examples/l3fwd/main.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2015 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_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_eal.h>
  53 #include <rte_per_lcore.h>
  54 #include <rte_launch.h>
  55 #include <rte_atomic.h>
  56 #include <rte_cycles.h>
  57 #include <rte_prefetch.h>
  58 #include <rte_lcore.h>
  59 #include <rte_per_lcore.h>
  60 #include <rte_branch_prediction.h>
  61 #include <rte_interrupts.h>
  62 #include <rte_pci.h>
  63 #include <rte_random.h>
  64 #include <rte_debug.h>
  65 #include <rte_ether.h>
  66 #include <rte_ethdev.h>
  67 #include <rte_ring.h>
  68 #include <rte_mempool.h>
  69 #include <rte_mbuf.h>
  70 #include <rte_ip.h>
  71 #include <rte_tcp.h>
  72 #include <rte_udp.h>
  73 #include <rte_string_fns.h>
  74
  75 #include <cmdline_parse.h>
  76 #include <cmdline_parse_etheraddr.h>
  77
  78 #define APP_LOOKUP_EXACT_MATCH          0
  79 #define APP_LOOKUP_LPM                  1
  80 #define DO_RFC_1812_CHECKS
  81
  82 #ifndef APP_LOOKUP_METHOD
  83 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  84 #endif
  85
  86 /*
  87  *  When set to zero, simple forwaring path is eanbled.
  88  *  When set to one, optimized forwarding path is enabled.
  89  *  Note that LPM optimisation path uses SSE4.1 instructions.
  90  */
  91 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && !defined(__SSE4_1__))
  92 #define ENABLE_MULTI_BUFFER_OPTIMIZE    0
  93 #else
  94 #define ENABLE_MULTI_BUFFER_OPTIMIZE    1
  95 #endif
  96
  97 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  98 #include <rte_hash.h>
  99 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 100 #include <rte_lpm.h>
 101 #include <rte_lpm6.h>
 102 #else
 103 #error "APP_LOOKUP_METHOD set to incorrect value"
 104 #endif
 105
 106 #ifndef IPv6_BYTES
 107 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 108                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 109 #define IPv6_BYTES(addr) \
 110         addr[0],  addr[1], addr[2],  addr[3], \
 111         addr[4],  addr[5], addr[6],  addr[7], \
 112         addr[8],  addr[9], addr[10], addr[11],\
 113         addr[12], addr[13],addr[14], addr[15]
 114 #endif
 115
 116
 117 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
 118
 119 #define MAX_JUMBO_PKT_LEN  9600
 120
 121 #define IPV6_ADDR_LEN 16
 122
 123 #define MEMPOOL_CACHE_SIZE 256
 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 uint64_t dest_eth_addr[RTE_MAX_ETHPORTS];
 166 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 167
 168 static __m128i val_eth[RTE_MAX_ETHPORTS];
 169
 170 /* replace first 12B of the ethernet header. */
 171 #define MASK_ETH        0x3f
 172
 173 /* mask of enabled ports */
 174 static uint32_t enabled_port_mask = 0;
 175 static int promiscuous_on = 0; /**< Ports set in promiscuous mode off by default. */
 176 static int numa_on = 1; /**< NUMA is enabled by default. */
 177
 178 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 179 static int ipv6 = 0; /**< ipv6 is false by default. */
 180 #endif
 181
 182 struct mbuf_table {
 183         uint16_t len;
 184         struct rte_mbuf *m_table[MAX_PKT_BURST];
 185 };
 186
 187 struct lcore_rx_queue {
 188         uint8_t port_id;
 189         uint8_t queue_id;
 190 } __rte_cache_aligned;
 191
 192 #define MAX_RX_QUEUE_PER_LCORE 16
 193 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 194 #define MAX_RX_QUEUE_PER_PORT 128
 195
 196 #define MAX_LCORE_PARAMS 1024
 197 struct lcore_params {
 198         uint8_t port_id;
 199         uint8_t queue_id;
 200         uint8_t lcore_id;
 201 } __rte_cache_aligned;
 202
 203 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 204 static struct lcore_params lcore_params_array_default[] = {
 205         {0, 0, 2},
 206         {0, 1, 2},
 207         {0, 2, 2},
 208         {1, 0, 2},
 209         {1, 1, 2},
 210         {1, 2, 2},
 211         {2, 0, 2},
 212         {3, 0, 3},
 213         {3, 1, 3},
 214 };
 215
 216 static struct lcore_params * lcore_params = lcore_params_array_default;
 217 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 218                                 sizeof(lcore_params_array_default[0]);
 219
 220 static struct rte_eth_conf port_conf = {
 221         .rxmode = {
 222                 .mq_mode = ETH_MQ_RX_RSS,
 223                 .max_rx_pkt_len = ETHER_MAX_LEN,
 224                 .split_hdr_size = 0,
 225                 .header_split   = 0, /**< Header Split disabled */
 226                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 227                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 228                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 229                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 230         },
 231         .rx_adv_conf = {
 232                 .rss_conf = {
 233                         .rss_key = NULL,
 234                         .rss_hf = ETH_RSS_IP,
 235                 },
 236         },
 237         .txmode = {
 238                 .mq_mode = ETH_MQ_TX_NONE,
 239         },
 240 };
 241
 242 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 243
 244 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 245
 246 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 247 #include <rte_hash_crc.h>
 248 #define DEFAULT_HASH_FUNC       rte_hash_crc
 249 #else
 250 #include <rte_jhash.h>
 251 #define DEFAULT_HASH_FUNC       rte_jhash
 252 #endif
 253
 254 struct ipv4_5tuple {
 255         uint32_t ip_dst;
 256         uint32_t ip_src;
 257         uint16_t port_dst;
 258         uint16_t port_src;
 259         uint8_t  proto;
 260 } __attribute__((__packed__));
 261
 262 union ipv4_5tuple_host {
 263         struct {
 264                 uint8_t  pad0;
 265                 uint8_t  proto;
 266                 uint16_t pad1;
 267                 uint32_t ip_src;
 268                 uint32_t ip_dst;
 269                 uint16_t port_src;
 270                 uint16_t port_dst;
 271         };
 272         __m128i xmm;
 273 };
 274
 275 #define XMM_NUM_IN_IPV6_5TUPLE 3
 276
 277 struct ipv6_5tuple {
 278         uint8_t  ip_dst[IPV6_ADDR_LEN];
 279         uint8_t  ip_src[IPV6_ADDR_LEN];
 280         uint16_t port_dst;
 281         uint16_t port_src;
 282         uint8_t  proto;
 283 } __attribute__((__packed__));
 284
 285 union ipv6_5tuple_host {
 286         struct {
 287                 uint16_t pad0;
 288                 uint8_t  proto;
 289                 uint8_t  pad1;
 290                 uint8_t  ip_src[IPV6_ADDR_LEN];
 291                 uint8_t  ip_dst[IPV6_ADDR_LEN];
 292                 uint16_t port_src;
 293                 uint16_t port_dst;
 294                 uint64_t reserve;
 295         };
 296         __m128i xmm[XMM_NUM_IN_IPV6_5TUPLE];
 297 };
 298
 299 struct ipv4_l3fwd_route {
 300         struct ipv4_5tuple key;
 301         uint8_t if_out;
 302 };
 303
 304 struct ipv6_l3fwd_route {
 305         struct ipv6_5tuple key;
 306         uint8_t if_out;
 307 };
 308
 309 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 310         {{IPv4(101,0,0,0), IPv4(100,10,0,1),  101, 11, IPPROTO_TCP}, 0},
 311         {{IPv4(201,0,0,0), IPv4(200,20,0,1),  102, 12, IPPROTO_TCP}, 1},
 312         {{IPv4(111,0,0,0), IPv4(100,30,0,1),  101, 11, IPPROTO_TCP}, 2},
 313         {{IPv4(211,0,0,0), IPv4(200,40,0,1),  102, 12, IPPROTO_TCP}, 3},
 314 };
 315
 316 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 317         {{
 318         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 319         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 320         101, 11, IPPROTO_TCP}, 0},
 321
 322         {{
 323         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 324         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 325         102, 12, IPPROTO_TCP}, 1},
 326
 327         {{
 328         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 329         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 330         101, 11, IPPROTO_TCP}, 2},
 331
 332         {{
 333         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 334         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 335         102, 12, IPPROTO_TCP}, 3},
 336 };
 337
 338 typedef struct rte_hash lookup_struct_t;
 339 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 340 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 341
 342 #ifdef RTE_ARCH_X86_64
 343 /* default to 4 million hash entries (approx) */
 344 #define L3FWD_HASH_ENTRIES              1024*1024*4
 345 #else
 346 /* 32-bit has less address-space for hugepage memory, limit to 1M entries */
 347 #define L3FWD_HASH_ENTRIES              1024*1024*1
 348 #endif
 349 #define HASH_ENTRY_NUMBER_DEFAULT       4
 350
 351 static uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
 352
 353 static inline uint32_t
 354 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
 355         uint32_t init_val)
 356 {
 357         const union ipv4_5tuple_host *k;
 358         uint32_t t;
 359         const uint32_t *p;
 360
 361         k = data;
 362         t = k->proto;
 363         p = (const uint32_t *)&k->port_src;
 364
 365 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 366         init_val = rte_hash_crc_4byte(t, init_val);
 367         init_val = rte_hash_crc_4byte(k->ip_src, init_val);
 368         init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
 369         init_val = rte_hash_crc_4byte(*p, init_val);
 370 #else /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 371         init_val = rte_jhash_1word(t, init_val);
 372         init_val = rte_jhash_1word(k->ip_src, init_val);
 373         init_val = rte_jhash_1word(k->ip_dst, init_val);
 374         init_val = rte_jhash_1word(*p, init_val);
 375 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 376         return (init_val);
 377 }
 378
 379 static inline uint32_t
 380 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len, uint32_t init_val)
 381 {
 382         const union ipv6_5tuple_host *k;
 383         uint32_t t;
 384         const uint32_t *p;
 385 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 386         const uint32_t  *ip_src0, *ip_src1, *ip_src2, *ip_src3;
 387         const uint32_t  *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
 388 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 389
 390         k = data;
 391         t = k->proto;
 392         p = (const uint32_t *)&k->port_src;
 393
 394 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 395         ip_src0 = (const uint32_t *) k->ip_src;
 396         ip_src1 = (const uint32_t *)(k->ip_src+4);
 397         ip_src2 = (const uint32_t *)(k->ip_src+8);
 398         ip_src3 = (const uint32_t *)(k->ip_src+12);
 399         ip_dst0 = (const uint32_t *) k->ip_dst;
 400         ip_dst1 = (const uint32_t *)(k->ip_dst+4);
 401         ip_dst2 = (const uint32_t *)(k->ip_dst+8);
 402         ip_dst3 = (const uint32_t *)(k->ip_dst+12);
 403         init_val = rte_hash_crc_4byte(t, init_val);
 404         init_val = rte_hash_crc_4byte(*ip_src0, init_val);
 405         init_val = rte_hash_crc_4byte(*ip_src1, init_val);
 406         init_val = rte_hash_crc_4byte(*ip_src2, init_val);
 407         init_val = rte_hash_crc_4byte(*ip_src3, init_val);
 408         init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
 409         init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
 410         init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
 411         init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
 412         init_val = rte_hash_crc_4byte(*p, init_val);
 413 #else /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 414         init_val = rte_jhash_1word(t, init_val);
 415         init_val = rte_jhash(k->ip_src, sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
 416         init_val = rte_jhash(k->ip_dst, sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
 417         init_val = rte_jhash_1word(*p, init_val);
 418 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 419         return (init_val);
 420 }
 421
 422 #define IPV4_L3FWD_NUM_ROUTES \
 423         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 424
 425 #define IPV6_L3FWD_NUM_ROUTES \
 426         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 427
 428 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 429 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 430
 431 #endif
 432
 433 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 434 struct ipv4_l3fwd_route {
 435         uint32_t ip;
 436         uint8_t  depth;
 437         uint8_t  if_out;
 438 };
 439
 440 struct ipv6_l3fwd_route {
 441         uint8_t ip[16];
 442         uint8_t  depth;
 443         uint8_t  if_out;
 444 };
 445
 446 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 447         {IPv4(1,1,1,0), 24, 0},
 448         {IPv4(2,1,1,0), 24, 1},
 449         {IPv4(3,1,1,0), 24, 2},
 450         {IPv4(4,1,1,0), 24, 3},
 451         {IPv4(5,1,1,0), 24, 4},
 452         {IPv4(6,1,1,0), 24, 5},
 453         {IPv4(7,1,1,0), 24, 6},
 454         {IPv4(8,1,1,0), 24, 7},
 455 };
 456
 457 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 458         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
 459         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
 460         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
 461         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
 462         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
 463         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
 464         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
 465         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
 466 };
 467
 468 #define IPV4_L3FWD_NUM_ROUTES \
 469         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 470 #define IPV6_L3FWD_NUM_ROUTES \
 471         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 472
 473 #define IPV4_L3FWD_LPM_MAX_RULES         1024
 474 #define IPV6_L3FWD_LPM_MAX_RULES         1024
 475 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
 476
 477 typedef struct rte_lpm lookup_struct_t;
 478 typedef struct rte_lpm6 lookup6_struct_t;
 479 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 480 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 481 #endif
 482
 483 struct lcore_conf {
 484         uint16_t n_rx_queue;
 485         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 486         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 487         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 488         lookup_struct_t * ipv4_lookup_struct;
 489 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 490         lookup6_struct_t * ipv6_lookup_struct;
 491 #else
 492         lookup_struct_t * ipv6_lookup_struct;
 493 #endif
 494 } __rte_cache_aligned;
 495
 496 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 497
 498 /* Send burst of packets on an output interface */
 499 static inline int
 500 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 501 {
 502         struct rte_mbuf **m_table;
 503         int ret;
 504         uint16_t queueid;
 505
 506         queueid = qconf->tx_queue_id[port];
 507         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 508
 509         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 510         if (unlikely(ret < n)) {
 511                 do {
 512                         rte_pktmbuf_free(m_table[ret]);
 513                 } while (++ret < n);
 514         }
 515
 516         return 0;
 517 }
 518
 519 /* Enqueue a single packet, and send burst if queue is filled */
 520 static inline int
 521 send_single_packet(struct rte_mbuf *m, uint8_t port)
 522 {
 523         uint32_t lcore_id;
 524         uint16_t len;
 525         struct lcore_conf *qconf;
 526
 527         lcore_id = rte_lcore_id();
 528
 529         qconf = &lcore_conf[lcore_id];
 530         len = qconf->tx_mbufs[port].len;
 531         qconf->tx_mbufs[port].m_table[len] = m;
 532         len++;
 533
 534         /* enough pkts to be sent */
 535         if (unlikely(len == MAX_PKT_BURST)) {
 536                 send_burst(qconf, MAX_PKT_BURST, port);
 537                 len = 0;
 538         }
 539
 540         qconf->tx_mbufs[port].len = len;
 541         return 0;
 542 }
 543
 544 static inline __attribute__((always_inline)) void
 545 send_packetsx4(struct lcore_conf *qconf, uint8_t port,
 546         struct rte_mbuf *m[], uint32_t num)
 547 {
 548         uint32_t len, j, n;
 549
 550         len = qconf->tx_mbufs[port].len;
 551
 552         /*
 553          * If TX buffer for that queue is empty, and we have enough packets,
 554          * then send them straightway.
 555          */
 556         if (num >= MAX_TX_BURST && len == 0) {
 557                 n = rte_eth_tx_burst(port, qconf->tx_queue_id[port], m, num);
 558                 if (unlikely(n < num)) {
 559                         do {
 560                                 rte_pktmbuf_free(m[n]);
 561                         } while (++n < num);
 562                 }
 563                 return;
 564         }
 565
 566         /*
 567          * Put packets into TX buffer for that queue.
 568          */
 569
 570         n = len + num;
 571         n = (n > MAX_PKT_BURST) ? MAX_PKT_BURST - len : num;
 572
 573         j = 0;
 574         switch (n % FWDSTEP) {
 575         while (j < n) {
 576         case 0:
 577                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 578                 j++;
 579         case 3:
 580                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 581                 j++;
 582         case 2:
 583                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 584                 j++;
 585         case 1:
 586                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 587                 j++;
 588         }
 589         }
 590
 591         len += n;
 592
 593         /* enough pkts to be sent */
 594         if (unlikely(len == MAX_PKT_BURST)) {
 595
 596                 send_burst(qconf, MAX_PKT_BURST, port);
 597
 598                 /* copy rest of the packets into the TX buffer. */
 599                 len = num - n;
 600                 j = 0;
 601                 switch (len % FWDSTEP) {
 602                 while (j < len) {
 603                 case 0:
 604                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 605                         j++;
 606                 case 3:
 607                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 608                         j++;
 609                 case 2:
 610                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 611                         j++;
 612                 case 1:
 613                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 614                         j++;
 615                 }
 616                 }
 617         }
 618
 619         qconf->tx_mbufs[port].len = len;
 620 }
 621
 622 #ifdef DO_RFC_1812_CHECKS
 623 static inline int
 624 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 625 {
 626         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 627         /*
 628          * 1. The packet length reported by the Link Layer must be large
 629          * enough to hold the minimum length legal IP datagram (20 bytes).
 630          */
 631         if (link_len < sizeof(struct ipv4_hdr))
 632                 return -1;
 633
 634         /* 2. The IP checksum must be correct. */
 635         /* this is checked in H/W */
 636
 637         /*
 638          * 3. The IP version number must be 4. If the version number is not 4
 639          * then the packet may be another version of IP, such as IPng or
 640          * ST-II.
 641          */
 642         if (((pkt->version_ihl) >> 4) != 4)
 643                 return -3;
 644         /*
 645          * 4. The IP header length field must be large enough to hold the
 646          * minimum length legal IP datagram (20 bytes = 5 words).
 647          */
 648         if ((pkt->version_ihl & 0xf) < 5)
 649                 return -4;
 650
 651         /*
 652          * 5. The IP total length field must be large enough to hold the IP
 653          * datagram header, whose length is specified in the IP header length
 654          * field.
 655          */
 656         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 657                 return -5;
 658
 659         return 0;
 660 }
 661 #endif
 662
 663 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 664
 665 static __m128i mask0;
 666 static __m128i mask1;
 667 static __m128i mask2;
 668 static inline uint8_t
 669 get_ipv4_dst_port(void *ipv4_hdr, uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
 670 {
 671         int ret = 0;
 672         union ipv4_5tuple_host key;
 673
 674         ipv4_hdr = (uint8_t *)ipv4_hdr + offsetof(struct ipv4_hdr, time_to_live);
 675         __m128i data = _mm_loadu_si128((__m128i*)(ipv4_hdr));
 676         /* Get 5 tuple: dst port, src port, dst IP address, src IP address and protocol */
 677         key.xmm = _mm_and_si128(data, mask0);
 678         /* Find destination port */
 679         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 680         return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
 681 }
 682
 683 static inline uint8_t
 684 get_ipv6_dst_port(void *ipv6_hdr,  uint8_t portid, lookup_struct_t * ipv6_l3fwd_lookup_struct)
 685 {
 686         int ret = 0;
 687         union ipv6_5tuple_host key;
 688
 689         ipv6_hdr = (uint8_t *)ipv6_hdr + offsetof(struct ipv6_hdr, payload_len);
 690         __m128i data0 = _mm_loadu_si128((__m128i*)(ipv6_hdr));
 691         __m128i data1 = _mm_loadu_si128((__m128i*)(((uint8_t*)ipv6_hdr)+sizeof(__m128i)));
 692         __m128i data2 = _mm_loadu_si128((__m128i*)(((uint8_t*)ipv6_hdr)+sizeof(__m128i)+sizeof(__m128i)));
 693         /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
 694         key.xmm[0] = _mm_and_si128(data0, mask1);
 695         /* Get part of 5 tuple: dst IP address lower 96 bits and src IP address higher 32 bits */
 696         key.xmm[1] = data1;
 697         /* Get part of 5 tuple: dst port and src port and dst IP address higher 32 bits */
 698         key.xmm[2] = _mm_and_si128(data2, mask2);
 699
 700         /* Find destination port */
 701         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 702         return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
 703 }
 704 #endif
 705
 706 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 707
 708 static inline uint8_t
 709 get_ipv4_dst_port(void *ipv4_hdr,  uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
 710 {
 711         uint8_t next_hop;
 712
 713         return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 714                 rte_be_to_cpu_32(((struct ipv4_hdr *)ipv4_hdr)->dst_addr),
 715                 &next_hop) == 0) ? next_hop : portid);
 716 }
 717
 718 static inline uint8_t
 719 get_ipv6_dst_port(void *ipv6_hdr,  uint8_t portid, lookup6_struct_t * ipv6_l3fwd_lookup_struct)
 720 {
 721         uint8_t next_hop;
 722         return (uint8_t) ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
 723                         ((struct ipv6_hdr*)ipv6_hdr)->dst_addr, &next_hop) == 0)?
 724                         next_hop : portid);
 725 }
 726 #endif
 727
 728 static inline void l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
 729         struct lcore_conf *qconf)  __attribute__((unused));
 730
 731 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) && \
 732         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
 733
 734 #define MASK_ALL_PKTS    0xff
 735 #define EXCLUDE_1ST_PKT 0xfe
 736 #define EXCLUDE_2ND_PKT 0xfd
 737 #define EXCLUDE_3RD_PKT 0xfb
 738 #define EXCLUDE_4TH_PKT 0xf7
 739 #define EXCLUDE_5TH_PKT 0xef
 740 #define EXCLUDE_6TH_PKT 0xdf
 741 #define EXCLUDE_7TH_PKT 0xbf
 742 #define EXCLUDE_8TH_PKT 0x7f
 743
 744 static inline void
 745 simple_ipv4_fwd_8pkts(struct rte_mbuf *m[8], uint8_t portid, struct lcore_conf *qconf)
 746 {
 747         struct ether_hdr *eth_hdr[8];
 748         struct ipv4_hdr *ipv4_hdr[8];
 749         uint8_t dst_port[8];
 750         int32_t ret[8];
 751         union ipv4_5tuple_host key[8];
 752         __m128i data[8];
 753
 754         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
 755         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
 756         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
 757         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
 758         eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *);
 759         eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *);
 760         eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *);
 761         eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *);
 762
 763         /* Handle IPv4 headers.*/
 764         ipv4_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv4_hdr *,
 765                                               sizeof(struct ether_hdr));
 766         ipv4_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv4_hdr *,
 767                                               sizeof(struct ether_hdr));
 768         ipv4_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv4_hdr *,
 769                                               sizeof(struct ether_hdr));
 770         ipv4_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv4_hdr *,
 771                                               sizeof(struct ether_hdr));
 772         ipv4_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv4_hdr *,
 773                                               sizeof(struct ether_hdr));
 774         ipv4_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv4_hdr *,
 775                                               sizeof(struct ether_hdr));
 776         ipv4_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv4_hdr *,
 777                                               sizeof(struct ether_hdr));
 778         ipv4_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv4_hdr *,
 779                                               sizeof(struct ether_hdr));
 780
 781 #ifdef DO_RFC_1812_CHECKS
 782         /* Check to make sure the packet is valid (RFC1812) */
 783         uint8_t valid_mask = MASK_ALL_PKTS;
 784         if (is_valid_ipv4_pkt(ipv4_hdr[0], m[0]->pkt_len) < 0) {
 785                 rte_pktmbuf_free(m[0]);
 786                 valid_mask &= EXCLUDE_1ST_PKT;
 787         }
 788         if (is_valid_ipv4_pkt(ipv4_hdr[1], m[1]->pkt_len) < 0) {
 789                 rte_pktmbuf_free(m[1]);
 790                 valid_mask &= EXCLUDE_2ND_PKT;
 791         }
 792         if (is_valid_ipv4_pkt(ipv4_hdr[2], m[2]->pkt_len) < 0) {
 793                 rte_pktmbuf_free(m[2]);
 794                 valid_mask &= EXCLUDE_3RD_PKT;
 795         }
 796         if (is_valid_ipv4_pkt(ipv4_hdr[3], m[3]->pkt_len) < 0) {
 797                 rte_pktmbuf_free(m[3]);
 798                 valid_mask &= EXCLUDE_4TH_PKT;
 799         }
 800         if (is_valid_ipv4_pkt(ipv4_hdr[4], m[4]->pkt_len) < 0) {
 801                 rte_pktmbuf_free(m[4]);
 802                 valid_mask &= EXCLUDE_5TH_PKT;
 803         }
 804         if (is_valid_ipv4_pkt(ipv4_hdr[5], m[5]->pkt_len) < 0) {
 805                 rte_pktmbuf_free(m[5]);
 806                 valid_mask &= EXCLUDE_6TH_PKT;
 807         }
 808         if (is_valid_ipv4_pkt(ipv4_hdr[6], m[6]->pkt_len) < 0) {
 809                 rte_pktmbuf_free(m[6]);
 810                 valid_mask &= EXCLUDE_7TH_PKT;
 811         }
 812         if (is_valid_ipv4_pkt(ipv4_hdr[7], m[7]->pkt_len) < 0) {
 813                 rte_pktmbuf_free(m[7]);
 814                 valid_mask &= EXCLUDE_8TH_PKT;
 815         }
 816         if (unlikely(valid_mask != MASK_ALL_PKTS)) {
 817                 if (valid_mask == 0){
 818                         return;
 819                 } else {
 820                         uint8_t i = 0;
 821                         for (i = 0; i < 8; i++) {
 822                                 if ((0x1 << i) & valid_mask) {
 823                                         l3fwd_simple_forward(m[i], portid, qconf);
 824                                 }
 825                         }
 826                         return;
 827                 }
 828         }
 829 #endif // End of #ifdef DO_RFC_1812_CHECKS
 830
 831         data[0] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[0], __m128i *,
 832                                         sizeof(struct ether_hdr) +
 833                                         offsetof(struct ipv4_hdr, time_to_live)));
 834         data[1] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[1], __m128i *,
 835                                         sizeof(struct ether_hdr) +
 836                                         offsetof(struct ipv4_hdr, time_to_live)));
 837         data[2] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[2], __m128i *,
 838                                         sizeof(struct ether_hdr) +
 839                                         offsetof(struct ipv4_hdr, time_to_live)));
 840         data[3] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[3], __m128i *,
 841                                         sizeof(struct ether_hdr) +
 842                                         offsetof(struct ipv4_hdr, time_to_live)));
 843         data[4] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[4], __m128i *,
 844                                         sizeof(struct ether_hdr) +
 845                                         offsetof(struct ipv4_hdr, time_to_live)));
 846         data[5] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[5], __m128i *,
 847                                         sizeof(struct ether_hdr) +
 848                                         offsetof(struct ipv4_hdr, time_to_live)));
 849         data[6] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[6], __m128i *,
 850                                         sizeof(struct ether_hdr) +
 851                                         offsetof(struct ipv4_hdr, time_to_live)));
 852         data[7] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[7], __m128i *,
 853                                         sizeof(struct ether_hdr) +
 854                                         offsetof(struct ipv4_hdr, time_to_live)));
 855
 856         key[0].xmm = _mm_and_si128(data[0], mask0);
 857         key[1].xmm = _mm_and_si128(data[1], mask0);
 858         key[2].xmm = _mm_and_si128(data[2], mask0);
 859         key[3].xmm = _mm_and_si128(data[3], mask0);
 860         key[4].xmm = _mm_and_si128(data[4], mask0);
 861         key[5].xmm = _mm_and_si128(data[5], mask0);
 862         key[6].xmm = _mm_and_si128(data[6], mask0);
 863         key[7].xmm = _mm_and_si128(data[7], mask0);
 864
 865         const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
 866                                 &key[4], &key[5], &key[6], &key[7]};
 867
 868         rte_hash_lookup_multi(qconf->ipv4_lookup_struct, &key_array[0], 8, ret);
 869         dst_port[0] = (uint8_t) ((ret[0] < 0) ? portid : ipv4_l3fwd_out_if[ret[0]]);
 870         dst_port[1] = (uint8_t) ((ret[1] < 0) ? portid : ipv4_l3fwd_out_if[ret[1]]);
 871         dst_port[2] = (uint8_t) ((ret[2] < 0) ? portid : ipv4_l3fwd_out_if[ret[2]]);
 872         dst_port[3] = (uint8_t) ((ret[3] < 0) ? portid : ipv4_l3fwd_out_if[ret[3]]);
 873         dst_port[4] = (uint8_t) ((ret[4] < 0) ? portid : ipv4_l3fwd_out_if[ret[4]]);
 874         dst_port[5] = (uint8_t) ((ret[5] < 0) ? portid : ipv4_l3fwd_out_if[ret[5]]);
 875         dst_port[6] = (uint8_t) ((ret[6] < 0) ? portid : ipv4_l3fwd_out_if[ret[6]]);
 876         dst_port[7] = (uint8_t) ((ret[7] < 0) ? portid : ipv4_l3fwd_out_if[ret[7]]);
 877
 878         if (dst_port[0] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[0]) == 0)
 879                 dst_port[0] = portid;
 880         if (dst_port[1] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[1]) == 0)
 881                 dst_port[1] = portid;
 882         if (dst_port[2] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[2]) == 0)
 883                 dst_port[2] = portid;
 884         if (dst_port[3] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[3]) == 0)
 885                 dst_port[3] = portid;
 886         if (dst_port[4] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[4]) == 0)
 887                 dst_port[4] = portid;
 888         if (dst_port[5] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[5]) == 0)
 889                 dst_port[5] = portid;
 890         if (dst_port[6] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[6]) == 0)
 891                 dst_port[6] = portid;
 892         if (dst_port[7] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[7]) == 0)
 893                 dst_port[7] = portid;
 894
 895 #ifdef DO_RFC_1812_CHECKS
 896         /* Update time to live and header checksum */
 897         --(ipv4_hdr[0]->time_to_live);
 898         --(ipv4_hdr[1]->time_to_live);
 899         --(ipv4_hdr[2]->time_to_live);
 900         --(ipv4_hdr[3]->time_to_live);
 901         ++(ipv4_hdr[0]->hdr_checksum);
 902         ++(ipv4_hdr[1]->hdr_checksum);
 903         ++(ipv4_hdr[2]->hdr_checksum);
 904         ++(ipv4_hdr[3]->hdr_checksum);
 905         --(ipv4_hdr[4]->time_to_live);
 906         --(ipv4_hdr[5]->time_to_live);
 907         --(ipv4_hdr[6]->time_to_live);
 908         --(ipv4_hdr[7]->time_to_live);
 909         ++(ipv4_hdr[4]->hdr_checksum);
 910         ++(ipv4_hdr[5]->hdr_checksum);
 911         ++(ipv4_hdr[6]->hdr_checksum);
 912         ++(ipv4_hdr[7]->hdr_checksum);
 913 #endif
 914
 915         /* dst addr */
 916         *(uint64_t *)&eth_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
 917         *(uint64_t *)&eth_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
 918         *(uint64_t *)&eth_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
 919         *(uint64_t *)&eth_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
 920         *(uint64_t *)&eth_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
 921         *(uint64_t *)&eth_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
 922         *(uint64_t *)&eth_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
 923         *(uint64_t *)&eth_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
 924
 925         /* src addr */
 926         ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
 927         ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
 928         ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
 929         ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
 930         ether_addr_copy(&ports_eth_addr[dst_port[4]], &eth_hdr[4]->s_addr);
 931         ether_addr_copy(&ports_eth_addr[dst_port[5]], &eth_hdr[5]->s_addr);
 932         ether_addr_copy(&ports_eth_addr[dst_port[6]], &eth_hdr[6]->s_addr);
 933         ether_addr_copy(&ports_eth_addr[dst_port[7]], &eth_hdr[7]->s_addr);
 934
 935         send_single_packet(m[0], (uint8_t)dst_port[0]);
 936         send_single_packet(m[1], (uint8_t)dst_port[1]);
 937         send_single_packet(m[2], (uint8_t)dst_port[2]);
 938         send_single_packet(m[3], (uint8_t)dst_port[3]);
 939         send_single_packet(m[4], (uint8_t)dst_port[4]);
 940         send_single_packet(m[5], (uint8_t)dst_port[5]);
 941         send_single_packet(m[6], (uint8_t)dst_port[6]);
 942         send_single_packet(m[7], (uint8_t)dst_port[7]);
 943
 944 }
 945
 946 static inline void get_ipv6_5tuple(struct rte_mbuf* m0, __m128i mask0, __m128i mask1,
 947                                  union ipv6_5tuple_host * key)
 948 {
 949         __m128i tmpdata0 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0, __m128i *, sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len)));
 950         __m128i tmpdata1 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0, __m128i *, sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i)));
 951         __m128i tmpdata2 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0, __m128i *, sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i) + sizeof(__m128i)));
 952         key->xmm[0] = _mm_and_si128(tmpdata0, mask0);
 953         key->xmm[1] = tmpdata1;
 954         key->xmm[2] = _mm_and_si128(tmpdata2, mask1);
 955         return;
 956 }
 957
 958 static inline void
 959 simple_ipv6_fwd_8pkts(struct rte_mbuf *m[8], uint8_t portid, struct lcore_conf *qconf)
 960 {
 961         struct ether_hdr *eth_hdr[8];
 962         __attribute__((unused)) struct ipv6_hdr *ipv6_hdr[8];
 963         uint8_t dst_port[8];
 964         int32_t ret[8];
 965         union ipv6_5tuple_host key[8];
 966
 967         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
 968         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
 969         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
 970         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
 971         eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *);
 972         eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *);
 973         eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *);
 974         eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *);
 975
 976         /* Handle IPv6 headers.*/
 977         ipv6_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv6_hdr *,
 978                                               sizeof(struct ether_hdr));
 979         ipv6_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv6_hdr *,
 980                                               sizeof(struct ether_hdr));
 981         ipv6_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv6_hdr *,
 982                                               sizeof(struct ether_hdr));
 983         ipv6_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv6_hdr *,
 984                                               sizeof(struct ether_hdr));
 985         ipv6_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv6_hdr *,
 986                                               sizeof(struct ether_hdr));
 987         ipv6_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv6_hdr *,
 988                                               sizeof(struct ether_hdr));
 989         ipv6_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv6_hdr *,
 990                                               sizeof(struct ether_hdr));
 991         ipv6_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv6_hdr *,
 992                                               sizeof(struct ether_hdr));
 993
 994         get_ipv6_5tuple(m[0], mask1, mask2, &key[0]);
 995         get_ipv6_5tuple(m[1], mask1, mask2, &key[1]);
 996         get_ipv6_5tuple(m[2], mask1, mask2, &key[2]);
 997         get_ipv6_5tuple(m[3], mask1, mask2, &key[3]);
 998         get_ipv6_5tuple(m[4], mask1, mask2, &key[4]);
 999         get_ipv6_5tuple(m[5], mask1, mask2, &key[5]);
1000         get_ipv6_5tuple(m[6], mask1, mask2, &key[6]);
1001         get_ipv6_5tuple(m[7], mask1, mask2, &key[7]);
1002
1003         const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1004                                 &key[4], &key[5], &key[6], &key[7]};
1005
1006         rte_hash_lookup_multi(qconf->ipv6_lookup_struct, &key_array[0], 4, ret);
1007         dst_port[0] = (uint8_t) ((ret[0] < 0) ? portid:ipv6_l3fwd_out_if[ret[0]]);
1008         dst_port[1] = (uint8_t) ((ret[1] < 0) ? portid:ipv6_l3fwd_out_if[ret[1]]);
1009         dst_port[2] = (uint8_t) ((ret[2] < 0) ? portid:ipv6_l3fwd_out_if[ret[2]]);
1010         dst_port[3] = (uint8_t) ((ret[3] < 0) ? portid:ipv6_l3fwd_out_if[ret[3]]);
1011         dst_port[4] = (uint8_t) ((ret[4] < 0) ? portid:ipv6_l3fwd_out_if[ret[4]]);
1012         dst_port[5] = (uint8_t) ((ret[5] < 0) ? portid:ipv6_l3fwd_out_if[ret[5]]);
1013         dst_port[6] = (uint8_t) ((ret[6] < 0) ? portid:ipv6_l3fwd_out_if[ret[6]]);
1014         dst_port[7] = (uint8_t) ((ret[7] < 0) ? portid:ipv6_l3fwd_out_if[ret[7]]);
1015
1016         if (dst_port[0] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[0]) == 0)
1017                 dst_port[0] = portid;
1018         if (dst_port[1] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[1]) == 0)
1019                 dst_port[1] = portid;
1020         if (dst_port[2] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[2]) == 0)
1021                 dst_port[2] = portid;
1022         if (dst_port[3] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[3]) == 0)
1023                 dst_port[3] = portid;
1024         if (dst_port[4] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[4]) == 0)
1025                 dst_port[4] = portid;
1026         if (dst_port[5] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[5]) == 0)
1027                 dst_port[5] = portid;
1028         if (dst_port[6] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[6]) == 0)
1029                 dst_port[6] = portid;
1030         if (dst_port[7] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[7]) == 0)
1031                 dst_port[7] = portid;
1032
1033         /* dst addr */
1034         *(uint64_t *)&eth_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1035         *(uint64_t *)&eth_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1036         *(uint64_t *)&eth_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1037         *(uint64_t *)&eth_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1038         *(uint64_t *)&eth_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1039         *(uint64_t *)&eth_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1040         *(uint64_t *)&eth_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1041         *(uint64_t *)&eth_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1042
1043         /* src addr */
1044         ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
1045         ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
1046         ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
1047         ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
1048         ether_addr_copy(&ports_eth_addr[dst_port[4]], &eth_hdr[4]->s_addr);
1049         ether_addr_copy(&ports_eth_addr[dst_port[5]], &eth_hdr[5]->s_addr);
1050         ether_addr_copy(&ports_eth_addr[dst_port[6]], &eth_hdr[6]->s_addr);
1051         ether_addr_copy(&ports_eth_addr[dst_port[7]], &eth_hdr[7]->s_addr);
1052
1053         send_single_packet(m[0], (uint8_t)dst_port[0]);
1054         send_single_packet(m[1], (uint8_t)dst_port[1]);
1055         send_single_packet(m[2], (uint8_t)dst_port[2]);
1056         send_single_packet(m[3], (uint8_t)dst_port[3]);
1057         send_single_packet(m[4], (uint8_t)dst_port[4]);
1058         send_single_packet(m[5], (uint8_t)dst_port[5]);
1059         send_single_packet(m[6], (uint8_t)dst_port[6]);
1060         send_single_packet(m[7], (uint8_t)dst_port[7]);
1061
1062 }
1063 #endif /* APP_LOOKUP_METHOD */
1064
1065 static inline __attribute__((always_inline)) void
1066 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, struct lcore_conf *qconf)
1067 {
1068         struct ether_hdr *eth_hdr;
1069         struct ipv4_hdr *ipv4_hdr;
1070         uint8_t dst_port;
1071
1072         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
1073
1074 #ifdef RTE_NEXT_ABI
1075         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
1076 #else
1077         if (m->ol_flags & PKT_RX_IPV4_HDR) {
1078 #endif
1079                 /* Handle IPv4 headers.*/
1080                 ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
1081                                                    sizeof(struct ether_hdr));
1082
1083 #ifdef DO_RFC_1812_CHECKS
1084                 /* Check to make sure the packet is valid (RFC1812) */
1085                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
1086                         rte_pktmbuf_free(m);
1087                         return;
1088                 }
1089 #endif
1090
1091                  dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
1092                         qconf->ipv4_lookup_struct);
1093                 if (dst_port >= RTE_MAX_ETHPORTS ||
1094                                 (enabled_port_mask & 1 << dst_port) == 0)
1095                         dst_port = portid;
1096
1097 #ifdef DO_RFC_1812_CHECKS
1098                 /* Update time to live and header checksum */
1099                 --(ipv4_hdr->time_to_live);
1100                 ++(ipv4_hdr->hdr_checksum);
1101 #endif
1102                 /* dst addr */
1103                 *(uint64_t *)&eth_hdr->d_addr = dest_eth_addr[dst_port];
1104
1105                 /* src addr */
1106                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
1107
1108                 send_single_packet(m, dst_port);
1109 #ifdef RTE_NEXT_ABI
1110         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
1111 #else
1112         } else {
1113 #endif
1114                 /* Handle IPv6 headers.*/
1115                 struct ipv6_hdr *ipv6_hdr;
1116
1117                 ipv6_hdr = rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
1118                                                    sizeof(struct ether_hdr));
1119
1120                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid, qconf->ipv6_lookup_struct);
1121
1122                 if (dst_port >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port) == 0)
1123                         dst_port = portid;
1124
1125                 /* dst addr */
1126                 *(uint64_t *)&eth_hdr->d_addr = dest_eth_addr[dst_port];
1127
1128                 /* src addr */
1129                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
1130
1131                 send_single_packet(m, dst_port);
1132 #ifdef RTE_NEXT_ABI
1133         } else
1134                 /* Free the mbuf that contains non-IPV4/IPV6 packet */
1135                 rte_pktmbuf_free(m);
1136 #else
1137         }
1138 #endif
1139 }
1140
1141 #ifdef DO_RFC_1812_CHECKS
1142
1143 #define IPV4_MIN_VER_IHL        0x45
1144 #define IPV4_MAX_VER_IHL        0x4f
1145 #define IPV4_MAX_VER_IHL_DIFF   (IPV4_MAX_VER_IHL - IPV4_MIN_VER_IHL)
1146
1147 /* Minimum value of IPV4 total length (20B) in network byte order. */
1148 #define IPV4_MIN_LEN_BE (sizeof(struct ipv4_hdr) << 8)
1149
1150 /*
1151  * From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2:
1152  * - The IP version number must be 4.
1153  * - The IP header length field must be large enough to hold the
1154  *    minimum length legal IP datagram (20 bytes = 5 words).
1155  * - The IP total length field must be large enough to hold the IP
1156  *   datagram header, whose length is specified in the IP header length
1157  *   field.
1158  * If we encounter invalid IPV4 packet, then set destination port for it
1159  * to BAD_PORT value.
1160  */
1161 static inline __attribute__((always_inline)) void
1162 #ifdef RTE_NEXT_ABI
1163 rfc1812_process(struct ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t ptype)
1164 #else
1165 rfc1812_process(struct ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t flags)
1166 #endif
1167 {
1168         uint8_t ihl;
1169
1170 #ifdef RTE_NEXT_ABI
1171         if (RTE_ETH_IS_IPV4_HDR(ptype)) {
1172 #else
1173         if ((flags & PKT_RX_IPV4_HDR) != 0) {
1174 #endif
1175                 ihl = ipv4_hdr->version_ihl - IPV4_MIN_VER_IHL;
1176
1177                 ipv4_hdr->time_to_live--;
1178                 ipv4_hdr->hdr_checksum++;
1179
1180                 if (ihl > IPV4_MAX_VER_IHL_DIFF ||
1181                                 ((uint8_t)ipv4_hdr->total_length == 0 &&
1182                                 ipv4_hdr->total_length < IPV4_MIN_LEN_BE)) {
1183                         dp[0] = BAD_PORT;
1184                 }
1185         }
1186 }
1187
1188 #else
1189 #define rfc1812_process(mb, dp) do { } while (0)
1190 #endif /* DO_RFC_1812_CHECKS */
1191
1192
1193 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1194         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1195
1196 static inline __attribute__((always_inline)) uint16_t
1197 get_dst_port(const struct lcore_conf *qconf, struct rte_mbuf *pkt,
1198         uint32_t dst_ipv4, uint8_t portid)
1199 {
1200         uint8_t next_hop;
1201         struct ipv6_hdr *ipv6_hdr;
1202         struct ether_hdr *eth_hdr;
1203
1204 #ifdef RTE_NEXT_ABI
1205         if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
1206 #else
1207         if (pkt->ol_flags & PKT_RX_IPV4_HDR) {
1208 #endif
1209                 if (rte_lpm_lookup(qconf->ipv4_lookup_struct, dst_ipv4,
1210                                 &next_hop) != 0)
1211                         next_hop = portid;
1212 #ifdef RTE_NEXT_ABI
1213         } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
1214 #else
1215         } else if (pkt->ol_flags & PKT_RX_IPV6_HDR) {
1216 #endif
1217                 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1218                 ipv6_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
1219                 if (rte_lpm6_lookup(qconf->ipv6_lookup_struct,
1220                                 ipv6_hdr->dst_addr, &next_hop) != 0)
1221                         next_hop = portid;
1222         } else {
1223                 next_hop = portid;
1224         }
1225
1226         return next_hop;
1227 }
1228
1229 static inline void
1230 process_packet(struct lcore_conf *qconf, struct rte_mbuf *pkt,
1231         uint16_t *dst_port, uint8_t portid)
1232 {
1233         struct ether_hdr *eth_hdr;
1234         struct ipv4_hdr *ipv4_hdr;
1235         uint32_t dst_ipv4;
1236         uint16_t dp;
1237         __m128i te, ve;
1238
1239         eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1240         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1241
1242         dst_ipv4 = ipv4_hdr->dst_addr;
1243         dst_ipv4 = rte_be_to_cpu_32(dst_ipv4);
1244         dp = get_dst_port(qconf, pkt, dst_ipv4, portid);
1245
1246         te = _mm_load_si128((__m128i *)eth_hdr);
1247         ve = val_eth[dp];
1248
1249         dst_port[0] = dp;
1250 #ifdef RTE_NEXT_ABI
1251         rfc1812_process(ipv4_hdr, dst_port, pkt->packet_type);
1252 #else
1253         rfc1812_process(ipv4_hdr, dst_port, pkt->ol_flags);
1254 #endif
1255
1256         te =  _mm_blend_epi16(te, ve, MASK_ETH);
1257         _mm_store_si128((__m128i *)eth_hdr, te);
1258 }
1259
1260 #ifdef RTE_NEXT_ABI
1261 /*
1262  * Read packet_type and destination IPV4 addresses from 4 mbufs.
1263  */
1264 static inline void
1265 processx4_step1(struct rte_mbuf *pkt[FWDSTEP],
1266                 __m128i *dip,
1267                 uint32_t *ipv4_flag)
1268 {
1269         struct ipv4_hdr *ipv4_hdr;
1270         struct ether_hdr *eth_hdr;
1271         uint32_t x0, x1, x2, x3;
1272
1273         eth_hdr = rte_pktmbuf_mtod(pkt[0], struct ether_hdr *);
1274         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1275         x0 = ipv4_hdr->dst_addr;
1276         ipv4_flag[0] = pkt[0]->packet_type & RTE_PTYPE_L3_IPV4;
1277
1278         eth_hdr = rte_pktmbuf_mtod(pkt[1], struct ether_hdr *);
1279         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1280         x1 = ipv4_hdr->dst_addr;
1281         ipv4_flag[0] &= pkt[1]->packet_type;
1282
1283         eth_hdr = rte_pktmbuf_mtod(pkt[2], struct ether_hdr *);
1284         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1285         x2 = ipv4_hdr->dst_addr;
1286         ipv4_flag[0] &= pkt[2]->packet_type;
1287
1288         eth_hdr = rte_pktmbuf_mtod(pkt[3], struct ether_hdr *);
1289         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1290         x3 = ipv4_hdr->dst_addr;
1291         ipv4_flag[0] &= pkt[3]->packet_type;
1292
1293         dip[0] = _mm_set_epi32(x3, x2, x1, x0);
1294 }
1295 #else /* RTE_NEXT_ABI */
1296 /*
1297  * Read ol_flags and destination IPV4 addresses from 4 mbufs.
1298  */
1299 static inline void
1300 processx4_step1(struct rte_mbuf *pkt[FWDSTEP], __m128i *dip, uint32_t *flag)
1301 {
1302         struct ipv4_hdr *ipv4_hdr;
1303         struct ether_hdr *eth_hdr;
1304         uint32_t x0, x1, x2, x3;
1305
1306         eth_hdr = rte_pktmbuf_mtod(pkt[0], struct ether_hdr *);
1307         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1308         x0 = ipv4_hdr->dst_addr;
1309         flag[0] = pkt[0]->ol_flags & PKT_RX_IPV4_HDR;
1310
1311         eth_hdr = rte_pktmbuf_mtod(pkt[1], struct ether_hdr *);
1312         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1313         x1 = ipv4_hdr->dst_addr;
1314         flag[0] &= pkt[1]->ol_flags;
1315
1316         eth_hdr = rte_pktmbuf_mtod(pkt[2], struct ether_hdr *);
1317         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1318         x2 = ipv4_hdr->dst_addr;
1319         flag[0] &= pkt[2]->ol_flags;
1320
1321         eth_hdr = rte_pktmbuf_mtod(pkt[3], struct ether_hdr *);
1322         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1323         x3 = ipv4_hdr->dst_addr;
1324         flag[0] &= pkt[3]->ol_flags;
1325
1326         dip[0] = _mm_set_epi32(x3, x2, x1, x0);
1327 }
1328 #endif /* RTE_NEXT_ABI */
1329
1330 /*
1331  * Lookup into LPM for destination port.
1332  * If lookup fails, use incoming port (portid) as destination port.
1333  */
1334 static inline void
1335 #ifdef RTE_NEXT_ABI
1336 processx4_step2(const struct lcore_conf *qconf,
1337                 __m128i dip,
1338                 uint32_t ipv4_flag,
1339                 uint8_t portid,
1340                 struct rte_mbuf *pkt[FWDSTEP],
1341                 uint16_t dprt[FWDSTEP])
1342 #else
1343 processx4_step2(const struct lcore_conf *qconf, __m128i dip, uint32_t flag,
1344         uint8_t portid, struct rte_mbuf *pkt[FWDSTEP], uint16_t dprt[FWDSTEP])
1345 #endif /* RTE_NEXT_ABI */
1346 {
1347         rte_xmm_t dst;
1348         const  __m128i bswap_mask = _mm_set_epi8(12, 13, 14, 15, 8, 9, 10, 11,
1349                                                 4, 5, 6, 7, 0, 1, 2, 3);
1350
1351         /* Byte swap 4 IPV4 addresses. */
1352         dip = _mm_shuffle_epi8(dip, bswap_mask);
1353
1354         /* if all 4 packets are IPV4. */
1355 #ifdef RTE_NEXT_ABI
1356         if (likely(ipv4_flag)) {
1357 #else
1358         if (likely(flag != 0)) {
1359 #endif
1360                 rte_lpm_lookupx4(qconf->ipv4_lookup_struct, dip, dprt, portid);
1361         } else {
1362                 dst.x = dip;
1363                 dprt[0] = get_dst_port(qconf, pkt[0], dst.u32[0], portid);
1364                 dprt[1] = get_dst_port(qconf, pkt[1], dst.u32[1], portid);
1365                 dprt[2] = get_dst_port(qconf, pkt[2], dst.u32[2], portid);
1366                 dprt[3] = get_dst_port(qconf, pkt[3], dst.u32[3], portid);
1367         }
1368 }
1369
1370 /*
1371  * Update source and destination MAC addresses in the ethernet header.
1372  * Perform RFC1812 checks and updates for IPV4 packets.
1373  */
1374 static inline void
1375 processx4_step3(struct rte_mbuf *pkt[FWDSTEP], uint16_t dst_port[FWDSTEP])
1376 {
1377         __m128i te[FWDSTEP];
1378         __m128i ve[FWDSTEP];
1379         __m128i *p[FWDSTEP];
1380
1381         p[0] = rte_pktmbuf_mtod(pkt[0], __m128i *);
1382         p[1] = rte_pktmbuf_mtod(pkt[1], __m128i *);
1383         p[2] = rte_pktmbuf_mtod(pkt[2], __m128i *);
1384         p[3] = rte_pktmbuf_mtod(pkt[3], __m128i *);
1385
1386         ve[0] = val_eth[dst_port[0]];
1387         te[0] = _mm_load_si128(p[0]);
1388
1389         ve[1] = val_eth[dst_port[1]];
1390         te[1] = _mm_load_si128(p[1]);
1391
1392         ve[2] = val_eth[dst_port[2]];
1393         te[2] = _mm_load_si128(p[2]);
1394
1395         ve[3] = val_eth[dst_port[3]];
1396         te[3] = _mm_load_si128(p[3]);
1397
1398         /* Update first 12 bytes, keep rest bytes intact. */
1399         te[0] =  _mm_blend_epi16(te[0], ve[0], MASK_ETH);
1400         te[1] =  _mm_blend_epi16(te[1], ve[1], MASK_ETH);
1401         te[2] =  _mm_blend_epi16(te[2], ve[2], MASK_ETH);
1402         te[3] =  _mm_blend_epi16(te[3], ve[3], MASK_ETH);
1403
1404         _mm_store_si128(p[0], te[0]);
1405         _mm_store_si128(p[1], te[1]);
1406         _mm_store_si128(p[2], te[2]);
1407         _mm_store_si128(p[3], te[3]);
1408
1409 #ifdef RTE_NEXT_ABI
1410         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[0] + 1),
1411                 &dst_port[0], pkt[0]->packet_type);
1412         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[1] + 1),
1413                 &dst_port[1], pkt[1]->packet_type);
1414         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[2] + 1),
1415                 &dst_port[2], pkt[2]->packet_type);
1416         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[3] + 1),
1417                 &dst_port[3], pkt[3]->packet_type);
1418 #else /* RTE_NEXT_ABI */
1419         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[0] + 1),
1420                 &dst_port[0], pkt[0]->ol_flags);
1421         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[1] + 1),
1422                 &dst_port[1], pkt[1]->ol_flags);
1423         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[2] + 1),
1424                 &dst_port[2], pkt[2]->ol_flags);
1425         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[3] + 1),
1426                 &dst_port[3], pkt[3]->ol_flags);
1427 #endif /* RTE_NEXT_ABI */
1428 }
1429
1430 /*
1431  * We group consecutive packets with the same destionation port into one burst.
1432  * To avoid extra latency this is done together with some other packet
1433  * processing, but after we made a final decision about packet's destination.
1434  * To do this we maintain:
1435  * pnum - array of number of consecutive packets with the same dest port for
1436  * each packet in the input burst.
1437  * lp - pointer to the last updated element in the pnum.
1438  * dlp - dest port value lp corresponds to.
1439  */
1440
1441 #define GRPSZ   (1 << FWDSTEP)
1442 #define GRPMSK  (GRPSZ - 1)
1443
1444 #define GROUP_PORT_STEP(dlp, dcp, lp, pn, idx)  do { \
1445         if (likely((dlp) == (dcp)[(idx)])) {         \
1446                 (lp)[0]++;                           \
1447         } else {                                     \
1448                 (dlp) = (dcp)[idx];                  \
1449                 (lp) = (pn) + (idx);                 \
1450                 (lp)[0] = 1;                         \
1451         }                                            \
1452 } while (0)
1453
1454 /*
1455  * Group consecutive packets with the same destination port in bursts of 4.
1456  * Suppose we have array of destionation ports:
1457  * dst_port[] = {a, b, c, d,, e, ... }
1458  * dp1 should contain: <a, b, c, d>, dp2: <b, c, d, e>.
1459  * We doing 4 comparisions at once and the result is 4 bit mask.
1460  * This mask is used as an index into prebuild array of pnum values.
1461  */
1462 static inline uint16_t *
1463 port_groupx4(uint16_t pn[FWDSTEP + 1], uint16_t *lp, __m128i dp1, __m128i dp2)
1464 {
1465         static const struct {
1466                 uint64_t pnum; /* prebuild 4 values for pnum[]. */
1467                 int32_t  idx;  /* index for new last updated elemnet. */
1468                 uint16_t lpv;  /* add value to the last updated element. */
1469         } gptbl[GRPSZ] = {
1470         {
1471                 /* 0: a != b, b != c, c != d, d != e */
1472                 .pnum = UINT64_C(0x0001000100010001),
1473                 .idx = 4,
1474                 .lpv = 0,
1475         },
1476         {
1477                 /* 1: a == b, b != c, c != d, d != e */
1478                 .pnum = UINT64_C(0x0001000100010002),
1479                 .idx = 4,
1480                 .lpv = 1,
1481         },
1482         {
1483                 /* 2: a != b, b == c, c != d, d != e */
1484                 .pnum = UINT64_C(0x0001000100020001),
1485                 .idx = 4,
1486                 .lpv = 0,
1487         },
1488         {
1489                 /* 3: a == b, b == c, c != d, d != e */
1490                 .pnum = UINT64_C(0x0001000100020003),
1491                 .idx = 4,
1492                 .lpv = 2,
1493         },
1494         {
1495                 /* 4: a != b, b != c, c == d, d != e */
1496                 .pnum = UINT64_C(0x0001000200010001),
1497                 .idx = 4,
1498                 .lpv = 0,
1499         },
1500         {
1501                 /* 5: a == b, b != c, c == d, d != e */
1502                 .pnum = UINT64_C(0x0001000200010002),
1503                 .idx = 4,
1504                 .lpv = 1,
1505         },
1506         {
1507                 /* 6: a != b, b == c, c == d, d != e */
1508                 .pnum = UINT64_C(0x0001000200030001),
1509                 .idx = 4,
1510                 .lpv = 0,
1511         },
1512         {
1513                 /* 7: a == b, b == c, c == d, d != e */
1514                 .pnum = UINT64_C(0x0001000200030004),
1515                 .idx = 4,
1516                 .lpv = 3,
1517         },
1518         {
1519                 /* 8: a != b, b != c, c != d, d == e */
1520                 .pnum = UINT64_C(0x0002000100010001),
1521                 .idx = 3,
1522                 .lpv = 0,
1523         },
1524         {
1525                 /* 9: a == b, b != c, c != d, d == e */
1526                 .pnum = UINT64_C(0x0002000100010002),
1527                 .idx = 3,
1528                 .lpv = 1,
1529         },
1530         {
1531                 /* 0xa: a != b, b == c, c != d, d == e */
1532                 .pnum = UINT64_C(0x0002000100020001),
1533                 .idx = 3,
1534                 .lpv = 0,
1535         },
1536         {
1537                 /* 0xb: a == b, b == c, c != d, d == e */
1538                 .pnum = UINT64_C(0x0002000100020003),
1539                 .idx = 3,
1540                 .lpv = 2,
1541         },
1542         {
1543                 /* 0xc: a != b, b != c, c == d, d == e */
1544                 .pnum = UINT64_C(0x0002000300010001),
1545                 .idx = 2,
1546                 .lpv = 0,
1547         },
1548         {
1549                 /* 0xd: a == b, b != c, c == d, d == e */
1550                 .pnum = UINT64_C(0x0002000300010002),
1551                 .idx = 2,
1552                 .lpv = 1,
1553         },
1554         {
1555                 /* 0xe: a != b, b == c, c == d, d == e */
1556                 .pnum = UINT64_C(0x0002000300040001),
1557                 .idx = 1,
1558                 .lpv = 0,
1559         },
1560         {
1561                 /* 0xf: a == b, b == c, c == d, d == e */
1562                 .pnum = UINT64_C(0x0002000300040005),
1563                 .idx = 0,
1564                 .lpv = 4,
1565         },
1566         };
1567
1568         union {
1569                 uint16_t u16[FWDSTEP + 1];
1570                 uint64_t u64;
1571         } *pnum = (void *)pn;
1572
1573         int32_t v;
1574
1575         dp1 = _mm_cmpeq_epi16(dp1, dp2);
1576         dp1 = _mm_unpacklo_epi16(dp1, dp1);
1577         v = _mm_movemask_ps((__m128)dp1);
1578
1579         /* update last port counter. */
1580         lp[0] += gptbl[v].lpv;
1581
1582         /* if dest port value has changed. */
1583         if (v != GRPMSK) {
1584                 lp = pnum->u16 + gptbl[v].idx;
1585                 lp[0] = 1;
1586                 pnum->u64 = gptbl[v].pnum;
1587         }
1588
1589         return lp;
1590 }
1591
1592 #endif /* APP_LOOKUP_METHOD */
1593
1594 /* main processing loop */
1595 static int
1596 main_loop(__attribute__((unused)) void *dummy)
1597 {
1598         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1599         unsigned lcore_id;
1600         uint64_t prev_tsc, diff_tsc, cur_tsc;
1601         int i, j, nb_rx;
1602         uint8_t portid, queueid;
1603         struct lcore_conf *qconf;
1604         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1605                 US_PER_S * BURST_TX_DRAIN_US;
1606
1607 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1608         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1609         int32_t k;
1610         uint16_t dlp;
1611         uint16_t *lp;
1612         uint16_t dst_port[MAX_PKT_BURST];
1613         __m128i dip[MAX_PKT_BURST / FWDSTEP];
1614 #ifdef RTE_NEXT_ABI
1615         uint32_t ipv4_flag[MAX_PKT_BURST / FWDSTEP];
1616 #else
1617         uint32_t flag[MAX_PKT_BURST / FWDSTEP];
1618 #endif
1619         uint16_t pnum[MAX_PKT_BURST + 1];
1620 #endif
1621
1622         prev_tsc = 0;
1623
1624         lcore_id = rte_lcore_id();
1625         qconf = &lcore_conf[lcore_id];
1626
1627         if (qconf->n_rx_queue == 0) {
1628                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
1629                 return 0;
1630         }
1631
1632         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
1633
1634         for (i = 0; i < qconf->n_rx_queue; i++) {
1635
1636                 portid = qconf->rx_queue_list[i].port_id;
1637                 queueid = qconf->rx_queue_list[i].queue_id;
1638                 RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n", lcore_id,
1639                         portid, queueid);
1640         }
1641
1642         while (1) {
1643
1644                 cur_tsc = rte_rdtsc();
1645
1646                 /*
1647                  * TX burst queue drain
1648                  */
1649                 diff_tsc = cur_tsc - prev_tsc;
1650                 if (unlikely(diff_tsc > drain_tsc)) {
1651
1652                         /*
1653                          * This could be optimized (use queueid instead of
1654                          * portid), but it is not called so often
1655                          */
1656                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
1657                                 if (qconf->tx_mbufs[portid].len == 0)
1658                                         continue;
1659                                 send_burst(qconf,
1660                                         qconf->tx_mbufs[portid].len,
1661                                         portid);
1662                                 qconf->tx_mbufs[portid].len = 0;
1663                         }
1664
1665                         prev_tsc = cur_tsc;
1666                 }
1667
1668                 /*
1669                  * Read packet from RX queues
1670                  */
1671                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1672                         portid = qconf->rx_queue_list[i].port_id;
1673                         queueid = qconf->rx_queue_list[i].queue_id;
1674                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1675                                 MAX_PKT_BURST);
1676                         if (nb_rx == 0)
1677                                 continue;
1678
1679 #if (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
1680 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1681                         {
1682                                 /*
1683                                  * Send nb_rx - nb_rx%8 packets
1684                                  * in groups of 8.
1685                                  */
1686                                 int32_t n = RTE_ALIGN_FLOOR(nb_rx, 8);
1687                                 for (j = 0; j < n; j += 8) {
1688 #ifdef RTE_NEXT_ABI
1689                                         uint32_t pkt_type =
1690                                                 pkts_burst[j]->packet_type &
1691                                                 pkts_burst[j+1]->packet_type &
1692                                                 pkts_burst[j+2]->packet_type &
1693                                                 pkts_burst[j+3]->packet_type &
1694                                                 pkts_burst[j+4]->packet_type &
1695                                                 pkts_burst[j+5]->packet_type &
1696                                                 pkts_burst[j+6]->packet_type &
1697                                                 pkts_burst[j+7]->packet_type;
1698                                         if (pkt_type & RTE_PTYPE_L3_IPV4) {
1699                                                 simple_ipv4_fwd_8pkts(
1700                                                 &pkts_burst[j], portid, qconf);
1701                                         } else if (pkt_type &
1702                                                 RTE_PTYPE_L3_IPV6) {
1703 #else /* RTE_NEXT_ABI */
1704                                         uint32_t ol_flag = pkts_burst[j]->ol_flags
1705                                                         & pkts_burst[j+1]->ol_flags
1706                                                         & pkts_burst[j+2]->ol_flags
1707                                                         & pkts_burst[j+3]->ol_flags
1708                                                         & pkts_burst[j+4]->ol_flags
1709                                                         & pkts_burst[j+5]->ol_flags
1710                                                         & pkts_burst[j+6]->ol_flags
1711                                                         & pkts_burst[j+7]->ol_flags;
1712                                         if (ol_flag & PKT_RX_IPV4_HDR ) {
1713                                                 simple_ipv8_fwd_4pkts(&pkts_burst[j],
1714                                                                         portid, qconf);
1715                                         } else if (ol_flag & PKT_RX_IPV6_HDR) {
1716 #endif /* RTE_NEXT_ABI */
1717                                                 simple_ipv6_fwd_4pkts(&pkts_burst[j],
1718                                                                         portid, qconf);
1719                                         } else {
1720                                                 l3fwd_simple_forward(pkts_burst[j],
1721                                                                         portid, qconf);
1722                                                 l3fwd_simple_forward(pkts_burst[j+1],
1723                                                                         portid, qconf);
1724                                                 l3fwd_simple_forward(pkts_burst[j+2],
1725                                                                         portid, qconf);
1726                                                 l3fwd_simple_forward(pkts_burst[j+3],
1727                                                                         portid, qconf);
1728                                                 l3fwd_simple_forward(pkts_burst[j+4],
1729                                                                         portid, qconf);
1730                                                 l3fwd_simple_forward(pkts_burst[j+5],
1731                                                                         portid, qconf);
1732                                                 l3fwd_simple_forward(pkts_burst[j+6],
1733                                                                         portid, qconf);
1734                                                 l3fwd_simple_forward(pkts_burst[j+7],
1735                                                                         portid, qconf);
1736                                         }
1737                                 }
1738                                 for (; j < nb_rx ; j++) {
1739                                         l3fwd_simple_forward(pkts_burst[j],
1740                                                                 portid, qconf);
1741                                 }
1742                         }
1743 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1744
1745                         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1746                         for (j = 0; j != k; j += FWDSTEP) {
1747                                 processx4_step1(&pkts_burst[j],
1748                                         &dip[j / FWDSTEP],
1749 #ifdef RTE_NEXT_ABI
1750                                         &ipv4_flag[j / FWDSTEP]);
1751 #else
1752                                         &flag[j / FWDSTEP]);
1753 #endif
1754                         }
1755
1756                         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1757                         for (j = 0; j != k; j += FWDSTEP) {
1758                                 processx4_step2(qconf, dip[j / FWDSTEP],
1759 #ifdef RTE_NEXT_ABI
1760                                         ipv4_flag[j / FWDSTEP], portid,
1761 #else
1762                                         flag[j / FWDSTEP], portid,
1763 #endif
1764                                         &pkts_burst[j], &dst_port[j]);
1765                         }
1766
1767                         /*
1768                          * Finish packet processing and group consecutive
1769                          * packets with the same destination port.
1770                          */
1771                         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1772                         if (k != 0) {
1773                                 __m128i dp1, dp2;
1774
1775                                 lp = pnum;
1776                                 lp[0] = 1;
1777
1778                                 processx4_step3(pkts_burst, dst_port);
1779
1780                                 /* dp1: <d[0], d[1], d[2], d[3], ... > */
1781                                 dp1 = _mm_loadu_si128((__m128i *)dst_port);
1782
1783                                 for (j = FWDSTEP; j != k; j += FWDSTEP) {
1784                                         processx4_step3(&pkts_burst[j],
1785                                                 &dst_port[j]);
1786
1787                                         /*
1788                                          * dp2:
1789                                          * <d[j-3], d[j-2], d[j-1], d[j], ... >
1790                                          */
1791                                         dp2 = _mm_loadu_si128((__m128i *)
1792                                                 &dst_port[j - FWDSTEP + 1]);
1793                                         lp  = port_groupx4(&pnum[j - FWDSTEP],
1794                                                 lp, dp1, dp2);
1795
1796                                         /*
1797                                          * dp1:
1798                                          * <d[j], d[j+1], d[j+2], d[j+3], ... >
1799                                          */
1800                                         dp1 = _mm_srli_si128(dp2,
1801                                                 (FWDSTEP - 1) *
1802                                                 sizeof(dst_port[0]));
1803                                 }
1804
1805                                 /*
1806                                  * dp2: <d[j-3], d[j-2], d[j-1], d[j-1], ... >
1807                                  */
1808                                 dp2 = _mm_shufflelo_epi16(dp1, 0xf9);
1809                                 lp  = port_groupx4(&pnum[j - FWDSTEP], lp,
1810                                         dp1, dp2);
1811
1812                                 /*
1813                                  * remove values added by the last repeated
1814                                  * dst port.
1815                                  */
1816                                 lp[0]--;
1817                                 dlp = dst_port[j - 1];
1818                         } else {
1819                                 /* set dlp and lp to the never used values. */
1820                                 dlp = BAD_PORT - 1;
1821                                 lp = pnum + MAX_PKT_BURST;
1822                         }
1823
1824                         /* Process up to last 3 packets one by one. */
1825                         switch (nb_rx % FWDSTEP) {
1826                         case 3:
1827                                 process_packet(qconf, pkts_burst[j],
1828                                         dst_port + j, portid);
1829                                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1830                                 j++;
1831                         case 2:
1832                                 process_packet(qconf, pkts_burst[j],
1833                                         dst_port + j, portid);
1834                                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1835                                 j++;
1836                         case 1:
1837                                 process_packet(qconf, pkts_burst[j],
1838                                         dst_port + j, portid);
1839                                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1840                                 j++;
1841                         }
1842
1843                         /*
1844                          * Send packets out, through destination port.
1845                          * Consecuteve pacekts with the same destination port
1846                          * are already grouped together.
1847                          * If destination port for the packet equals BAD_PORT,
1848                          * then free the packet without sending it out.
1849                          */
1850                         for (j = 0; j < nb_rx; j += k) {
1851
1852                                 int32_t m;
1853                                 uint16_t pn;
1854
1855                                 pn = dst_port[j];
1856                                 k = pnum[j];
1857
1858                                 if (likely(pn != BAD_PORT)) {
1859                                         send_packetsx4(qconf, pn,
1860                                                 pkts_burst + j, k);
1861                                 } else {
1862                                         for (m = j; m != j + k; m++)
1863                                                 rte_pktmbuf_free(pkts_burst[m]);
1864                                 }
1865                         }
1866
1867 #endif /* APP_LOOKUP_METHOD */
1868 #else /* ENABLE_MULTI_BUFFER_OPTIMIZE == 0 */
1869
1870                         /* Prefetch first packets */
1871                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1872                                 rte_prefetch0(rte_pktmbuf_mtod(
1873                                                 pkts_burst[j], void *));
1874                         }
1875
1876                         /* Prefetch and forward already prefetched packets */
1877                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1878                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1879                                                 j + PREFETCH_OFFSET], void *));
1880                                 l3fwd_simple_forward(pkts_burst[j], portid,
1881                                         qconf);
1882                         }
1883
1884                         /* Forward remaining prefetched packets */
1885                         for (; j < nb_rx; j++) {
1886                                 l3fwd_simple_forward(pkts_burst[j], portid,
1887                                         qconf);
1888                         }
1889 #endif /* ENABLE_MULTI_BUFFER_OPTIMIZE */
1890
1891                 }
1892         }
1893 }
1894
1895 static int
1896 check_lcore_params(void)
1897 {
1898         uint8_t queue, lcore;
1899         uint16_t i;
1900         int socketid;
1901
1902         for (i = 0; i < nb_lcore_params; ++i) {
1903                 queue = lcore_params[i].queue_id;
1904                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1905                         printf("invalid queue number: %hhu\n", queue);
1906                         return -1;
1907                 }
1908                 lcore = lcore_params[i].lcore_id;
1909                 if (!rte_lcore_is_enabled(lcore)) {
1910                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
1911                         return -1;
1912                 }
1913                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1914                         (numa_on == 0)) {
1915                         printf("warning: lcore %hhu is on socket %d with numa off \n",
1916                                 lcore, socketid);
1917                 }
1918         }
1919         return 0;
1920 }
1921
1922 static int
1923 check_port_config(const unsigned nb_ports)
1924 {
1925         unsigned portid;
1926         uint16_t i;
1927
1928         for (i = 0; i < nb_lcore_params; ++i) {
1929                 portid = lcore_params[i].port_id;
1930                 if ((enabled_port_mask & (1 << portid)) == 0) {
1931                         printf("port %u is not enabled in port mask\n", portid);
1932                         return -1;
1933                 }
1934                 if (portid >= nb_ports) {
1935                         printf("port %u is not present on the board\n", portid);
1936                         return -1;
1937                 }
1938         }
1939         return 0;
1940 }
1941
1942 static uint8_t
1943 get_port_n_rx_queues(const uint8_t port)
1944 {
1945         int queue = -1;
1946         uint16_t i;
1947
1948         for (i = 0; i < nb_lcore_params; ++i) {
1949                 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
1950                         queue = lcore_params[i].queue_id;
1951         }
1952         return (uint8_t)(++queue);
1953 }
1954
1955 static int
1956 init_lcore_rx_queues(void)
1957 {
1958         uint16_t i, nb_rx_queue;
1959         uint8_t lcore;
1960
1961         for (i = 0; i < nb_lcore_params; ++i) {
1962                 lcore = lcore_params[i].lcore_id;
1963                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1964                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1965                         printf("error: too many queues (%u) for lcore: %u\n",
1966                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1967                         return -1;
1968                 } else {
1969                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1970                                 lcore_params[i].port_id;
1971                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1972                                 lcore_params[i].queue_id;
1973                         lcore_conf[lcore].n_rx_queue++;
1974                 }
1975         }
1976         return 0;
1977 }
1978
1979 /* display usage */
1980 static void
1981 print_usage(const char *prgname)
1982 {
1983         printf ("%s [EAL options] -- -p PORTMASK -P"
1984                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1985                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1986                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1987                 "  -P : enable promiscuous mode\n"
1988                 "  --config (port,queue,lcore): rx queues configuration\n"
1989                 "  --eth-dest=X,MM:MM:MM:MM:MM:MM: optional, ethernet destination for port X\n"
1990                 "  --no-numa: optional, disable numa awareness\n"
1991                 "  --ipv6: optional, specify it if running ipv6 packets\n"
1992                 "  --enable-jumbo: enable jumbo frame"
1993                 " which max packet len is PKTLEN in decimal (64-9600)\n"
1994                 "  --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n",
1995                 prgname);
1996 }
1997
1998 static int parse_max_pkt_len(const char *pktlen)
1999 {
2000         char *end = NULL;
2001         unsigned long len;
2002
2003         /* parse decimal string */
2004         len = strtoul(pktlen, &end, 10);
2005         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
2006                 return -1;
2007
2008         if (len == 0)
2009                 return -1;
2010
2011         return len;
2012 }
2013
2014 static int
2015 parse_portmask(const char *portmask)
2016 {
2017         char *end = NULL;
2018         unsigned long pm;
2019
2020         /* parse hexadecimal string */
2021         pm = strtoul(portmask, &end, 16);
2022         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
2023                 return -1;
2024
2025         if (pm == 0)
2026                 return -1;
2027
2028         return pm;
2029 }
2030
2031 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2032 static int
2033 parse_hash_entry_number(const char *hash_entry_num)
2034 {
2035         char *end = NULL;
2036         unsigned long hash_en;
2037         /* parse hexadecimal string */
2038         hash_en = strtoul(hash_entry_num, &end, 16);
2039         if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
2040                 return -1;
2041
2042         if (hash_en == 0)
2043                 return -1;
2044
2045         return hash_en;
2046 }
2047 #endif
2048
2049 static int
2050 parse_config(const char *q_arg)
2051 {
2052         char s[256];
2053         const char *p, *p0 = q_arg;
2054         char *end;
2055         enum fieldnames {
2056                 FLD_PORT = 0,
2057                 FLD_QUEUE,
2058                 FLD_LCORE,
2059                 _NUM_FLD
2060         };
2061         unsigned long int_fld[_NUM_FLD];
2062         char *str_fld[_NUM_FLD];
2063         int i;
2064         unsigned size;
2065
2066         nb_lcore_params = 0;
2067
2068         while ((p = strchr(p0,'(')) != NULL) {
2069                 ++p;
2070                 if((p0 = strchr(p,')')) == NULL)
2071                         return -1;
2072
2073                 size = p0 - p;
2074                 if(size >= sizeof(s))
2075                         return -1;
2076
2077                 snprintf(s, sizeof(s), "%.*s", size, p);
2078                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2079                         return -1;
2080                 for (i = 0; i < _NUM_FLD; i++){
2081                         errno = 0;
2082                         int_fld[i] = strtoul(str_fld[i], &end, 0);
2083                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2084                                 return -1;
2085                 }
2086                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
2087                         printf("exceeded max number of lcore params: %hu\n",
2088                                 nb_lcore_params);
2089                         return -1;
2090                 }
2091                 lcore_params_array[nb_lcore_params].port_id = (uint8_t)int_fld[FLD_PORT];
2092                 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
2093                 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
2094                 ++nb_lcore_params;
2095         }
2096         lcore_params = lcore_params_array;
2097         return 0;
2098 }
2099
2100 static void
2101 parse_eth_dest(const char *optarg)
2102 {
2103         uint8_t portid;
2104         char *port_end;
2105         uint8_t c, *dest, peer_addr[6];
2106
2107         errno = 0;
2108         portid = strtoul(optarg, &port_end, 10);
2109         if (errno != 0 || port_end == optarg || *port_end++ != ',')
2110                 rte_exit(EXIT_FAILURE,
2111                 "Invalid eth-dest: %s", optarg);
2112         if (portid >= RTE_MAX_ETHPORTS)
2113                 rte_exit(EXIT_FAILURE,
2114                 "eth-dest: port %d >= RTE_MAX_ETHPORTS(%d)\n",
2115                 portid, RTE_MAX_ETHPORTS);
2116
2117         if (cmdline_parse_etheraddr(NULL, port_end,
2118                 &peer_addr, sizeof(peer_addr)) < 0)
2119                 rte_exit(EXIT_FAILURE,
2120                 "Invalid ethernet address: %s\n",
2121                 port_end);
2122         dest = (uint8_t *)&dest_eth_addr[portid];
2123         for (c = 0; c < 6; c++)
2124                 dest[c] = peer_addr[c];
2125         *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
2126 }
2127
2128 #define CMD_LINE_OPT_CONFIG "config"
2129 #define CMD_LINE_OPT_ETH_DEST "eth-dest"
2130 #define CMD_LINE_OPT_NO_NUMA "no-numa"
2131 #define CMD_LINE_OPT_IPV6 "ipv6"
2132 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
2133 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
2134
2135 /* Parse the argument given in the command line of the application */
2136 static int
2137 parse_args(int argc, char **argv)
2138 {
2139         int opt, ret;
2140         char **argvopt;
2141         int option_index;
2142         char *prgname = argv[0];
2143         static struct option lgopts[] = {
2144                 {CMD_LINE_OPT_CONFIG, 1, 0, 0},
2145                 {CMD_LINE_OPT_ETH_DEST, 1, 0, 0},
2146                 {CMD_LINE_OPT_NO_NUMA, 0, 0, 0},
2147                 {CMD_LINE_OPT_IPV6, 0, 0, 0},
2148                 {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, 0},
2149                 {CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, 0},
2150                 {NULL, 0, 0, 0}
2151         };
2152
2153         argvopt = argv;
2154
2155         while ((opt = getopt_long(argc, argvopt, "p:P",
2156                                 lgopts, &option_index)) != EOF) {
2157
2158                 switch (opt) {
2159                 /* portmask */
2160                 case 'p':
2161                         enabled_port_mask = parse_portmask(optarg);
2162                         if (enabled_port_mask == 0) {
2163                                 printf("invalid portmask\n");
2164                                 print_usage(prgname);
2165                                 return -1;
2166                         }
2167                         break;
2168                 case 'P':
2169                         printf("Promiscuous mode selected\n");
2170                         promiscuous_on = 1;
2171                         break;
2172
2173                 /* long options */
2174                 case 0:
2175                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_CONFIG,
2176                                 sizeof (CMD_LINE_OPT_CONFIG))) {
2177                                 ret = parse_config(optarg);
2178                                 if (ret) {
2179                                         printf("invalid config\n");
2180                                         print_usage(prgname);
2181                                         return -1;
2182                                 }
2183                         }
2184
2185                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ETH_DEST,
2186                                 sizeof(CMD_LINE_OPT_CONFIG))) {
2187                                         parse_eth_dest(optarg);
2188                         }
2189
2190                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_NUMA,
2191                                 sizeof(CMD_LINE_OPT_NO_NUMA))) {
2192                                 printf("numa is disabled \n");
2193                                 numa_on = 0;
2194                         }
2195
2196 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2197                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_IPV6,
2198                                 sizeof(CMD_LINE_OPT_IPV6))) {
2199                                 printf("ipv6 is specified \n");
2200                                 ipv6 = 1;
2201                         }
2202 #endif
2203
2204                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ENABLE_JUMBO,
2205                                 sizeof (CMD_LINE_OPT_ENABLE_JUMBO))) {
2206                                 struct option lenopts = {"max-pkt-len", required_argument, 0, 0};
2207
2208                                 printf("jumbo frame is enabled - disabling simple TX path\n");
2209                                 port_conf.rxmode.jumbo_frame = 1;
2210
2211                                 /* if no max-pkt-len set, use the default value ETHER_MAX_LEN */
2212                                 if (0 == getopt_long(argc, argvopt, "", &lenopts, &option_index)) {
2213                                         ret = parse_max_pkt_len(optarg);
2214                                         if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)){
2215                                                 printf("invalid packet length\n");
2216                                                 print_usage(prgname);
2217                                                 return -1;
2218                                         }
2219                                         port_conf.rxmode.max_rx_pkt_len = ret;
2220                                 }
2221                                 printf("set jumbo frame max packet length to %u\n",
2222                                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
2223                         }
2224 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2225                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_HASH_ENTRY_NUM,
2226                                 sizeof(CMD_LINE_OPT_HASH_ENTRY_NUM))) {
2227                                 ret = parse_hash_entry_number(optarg);
2228                                 if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
2229                                         hash_entry_number = ret;
2230                                 } else {
2231                                         printf("invalid hash entry number\n");
2232                                         print_usage(prgname);
2233                                         return -1;
2234                                 }
2235                         }
2236 #endif
2237                         break;
2238
2239                 default:
2240                         print_usage(prgname);
2241                         return -1;
2242                 }
2243         }
2244
2245         if (optind >= 0)
2246                 argv[optind-1] = prgname;
2247
2248         ret = optind-1;
2249         optind = 0; /* reset getopt lib */
2250         return ret;
2251 }
2252
2253 static void
2254 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
2255 {
2256         char buf[ETHER_ADDR_FMT_SIZE];
2257         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
2258         printf("%s%s", name, buf);
2259 }
2260
2261 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2262
2263 static void convert_ipv4_5tuple(struct ipv4_5tuple* key1,
2264                 union ipv4_5tuple_host* key2)
2265 {
2266         key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
2267         key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
2268         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
2269         key2->port_src = rte_cpu_to_be_16(key1->port_src);
2270         key2->proto = key1->proto;
2271         key2->pad0 = 0;
2272         key2->pad1 = 0;
2273         return;
2274 }
2275
2276 static void convert_ipv6_5tuple(struct ipv6_5tuple* key1,
2277                 union ipv6_5tuple_host* key2)
2278 {
2279         uint32_t i;
2280         for (i = 0; i < 16; i++)
2281         {
2282                 key2->ip_dst[i] = key1->ip_dst[i];
2283                 key2->ip_src[i] = key1->ip_src[i];
2284         }
2285         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
2286         key2->port_src = rte_cpu_to_be_16(key1->port_src);
2287         key2->proto = key1->proto;
2288         key2->pad0 = 0;
2289         key2->pad1 = 0;
2290         key2->reserve = 0;
2291         return;
2292 }
2293
2294 #define BYTE_VALUE_MAX 256
2295 #define ALL_32_BITS 0xffffffff
2296 #define BIT_8_TO_15 0x0000ff00
2297 static inline void
2298 populate_ipv4_few_flow_into_table(const struct rte_hash* h)
2299 {
2300         uint32_t i;
2301         int32_t ret;
2302         uint32_t array_len = sizeof(ipv4_l3fwd_route_array)/sizeof(ipv4_l3fwd_route_array[0]);
2303
2304         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
2305         for (i = 0; i < array_len; i++) {
2306                 struct ipv4_l3fwd_route  entry;
2307                 union ipv4_5tuple_host newkey;
2308                 entry = ipv4_l3fwd_route_array[i];
2309                 convert_ipv4_5tuple(&entry.key, &newkey);
2310                 ret = rte_hash_add_key (h,(void *) &newkey);
2311                 if (ret < 0) {
2312                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
2313                                 " to the l3fwd hash.\n", i);
2314                 }
2315                 ipv4_l3fwd_out_if[ret] = entry.if_out;
2316         }
2317         printf("Hash: Adding 0x%" PRIx32 " keys\n", array_len);
2318 }
2319
2320 #define BIT_16_TO_23 0x00ff0000
2321 static inline void
2322 populate_ipv6_few_flow_into_table(const struct rte_hash* h)
2323 {
2324         uint32_t i;
2325         int32_t ret;
2326         uint32_t array_len = sizeof(ipv6_l3fwd_route_array)/sizeof(ipv6_l3fwd_route_array[0]);
2327
2328         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
2329         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
2330         for (i = 0; i < array_len; i++) {
2331                 struct ipv6_l3fwd_route entry;
2332                 union ipv6_5tuple_host newkey;
2333                 entry = ipv6_l3fwd_route_array[i];
2334                 convert_ipv6_5tuple(&entry.key, &newkey);
2335                 ret = rte_hash_add_key (h, (void *) &newkey);
2336                 if (ret < 0) {
2337                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
2338                                 " to the l3fwd hash.\n", i);
2339                 }
2340                 ipv6_l3fwd_out_if[ret] = entry.if_out;
2341         }
2342         printf("Hash: Adding 0x%" PRIx32 "keys\n", array_len);
2343 }
2344
2345 #define NUMBER_PORT_USED 4
2346 static inline void
2347 populate_ipv4_many_flow_into_table(const struct rte_hash* h,
2348                 unsigned int nr_flow)
2349 {
2350         unsigned i;
2351         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
2352         for (i = 0; i < nr_flow; i++) {
2353                 struct ipv4_l3fwd_route entry;
2354                 union ipv4_5tuple_host newkey;
2355                 uint8_t a = (uint8_t) ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
2356                 uint8_t b = (uint8_t) (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
2357                 uint8_t c = (uint8_t) ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
2358                 /* Create the ipv4 exact match flow */
2359                 memset(&entry, 0, sizeof(entry));
2360                 switch (i & (NUMBER_PORT_USED -1)) {
2361                 case 0:
2362                         entry = ipv4_l3fwd_route_array[0];
2363                         entry.key.ip_dst = IPv4(101,c,b,a);
2364                         break;
2365                 case 1:
2366                         entry = ipv4_l3fwd_route_array[1];
2367                         entry.key.ip_dst = IPv4(201,c,b,a);
2368                         break;
2369                 case 2:
2370                         entry = ipv4_l3fwd_route_array[2];
2371                         entry.key.ip_dst = IPv4(111,c,b,a);
2372                         break;
2373                 case 3:
2374                         entry = ipv4_l3fwd_route_array[3];
2375                         entry.key.ip_dst = IPv4(211,c,b,a);
2376                         break;
2377                 };
2378                 convert_ipv4_5tuple(&entry.key, &newkey);
2379                 int32_t ret = rte_hash_add_key(h,(void *) &newkey);
2380                 if (ret < 0) {
2381                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
2382                 }
2383                 ipv4_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
2384
2385         }
2386         printf("Hash: Adding 0x%x keys\n", nr_flow);
2387 }
2388
2389 static inline void
2390 populate_ipv6_many_flow_into_table(const struct rte_hash* h,
2391                 unsigned int nr_flow)
2392 {
2393         unsigned i;
2394         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
2395         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
2396         for (i = 0; i < nr_flow; i++) {
2397                 struct ipv6_l3fwd_route entry;
2398                 union ipv6_5tuple_host newkey;
2399                 uint8_t a = (uint8_t) ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
2400                 uint8_t b = (uint8_t) (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
2401                 uint8_t c = (uint8_t) ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
2402                 /* Create the ipv6 exact match flow */
2403                 memset(&entry, 0, sizeof(entry));
2404                 switch (i & (NUMBER_PORT_USED - 1)) {
2405                 case 0: entry = ipv6_l3fwd_route_array[0]; break;
2406                 case 1: entry = ipv6_l3fwd_route_array[1]; break;
2407                 case 2: entry = ipv6_l3fwd_route_array[2]; break;
2408                 case 3: entry = ipv6_l3fwd_route_array[3]; break;
2409                 };
2410                 entry.key.ip_dst[13] = c;
2411                 entry.key.ip_dst[14] = b;
2412                 entry.key.ip_dst[15] = a;
2413                 convert_ipv6_5tuple(&entry.key, &newkey);
2414                 int32_t ret = rte_hash_add_key(h,(void *) &newkey);
2415                 if (ret < 0) {
2416                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
2417                 }
2418                 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
2419
2420         }
2421         printf("Hash: Adding 0x%x keys\n", nr_flow);
2422 }
2423
2424 static void
2425 setup_hash(int socketid)
2426 {
2427     struct rte_hash_parameters ipv4_l3fwd_hash_params = {
2428         .name = NULL,
2429         .entries = L3FWD_HASH_ENTRIES,
2430         .key_len = sizeof(union ipv4_5tuple_host),
2431         .hash_func = ipv4_hash_crc,
2432         .hash_func_init_val = 0,
2433     };
2434
2435     struct rte_hash_parameters ipv6_l3fwd_hash_params = {
2436         .name = NULL,
2437         .entries = L3FWD_HASH_ENTRIES,
2438         .key_len = sizeof(union ipv6_5tuple_host),
2439         .hash_func = ipv6_hash_crc,
2440         .hash_func_init_val = 0,
2441     };
2442
2443     char s[64];
2444
2445         /* create ipv4 hash */
2446         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
2447         ipv4_l3fwd_hash_params.name = s;
2448         ipv4_l3fwd_hash_params.socket_id = socketid;
2449         ipv4_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv4_l3fwd_hash_params);
2450         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2451                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2452                                 "socket %d\n", socketid);
2453
2454         /* create ipv6 hash */
2455         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
2456         ipv6_l3fwd_hash_params.name = s;
2457         ipv6_l3fwd_hash_params.socket_id = socketid;
2458         ipv6_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv6_l3fwd_hash_params);
2459         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
2460                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2461                                 "socket %d\n", socketid);
2462
2463         if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
2464                 /* For testing hash matching with a large number of flows we
2465                  * generate millions of IP 5-tuples with an incremented dst
2466                  * address to initialize the hash table. */
2467                 if (ipv6 == 0) {
2468                         /* populate the ipv4 hash */
2469                         populate_ipv4_many_flow_into_table(
2470                                 ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
2471                 } else {
2472                         /* populate the ipv6 hash */
2473                         populate_ipv6_many_flow_into_table(
2474                                 ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
2475                 }
2476         } else {
2477                 /* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize the hash table */
2478                 if (ipv6 == 0) {
2479                         /* populate the ipv4 hash */
2480                         populate_ipv4_few_flow_into_table(ipv4_l3fwd_lookup_struct[socketid]);
2481                 } else {
2482                         /* populate the ipv6 hash */
2483                         populate_ipv6_few_flow_into_table(ipv6_l3fwd_lookup_struct[socketid]);
2484                 }
2485         }
2486 }
2487 #endif
2488
2489 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2490 static void
2491 setup_lpm(int socketid)
2492 {
2493         struct rte_lpm6_config config;
2494         unsigned i;
2495         int ret;
2496         char s[64];
2497
2498         /* create the LPM table */
2499         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
2500         ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
2501                                 IPV4_L3FWD_LPM_MAX_RULES, 0);
2502         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2503                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
2504                                 " on socket %d\n", socketid);
2505
2506         /* populate the LPM table */
2507         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
2508
2509                 /* skip unused ports */
2510                 if ((1 << ipv4_l3fwd_route_array[i].if_out &
2511                                 enabled_port_mask) == 0)
2512                         continue;
2513
2514                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
2515                         ipv4_l3fwd_route_array[i].ip,
2516                         ipv4_l3fwd_route_array[i].depth,
2517                         ipv4_l3fwd_route_array[i].if_out);
2518
2519                 if (ret < 0) {
2520                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
2521                                 "l3fwd LPM table on socket %d\n",
2522                                 i, socketid);
2523                 }
2524
2525                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
2526                         (unsigned)ipv4_l3fwd_route_array[i].ip,
2527                         ipv4_l3fwd_route_array[i].depth,
2528                         ipv4_l3fwd_route_array[i].if_out);
2529         }
2530
2531         /* create the LPM6 table */
2532         snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
2533
2534         config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
2535         config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
2536         config.flags = 0;
2537         ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
2538                                 &config);
2539         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
2540                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
2541                                 " on socket %d\n", socketid);
2542
2543         /* populate the LPM table */
2544         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
2545
2546                 /* skip unused ports */
2547                 if ((1 << ipv6_l3fwd_route_array[i].if_out &
2548                                 enabled_port_mask) == 0)
2549                         continue;
2550
2551                 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
2552                         ipv6_l3fwd_route_array[i].ip,
2553                         ipv6_l3fwd_route_array[i].depth,
2554                         ipv6_l3fwd_route_array[i].if_out);
2555
2556                 if (ret < 0) {
2557                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
2558                                 "l3fwd LPM table on socket %d\n",
2559                                 i, socketid);
2560                 }
2561
2562                 printf("LPM: Adding route %s / %d (%d)\n",
2563                         "IPV6",
2564                         ipv6_l3fwd_route_array[i].depth,
2565                         ipv6_l3fwd_route_array[i].if_out);
2566         }
2567 }
2568 #endif
2569
2570 static int
2571 init_mem(unsigned nb_mbuf)
2572 {
2573         struct lcore_conf *qconf;
2574         int socketid;
2575         unsigned lcore_id;
2576         char s[64];
2577
2578         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2579                 if (rte_lcore_is_enabled(lcore_id) == 0)
2580                         continue;
2581
2582                 if (numa_on)
2583                         socketid = rte_lcore_to_socket_id(lcore_id);
2584                 else
2585                         socketid = 0;
2586
2587                 if (socketid >= NB_SOCKETS) {
2588                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
2589                                 socketid, lcore_id, NB_SOCKETS);
2590                 }
2591                 if (pktmbuf_pool[socketid] == NULL) {
2592                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
2593                         pktmbuf_pool[socketid] =
2594                                 rte_pktmbuf_pool_create(s, nb_mbuf,
2595                                         MEMPOOL_CACHE_SIZE, 0,
2596                                         RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
2597                         if (pktmbuf_pool[socketid] == NULL)
2598                                 rte_exit(EXIT_FAILURE,
2599                                                 "Cannot init mbuf pool on socket %d\n", socketid);
2600                         else
2601                                 printf("Allocated mbuf pool on socket %d\n", socketid);
2602
2603 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2604                         setup_lpm(socketid);
2605 #else
2606                         setup_hash(socketid);
2607 #endif
2608                 }
2609                 qconf = &lcore_conf[lcore_id];
2610                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
2611                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
2612         }
2613         return 0;
2614 }
2615
2616 /* Check the link status of all ports in up to 9s, and print them finally */
2617 static void
2618 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
2619 {
2620 #define CHECK_INTERVAL 100 /* 100ms */
2621 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
2622         uint8_t portid, count, all_ports_up, print_flag = 0;
2623         struct rte_eth_link link;
2624
2625         printf("\nChecking link status");
2626         fflush(stdout);
2627         for (count = 0; count <= MAX_CHECK_TIME; count++) {
2628                 all_ports_up = 1;
2629                 for (portid = 0; portid < port_num; portid++) {
2630                         if ((port_mask & (1 << portid)) == 0)
2631                                 continue;
2632                         memset(&link, 0, sizeof(link));
2633                         rte_eth_link_get_nowait(portid, &link);
2634                         /* print link status if flag set */
2635                         if (print_flag == 1) {
2636                                 if (link.link_status)
2637                                         printf("Port %d Link Up - speed %u "
2638                                                 "Mbps - %s\n", (uint8_t)portid,
2639                                                 (unsigned)link.link_speed,
2640                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
2641                                         ("full-duplex") : ("half-duplex\n"));
2642                                 else
2643                                         printf("Port %d Link Down\n",
2644                                                 (uint8_t)portid);
2645                                 continue;
2646                         }
2647                         /* clear all_ports_up flag if any link down */
2648                         if (link.link_status == 0) {
2649                                 all_ports_up = 0;
2650                                 break;
2651                         }
2652                 }
2653                 /* after finally printing all link status, get out */
2654                 if (print_flag == 1)
2655                         break;
2656
2657                 if (all_ports_up == 0) {
2658                         printf(".");
2659                         fflush(stdout);
2660                         rte_delay_ms(CHECK_INTERVAL);
2661                 }
2662
2663                 /* set the print_flag if all ports up or timeout */
2664                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
2665                         print_flag = 1;
2666                         printf("done\n");
2667                 }
2668         }
2669 }
2670
2671 int
2672 main(int argc, char **argv)
2673 {
2674         struct lcore_conf *qconf;
2675         struct rte_eth_dev_info dev_info;
2676         struct rte_eth_txconf *txconf;
2677         int ret;
2678         unsigned nb_ports;
2679         uint16_t queueid;
2680         unsigned lcore_id;
2681         uint32_t n_tx_queue, nb_lcores;
2682         uint8_t portid, nb_rx_queue, queue, socketid;
2683
2684         /* init EAL */
2685         ret = rte_eal_init(argc, argv);
2686         if (ret < 0)
2687                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2688         argc -= ret;
2689         argv += ret;
2690
2691         /* pre-init dst MACs for all ports to 02:00:00:00:00:xx */
2692         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2693                 dest_eth_addr[portid] = ETHER_LOCAL_ADMIN_ADDR + ((uint64_t)portid << 40);
2694                 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
2695         }
2696
2697         /* parse application arguments (after the EAL ones) */
2698         ret = parse_args(argc, argv);
2699         if (ret < 0)
2700                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
2701
2702         if (check_lcore_params() < 0)
2703                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
2704
2705         ret = init_lcore_rx_queues();
2706         if (ret < 0)
2707                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2708
2709         nb_ports = rte_eth_dev_count();
2710         if (nb_ports > RTE_MAX_ETHPORTS)
2711                 nb_ports = RTE_MAX_ETHPORTS;
2712
2713         if (check_port_config(nb_ports) < 0)
2714                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
2715
2716         nb_lcores = rte_lcore_count();
2717
2718         /* initialize all ports */
2719         for (portid = 0; portid < nb_ports; portid++) {
2720                 /* skip ports that are not enabled */
2721                 if ((enabled_port_mask & (1 << portid)) == 0) {
2722                         printf("\nSkipping disabled port %d\n", portid);
2723                         continue;
2724                 }
2725
2726                 /* init port */
2727                 printf("Initializing port %d ... ", portid );
2728                 fflush(stdout);
2729
2730                 nb_rx_queue = get_port_n_rx_queues(portid);
2731                 n_tx_queue = nb_lcores;
2732                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
2733                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
2734                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
2735                         nb_rx_queue, (unsigned)n_tx_queue );
2736                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
2737                                         (uint16_t)n_tx_queue, &port_conf);
2738                 if (ret < 0)
2739                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
2740                                 ret, portid);
2741
2742                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
2743                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
2744                 printf(", ");
2745                 print_ethaddr("Destination:",
2746                         (const struct ether_addr *)&dest_eth_addr[portid]);
2747                 printf(", ");
2748
2749                 /*
2750                  * prepare src MACs for each port.
2751                  */
2752                 ether_addr_copy(&ports_eth_addr[portid],
2753                         (struct ether_addr *)(val_eth + portid) + 1);
2754
2755                 /* init memory */
2756                 ret = init_mem(NB_MBUF);
2757                 if (ret < 0)
2758                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
2759
2760                 /* init one TX queue per couple (lcore,port) */
2761                 queueid = 0;
2762                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2763                         if (rte_lcore_is_enabled(lcore_id) == 0)
2764                                 continue;
2765
2766                         if (numa_on)
2767                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2768                         else
2769                                 socketid = 0;
2770
2771                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
2772                         fflush(stdout);
2773
2774                         rte_eth_dev_info_get(portid, &dev_info);
2775                         txconf = &dev_info.default_txconf;
2776                         if (port_conf.rxmode.jumbo_frame)
2777                                 txconf->txq_flags = 0;
2778                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
2779                                                      socketid, txconf);
2780                         if (ret < 0)
2781                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
2782                                         "port=%d\n", ret, portid);
2783
2784                         qconf = &lcore_conf[lcore_id];
2785                         qconf->tx_queue_id[portid] = queueid;
2786                         queueid++;
2787                 }
2788                 printf("\n");
2789         }
2790
2791         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2792                 if (rte_lcore_is_enabled(lcore_id) == 0)
2793                         continue;
2794                 qconf = &lcore_conf[lcore_id];
2795                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
2796                 fflush(stdout);
2797                 /* init RX queues */
2798                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
2799                         portid = qconf->rx_queue_list[queue].port_id;
2800                         queueid = qconf->rx_queue_list[queue].queue_id;
2801
2802                         if (numa_on)
2803                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2804                         else
2805                                 socketid = 0;
2806
2807                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
2808                         fflush(stdout);
2809
2810                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
2811                                         socketid,
2812                                         NULL,
2813                                         pktmbuf_pool[socketid]);
2814                         if (ret < 0)
2815                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
2816                                                 "port=%d\n", ret, portid);
2817                 }
2818         }
2819
2820         printf("\n");
2821
2822         /* start ports */
2823         for (portid = 0; portid < nb_ports; portid++) {
2824                 if ((enabled_port_mask & (1 << portid)) == 0) {
2825                         continue;
2826                 }
2827                 /* Start device */
2828                 ret = rte_eth_dev_start(portid);
2829                 if (ret < 0)
2830                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
2831                                 ret, portid);
2832
2833                 /*
2834                  * If enabled, put device in promiscuous mode.
2835                  * This allows IO forwarding mode to forward packets
2836                  * to itself through 2 cross-connected  ports of the
2837                  * target machine.
2838                  */
2839                 if (promiscuous_on)
2840                         rte_eth_promiscuous_enable(portid);
2841         }
2842
2843         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
2844
2845         /* launch per-lcore init on every lcore */
2846         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2847         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2848                 if (rte_eal_wait_lcore(lcore_id) < 0)
2849                         return -1;
2850         }
2851
2852         return 0;
2853 }