[i].if_out &
2273                                 enabled_port_mask) == 0)
2274                         continue;
2275
2276                 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
2277                         ipv6_l3fwd_route_array[i].ip,
2278                         ipv6_l3fwd_route_array[i].depth,
2279                         ipv6_l3fwd_route_array[i].if_out);
2280
2281                 if (ret < 0) {
2282                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
2283                                 "l3fwd LPM table on socket %d\n",
2284                                 i, socketid);
2285                 }
2286
2287                 printf("LPM: Adding route %s / %d (%d)\n",
2288                         "IPV6",
2289                         ipv6_l3fwd_route_array[i].depth,
2290                         ipv6_l3fwd_route_array[i].if_out);
2291         }
2292 }
2293 #endif
2294
2295 static int
2296 init_mem(unsigned nb_mbuf)
2297 {
2298         struct lcore_conf *qconf;
2299         int socketid;
2300         unsigned lcore_id;
2301         char s[64];
2302
2303         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2304                 if (rte_lcore_is_enabled(lcore_id) == 0)
2305                         continue;
2306
2307                 if (numa_on)
2308                         socketid = rte_lcore_to_socket_id(lcore_id);
2309                 else
2310                         socketid = 0;
2311
2312                 if (socketid >= NB_SOCKETS) {
2313                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
2314                                 socketid, lcore_id, NB_SOCKETS);
2315                 }
2316                 if (pktmbuf_pool[socketid] == NULL) {
2317                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
2318                         pktmbuf_pool[socketid] =
2319                                 rte_mempool_create(s, nb_mbuf, MBUF_SIZE, MEMPOOL_CACHE_SIZE,
2320                                         sizeof(struct rte_pktmbuf_pool_private),
2321                                         rte_pktmbuf_pool_init, NULL,
2322                                         rte_pktmbuf_init, NULL,
2323                                         socketid, 0);
2324                         if (pktmbuf_pool[socketid] == NULL)
2325                                 rte_exit(EXIT_FAILURE,
2326                                                 "Cannot init mbuf pool on socket %d\n", socketid);
2327                         else
2328                                 printf("Allocated mbuf pool on socket %d\n", socketid);
2329
2330 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2331                         setup_lpm(socketid);
2332 #else
2333                         setup_hash(socketid);
2334 #endif
2335                 }
2336                 qconf = &lcore_conf[lcore_id];
2337                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
2338                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
2339         }
2340         return 0;
2341 }
2342
2343 /* Check the link status of all ports in up to 9s, and print them finally */
2344 static void
2345 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
2346 {
2347 #define CHECK_INTERVAL 100 /* 100ms */
2348 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
2349         uint8_t portid, count, all_ports_up, print_flag = 0;
2350         struct rte_eth_link link;
2351
2352         printf("\nChecking link status");
2353         fflush(stdout);
2354         for (count = 0; count <= MAX_CHECK_TIME; count++) {
2355                 all_ports_up = 1;
2356                 for (portid = 0; portid < port_num; portid++) {
2357                         if ((port_mask & (1 << portid)) == 0)
2358                                 continue;
2359                         memset(&link, 0, sizeof(link));
2360                         rte_eth_link_get_nowait(portid, &link);
2361                         /* print link status if flag set */
2362                         if (print_flag == 1) {
2363                                 if (link.link_status)
2364                                         printf("Port %d Link Up - speed %u "
2365                                                 "Mbps - %s\n", (uint8_t)portid,
2366                                                 (unsigned)link.link_speed,
2367                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
2368                                         ("full-duplex") : ("half-duplex\n"));
2369                                 else
2370                                         printf("Port %d Link Down\n",
2371                                                 (uint8_t)portid);
2372                                 continue;
2373                         }
2374                         /* clear all_ports_up flag if any link down */
2375                         if (link.link_status == 0) {
2376                                 all_ports_up = 0;
2377                                 break;
2378                         }
2379                 }
2380                 /* after finally printing all link status, get out */
2381                 if (print_flag == 1)
2382                         break;
2383
2384                 if (all_ports_up == 0) {
2385                         printf(".");
2386                         fflush(stdout);
2387                         rte_delay_ms(CHECK_INTERVAL);
2388                 }
2389
2390                 /* set the print_flag if all ports up or timeout */
2391                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
2392                         print_flag = 1;
2393                         printf("done\n");
2394                 }
2395         }
2396 }
2397
2398 int
2399 main(int argc, char **argv)
2400 {
2401         struct lcore_conf *qconf;
2402         struct rte_eth_dev_info dev_info;
2403         struct rte_eth_txconf *txconf;
2404         int ret;
2405         unsigned nb_ports;
2406         uint16_t queueid;
2407         unsigned lcore_id;
2408         uint32_t n_tx_queue, nb_lcores;
2409         uint8_t portid, nb_rx_queue, queue, socketid;
2410
2411         /* init EAL */
2412         ret = rte_eal_init(argc, argv);
2413         if (ret < 0)
2414                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2415         argc -= ret;
2416         argv += ret;
2417
2418         /* parse application arguments (after the EAL ones) */
2419         ret = parse_args(argc, argv);
2420         if (ret < 0)
2421                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
2422
2423         if (check_lcore_params() < 0)
2424                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
2425
2426         ret = init_lcore_rx_queues();
2427         if (ret < 0)
2428                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2429
2430         nb_ports = rte_eth_dev_count();
2431         if (nb_ports > RTE_MAX_ETHPORTS)
2432                 nb_ports = RTE_MAX_ETHPORTS;
2433
2434         if (check_port_config(nb_ports) < 0)
2435                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
2436
2437         nb_lcores = rte_lcore_count();
2438
2439         /* initialize all ports */
2440         for (portid = 0; portid < nb_ports; portid++) {
2441                 /* skip ports that are not enabled */
2442                 if ((enabled_port_mask & (1 << portid)) == 0) {
2443                         printf("\nSkipping disabled port %d\n", portid);
2444                         continue;
2445                 }
2446
2447                 /* init port */
2448                 printf("Initializing port %d ... ", portid );
2449                 fflush(stdout);
2450
2451                 nb_rx_queue = get_port_n_rx_queues(portid);
2452                 n_tx_queue = nb_lcores;
2453                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
2454                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
2455                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
2456                         nb_rx_queue, (unsigned)n_tx_queue );
2457                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
2458                                         (uint16_t)n_tx_queue, &port_conf);
2459                 if (ret < 0)
2460                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
2461                                 ret, portid);
2462
2463                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
2464                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
2465                 printf(", ");
2466
2467                 /*
2468                  * prepare dst and src MACs for each port.
2469                  */
2470                 *(uint64_t *)(val_eth + portid) =
2471                         ETHER_LOCAL_ADMIN_ADDR + ((uint64_t)portid << 40);
2472                 ether_addr_copy(&ports_eth_addr[portid],
2473                         (struct ether_addr *)(val_eth + portid) + 1);
2474
2475                 /* init memory */
2476                 ret = init_mem(NB_MBUF);
2477                 if (ret < 0)
2478                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
2479
2480                 /* init one TX queue per couple (lcore,port) */
2481                 queueid = 0;
2482                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2483                         if (rte_lcore_is_enabled(lcore_id) == 0)
2484                                 continue;
2485
2486                         if (numa_on)
2487                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2488                         else
2489                                 socketid = 0;
2490
2491                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
2492                         fflush(stdout);
2493
2494                         rte_eth_dev_info_get(portid, &dev_info);
2495                         txconf = &dev_info.default_txconf;
2496                         if (port_conf.rxmode.jumbo_frame)
2497                                 txconf->txq_flags = 0;
2498                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
2499                                                      socketid, txconf);
2500                         if (ret < 0)
2501                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
2502                                         "port=%d\n", ret, portid);
2503
2504                         qconf = &lcore_conf[lcore_id];
2505                         qconf->tx_queue_id[portid] = queueid;
2506                         queueid++;
2507                 }
2508                 printf("\n");
2509         }
2510
2511         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2512                 if (rte_lcore_is_enabled(lcore_id) == 0)
2513                         continue;
2514                 qconf = &lcore_conf[lcore_id];
2515                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
2516                 fflush(stdout);
2517                 /* init RX queues */
2518                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
2519                         portid = qconf->rx_queue_list[queue].port_id;
2520                         queueid = qconf->rx_queue_list[queue].queue_id;
2521
2522                         if (numa_on)
2523                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2524                         else
2525                                 socketid = 0;
2526
2527                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
2528                         fflush(stdout);
2529
2530                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
2531                                         socketid,
2532                                         NULL,
2533                                         pktmbuf_pool[socketid]);
2534                         if (ret < 0)
2535                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
2536                                                 "port=%d\n", ret, portid);
2537                 }
2538         }
2539
2540         printf("\n");
2541
2542         /* start ports */
2543         for (portid = 0; portid < nb_ports; portid++) {
2544                 if ((enabled_port_mask & (1 << portid)) == 0) {
2545                         continue;
2546                 }
2547                 /* Start device */
2548                 ret = rte_eth_dev_start(portid);
2549                 if (ret < 0)
2550                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
2551                                 ret, portid);
2552
2553                 /*
2554                  * If enabled, put device in promiscuous mode.
2555                  * This allows IO forwarding mode to forward packets
2556                  * to itself through 2 cross-connected  ports of the
2557                  * target machine.
2558                  */
2559                 if (promiscuous_on)
2560                         rte_eth_promiscuous_enable(portid);
2561         }
2562
2563         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
2564
2565         /* launch per-lcore init on every lcore */
2566         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2567         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2568                 if (rte_eal_wait_lcore(lcore_id) < 0)
2569                         return -1;
2570         }
2571
2572         return 0;
2573 }