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 #include <signal.h>
  45 #include <stdbool.h>
  46
  47 #include <rte_common.h>
  48 #include <rte_vect.h>
  49 #include <rte_byteorder.h>
  50 #include <rte_log.h>
  51 #include <rte_memory.h>
  52 #include <rte_memcpy.h>
  53 #include <rte_memzone.h>
  54 #include <rte_eal.h>
  55 #include <rte_per_lcore.h>
  56 #include <rte_launch.h>
  57 #include <rte_atomic.h>
  58 #include <rte_cycles.h>
  59 #include <rte_prefetch.h>
  60 #include <rte_lcore.h>
  61 #include <rte_per_lcore.h>
  62 #include <rte_branch_prediction.h>
  63 #include <rte_interrupts.h>
  64 #include <rte_pci.h>
  65 #include <rte_random.h>
  66 #include <rte_debug.h>
  67 #include <rte_ether.h>
  68 #include <rte_ethdev.h>
  69 #include <rte_ring.h>
  70 #include <rte_mempool.h>
  71 #include <rte_mbuf.h>
  72 #include <rte_ip.h>
  73 #include <rte_tcp.h>
  74 #include <rte_udp.h>
  75 #include <rte_string_fns.h>
  76
  77 #include <cmdline_parse.h>
  78 #include <cmdline_parse_etheraddr.h>
  79
  80 static volatile bool force_quit;
  81
  82 #define APP_LOOKUP_EXACT_MATCH          0
  83 #define APP_LOOKUP_LPM                  1
  84 #define DO_RFC_1812_CHECKS
  85
  86 #ifndef APP_LOOKUP_METHOD
  87 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  88 #endif
  89
  90 /*
  91  *  When set to zero, simple forwaring path is eanbled.
  92  *  When set to one, optimized forwarding path is enabled.
  93  *  Note that LPM optimisation path uses SSE4.1 instructions.
  94  */
  95 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && !defined(__SSE4_1__))
  96 #define ENABLE_MULTI_BUFFER_OPTIMIZE    0
  97 #else
  98 #define ENABLE_MULTI_BUFFER_OPTIMIZE    1
  99 #endif
 100
 101 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 102 #include <rte_hash.h>
 103 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 104 #include <rte_lpm.h>
 105 #include <rte_lpm6.h>
 106 #else
 107 #error "APP_LOOKUP_METHOD set to incorrect value"
 108 #endif
 109
 110 #ifndef IPv6_BYTES
 111 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 112                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 113 #define IPv6_BYTES(addr) \
 114         addr[0],  addr[1], addr[2],  addr[3], \
 115         addr[4],  addr[5], addr[6],  addr[7], \
 116         addr[8],  addr[9], addr[10], addr[11],\
 117         addr[12], addr[13],addr[14], addr[15]
 118 #endif
 119
 120
 121 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
 122
 123 #define MAX_JUMBO_PKT_LEN  9600
 124
 125 #define IPV6_ADDR_LEN 16
 126
 127 #define MEMPOOL_CACHE_SIZE 256
 128
 129 /*
 130  * This expression is used to calculate the number of mbufs needed depending on user input, taking
 131  *  into account memory for rx and tx hardware rings, cache per lcore and mtable per port per lcore.
 132  *  RTE_MAX is used to ensure that NB_MBUF never goes below a minimum value of 8192
 133  */
 134
 135 #define NB_MBUF RTE_MAX (                                                                                                                                       \
 136                                 (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT +                                                        \
 137                                 nb_ports*nb_lcores*MAX_PKT_BURST +                                                                                      \
 138                                 nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT +                                                          \
 139                                 nb_lcores*MEMPOOL_CACHE_SIZE),                                                                                          \
 140                                 (unsigned)8192)
 141
 142 #define MAX_PKT_BURST     32
 143 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 144
 145 /*
 146  * Try to avoid TX buffering if we have at least MAX_TX_BURST packets to send.
 147  */
 148 #define MAX_TX_BURST    (MAX_PKT_BURST / 2)
 149
 150 #define NB_SOCKETS 8
 151
 152 /* Configure how many packets ahead to prefetch, when reading packets */
 153 #define PREFETCH_OFFSET 3
 154
 155 /* Used to mark destination port as 'invalid'. */
 156 #define BAD_PORT        ((uint16_t)-1)
 157
 158 #define FWDSTEP 4
 159
 160 /*
 161  * Configurable number of RX/TX ring descriptors
 162  */
 163 #define RTE_TEST_RX_DESC_DEFAULT 128
 164 #define RTE_TEST_TX_DESC_DEFAULT 512
 165 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 166 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 167
 168 /* ethernet addresses of ports */
 169 static uint64_t dest_eth_addr[RTE_MAX_ETHPORTS];
 170 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 171
 172 static __m128i val_eth[RTE_MAX_ETHPORTS];
 173
 174 /* replace first 12B of the ethernet header. */
 175 #define MASK_ETH        0x3f
 176
 177 /* mask of enabled ports */
 178 static uint32_t enabled_port_mask = 0;
 179 static int promiscuous_on = 0; /**< Ports set in promiscuous mode off by default. */
 180 static int numa_on = 1; /**< NUMA is enabled by default. */
 181
 182 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 183 static int ipv6 = 0; /**< ipv6 is false by default. */
 184 #endif
 185
 186 struct mbuf_table {
 187         uint16_t len;
 188         struct rte_mbuf *m_table[MAX_PKT_BURST];
 189 };
 190
 191 struct lcore_rx_queue {
 192         uint8_t port_id;
 193         uint8_t queue_id;
 194 } __rte_cache_aligned;
 195
 196 #define MAX_RX_QUEUE_PER_LCORE 16
 197 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 198 #define MAX_RX_QUEUE_PER_PORT 128
 199
 200 #define MAX_LCORE_PARAMS 1024
 201 struct lcore_params {
 202         uint8_t port_id;
 203         uint8_t queue_id;
 204         uint8_t lcore_id;
 205 } __rte_cache_aligned;
 206
 207 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 208 static struct lcore_params lcore_params_array_default[] = {
 209         {0, 0, 2},
 210         {0, 1, 2},
 211         {0, 2, 2},
 212         {1, 0, 2},
 213         {1, 1, 2},
 214         {1, 2, 2},
 215         {2, 0, 2},
 216         {3, 0, 3},
 217         {3, 1, 3},
 218 };
 219
 220 static struct lcore_params * lcore_params = lcore_params_array_default;
 221 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 222                                 sizeof(lcore_params_array_default[0]);
 223
 224 static struct rte_eth_conf port_conf = {
 225         .rxmode = {
 226                 .mq_mode = ETH_MQ_RX_RSS,
 227                 .max_rx_pkt_len = ETHER_MAX_LEN,
 228                 .split_hdr_size = 0,
 229                 .header_split   = 0, /**< Header Split disabled */
 230                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 231                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 232                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 233                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 234         },
 235         .rx_adv_conf = {
 236                 .rss_conf = {
 237                         .rss_key = NULL,
 238                         .rss_hf = ETH_RSS_IP,
 239                 },
 240         },
 241         .txmode = {
 242                 .mq_mode = ETH_MQ_TX_NONE,
 243         },
 244 };
 245
 246 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 247
 248 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 249
 250 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 251 #include <rte_hash_crc.h>
 252 #define DEFAULT_HASH_FUNC       rte_hash_crc
 253 #else
 254 #include <rte_jhash.h>
 255 #define DEFAULT_HASH_FUNC       rte_jhash
 256 #endif
 257
 258 struct ipv4_5tuple {
 259         uint32_t ip_dst;
 260         uint32_t ip_src;
 261         uint16_t port_dst;
 262         uint16_t port_src;
 263         uint8_t  proto;
 264 } __attribute__((__packed__));
 265
 266 union ipv4_5tuple_host {
 267         struct {
 268                 uint8_t  pad0;
 269                 uint8_t  proto;
 270                 uint16_t pad1;
 271                 uint32_t ip_src;
 272                 uint32_t ip_dst;
 273                 uint16_t port_src;
 274                 uint16_t port_dst;
 275         };
 276         __m128i xmm;
 277 };
 278
 279 #define XMM_NUM_IN_IPV6_5TUPLE 3
 280
 281 struct ipv6_5tuple {
 282         uint8_t  ip_dst[IPV6_ADDR_LEN];
 283         uint8_t  ip_src[IPV6_ADDR_LEN];
 284         uint16_t port_dst;
 285         uint16_t port_src;
 286         uint8_t  proto;
 287 } __attribute__((__packed__));
 288
 289 union ipv6_5tuple_host {
 290         struct {
 291                 uint16_t pad0;
 292                 uint8_t  proto;
 293                 uint8_t  pad1;
 294                 uint8_t  ip_src[IPV6_ADDR_LEN];
 295                 uint8_t  ip_dst[IPV6_ADDR_LEN];
 296                 uint16_t port_src;
 297                 uint16_t port_dst;
 298                 uint64_t reserve;
 299         };
 300         __m128i xmm[XMM_NUM_IN_IPV6_5TUPLE];
 301 };
 302
 303 struct ipv4_l3fwd_route {
 304         struct ipv4_5tuple key;
 305         uint8_t if_out;
 306 };
 307
 308 struct ipv6_l3fwd_route {
 309         struct ipv6_5tuple key;
 310         uint8_t if_out;
 311 };
 312
 313 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 314         {{IPv4(101,0,0,0), IPv4(100,10,0,1),  101, 11, IPPROTO_TCP}, 0},
 315         {{IPv4(201,0,0,0), IPv4(200,20,0,1),  102, 12, IPPROTO_TCP}, 1},
 316         {{IPv4(111,0,0,0), IPv4(100,30,0,1),  101, 11, IPPROTO_TCP}, 2},
 317         {{IPv4(211,0,0,0), IPv4(200,40,0,1),  102, 12, IPPROTO_TCP}, 3},
 318 };
 319
 320 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 321         {{
 322         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 323         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 324         101, 11, IPPROTO_TCP}, 0},
 325
 326         {{
 327         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 328         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 329         102, 12, IPPROTO_TCP}, 1},
 330
 331         {{
 332         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 333         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 334         101, 11, IPPROTO_TCP}, 2},
 335
 336         {{
 337         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 338         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 339         102, 12, IPPROTO_TCP}, 3},
 340 };
 341
 342 typedef struct rte_hash lookup_struct_t;
 343 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 344 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 345
 346 #ifdef RTE_ARCH_X86_64
 347 /* default to 4 million hash entries (approx) */
 348 #define L3FWD_HASH_ENTRIES              1024*1024*4
 349 #else
 350 /* 32-bit has less address-space for hugepage memory, limit to 1M entries */
 351 #define L3FWD_HASH_ENTRIES              1024*1024*1
 352 #endif
 353 #define HASH_ENTRY_NUMBER_DEFAULT       4
 354
 355 static uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
 356
 357 static inline uint32_t
 358 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
 359         uint32_t init_val)
 360 {
 361         const union ipv4_5tuple_host *k;
 362         uint32_t t;
 363         const uint32_t *p;
 364
 365         k = data;
 366         t = k->proto;
 367         p = (const uint32_t *)&k->port_src;
 368
 369 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 370         init_val = rte_hash_crc_4byte(t, init_val);
 371         init_val = rte_hash_crc_4byte(k->ip_src, init_val);
 372         init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
 373         init_val = rte_hash_crc_4byte(*p, init_val);
 374 #else /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 375         init_val = rte_jhash_1word(t, init_val);
 376         init_val = rte_jhash_1word(k->ip_src, init_val);
 377         init_val = rte_jhash_1word(k->ip_dst, init_val);
 378         init_val = rte_jhash_1word(*p, init_val);
 379 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 380         return init_val;
 381 }
 382
 383 static inline uint32_t
 384 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len, uint32_t init_val)
 385 {
 386         const union ipv6_5tuple_host *k;
 387         uint32_t t;
 388         const uint32_t *p;
 389 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 390         const uint32_t  *ip_src0, *ip_src1, *ip_src2, *ip_src3;
 391         const uint32_t  *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
 392 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 393
 394         k = data;
 395         t = k->proto;
 396         p = (const uint32_t *)&k->port_src;
 397
 398 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 399         ip_src0 = (const uint32_t *) k->ip_src;
 400         ip_src1 = (const uint32_t *)(k->ip_src+4);
 401         ip_src2 = (const uint32_t *)(k->ip_src+8);
 402         ip_src3 = (const uint32_t *)(k->ip_src+12);
 403         ip_dst0 = (const uint32_t *) k->ip_dst;
 404         ip_dst1 = (const uint32_t *)(k->ip_dst+4);
 405         ip_dst2 = (const uint32_t *)(k->ip_dst+8);
 406         ip_dst3 = (const uint32_t *)(k->ip_dst+12);
 407         init_val = rte_hash_crc_4byte(t, init_val);
 408         init_val = rte_hash_crc_4byte(*ip_src0, init_val);
 409         init_val = rte_hash_crc_4byte(*ip_src1, init_val);
 410         init_val = rte_hash_crc_4byte(*ip_src2, init_val);
 411         init_val = rte_hash_crc_4byte(*ip_src3, init_val);
 412         init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
 413         init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
 414         init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
 415         init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
 416         init_val = rte_hash_crc_4byte(*p, init_val);
 417 #else /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 418         init_val = rte_jhash_1word(t, init_val);
 419         init_val = rte_jhash(k->ip_src, sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
 420         init_val = rte_jhash(k->ip_dst, sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
 421         init_val = rte_jhash_1word(*p, init_val);
 422 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 423         return init_val;
 424 }
 425
 426 #define IPV4_L3FWD_NUM_ROUTES \
 427         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 428
 429 #define IPV6_L3FWD_NUM_ROUTES \
 430         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 431
 432 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 433 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 434
 435 #endif
 436
 437 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 438 struct ipv4_l3fwd_route {
 439         uint32_t ip;
 440         uint8_t  depth;
 441         uint8_t  if_out;
 442 };
 443
 444 struct ipv6_l3fwd_route {
 445         uint8_t ip[16];
 446         uint8_t  depth;
 447         uint8_t  if_out;
 448 };
 449
 450 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 451         {IPv4(1,1,1,0), 24, 0},
 452         {IPv4(2,1,1,0), 24, 1},
 453         {IPv4(3,1,1,0), 24, 2},
 454         {IPv4(4,1,1,0), 24, 3},
 455         {IPv4(5,1,1,0), 24, 4},
 456         {IPv4(6,1,1,0), 24, 5},
 457         {IPv4(7,1,1,0), 24, 6},
 458         {IPv4(8,1,1,0), 24, 7},
 459 };
 460
 461 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 462         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
 463         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
 464         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
 465         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
 466         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
 467         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
 468         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
 469         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
 470 };
 471
 472 #define IPV4_L3FWD_NUM_ROUTES \
 473         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 474 #define IPV6_L3FWD_NUM_ROUTES \
 475         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 476
 477 #define IPV4_L3FWD_LPM_MAX_RULES         1024
 478 #define IPV6_L3FWD_LPM_MAX_RULES         1024
 479 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
 480
 481 typedef struct rte_lpm lookup_struct_t;
 482 typedef struct rte_lpm6 lookup6_struct_t;
 483 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 484 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 485 #endif
 486
 487 struct lcore_conf {
 488         uint16_t n_rx_queue;
 489         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 490         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 491         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 492         lookup_struct_t * ipv4_lookup_struct;
 493 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 494         lookup6_struct_t * ipv6_lookup_struct;
 495 #else
 496         lookup_struct_t * ipv6_lookup_struct;
 497 #endif
 498 } __rte_cache_aligned;
 499
 500 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 501
 502 /* Send burst of packets on an output interface */
 503 static inline int
 504 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 505 {
 506         struct rte_mbuf **m_table;
 507         int ret;
 508         uint16_t queueid;
 509
 510         queueid = qconf->tx_queue_id[port];
 511         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 512
 513         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 514         if (unlikely(ret < n)) {
 515                 do {
 516                         rte_pktmbuf_free(m_table[ret]);
 517                 } while (++ret < n);
 518         }
 519
 520         return 0;
 521 }
 522
 523 /* Enqueue a single packet, and send burst if queue is filled */
 524 static inline int
 525 send_single_packet(struct rte_mbuf *m, uint8_t port)
 526 {
 527         uint32_t lcore_id;
 528         uint16_t len;
 529         struct lcore_conf *qconf;
 530
 531         lcore_id = rte_lcore_id();
 532
 533         qconf = &lcore_conf[lcore_id];
 534         len = qconf->tx_mbufs[port].len;
 535         qconf->tx_mbufs[port].m_table[len] = m;
 536         len++;
 537
 538         /* enough pkts to be sent */
 539         if (unlikely(len == MAX_PKT_BURST)) {
 540                 send_burst(qconf, MAX_PKT_BURST, port);
 541                 len = 0;
 542         }
 543
 544         qconf->tx_mbufs[port].len = len;
 545         return 0;
 546 }
 547
 548 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
 549         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
 550 static inline __attribute__((always_inline)) void
 551 send_packetsx4(struct lcore_conf *qconf, uint8_t port,
 552         struct rte_mbuf *m[], uint32_t num)
 553 {
 554         uint32_t len, j, n;
 555
 556         len = qconf->tx_mbufs[port].len;
 557
 558         /*
 559          * If TX buffer for that queue is empty, and we have enough packets,
 560          * then send them straightway.
 561          */
 562         if (num >= MAX_TX_BURST && len == 0) {
 563                 n = rte_eth_tx_burst(port, qconf->tx_queue_id[port], m, num);
 564                 if (unlikely(n < num)) {
 565                         do {
 566                                 rte_pktmbuf_free(m[n]);
 567                         } while (++n < num);
 568                 }
 569                 return;
 570         }
 571
 572         /*
 573          * Put packets into TX buffer for that queue.
 574          */
 575
 576         n = len + num;
 577         n = (n > MAX_PKT_BURST) ? MAX_PKT_BURST - len : num;
 578
 579         j = 0;
 580         switch (n % FWDSTEP) {
 581         while (j < n) {
 582         case 0:
 583                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 584                 j++;
 585         case 3:
 586                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 587                 j++;
 588         case 2:
 589                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 590                 j++;
 591         case 1:
 592                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 593                 j++;
 594         }
 595         }
 596
 597         len += n;
 598
 599         /* enough pkts to be sent */
 600         if (unlikely(len == MAX_PKT_BURST)) {
 601
 602                 send_burst(qconf, MAX_PKT_BURST, port);
 603
 604                 /* copy rest of the packets into the TX buffer. */
 605                 len = num - n;
 606                 j = 0;
 607                 switch (len % FWDSTEP) {
 608                 while (j < len) {
 609                 case 0:
 610                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 611                         j++;
 612                 case 3:
 613                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 614                         j++;
 615                 case 2:
 616                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 617                         j++;
 618                 case 1:
 619                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 620                         j++;
 621                 }
 622                 }
 623         }
 624
 625         qconf->tx_mbufs[port].len = len;
 626 }
 627 #endif /* APP_LOOKUP_LPM */
 628
 629 #ifdef DO_RFC_1812_CHECKS
 630 static inline int
 631 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 632 {
 633         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 634         /*
 635          * 1. The packet length reported by the Link Layer must be large
 636          * enough to hold the minimum length legal IP datagram (20 bytes).
 637          */
 638         if (link_len < sizeof(struct ipv4_hdr))
 639                 return -1;
 640
 641         /* 2. The IP checksum must be correct. */
 642         /* this is checked in H/W */
 643
 644         /*
 645          * 3. The IP version number must be 4. If the version number is not 4
 646          * then the packet may be another version of IP, such as IPng or
 647          * ST-II.
 648          */
 649         if (((pkt->version_ihl) >> 4) != 4)
 650                 return -3;
 651         /*
 652          * 4. The IP header length field must be large enough to hold the
 653          * minimum length legal IP datagram (20 bytes = 5 words).
 654          */
 655         if ((pkt->version_ihl & 0xf) < 5)
 656                 return -4;
 657
 658         /*
 659          * 5. The IP total length field must be large enough to hold the IP
 660          * datagram header, whose length is specified in the IP header length
 661          * field.
 662          */
 663         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 664                 return -5;
 665
 666         return 0;
 667 }
 668 #endif
 669
 670 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 671
 672 static __m128i mask0;
 673 static __m128i mask1;
 674 static __m128i mask2;
 675 static inline uint8_t
 676 get_ipv4_dst_port(void *ipv4_hdr, uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
 677 {
 678         int ret = 0;
 679         union ipv4_5tuple_host key;
 680
 681         ipv4_hdr = (uint8_t *)ipv4_hdr + offsetof(struct ipv4_hdr, time_to_live);
 682         __m128i data = _mm_loadu_si128((__m128i*)(ipv4_hdr));
 683         /* Get 5 tuple: dst port, src port, dst IP address, src IP address and protocol */
 684         key.xmm = _mm_and_si128(data, mask0);
 685         /* Find destination port */
 686         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 687         return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
 688 }
 689
 690 static inline uint8_t
 691 get_ipv6_dst_port(void *ipv6_hdr,  uint8_t portid, lookup_struct_t * ipv6_l3fwd_lookup_struct)
 692 {
 693         int ret = 0;
 694         union ipv6_5tuple_host key;
 695
 696         ipv6_hdr = (uint8_t *)ipv6_hdr + offsetof(struct ipv6_hdr, payload_len);
 697         __m128i data0 = _mm_loadu_si128((__m128i*)(ipv6_hdr));
 698         __m128i data1 = _mm_loadu_si128((__m128i*)(((uint8_t*)ipv6_hdr)+sizeof(__m128i)));
 699         __m128i data2 = _mm_loadu_si128((__m128i*)(((uint8_t*)ipv6_hdr)+sizeof(__m128i)+sizeof(__m128i)));
 700         /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
 701         key.xmm[0] = _mm_and_si128(data0, mask1);
 702         /* Get part of 5 tuple: dst IP address lower 96 bits and src IP address higher 32 bits */
 703         key.xmm[1] = data1;
 704         /* Get part of 5 tuple: dst port and src port and dst IP address higher 32 bits */
 705         key.xmm[2] = _mm_and_si128(data2, mask2);
 706
 707         /* Find destination port */
 708         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 709         return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
 710 }
 711 #endif
 712
 713 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 714
 715 static inline uint8_t
 716 get_ipv4_dst_port(void *ipv4_hdr,  uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
 717 {
 718         uint8_t next_hop;
 719
 720         return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 721                 rte_be_to_cpu_32(((struct ipv4_hdr *)ipv4_hdr)->dst_addr),
 722                 &next_hop) == 0) ? next_hop : portid);
 723 }
 724
 725 static inline uint8_t
 726 get_ipv6_dst_port(void *ipv6_hdr,  uint8_t portid, lookup6_struct_t * ipv6_l3fwd_lookup_struct)
 727 {
 728         uint8_t next_hop;
 729         return (uint8_t) ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
 730                         ((struct ipv6_hdr*)ipv6_hdr)->dst_addr, &next_hop) == 0)?
 731                         next_hop : portid);
 732 }
 733 #endif
 734
 735 static inline void l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
 736         struct lcore_conf *qconf)  __attribute__((unused));
 737
 738 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) && \
 739         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
 740
 741 #define MASK_ALL_PKTS    0xff
 742 #define EXCLUDE_1ST_PKT 0xfe
 743 #define EXCLUDE_2ND_PKT 0xfd
 744 #define EXCLUDE_3RD_PKT 0xfb
 745 #define EXCLUDE_4TH_PKT 0xf7
 746 #define EXCLUDE_5TH_PKT 0xef
 747 #define EXCLUDE_6TH_PKT 0xdf
 748 #define EXCLUDE_7TH_PKT 0xbf
 749 #define EXCLUDE_8TH_PKT 0x7f
 750
 751 static inline void
 752 simple_ipv4_fwd_8pkts(struct rte_mbuf *m[8], uint8_t portid, struct lcore_conf *qconf)
 753 {
 754         struct ether_hdr *eth_hdr[8];
 755         struct ipv4_hdr *ipv4_hdr[8];
 756         uint8_t dst_port[8];
 757         int32_t ret[8];
 758         union ipv4_5tuple_host key[8];
 759         __m128i data[8];
 760
 761         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
 762         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
 763         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
 764         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
 765         eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *);
 766         eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *);
 767         eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *);
 768         eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *);
 769
 770         /* Handle IPv4 headers.*/
 771         ipv4_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv4_hdr *,
 772                                               sizeof(struct ether_hdr));
 773         ipv4_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv4_hdr *,
 774                                               sizeof(struct ether_hdr));
 775         ipv4_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv4_hdr *,
 776                                               sizeof(struct ether_hdr));
 777         ipv4_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv4_hdr *,
 778                                               sizeof(struct ether_hdr));
 779         ipv4_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv4_hdr *,
 780                                               sizeof(struct ether_hdr));
 781         ipv4_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv4_hdr *,
 782                                               sizeof(struct ether_hdr));
 783         ipv4_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv4_hdr *,
 784                                               sizeof(struct ether_hdr));
 785         ipv4_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv4_hdr *,
 786                                               sizeof(struct ether_hdr));
 787
 788 #ifdef DO_RFC_1812_CHECKS
 789         /* Check to make sure the packet is valid (RFC1812) */
 790         uint8_t valid_mask = MASK_ALL_PKTS;
 791         if (is_valid_ipv4_pkt(ipv4_hdr[0], m[0]->pkt_len) < 0) {
 792                 rte_pktmbuf_free(m[0]);
 793                 valid_mask &= EXCLUDE_1ST_PKT;
 794         }
 795         if (is_valid_ipv4_pkt(ipv4_hdr[1], m[1]->pkt_len) < 0) {
 796                 rte_pktmbuf_free(m[1]);
 797                 valid_mask &= EXCLUDE_2ND_PKT;
 798         }
 799         if (is_valid_ipv4_pkt(ipv4_hdr[2], m[2]->pkt_len) < 0) {
 800                 rte_pktmbuf_free(m[2]);
 801                 valid_mask &= EXCLUDE_3RD_PKT;
 802         }
 803         if (is_valid_ipv4_pkt(ipv4_hdr[3], m[3]->pkt_len) < 0) {
 804                 rte_pktmbuf_free(m[3]);
 805                 valid_mask &= EXCLUDE_4TH_PKT;
 806         }
 807         if (is_valid_ipv4_pkt(ipv4_hdr[4], m[4]->pkt_len) < 0) {
 808                 rte_pktmbuf_free(m[4]);
 809                 valid_mask &= EXCLUDE_5TH_PKT;
 810         }
 811         if (is_valid_ipv4_pkt(ipv4_hdr[5], m[5]->pkt_len) < 0) {
 812                 rte_pktmbuf_free(m[5]);
 813                 valid_mask &= EXCLUDE_6TH_PKT;
 814         }
 815         if (is_valid_ipv4_pkt(ipv4_hdr[6], m[6]->pkt_len) < 0) {
 816                 rte_pktmbuf_free(m[6]);
 817                 valid_mask &= EXCLUDE_7TH_PKT;
 818         }
 819         if (is_valid_ipv4_pkt(ipv4_hdr[7], m[7]->pkt_len) < 0) {
 820                 rte_pktmbuf_free(m[7]);
 821                 valid_mask &= EXCLUDE_8TH_PKT;
 822         }
 823         if (unlikely(valid_mask != MASK_ALL_PKTS)) {
 824                 if (valid_mask == 0){
 825                         return;
 826                 } else {
 827                         uint8_t i = 0;
 828                         for (i = 0; i < 8; i++) {
 829                                 if ((0x1 << i) & valid_mask) {
 830                                         l3fwd_simple_forward(m[i], portid, qconf);
 831                                 }
 832                         }
 833                         return;
 834                 }
 835         }
 836 #endif // End of #ifdef DO_RFC_1812_CHECKS
 837
 838         data[0] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[0], __m128i *,
 839                                         sizeof(struct ether_hdr) +
 840                                         offsetof(struct ipv4_hdr, time_to_live)));
 841         data[1] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[1], __m128i *,
 842                                         sizeof(struct ether_hdr) +
 843                                         offsetof(struct ipv4_hdr, time_to_live)));
 844         data[2] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[2], __m128i *,
 845                                         sizeof(struct ether_hdr) +
 846                                         offsetof(struct ipv4_hdr, time_to_live)));
 847         data[3] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[3], __m128i *,
 848                                         sizeof(struct ether_hdr) +
 849                                         offsetof(struct ipv4_hdr, time_to_live)));
 850         data[4] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[4], __m128i *,
 851                                         sizeof(struct ether_hdr) +
 852                                         offsetof(struct ipv4_hdr, time_to_live)));
 853         data[5] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[5], __m128i *,
 854                                         sizeof(struct ether_hdr) +
 855                                         offsetof(struct ipv4_hdr, time_to_live)));
 856         data[6] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[6], __m128i *,
 857                                         sizeof(struct ether_hdr) +
 858                                         offsetof(struct ipv4_hdr, time_to_live)));
 859         data[7] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[7], __m128i *,
 860                                         sizeof(struct ether_hdr) +
 861                                         offsetof(struct ipv4_hdr, time_to_live)));
 862
 863         key[0].xmm = _mm_and_si128(data[0], mask0);
 864         key[1].xmm = _mm_and_si128(data[1], mask0);
 865         key[2].xmm = _mm_and_si128(data[2], mask0);
 866         key[3].xmm = _mm_and_si128(data[3], mask0);
 867         key[4].xmm = _mm_and_si128(data[4], mask0);
 868         key[5].xmm = _mm_and_si128(data[5], mask0);
 869         key[6].xmm = _mm_and_si128(data[6], mask0);
 870         key[7].xmm = _mm_and_si128(data[7], mask0);
 871
 872         const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
 873                                 &key[4], &key[5], &key[6], &key[7]};
 874
 875         rte_hash_lookup_multi(qconf->ipv4_lookup_struct, &key_array[0], 8, ret);
 876         dst_port[0] = (uint8_t) ((ret[0] < 0) ? portid : ipv4_l3fwd_out_if[ret[0]]);
 877         dst_port[1] = (uint8_t) ((ret[1] < 0) ? portid : ipv4_l3fwd_out_if[ret[1]]);
 878         dst_port[2] = (uint8_t) ((ret[2] < 0) ? portid : ipv4_l3fwd_out_if[ret[2]]);
 879         dst_port[3] = (uint8_t) ((ret[3] < 0) ? portid : ipv4_l3fwd_out_if[ret[3]]);
 880         dst_port[4] = (uint8_t) ((ret[4] < 0) ? portid : ipv4_l3fwd_out_if[ret[4]]);
 881         dst_port[5] = (uint8_t) ((ret[5] < 0) ? portid : ipv4_l3fwd_out_if[ret[5]]);
 882         dst_port[6] = (uint8_t) ((ret[6] < 0) ? portid : ipv4_l3fwd_out_if[ret[6]]);
 883         dst_port[7] = (uint8_t) ((ret[7] < 0) ? portid : ipv4_l3fwd_out_if[ret[7]]);
 884
 885         if (dst_port[0] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[0]) == 0)
 886                 dst_port[0] = portid;
 887         if (dst_port[1] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[1]) == 0)
 888                 dst_port[1] = portid;
 889         if (dst_port[2] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[2]) == 0)
 890                 dst_port[2] = portid;
 891         if (dst_port[3] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[3]) == 0)
 892                 dst_port[3] = portid;
 893         if (dst_port[4] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[4]) == 0)
 894                 dst_port[4] = portid;
 895         if (dst_port[5] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[5]) == 0)
 896                 dst_port[5] = portid;
 897         if (dst_port[6] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[6]) == 0)
 898                 dst_port[6] = portid;
 899         if (dst_port[7] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[7]) == 0)
 900                 dst_port[7] = portid;
 901
 902 #ifdef DO_RFC_1812_CHECKS
 903         /* Update time to live and header checksum */
 904         --(ipv4_hdr[0]->time_to_live);
 905         --(ipv4_hdr[1]->time_to_live);
 906         --(ipv4_hdr[2]->time_to_live);
 907         --(ipv4_hdr[3]->time_to_live);
 908         ++(ipv4_hdr[0]->hdr_checksum);
 909         ++(ipv4_hdr[1]->hdr_checksum);
 910         ++(ipv4_hdr[2]->hdr_checksum);
 911         ++(ipv4_hdr[3]->hdr_checksum);
 912         --(ipv4_hdr[4]->time_to_live);
 913         --(ipv4_hdr[5]->time_to_live);
 914         --(ipv4_hdr[6]->time_to_live);
 915         --(ipv4_hdr[7]->time_to_live);
 916         ++(ipv4_hdr[4]->hdr_checksum);
 917         ++(ipv4_hdr[5]->hdr_checksum);
 918         ++(ipv4_hdr[6]->hdr_checksum);
 919         ++(ipv4_hdr[7]->hdr_checksum);
 920 #endif
 921
 922         /* dst addr */
 923         *(uint64_t *)&eth_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
 924         *(uint64_t *)&eth_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
 925         *(uint64_t *)&eth_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
 926         *(uint64_t *)&eth_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
 927         *(uint64_t *)&eth_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
 928         *(uint64_t *)&eth_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
 929         *(uint64_t *)&eth_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
 930         *(uint64_t *)&eth_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
 931
 932         /* src addr */
 933         ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
 934         ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
 935         ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
 936         ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
 937         ether_addr_copy(&ports_eth_addr[dst_port[4]], &eth_hdr[4]->s_addr);
 938         ether_addr_copy(&ports_eth_addr[dst_port[5]], &eth_hdr[5]->s_addr);
 939         ether_addr_copy(&ports_eth_addr[dst_port[6]], &eth_hdr[6]->s_addr);
 940         ether_addr_copy(&ports_eth_addr[dst_port[7]], &eth_hdr[7]->s_addr);
 941
 942         send_single_packet(m[0], (uint8_t)dst_port[0]);
 943         send_single_packet(m[1], (uint8_t)dst_port[1]);
 944         send_single_packet(m[2], (uint8_t)dst_port[2]);
 945         send_single_packet(m[3], (uint8_t)dst_port[3]);
 946         send_single_packet(m[4], (uint8_t)dst_port[4]);
 947         send_single_packet(m[5], (uint8_t)dst_port[5]);
 948         send_single_packet(m[6], (uint8_t)dst_port[6]);
 949         send_single_packet(m[7], (uint8_t)dst_port[7]);
 950
 951 }
 952
 953 static inline void get_ipv6_5tuple(struct rte_mbuf* m0, __m128i mask0, __m128i mask1,
 954                                  union ipv6_5tuple_host * key)
 955 {
 956         __m128i tmpdata0 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0, __m128i *, sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len)));
 957         __m128i tmpdata1 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0, __m128i *, sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i)));
 958         __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)));
 959         key->xmm[0] = _mm_and_si128(tmpdata0, mask0);
 960         key->xmm[1] = tmpdata1;
 961         key->xmm[2] = _mm_and_si128(tmpdata2, mask1);
 962         return;
 963 }
 964
 965 static inline void
 966 simple_ipv6_fwd_8pkts(struct rte_mbuf *m[8], uint8_t portid, struct lcore_conf *qconf)
 967 {
 968         struct ether_hdr *eth_hdr[8];
 969         __attribute__((unused)) struct ipv6_hdr *ipv6_hdr[8];
 970         uint8_t dst_port[8];
 971         int32_t ret[8];
 972         union ipv6_5tuple_host key[8];
 973
 974         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
 975         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
 976         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
 977         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
 978         eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *);
 979         eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *);
 980         eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *);
 981         eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *);
 982
 983         /* Handle IPv6 headers.*/
 984         ipv6_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv6_hdr *,
 985                                               sizeof(struct ether_hdr));
 986         ipv6_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv6_hdr *,
 987                                               sizeof(struct ether_hdr));
 988         ipv6_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv6_hdr *,
 989                                               sizeof(struct ether_hdr));
 990         ipv6_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv6_hdr *,
 991                                               sizeof(struct ether_hdr));
 992         ipv6_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv6_hdr *,
 993                                               sizeof(struct ether_hdr));
 994         ipv6_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv6_hdr *,
 995                                               sizeof(struct ether_hdr));
 996         ipv6_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv6_hdr *,
 997                                               sizeof(struct ether_hdr));
 998         ipv6_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv6_hdr *,
 999                                               sizeof(struct ether_hdr));
1000
1001         get_ipv6_5tuple(m[0], mask1, mask2, &key[0]);
1002         get_ipv6_5tuple(m[1], mask1, mask2, &key[1]);
1003         get_ipv6_5tuple(m[2], mask1, mask2, &key[2]);
1004         get_ipv6_5tuple(m[3], mask1, mask2, &key[3]);
1005         get_ipv6_5tuple(m[4], mask1, mask2, &key[4]);
1006         get_ipv6_5tuple(m[5], mask1, mask2, &key[5]);
1007         get_ipv6_5tuple(m[6], mask1, mask2, &key[6]);
1008         get_ipv6_5tuple(m[7], mask1, mask2, &key[7]);
1009
1010         const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1011                                 &key[4], &key[5], &key[6], &key[7]};
1012
1013         rte_hash_lookup_multi(qconf->ipv6_lookup_struct, &key_array[0], 8, ret);
1014         dst_port[0] = (uint8_t) ((ret[0] < 0) ? portid:ipv6_l3fwd_out_if[ret[0]]);
1015         dst_port[1] = (uint8_t) ((ret[1] < 0) ? portid:ipv6_l3fwd_out_if[ret[1]]);
1016         dst_port[2] = (uint8_t) ((ret[2] < 0) ? portid:ipv6_l3fwd_out_if[ret[2]]);
1017         dst_port[3] = (uint8_t) ((ret[3] < 0) ? portid:ipv6_l3fwd_out_if[ret[3]]);
1018         dst_port[4] = (uint8_t) ((ret[4] < 0) ? portid:ipv6_l3fwd_out_if[ret[4]]);
1019         dst_port[5] = (uint8_t) ((ret[5] < 0) ? portid:ipv6_l3fwd_out_if[ret[5]]);
1020         dst_port[6] = (uint8_t) ((ret[6] < 0) ? portid:ipv6_l3fwd_out_if[ret[6]]);
1021         dst_port[7] = (uint8_t) ((ret[7] < 0) ? portid:ipv6_l3fwd_out_if[ret[7]]);
1022
1023         if (dst_port[0] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[0]) == 0)
1024                 dst_port[0] = portid;
1025         if (dst_port[1] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[1]) == 0)
1026                 dst_port[1] = portid;
1027         if (dst_port[2] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[2]) == 0)
1028                 dst_port[2] = portid;
1029         if (dst_port[3] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[3]) == 0)
1030                 dst_port[3] = portid;
1031         if (dst_port[4] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[4]) == 0)
1032                 dst_port[4] = portid;
1033         if (dst_port[5] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[5]) == 0)
1034                 dst_port[5] = portid;
1035         if (dst_port[6] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[6]) == 0)
1036                 dst_port[6] = portid;
1037         if (dst_port[7] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[7]) == 0)
1038                 dst_port[7] = portid;
1039
1040         /* dst addr */
1041         *(uint64_t *)&eth_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1042         *(uint64_t *)&eth_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1043         *(uint64_t *)&eth_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1044         *(uint64_t *)&eth_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1045         *(uint64_t *)&eth_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1046         *(uint64_t *)&eth_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1047         *(uint64_t *)&eth_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1048         *(uint64_t *)&eth_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1049
1050         /* src addr */
1051         ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
1052         ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
1053         ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
1054         ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
1055         ether_addr_copy(&ports_eth_addr[dst_port[4]], &eth_hdr[4]->s_addr);
1056         ether_addr_copy(&ports_eth_addr[dst_port[5]], &eth_hdr[5]->s_addr);
1057         ether_addr_copy(&ports_eth_addr[dst_port[6]], &eth_hdr[6]->s_addr);
1058         ether_addr_copy(&ports_eth_addr[dst_port[7]], &eth_hdr[7]->s_addr);
1059
1060         send_single_packet(m[0], (uint8_t)dst_port[0]);
1061         send_single_packet(m[1], (uint8_t)dst_port[1]);
1062         send_single_packet(m[2], (uint8_t)dst_port[2]);
1063         send_single_packet(m[3], (uint8_t)dst_port[3]);
1064         send_single_packet(m[4], (uint8_t)dst_port[4]);
1065         send_single_packet(m[5], (uint8_t)dst_port[5]);
1066         send_single_packet(m[6], (uint8_t)dst_port[6]);
1067         send_single_packet(m[7], (uint8_t)dst_port[7]);
1068
1069 }
1070 #endif /* APP_LOOKUP_METHOD */
1071
1072 static inline __attribute__((always_inline)) void
1073 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, struct lcore_conf *qconf)
1074 {
1075         struct ether_hdr *eth_hdr;
1076         struct ipv4_hdr *ipv4_hdr;
1077         uint8_t dst_port;
1078
1079         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
1080
1081         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
1082                 /* Handle IPv4 headers.*/
1083                 ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
1084                                                    sizeof(struct ether_hdr));
1085
1086 #ifdef DO_RFC_1812_CHECKS
1087                 /* Check to make sure the packet is valid (RFC1812) */
1088                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
1089                         rte_pktmbuf_free(m);
1090                         return;
1091                 }
1092 #endif
1093
1094                  dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
1095                         qconf->ipv4_lookup_struct);
1096                 if (dst_port >= RTE_MAX_ETHPORTS ||
1097                                 (enabled_port_mask & 1 << dst_port) == 0)
1098                         dst_port = portid;
1099
1100 #ifdef DO_RFC_1812_CHECKS
1101                 /* Update time to live and header checksum */
1102                 --(ipv4_hdr->time_to_live);
1103                 ++(ipv4_hdr->hdr_checksum);
1104 #endif
1105                 /* dst addr */
1106                 *(uint64_t *)&eth_hdr->d_addr = dest_eth_addr[dst_port];
1107
1108                 /* src addr */
1109                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
1110
1111                 send_single_packet(m, dst_port);
1112         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
1113                 /* Handle IPv6 headers.*/
1114                 struct ipv6_hdr *ipv6_hdr;
1115
1116                 ipv6_hdr = rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
1117                                                    sizeof(struct ether_hdr));
1118
1119                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid, qconf->ipv6_lookup_struct);
1120
1121                 if (dst_port >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port) == 0)
1122                         dst_port = portid;
1123
1124                 /* dst addr */
1125                 *(uint64_t *)&eth_hdr->d_addr = dest_eth_addr[dst_port];
1126
1127                 /* src addr */
1128                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
1129
1130                 send_single_packet(m, dst_port);
1131         } else
1132                 /* Free the mbuf that contains non-IPV4/IPV6 packet */
1133                 rte_pktmbuf_free(m);
1134 }
1135
1136 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1137         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1138 #ifdef DO_RFC_1812_CHECKS
1139
1140 #define IPV4_MIN_VER_IHL        0x45
1141 #define IPV4_MAX_VER_IHL        0x4f
1142 #define IPV4_MAX_VER_IHL_DIFF   (IPV4_MAX_VER_IHL - IPV4_MIN_VER_IHL)
1143
1144 /* Minimum value of IPV4 total length (20B) in network byte order. */
1145 #define IPV4_MIN_LEN_BE (sizeof(struct ipv4_hdr) << 8)
1146
1147 /*
1148  * From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2:
1149  * - The IP version number must be 4.
1150  * - The IP header length field must be large enough to hold the
1151  *    minimum length legal IP datagram (20 bytes = 5 words).
1152  * - The IP total length field must be large enough to hold the IP
1153  *   datagram header, whose length is specified in the IP header length
1154  *   field.
1155  * If we encounter invalid IPV4 packet, then set destination port for it
1156  * to BAD_PORT value.
1157  */
1158 static inline __attribute__((always_inline)) void
1159 rfc1812_process(struct ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t ptype)
1160 {
1161         uint8_t ihl;
1162
1163         if (RTE_ETH_IS_IPV4_HDR(ptype)) {
1164                 ihl = ipv4_hdr->version_ihl - IPV4_MIN_VER_IHL;
1165
1166                 ipv4_hdr->time_to_live--;
1167                 ipv4_hdr->hdr_checksum++;
1168
1169                 if (ihl > IPV4_MAX_VER_IHL_DIFF ||
1170                                 ((uint8_t)ipv4_hdr->total_length == 0 &&
1171                                 ipv4_hdr->total_length < IPV4_MIN_LEN_BE)) {
1172                         dp[0] = BAD_PORT;
1173                 }
1174         }
1175 }
1176
1177 #else
1178 #define rfc1812_process(mb, dp) do { } while (0)
1179 #endif /* DO_RFC_1812_CHECKS */
1180 #endif /* APP_LOOKUP_LPM && ENABLE_MULTI_BUFFER_OPTIMIZE */
1181
1182
1183 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1184         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1185
1186 static inline __attribute__((always_inline)) uint16_t
1187 get_dst_port(const struct lcore_conf *qconf, struct rte_mbuf *pkt,
1188         uint32_t dst_ipv4, uint8_t portid)
1189 {
1190         uint8_t next_hop;
1191         struct ipv6_hdr *ipv6_hdr;
1192         struct ether_hdr *eth_hdr;
1193
1194         if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
1195                 if (rte_lpm_lookup(qconf->ipv4_lookup_struct, dst_ipv4,
1196                                 &next_hop) != 0)
1197                         next_hop = portid;
1198         } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
1199                 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1200                 ipv6_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
1201                 if (rte_lpm6_lookup(qconf->ipv6_lookup_struct,
1202                                 ipv6_hdr->dst_addr, &next_hop) != 0)
1203                         next_hop = portid;
1204         } else {
1205                 next_hop = portid;
1206         }
1207
1208         return next_hop;
1209 }
1210
1211 static inline void
1212 process_packet(struct lcore_conf *qconf, struct rte_mbuf *pkt,
1213         uint16_t *dst_port, uint8_t portid)
1214 {
1215         struct ether_hdr *eth_hdr;
1216         struct ipv4_hdr *ipv4_hdr;
1217         uint32_t dst_ipv4;
1218         uint16_t dp;
1219         __m128i te, ve;
1220
1221         eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1222         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1223
1224         dst_ipv4 = ipv4_hdr->dst_addr;
1225         dst_ipv4 = rte_be_to_cpu_32(dst_ipv4);
1226         dp = get_dst_port(qconf, pkt, dst_ipv4, portid);
1227
1228         te = _mm_loadu_si128((__m128i *)eth_hdr);
1229         ve = val_eth[dp];
1230
1231         dst_port[0] = dp;
1232         rfc1812_process(ipv4_hdr, dst_port, pkt->packet_type);
1233
1234         te =  _mm_blend_epi16(te, ve, MASK_ETH);
1235         _mm_storeu_si128((__m128i *)eth_hdr, te);
1236 }
1237
1238 /*
1239  * Read packet_type and destination IPV4 addresses from 4 mbufs.
1240  */
1241 static inline void
1242 processx4_step1(struct rte_mbuf *pkt[FWDSTEP],
1243                 __m128i *dip,
1244                 uint32_t *ipv4_flag)
1245 {
1246         struct ipv4_hdr *ipv4_hdr;
1247         struct ether_hdr *eth_hdr;
1248         uint32_t x0, x1, x2, x3;
1249
1250         eth_hdr = rte_pktmbuf_mtod(pkt[0], struct ether_hdr *);
1251         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1252         x0 = ipv4_hdr->dst_addr;
1253         ipv4_flag[0] = pkt[0]->packet_type & RTE_PTYPE_L3_IPV4;
1254
1255         eth_hdr = rte_pktmbuf_mtod(pkt[1], struct ether_hdr *);
1256         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1257         x1 = ipv4_hdr->dst_addr;
1258         ipv4_flag[0] &= pkt[1]->packet_type;
1259
1260         eth_hdr = rte_pktmbuf_mtod(pkt[2], struct ether_hdr *);
1261         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1262         x2 = ipv4_hdr->dst_addr;
1263         ipv4_flag[0] &= pkt[2]->packet_type;
1264
1265         eth_hdr = rte_pktmbuf_mtod(pkt[3], struct ether_hdr *);
1266         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1267         x3 = ipv4_hdr->dst_addr;
1268         ipv4_flag[0] &= pkt[3]->packet_type;
1269
1270         dip[0] = _mm_set_epi32(x3, x2, x1, x0);
1271 }
1272
1273 /*
1274  * Lookup into LPM for destination port.
1275  * If lookup fails, use incoming port (portid) as destination port.
1276  */
1277 static inline void
1278 processx4_step2(const struct lcore_conf *qconf,
1279                 __m128i dip,
1280                 uint32_t ipv4_flag,
1281                 uint8_t portid,
1282                 struct rte_mbuf *pkt[FWDSTEP],
1283                 uint16_t dprt[FWDSTEP])
1284 {
1285         rte_xmm_t dst;
1286         const  __m128i bswap_mask = _mm_set_epi8(12, 13, 14, 15, 8, 9, 10, 11,
1287                                                 4, 5, 6, 7, 0, 1, 2, 3);
1288
1289         /* Byte swap 4 IPV4 addresses. */
1290         dip = _mm_shuffle_epi8(dip, bswap_mask);
1291
1292         /* if all 4 packets are IPV4. */
1293         if (likely(ipv4_flag)) {
1294                 rte_lpm_lookupx4(qconf->ipv4_lookup_struct, dip, dprt, portid);
1295         } else {
1296                 dst.x = dip;
1297                 dprt[0] = get_dst_port(qconf, pkt[0], dst.u32[0], portid);
1298                 dprt[1] = get_dst_port(qconf, pkt[1], dst.u32[1], portid);
1299                 dprt[2] = get_dst_port(qconf, pkt[2], dst.u32[2], portid);
1300                 dprt[3] = get_dst_port(qconf, pkt[3], dst.u32[3], portid);
1301         }
1302 }
1303
1304 /*
1305  * Update source and destination MAC addresses in the ethernet header.
1306  * Perform RFC1812 checks and updates for IPV4 packets.
1307  */
1308 static inline void
1309 processx4_step3(struct rte_mbuf *pkt[FWDSTEP], uint16_t dst_port[FWDSTEP])
1310 {
1311         __m128i te[FWDSTEP];
1312         __m128i ve[FWDSTEP];
1313         __m128i *p[FWDSTEP];
1314
1315         p[0] = rte_pktmbuf_mtod(pkt[0], __m128i *);
1316         p[1] = rte_pktmbuf_mtod(pkt[1], __m128i *);
1317         p[2] = rte_pktmbuf_mtod(pkt[2], __m128i *);
1318         p[3] = rte_pktmbuf_mtod(pkt[3], __m128i *);
1319
1320         ve[0] = val_eth[dst_port[0]];
1321         te[0] = _mm_loadu_si128(p[0]);
1322
1323         ve[1] = val_eth[dst_port[1]];
1324         te[1] = _mm_loadu_si128(p[1]);
1325
1326         ve[2] = val_eth[dst_port[2]];
1327         te[2] = _mm_loadu_si128(p[2]);
1328
1329         ve[3] = val_eth[dst_port[3]];
1330         te[3] = _mm_loadu_si128(p[3]);
1331
1332         /* Update first 12 bytes, keep rest bytes intact. */
1333         te[0] =  _mm_blend_epi16(te[0], ve[0], MASK_ETH);
1334         te[1] =  _mm_blend_epi16(te[1], ve[1], MASK_ETH);
1335         te[2] =  _mm_blend_epi16(te[2], ve[2], MASK_ETH);
1336         te[3] =  _mm_blend_epi16(te[3], ve[3], MASK_ETH);
1337
1338         _mm_storeu_si128(p[0], te[0]);
1339         _mm_storeu_si128(p[1], te[1]);
1340         _mm_storeu_si128(p[2], te[2]);
1341         _mm_storeu_si128(p[3], te[3]);
1342
1343         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[0] + 1),
1344                 &dst_port[0], pkt[0]->packet_type);
1345         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[1] + 1),
1346                 &dst_port[1], pkt[1]->packet_type);
1347         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[2] + 1),
1348                 &dst_port[2], pkt[2]->packet_type);
1349         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[3] + 1),
1350                 &dst_port[3], pkt[3]->packet_type);
1351 }
1352
1353 /*
1354  * We group consecutive packets with the same destionation port into one burst.
1355  * To avoid extra latency this is done together with some other packet
1356  * processing, but after we made a final decision about packet's destination.
1357  * To do this we maintain:
1358  * pnum - array of number of consecutive packets with the same dest port for
1359  * each packet in the input burst.
1360  * lp - pointer to the last updated element in the pnum.
1361  * dlp - dest port value lp corresponds to.
1362  */
1363
1364 #define GRPSZ   (1 << FWDSTEP)
1365 #define GRPMSK  (GRPSZ - 1)
1366
1367 #define GROUP_PORT_STEP(dlp, dcp, lp, pn, idx)  do { \
1368         if (likely((dlp) == (dcp)[(idx)])) {         \
1369                 (lp)[0]++;                           \
1370         } else {                                     \
1371                 (dlp) = (dcp)[idx];                  \
1372                 (lp) = (pn) + (idx);                 \
1373                 (lp)[0] = 1;                         \
1374         }                                            \
1375 } while (0)
1376
1377 /*
1378  * Group consecutive packets with the same destination port in bursts of 4.
1379  * Suppose we have array of destionation ports:
1380  * dst_port[] = {a, b, c, d,, e, ... }
1381  * dp1 should contain: <a, b, c, d>, dp2: <b, c, d, e>.
1382  * We doing 4 comparisions at once and the result is 4 bit mask.
1383  * This mask is used as an index into prebuild array of pnum values.
1384  */
1385 static inline uint16_t *
1386 port_groupx4(uint16_t pn[FWDSTEP + 1], uint16_t *lp, __m128i dp1, __m128i dp2)
1387 {
1388         static const struct {
1389                 uint64_t pnum; /* prebuild 4 values for pnum[]. */
1390                 int32_t  idx;  /* index for new last updated elemnet. */
1391                 uint16_t lpv;  /* add value to the last updated element. */
1392         } gptbl[GRPSZ] = {
1393         {
1394                 /* 0: a != b, b != c, c != d, d != e */
1395                 .pnum = UINT64_C(0x0001000100010001),
1396                 .idx = 4,
1397                 .lpv = 0,
1398         },
1399         {
1400                 /* 1: a == b, b != c, c != d, d != e */
1401                 .pnum = UINT64_C(0x0001000100010002),
1402                 .idx = 4,
1403                 .lpv = 1,
1404         },
1405         {
1406                 /* 2: a != b, b == c, c != d, d != e */
1407                 .pnum = UINT64_C(0x0001000100020001),
1408                 .idx = 4,
1409                 .lpv = 0,
1410         },
1411         {
1412                 /* 3: a == b, b == c, c != d, d != e */
1413                 .pnum = UINT64_C(0x0001000100020003),
1414                 .idx = 4,
1415                 .lpv = 2,
1416         },
1417         {
1418                 /* 4: a != b, b != c, c == d, d != e */
1419                 .pnum = UINT64_C(0x0001000200010001),
1420                 .idx = 4,
1421                 .lpv = 0,
1422         },
1423         {
1424                 /* 5: a == b, b != c, c == d, d != e */
1425                 .pnum = UINT64_C(0x0001000200010002),
1426                 .idx = 4,
1427                 .lpv = 1,
1428         },
1429         {
1430                 /* 6: a != b, b == c, c == d, d != e */
1431                 .pnum = UINT64_C(0x0001000200030001),
1432                 .idx = 4,
1433                 .lpv = 0,
1434         },
1435         {
1436                 /* 7: a == b, b == c, c == d, d != e */
1437                 .pnum = UINT64_C(0x0001000200030004),
1438                 .idx = 4,
1439                 .lpv = 3,
1440         },
1441         {
1442                 /* 8: a != b, b != c, c != d, d == e */
1443                 .pnum = UINT64_C(0x0002000100010001),
1444                 .idx = 3,
1445                 .lpv = 0,
1446         },
1447         {
1448                 /* 9: a == b, b != c, c != d, d == e */
1449                 .pnum = UINT64_C(0x0002000100010002),
1450                 .idx = 3,
1451                 .lpv = 1,
1452         },
1453         {
1454                 /* 0xa: a != b, b == c, c != d, d == e */
1455                 .pnum = UINT64_C(0x0002000100020001),
1456                 .idx = 3,
1457                 .lpv = 0,
1458         },
1459         {
1460                 /* 0xb: a == b, b == c, c != d, d == e */
1461                 .pnum = UINT64_C(0x0002000100020003),
1462                 .idx = 3,
1463                 .lpv = 2,
1464         },
1465         {
1466                 /* 0xc: a != b, b != c, c == d, d == e */
1467                 .pnum = UINT64_C(0x0002000300010001),
1468                 .idx = 2,
1469                 .lpv = 0,
1470         },
1471         {
1472                 /* 0xd: a == b, b != c, c == d, d == e */
1473                 .pnum = UINT64_C(0x0002000300010002),
1474                 .idx = 2,
1475                 .lpv = 1,
1476         },
1477         {
1478                 /* 0xe: a != b, b == c, c == d, d == e */
1479                 .pnum = UINT64_C(0x0002000300040001),
1480                 .idx = 1,
1481                 .lpv = 0,
1482         },
1483         {
1484                 /* 0xf: a == b, b == c, c == d, d == e */
1485                 .pnum = UINT64_C(0x0002000300040005),
1486                 .idx = 0,
1487                 .lpv = 4,
1488         },
1489         };
1490
1491         union {
1492                 uint16_t u16[FWDSTEP + 1];
1493                 uint64_t u64;
1494         } *pnum = (void *)pn;
1495
1496         int32_t v;
1497
1498         dp1 = _mm_cmpeq_epi16(dp1, dp2);
1499         dp1 = _mm_unpacklo_epi16(dp1, dp1);
1500         v = _mm_movemask_ps((__m128)dp1);
1501
1502         /* update last port counter. */
1503         lp[0] += gptbl[v].lpv;
1504
1505         /* if dest port value has changed. */
1506         if (v != GRPMSK) {
1507                 lp = pnum->u16 + gptbl[v].idx;
1508                 lp[0] = 1;
1509                 pnum->u64 = gptbl[v].pnum;
1510         }
1511
1512         return lp;
1513 }
1514
1515 #endif /* APP_LOOKUP_METHOD */
1516
1517 /* main processing loop */
1518 static int
1519 main_loop(__attribute__((unused)) void *dummy)
1520 {
1521         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1522         unsigned lcore_id;
1523         uint64_t prev_tsc, diff_tsc, cur_tsc;
1524         int i, j, nb_rx;
1525         uint8_t portid, queueid;
1526         struct lcore_conf *qconf;
1527         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1528                 US_PER_S * BURST_TX_DRAIN_US;
1529
1530 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1531         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1532         int32_t k;
1533         uint16_t dlp;
1534         uint16_t *lp;
1535         uint16_t dst_port[MAX_PKT_BURST];
1536         __m128i dip[MAX_PKT_BURST / FWDSTEP];
1537         uint32_t ipv4_flag[MAX_PKT_BURST / FWDSTEP];
1538         uint16_t pnum[MAX_PKT_BURST + 1];
1539 #endif
1540
1541         prev_tsc = 0;
1542
1543         lcore_id = rte_lcore_id();
1544         qconf = &lcore_conf[lcore_id];
1545
1546         if (qconf->n_rx_queue == 0) {
1547                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
1548                 return 0;
1549         }
1550
1551         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
1552
1553         for (i = 0; i < qconf->n_rx_queue; i++) {
1554
1555                 portid = qconf->rx_queue_list[i].port_id;
1556                 queueid = qconf->rx_queue_list[i].queue_id;
1557                 RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n", lcore_id,
1558                         portid, queueid);
1559         }
1560
1561         while (!force_quit) {
1562
1563                 cur_tsc = rte_rdtsc();
1564
1565                 /*
1566                  * TX burst queue drain
1567                  */
1568                 diff_tsc = cur_tsc - prev_tsc;
1569                 if (unlikely(diff_tsc > drain_tsc)) {
1570
1571                         /*
1572                          * This could be optimized (use queueid instead of
1573                          * portid), but it is not called so often
1574                          */
1575                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
1576                                 if (qconf->tx_mbufs[portid].len == 0)
1577                                         continue;
1578                                 send_burst(qconf,
1579                                         qconf->tx_mbufs[portid].len,
1580                                         portid);
1581                                 qconf->tx_mbufs[portid].len = 0;
1582                         }
1583
1584                         prev_tsc = cur_tsc;
1585                 }
1586
1587                 /*
1588                  * Read packet from RX queues
1589                  */
1590                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1591                         portid = qconf->rx_queue_list[i].port_id;
1592                         queueid = qconf->rx_queue_list[i].queue_id;
1593                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1594                                 MAX_PKT_BURST);
1595                         if (nb_rx == 0)
1596                                 continue;
1597
1598 #if (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
1599 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1600                         {
1601                                 /*
1602                                  * Send nb_rx - nb_rx%8 packets
1603                                  * in groups of 8.
1604                                  */
1605                                 int32_t n = RTE_ALIGN_FLOOR(nb_rx, 8);
1606                                 for (j = 0; j < n; j += 8) {
1607                                         uint32_t pkt_type =
1608                                                 pkts_burst[j]->packet_type &
1609                                                 pkts_burst[j+1]->packet_type &
1610                                                 pkts_burst[j+2]->packet_type &
1611                                                 pkts_burst[j+3]->packet_type &
1612                                                 pkts_burst[j+4]->packet_type &
1613                                                 pkts_burst[j+5]->packet_type &
1614                                                 pkts_burst[j+6]->packet_type &
1615                                                 pkts_burst[j+7]->packet_type;
1616                                         if (pkt_type & RTE_PTYPE_L3_IPV4) {
1617                                                 simple_ipv4_fwd_8pkts(
1618                                                 &pkts_burst[j], portid, qconf);
1619                                         } else if (pkt_type &
1620                                                 RTE_PTYPE_L3_IPV6) {
1621                                                 simple_ipv6_fwd_8pkts(&pkts_burst[j],
1622                                                                         portid, qconf);
1623                                         } else {
1624                                                 l3fwd_simple_forward(pkts_burst[j],
1625                                                                         portid, qconf);
1626                                                 l3fwd_simple_forward(pkts_burst[j+1],
1627                                                                         portid, qconf);
1628                                                 l3fwd_simple_forward(pkts_burst[j+2],
1629                                                                         portid, qconf);
1630                                                 l3fwd_simple_forward(pkts_burst[j+3],
1631                                                                         portid, qconf);
1632                                                 l3fwd_simple_forward(pkts_burst[j+4],
1633                                                                         portid, qconf);
1634                                                 l3fwd_simple_forward(pkts_burst[j+5],
1635                                                                         portid, qconf);
1636                                                 l3fwd_simple_forward(pkts_burst[j+6],
1637                                                                         portid, qconf);
1638                                                 l3fwd_simple_forward(pkts_burst[j+7],
1639                                                                         portid, qconf);
1640                                         }
1641                                 }
1642                                 for (; j < nb_rx ; j++) {
1643                                         l3fwd_simple_forward(pkts_burst[j],
1644                                                                 portid, qconf);
1645                                 }
1646                         }
1647 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1648
1649                         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1650                         for (j = 0; j != k; j += FWDSTEP) {
1651                                 processx4_step1(&pkts_burst[j],
1652                                         &dip[j / FWDSTEP],
1653                                         &ipv4_flag[j / FWDSTEP]);
1654                         }
1655
1656                         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1657                         for (j = 0; j != k; j += FWDSTEP) {
1658                                 processx4_step2(qconf, dip[j / FWDSTEP],
1659                                         ipv4_flag[j / FWDSTEP], portid,
1660                                         &pkts_burst[j], &dst_port[j]);
1661                         }
1662
1663                         /*
1664                          * Finish packet processing and group consecutive
1665                          * packets with the same destination port.
1666                          */
1667                         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1668                         if (k != 0) {
1669                                 __m128i dp1, dp2;
1670
1671                                 lp = pnum;
1672                                 lp[0] = 1;
1673
1674                                 processx4_step3(pkts_burst, dst_port);
1675
1676                                 /* dp1: <d[0], d[1], d[2], d[3], ... > */
1677                                 dp1 = _mm_loadu_si128((__m128i *)dst_port);
1678
1679                                 for (j = FWDSTEP; j != k; j += FWDSTEP) {
1680                                         processx4_step3(&pkts_burst[j],
1681                                                 &dst_port[j]);
1682
1683                                         /*
1684                                          * dp2:
1685                                          * <d[j-3], d[j-2], d[j-1], d[j], ... >
1686                                          */
1687                                         dp2 = _mm_loadu_si128((__m128i *)
1688                                                 &dst_port[j - FWDSTEP + 1]);
1689                                         lp  = port_groupx4(&pnum[j - FWDSTEP],
1690                                                 lp, dp1, dp2);
1691
1692                                         /*
1693                                          * dp1:
1694                                          * <d[j], d[j+1], d[j+2], d[j+3], ... >
1695                                          */
1696                                         dp1 = _mm_srli_si128(dp2,
1697                                                 (FWDSTEP - 1) *
1698                                                 sizeof(dst_port[0]));
1699                                 }
1700
1701                                 /*
1702                                  * dp2: <d[j-3], d[j-2], d[j-1], d[j-1], ... >
1703                                  */
1704                                 dp2 = _mm_shufflelo_epi16(dp1, 0xf9);
1705                                 lp  = port_groupx4(&pnum[j - FWDSTEP], lp,
1706                                         dp1, dp2);
1707
1708                                 /*
1709                                  * remove values added by the last repeated
1710                                  * dst port.
1711                                  */
1712                                 lp[0]--;
1713                                 dlp = dst_port[j - 1];
1714                         } else {
1715                                 /* set dlp and lp to the never used values. */
1716                                 dlp = BAD_PORT - 1;
1717                                 lp = pnum + MAX_PKT_BURST;
1718                         }
1719
1720                         /* Process up to last 3 packets one by one. */
1721                         switch (nb_rx % FWDSTEP) {
1722                         case 3:
1723                                 process_packet(qconf, pkts_burst[j],
1724                                         dst_port + j, portid);
1725                                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1726                                 j++;
1727                         case 2:
1728                                 process_packet(qconf, pkts_burst[j],
1729                                         dst_port + j, portid);
1730                                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1731                                 j++;
1732                         case 1:
1733                                 process_packet(qconf, pkts_burst[j],
1734                                         dst_port + j, portid);
1735                                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1736                                 j++;
1737                         }
1738
1739                         /*
1740                          * Send packets out, through destination port.
1741                          * Consecuteve pacekts with the same destination port
1742                          * are already grouped together.
1743                          * If destination port for the packet equals BAD_PORT,
1744                          * then free the packet without sending it out.
1745                          */
1746                         for (j = 0; j < nb_rx; j += k) {
1747
1748                                 int32_t m;
1749                                 uint16_t pn;
1750
1751                                 pn = dst_port[j];
1752                                 k = pnum[j];
1753
1754                                 if (likely(pn != BAD_PORT)) {
1755                                         send_packetsx4(qconf, pn,
1756                                                 pkts_burst + j, k);
1757                                 } else {
1758                                         for (m = j; m != j + k; m++)
1759                                                 rte_pktmbuf_free(pkts_burst[m]);
1760                                 }
1761                         }
1762
1763 #endif /* APP_LOOKUP_METHOD */
1764 #else /* ENABLE_MULTI_BUFFER_OPTIMIZE == 0 */
1765
1766                         /* Prefetch first packets */
1767                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1768                                 rte_prefetch0(rte_pktmbuf_mtod(
1769                                                 pkts_burst[j], void *));
1770                         }
1771
1772                         /* Prefetch and forward already prefetched packets */
1773                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1774                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1775                                                 j + PREFETCH_OFFSET], void *));
1776                                 l3fwd_simple_forward(pkts_burst[j], portid,
1777                                         qconf);
1778                         }
1779
1780                         /* Forward remaining prefetched packets */
1781                         for (; j < nb_rx; j++) {
1782                                 l3fwd_simple_forward(pkts_burst[j], portid,
1783                                         qconf);
1784                         }
1785 #endif /* ENABLE_MULTI_BUFFER_OPTIMIZE */
1786
1787                 }
1788         }
1789
1790         return 0;
1791 }
1792
1793 static int
1794 check_lcore_params(void)
1795 {
1796         uint8_t queue, lcore;
1797         uint16_t i;
1798         int socketid;
1799
1800         for (i = 0; i < nb_lcore_params; ++i) {
1801                 queue = lcore_params[i].queue_id;
1802                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1803                         printf("invalid queue number: %hhu\n", queue);
1804                         return -1;
1805                 }
1806                 lcore = lcore_params[i].lcore_id;
1807                 if (!rte_lcore_is_enabled(lcore)) {
1808                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
1809                         return -1;
1810                 }
1811                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1812                         (numa_on == 0)) {
1813                         printf("warning: lcore %hhu is on socket %d with numa off \n",
1814                                 lcore, socketid);
1815                 }
1816         }
1817         return 0;
1818 }
1819
1820 static int
1821 check_port_config(const unsigned nb_ports)
1822 {
1823         unsigned portid;
1824         uint16_t i;
1825
1826         for (i = 0; i < nb_lcore_params; ++i) {
1827                 portid = lcore_params[i].port_id;
1828                 if ((enabled_port_mask & (1 << portid)) == 0) {
1829                         printf("port %u is not enabled in port mask\n", portid);
1830                         return -1;
1831                 }
1832                 if (portid >= nb_ports) {
1833                         printf("port %u is not present on the board\n", portid);
1834                         return -1;
1835                 }
1836         }
1837         return 0;
1838 }
1839
1840 static uint8_t
1841 get_port_n_rx_queues(const uint8_t port)
1842 {
1843         int queue = -1;
1844         uint16_t i;
1845
1846         for (i = 0; i < nb_lcore_params; ++i) {
1847                 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
1848                         queue = lcore_params[i].queue_id;
1849         }
1850         return (uint8_t)(++queue);
1851 }
1852
1853 static int
1854 init_lcore_rx_queues(void)
1855 {
1856         uint16_t i, nb_rx_queue;
1857         uint8_t lcore;
1858
1859         for (i = 0; i < nb_lcore_params; ++i) {
1860                 lcore = lcore_params[i].lcore_id;
1861                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1862                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1863                         printf("error: too many queues (%u) for lcore: %u\n",
1864                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1865                         return -1;
1866                 } else {
1867                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1868                                 lcore_params[i].port_id;
1869                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1870                                 lcore_params[i].queue_id;
1871                         lcore_conf[lcore].n_rx_queue++;
1872                 }
1873         }
1874         return 0;
1875 }
1876
1877 /* display usage */
1878 static void
1879 print_usage(const char *prgname)
1880 {
1881         printf ("%s [EAL options] -- -p PORTMASK -P"
1882                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1883                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1884                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1885                 "  -P : enable promiscuous mode\n"
1886                 "  --config (port,queue,lcore): rx queues configuration\n"
1887                 "  --eth-dest=X,MM:MM:MM:MM:MM:MM: optional, ethernet destination for port X\n"
1888                 "  --no-numa: optional, disable numa awareness\n"
1889                 "  --ipv6: optional, specify it if running ipv6 packets\n"
1890                 "  --enable-jumbo: enable jumbo frame"
1891                 " which max packet len is PKTLEN in decimal (64-9600)\n"
1892                 "  --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n",
1893                 prgname);
1894 }
1895
1896 static int parse_max_pkt_len(const char *pktlen)
1897 {
1898         char *end = NULL;
1899         unsigned long len;
1900
1901         /* parse decimal string */
1902         len = strtoul(pktlen, &end, 10);
1903         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1904                 return -1;
1905
1906         if (len == 0)
1907                 return -1;
1908
1909         return len;
1910 }
1911
1912 static int
1913 parse_portmask(const char *portmask)
1914 {
1915         char *end = NULL;
1916         unsigned long pm;
1917
1918         /* parse hexadecimal string */
1919         pm = strtoul(portmask, &end, 16);
1920         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1921                 return -1;
1922
1923         if (pm == 0)
1924                 return -1;
1925
1926         return pm;
1927 }
1928
1929 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1930 static int
1931 parse_hash_entry_number(const char *hash_entry_num)
1932 {
1933         char *end = NULL;
1934         unsigned long hash_en;
1935         /* parse hexadecimal string */
1936         hash_en = strtoul(hash_entry_num, &end, 16);
1937         if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
1938                 return -1;
1939
1940         if (hash_en == 0)
1941                 return -1;
1942
1943         return hash_en;
1944 }
1945 #endif
1946
1947 static int
1948 parse_config(const char *q_arg)
1949 {
1950         char s[256];
1951         const char *p, *p0 = q_arg;
1952         char *end;
1953         enum fieldnames {
1954                 FLD_PORT = 0,
1955                 FLD_QUEUE,
1956                 FLD_LCORE,
1957                 _NUM_FLD
1958         };
1959         unsigned long int_fld[_NUM_FLD];
1960         char *str_fld[_NUM_FLD];
1961         int i;
1962         unsigned size;
1963
1964         nb_lcore_params = 0;
1965
1966         while ((p = strchr(p0,'(')) != NULL) {
1967                 ++p;
1968                 if((p0 = strchr(p,')')) == NULL)
1969                         return -1;
1970
1971                 size = p0 - p;
1972                 if(size >= sizeof(s))
1973                         return -1;
1974
1975                 snprintf(s, sizeof(s), "%.*s", size, p);
1976                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
1977                         return -1;
1978                 for (i = 0; i < _NUM_FLD; i++){
1979                         errno = 0;
1980                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1981                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1982                                 return -1;
1983                 }
1984                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1985                         printf("exceeded max number of lcore params: %hu\n",
1986                                 nb_lcore_params);
1987                         return -1;
1988                 }
1989                 lcore_params_array[nb_lcore_params].port_id = (uint8_t)int_fld[FLD_PORT];
1990                 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
1991                 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
1992                 ++nb_lcore_params;
1993         }
1994         lcore_params = lcore_params_array;
1995         return 0;
1996 }
1997
1998 static void
1999 parse_eth_dest(const char *optarg)
2000 {
2001         uint8_t portid;
2002         char *port_end;
2003         uint8_t c, *dest, peer_addr[6];
2004
2005         errno = 0;
2006         portid = strtoul(optarg, &port_end, 10);
2007         if (errno != 0 || port_end == optarg || *port_end++ != ',')
2008                 rte_exit(EXIT_FAILURE,
2009                 "Invalid eth-dest: %s", optarg);
2010         if (portid >= RTE_MAX_ETHPORTS)
2011                 rte_exit(EXIT_FAILURE,
2012                 "eth-dest: port %d >= RTE_MAX_ETHPORTS(%d)\n",
2013                 portid, RTE_MAX_ETHPORTS);
2014
2015         if (cmdline_parse_etheraddr(NULL, port_end,
2016                 &peer_addr, sizeof(peer_addr)) < 0)
2017                 rte_exit(EXIT_FAILURE,
2018                 "Invalid ethernet address: %s\n",
2019                 port_end);
2020         dest = (uint8_t *)&dest_eth_addr[portid];
2021         for (c = 0; c < 6; c++)
2022                 dest[c] = peer_addr[c];
2023         *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
2024 }
2025
2026 #define CMD_LINE_OPT_CONFIG "config"
2027 #define CMD_LINE_OPT_ETH_DEST "eth-dest"
2028 #define CMD_LINE_OPT_NO_NUMA "no-numa"
2029 #define CMD_LINE_OPT_IPV6 "ipv6"
2030 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
2031 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
2032
2033 /* Parse the argument given in the command line of the application */
2034 static int
2035 parse_args(int argc, char **argv)
2036 {
2037         int opt, ret;
2038         char **argvopt;
2039         int option_index;
2040         char *prgname = argv[0];
2041         static struct option lgopts[] = {
2042                 {CMD_LINE_OPT_CONFIG, 1, 0, 0},
2043                 {CMD_LINE_OPT_ETH_DEST, 1, 0, 0},
2044                 {CMD_LINE_OPT_NO_NUMA, 0, 0, 0},
2045                 {CMD_LINE_OPT_IPV6, 0, 0, 0},
2046                 {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, 0},
2047                 {CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, 0},
2048                 {NULL, 0, 0, 0}
2049         };
2050
2051         argvopt = argv;
2052
2053         while ((opt = getopt_long(argc, argvopt, "p:P",
2054                                 lgopts, &option_index)) != EOF) {
2055
2056                 switch (opt) {
2057                 /* portmask */
2058                 case 'p':
2059                         enabled_port_mask = parse_portmask(optarg);
2060                         if (enabled_port_mask == 0) {
2061                                 printf("invalid portmask\n");
2062                                 print_usage(prgname);
2063                                 return -1;
2064                         }
2065                         break;
2066                 case 'P':
2067                         printf("Promiscuous mode selected\n");
2068                         promiscuous_on = 1;
2069                         break;
2070
2071                 /* long options */
2072                 case 0:
2073                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_CONFIG,
2074                                 sizeof (CMD_LINE_OPT_CONFIG))) {
2075                                 ret = parse_config(optarg);
2076                                 if (ret) {
2077                                         printf("invalid config\n");
2078                                         print_usage(prgname);
2079                                         return -1;
2080                                 }
2081                         }
2082
2083                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ETH_DEST,
2084                                 sizeof(CMD_LINE_OPT_ETH_DEST))) {
2085                                         parse_eth_dest(optarg);
2086                         }
2087
2088                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_NUMA,
2089                                 sizeof(CMD_LINE_OPT_NO_NUMA))) {
2090                                 printf("numa is disabled \n");
2091                                 numa_on = 0;
2092                         }
2093
2094 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2095                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_IPV6,
2096                                 sizeof(CMD_LINE_OPT_IPV6))) {
2097                                 printf("ipv6 is specified \n");
2098                                 ipv6 = 1;
2099                         }
2100 #endif
2101
2102                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ENABLE_JUMBO,
2103                                 sizeof (CMD_LINE_OPT_ENABLE_JUMBO))) {
2104                                 struct option lenopts = {"max-pkt-len", required_argument, 0, 0};
2105
2106                                 printf("jumbo frame is enabled - disabling simple TX path\n");
2107                                 port_conf.rxmode.jumbo_frame = 1;
2108
2109                                 /* if no max-pkt-len set, use the default value ETHER_MAX_LEN */
2110                                 if (0 == getopt_long(argc, argvopt, "", &lenopts, &option_index)) {
2111                                         ret = parse_max_pkt_len(optarg);
2112                                         if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)){
2113                                                 printf("invalid packet length\n");
2114                                                 print_usage(prgname);
2115                                                 return -1;
2116                                         }
2117                                         port_conf.rxmode.max_rx_pkt_len = ret;
2118                                 }
2119                                 printf("set jumbo frame max packet length to %u\n",
2120                                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
2121                         }
2122 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2123                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_HASH_ENTRY_NUM,
2124                                 sizeof(CMD_LINE_OPT_HASH_ENTRY_NUM))) {
2125                                 ret = parse_hash_entry_number(optarg);
2126                                 if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
2127                                         hash_entry_number = ret;
2128                                 } else {
2129                                         printf("invalid hash entry number\n");
2130                                         print_usage(prgname);
2131                                         return -1;
2132                                 }
2133                         }
2134 #endif
2135                         break;
2136
2137                 default:
2138                         print_usage(prgname);
2139                         return -1;
2140                 }
2141         }
2142
2143         if (optind >= 0)
2144                 argv[optind-1] = prgname;
2145
2146         ret = optind-1;
2147         optind = 0; /* reset getopt lib */
2148         return ret;
2149 }
2150
2151 static void
2152 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
2153 {
2154         char buf[ETHER_ADDR_FMT_SIZE];
2155         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
2156         printf("%s%s", name, buf);
2157 }
2158
2159 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2160
2161 static void convert_ipv4_5tuple(struct ipv4_5tuple* key1,
2162                 union ipv4_5tuple_host* key2)
2163 {
2164         key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
2165         key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
2166         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
2167         key2->port_src = rte_cpu_to_be_16(key1->port_src);
2168         key2->proto = key1->proto;
2169         key2->pad0 = 0;
2170         key2->pad1 = 0;
2171         return;
2172 }
2173
2174 static void convert_ipv6_5tuple(struct ipv6_5tuple* key1,
2175                 union ipv6_5tuple_host* key2)
2176 {
2177         uint32_t i;
2178         for (i = 0; i < 16; i++)
2179         {
2180                 key2->ip_dst[i] = key1->ip_dst[i];
2181                 key2->ip_src[i] = key1->ip_src[i];
2182         }
2183         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
2184         key2->port_src = rte_cpu_to_be_16(key1->port_src);
2185         key2->proto = key1->proto;
2186         key2->pad0 = 0;
2187         key2->pad1 = 0;
2188         key2->reserve = 0;
2189         return;
2190 }
2191
2192 #define BYTE_VALUE_MAX 256
2193 #define ALL_32_BITS 0xffffffff
2194 #define BIT_8_TO_15 0x0000ff00
2195 static inline void
2196 populate_ipv4_few_flow_into_table(const struct rte_hash* h)
2197 {
2198         uint32_t i;
2199         int32_t ret;
2200         uint32_t array_len = sizeof(ipv4_l3fwd_route_array)/sizeof(ipv4_l3fwd_route_array[0]);
2201
2202         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
2203         for (i = 0; i < array_len; i++) {
2204                 struct ipv4_l3fwd_route  entry;
2205                 union ipv4_5tuple_host newkey;
2206                 entry = ipv4_l3fwd_route_array[i];
2207                 convert_ipv4_5tuple(&entry.key, &newkey);
2208                 ret = rte_hash_add_key (h,(void *) &newkey);
2209                 if (ret < 0) {
2210                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
2211                                 " to the l3fwd hash.\n", i);
2212                 }
2213                 ipv4_l3fwd_out_if[ret] = entry.if_out;
2214         }
2215         printf("Hash: Adding 0x%" PRIx32 " keys\n", array_len);
2216 }
2217
2218 #define BIT_16_TO_23 0x00ff0000
2219 static inline void
2220 populate_ipv6_few_flow_into_table(const struct rte_hash* h)
2221 {
2222         uint32_t i;
2223         int32_t ret;
2224         uint32_t array_len = sizeof(ipv6_l3fwd_route_array)/sizeof(ipv6_l3fwd_route_array[0]);
2225
2226         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
2227         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
2228         for (i = 0; i < array_len; i++) {
2229                 struct ipv6_l3fwd_route entry;
2230                 union ipv6_5tuple_host newkey;
2231                 entry = ipv6_l3fwd_route_array[i];
2232                 convert_ipv6_5tuple(&entry.key, &newkey);
2233                 ret = rte_hash_add_key (h, (void *) &newkey);
2234                 if (ret < 0) {
2235                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
2236                                 " to the l3fwd hash.\n", i);
2237                 }
2238                 ipv6_l3fwd_out_if[ret] = entry.if_out;
2239         }
2240         printf("Hash: Adding 0x%" PRIx32 "keys\n", array_len);
2241 }
2242
2243 #define NUMBER_PORT_USED 4
2244 static inline void
2245 populate_ipv4_many_flow_into_table(const struct rte_hash* h,
2246                 unsigned int nr_flow)
2247 {
2248         unsigned i;
2249         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
2250         for (i = 0; i < nr_flow; i++) {
2251                 struct ipv4_l3fwd_route entry;
2252                 union ipv4_5tuple_host newkey;
2253                 uint8_t a = (uint8_t) ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
2254                 uint8_t b = (uint8_t) (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
2255                 uint8_t c = (uint8_t) ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
2256                 /* Create the ipv4 exact match flow */
2257                 memset(&entry, 0, sizeof(entry));
2258                 switch (i & (NUMBER_PORT_USED -1)) {
2259                 case 0:
2260                         entry = ipv4_l3fwd_route_array[0];
2261                         entry.key.ip_dst = IPv4(101,c,b,a);
2262                         break;
2263                 case 1:
2264                         entry = ipv4_l3fwd_route_array[1];
2265                         entry.key.ip_dst = IPv4(201,c,b,a);
2266                         break;
2267                 case 2:
2268                         entry = ipv4_l3fwd_route_array[2];
2269                         entry.key.ip_dst = IPv4(111,c,b,a);
2270                         break;
2271                 case 3:
2272                         entry = ipv4_l3fwd_route_array[3];
2273                         entry.key.ip_dst = IPv4(211,c,b,a);
2274                         break;
2275                 };
2276                 convert_ipv4_5tuple(&entry.key, &newkey);
2277                 int32_t ret = rte_hash_add_key(h,(void *) &newkey);
2278                 if (ret < 0) {
2279                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
2280                 }
2281                 ipv4_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
2282
2283         }
2284         printf("Hash: Adding 0x%x keys\n", nr_flow);
2285 }
2286
2287 static inline void
2288 populate_ipv6_many_flow_into_table(const struct rte_hash* h,
2289                 unsigned int nr_flow)
2290 {
2291         unsigned i;
2292         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
2293         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
2294         for (i = 0; i < nr_flow; i++) {
2295                 struct ipv6_l3fwd_route entry;
2296                 union ipv6_5tuple_host newkey;
2297                 uint8_t a = (uint8_t) ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
2298                 uint8_t b = (uint8_t) (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
2299                 uint8_t c = (uint8_t) ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
2300                 /* Create the ipv6 exact match flow */
2301                 memset(&entry, 0, sizeof(entry));
2302                 switch (i & (NUMBER_PORT_USED - 1)) {
2303                 case 0: entry = ipv6_l3fwd_route_array[0]; break;
2304                 case 1: entry = ipv6_l3fwd_route_array[1]; break;
2305                 case 2: entry = ipv6_l3fwd_route_array[2]; break;
2306                 case 3: entry = ipv6_l3fwd_route_array[3]; break;
2307                 };
2308                 entry.key.ip_dst[13] = c;
2309                 entry.key.ip_dst[14] = b;
2310                 entry.key.ip_dst[15] = a;
2311                 convert_ipv6_5tuple(&entry.key, &newkey);
2312                 int32_t ret = rte_hash_add_key(h,(void *) &newkey);
2313                 if (ret < 0) {
2314                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
2315                 }
2316                 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
2317
2318         }
2319         printf("Hash: Adding 0x%x keys\n", nr_flow);
2320 }
2321
2322 static void
2323 setup_hash(int socketid)
2324 {
2325     struct rte_hash_parameters ipv4_l3fwd_hash_params = {
2326         .name = NULL,
2327         .entries = L3FWD_HASH_ENTRIES,
2328         .key_len = sizeof(union ipv4_5tuple_host),
2329         .hash_func = ipv4_hash_crc,
2330         .hash_func_init_val = 0,
2331     };
2332
2333     struct rte_hash_parameters ipv6_l3fwd_hash_params = {
2334         .name = NULL,
2335         .entries = L3FWD_HASH_ENTRIES,
2336         .key_len = sizeof(union ipv6_5tuple_host),
2337         .hash_func = ipv6_hash_crc,
2338         .hash_func_init_val = 0,
2339     };
2340
2341     char s[64];
2342
2343         /* create ipv4 hash */
2344         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
2345         ipv4_l3fwd_hash_params.name = s;
2346         ipv4_l3fwd_hash_params.socket_id = socketid;
2347         ipv4_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv4_l3fwd_hash_params);
2348         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2349                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2350                                 "socket %d\n", socketid);
2351
2352         /* create ipv6 hash */
2353         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
2354         ipv6_l3fwd_hash_params.name = s;
2355         ipv6_l3fwd_hash_params.socket_id = socketid;
2356         ipv6_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv6_l3fwd_hash_params);
2357         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
2358                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2359                                 "socket %d\n", socketid);
2360
2361         if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
2362                 /* For testing hash matching with a large number of flows we
2363                  * generate millions of IP 5-tuples with an incremented dst
2364                  * address to initialize the hash table. */
2365                 if (ipv6 == 0) {
2366                         /* populate the ipv4 hash */
2367                         populate_ipv4_many_flow_into_table(
2368                                 ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
2369                 } else {
2370                         /* populate the ipv6 hash */
2371                         populate_ipv6_many_flow_into_table(
2372                                 ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
2373                 }
2374         } else {
2375                 /* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize the hash table */
2376                 if (ipv6 == 0) {
2377                         /* populate the ipv4 hash */
2378                         populate_ipv4_few_flow_into_table(ipv4_l3fwd_lookup_struct[socketid]);
2379                 } else {
2380                         /* populate the ipv6 hash */
2381                         populate_ipv6_few_flow_into_table(ipv6_l3fwd_lookup_struct[socketid]);
2382                 }
2383         }
2384 }
2385 #endif
2386
2387 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2388 static void
2389 setup_lpm(int socketid)
2390 {
2391         struct rte_lpm6_config config;
2392         unsigned i;
2393         int ret;
2394         char s[64];
2395
2396         /* create the LPM table */
2397         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
2398         ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
2399                                 IPV4_L3FWD_LPM_MAX_RULES, 0);
2400         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2401                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
2402                                 " on socket %d\n", socketid);
2403
2404         /* populate the LPM table */
2405         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
2406
2407                 /* skip unused ports */
2408                 if ((1 << ipv4_l3fwd_route_array[i].if_out &
2409                                 enabled_port_mask) == 0)
2410                         continue;
2411
2412                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
2413                         ipv4_l3fwd_route_array[i].ip,
2414                         ipv4_l3fwd_route_array[i].depth,
2415                         ipv4_l3fwd_route_array[i].if_out);
2416
2417                 if (ret < 0) {
2418                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
2419                                 "l3fwd LPM table on socket %d\n",
2420                                 i, socketid);
2421                 }
2422
2423                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
2424                         (unsigned)ipv4_l3fwd_route_array[i].ip,
2425                         ipv4_l3fwd_route_array[i].depth,
2426                         ipv4_l3fwd_route_array[i].if_out);
2427         }
2428
2429         /* create the LPM6 table */
2430         snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
2431
2432         config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
2433         config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
2434         config.flags = 0;
2435         ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
2436                                 &config);
2437         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
2438                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
2439                                 " on socket %d\n", socketid);
2440
2441         /* populate the LPM table */
2442         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
2443
2444                 /* skip unused ports */
2445                 if ((1 << ipv6_l3fwd_route_array[i].if_out &
2446                                 enabled_port_mask) == 0)
2447                         continue;
2448
2449                 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
2450                         ipv6_l3fwd_route_array[i].ip,
2451                         ipv6_l3fwd_route_array[i].depth,
2452                         ipv6_l3fwd_route_array[i].if_out);
2453
2454                 if (ret < 0) {
2455                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
2456                                 "l3fwd LPM table on socket %d\n",
2457                                 i, socketid);
2458                 }
2459
2460                 printf("LPM: Adding route %s / %d (%d)\n",
2461                         "IPV6",
2462                         ipv6_l3fwd_route_array[i].depth,
2463                         ipv6_l3fwd_route_array[i].if_out);
2464         }
2465 }
2466 #endif
2467
2468 static int
2469 init_mem(unsigned nb_mbuf)
2470 {
2471         struct lcore_conf *qconf;
2472         int socketid;
2473         unsigned lcore_id;
2474         char s[64];
2475
2476         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2477                 if (rte_lcore_is_enabled(lcore_id) == 0)
2478                         continue;
2479
2480                 if (numa_on)
2481                         socketid = rte_lcore_to_socket_id(lcore_id);
2482                 else
2483                         socketid = 0;
2484
2485                 if (socketid >= NB_SOCKETS) {
2486                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
2487                                 socketid, lcore_id, NB_SOCKETS);
2488                 }
2489                 if (pktmbuf_pool[socketid] == NULL) {
2490                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
2491                         pktmbuf_pool[socketid] =
2492                                 rte_pktmbuf_pool_create(s, nb_mbuf,
2493                                         MEMPOOL_CACHE_SIZE, 0,
2494                                         RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
2495                         if (pktmbuf_pool[socketid] == NULL)
2496                                 rte_exit(EXIT_FAILURE,
2497                                                 "Cannot init mbuf pool on socket %d\n", socketid);
2498                         else
2499                                 printf("Allocated mbuf pool on socket %d\n", socketid);
2500
2501 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2502                         setup_lpm(socketid);
2503 #else
2504                         setup_hash(socketid);
2505 #endif
2506                 }
2507                 qconf = &lcore_conf[lcore_id];
2508                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
2509                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
2510         }
2511         return 0;
2512 }
2513
2514 /* Check the link status of all ports in up to 9s, and print them finally */
2515 static void
2516 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
2517 {
2518 #define CHECK_INTERVAL 100 /* 100ms */
2519 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
2520         uint8_t portid, count, all_ports_up, print_flag = 0;
2521         struct rte_eth_link link;
2522
2523         printf("\nChecking link status");
2524         fflush(stdout);
2525         for (count = 0; count <= MAX_CHECK_TIME; count++) {
2526                 if (force_quit)
2527                         return;
2528                 all_ports_up = 1;
2529                 for (portid = 0; portid < port_num; portid++) {
2530                         if (force_quit)
2531                                 return;
2532                         if ((port_mask & (1 << portid)) == 0)
2533                                 continue;
2534                         memset(&link, 0, sizeof(link));
2535                         rte_eth_link_get_nowait(portid, &link);
2536                         /* print link status if flag set */
2537                         if (print_flag == 1) {
2538                                 if (link.link_status)
2539                                         printf("Port %d Link Up - speed %u "
2540                                                 "Mbps - %s\n", (uint8_t)portid,
2541                                                 (unsigned)link.link_speed,
2542                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
2543                                         ("full-duplex") : ("half-duplex\n"));
2544                                 else
2545                                         printf("Port %d Link Down\n",
2546                                                 (uint8_t)portid);
2547                                 continue;
2548                         }
2549                         /* clear all_ports_up flag if any link down */
2550                         if (link.link_status == 0) {
2551                                 all_ports_up = 0;
2552                                 break;
2553                         }
2554                 }
2555                 /* after finally printing all link status, get out */
2556                 if (print_flag == 1)
2557                         break;
2558
2559                 if (all_ports_up == 0) {
2560                         printf(".");
2561                         fflush(stdout);
2562                         rte_delay_ms(CHECK_INTERVAL);
2563                 }
2564
2565                 /* set the print_flag if all ports up or timeout */
2566                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
2567                         print_flag = 1;
2568                         printf("done\n");
2569                 }
2570         }
2571 }
2572
2573 static void
2574 signal_handler(int signum)
2575 {
2576         if (signum == SIGINT || signum == SIGTERM) {
2577                 printf("\n\nSignal %d received, preparing to exit...\n",
2578                                 signum);
2579                 force_quit = true;
2580         }
2581 }
2582
2583 int
2584 main(int argc, char **argv)
2585 {
2586         struct lcore_conf *qconf;
2587         struct rte_eth_dev_info dev_info;
2588         struct rte_eth_txconf *txconf;
2589         int ret;
2590         unsigned nb_ports;
2591         uint16_t queueid;
2592         unsigned lcore_id;
2593         uint32_t n_tx_queue, nb_lcores;
2594         uint8_t portid, nb_rx_queue, queue, socketid;
2595
2596         /* init EAL */
2597         ret = rte_eal_init(argc, argv);
2598         if (ret < 0)
2599                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2600         argc -= ret;
2601         argv += ret;
2602
2603         force_quit = false;
2604         signal(SIGINT, signal_handler);
2605         signal(SIGTERM, signal_handler);
2606
2607         /* pre-init dst MACs for all ports to 02:00:00:00:00:xx */
2608         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2609                 dest_eth_addr[portid] = ETHER_LOCAL_ADMIN_ADDR + ((uint64_t)portid << 40);
2610                 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
2611         }
2612
2613         /* parse application arguments (after the EAL ones) */
2614         ret = parse_args(argc, argv);
2615         if (ret < 0)
2616                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
2617
2618         if (check_lcore_params() < 0)
2619                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
2620
2621         ret = init_lcore_rx_queues();
2622         if (ret < 0)
2623                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2624
2625         nb_ports = rte_eth_dev_count();
2626         if (nb_ports > RTE_MAX_ETHPORTS)
2627                 nb_ports = RTE_MAX_ETHPORTS;
2628
2629         if (check_port_config(nb_ports) < 0)
2630                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
2631
2632         nb_lcores = rte_lcore_count();
2633
2634         /* initialize all ports */
2635         for (portid = 0; portid < nb_ports; portid++) {
2636                 /* skip ports that are not enabled */
2637                 if ((enabled_port_mask & (1 << portid)) == 0) {
2638                         printf("\nSkipping disabled port %d\n", portid);
2639                         continue;
2640                 }
2641
2642                 /* init port */
2643                 printf("Initializing port %d ... ", portid );
2644                 fflush(stdout);
2645
2646                 nb_rx_queue = get_port_n_rx_queues(portid);
2647                 n_tx_queue = nb_lcores;
2648                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
2649                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
2650                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
2651                         nb_rx_queue, (unsigned)n_tx_queue );
2652                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
2653                                         (uint16_t)n_tx_queue, &port_conf);
2654                 if (ret < 0)
2655                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
2656                                 ret, portid);
2657
2658                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
2659                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
2660                 printf(", ");
2661                 print_ethaddr("Destination:",
2662                         (const struct ether_addr *)&dest_eth_addr[portid]);
2663                 printf(", ");
2664
2665                 /*
2666                  * prepare src MACs for each port.
2667                  */
2668                 ether_addr_copy(&ports_eth_addr[portid],
2669                         (struct ether_addr *)(val_eth + portid) + 1);
2670
2671                 /* init memory */
2672                 ret = init_mem(NB_MBUF);
2673                 if (ret < 0)
2674                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
2675
2676                 /* init one TX queue per couple (lcore,port) */
2677                 queueid = 0;
2678                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2679                         if (rte_lcore_is_enabled(lcore_id) == 0)
2680                                 continue;
2681
2682                         if (numa_on)
2683                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2684                         else
2685                                 socketid = 0;
2686
2687                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
2688                         fflush(stdout);
2689
2690                         rte_eth_dev_info_get(portid, &dev_info);
2691                         txconf = &dev_info.default_txconf;
2692                         if (port_conf.rxmode.jumbo_frame)
2693                                 txconf->txq_flags = 0;
2694                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
2695                                                      socketid, txconf);
2696                         if (ret < 0)
2697                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
2698                                         "port=%d\n", ret, portid);
2699
2700                         qconf = &lcore_conf[lcore_id];
2701                         qconf->tx_queue_id[portid] = queueid;
2702                         queueid++;
2703                 }
2704                 printf("\n");
2705         }
2706
2707         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2708                 if (rte_lcore_is_enabled(lcore_id) == 0)
2709                         continue;
2710                 qconf = &lcore_conf[lcore_id];
2711                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
2712                 fflush(stdout);
2713                 /* init RX queues */
2714                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
2715                         portid = qconf->rx_queue_list[queue].port_id;
2716                         queueid = qconf->rx_queue_list[queue].queue_id;
2717
2718                         if (numa_on)
2719                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2720                         else
2721                                 socketid = 0;
2722
2723                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
2724                         fflush(stdout);
2725
2726                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
2727                                         socketid,
2728                                         NULL,
2729                                         pktmbuf_pool[socketid]);
2730                         if (ret < 0)
2731                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
2732                                                 "port=%d\n", ret, portid);
2733                 }
2734         }
2735
2736         printf("\n");
2737
2738         /* start ports */
2739         for (portid = 0; portid < nb_ports; portid++) {
2740                 if ((enabled_port_mask & (1 << portid)) == 0) {
2741                         continue;
2742                 }
2743                 /* Start device */
2744                 ret = rte_eth_dev_start(portid);
2745                 if (ret < 0)
2746                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
2747                                 ret, portid);
2748
2749                 /*
2750                  * If enabled, put device in promiscuous mode.
2751                  * This allows IO forwarding mode to forward packets
2752                  * to itself through 2 cross-connected  ports of the
2753                  * target machine.
2754                  */
2755                 if (promiscuous_on)
2756                         rte_eth_promiscuous_enable(portid);
2757         }
2758
2759         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
2760
2761         ret = 0;
2762         /* launch per-lcore init on every lcore */
2763         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2764         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2765                 if (rte_eal_wait_lcore(lcore_id) < 0) {
2766                         ret = -1;
2767                         break;
2768                 }
2769         }
2770
2771         /* stop ports */
2772         for (portid = 0; portid < nb_ports; portid++) {
2773                 if ((enabled_port_mask & (1 << portid)) == 0)
2774                         continue;
2775                 printf("Closing port %d...", portid);
2776                 rte_eth_dev_stop(portid);
2777                 rte_eth_dev_close(portid);
2778                 printf(" Done\n");
2779         }
2780         printf("Bye...\n");
2781
2782         return ret;
2783 }