examples/l3fwd/main.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <stdio.h>
  35 #include <stdlib.h>
  36 #include <stdint.h>
  37 #include <inttypes.h>
  38 #include <sys/types.h>
  39 #include <string.h>
  40 #include <sys/queue.h>
  41 #include <stdarg.h>
  42 #include <errno.h>
  43 #include <getopt.h>
  44
  45 #include <tmmintrin.h>
  46 #include <rte_common.h>
  47 #include <rte_byteorder.h>
  48 #include <rte_log.h>
  49 #include <rte_memory.h>
  50 #include <rte_memcpy.h>
  51 #include <rte_memzone.h>
  52 #include <rte_tailq.h>
  53 #include <rte_eal.h>
  54 #include <rte_per_lcore.h>
  55 #include <rte_launch.h>
  56 #include <rte_atomic.h>
  57 #include <rte_cycles.h>
  58 #include <rte_prefetch.h>
  59 #include <rte_lcore.h>
  60 #include <rte_per_lcore.h>
  61 #include <rte_branch_prediction.h>
  62 #include <rte_interrupts.h>
  63 #include <rte_pci.h>
  64 #include <rte_random.h>
  65 #include <rte_debug.h>
  66 #include <rte_ether.h>
  67 #include <rte_ethdev.h>
  68 #include <rte_ring.h>
  69 #include <rte_mempool.h>
  70 #include <rte_mbuf.h>
  71 #include <rte_ip.h>
  72 #include <rte_tcp.h>
  73 #include <rte_udp.h>
  74 #include <rte_string_fns.h>
  75
  76 #include "main.h"
  77
  78 #define APP_LOOKUP_EXACT_MATCH          0
  79 #define APP_LOOKUP_LPM                  1
  80 #define DO_RFC_1812_CHECKS
  81
  82 #ifndef APP_LOOKUP_METHOD
  83 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  84 #endif
  85
  86 #define ENABLE_MULTI_BUFFER_OPTIMIZE    1
  87
  88 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  89 #include <rte_hash.h>
  90 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
  91 #include <rte_lpm.h>
  92 #include <rte_lpm6.h>
  93 #else
  94 #error "APP_LOOKUP_METHOD set to incorrect value"
  95 #endif
  96
  97 #ifndef IPv6_BYTES
  98 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
  99                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 100 #define IPv6_BYTES(addr) \
 101         addr[0],  addr[1], addr[2],  addr[3], \
 102         addr[4],  addr[5], addr[6],  addr[7], \
 103         addr[8],  addr[9], addr[10], addr[11],\
 104         addr[12], addr[13],addr[14], addr[15]
 105 #endif
 106
 107
 108 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
 109
 110 #define MAX_JUMBO_PKT_LEN  9600
 111
 112 #define IPV6_ADDR_LEN 16
 113
 114 #define MEMPOOL_CACHE_SIZE 256
 115
 116 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
 117
 118 /*
 119  * This expression is used to calculate the number of mbufs needed depending on user input, taking
 120  *  into account memory for rx and tx hardware rings, cache per lcore and mtable per port per lcore.
 121  *  RTE_MAX is used to ensure that NB_MBUF never goes below a minimum value of 8192
 122  */
 123
 124 #define NB_MBUF RTE_MAX (                                                                                                                                       \
 125                                 (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT +                                                        \
 126                                 nb_ports*nb_lcores*MAX_PKT_BURST +                                                                                      \
 127                                 nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT +                                                          \
 128                                 nb_lcores*MEMPOOL_CACHE_SIZE),                                                                                          \
 129                                 (unsigned)8192)
 130
 131 /*
 132  * RX and TX Prefetch, Host, and Write-back threshold values should be
 133  * carefully set for optimal performance. Consult the network
 134  * controller's datasheet and supporting DPDK documentation for guidance
 135  * on how these parameters should be set.
 136  */
 137 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
 138 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
 139 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
 140
 141 /*
 142  * These default values are optimized for use with the Intel(R) 82599 10 GbE
 143  * Controller and the DPDK ixgbe PMD. Consider using other values for other
 144  * network controllers and/or network drivers.
 145  */
 146 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
 147 #define TX_HTHRESH 0  /**< Default values of TX host threshold reg. */
 148 #define TX_WTHRESH 0  /**< Default values of TX write-back threshold reg. */
 149
 150 #define MAX_PKT_BURST     32
 151 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 152
 153 #define NB_SOCKETS 8
 154
 155 /* Configure how many packets ahead to prefetch, when reading packets */
 156 #define PREFETCH_OFFSET 3
 157
 158 /*
 159  * Configurable number of RX/TX ring descriptors
 160  */
 161 #define RTE_TEST_RX_DESC_DEFAULT 128
 162 #define RTE_TEST_TX_DESC_DEFAULT 512
 163 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 164 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 165
 166 /* ethernet addresses of ports */
 167 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 168
 169 /* mask of enabled ports */
 170 static uint32_t enabled_port_mask = 0;
 171 static int promiscuous_on = 0; /**< Ports set in promiscuous mode off by default. */
 172 static int numa_on = 1; /**< NUMA is enabled by default. */
 173
 174 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 175 static int ipv6 = 0; /**< ipv6 is false by default. */
 176 #endif
 177
 178 struct mbuf_table {
 179         uint16_t len;
 180         struct rte_mbuf *m_table[MAX_PKT_BURST];
 181 };
 182
 183 struct lcore_rx_queue {
 184         uint8_t port_id;
 185         uint8_t queue_id;
 186 } __rte_cache_aligned;
 187
 188 #define MAX_RX_QUEUE_PER_LCORE 16
 189 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 190 #define MAX_RX_QUEUE_PER_PORT 128
 191
 192 #define MAX_LCORE_PARAMS 1024
 193 struct lcore_params {
 194         uint8_t port_id;
 195         uint8_t queue_id;
 196         uint8_t lcore_id;
 197 } __rte_cache_aligned;
 198
 199 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 200 static struct lcore_params lcore_params_array_default[] = {
 201         {0, 0, 2},
 202         {0, 1, 2},
 203         {0, 2, 2},
 204         {1, 0, 2},
 205         {1, 1, 2},
 206         {1, 2, 2},
 207         {2, 0, 2},
 208         {3, 0, 3},
 209         {3, 1, 3},
 210 };
 211
 212 static struct lcore_params * lcore_params = lcore_params_array_default;
 213 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 214                                 sizeof(lcore_params_array_default[0]);
 215
 216 static struct rte_eth_conf port_conf = {
 217         .rxmode = {
 218                 .max_rx_pkt_len = ETHER_MAX_LEN,
 219                 .split_hdr_size = 0,
 220                 .header_split   = 0, /**< Header Split disabled */
 221                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 222                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 223                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 224                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 225         },
 226         .rx_adv_conf = {
 227                 .rss_conf = {
 228                         .rss_key = NULL,
 229                         .rss_hf = ETH_RSS_IPV4 | ETH_RSS_IPV6,
 230                 },
 231         },
 232         .txmode = {
 233                 .mq_mode = ETH_MQ_TX_NONE,
 234         },
 235 };
 236
 237 static const struct rte_eth_rxconf rx_conf = {
 238         .rx_thresh = {
 239                 .pthresh = RX_PTHRESH,
 240                 .hthresh = RX_HTHRESH,
 241                 .wthresh = RX_WTHRESH,
 242         },
 243         .rx_free_thresh = 32,
 244 };
 245
 246 static struct rte_eth_txconf tx_conf = {
 247         .tx_thresh = {
 248                 .pthresh = TX_PTHRESH,
 249                 .hthresh = TX_HTHRESH,
 250                 .wthresh = TX_WTHRESH,
 251         },
 252         .tx_free_thresh = 0, /* Use PMD default values */
 253         .tx_rs_thresh = 0, /* Use PMD default values */
 254         .txq_flags = (ETH_TXQ_FLAGS_NOMULTSEGS |
 255                         ETH_TXQ_FLAGS_NOVLANOFFL |
 256                         ETH_TXQ_FLAGS_NOXSUMSCTP |
 257                         ETH_TXQ_FLAGS_NOXSUMUDP |
 258                         ETH_TXQ_FLAGS_NOXSUMTCP)
 259
 260 };
 261
 262 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 263
 264 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 265
 266 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 267 #include <rte_hash_crc.h>
 268 #define DEFAULT_HASH_FUNC       rte_hash_crc
 269 #else
 270 #include <rte_jhash.h>
 271 #define DEFAULT_HASH_FUNC       rte_jhash
 272 #endif
 273
 274 struct ipv4_5tuple {
 275         uint32_t ip_dst;
 276         uint32_t ip_src;
 277         uint16_t port_dst;
 278         uint16_t port_src;
 279         uint8_t  proto;
 280 } __attribute__((__packed__));
 281
 282 union ipv4_5tuple_host {
 283         struct {
 284                 uint8_t  pad0;
 285                 uint8_t  proto;
 286                 uint16_t pad1;
 287                 uint32_t ip_src;
 288                 uint32_t ip_dst;
 289                 uint16_t port_src;
 290                 uint16_t port_dst;
 291         };
 292         __m128i xmm;
 293 };
 294
 295 #define XMM_NUM_IN_IPV6_5TUPLE 3
 296
 297 struct ipv6_5tuple {
 298         uint8_t  ip_dst[IPV6_ADDR_LEN];
 299         uint8_t  ip_src[IPV6_ADDR_LEN];
 300         uint16_t port_dst;
 301         uint16_t port_src;
 302         uint8_t  proto;
 303 } __attribute__((__packed__));
 304
 305 union ipv6_5tuple_host {
 306         struct {
 307                 uint16_t pad0;
 308                 uint8_t  proto;
 309                 uint8_t  pad1;
 310                 uint8_t  ip_src[IPV6_ADDR_LEN];
 311                 uint8_t  ip_dst[IPV6_ADDR_LEN];
 312                 uint16_t port_src;
 313                 uint16_t port_dst;
 314                 uint64_t reserve;
 315         };
 316         __m128i xmm[XMM_NUM_IN_IPV6_5TUPLE];
 317 };
 318
 319 struct ipv4_l3fwd_route {
 320         struct ipv4_5tuple key;
 321         uint8_t if_out;
 322 };
 323
 324 struct ipv6_l3fwd_route {
 325         struct ipv6_5tuple key;
 326         uint8_t if_out;
 327 };
 328
 329 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 330         {{IPv4(101,0,0,0), IPv4(100,10,0,1),  101, 11, IPPROTO_TCP}, 0},
 331         {{IPv4(201,0,0,0), IPv4(200,20,0,1),  102, 12, IPPROTO_TCP}, 1},
 332         {{IPv4(111,0,0,0), IPv4(100,30,0,1),  101, 11, IPPROTO_TCP}, 2},
 333         {{IPv4(211,0,0,0), IPv4(200,40,0,1),  102, 12, IPPROTO_TCP}, 3},
 334 };
 335
 336 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 337         {{
 338         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 339         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 340         101, 11, IPPROTO_TCP}, 0},
 341
 342         {{
 343         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 344         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 345         102, 12, IPPROTO_TCP}, 1},
 346
 347         {{
 348         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 349         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 350         101, 11, IPPROTO_TCP}, 2},
 351
 352         {{
 353         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 354         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 355         102, 12, IPPROTO_TCP}, 3},
 356 };
 357
 358 typedef struct rte_hash lookup_struct_t;
 359 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 360 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 361
 362 #ifdef RTE_ARCH_X86_64
 363 /* default to 4 million hash entries (approx) */
 364 #define L3FWD_HASH_ENTRIES              1024*1024*4
 365 #else
 366 /* 32-bit has less address-space for hugepage memory, limit to 1M entries */
 367 #define L3FWD_HASH_ENTRIES              1024*1024*1
 368 #endif
 369 #define HASH_ENTRY_NUMBER_DEFAULT       4
 370
 371 static uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
 372
 373 static inline uint32_t
 374 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
 375         uint32_t init_val)
 376 {
 377         const union ipv4_5tuple_host *k;
 378         uint32_t t;
 379         const uint32_t *p;
 380
 381         k = data;
 382         t = k->proto;
 383         p = (const uint32_t *)&k->port_src;
 384
 385 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 386         init_val = rte_hash_crc_4byte(t, init_val);
 387         init_val = rte_hash_crc_4byte(k->ip_src, init_val);
 388         init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
 389         init_val = rte_hash_crc_4byte(*p, init_val);
 390 #else /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 391         init_val = rte_jhash_1word(t, init_val);
 392         init_val = rte_jhash_1word(k->ip_src, init_val);
 393         init_val = rte_jhash_1word(k->ip_dst, init_val);
 394         init_val = rte_jhash_1word(*p, init_val);
 395 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 396         return (init_val);
 397 }
 398
 399 static inline uint32_t
 400 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len, uint32_t init_val)
 401 {
 402         const union ipv6_5tuple_host *k;
 403         uint32_t t;
 404         const uint32_t *p;
 405 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 406         const uint32_t  *ip_src0, *ip_src1, *ip_src2, *ip_src3;
 407         const uint32_t  *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
 408 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 409
 410         k = data;
 411         t = k->proto;
 412         p = (const uint32_t *)&k->port_src;
 413
 414 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 415         ip_src0 = (const uint32_t *) k->ip_src;
 416         ip_src1 = (const uint32_t *)(k->ip_src+4);
 417         ip_src2 = (const uint32_t *)(k->ip_src+8);
 418         ip_src3 = (const uint32_t *)(k->ip_src+12);
 419         ip_dst0 = (const uint32_t *) k->ip_dst;
 420         ip_dst1 = (const uint32_t *)(k->ip_dst+4);
 421         ip_dst2 = (const uint32_t *)(k->ip_dst+8);
 422         ip_dst3 = (const uint32_t *)(k->ip_dst+12);
 423         init_val = rte_hash_crc_4byte(t, init_val);
 424         init_val = rte_hash_crc_4byte(*ip_src0, init_val);
 425         init_val = rte_hash_crc_4byte(*ip_src1, init_val);
 426         init_val = rte_hash_crc_4byte(*ip_src2, init_val);
 427         init_val = rte_hash_crc_4byte(*ip_src3, init_val);
 428         init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
 429         init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
 430         init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
 431         init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
 432         init_val = rte_hash_crc_4byte(*p, init_val);
 433 #else /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 434         init_val = rte_jhash_1word(t, init_val);
 435         init_val = rte_jhash(k->ip_src, sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
 436         init_val = rte_jhash(k->ip_dst, sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
 437         init_val = rte_jhash_1word(*p, init_val);
 438 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 439         return (init_val);
 440 }
 441
 442 #define IPV4_L3FWD_NUM_ROUTES \
 443         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 444
 445 #define IPV6_L3FWD_NUM_ROUTES \
 446         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 447
 448 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 449 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 450
 451 #endif
 452
 453 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 454 struct ipv4_l3fwd_route {
 455         uint32_t ip;
 456         uint8_t  depth;
 457         uint8_t  if_out;
 458 };
 459
 460 struct ipv6_l3fwd_route {
 461         uint8_t ip[16];
 462         uint8_t  depth;
 463         uint8_t  if_out;
 464 };
 465
 466 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 467         {IPv4(1,1,1,0), 24, 0},
 468         {IPv4(2,1,1,0), 24, 1},
 469         {IPv4(3,1,1,0), 24, 2},
 470         {IPv4(4,1,1,0), 24, 3},
 471         {IPv4(5,1,1,0), 24, 4},
 472         {IPv4(6,1,1,0), 24, 5},
 473         {IPv4(7,1,1,0), 24, 6},
 474         {IPv4(8,1,1,0), 24, 7},
 475 };
 476
 477 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 478         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
 479         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
 480         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
 481         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
 482         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
 483         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
 484         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
 485         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
 486 };
 487
 488 #define IPV4_L3FWD_NUM_ROUTES \
 489         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 490 #define IPV6_L3FWD_NUM_ROUTES \
 491         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 492
 493 #define IPV4_L3FWD_LPM_MAX_RULES         1024
 494 #define IPV6_L3FWD_LPM_MAX_RULES         1024
 495 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
 496
 497 typedef struct rte_lpm lookup_struct_t;
 498 typedef struct rte_lpm6 lookup6_struct_t;
 499 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 500 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 501 #endif
 502
 503 struct lcore_conf {
 504         uint16_t n_rx_queue;
 505         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 506         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 507         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 508         lookup_struct_t * ipv4_lookup_struct;
 509 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 510         lookup6_struct_t * ipv6_lookup_struct;
 511 #else
 512         lookup_struct_t * ipv6_lookup_struct;
 513 #endif
 514 } __rte_cache_aligned;
 515
 516 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 517
 518 /* Send burst of packets on an output interface */
 519 static inline int
 520 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 521 {
 522         struct rte_mbuf **m_table;
 523         int ret;
 524         uint16_t queueid;
 525
 526         queueid = qconf->tx_queue_id[port];
 527         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 528
 529         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 530         if (unlikely(ret < n)) {
 531                 do {
 532                         rte_pktmbuf_free(m_table[ret]);
 533                 } while (++ret < n);
 534         }
 535
 536         return 0;
 537 }
 538
 539 /* Enqueue a single packet, and send burst if queue is filled */
 540 static inline int
 541 send_single_packet(struct rte_mbuf *m, uint8_t port)
 542 {
 543         uint32_t lcore_id;
 544         uint16_t len;
 545         struct lcore_conf *qconf;
 546
 547         lcore_id = rte_lcore_id();
 548
 549         qconf = &lcore_conf[lcore_id];
 550         len = qconf->tx_mbufs[port].len;
 551         qconf->tx_mbufs[port].m_table[len] = m;
 552         len++;
 553
 554         /* enough pkts to be sent */
 555         if (unlikely(len == MAX_PKT_BURST)) {
 556                 send_burst(qconf, MAX_PKT_BURST, port);
 557                 len = 0;
 558         }
 559
 560         qconf->tx_mbufs[port].len = len;
 561         return 0;
 562 }
 563
 564 #ifdef DO_RFC_1812_CHECKS
 565 static inline int
 566 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 567 {
 568         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 569         /*
 570          * 1. The packet length reported by the Link Layer must be large
 571          * enough to hold the minimum length legal IP datagram (20 bytes).
 572          */
 573         if (link_len < sizeof(struct ipv4_hdr))
 574                 return -1;
 575
 576         /* 2. The IP checksum must be correct. */
 577         /* this is checked in H/W */
 578
 579         /*
 580          * 3. The IP version number must be 4. If the version number is not 4
 581          * then the packet may be another version of IP, such as IPng or
 582          * ST-II.
 583          */
 584         if (((pkt->version_ihl) >> 4) != 4)
 585                 return -3;
 586         /*
 587          * 4. The IP header length field must be large enough to hold the
 588          * minimum length legal IP datagram (20 bytes = 5 words).
 589          */
 590         if ((pkt->version_ihl & 0xf) < 5)
 591                 return -4;
 592
 593         /*
 594          * 5. The IP total length field must be large enough to hold the IP
 595          * datagram header, whose length is specified in the IP header length
 596          * field.
 597          */
 598         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 599                 return -5;
 600
 601         return 0;
 602 }
 603 #endif
 604
 605 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 606
 607 static __m128i mask0;
 608 static __m128i mask1;
 609 static __m128i mask2;
 610 static inline uint8_t
 611 get_ipv4_dst_port(void *ipv4_hdr, uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
 612 {
 613         int ret = 0;
 614         union ipv4_5tuple_host key;
 615
 616         ipv4_hdr = (uint8_t *)ipv4_hdr + offsetof(struct ipv4_hdr, time_to_live);
 617         __m128i data = _mm_loadu_si128((__m128i*)(ipv4_hdr));
 618         /* Get 5 tuple: dst port, src port, dst IP address, src IP address and protocol */
 619         key.xmm = _mm_and_si128(data, mask0);
 620         /* Find destination port */
 621         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 622         return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
 623 }
 624
 625 static inline uint8_t
 626 get_ipv6_dst_port(void *ipv6_hdr,  uint8_t portid, lookup_struct_t * ipv6_l3fwd_lookup_struct)
 627 {
 628         int ret = 0;
 629         union ipv6_5tuple_host key;
 630
 631         ipv6_hdr = (uint8_t *)ipv6_hdr + offsetof(struct ipv6_hdr, payload_len);
 632         __m128i data0 = _mm_loadu_si128((__m128i*)(ipv6_hdr));
 633         __m128i data1 = _mm_loadu_si128((__m128i*)(((uint8_t*)ipv6_hdr)+sizeof(__m128i)));
 634         __m128i data2 = _mm_loadu_si128((__m128i*)(((uint8_t*)ipv6_hdr)+sizeof(__m128i)+sizeof(__m128i)));
 635         /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
 636         key.xmm[0] = _mm_and_si128(data0, mask1);
 637         /* Get part of 5 tuple: dst IP address lower 96 bits and src IP address higher 32 bits */
 638         key.xmm[1] = data1;
 639         /* Get part of 5 tuple: dst port and src port and dst IP address higher 32 bits */
 640         key.xmm[2] = _mm_and_si128(data2, mask2);
 641
 642         /* Find destination port */
 643         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 644         return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
 645 }
 646 #endif
 647
 648 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 649 static inline uint8_t
 650 get_ipv4_dst_port(void *ipv4_hdr,  uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
 651 {
 652         uint8_t next_hop;
 653
 654         return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 655                         rte_be_to_cpu_32(((struct ipv4_hdr*)ipv4_hdr)->dst_addr), &next_hop) == 0)?
 656                         next_hop : portid);
 657 }
 658
 659 static inline uint8_t
 660 get_ipv6_dst_port(void *ipv6_hdr,  uint8_t portid, lookup6_struct_t * ipv6_l3fwd_lookup_struct)
 661 {
 662         uint8_t next_hop;
 663         return (uint8_t) ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
 664                         ((struct ipv6_hdr*)ipv6_hdr)->dst_addr, &next_hop) == 0)?
 665                         next_hop : portid);
 666 }
 667 #endif
 668
 669 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) & (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
 670 static inline void l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, struct lcore_conf *qconf);
 671
 672 #define MASK_ALL_PKTS    0xf
 673 #define EXECLUDE_1ST_PKT 0xe
 674 #define EXECLUDE_2ND_PKT 0xd
 675 #define EXECLUDE_3RD_PKT 0xb
 676 #define EXECLUDE_4TH_PKT 0x7
 677
 678 static inline void
 679 simple_ipv4_fwd_4pkts(struct rte_mbuf* m[4], uint8_t portid, struct lcore_conf *qconf)
 680 {
 681         struct ether_hdr *eth_hdr[4];
 682         struct ipv4_hdr *ipv4_hdr[4];
 683         void *d_addr_bytes[4];
 684         uint8_t dst_port[4];
 685         int32_t ret[4];
 686         union ipv4_5tuple_host key[4];
 687         __m128i data[4];
 688
 689         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
 690         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
 691         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
 692         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
 693
 694         /* Handle IPv4 headers.*/
 695         ipv4_hdr[0] = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m[0], unsigned char *) +
 696                         sizeof(struct ether_hdr));
 697         ipv4_hdr[1] = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m[1], unsigned char *) +
 698                         sizeof(struct ether_hdr));
 699         ipv4_hdr[2] = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m[2], unsigned char *) +
 700                         sizeof(struct ether_hdr));
 701         ipv4_hdr[3] = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m[3], unsigned char *) +
 702                         sizeof(struct ether_hdr));
 703
 704 #ifdef DO_RFC_1812_CHECKS
 705         /* Check to make sure the packet is valid (RFC1812) */
 706         uint8_t valid_mask = MASK_ALL_PKTS;
 707         if (is_valid_ipv4_pkt(ipv4_hdr[0], m[0]->pkt.pkt_len) < 0) {
 708                 rte_pktmbuf_free(m[0]);
 709                 valid_mask &= EXECLUDE_1ST_PKT;
 710         }
 711         if (is_valid_ipv4_pkt(ipv4_hdr[1], m[1]->pkt.pkt_len) < 0) {
 712                 rte_pktmbuf_free(m[1]);
 713                 valid_mask &= EXECLUDE_2ND_PKT;
 714         }
 715         if (is_valid_ipv4_pkt(ipv4_hdr[2], m[2]->pkt.pkt_len) < 0) {
 716                 rte_pktmbuf_free(m[2]);
 717                 valid_mask &= EXECLUDE_3RD_PKT;
 718         }
 719         if (is_valid_ipv4_pkt(ipv4_hdr[3], m[3]->pkt.pkt_len) < 0) {
 720                 rte_pktmbuf_free(m[3]);
 721                 valid_mask &= EXECLUDE_4TH_PKT;
 722         }
 723         if (unlikely(valid_mask != MASK_ALL_PKTS)) {
 724                 if (valid_mask == 0){
 725                         return;
 726                 } else {
 727                         uint8_t i = 0;
 728                         for (i = 0; i < 4; i++) {
 729                                 if ((0x1 << i) & valid_mask) {
 730                                         l3fwd_simple_forward(m[i], portid, qconf);
 731                                 }
 732                         }
 733                         return;
 734                 }
 735         }
 736 #endif // End of #ifdef DO_RFC_1812_CHECKS
 737
 738         data[0] = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m[0], unsigned char *) +
 739                 sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
 740         data[1] = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m[1], unsigned char *) +
 741                 sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
 742         data[2] = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m[2], unsigned char *) +
 743                 sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
 744         data[3] = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m[3], unsigned char *) +
 745                 sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
 746
 747         key[0].xmm = _mm_and_si128(data[0], mask0);
 748         key[1].xmm = _mm_and_si128(data[1], mask0);
 749         key[2].xmm = _mm_and_si128(data[2], mask0);
 750         key[3].xmm = _mm_and_si128(data[3], mask0);
 751
 752         const void *key_array[4] = {&key[0], &key[1], &key[2],&key[3]};
 753         rte_hash_lookup_multi(qconf->ipv4_lookup_struct, &key_array[0], 4, ret);
 754         dst_port[0] = (uint8_t) ((ret[0] < 0) ? portid : ipv4_l3fwd_out_if[ret[0]]);
 755         dst_port[1] = (uint8_t) ((ret[1] < 0) ? portid : ipv4_l3fwd_out_if[ret[1]]);
 756         dst_port[2] = (uint8_t) ((ret[2] < 0) ? portid : ipv4_l3fwd_out_if[ret[2]]);
 757         dst_port[3] = (uint8_t) ((ret[3] < 0) ? portid : ipv4_l3fwd_out_if[ret[3]]);
 758
 759         if (dst_port[0] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[0]) == 0)
 760                 dst_port[0] = portid;
 761         if (dst_port[1] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[1]) == 0)
 762                 dst_port[1] = portid;
 763         if (dst_port[2] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[2]) == 0)
 764                 dst_port[2] = portid;
 765         if (dst_port[3] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[3]) == 0)
 766                 dst_port[3] = portid;
 767
 768         /* 02:00:00:00:00:xx */
 769         d_addr_bytes[0] = &eth_hdr[0]->d_addr.addr_bytes[0];
 770         d_addr_bytes[1] = &eth_hdr[1]->d_addr.addr_bytes[0];
 771         d_addr_bytes[2] = &eth_hdr[2]->d_addr.addr_bytes[0];
 772         d_addr_bytes[3] = &eth_hdr[3]->d_addr.addr_bytes[0];
 773         *((uint64_t *)d_addr_bytes[0]) = 0x000000000002 + ((uint64_t)dst_port[0] << 40);
 774         *((uint64_t *)d_addr_bytes[1]) = 0x000000000002 + ((uint64_t)dst_port[1] << 40);
 775         *((uint64_t *)d_addr_bytes[2]) = 0x000000000002 + ((uint64_t)dst_port[2] << 40);
 776         *((uint64_t *)d_addr_bytes[3]) = 0x000000000002 + ((uint64_t)dst_port[3] << 40);
 777
 778 #ifdef DO_RFC_1812_CHECKS
 779         /* Update time to live and header checksum */
 780         --(ipv4_hdr[0]->time_to_live);
 781         --(ipv4_hdr[1]->time_to_live);
 782         --(ipv4_hdr[2]->time_to_live);
 783         --(ipv4_hdr[3]->time_to_live);
 784         ++(ipv4_hdr[0]->hdr_checksum);
 785         ++(ipv4_hdr[1]->hdr_checksum);
 786         ++(ipv4_hdr[2]->hdr_checksum);
 787         ++(ipv4_hdr[3]->hdr_checksum);
 788 #endif
 789
 790         /* src addr */
 791         ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
 792         ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
 793         ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
 794         ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
 795
 796         send_single_packet(m[0], (uint8_t)dst_port[0]);
 797         send_single_packet(m[1], (uint8_t)dst_port[1]);
 798         send_single_packet(m[2], (uint8_t)dst_port[2]);
 799         send_single_packet(m[3], (uint8_t)dst_port[3]);
 800
 801 }
 802
 803 static inline void get_ipv6_5tuple(struct rte_mbuf* m0, __m128i mask0, __m128i mask1,
 804                                  union ipv6_5tuple_host * key)
 805 {
 806         __m128i tmpdata0 = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m0, unsigned char *)
 807                         + sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len)));
 808         __m128i tmpdata1 = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m0, unsigned char *)
 809                         + sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len)
 810                         +  sizeof(__m128i)));
 811         __m128i tmpdata2 = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m0, unsigned char *)
 812                         + sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len)
 813                         + sizeof(__m128i) + sizeof(__m128i)));
 814         key->xmm[0] = _mm_and_si128(tmpdata0, mask0);
 815         key->xmm[1] = tmpdata1;
 816         key->xmm[2] = _mm_and_si128(tmpdata2, mask1);
 817         return;
 818 }
 819
 820 static inline void
 821 simple_ipv6_fwd_4pkts(struct rte_mbuf* m[4], uint8_t portid, struct lcore_conf *qconf)
 822 {
 823         struct ether_hdr *eth_hdr[4];
 824         __attribute__((unused)) struct ipv6_hdr *ipv6_hdr[4];
 825         void *d_addr_bytes[4];
 826         uint8_t dst_port[4];
 827         int32_t ret[4];
 828         union ipv6_5tuple_host key[4];
 829
 830         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
 831         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
 832         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
 833         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
 834
 835         /* Handle IPv6 headers.*/
 836         ipv6_hdr[0] = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m[0], unsigned char *) +
 837                         sizeof(struct ether_hdr));
 838         ipv6_hdr[1] = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m[1], unsigned char *) +
 839                         sizeof(struct ether_hdr));
 840         ipv6_hdr[2] = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m[2], unsigned char *) +
 841                         sizeof(struct ether_hdr));
 842         ipv6_hdr[3] = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m[3], unsigned char *) +
 843                         sizeof(struct ether_hdr));
 844
 845         get_ipv6_5tuple(m[0], mask1, mask2, &key[0]);
 846         get_ipv6_5tuple(m[1], mask1, mask2, &key[1]);
 847         get_ipv6_5tuple(m[2], mask1, mask2, &key[2]);
 848         get_ipv6_5tuple(m[3], mask1, mask2, &key[3]);
 849
 850         const void *key_array[4] = {&key[0], &key[1], &key[2],&key[3]};
 851         rte_hash_lookup_multi(qconf->ipv6_lookup_struct, &key_array[0], 4, ret);
 852         dst_port[0] = (uint8_t) ((ret[0] < 0)? portid:ipv6_l3fwd_out_if[ret[0]]);
 853         dst_port[1] = (uint8_t) ((ret[1] < 0)? portid:ipv6_l3fwd_out_if[ret[1]]);
 854         dst_port[2] = (uint8_t) ((ret[2] < 0)? portid:ipv6_l3fwd_out_if[ret[2]]);
 855         dst_port[3] = (uint8_t) ((ret[3] < 0)? portid:ipv6_l3fwd_out_if[ret[3]]);
 856
 857         if (dst_port[0] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[0]) == 0)
 858                 dst_port[0] = portid;
 859         if (dst_port[1] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[1]) == 0)
 860                 dst_port[1] = portid;
 861         if (dst_port[2] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[2]) == 0)
 862                 dst_port[2] = portid;
 863         if (dst_port[3] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[3]) == 0)
 864                 dst_port[3] = portid;
 865
 866         /* 02:00:00:00:00:xx */
 867         d_addr_bytes[0] = &eth_hdr[0]->d_addr.addr_bytes[0];
 868         d_addr_bytes[1] = &eth_hdr[1]->d_addr.addr_bytes[0];
 869         d_addr_bytes[2] = &eth_hdr[2]->d_addr.addr_bytes[0];
 870         d_addr_bytes[3] = &eth_hdr[3]->d_addr.addr_bytes[0];
 871         *((uint64_t *)d_addr_bytes[0]) = 0x000000000002 + ((uint64_t)dst_port[0] << 40);
 872         *((uint64_t *)d_addr_bytes[1]) = 0x000000000002 + ((uint64_t)dst_port[1] << 40);
 873         *((uint64_t *)d_addr_bytes[2]) = 0x000000000002 + ((uint64_t)dst_port[2] << 40);
 874         *((uint64_t *)d_addr_bytes[3]) = 0x000000000002 + ((uint64_t)dst_port[3] << 40);
 875
 876         /* src addr */
 877         ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
 878         ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
 879         ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
 880         ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
 881
 882         send_single_packet(m[0], (uint8_t)dst_port[0]);
 883         send_single_packet(m[1], (uint8_t)dst_port[1]);
 884         send_single_packet(m[2], (uint8_t)dst_port[2]);
 885         send_single_packet(m[3], (uint8_t)dst_port[3]);
 886
 887 }
 888 #endif // End of #if(APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)&(ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
 889
 890 static inline __attribute__((always_inline)) void
 891 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, struct lcore_conf *qconf)
 892 {
 893         struct ether_hdr *eth_hdr;
 894         struct ipv4_hdr *ipv4_hdr;
 895         void *d_addr_bytes;
 896         uint8_t dst_port;
 897
 898         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 899
 900         if (m->ol_flags & PKT_RX_IPV4_HDR) {
 901                 /* Handle IPv4 headers.*/
 902                 ipv4_hdr = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
 903                                 sizeof(struct ether_hdr));
 904
 905 #ifdef DO_RFC_1812_CHECKS
 906                 /* Check to make sure the packet is valid (RFC1812) */
 907                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt.pkt_len) < 0) {
 908                         rte_pktmbuf_free(m);
 909                         return;
 910                 }
 911 #endif
 912
 913                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid, qconf->ipv4_lookup_struct);
 914                 if (dst_port >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port) == 0)
 915                         dst_port = portid;
 916
 917                 /* 02:00:00:00:00:xx */
 918                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 919                 *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
 920
 921 #ifdef DO_RFC_1812_CHECKS
 922                 /* Update time to live and header checksum */
 923                 --(ipv4_hdr->time_to_live);
 924                 ++(ipv4_hdr->hdr_checksum);
 925 #endif
 926
 927                 /* src addr */
 928                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 929
 930                 send_single_packet(m, dst_port);
 931
 932         } else {
 933                 /* Handle IPv6 headers.*/
 934                 struct ipv6_hdr *ipv6_hdr;
 935
 936                 ipv6_hdr = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
 937                                 sizeof(struct ether_hdr));
 938
 939                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid, qconf->ipv6_lookup_struct);
 940
 941                 if (dst_port >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port) == 0)
 942                         dst_port = portid;
 943
 944                 /* 02:00:00:00:00:xx */
 945                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 946                 *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
 947
 948                 /* src addr */
 949                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 950
 951                 send_single_packet(m, dst_port);
 952         }
 953
 954 }
 955
 956 /* main processing loop */
 957 static int
 958 main_loop(__attribute__((unused)) void *dummy)
 959 {
 960         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 961         unsigned lcore_id;
 962         uint64_t prev_tsc, diff_tsc, cur_tsc;
 963         int i, j, nb_rx;
 964         uint8_t portid, queueid;
 965         struct lcore_conf *qconf;
 966         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 967
 968         prev_tsc = 0;
 969
 970         lcore_id = rte_lcore_id();
 971         qconf = &lcore_conf[lcore_id];
 972
 973         if (qconf->n_rx_queue == 0) {
 974                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
 975                 return 0;
 976         }
 977
 978         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
 979
 980         for (i = 0; i < qconf->n_rx_queue; i++) {
 981
 982                 portid = qconf->rx_queue_list[i].port_id;
 983                 queueid = qconf->rx_queue_list[i].queue_id;
 984                 RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n", lcore_id,
 985                         portid, queueid);
 986         }
 987
 988         while (1) {
 989
 990                 cur_tsc = rte_rdtsc();
 991
 992                 /*
 993                  * TX burst queue drain
 994                  */
 995                 diff_tsc = cur_tsc - prev_tsc;
 996                 if (unlikely(diff_tsc > drain_tsc)) {
 997
 998                         /*
 999                          * This could be optimized (use queueid instead of
1000                          * portid), but it is not called so often
1001                          */
1002                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
1003                                 if (qconf->tx_mbufs[portid].len == 0)
1004                                         continue;
1005                                 send_burst(&lcore_conf[lcore_id],
1006                                         qconf->tx_mbufs[portid].len,
1007                                         portid);
1008                                 qconf->tx_mbufs[portid].len = 0;
1009                         }
1010
1011                         prev_tsc = cur_tsc;
1012                 }
1013
1014                 /*
1015                  * Read packet from RX queues
1016                  */
1017                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1018                         portid = qconf->rx_queue_list[i].port_id;
1019                         queueid = qconf->rx_queue_list[i].queue_id;
1020                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst, MAX_PKT_BURST);
1021 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) & (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
1022                         {
1023                                 /* Send nb_rx - nb_rx%4 packets in groups of 4.*/
1024                                 int32_t n = RTE_ALIGN_FLOOR(nb_rx, 4);
1025                                 for (j = 0; j < n ; j+=4) {
1026                                         uint32_t ol_flag = pkts_burst[j]->ol_flags
1027                                                         & pkts_burst[j+1]->ol_flags
1028                                                         & pkts_burst[j+2]->ol_flags
1029                                                         & pkts_burst[j+3]->ol_flags;
1030                                         if (ol_flag & PKT_RX_IPV4_HDR ) {
1031                                                 simple_ipv4_fwd_4pkts(&pkts_burst[j],
1032                                                                         portid, qconf);
1033                                         } else if (ol_flag & PKT_RX_IPV6_HDR) {
1034                                                 simple_ipv6_fwd_4pkts(&pkts_burst[j],
1035                                                                         portid, qconf);
1036                                         } else {
1037                                                 l3fwd_simple_forward(pkts_burst[j],
1038                                                                         portid, qconf);
1039                                                 l3fwd_simple_forward(pkts_burst[j+1],
1040                                                                         portid, qconf);
1041                                                 l3fwd_simple_forward(pkts_burst[j+2],
1042                                                                         portid, qconf);
1043                                                 l3fwd_simple_forward(pkts_burst[j+3],
1044                                                                         portid, qconf);
1045                                         }
1046                                 }
1047                                 for (; j < nb_rx ; j++) {
1048                                         l3fwd_simple_forward(pkts_burst[j],
1049                                                                 portid, qconf);
1050                                 }
1051                         }
1052 #else
1053                         /* Prefetch first packets */
1054                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1055                                 rte_prefetch0(rte_pktmbuf_mtod(
1056                                                 pkts_burst[j], void *));
1057                         }
1058
1059                         /* Prefetch and forward already prefetched packets */
1060                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1061                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1062                                                 j + PREFETCH_OFFSET], void *));
1063                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf);
1064                         }
1065
1066                         /* Forward remaining prefetched packets */
1067                         for (; j < nb_rx; j++) {
1068                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf);
1069                         }
1070 #endif // End of #if((ENABLE_MULTI_BUFFER_OPTIMIZE == 1)&(APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH))
1071                 }
1072         }
1073 }
1074
1075 static int
1076 check_lcore_params(void)
1077 {
1078         uint8_t queue, lcore;
1079         uint16_t i;
1080         int socketid;
1081
1082         for (i = 0; i < nb_lcore_params; ++i) {
1083                 queue = lcore_params[i].queue_id;
1084                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1085                         printf("invalid queue number: %hhu\n", queue);
1086                         return -1;
1087                 }
1088                 lcore = lcore_params[i].lcore_id;
1089                 if (!rte_lcore_is_enabled(lcore)) {
1090                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
1091                         return -1;
1092                 }
1093                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1094                         (numa_on == 0)) {
1095                         printf("warning: lcore %hhu is on socket %d with numa off \n",
1096                                 lcore, socketid);
1097                 }
1098         }
1099         return 0;
1100 }
1101
1102 static int
1103 check_port_config(const unsigned nb_ports)
1104 {
1105         unsigned portid;
1106         uint16_t i;
1107
1108         for (i = 0; i < nb_lcore_params; ++i) {
1109                 portid = lcore_params[i].port_id;
1110                 if ((enabled_port_mask & (1 << portid)) == 0) {
1111                         printf("port %u is not enabled in port mask\n", portid);
1112                         return -1;
1113                 }
1114                 if (portid >= nb_ports) {
1115                         printf("port %u is not present on the board\n", portid);
1116                         return -1;
1117                 }
1118         }
1119         return 0;
1120 }
1121
1122 static uint8_t
1123 get_port_n_rx_queues(const uint8_t port)
1124 {
1125         int queue = -1;
1126         uint16_t i;
1127
1128         for (i = 0; i < nb_lcore_params; ++i) {
1129                 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
1130                         queue = lcore_params[i].queue_id;
1131         }
1132         return (uint8_t)(++queue);
1133 }
1134
1135 static int
1136 init_lcore_rx_queues(void)
1137 {
1138         uint16_t i, nb_rx_queue;
1139         uint8_t lcore;
1140
1141         for (i = 0; i < nb_lcore_params; ++i) {
1142                 lcore = lcore_params[i].lcore_id;
1143                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1144                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1145                         printf("error: too many queues (%u) for lcore: %u\n",
1146                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1147                         return -1;
1148                 } else {
1149                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1150                                 lcore_params[i].port_id;
1151                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1152                                 lcore_params[i].queue_id;
1153                         lcore_conf[lcore].n_rx_queue++;
1154                 }
1155         }
1156         return 0;
1157 }
1158
1159 /* display usage */
1160 static void
1161 print_usage(const char *prgname)
1162 {
1163         printf ("%s [EAL options] -- -p PORTMASK -P"
1164                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1165                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1166                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1167                 "  -P : enable promiscuous mode\n"
1168                 "  --config (port,queue,lcore): rx queues configuration\n"
1169                 "  --no-numa: optional, disable numa awareness\n"
1170                 "  --ipv6: optional, specify it if running ipv6 packets\n"
1171                 "  --enable-jumbo: enable jumbo frame"
1172                 " which max packet len is PKTLEN in decimal (64-9600)\n"
1173                 "  --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n",
1174                 prgname);
1175 }
1176
1177 static int parse_max_pkt_len(const char *pktlen)
1178 {
1179         char *end = NULL;
1180         unsigned long len;
1181
1182         /* parse decimal string */
1183         len = strtoul(pktlen, &end, 10);
1184         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1185                 return -1;
1186
1187         if (len == 0)
1188                 return -1;
1189
1190         return len;
1191 }
1192
1193 static int
1194 parse_portmask(const char *portmask)
1195 {
1196         char *end = NULL;
1197         unsigned long pm;
1198
1199         /* parse hexadecimal string */
1200         pm = strtoul(portmask, &end, 16);
1201         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1202                 return -1;
1203
1204         if (pm == 0)
1205                 return -1;
1206
1207         return pm;
1208 }
1209
1210 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1211 static int
1212 parse_hash_entry_number(const char *hash_entry_num)
1213 {
1214         char *end = NULL;
1215         unsigned long hash_en;
1216         /* parse hexadecimal string */
1217         hash_en = strtoul(hash_entry_num, &end, 16);
1218         if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
1219                 return -1;
1220
1221         if (hash_en == 0)
1222                 return -1;
1223
1224         return hash_en;
1225 }
1226 #endif
1227
1228 static int
1229 parse_config(const char *q_arg)
1230 {
1231         char s[256];
1232         const char *p, *p0 = q_arg;
1233         char *end;
1234         enum fieldnames {
1235                 FLD_PORT = 0,
1236                 FLD_QUEUE,
1237                 FLD_LCORE,
1238                 _NUM_FLD
1239         };
1240         unsigned long int_fld[_NUM_FLD];
1241         char *str_fld[_NUM_FLD];
1242         int i;
1243         unsigned size;
1244
1245         nb_lcore_params = 0;
1246
1247         while ((p = strchr(p0,'(')) != NULL) {
1248                 ++p;
1249                 if((p0 = strchr(p,')')) == NULL)
1250                         return -1;
1251
1252                 size = p0 - p;
1253                 if(size >= sizeof(s))
1254                         return -1;
1255
1256                 rte_snprintf(s, sizeof(s), "%.*s", size, p);
1257                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
1258                         return -1;
1259                 for (i = 0; i < _NUM_FLD; i++){
1260                         errno = 0;
1261                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1262                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1263                                 return -1;
1264                 }
1265                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1266                         printf("exceeded max number of lcore params: %hu\n",
1267                                 nb_lcore_params);
1268                         return -1;
1269                 }
1270                 lcore_params_array[nb_lcore_params].port_id = (uint8_t)int_fld[FLD_PORT];
1271                 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
1272                 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
1273                 ++nb_lcore_params;
1274         }
1275         lcore_params = lcore_params_array;
1276         return 0;
1277 }
1278
1279 #define CMD_LINE_OPT_CONFIG "config"
1280 #define CMD_LINE_OPT_NO_NUMA "no-numa"
1281 #define CMD_LINE_OPT_IPV6 "ipv6"
1282 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
1283 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
1284
1285 /* Parse the argument given in the command line of the application */
1286 static int
1287 parse_args(int argc, char **argv)
1288 {
1289         int opt, ret;
1290         char **argvopt;
1291         int option_index;
1292         char *prgname = argv[0];
1293         static struct option lgopts[] = {
1294                 {CMD_LINE_OPT_CONFIG, 1, 0, 0},
1295                 {CMD_LINE_OPT_NO_NUMA, 0, 0, 0},
1296                 {CMD_LINE_OPT_IPV6, 0, 0, 0},
1297                 {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, 0},
1298                 {CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, 0},
1299                 {NULL, 0, 0, 0}
1300         };
1301
1302         argvopt = argv;
1303
1304         while ((opt = getopt_long(argc, argvopt, "p:P",
1305                                 lgopts, &option_index)) != EOF) {
1306
1307                 switch (opt) {
1308                 /* portmask */
1309                 case 'p':
1310                         enabled_port_mask = parse_portmask(optarg);
1311                         if (enabled_port_mask == 0) {
1312                                 printf("invalid portmask\n");
1313                                 print_usage(prgname);
1314                                 return -1;
1315                         }
1316                         break;
1317                 case 'P':
1318                         printf("Promiscuous mode selected\n");
1319                         promiscuous_on = 1;
1320                         break;
1321
1322                 /* long options */
1323                 case 0:
1324                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_CONFIG,
1325                                 sizeof (CMD_LINE_OPT_CONFIG))) {
1326                                 ret = parse_config(optarg);
1327                                 if (ret) {
1328                                         printf("invalid config\n");
1329                                         print_usage(prgname);
1330                                         return -1;
1331                                 }
1332                         }
1333
1334                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_NUMA,
1335                                 sizeof(CMD_LINE_OPT_NO_NUMA))) {
1336                                 printf("numa is disabled \n");
1337                                 numa_on = 0;
1338                         }
1339
1340 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1341                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_IPV6,
1342                                 sizeof(CMD_LINE_OPT_IPV6))) {
1343                                 printf("ipv6 is specified \n");
1344                                 ipv6 = 1;
1345                         }
1346 #endif
1347
1348                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ENABLE_JUMBO,
1349                                 sizeof (CMD_LINE_OPT_ENABLE_JUMBO))) {
1350                                 struct option lenopts = {"max-pkt-len", required_argument, 0, 0};
1351
1352                                 printf("jumbo frame is enabled - disabling simple TX path\n");
1353                                 port_conf.rxmode.jumbo_frame = 1;
1354                                 tx_conf.txq_flags = 0;
1355
1356                                 /* if no max-pkt-len set, use the default value ETHER_MAX_LEN */
1357                                 if (0 == getopt_long(argc, argvopt, "", &lenopts, &option_index)) {
1358                                         ret = parse_max_pkt_len(optarg);
1359                                         if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)){
1360                                                 printf("invalid packet length\n");
1361                                                 print_usage(prgname);
1362                                                 return -1;
1363                                         }
1364                                         port_conf.rxmode.max_rx_pkt_len = ret;
1365                                 }
1366                                 printf("set jumbo frame max packet length to %u\n",
1367                                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1368                         }
1369 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1370                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_HASH_ENTRY_NUM,
1371                                 sizeof(CMD_LINE_OPT_HASH_ENTRY_NUM))) {
1372                                 ret = parse_hash_entry_number(optarg);
1373                                 if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
1374                                         hash_entry_number = ret;
1375                                 } else {
1376                                         printf("invalid hash entry number\n");
1377                                         print_usage(prgname);
1378                                         return -1;
1379                                 }
1380                         }
1381 #endif
1382                         break;
1383
1384                 default:
1385                         print_usage(prgname);
1386                         return -1;
1387                 }
1388         }
1389
1390         if (optind >= 0)
1391                 argv[optind-1] = prgname;
1392
1393         ret = optind-1;
1394         optind = 0; /* reset getopt lib */
1395         return ret;
1396 }
1397
1398 static void
1399 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1400 {
1401         printf ("%s%02X:%02X:%02X:%02X:%02X:%02X", name,
1402                 eth_addr->addr_bytes[0],
1403                 eth_addr->addr_bytes[1],
1404                 eth_addr->addr_bytes[2],
1405                 eth_addr->addr_bytes[3],
1406                 eth_addr->addr_bytes[4],
1407                 eth_addr->addr_bytes[5]);
1408 }
1409
1410 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1411
1412 static void convert_ipv4_5tuple(struct ipv4_5tuple* key1,
1413                 union ipv4_5tuple_host* key2)
1414 {
1415         key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
1416         key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
1417         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
1418         key2->port_src = rte_cpu_to_be_16(key1->port_src);
1419         key2->proto = key1->proto;
1420         key2->pad0 = 0;
1421         key2->pad1 = 0;
1422         return;
1423 }
1424
1425 static void convert_ipv6_5tuple(struct ipv6_5tuple* key1,
1426                 union ipv6_5tuple_host* key2)
1427 {
1428         uint32_t i;
1429         for (i = 0; i < 16; i++)
1430         {
1431                 key2->ip_dst[i] = key1->ip_dst[i];
1432                 key2->ip_src[i] = key1->ip_src[i];
1433         }
1434         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
1435         key2->port_src = rte_cpu_to_be_16(key1->port_src);
1436         key2->proto = key1->proto;
1437         key2->pad0 = 0;
1438         key2->pad1 = 0;
1439         key2->reserve = 0;
1440         return;
1441 }
1442
1443 #define BYTE_VALUE_MAX 256
1444 #define ALL_32_BITS 0xffffffff
1445 #define BIT_8_TO_15 0x0000ff00
1446 static inline void
1447 populate_ipv4_few_flow_into_table(const struct rte_hash* h)
1448 {
1449         uint32_t i;
1450         int32_t ret;
1451         uint32_t array_len = sizeof(ipv4_l3fwd_route_array)/sizeof(ipv4_l3fwd_route_array[0]);
1452
1453         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
1454         for (i = 0; i < array_len; i++) {
1455                 struct ipv4_l3fwd_route  entry;
1456                 union ipv4_5tuple_host newkey;
1457                 entry = ipv4_l3fwd_route_array[i];
1458                 convert_ipv4_5tuple(&entry.key, &newkey);
1459                 ret = rte_hash_add_key (h,(void *) &newkey);
1460                 if (ret < 0) {
1461                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1462                                 "l3fwd hash.\n", i);
1463                 }
1464                 ipv4_l3fwd_out_if[ret] = entry.if_out;
1465         }
1466         printf("Hash: Adding 0x%x keys\n", array_len);
1467 }
1468
1469 #define BIT_16_TO_23 0x00ff0000
1470 static inline void
1471 populate_ipv6_few_flow_into_table(const struct rte_hash* h)
1472 {
1473         uint32_t i;
1474         int32_t ret;
1475         uint32_t array_len = sizeof(ipv6_l3fwd_route_array)/sizeof(ipv6_l3fwd_route_array[0]);
1476
1477         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
1478         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
1479         for (i = 0; i < array_len; i++) {
1480                 struct ipv6_l3fwd_route entry;
1481                 union ipv6_5tuple_host newkey;
1482                 entry = ipv6_l3fwd_route_array[i];
1483                 convert_ipv6_5tuple(&entry.key, &newkey);
1484                 ret = rte_hash_add_key (h, (void *) &newkey);
1485                 if (ret < 0) {
1486                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1487                                 "l3fwd hash.\n", i);
1488                 }
1489                 ipv6_l3fwd_out_if[ret] = entry.if_out;
1490         }
1491         printf("Hash: Adding 0x%xkeys\n", array_len);
1492 }
1493
1494 #define NUMBER_PORT_USED 4
1495 static inline void
1496 populate_ipv4_many_flow_into_table(const struct rte_hash* h,
1497                 unsigned int nr_flow)
1498 {
1499         unsigned i;
1500         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
1501         for (i = 0; i < nr_flow; i++) {
1502                 struct ipv4_l3fwd_route entry;
1503                 union ipv4_5tuple_host newkey;
1504                 uint8_t a = (uint8_t) ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
1505                 uint8_t b = (uint8_t) (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
1506                 uint8_t c = (uint8_t) ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
1507                 /* Create the ipv4 exact match flow */
1508                 switch (i & (NUMBER_PORT_USED -1)) {
1509                 case 0:
1510                         entry = ipv4_l3fwd_route_array[0];
1511                         entry.key.ip_dst = IPv4(101,c,b,a);
1512                         break;
1513                 case 1:
1514                         entry = ipv4_l3fwd_route_array[1];
1515                         entry.key.ip_dst = IPv4(201,c,b,a);
1516                         break;
1517                 case 2:
1518                         entry = ipv4_l3fwd_route_array[2];
1519                         entry.key.ip_dst = IPv4(111,c,b,a);
1520                         break;
1521                 case 3:
1522                         entry = ipv4_l3fwd_route_array[3];
1523                         entry.key.ip_dst = IPv4(211,c,b,a);
1524                         break;
1525                 };
1526                 convert_ipv4_5tuple(&entry.key, &newkey);
1527                 int32_t ret = rte_hash_add_key(h,(void *) &newkey);
1528                 if (ret < 0) {
1529                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
1530                 }
1531                 ipv4_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
1532
1533         }
1534         printf("Hash: Adding 0x%x keys\n", nr_flow);
1535 }
1536
1537 static inline void
1538 populate_ipv6_many_flow_into_table(const struct rte_hash* h,
1539                 unsigned int nr_flow)
1540 {
1541         unsigned i;
1542         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
1543         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
1544         for (i = 0; i < nr_flow; i++) {
1545                 struct ipv6_l3fwd_route entry;
1546                 union ipv6_5tuple_host newkey;
1547                 uint8_t a = (uint8_t) ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
1548                 uint8_t b = (uint8_t) (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
1549                 uint8_t c = (uint8_t) ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
1550                 /* Create the ipv6 exact match flow */
1551                 switch (i & (NUMBER_PORT_USED - 1)) {
1552                 case 0: entry = ipv6_l3fwd_route_array[0]; break;
1553                 case 1: entry = ipv6_l3fwd_route_array[1]; break;
1554                 case 2: entry = ipv6_l3fwd_route_array[2]; break;
1555                 case 3: entry = ipv6_l3fwd_route_array[3]; break;
1556                 };
1557                 entry.key.ip_dst[13] = c;
1558                 entry.key.ip_dst[14] = b;
1559                 entry.key.ip_dst[15] = a;
1560                 convert_ipv6_5tuple(&entry.key, &newkey);
1561                 int32_t ret = rte_hash_add_key(h,(void *) &newkey);
1562                 if (ret < 0) {
1563                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
1564                 }
1565                 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
1566
1567         }
1568         printf("Hash: Adding 0x%x keys\n", nr_flow);
1569 }
1570
1571 static void
1572 setup_hash(int socketid)
1573 {
1574     struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1575         .name = NULL,
1576         .entries = L3FWD_HASH_ENTRIES,
1577         .bucket_entries = 4,
1578         .key_len = sizeof(union ipv4_5tuple_host),
1579         .hash_func = ipv4_hash_crc,
1580         .hash_func_init_val = 0,
1581     };
1582
1583     struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1584         .name = NULL,
1585         .entries = L3FWD_HASH_ENTRIES,
1586         .bucket_entries = 4,
1587         .key_len = sizeof(union ipv6_5tuple_host),
1588         .hash_func = ipv6_hash_crc,
1589         .hash_func_init_val = 0,
1590     };
1591
1592     char s[64];
1593
1594         /* create ipv4 hash */
1595         rte_snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1596         ipv4_l3fwd_hash_params.name = s;
1597         ipv4_l3fwd_hash_params.socket_id = socketid;
1598         ipv4_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv4_l3fwd_hash_params);
1599         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1600                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1601                                 "socket %d\n", socketid);
1602
1603         /* create ipv6 hash */
1604         rte_snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1605         ipv6_l3fwd_hash_params.name = s;
1606         ipv6_l3fwd_hash_params.socket_id = socketid;
1607         ipv6_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv6_l3fwd_hash_params);
1608         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1609                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1610                                 "socket %d\n", socketid);
1611
1612         if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
1613                 /* For testing hash matching with a large number of flows we
1614                  * generate millions of IP 5-tuples with an incremented dst
1615                  * address to initialize the hash table. */
1616                 if (ipv6 == 0) {
1617                         /* populate the ipv4 hash */
1618                         populate_ipv4_many_flow_into_table(
1619                                 ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
1620                 } else {
1621                         /* populate the ipv6 hash */
1622                         populate_ipv6_many_flow_into_table(
1623                                 ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
1624                 }
1625         } else {
1626                 /* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize the hash table */
1627                 if (ipv6 == 0) {
1628                         /* populate the ipv4 hash */
1629                         populate_ipv4_few_flow_into_table(ipv4_l3fwd_lookup_struct[socketid]);
1630                 } else {
1631                         /* populate the ipv6 hash */
1632                         populate_ipv6_few_flow_into_table(ipv6_l3fwd_lookup_struct[socketid]);
1633                 }
1634         }
1635 }
1636 #endif
1637
1638 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1639 static void
1640 setup_lpm(int socketid)
1641 {
1642         struct rte_lpm6_config config;
1643         unsigned i;
1644         int ret;
1645         char s[64];
1646
1647         /* create the LPM table */
1648         rte_snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1649         ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1650                                 IPV4_L3FWD_LPM_MAX_RULES, 0);
1651         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1652                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1653                                 " on socket %d\n", socketid);
1654
1655         /* populate the LPM table */
1656         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1657                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1658                         ipv4_l3fwd_route_array[i].ip,
1659                         ipv4_l3fwd_route_array[i].depth,
1660                         ipv4_l3fwd_route_array[i].if_out);
1661
1662                 if (ret < 0) {
1663                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1664                                 "l3fwd LPM table on socket %d\n",
1665                                 i, socketid);
1666                 }
1667
1668                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1669                         (unsigned)ipv4_l3fwd_route_array[i].ip,
1670                         ipv4_l3fwd_route_array[i].depth,
1671                         ipv4_l3fwd_route_array[i].if_out);
1672         }
1673
1674         /* create the LPM6 table */
1675         rte_snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
1676
1677         config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
1678         config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
1679         config.flags = 0;
1680         ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
1681                                 &config);
1682         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1683                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1684                                 " on socket %d\n", socketid);
1685
1686         /* populate the LPM table */
1687         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1688                 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
1689                         ipv6_l3fwd_route_array[i].ip,
1690                         ipv6_l3fwd_route_array[i].depth,
1691                         ipv6_l3fwd_route_array[i].if_out);
1692
1693                 if (ret < 0) {
1694                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1695                                 "l3fwd LPM table on socket %d\n",
1696                                 i, socketid);
1697                 }
1698
1699                 printf("LPM: Adding route %s / %d (%d)\n",
1700                         "IPV6",
1701                         ipv6_l3fwd_route_array[i].depth,
1702                         ipv6_l3fwd_route_array[i].if_out);
1703         }
1704 }
1705 #endif
1706
1707 static int
1708 init_mem(unsigned nb_mbuf)
1709 {
1710         struct lcore_conf *qconf;
1711         int socketid;
1712         unsigned lcore_id;
1713         char s[64];
1714
1715         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1716                 if (rte_lcore_is_enabled(lcore_id) == 0)
1717                         continue;
1718
1719                 if (numa_on)
1720                         socketid = rte_lcore_to_socket_id(lcore_id);
1721                 else
1722                         socketid = 0;
1723
1724                 if (socketid >= NB_SOCKETS) {
1725                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
1726                                 socketid, lcore_id, NB_SOCKETS);
1727                 }
1728                 if (pktmbuf_pool[socketid] == NULL) {
1729                         rte_snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1730                         pktmbuf_pool[socketid] =
1731                                 rte_mempool_create(s, nb_mbuf, MBUF_SIZE, MEMPOOL_CACHE_SIZE,
1732                                         sizeof(struct rte_pktmbuf_pool_private),
1733                                         rte_pktmbuf_pool_init, NULL,
1734                                         rte_pktmbuf_init, NULL,
1735                                         socketid, 0);
1736                         if (pktmbuf_pool[socketid] == NULL)
1737                                 rte_exit(EXIT_FAILURE,
1738                                                 "Cannot init mbuf pool on socket %d\n", socketid);
1739                         else
1740                                 printf("Allocated mbuf pool on socket %d\n", socketid);
1741
1742 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1743                         setup_lpm(socketid);
1744 #else
1745                         setup_hash(socketid);
1746 #endif
1747                 }
1748                 qconf = &lcore_conf[lcore_id];
1749                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1750                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1751         }
1752         return 0;
1753 }
1754
1755 /* Check the link status of all ports in up to 9s, and print them finally */
1756 static void
1757 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1758 {
1759 #define CHECK_INTERVAL 100 /* 100ms */
1760 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1761         uint8_t portid, count, all_ports_up, print_flag = 0;
1762         struct rte_eth_link link;
1763
1764         printf("\nChecking link status");
1765         fflush(stdout);
1766         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1767                 all_ports_up = 1;
1768                 for (portid = 0; portid < port_num; portid++) {
1769                         if ((port_mask & (1 << portid)) == 0)
1770                                 continue;
1771                         memset(&link, 0, sizeof(link));
1772                         rte_eth_link_get_nowait(portid, &link);
1773                         /* print link status if flag set */
1774                         if (print_flag == 1) {
1775                                 if (link.link_status)
1776                                         printf("Port %d Link Up - speed %u "
1777                                                 "Mbps - %s\n", (uint8_t)portid,
1778                                                 (unsigned)link.link_speed,
1779                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1780                                         ("full-duplex") : ("half-duplex\n"));
1781                                 else
1782                                         printf("Port %d Link Down\n",
1783                                                 (uint8_t)portid);
1784                                 continue;
1785                         }
1786                         /* clear all_ports_up flag if any link down */
1787                         if (link.link_status == 0) {
1788                                 all_ports_up = 0;
1789                                 break;
1790                         }
1791                 }
1792                 /* after finally printing all link status, get out */
1793                 if (print_flag == 1)
1794                         break;
1795
1796                 if (all_ports_up == 0) {
1797                         printf(".");
1798                         fflush(stdout);
1799                         rte_delay_ms(CHECK_INTERVAL);
1800                 }
1801
1802                 /* set the print_flag if all ports up or timeout */
1803                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1804                         print_flag = 1;
1805                         printf("done\n");
1806                 }
1807         }
1808 }
1809
1810 int
1811 MAIN(int argc, char **argv)
1812 {
1813         struct lcore_conf *qconf;
1814         int ret;
1815         unsigned nb_ports;
1816         uint16_t queueid;
1817         unsigned lcore_id;
1818         uint32_t n_tx_queue, nb_lcores;
1819         uint8_t portid, nb_rx_queue, queue, socketid;
1820
1821         /* init EAL */
1822         ret = rte_eal_init(argc, argv);
1823         if (ret < 0)
1824                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1825         argc -= ret;
1826         argv += ret;
1827
1828         /* parse application arguments (after the EAL ones) */
1829         ret = parse_args(argc, argv);
1830         if (ret < 0)
1831                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1832
1833         if (check_lcore_params() < 0)
1834                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1835
1836         ret = init_lcore_rx_queues();
1837         if (ret < 0)
1838                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1839
1840
1841         /* init driver(s) */
1842         if (rte_pmd_init_all() < 0)
1843                 rte_exit(EXIT_FAILURE, "Cannot init pmd\n");
1844
1845         if (rte_eal_pci_probe() < 0)
1846                 rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");
1847
1848         nb_ports = rte_eth_dev_count();
1849         if (nb_ports > RTE_MAX_ETHPORTS)
1850                 nb_ports = RTE_MAX_ETHPORTS;
1851
1852         if (check_port_config(nb_ports) < 0)
1853                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1854
1855         nb_lcores = rte_lcore_count();
1856
1857         /* initialize all ports */
1858         for (portid = 0; portid < nb_ports; portid++) {
1859                 /* skip ports that are not enabled */
1860                 if ((enabled_port_mask & (1 << portid)) == 0) {
1861                         printf("\nSkipping disabled port %d\n", portid);
1862                         continue;
1863                 }
1864
1865                 /* init port */
1866                 printf("Initializing port %d ... ", portid );
1867                 fflush(stdout);
1868
1869                 nb_rx_queue = get_port_n_rx_queues(portid);
1870                 n_tx_queue = nb_lcores;
1871                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1872                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1873                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1874                         nb_rx_queue, (unsigned)n_tx_queue );
1875                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1876                                         (uint16_t)n_tx_queue, &port_conf);
1877                 if (ret < 0)
1878                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
1879                                 ret, portid);
1880
1881                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1882                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1883                 printf(", ");
1884
1885                 /* init memory */
1886                 ret = init_mem(NB_MBUF);
1887                 if (ret < 0)
1888                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1889
1890                 /* init one TX queue per couple (lcore,port) */
1891                 queueid = 0;
1892                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1893                         if (rte_lcore_is_enabled(lcore_id) == 0)
1894                                 continue;
1895
1896                         if (numa_on)
1897                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1898                         else
1899                                 socketid = 0;
1900
1901                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1902                         fflush(stdout);
1903                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1904                                                      socketid, &tx_conf);
1905                         if (ret < 0)
1906                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1907                                         "port=%d\n", ret, portid);
1908
1909                         qconf = &lcore_conf[lcore_id];
1910                         qconf->tx_queue_id[portid] = queueid;
1911                         queueid++;
1912                 }
1913                 printf("\n");
1914         }
1915
1916         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1917                 if (rte_lcore_is_enabled(lcore_id) == 0)
1918                         continue;
1919                 qconf = &lcore_conf[lcore_id];
1920                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1921                 fflush(stdout);
1922                 /* init RX queues */
1923                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1924                         portid = qconf->rx_queue_list[queue].port_id;
1925                         queueid = qconf->rx_queue_list[queue].queue_id;
1926
1927                         if (numa_on)
1928                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1929                         else
1930                                 socketid = 0;
1931
1932                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1933                         fflush(stdout);
1934
1935                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1936                                         socketid, &rx_conf, pktmbuf_pool[socketid]);
1937                         if (ret < 0)
1938                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
1939                                                 "port=%d\n", ret, portid);
1940                 }
1941         }
1942
1943         printf("\n");
1944
1945         /* start ports */
1946         for (portid = 0; portid < nb_ports; portid++) {
1947                 if ((enabled_port_mask & (1 << portid)) == 0) {
1948                         continue;
1949                 }
1950                 /* Start device */
1951                 ret = rte_eth_dev_start(portid);
1952                 if (ret < 0)
1953                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1954                                 ret, portid);
1955
1956                 /*
1957                  * If enabled, put device in promiscuous mode.
1958                  * This allows IO forwarding mode to forward packets
1959                  * to itself through 2 cross-connected  ports of the
1960                  * target machine.
1961                  */
1962                 if (promiscuous_on)
1963                         rte_eth_promiscuous_enable(portid);
1964         }
1965
1966         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1967
1968         /* launch per-lcore init on every lcore */
1969         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1970         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1971                 if (rte_eal_wait_lcore(lcore_id) < 0)
1972                         return -1;
1973         }
1974
1975         return 0;
1976 }