examples/l3fwd/main.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <stdio.h>
  35 #include <stdlib.h>
  36 #include <stdint.h>
  37 #include <inttypes.h>
  38 #include <sys/types.h>
  39 #include <string.h>
  40 #include <sys/queue.h>
  41 #include <stdarg.h>
  42 #include <errno.h>
  43 #include <getopt.h>
  44
  45 #include <rte_common.h>
  46 #include <rte_common_vect.h>
  47 #include <rte_byteorder.h>
  48 #include <rte_log.h>
  49 #include <rte_memory.h>
  50 #include <rte_memcpy.h>
  51 #include <rte_memzone.h>
  52 #include <rte_tailq.h>
  53 #include <rte_eal.h>
  54 #include <rte_per_lcore.h>
  55 #include <rte_launch.h>
  56 #include <rte_atomic.h>
  57 #include <rte_cycles.h>
  58 #include <rte_prefetch.h>
  59 #include <rte_lcore.h>
  60 #include <rte_per_lcore.h>
  61 #include <rte_branch_prediction.h>
  62 #include <rte_interrupts.h>
  63 #include <rte_pci.h>
  64 #include <rte_random.h>
  65 #include <rte_debug.h>
  66 #include <rte_ether.h>
  67 #include <rte_ethdev.h>
  68 #include <rte_ring.h>
  69 #include <rte_mempool.h>
  70 #include <rte_mbuf.h>
  71 #include <rte_ip.h>
  72 #include <rte_tcp.h>
  73 #include <rte_udp.h>
  74 #include <rte_string_fns.h>
  75
  76 #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 /*
  87  *  When set to zero, simple forwaring path is eanbled.
  88  *  When set to one, optimized forwarding path is enabled.
  89  *  Note that LPM optimisation path uses SSE4.1 instructions.
  90  */
  91 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && !defined(__SSE4_1__))
  92 #define ENABLE_MULTI_BUFFER_OPTIMIZE    0
  93 #else
  94 #define ENABLE_MULTI_BUFFER_OPTIMIZE    1
  95 #endif
  96
  97 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  98 #include <rte_hash.h>
  99 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 100 #include <rte_lpm.h>
 101 #include <rte_lpm6.h>
 102 #else
 103 #error "APP_LOOKUP_METHOD set to incorrect value"
 104 #endif
 105
 106 #ifndef IPv6_BYTES
 107 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 108                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 109 #define IPv6_BYTES(addr) \
 110         addr[0],  addr[1], addr[2],  addr[3], \
 111         addr[4],  addr[5], addr[6],  addr[7], \
 112         addr[8],  addr[9], addr[10], addr[11],\
 113         addr[12], addr[13],addr[14], addr[15]
 114 #endif
 115
 116
 117 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
 118
 119 #define MAX_JUMBO_PKT_LEN  9600
 120
 121 #define IPV6_ADDR_LEN 16
 122
 123 #define MEMPOOL_CACHE_SIZE 256
 124
 125 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
 126
 127 /*
 128  * This expression is used to calculate the number of mbufs needed depending on user input, taking
 129  *  into account memory for rx and tx hardware rings, cache per lcore and mtable per port per lcore.
 130  *  RTE_MAX is used to ensure that NB_MBUF never goes below a minimum value of 8192
 131  */
 132
 133 #define NB_MBUF RTE_MAX (                                                                                                                                       \
 134                                 (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT +                                                        \
 135                                 nb_ports*nb_lcores*MAX_PKT_BURST +                                                                                      \
 136                                 nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT +                                                          \
 137                                 nb_lcores*MEMPOOL_CACHE_SIZE),                                                                                          \
 138                                 (unsigned)8192)
 139
 140 /*
 141  * RX and TX Prefetch, Host, and Write-back threshold values should be
 142  * carefully set for optimal performance. Consult the network
 143  * controller's datasheet and supporting DPDK documentation for guidance
 144  * on how these parameters should be set.
 145  */
 146 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
 147 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
 148 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
 149
 150 /*
 151  * These default values are optimized for use with the Intel(R) 82599 10 GbE
 152  * Controller and the DPDK ixgbe PMD. Consider using other values for other
 153  * network controllers and/or network drivers.
 154  */
 155 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
 156 #define TX_HTHRESH 0  /**< Default values of TX host threshold reg. */
 157 #define TX_WTHRESH 0  /**< Default values of TX write-back threshold reg. */
 158
 159 #define MAX_PKT_BURST     32
 160 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 161
 162 /*
 163  * Try to avoid TX buffering if we have at least MAX_TX_BURST packets to send.
 164  */
 165 #define MAX_TX_BURST    (MAX_PKT_BURST / 2)
 166
 167 #define NB_SOCKETS 8
 168
 169 /* Configure how many packets ahead to prefetch, when reading packets */
 170 #define PREFETCH_OFFSET 3
 171
 172 /* Used to mark destination port as 'invalid'. */
 173 #define BAD_PORT        ((uint16_t)-1)
 174
 175 #define FWDSTEP 4
 176
 177 /*
 178  * Configurable number of RX/TX ring descriptors
 179  */
 180 #define RTE_TEST_RX_DESC_DEFAULT 128
 181 #define RTE_TEST_TX_DESC_DEFAULT 512
 182 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 183 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 184
 185 /* ethernet addresses of ports */
 186 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 187
 188 static __m128i val_eth[RTE_MAX_ETHPORTS];
 189
 190 /* replace first 12B of the ethernet header. */
 191 #define MASK_ETH        0x3f
 192
 193 /* mask of enabled ports */
 194 static uint32_t enabled_port_mask = 0;
 195 static int promiscuous_on = 0; /**< Ports set in promiscuous mode off by default. */
 196 static int numa_on = 1; /**< NUMA is enabled by default. */
 197
 198 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 199 static int ipv6 = 0; /**< ipv6 is false by default. */
 200 #endif
 201
 202 struct mbuf_table {
 203         uint16_t len;
 204         struct rte_mbuf *m_table[MAX_PKT_BURST];
 205 };
 206
 207 struct lcore_rx_queue {
 208         uint8_t port_id;
 209         uint8_t queue_id;
 210 } __rte_cache_aligned;
 211
 212 #define MAX_RX_QUEUE_PER_LCORE 16
 213 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 214 #define MAX_RX_QUEUE_PER_PORT 128
 215
 216 #define MAX_LCORE_PARAMS 1024
 217 struct lcore_params {
 218         uint8_t port_id;
 219         uint8_t queue_id;
 220         uint8_t lcore_id;
 221 } __rte_cache_aligned;
 222
 223 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 224 static struct lcore_params lcore_params_array_default[] = {
 225         {0, 0, 2},
 226         {0, 1, 2},
 227         {0, 2, 2},
 228         {1, 0, 2},
 229         {1, 1, 2},
 230         {1, 2, 2},
 231         {2, 0, 2},
 232         {3, 0, 3},
 233         {3, 1, 3},
 234 };
 235
 236 static struct lcore_params * lcore_params = lcore_params_array_default;
 237 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 238                                 sizeof(lcore_params_array_default[0]);
 239
 240 static struct rte_eth_conf port_conf = {
 241         .rxmode = {
 242                 .mq_mode = ETH_MQ_RX_RSS,
 243                 .max_rx_pkt_len = ETHER_MAX_LEN,
 244                 .split_hdr_size = 0,
 245                 .header_split   = 0, /**< Header Split disabled */
 246                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 247                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 248                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 249                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 250         },
 251         .rx_adv_conf = {
 252                 .rss_conf = {
 253                         .rss_key = NULL,
 254                         .rss_hf = ETH_RSS_IPV4 | ETH_RSS_IPV6,
 255                 },
 256         },
 257         .txmode = {
 258                 .mq_mode = ETH_MQ_TX_NONE,
 259         },
 260 };
 261
 262 static const struct rte_eth_rxconf rx_conf = {
 263         .rx_thresh = {
 264                 .pthresh = RX_PTHRESH,
 265                 .hthresh = RX_HTHRESH,
 266                 .wthresh = RX_WTHRESH,
 267         },
 268         .rx_free_thresh = 32,
 269 };
 270
 271 static struct rte_eth_txconf tx_conf = {
 272         .tx_thresh = {
 273                 .pthresh = TX_PTHRESH,
 274                 .hthresh = TX_HTHRESH,
 275                 .wthresh = TX_WTHRESH,
 276         },
 277         .tx_free_thresh = 0, /* Use PMD default values */
 278         .tx_rs_thresh = 0, /* Use PMD default values */
 279         .txq_flags = (ETH_TXQ_FLAGS_NOMULTSEGS |
 280                         ETH_TXQ_FLAGS_NOVLANOFFL |
 281                         ETH_TXQ_FLAGS_NOXSUMSCTP |
 282                         ETH_TXQ_FLAGS_NOXSUMUDP |
 283                         ETH_TXQ_FLAGS_NOXSUMTCP)
 284
 285 };
 286
 287 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 288
 289 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 290
 291 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 292 #include <rte_hash_crc.h>
 293 #define DEFAULT_HASH_FUNC       rte_hash_crc
 294 #else
 295 #include <rte_jhash.h>
 296 #define DEFAULT_HASH_FUNC       rte_jhash
 297 #endif
 298
 299 struct ipv4_5tuple {
 300         uint32_t ip_dst;
 301         uint32_t ip_src;
 302         uint16_t port_dst;
 303         uint16_t port_src;
 304         uint8_t  proto;
 305 } __attribute__((__packed__));
 306
 307 union ipv4_5tuple_host {
 308         struct {
 309                 uint8_t  pad0;
 310                 uint8_t  proto;
 311                 uint16_t pad1;
 312                 uint32_t ip_src;
 313                 uint32_t ip_dst;
 314                 uint16_t port_src;
 315                 uint16_t port_dst;
 316         };
 317         __m128i xmm;
 318 };
 319
 320 #define XMM_NUM_IN_IPV6_5TUPLE 3
 321
 322 struct ipv6_5tuple {
 323         uint8_t  ip_dst[IPV6_ADDR_LEN];
 324         uint8_t  ip_src[IPV6_ADDR_LEN];
 325         uint16_t port_dst;
 326         uint16_t port_src;
 327         uint8_t  proto;
 328 } __attribute__((__packed__));
 329
 330 union ipv6_5tuple_host {
 331         struct {
 332                 uint16_t pad0;
 333                 uint8_t  proto;
 334                 uint8_t  pad1;
 335                 uint8_t  ip_src[IPV6_ADDR_LEN];
 336                 uint8_t  ip_dst[IPV6_ADDR_LEN];
 337                 uint16_t port_src;
 338                 uint16_t port_dst;
 339                 uint64_t reserve;
 340         };
 341         __m128i xmm[XMM_NUM_IN_IPV6_5TUPLE];
 342 };
 343
 344 struct ipv4_l3fwd_route {
 345         struct ipv4_5tuple key;
 346         uint8_t if_out;
 347 };
 348
 349 struct ipv6_l3fwd_route {
 350         struct ipv6_5tuple key;
 351         uint8_t if_out;
 352 };
 353
 354 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 355         {{IPv4(101,0,0,0), IPv4(100,10,0,1),  101, 11, IPPROTO_TCP}, 0},
 356         {{IPv4(201,0,0,0), IPv4(200,20,0,1),  102, 12, IPPROTO_TCP}, 1},
 357         {{IPv4(111,0,0,0), IPv4(100,30,0,1),  101, 11, IPPROTO_TCP}, 2},
 358         {{IPv4(211,0,0,0), IPv4(200,40,0,1),  102, 12, IPPROTO_TCP}, 3},
 359 };
 360
 361 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 362         {{
 363         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 364         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 365         101, 11, IPPROTO_TCP}, 0},
 366
 367         {{
 368         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 369         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 370         102, 12, IPPROTO_TCP}, 1},
 371
 372         {{
 373         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 374         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 375         101, 11, IPPROTO_TCP}, 2},
 376
 377         {{
 378         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 379         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 380         102, 12, IPPROTO_TCP}, 3},
 381 };
 382
 383 typedef struct rte_hash lookup_struct_t;
 384 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 385 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 386
 387 #ifdef RTE_ARCH_X86_64
 388 /* default to 4 million hash entries (approx) */
 389 #define L3FWD_HASH_ENTRIES              1024*1024*4
 390 #else
 391 /* 32-bit has less address-space for hugepage memory, limit to 1M entries */
 392 #define L3FWD_HASH_ENTRIES              1024*1024*1
 393 #endif
 394 #define HASH_ENTRY_NUMBER_DEFAULT       4
 395
 396 static uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
 397
 398 static inline uint32_t
 399 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
 400         uint32_t init_val)
 401 {
 402         const union ipv4_5tuple_host *k;
 403         uint32_t t;
 404         const uint32_t *p;
 405
 406         k = data;
 407         t = k->proto;
 408         p = (const uint32_t *)&k->port_src;
 409
 410 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 411         init_val = rte_hash_crc_4byte(t, init_val);
 412         init_val = rte_hash_crc_4byte(k->ip_src, init_val);
 413         init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
 414         init_val = rte_hash_crc_4byte(*p, init_val);
 415 #else /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 416         init_val = rte_jhash_1word(t, init_val);
 417         init_val = rte_jhash_1word(k->ip_src, init_val);
 418         init_val = rte_jhash_1word(k->ip_dst, init_val);
 419         init_val = rte_jhash_1word(*p, init_val);
 420 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 421         return (init_val);
 422 }
 423
 424 static inline uint32_t
 425 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len, uint32_t init_val)
 426 {
 427         const union ipv6_5tuple_host *k;
 428         uint32_t t;
 429         const uint32_t *p;
 430 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 431         const uint32_t  *ip_src0, *ip_src1, *ip_src2, *ip_src3;
 432         const uint32_t  *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
 433 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 434
 435         k = data;
 436         t = k->proto;
 437         p = (const uint32_t *)&k->port_src;
 438
 439 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 440         ip_src0 = (const uint32_t *) k->ip_src;
 441         ip_src1 = (const uint32_t *)(k->ip_src+4);
 442         ip_src2 = (const uint32_t *)(k->ip_src+8);
 443         ip_src3 = (const uint32_t *)(k->ip_src+12);
 444         ip_dst0 = (const uint32_t *) k->ip_dst;
 445         ip_dst1 = (const uint32_t *)(k->ip_dst+4);
 446         ip_dst2 = (const uint32_t *)(k->ip_dst+8);
 447         ip_dst3 = (const uint32_t *)(k->ip_dst+12);
 448         init_val = rte_hash_crc_4byte(t, init_val);
 449         init_val = rte_hash_crc_4byte(*ip_src0, init_val);
 450         init_val = rte_hash_crc_4byte(*ip_src1, init_val);
 451         init_val = rte_hash_crc_4byte(*ip_src2, init_val);
 452         init_val = rte_hash_crc_4byte(*ip_src3, init_val);
 453         init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
 454         init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
 455         init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
 456         init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
 457         init_val = rte_hash_crc_4byte(*p, init_val);
 458 #else /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 459         init_val = rte_jhash_1word(t, init_val);
 460         init_val = rte_jhash(k->ip_src, sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
 461         init_val = rte_jhash(k->ip_dst, sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
 462         init_val = rte_jhash_1word(*p, init_val);
 463 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 464         return (init_val);
 465 }
 466
 467 #define IPV4_L3FWD_NUM_ROUTES \
 468         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 469
 470 #define IPV6_L3FWD_NUM_ROUTES \
 471         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 472
 473 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 474 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 475
 476 #endif
 477
 478 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 479 struct ipv4_l3fwd_route {
 480         uint32_t ip;
 481         uint8_t  depth;
 482         uint8_t  if_out;
 483 };
 484
 485 struct ipv6_l3fwd_route {
 486         uint8_t ip[16];
 487         uint8_t  depth;
 488         uint8_t  if_out;
 489 };
 490
 491 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 492         {IPv4(1,1,1,0), 24, 0},
 493         {IPv4(2,1,1,0), 24, 1},
 494         {IPv4(3,1,1,0), 24, 2},
 495         {IPv4(4,1,1,0), 24, 3},
 496         {IPv4(5,1,1,0), 24, 4},
 497         {IPv4(6,1,1,0), 24, 5},
 498         {IPv4(7,1,1,0), 24, 6},
 499         {IPv4(8,1,1,0), 24, 7},
 500 };
 501
 502 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 503         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
 504         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
 505         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
 506         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
 507         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
 508         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
 509         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
 510         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
 511 };
 512
 513 #define IPV4_L3FWD_NUM_ROUTES \
 514         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 515 #define IPV6_L3FWD_NUM_ROUTES \
 516         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 517
 518 #define IPV4_L3FWD_LPM_MAX_RULES         1024
 519 #define IPV6_L3FWD_LPM_MAX_RULES         1024
 520 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
 521
 522 typedef struct rte_lpm lookup_struct_t;
 523 typedef struct rte_lpm6 lookup6_struct_t;
 524 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 525 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 526 #endif
 527
 528 struct lcore_conf {
 529         uint16_t n_rx_queue;
 530         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 531         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 532         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 533         lookup_struct_t * ipv4_lookup_struct;
 534 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 535         lookup6_struct_t * ipv6_lookup_struct;
 536 #else
 537         lookup_struct_t * ipv6_lookup_struct;
 538 #endif
 539 } __rte_cache_aligned;
 540
 541 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 542
 543 /* Send burst of packets on an output interface */
 544 static inline int
 545 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 546 {
 547         struct rte_mbuf **m_table;
 548         int ret;
 549         uint16_t queueid;
 550
 551         queueid = qconf->tx_queue_id[port];
 552         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 553
 554         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 555         if (unlikely(ret < n)) {
 556                 do {
 557                         rte_pktmbuf_free(m_table[ret]);
 558                 } while (++ret < n);
 559         }
 560
 561         return 0;
 562 }
 563
 564 /* Enqueue a single packet, and send burst if queue is filled */
 565 static inline int
 566 send_single_packet(struct rte_mbuf *m, uint8_t port)
 567 {
 568         uint32_t lcore_id;
 569         uint16_t len;
 570         struct lcore_conf *qconf;
 571
 572         lcore_id = rte_lcore_id();
 573
 574         qconf = &lcore_conf[lcore_id];
 575         len = qconf->tx_mbufs[port].len;
 576         qconf->tx_mbufs[port].m_table[len] = m;
 577         len++;
 578
 579         /* enough pkts to be sent */
 580         if (unlikely(len == MAX_PKT_BURST)) {
 581                 send_burst(qconf, MAX_PKT_BURST, port);
 582                 len = 0;
 583         }
 584
 585         qconf->tx_mbufs[port].len = len;
 586         return 0;
 587 }
 588
 589 static inline __attribute__((always_inline)) void
 590 send_packetsx4(struct lcore_conf *qconf, uint8_t port,
 591         struct rte_mbuf *m[], uint32_t num)
 592 {
 593         uint32_t len, j, n;
 594
 595         len = qconf->tx_mbufs[port].len;
 596
 597         /*
 598          * If TX buffer for that queue is empty, and we have enough packets,
 599          * then send them straightway.
 600          */
 601         if (num >= MAX_TX_BURST && len == 0) {
 602                 n = rte_eth_tx_burst(port, qconf->tx_queue_id[port], m, num);
 603                 if (unlikely(n < num)) {
 604                         do {
 605                                 rte_pktmbuf_free(m[n]);
 606                         } while (++n < num);
 607                 }
 608                 return;
 609         }
 610
 611         /*
 612          * Put packets into TX buffer for that queue.
 613          */
 614
 615         n = len + num;
 616         n = (n > MAX_PKT_BURST) ? MAX_PKT_BURST - len : num;
 617
 618         j = 0;
 619         switch (n % FWDSTEP) {
 620         while (j < n) {
 621         case 0:
 622                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 623                 j++;
 624         case 3:
 625                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 626                 j++;
 627         case 2:
 628                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 629                 j++;
 630         case 1:
 631                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 632                 j++;
 633         }
 634         }
 635
 636         len += n;
 637
 638         /* enough pkts to be sent */
 639         if (unlikely(len == MAX_PKT_BURST)) {
 640
 641                 send_burst(qconf, MAX_PKT_BURST, port);
 642
 643                 /* copy rest of the packets into the TX buffer. */
 644                 len = num - n;
 645                 j = 0;
 646                 switch (len % FWDSTEP) {
 647                 while (j < len) {
 648                 case 0:
 649                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 650                         j++;
 651                 case 3:
 652                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 653                         j++;
 654                 case 2:
 655                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 656                         j++;
 657                 case 1:
 658                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 659                         j++;
 660                 }
 661                 }
 662         }
 663
 664         qconf->tx_mbufs[port].len = len;
 665 }
 666
 667 #ifdef DO_RFC_1812_CHECKS
 668 static inline int
 669 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 670 {
 671         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 672         /*
 673          * 1. The packet length reported by the Link Layer must be large
 674          * enough to hold the minimum length legal IP datagram (20 bytes).
 675          */
 676         if (link_len < sizeof(struct ipv4_hdr))
 677                 return -1;
 678
 679         /* 2. The IP checksum must be correct. */
 680         /* this is checked in H/W */
 681
 682         /*
 683          * 3. The IP version number must be 4. If the version number is not 4
 684          * then the packet may be another version of IP, such as IPng or
 685          * ST-II.
 686          */
 687         if (((pkt->version_ihl) >> 4) != 4)
 688                 return -3;
 689         /*
 690          * 4. The IP header length field must be large enough to hold the
 691          * minimum length legal IP datagram (20 bytes = 5 words).
 692          */
 693         if ((pkt->version_ihl & 0xf) < 5)
 694                 return -4;
 695
 696         /*
 697          * 5. The IP total length field must be large enough to hold the IP
 698          * datagram header, whose length is specified in the IP header length
 699          * field.
 700          */
 701         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 702                 return -5;
 703
 704         return 0;
 705 }
 706 #endif
 707
 708 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 709
 710 static __m128i mask0;
 711 static __m128i mask1;
 712 static __m128i mask2;
 713 static inline uint8_t
 714 get_ipv4_dst_port(void *ipv4_hdr, uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
 715 {
 716         int ret = 0;
 717         union ipv4_5tuple_host key;
 718
 719         ipv4_hdr = (uint8_t *)ipv4_hdr + offsetof(struct ipv4_hdr, time_to_live);
 720         __m128i data = _mm_loadu_si128((__m128i*)(ipv4_hdr));
 721         /* Get 5 tuple: dst port, src port, dst IP address, src IP address and protocol */
 722         key.xmm = _mm_and_si128(data, mask0);
 723         /* Find destination port */
 724         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 725         return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
 726 }
 727
 728 static inline uint8_t
 729 get_ipv6_dst_port(void *ipv6_hdr,  uint8_t portid, lookup_struct_t * ipv6_l3fwd_lookup_struct)
 730 {
 731         int ret = 0;
 732         union ipv6_5tuple_host key;
 733
 734         ipv6_hdr = (uint8_t *)ipv6_hdr + offsetof(struct ipv6_hdr, payload_len);
 735         __m128i data0 = _mm_loadu_si128((__m128i*)(ipv6_hdr));
 736         __m128i data1 = _mm_loadu_si128((__m128i*)(((uint8_t*)ipv6_hdr)+sizeof(__m128i)));
 737         __m128i data2 = _mm_loadu_si128((__m128i*)(((uint8_t*)ipv6_hdr)+sizeof(__m128i)+sizeof(__m128i)));
 738         /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
 739         key.xmm[0] = _mm_and_si128(data0, mask1);
 740         /* Get part of 5 tuple: dst IP address lower 96 bits and src IP address higher 32 bits */
 741         key.xmm[1] = data1;
 742         /* Get part of 5 tuple: dst port and src port and dst IP address higher 32 bits */
 743         key.xmm[2] = _mm_and_si128(data2, mask2);
 744
 745         /* Find destination port */
 746         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 747         return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
 748 }
 749 #endif
 750
 751 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 752
 753 static inline uint8_t
 754 get_ipv4_dst_port(void *ipv4_hdr,  uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
 755 {
 756         uint8_t next_hop;
 757
 758         return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 759                 rte_be_to_cpu_32(((struct ipv4_hdr *)ipv4_hdr)->dst_addr),
 760                 &next_hop) == 0) ? next_hop : portid);
 761 }
 762
 763 static inline uint8_t
 764 get_ipv6_dst_port(void *ipv6_hdr,  uint8_t portid, lookup6_struct_t * ipv6_l3fwd_lookup_struct)
 765 {
 766         uint8_t next_hop;
 767         return (uint8_t) ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
 768                         ((struct ipv6_hdr*)ipv6_hdr)->dst_addr, &next_hop) == 0)?
 769                         next_hop : portid);
 770 }
 771 #endif
 772
 773 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) && \
 774         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
 775 static inline void l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, struct lcore_conf *qconf);
 776
 777 #define MASK_ALL_PKTS    0xf
 778 #define EXECLUDE_1ST_PKT 0xe
 779 #define EXECLUDE_2ND_PKT 0xd
 780 #define EXECLUDE_3RD_PKT 0xb
 781 #define EXECLUDE_4TH_PKT 0x7
 782
 783 static inline void
 784 simple_ipv4_fwd_4pkts(struct rte_mbuf* m[4], uint8_t portid, struct lcore_conf *qconf)
 785 {
 786         struct ether_hdr *eth_hdr[4];
 787         struct ipv4_hdr *ipv4_hdr[4];
 788         void *d_addr_bytes[4];
 789         uint8_t dst_port[4];
 790         int32_t ret[4];
 791         union ipv4_5tuple_host key[4];
 792         __m128i data[4];
 793
 794         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
 795         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
 796         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
 797         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
 798
 799         /* Handle IPv4 headers.*/
 800         ipv4_hdr[0] = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m[0], unsigned char *) +
 801                         sizeof(struct ether_hdr));
 802         ipv4_hdr[1] = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m[1], unsigned char *) +
 803                         sizeof(struct ether_hdr));
 804         ipv4_hdr[2] = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m[2], unsigned char *) +
 805                         sizeof(struct ether_hdr));
 806         ipv4_hdr[3] = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m[3], unsigned char *) +
 807                         sizeof(struct ether_hdr));
 808
 809 #ifdef DO_RFC_1812_CHECKS
 810         /* Check to make sure the packet is valid (RFC1812) */
 811         uint8_t valid_mask = MASK_ALL_PKTS;
 812         if (is_valid_ipv4_pkt(ipv4_hdr[0], m[0]->pkt.pkt_len) < 0) {
 813                 rte_pktmbuf_free(m[0]);
 814                 valid_mask &= EXECLUDE_1ST_PKT;
 815         }
 816         if (is_valid_ipv4_pkt(ipv4_hdr[1], m[1]->pkt.pkt_len) < 0) {
 817                 rte_pktmbuf_free(m[1]);
 818                 valid_mask &= EXECLUDE_2ND_PKT;
 819         }
 820         if (is_valid_ipv4_pkt(ipv4_hdr[2], m[2]->pkt.pkt_len) < 0) {
 821                 rte_pktmbuf_free(m[2]);
 822                 valid_mask &= EXECLUDE_3RD_PKT;
 823         }
 824         if (is_valid_ipv4_pkt(ipv4_hdr[3], m[3]->pkt.pkt_len) < 0) {
 825                 rte_pktmbuf_free(m[3]);
 826                 valid_mask &= EXECLUDE_4TH_PKT;
 827         }
 828         if (unlikely(valid_mask != MASK_ALL_PKTS)) {
 829                 if (valid_mask == 0){
 830                         return;
 831                 } else {
 832                         uint8_t i = 0;
 833                         for (i = 0; i < 4; i++) {
 834                                 if ((0x1 << i) & valid_mask) {
 835                                         l3fwd_simple_forward(m[i], portid, qconf);
 836                                 }
 837                         }
 838                         return;
 839                 }
 840         }
 841 #endif // End of #ifdef DO_RFC_1812_CHECKS
 842
 843         data[0] = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m[0], unsigned char *) +
 844                 sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
 845         data[1] = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m[1], unsigned char *) +
 846                 sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
 847         data[2] = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m[2], unsigned char *) +
 848                 sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
 849         data[3] = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m[3], unsigned char *) +
 850                 sizeof(struct ether_hdr) + offsetof(struct ipv4_hdr, time_to_live)));
 851
 852         key[0].xmm = _mm_and_si128(data[0], mask0);
 853         key[1].xmm = _mm_and_si128(data[1], mask0);
 854         key[2].xmm = _mm_and_si128(data[2], mask0);
 855         key[3].xmm = _mm_and_si128(data[3], mask0);
 856
 857         const void *key_array[4] = {&key[0], &key[1], &key[2],&key[3]};
 858         rte_hash_lookup_multi(qconf->ipv4_lookup_struct, &key_array[0], 4, ret);
 859         dst_port[0] = (uint8_t) ((ret[0] < 0) ? portid : ipv4_l3fwd_out_if[ret[0]]);
 860         dst_port[1] = (uint8_t) ((ret[1] < 0) ? portid : ipv4_l3fwd_out_if[ret[1]]);
 861         dst_port[2] = (uint8_t) ((ret[2] < 0) ? portid : ipv4_l3fwd_out_if[ret[2]]);
 862         dst_port[3] = (uint8_t) ((ret[3] < 0) ? portid : ipv4_l3fwd_out_if[ret[3]]);
 863
 864         if (dst_port[0] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[0]) == 0)
 865                 dst_port[0] = portid;
 866         if (dst_port[1] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[1]) == 0)
 867                 dst_port[1] = portid;
 868         if (dst_port[2] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[2]) == 0)
 869                 dst_port[2] = portid;
 870         if (dst_port[3] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[3]) == 0)
 871                 dst_port[3] = portid;
 872
 873         /* 02:00:00:00:00:xx */
 874         d_addr_bytes[0] = &eth_hdr[0]->d_addr.addr_bytes[0];
 875         d_addr_bytes[1] = &eth_hdr[1]->d_addr.addr_bytes[0];
 876         d_addr_bytes[2] = &eth_hdr[2]->d_addr.addr_bytes[0];
 877         d_addr_bytes[3] = &eth_hdr[3]->d_addr.addr_bytes[0];
 878         *((uint64_t *)d_addr_bytes[0]) = 0x000000000002 + ((uint64_t)dst_port[0] << 40);
 879         *((uint64_t *)d_addr_bytes[1]) = 0x000000000002 + ((uint64_t)dst_port[1] << 40);
 880         *((uint64_t *)d_addr_bytes[2]) = 0x000000000002 + ((uint64_t)dst_port[2] << 40);
 881         *((uint64_t *)d_addr_bytes[3]) = 0x000000000002 + ((uint64_t)dst_port[3] << 40);
 882
 883 #ifdef DO_RFC_1812_CHECKS
 884         /* Update time to live and header checksum */
 885         --(ipv4_hdr[0]->time_to_live);
 886         --(ipv4_hdr[1]->time_to_live);
 887         --(ipv4_hdr[2]->time_to_live);
 888         --(ipv4_hdr[3]->time_to_live);
 889         ++(ipv4_hdr[0]->hdr_checksum);
 890         ++(ipv4_hdr[1]->hdr_checksum);
 891         ++(ipv4_hdr[2]->hdr_checksum);
 892         ++(ipv4_hdr[3]->hdr_checksum);
 893 #endif
 894
 895         /* src addr */
 896         ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
 897         ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
 898         ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
 899         ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
 900
 901         send_single_packet(m[0], (uint8_t)dst_port[0]);
 902         send_single_packet(m[1], (uint8_t)dst_port[1]);
 903         send_single_packet(m[2], (uint8_t)dst_port[2]);
 904         send_single_packet(m[3], (uint8_t)dst_port[3]);
 905
 906 }
 907
 908 static inline void get_ipv6_5tuple(struct rte_mbuf* m0, __m128i mask0, __m128i mask1,
 909                                  union ipv6_5tuple_host * key)
 910 {
 911         __m128i tmpdata0 = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m0, unsigned char *)
 912                         + sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len)));
 913         __m128i tmpdata1 = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m0, unsigned char *)
 914                         + sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len)
 915                         +  sizeof(__m128i)));
 916         __m128i tmpdata2 = _mm_loadu_si128((__m128i*)(rte_pktmbuf_mtod(m0, unsigned char *)
 917                         + sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len)
 918                         + sizeof(__m128i) + sizeof(__m128i)));
 919         key->xmm[0] = _mm_and_si128(tmpdata0, mask0);
 920         key->xmm[1] = tmpdata1;
 921         key->xmm[2] = _mm_and_si128(tmpdata2, mask1);
 922         return;
 923 }
 924
 925 static inline void
 926 simple_ipv6_fwd_4pkts(struct rte_mbuf* m[4], uint8_t portid, struct lcore_conf *qconf)
 927 {
 928         struct ether_hdr *eth_hdr[4];
 929         __attribute__((unused)) struct ipv6_hdr *ipv6_hdr[4];
 930         void *d_addr_bytes[4];
 931         uint8_t dst_port[4];
 932         int32_t ret[4];
 933         union ipv6_5tuple_host key[4];
 934
 935         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
 936         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
 937         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
 938         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
 939
 940         /* Handle IPv6 headers.*/
 941         ipv6_hdr[0] = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m[0], unsigned char *) +
 942                         sizeof(struct ether_hdr));
 943         ipv6_hdr[1] = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m[1], unsigned char *) +
 944                         sizeof(struct ether_hdr));
 945         ipv6_hdr[2] = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m[2], unsigned char *) +
 946                         sizeof(struct ether_hdr));
 947         ipv6_hdr[3] = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m[3], unsigned char *) +
 948                         sizeof(struct ether_hdr));
 949
 950         get_ipv6_5tuple(m[0], mask1, mask2, &key[0]);
 951         get_ipv6_5tuple(m[1], mask1, mask2, &key[1]);
 952         get_ipv6_5tuple(m[2], mask1, mask2, &key[2]);
 953         get_ipv6_5tuple(m[3], mask1, mask2, &key[3]);
 954
 955         const void *key_array[4] = {&key[0], &key[1], &key[2],&key[3]};
 956         rte_hash_lookup_multi(qconf->ipv6_lookup_struct, &key_array[0], 4, ret);
 957         dst_port[0] = (uint8_t) ((ret[0] < 0)? portid:ipv6_l3fwd_out_if[ret[0]]);
 958         dst_port[1] = (uint8_t) ((ret[1] < 0)? portid:ipv6_l3fwd_out_if[ret[1]]);
 959         dst_port[2] = (uint8_t) ((ret[2] < 0)? portid:ipv6_l3fwd_out_if[ret[2]]);
 960         dst_port[3] = (uint8_t) ((ret[3] < 0)? portid:ipv6_l3fwd_out_if[ret[3]]);
 961
 962         if (dst_port[0] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[0]) == 0)
 963                 dst_port[0] = portid;
 964         if (dst_port[1] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[1]) == 0)
 965                 dst_port[1] = portid;
 966         if (dst_port[2] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[2]) == 0)
 967                 dst_port[2] = portid;
 968         if (dst_port[3] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[3]) == 0)
 969                 dst_port[3] = portid;
 970
 971         /* 02:00:00:00:00:xx */
 972         d_addr_bytes[0] = &eth_hdr[0]->d_addr.addr_bytes[0];
 973         d_addr_bytes[1] = &eth_hdr[1]->d_addr.addr_bytes[0];
 974         d_addr_bytes[2] = &eth_hdr[2]->d_addr.addr_bytes[0];
 975         d_addr_bytes[3] = &eth_hdr[3]->d_addr.addr_bytes[0];
 976         *((uint64_t *)d_addr_bytes[0]) = 0x000000000002 + ((uint64_t)dst_port[0] << 40);
 977         *((uint64_t *)d_addr_bytes[1]) = 0x000000000002 + ((uint64_t)dst_port[1] << 40);
 978         *((uint64_t *)d_addr_bytes[2]) = 0x000000000002 + ((uint64_t)dst_port[2] << 40);
 979         *((uint64_t *)d_addr_bytes[3]) = 0x000000000002 + ((uint64_t)dst_port[3] << 40);
 980
 981         /* src addr */
 982         ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
 983         ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
 984         ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
 985         ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
 986
 987         send_single_packet(m[0], (uint8_t)dst_port[0]);
 988         send_single_packet(m[1], (uint8_t)dst_port[1]);
 989         send_single_packet(m[2], (uint8_t)dst_port[2]);
 990         send_single_packet(m[3], (uint8_t)dst_port[3]);
 991
 992 }
 993 #endif /* APP_LOOKUP_METHOD */
 994
 995 static inline __attribute__((always_inline)) void
 996 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, struct lcore_conf *qconf)
 997 {
 998         struct ether_hdr *eth_hdr;
 999         struct ipv4_hdr *ipv4_hdr;
1000         void *d_addr_bytes;
1001         uint8_t dst_port;
1002
1003         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
1004
1005         if (m->ol_flags & PKT_RX_IPV4_HDR) {
1006                 /* Handle IPv4 headers.*/
1007                 ipv4_hdr = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
1008                                 sizeof(struct ether_hdr));
1009
1010 #ifdef DO_RFC_1812_CHECKS
1011                 /* Check to make sure the packet is valid (RFC1812) */
1012                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt.pkt_len) < 0) {
1013                         rte_pktmbuf_free(m);
1014                         return;
1015                 }
1016 #endif
1017
1018                  dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
1019                         qconf->ipv4_lookup_struct);
1020                 if (dst_port >= RTE_MAX_ETHPORTS ||
1021                                 (enabled_port_mask & 1 << dst_port) == 0)
1022                         dst_port = portid;
1023
1024                 /* 02:00:00:00:00:xx */
1025                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
1026                 *((uint64_t *)d_addr_bytes) = ETHER_LOCAL_ADMIN_ADDR +
1027                         ((uint64_t)dst_port << 40);
1028
1029 #ifdef DO_RFC_1812_CHECKS
1030                 /* Update time to live and header checksum */
1031                 --(ipv4_hdr->time_to_live);
1032                 ++(ipv4_hdr->hdr_checksum);
1033 #endif
1034
1035                 /* src addr */
1036                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
1037
1038                 send_single_packet(m, dst_port);
1039
1040         } else {
1041                 /* Handle IPv6 headers.*/
1042                 struct ipv6_hdr *ipv6_hdr;
1043
1044                 ipv6_hdr = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
1045                                 sizeof(struct ether_hdr));
1046
1047                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid, qconf->ipv6_lookup_struct);
1048
1049                 if (dst_port >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port) == 0)
1050                         dst_port = portid;
1051
1052                 /* 02:00:00:00:00:xx */
1053                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
1054                 *((uint64_t *)d_addr_bytes) = ETHER_LOCAL_ADMIN_ADDR +
1055                         ((uint64_t)dst_port << 40);
1056
1057                 /* src addr */
1058                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
1059
1060                 send_single_packet(m, dst_port);
1061         }
1062
1063 }
1064
1065 #ifdef DO_RFC_1812_CHECKS
1066
1067 #define IPV4_MIN_VER_IHL        0x45
1068 #define IPV4_MAX_VER_IHL        0x4f
1069 #define IPV4_MAX_VER_IHL_DIFF   (IPV4_MAX_VER_IHL - IPV4_MIN_VER_IHL)
1070
1071 /* Minimum value of IPV4 total length (20B) in network byte order. */
1072 #define IPV4_MIN_LEN_BE (sizeof(struct ipv4_hdr) << 8)
1073
1074 /*
1075  * From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2:
1076  * - The IP version number must be 4.
1077  * - The IP header length field must be large enough to hold the
1078  *    minimum length legal IP datagram (20 bytes = 5 words).
1079  * - The IP total length field must be large enough to hold the IP
1080  *   datagram header, whose length is specified in the IP header length
1081  *   field.
1082  * If we encounter invalid IPV4 packet, then set destination port for it
1083  * to BAD_PORT value.
1084  */
1085 static inline __attribute__((always_inline)) void
1086 rfc1812_process(struct ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t flags)
1087 {
1088         uint8_t ihl;
1089
1090         if ((flags & PKT_RX_IPV4_HDR) != 0) {
1091
1092                 ihl = ipv4_hdr->version_ihl - IPV4_MIN_VER_IHL;
1093
1094                 ipv4_hdr->time_to_live--;
1095                 ipv4_hdr->hdr_checksum++;
1096
1097                 if (ihl > IPV4_MAX_VER_IHL_DIFF ||
1098                                 ((uint8_t)ipv4_hdr->total_length == 0 &&
1099                                 ipv4_hdr->total_length < IPV4_MIN_LEN_BE)) {
1100                         dp[0] = BAD_PORT;
1101                 }
1102         }
1103 }
1104
1105 #else
1106 #define rfc1812_process(mb, dp) do { } while (0)
1107 #endif /* DO_RFC_1812_CHECKS */
1108
1109
1110 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1111         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1112
1113 static inline __attribute__((always_inline)) uint16_t
1114 get_dst_port(const struct lcore_conf *qconf, struct rte_mbuf *pkt,
1115         uint32_t dst_ipv4, uint8_t portid)
1116 {
1117         uint8_t next_hop;
1118         struct ipv6_hdr *ipv6_hdr;
1119         struct ether_hdr *eth_hdr;
1120
1121         if (pkt->ol_flags & PKT_RX_IPV4_HDR) {
1122                 if (rte_lpm_lookup(qconf->ipv4_lookup_struct, dst_ipv4,
1123                                 &next_hop) != 0)
1124                         next_hop = portid;
1125         } else if (pkt->ol_flags & PKT_RX_IPV6_HDR) {
1126                 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1127                 ipv6_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
1128                 if (rte_lpm6_lookup(qconf->ipv6_lookup_struct,
1129                                 ipv6_hdr->dst_addr, &next_hop) != 0)
1130                         next_hop = portid;
1131         } else {
1132                 next_hop = portid;
1133         }
1134
1135         return next_hop;
1136 }
1137
1138 static inline void
1139 process_packet(struct lcore_conf *qconf, struct rte_mbuf *pkt,
1140         uint16_t *dst_port, uint8_t portid)
1141 {
1142         struct ether_hdr *eth_hdr;
1143         struct ipv4_hdr *ipv4_hdr;
1144         uint32_t dst_ipv4;
1145         uint16_t dp;
1146         __m128i te, ve;
1147
1148         eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1149         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1150
1151         dst_ipv4 = ipv4_hdr->dst_addr;
1152         dst_ipv4 = rte_be_to_cpu_32(dst_ipv4);
1153         dp = get_dst_port(qconf, pkt, dst_ipv4, portid);
1154
1155         te = _mm_load_si128((__m128i *)eth_hdr);
1156         ve = val_eth[dp];
1157
1158         dst_port[0] = dp;
1159         rfc1812_process(ipv4_hdr, dst_port, pkt->ol_flags);
1160
1161         te =  _mm_blend_epi16(te, ve, MASK_ETH);
1162         _mm_store_si128((__m128i *)eth_hdr, te);
1163 }
1164
1165 /*
1166  * Read ol_flags and destination IPV4 addresses from 4 mbufs.
1167  */
1168 static inline void
1169 processx4_step1(struct rte_mbuf *pkt[FWDSTEP], __m128i *dip, uint32_t *flag)
1170 {
1171         struct ipv4_hdr *ipv4_hdr;
1172         struct ether_hdr *eth_hdr;
1173         uint32_t x0, x1, x2, x3;
1174
1175         eth_hdr = rte_pktmbuf_mtod(pkt[0], struct ether_hdr *);
1176         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1177         x0 = ipv4_hdr->dst_addr;
1178         flag[0] = pkt[0]->ol_flags & PKT_RX_IPV4_HDR;
1179
1180         eth_hdr = rte_pktmbuf_mtod(pkt[1], struct ether_hdr *);
1181         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1182         x1 = ipv4_hdr->dst_addr;
1183         flag[0] &= pkt[1]->ol_flags;
1184
1185         eth_hdr = rte_pktmbuf_mtod(pkt[2], struct ether_hdr *);
1186         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1187         x2 = ipv4_hdr->dst_addr;
1188         flag[0] &= pkt[2]->ol_flags;
1189
1190         eth_hdr = rte_pktmbuf_mtod(pkt[3], struct ether_hdr *);
1191         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1192         x3 = ipv4_hdr->dst_addr;
1193         flag[0] &= pkt[3]->ol_flags;
1194
1195         dip[0] = _mm_set_epi32(x3, x2, x1, x0);
1196 }
1197
1198 /*
1199  * Lookup into LPM for destination port.
1200  * If lookup fails, use incoming port (portid) as destination port.
1201  */
1202 static inline void
1203 processx4_step2(const struct lcore_conf *qconf, __m128i dip, uint32_t flag,
1204         uint8_t portid, struct rte_mbuf *pkt[FWDSTEP], uint16_t dprt[FWDSTEP])
1205 {
1206         rte_xmm_t dst;
1207         const  __m128i bswap_mask = _mm_set_epi8(12, 13, 14, 15, 8, 9, 10, 11,
1208                                                 4, 5, 6, 7, 0, 1, 2, 3);
1209
1210         /* Byte swap 4 IPV4 addresses. */
1211         dip = _mm_shuffle_epi8(dip, bswap_mask);
1212
1213         /* if all 4 packets are IPV4. */
1214         if (likely(flag != 0)) {
1215                 rte_lpm_lookupx4(qconf->ipv4_lookup_struct, dip, dprt, portid);
1216         } else {
1217                 dst.m = dip;
1218                 dprt[0] = get_dst_port(qconf, pkt[0], dst.u32[0], portid);
1219                 dprt[1] = get_dst_port(qconf, pkt[1], dst.u32[1], portid);
1220                 dprt[2] = get_dst_port(qconf, pkt[2], dst.u32[2], portid);
1221                 dprt[3] = get_dst_port(qconf, pkt[3], dst.u32[3], portid);
1222         }
1223 }
1224
1225 /*
1226  * Update source and destination MAC addresses in the ethernet header.
1227  * Perform RFC1812 checks and updates for IPV4 packets.
1228  */
1229 static inline void
1230 processx4_step3(struct rte_mbuf *pkt[FWDSTEP], uint16_t dst_port[FWDSTEP])
1231 {
1232         __m128i te[FWDSTEP];
1233         __m128i ve[FWDSTEP];
1234         __m128i *p[FWDSTEP];
1235
1236         p[0] = (rte_pktmbuf_mtod(pkt[0], __m128i *));
1237         p[1] = (rte_pktmbuf_mtod(pkt[1], __m128i *));
1238         p[2] = (rte_pktmbuf_mtod(pkt[2], __m128i *));
1239         p[3] = (rte_pktmbuf_mtod(pkt[3], __m128i *));
1240
1241         ve[0] = val_eth[dst_port[0]];
1242         te[0] = _mm_load_si128(p[0]);
1243
1244         ve[1] = val_eth[dst_port[1]];
1245         te[1] = _mm_load_si128(p[1]);
1246
1247         ve[2] = val_eth[dst_port[2]];
1248         te[2] = _mm_load_si128(p[2]);
1249
1250         ve[3] = val_eth[dst_port[3]];
1251         te[3] = _mm_load_si128(p[3]);
1252
1253         /* Update first 12 bytes, keep rest bytes intact. */
1254         te[0] =  _mm_blend_epi16(te[0], ve[0], MASK_ETH);
1255         te[1] =  _mm_blend_epi16(te[1], ve[1], MASK_ETH);
1256         te[2] =  _mm_blend_epi16(te[2], ve[2], MASK_ETH);
1257         te[3] =  _mm_blend_epi16(te[3], ve[3], MASK_ETH);
1258
1259         _mm_store_si128(p[0], te[0]);
1260         _mm_store_si128(p[1], te[1]);
1261         _mm_store_si128(p[2], te[2]);
1262         _mm_store_si128(p[3], te[3]);
1263
1264         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[0] + 1),
1265                 &dst_port[0], pkt[0]->ol_flags);
1266         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[1] + 1),
1267                 &dst_port[1], pkt[1]->ol_flags);
1268         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[2] + 1),
1269                 &dst_port[2], pkt[2]->ol_flags);
1270         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[3] + 1),
1271                 &dst_port[3], pkt[3]->ol_flags);
1272 }
1273
1274 #endif /* APP_LOOKUP_METHOD */
1275
1276 /* main processing loop */
1277 static int
1278 main_loop(__attribute__((unused)) void *dummy)
1279 {
1280         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1281         unsigned lcore_id;
1282         uint64_t prev_tsc, diff_tsc, cur_tsc;
1283         int i, j, nb_rx;
1284         uint8_t portid, queueid;
1285         struct lcore_conf *qconf;
1286         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1287                 US_PER_S * BURST_TX_DRAIN_US;
1288
1289 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1290         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1291         int32_t k;
1292         uint16_t dst_port[MAX_PKT_BURST];
1293         __m128i dip[MAX_PKT_BURST / FWDSTEP];
1294         uint32_t flag[MAX_PKT_BURST / FWDSTEP];
1295 #endif
1296
1297         prev_tsc = 0;
1298
1299         lcore_id = rte_lcore_id();
1300         qconf = &lcore_conf[lcore_id];
1301
1302         if (qconf->n_rx_queue == 0) {
1303                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
1304                 return 0;
1305         }
1306
1307         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
1308
1309         for (i = 0; i < qconf->n_rx_queue; i++) {
1310
1311                 portid = qconf->rx_queue_list[i].port_id;
1312                 queueid = qconf->rx_queue_list[i].queue_id;
1313                 RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n", lcore_id,
1314                         portid, queueid);
1315         }
1316
1317         while (1) {
1318
1319                 cur_tsc = rte_rdtsc();
1320
1321                 /*
1322                  * TX burst queue drain
1323                  */
1324                 diff_tsc = cur_tsc - prev_tsc;
1325                 if (unlikely(diff_tsc > drain_tsc)) {
1326
1327                         /*
1328                          * This could be optimized (use queueid instead of
1329                          * portid), but it is not called so often
1330                          */
1331                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
1332                                 if (qconf->tx_mbufs[portid].len == 0)
1333                                         continue;
1334                                 send_burst(qconf,
1335                                         qconf->tx_mbufs[portid].len,
1336                                         portid);
1337                                 qconf->tx_mbufs[portid].len = 0;
1338                         }
1339
1340                         prev_tsc = cur_tsc;
1341                 }
1342
1343                 /*
1344                  * Read packet from RX queues
1345                  */
1346                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1347                         portid = qconf->rx_queue_list[i].port_id;
1348                         queueid = qconf->rx_queue_list[i].queue_id;
1349                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1350                                 MAX_PKT_BURST);
1351                         if (nb_rx == 0)
1352                                 continue;
1353
1354 #if (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
1355 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1356                         {
1357                                 /*
1358                                  * Send nb_rx - nb_rx%4 packets
1359                                  * in groups of 4.
1360                                  */
1361                                 int32_t n = RTE_ALIGN_FLOOR(nb_rx, 4);
1362                                 for (j = 0; j < n ; j+=4) {
1363                                         uint32_t ol_flag = pkts_burst[j]->ol_flags
1364                                                         & pkts_burst[j+1]->ol_flags
1365                                                         & pkts_burst[j+2]->ol_flags
1366                                                         & pkts_burst[j+3]->ol_flags;
1367                                         if (ol_flag & PKT_RX_IPV4_HDR ) {
1368                                                 simple_ipv4_fwd_4pkts(&pkts_burst[j],
1369                                                                         portid, qconf);
1370                                         } else if (ol_flag & PKT_RX_IPV6_HDR) {
1371                                                 simple_ipv6_fwd_4pkts(&pkts_burst[j],
1372                                                                         portid, qconf);
1373                                         } else {
1374                                                 l3fwd_simple_forward(pkts_burst[j],
1375                                                                         portid, qconf);
1376                                                 l3fwd_simple_forward(pkts_burst[j+1],
1377                                                                         portid, qconf);
1378                                                 l3fwd_simple_forward(pkts_burst[j+2],
1379                                                                         portid, qconf);
1380                                                 l3fwd_simple_forward(pkts_burst[j+3],
1381                                                                         portid, qconf);
1382                                         }
1383                                 }
1384                                 for (; j < nb_rx ; j++) {
1385                                         l3fwd_simple_forward(pkts_burst[j],
1386                                                                 portid, qconf);
1387                                 }
1388                         }
1389 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1390
1391                         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1392                         for (j = 0; j != k; j += FWDSTEP) {
1393                                 processx4_step1(&pkts_burst[j],
1394                                         &dip[j / FWDSTEP],
1395                                         &flag[j / FWDSTEP]);
1396                         }
1397
1398                         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1399                         for (j = 0; j != k; j += FWDSTEP) {
1400                                 processx4_step2(qconf, dip[j / FWDSTEP],
1401                                         flag[j / FWDSTEP], portid,
1402                                         &pkts_burst[j], &dst_port[j]);
1403                         }
1404
1405                         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1406                         for (j = 0; j != k; j += FWDSTEP) {
1407                                 processx4_step3(&pkts_burst[j], &dst_port[j]);
1408                         }
1409
1410                         /* Process up to last 3 packets one by one. */
1411                         switch (nb_rx % FWDSTEP) {
1412                         case 3:
1413                                 process_packet(qconf, pkts_burst[j],
1414                                         dst_port + j, portid);
1415                                 j++;
1416                         case 2:
1417                                 process_packet(qconf, pkts_burst[j],
1418                                         dst_port + j, portid);
1419                                 j++;
1420                         case 1:
1421                                 process_packet(qconf, pkts_burst[j],
1422                                         dst_port + j, portid);
1423                                 j++;
1424                         }
1425
1426                         /*
1427                          * Send packets out, through destination port.
1428                          * Try to group packets with the same destination port.
1429                          * If destination port for the packet equals BAD_PORT,
1430                          * then free the packet without sending it out.
1431                          */
1432                         for (j = 0; j < nb_rx; j = k) {
1433
1434                                 uint16_t cn, pn = dst_port[j];
1435
1436                                 k = j;
1437                                 do {
1438                                         cn = dst_port[k];
1439                                 } while (cn != BAD_PORT && pn == cn &&
1440                                                 ++k < nb_rx);
1441
1442                                 send_packetsx4(qconf, pn, pkts_burst + j,
1443                                         k - j);
1444
1445                                 if (cn == BAD_PORT) {
1446                                         rte_pktmbuf_free(pkts_burst[k]);
1447                                         k += 1;
1448                                 }
1449                         }
1450
1451 #endif /* APP_LOOKUP_METHOD */
1452 #else /* ENABLE_MULTI_BUFFER_OPTIMIZE == 0 */
1453
1454                         /* Prefetch first packets */
1455                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1456                                 rte_prefetch0(rte_pktmbuf_mtod(
1457                                                 pkts_burst[j], void *));
1458                         }
1459
1460                         /* Prefetch and forward already prefetched packets */
1461                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1462                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1463                                                 j + PREFETCH_OFFSET], void *));
1464                                 l3fwd_simple_forward(pkts_burst[j], portid,
1465                                         qconf);
1466                         }
1467
1468                         /* Forward remaining prefetched packets */
1469                         for (; j < nb_rx; j++) {
1470                                 l3fwd_simple_forward(pkts_burst[j], portid,
1471                                         qconf);
1472                         }
1473 #endif /* ENABLE_MULTI_BUFFER_OPTIMIZE */
1474
1475                 }
1476         }
1477 }
1478
1479 static int
1480 check_lcore_params(void)
1481 {
1482         uint8_t queue, lcore;
1483         uint16_t i;
1484         int socketid;
1485
1486         for (i = 0; i < nb_lcore_params; ++i) {
1487                 queue = lcore_params[i].queue_id;
1488                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1489                         printf("invalid queue number: %hhu\n", queue);
1490                         return -1;
1491                 }
1492                 lcore = lcore_params[i].lcore_id;
1493                 if (!rte_lcore_is_enabled(lcore)) {
1494                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
1495                         return -1;
1496                 }
1497                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1498                         (numa_on == 0)) {
1499                         printf("warning: lcore %hhu is on socket %d with numa off \n",
1500                                 lcore, socketid);
1501                 }
1502         }
1503         return 0;
1504 }
1505
1506 static int
1507 check_port_config(const unsigned nb_ports)
1508 {
1509         unsigned portid;
1510         uint16_t i;
1511
1512         for (i = 0; i < nb_lcore_params; ++i) {
1513                 portid = lcore_params[i].port_id;
1514                 if ((enabled_port_mask & (1 << portid)) == 0) {
1515                         printf("port %u is not enabled in port mask\n", portid);
1516                         return -1;
1517                 }
1518                 if (portid >= nb_ports) {
1519                         printf("port %u is not present on the board\n", portid);
1520                         return -1;
1521                 }
1522         }
1523         return 0;
1524 }
1525
1526 static uint8_t
1527 get_port_n_rx_queues(const uint8_t port)
1528 {
1529         int queue = -1;
1530         uint16_t i;
1531
1532         for (i = 0; i < nb_lcore_params; ++i) {
1533                 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
1534                         queue = lcore_params[i].queue_id;
1535         }
1536         return (uint8_t)(++queue);
1537 }
1538
1539 static int
1540 init_lcore_rx_queues(void)
1541 {
1542         uint16_t i, nb_rx_queue;
1543         uint8_t lcore;
1544
1545         for (i = 0; i < nb_lcore_params; ++i) {
1546                 lcore = lcore_params[i].lcore_id;
1547                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1548                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1549                         printf("error: too many queues (%u) for lcore: %u\n",
1550                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1551                         return -1;
1552                 } else {
1553                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1554                                 lcore_params[i].port_id;
1555                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1556                                 lcore_params[i].queue_id;
1557                         lcore_conf[lcore].n_rx_queue++;
1558                 }
1559         }
1560         return 0;
1561 }
1562
1563 /* display usage */
1564 static void
1565 print_usage(const char *prgname)
1566 {
1567         printf ("%s [EAL options] -- -p PORTMASK -P"
1568                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1569                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1570                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1571                 "  -P : enable promiscuous mode\n"
1572                 "  --config (port,queue,lcore): rx queues configuration\n"
1573                 "  --no-numa: optional, disable numa awareness\n"
1574                 "  --ipv6: optional, specify it if running ipv6 packets\n"
1575                 "  --enable-jumbo: enable jumbo frame"
1576                 " which max packet len is PKTLEN in decimal (64-9600)\n"
1577                 "  --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n",
1578                 prgname);
1579 }
1580
1581 static int parse_max_pkt_len(const char *pktlen)
1582 {
1583         char *end = NULL;
1584         unsigned long len;
1585
1586         /* parse decimal string */
1587         len = strtoul(pktlen, &end, 10);
1588         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1589                 return -1;
1590
1591         if (len == 0)
1592                 return -1;
1593
1594         return len;
1595 }
1596
1597 static int
1598 parse_portmask(const char *portmask)
1599 {
1600         char *end = NULL;
1601         unsigned long pm;
1602
1603         /* parse hexadecimal string */
1604         pm = strtoul(portmask, &end, 16);
1605         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1606                 return -1;
1607
1608         if (pm == 0)
1609                 return -1;
1610
1611         return pm;
1612 }
1613
1614 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1615 static int
1616 parse_hash_entry_number(const char *hash_entry_num)
1617 {
1618         char *end = NULL;
1619         unsigned long hash_en;
1620         /* parse hexadecimal string */
1621         hash_en = strtoul(hash_entry_num, &end, 16);
1622         if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
1623                 return -1;
1624
1625         if (hash_en == 0)
1626                 return -1;
1627
1628         return hash_en;
1629 }
1630 #endif
1631
1632 static int
1633 parse_config(const char *q_arg)
1634 {
1635         char s[256];
1636         const char *p, *p0 = q_arg;
1637         char *end;
1638         enum fieldnames {
1639                 FLD_PORT = 0,
1640                 FLD_QUEUE,
1641                 FLD_LCORE,
1642                 _NUM_FLD
1643         };
1644         unsigned long int_fld[_NUM_FLD];
1645         char *str_fld[_NUM_FLD];
1646         int i;
1647         unsigned size;
1648
1649         nb_lcore_params = 0;
1650
1651         while ((p = strchr(p0,'(')) != NULL) {
1652                 ++p;
1653                 if((p0 = strchr(p,')')) == NULL)
1654                         return -1;
1655
1656                 size = p0 - p;
1657                 if(size >= sizeof(s))
1658                         return -1;
1659
1660                 rte_snprintf(s, sizeof(s), "%.*s", size, p);
1661                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
1662                         return -1;
1663                 for (i = 0; i < _NUM_FLD; i++){
1664                         errno = 0;
1665                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1666                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1667                                 return -1;
1668                 }
1669                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1670                         printf("exceeded max number of lcore params: %hu\n",
1671                                 nb_lcore_params);
1672                         return -1;
1673                 }
1674                 lcore_params_array[nb_lcore_params].port_id = (uint8_t)int_fld[FLD_PORT];
1675                 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
1676                 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
1677                 ++nb_lcore_params;
1678         }
1679         lcore_params = lcore_params_array;
1680         return 0;
1681 }
1682
1683 #define CMD_LINE_OPT_CONFIG "config"
1684 #define CMD_LINE_OPT_NO_NUMA "no-numa"
1685 #define CMD_LINE_OPT_IPV6 "ipv6"
1686 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
1687 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
1688
1689 /* Parse the argument given in the command line of the application */
1690 static int
1691 parse_args(int argc, char **argv)
1692 {
1693         int opt, ret;
1694         char **argvopt;
1695         int option_index;
1696         char *prgname = argv[0];
1697         static struct option lgopts[] = {
1698                 {CMD_LINE_OPT_CONFIG, 1, 0, 0},
1699                 {CMD_LINE_OPT_NO_NUMA, 0, 0, 0},
1700                 {CMD_LINE_OPT_IPV6, 0, 0, 0},
1701                 {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, 0},
1702                 {CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, 0},
1703                 {NULL, 0, 0, 0}
1704         };
1705
1706         argvopt = argv;
1707
1708         while ((opt = getopt_long(argc, argvopt, "p:P",
1709                                 lgopts, &option_index)) != EOF) {
1710
1711                 switch (opt) {
1712                 /* portmask */
1713                 case 'p':
1714                         enabled_port_mask = parse_portmask(optarg);
1715                         if (enabled_port_mask == 0) {
1716                                 printf("invalid portmask\n");
1717                                 print_usage(prgname);
1718                                 return -1;
1719                         }
1720                         break;
1721                 case 'P':
1722                         printf("Promiscuous mode selected\n");
1723                         promiscuous_on = 1;
1724                         break;
1725
1726                 /* long options */
1727                 case 0:
1728                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_CONFIG,
1729                                 sizeof (CMD_LINE_OPT_CONFIG))) {
1730                                 ret = parse_config(optarg);
1731                                 if (ret) {
1732                                         printf("invalid config\n");
1733                                         print_usage(prgname);
1734                                         return -1;
1735                                 }
1736                         }
1737
1738                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_NUMA,
1739                                 sizeof(CMD_LINE_OPT_NO_NUMA))) {
1740                                 printf("numa is disabled \n");
1741                                 numa_on = 0;
1742                         }
1743
1744 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1745                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_IPV6,
1746                                 sizeof(CMD_LINE_OPT_IPV6))) {
1747                                 printf("ipv6 is specified \n");
1748                                 ipv6 = 1;
1749                         }
1750 #endif
1751
1752                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ENABLE_JUMBO,
1753                                 sizeof (CMD_LINE_OPT_ENABLE_JUMBO))) {
1754                                 struct option lenopts = {"max-pkt-len", required_argument, 0, 0};
1755
1756                                 printf("jumbo frame is enabled - disabling simple TX path\n");
1757                                 port_conf.rxmode.jumbo_frame = 1;
1758                                 tx_conf.txq_flags = 0;
1759
1760                                 /* if no max-pkt-len set, use the default value ETHER_MAX_LEN */
1761                                 if (0 == getopt_long(argc, argvopt, "", &lenopts, &option_index)) {
1762                                         ret = parse_max_pkt_len(optarg);
1763                                         if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)){
1764                                                 printf("invalid packet length\n");
1765                                                 print_usage(prgname);
1766                                                 return -1;
1767                                         }
1768                                         port_conf.rxmode.max_rx_pkt_len = ret;
1769                                 }
1770                                 printf("set jumbo frame max packet length to %u\n",
1771                                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1772                         }
1773 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1774                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_HASH_ENTRY_NUM,
1775                                 sizeof(CMD_LINE_OPT_HASH_ENTRY_NUM))) {
1776                                 ret = parse_hash_entry_number(optarg);
1777                                 if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
1778                                         hash_entry_number = ret;
1779                                 } else {
1780                                         printf("invalid hash entry number\n");
1781                                         print_usage(prgname);
1782                                         return -1;
1783                                 }
1784                         }
1785 #endif
1786                         break;
1787
1788                 default:
1789                         print_usage(prgname);
1790                         return -1;
1791                 }
1792         }
1793
1794         if (optind >= 0)
1795                 argv[optind-1] = prgname;
1796
1797         ret = optind-1;
1798         optind = 0; /* reset getopt lib */
1799         return ret;
1800 }
1801
1802 static void
1803 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1804 {
1805         printf ("%s%02X:%02X:%02X:%02X:%02X:%02X", name,
1806                 eth_addr->addr_bytes[0],
1807                 eth_addr->addr_bytes[1],
1808                 eth_addr->addr_bytes[2],
1809                 eth_addr->addr_bytes[3],
1810                 eth_addr->addr_bytes[4],
1811                 eth_addr->addr_bytes[5]);
1812 }
1813
1814 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1815
1816 static void convert_ipv4_5tuple(struct ipv4_5tuple* key1,
1817                 union ipv4_5tuple_host* key2)
1818 {
1819         key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
1820         key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
1821         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
1822         key2->port_src = rte_cpu_to_be_16(key1->port_src);
1823         key2->proto = key1->proto;
1824         key2->pad0 = 0;
1825         key2->pad1 = 0;
1826         return;
1827 }
1828
1829 static void convert_ipv6_5tuple(struct ipv6_5tuple* key1,
1830                 union ipv6_5tuple_host* key2)
1831 {
1832         uint32_t i;
1833         for (i = 0; i < 16; i++)
1834         {
1835                 key2->ip_dst[i] = key1->ip_dst[i];
1836                 key2->ip_src[i] = key1->ip_src[i];
1837         }
1838         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
1839         key2->port_src = rte_cpu_to_be_16(key1->port_src);
1840         key2->proto = key1->proto;
1841         key2->pad0 = 0;
1842         key2->pad1 = 0;
1843         key2->reserve = 0;
1844         return;
1845 }
1846
1847 #define BYTE_VALUE_MAX 256
1848 #define ALL_32_BITS 0xffffffff
1849 #define BIT_8_TO_15 0x0000ff00
1850 static inline void
1851 populate_ipv4_few_flow_into_table(const struct rte_hash* h)
1852 {
1853         uint32_t i;
1854         int32_t ret;
1855         uint32_t array_len = sizeof(ipv4_l3fwd_route_array)/sizeof(ipv4_l3fwd_route_array[0]);
1856
1857         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
1858         for (i = 0; i < array_len; i++) {
1859                 struct ipv4_l3fwd_route  entry;
1860                 union ipv4_5tuple_host newkey;
1861                 entry = ipv4_l3fwd_route_array[i];
1862                 convert_ipv4_5tuple(&entry.key, &newkey);
1863                 ret = rte_hash_add_key (h,(void *) &newkey);
1864                 if (ret < 0) {
1865                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
1866                                 " to the l3fwd hash.\n", i);
1867                 }
1868                 ipv4_l3fwd_out_if[ret] = entry.if_out;
1869         }
1870         printf("Hash: Adding 0x%" PRIx32 " keys\n", array_len);
1871 }
1872
1873 #define BIT_16_TO_23 0x00ff0000
1874 static inline void
1875 populate_ipv6_few_flow_into_table(const struct rte_hash* h)
1876 {
1877         uint32_t i;
1878         int32_t ret;
1879         uint32_t array_len = sizeof(ipv6_l3fwd_route_array)/sizeof(ipv6_l3fwd_route_array[0]);
1880
1881         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
1882         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
1883         for (i = 0; i < array_len; i++) {
1884                 struct ipv6_l3fwd_route entry;
1885                 union ipv6_5tuple_host newkey;
1886                 entry = ipv6_l3fwd_route_array[i];
1887                 convert_ipv6_5tuple(&entry.key, &newkey);
1888                 ret = rte_hash_add_key (h, (void *) &newkey);
1889                 if (ret < 0) {
1890                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
1891                                 " to the l3fwd hash.\n", i);
1892                 }
1893                 ipv6_l3fwd_out_if[ret] = entry.if_out;
1894         }
1895         printf("Hash: Adding 0x%" PRIx32 "keys\n", array_len);
1896 }
1897
1898 #define NUMBER_PORT_USED 4
1899 static inline void
1900 populate_ipv4_many_flow_into_table(const struct rte_hash* h,
1901                 unsigned int nr_flow)
1902 {
1903         unsigned i;
1904         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
1905         for (i = 0; i < nr_flow; i++) {
1906                 struct ipv4_l3fwd_route entry;
1907                 union ipv4_5tuple_host newkey;
1908                 uint8_t a = (uint8_t) ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
1909                 uint8_t b = (uint8_t) (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
1910                 uint8_t c = (uint8_t) ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
1911                 /* Create the ipv4 exact match flow */
1912                 memset(&entry, 0, sizeof(entry));
1913                 switch (i & (NUMBER_PORT_USED -1)) {
1914                 case 0:
1915                         entry = ipv4_l3fwd_route_array[0];
1916                         entry.key.ip_dst = IPv4(101,c,b,a);
1917                         break;
1918                 case 1:
1919                         entry = ipv4_l3fwd_route_array[1];
1920                         entry.key.ip_dst = IPv4(201,c,b,a);
1921                         break;
1922                 case 2:
1923                         entry = ipv4_l3fwd_route_array[2];
1924                         entry.key.ip_dst = IPv4(111,c,b,a);
1925                         break;
1926                 case 3:
1927                         entry = ipv4_l3fwd_route_array[3];
1928                         entry.key.ip_dst = IPv4(211,c,b,a);
1929                         break;
1930                 };
1931                 convert_ipv4_5tuple(&entry.key, &newkey);
1932                 int32_t ret = rte_hash_add_key(h,(void *) &newkey);
1933                 if (ret < 0) {
1934                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
1935                 }
1936                 ipv4_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
1937
1938         }
1939         printf("Hash: Adding 0x%x keys\n", nr_flow);
1940 }
1941
1942 static inline void
1943 populate_ipv6_many_flow_into_table(const struct rte_hash* h,
1944                 unsigned int nr_flow)
1945 {
1946         unsigned i;
1947         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
1948         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
1949         for (i = 0; i < nr_flow; i++) {
1950                 struct ipv6_l3fwd_route entry;
1951                 union ipv6_5tuple_host newkey;
1952                 uint8_t a = (uint8_t) ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
1953                 uint8_t b = (uint8_t) (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
1954                 uint8_t c = (uint8_t) ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
1955                 /* Create the ipv6 exact match flow */
1956                 memset(&entry, 0, sizeof(entry));
1957                 switch (i & (NUMBER_PORT_USED - 1)) {
1958                 case 0: entry = ipv6_l3fwd_route_array[0]; break;
1959                 case 1: entry = ipv6_l3fwd_route_array[1]; break;
1960                 case 2: entry = ipv6_l3fwd_route_array[2]; break;
1961                 case 3: entry = ipv6_l3fwd_route_array[3]; break;
1962                 };
1963                 entry.key.ip_dst[13] = c;
1964                 entry.key.ip_dst[14] = b;
1965                 entry.key.ip_dst[15] = a;
1966                 convert_ipv6_5tuple(&entry.key, &newkey);
1967                 int32_t ret = rte_hash_add_key(h,(void *) &newkey);
1968                 if (ret < 0) {
1969                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
1970                 }
1971                 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
1972
1973         }
1974         printf("Hash: Adding 0x%x keys\n", nr_flow);
1975 }
1976
1977 static void
1978 setup_hash(int socketid)
1979 {
1980     struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1981         .name = NULL,
1982         .entries = L3FWD_HASH_ENTRIES,
1983         .bucket_entries = 4,
1984         .key_len = sizeof(union ipv4_5tuple_host),
1985         .hash_func = ipv4_hash_crc,
1986         .hash_func_init_val = 0,
1987     };
1988
1989     struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1990         .name = NULL,
1991         .entries = L3FWD_HASH_ENTRIES,
1992         .bucket_entries = 4,
1993         .key_len = sizeof(union ipv6_5tuple_host),
1994         .hash_func = ipv6_hash_crc,
1995         .hash_func_init_val = 0,
1996     };
1997
1998     char s[64];
1999
2000         /* create ipv4 hash */
2001         rte_snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
2002         ipv4_l3fwd_hash_params.name = s;
2003         ipv4_l3fwd_hash_params.socket_id = socketid;
2004         ipv4_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv4_l3fwd_hash_params);
2005         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2006                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2007                                 "socket %d\n", socketid);
2008
2009         /* create ipv6 hash */
2010         rte_snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
2011         ipv6_l3fwd_hash_params.name = s;
2012         ipv6_l3fwd_hash_params.socket_id = socketid;
2013         ipv6_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv6_l3fwd_hash_params);
2014         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
2015                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2016                                 "socket %d\n", socketid);
2017
2018         if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
2019                 /* For testing hash matching with a large number of flows we
2020                  * generate millions of IP 5-tuples with an incremented dst
2021                  * address to initialize the hash table. */
2022                 if (ipv6 == 0) {
2023                         /* populate the ipv4 hash */
2024                         populate_ipv4_many_flow_into_table(
2025                                 ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
2026                 } else {
2027                         /* populate the ipv6 hash */
2028                         populate_ipv6_many_flow_into_table(
2029                                 ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
2030                 }
2031         } else {
2032                 /* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize the hash table */
2033                 if (ipv6 == 0) {
2034                         /* populate the ipv4 hash */
2035                         populate_ipv4_few_flow_into_table(ipv4_l3fwd_lookup_struct[socketid]);
2036                 } else {
2037                         /* populate the ipv6 hash */
2038                         populate_ipv6_few_flow_into_table(ipv6_l3fwd_lookup_struct[socketid]);
2039                 }
2040         }
2041 }
2042 #endif
2043
2044 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2045 static void
2046 setup_lpm(int socketid)
2047 {
2048         struct rte_lpm6_config config;
2049         unsigned i;
2050         int ret;
2051         char s[64];
2052
2053         /* create the LPM table */
2054         rte_snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
2055         ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
2056                                 IPV4_L3FWD_LPM_MAX_RULES, 0);
2057         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2058                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
2059                                 " on socket %d\n", socketid);
2060
2061         /* populate the LPM table */
2062         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
2063
2064                 /* skip unused ports */
2065                 if ((1 << ipv4_l3fwd_route_array[i].if_out &
2066                                 enabled_port_mask) == 0)
2067                         continue;
2068
2069                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
2070                         ipv4_l3fwd_route_array[i].ip,
2071                         ipv4_l3fwd_route_array[i].depth,
2072                         ipv4_l3fwd_route_array[i].if_out);
2073
2074                 if (ret < 0) {
2075                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
2076                                 "l3fwd LPM table on socket %d\n",
2077                                 i, socketid);
2078                 }
2079
2080                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
2081                         (unsigned)ipv4_l3fwd_route_array[i].ip,
2082                         ipv4_l3fwd_route_array[i].depth,
2083                         ipv4_l3fwd_route_array[i].if_out);
2084         }
2085
2086         /* create the LPM6 table */
2087         rte_snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
2088
2089         config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
2090         config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
2091         config.flags = 0;
2092         ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
2093                                 &config);
2094         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
2095                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
2096                                 " on socket %d\n", socketid);
2097
2098         /* populate the LPM table */
2099         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
2100
2101                 /* skip unused ports */
2102                 if ((1 << ipv6_l3fwd_route_array[i].if_out &
2103                                 enabled_port_mask) == 0)
2104                         continue;
2105
2106                 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
2107                         ipv6_l3fwd_route_array[i].ip,
2108                         ipv6_l3fwd_route_array[i].depth,
2109                         ipv6_l3fwd_route_array[i].if_out);
2110
2111                 if (ret < 0) {
2112                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
2113                                 "l3fwd LPM table on socket %d\n",
2114                                 i, socketid);
2115                 }
2116
2117                 printf("LPM: Adding route %s / %d (%d)\n",
2118                         "IPV6",
2119                         ipv6_l3fwd_route_array[i].depth,
2120                         ipv6_l3fwd_route_array[i].if_out);
2121         }
2122 }
2123 #endif
2124
2125 static int
2126 init_mem(unsigned nb_mbuf)
2127 {
2128         struct lcore_conf *qconf;
2129         int socketid;
2130         unsigned lcore_id;
2131         char s[64];
2132
2133         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2134                 if (rte_lcore_is_enabled(lcore_id) == 0)
2135                         continue;
2136
2137                 if (numa_on)
2138                         socketid = rte_lcore_to_socket_id(lcore_id);
2139                 else
2140                         socketid = 0;
2141
2142                 if (socketid >= NB_SOCKETS) {
2143                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
2144                                 socketid, lcore_id, NB_SOCKETS);
2145                 }
2146                 if (pktmbuf_pool[socketid] == NULL) {
2147                         rte_snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
2148                         pktmbuf_pool[socketid] =
2149                                 rte_mempool_create(s, nb_mbuf, MBUF_SIZE, MEMPOOL_CACHE_SIZE,
2150                                         sizeof(struct rte_pktmbuf_pool_private),
2151                                         rte_pktmbuf_pool_init, NULL,
2152                                         rte_pktmbuf_init, NULL,
2153                                         socketid, 0);
2154                         if (pktmbuf_pool[socketid] == NULL)
2155                                 rte_exit(EXIT_FAILURE,
2156                                                 "Cannot init mbuf pool on socket %d\n", socketid);
2157                         else
2158                                 printf("Allocated mbuf pool on socket %d\n", socketid);
2159
2160 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2161                         setup_lpm(socketid);
2162 #else
2163                         setup_hash(socketid);
2164 #endif
2165                 }
2166                 qconf = &lcore_conf[lcore_id];
2167                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
2168                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
2169         }
2170         return 0;
2171 }
2172
2173 /* Check the link status of all ports in up to 9s, and print them finally */
2174 static void
2175 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
2176 {
2177 #define CHECK_INTERVAL 100 /* 100ms */
2178 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
2179         uint8_t portid, count, all_ports_up, print_flag = 0;
2180         struct rte_eth_link link;
2181
2182         printf("\nChecking link status");
2183         fflush(stdout);
2184         for (count = 0; count <= MAX_CHECK_TIME; count++) {
2185                 all_ports_up = 1;
2186                 for (portid = 0; portid < port_num; portid++) {
2187                         if ((port_mask & (1 << portid)) == 0)
2188                                 continue;
2189                         memset(&link, 0, sizeof(link));
2190                         rte_eth_link_get_nowait(portid, &link);
2191                         /* print link status if flag set */
2192                         if (print_flag == 1) {
2193                                 if (link.link_status)
2194                                         printf("Port %d Link Up - speed %u "
2195                                                 "Mbps - %s\n", (uint8_t)portid,
2196                                                 (unsigned)link.link_speed,
2197                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
2198                                         ("full-duplex") : ("half-duplex\n"));
2199                                 else
2200                                         printf("Port %d Link Down\n",
2201                                                 (uint8_t)portid);
2202                                 continue;
2203                         }
2204                         /* clear all_ports_up flag if any link down */
2205                         if (link.link_status == 0) {
2206                                 all_ports_up = 0;
2207                                 break;
2208                         }
2209                 }
2210                 /* after finally printing all link status, get out */
2211                 if (print_flag == 1)
2212                         break;
2213
2214                 if (all_ports_up == 0) {
2215                         printf(".");
2216                         fflush(stdout);
2217                         rte_delay_ms(CHECK_INTERVAL);
2218                 }
2219
2220                 /* set the print_flag if all ports up or timeout */
2221                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
2222                         print_flag = 1;
2223                         printf("done\n");
2224                 }
2225         }
2226 }
2227
2228 int
2229 MAIN(int argc, char **argv)
2230 {
2231         struct lcore_conf *qconf;
2232         int ret;
2233         unsigned nb_ports;
2234         uint16_t queueid;
2235         unsigned lcore_id;
2236         uint32_t n_tx_queue, nb_lcores;
2237         uint8_t portid, nb_rx_queue, queue, socketid;
2238
2239         /* init EAL */
2240         ret = rte_eal_init(argc, argv);
2241         if (ret < 0)
2242                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2243         argc -= ret;
2244         argv += ret;
2245
2246         /* parse application arguments (after the EAL ones) */
2247         ret = parse_args(argc, argv);
2248         if (ret < 0)
2249                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
2250
2251         if (check_lcore_params() < 0)
2252                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
2253
2254         ret = init_lcore_rx_queues();
2255         if (ret < 0)
2256                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2257
2258
2259         if (rte_eal_pci_probe() < 0)
2260                 rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");
2261
2262         nb_ports = rte_eth_dev_count();
2263         if (nb_ports > RTE_MAX_ETHPORTS)
2264                 nb_ports = RTE_MAX_ETHPORTS;
2265
2266         if (check_port_config(nb_ports) < 0)
2267                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
2268
2269         nb_lcores = rte_lcore_count();
2270
2271         /* initialize all ports */
2272         for (portid = 0; portid < nb_ports; portid++) {
2273                 /* skip ports that are not enabled */
2274                 if ((enabled_port_mask & (1 << portid)) == 0) {
2275                         printf("\nSkipping disabled port %d\n", portid);
2276                         continue;
2277                 }
2278
2279                 /* init port */
2280                 printf("Initializing port %d ... ", portid );
2281                 fflush(stdout);
2282
2283                 nb_rx_queue = get_port_n_rx_queues(portid);
2284                 n_tx_queue = nb_lcores;
2285                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
2286                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
2287                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
2288                         nb_rx_queue, (unsigned)n_tx_queue );
2289                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
2290                                         (uint16_t)n_tx_queue, &port_conf);
2291                 if (ret < 0)
2292                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
2293                                 ret, portid);
2294
2295                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
2296                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
2297                 printf(", ");
2298
2299                 /*
2300                  * prepare dst and src MACs for each port.
2301                  */
2302                 *(uint64_t *)(val_eth + portid) =
2303                         ETHER_LOCAL_ADMIN_ADDR + ((uint64_t)portid << 40);
2304                 ether_addr_copy(&ports_eth_addr[portid],
2305                         (struct ether_addr *)(val_eth + portid) + 1);
2306
2307                 /* init memory */
2308                 ret = init_mem(NB_MBUF);
2309                 if (ret < 0)
2310                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
2311
2312                 /* init one TX queue per couple (lcore,port) */
2313                 queueid = 0;
2314                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2315                         if (rte_lcore_is_enabled(lcore_id) == 0)
2316                                 continue;
2317
2318                         if (numa_on)
2319                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2320                         else
2321                                 socketid = 0;
2322
2323                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
2324                         fflush(stdout);
2325                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
2326                                                      socketid, &tx_conf);
2327                         if (ret < 0)
2328                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
2329                                         "port=%d\n", ret, portid);
2330
2331                         qconf = &lcore_conf[lcore_id];
2332                         qconf->tx_queue_id[portid] = queueid;
2333                         queueid++;
2334                 }
2335                 printf("\n");
2336         }
2337
2338         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2339                 if (rte_lcore_is_enabled(lcore_id) == 0)
2340                         continue;
2341                 qconf = &lcore_conf[lcore_id];
2342                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
2343                 fflush(stdout);
2344                 /* init RX queues */
2345                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
2346                         portid = qconf->rx_queue_list[queue].port_id;
2347                         queueid = qconf->rx_queue_list[queue].queue_id;
2348
2349                         if (numa_on)
2350                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2351                         else
2352                                 socketid = 0;
2353
2354                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
2355                         fflush(stdout);
2356
2357                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
2358                                         socketid, &rx_conf, pktmbuf_pool[socketid]);
2359                         if (ret < 0)
2360                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
2361                                                 "port=%d\n", ret, portid);
2362                 }
2363         }
2364
2365         printf("\n");
2366
2367         /* start ports */
2368         for (portid = 0; portid < nb_ports; portid++) {
2369                 if ((enabled_port_mask & (1 << portid)) == 0) {
2370                         continue;
2371                 }
2372                 /* Start device */
2373                 ret = rte_eth_dev_start(portid);
2374                 if (ret < 0)
2375                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
2376                                 ret, portid);
2377
2378                 /*
2379                  * If enabled, put device in promiscuous mode.
2380                  * This allows IO forwarding mode to forward packets
2381                  * to itself through 2 cross-connected  ports of the
2382                  * target machine.
2383                  */
2384                 if (promiscuous_on)
2385                         rte_eth_promiscuous_enable(portid);
2386         }
2387
2388         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
2389
2390         /* launch per-lcore init on every lcore */
2391         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2392         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2393                 if (rte_eal_wait_lcore(lcore_id) < 0)
2394                         return -1;
2395         }
2396
2397         return 0;
2398 }