examples/l3fwd/main.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <stdio.h>
  35 #include <stdlib.h>
  36 #include <stdint.h>
  37 #include <inttypes.h>
  38 #include <sys/types.h>
  39 #include <string.h>
  40 #include <sys/queue.h>
  41 #include <stdarg.h>
  42 #include <errno.h>
  43 #include <getopt.h>
  44
  45 #include <rte_common.h>
  46 #include <rte_vect.h>
  47 #include <rte_byteorder.h>
  48 #include <rte_log.h>
  49 #include <rte_memory.h>
  50 #include <rte_memcpy.h>
  51 #include <rte_memzone.h>
  52 #include <rte_eal.h>
  53 #include <rte_per_lcore.h>
  54 #include <rte_launch.h>
  55 #include <rte_atomic.h>
  56 #include <rte_cycles.h>
  57 #include <rte_prefetch.h>
  58 #include <rte_lcore.h>
  59 #include <rte_per_lcore.h>
  60 #include <rte_branch_prediction.h>
  61 #include <rte_interrupts.h>
  62 #include <rte_pci.h>
  63 #include <rte_random.h>
  64 #include <rte_debug.h>
  65 #include <rte_ether.h>
  66 #include <rte_ethdev.h>
  67 #include <rte_ring.h>
  68 #include <rte_mempool.h>
  69 #include <rte_mbuf.h>
  70 #include <rte_ip.h>
  71 #include <rte_tcp.h>
  72 #include <rte_udp.h>
  73 #include <rte_string_fns.h>
  74
  75 #include <cmdline_parse.h>
  76 #include <cmdline_parse_etheraddr.h>
  77
  78 #define APP_LOOKUP_EXACT_MATCH          0
  79 #define APP_LOOKUP_LPM                  1
  80 #define DO_RFC_1812_CHECKS
  81
  82 #ifndef APP_LOOKUP_METHOD
  83 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  84 #endif
  85
  86 /*
  87  *  When set to zero, simple forwaring path is eanbled.
  88  *  When set to one, optimized forwarding path is enabled.
  89  *  Note that LPM optimisation path uses SSE4.1 instructions.
  90  */
  91 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && !defined(__SSE4_1__))
  92 #define ENABLE_MULTI_BUFFER_OPTIMIZE    0
  93 #else
  94 #define ENABLE_MULTI_BUFFER_OPTIMIZE    1
  95 #endif
  96
  97 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  98 #include <rte_hash.h>
  99 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 100 #include <rte_lpm.h>
 101 #include <rte_lpm6.h>
 102 #else
 103 #error "APP_LOOKUP_METHOD set to incorrect value"
 104 #endif
 105
 106 #ifndef IPv6_BYTES
 107 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 108                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 109 #define IPv6_BYTES(addr) \
 110         addr[0],  addr[1], addr[2],  addr[3], \
 111         addr[4],  addr[5], addr[6],  addr[7], \
 112         addr[8],  addr[9], addr[10], addr[11],\
 113         addr[12], addr[13],addr[14], addr[15]
 114 #endif
 115
 116
 117 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
 118
 119 #define MAX_JUMBO_PKT_LEN  9600
 120
 121 #define IPV6_ADDR_LEN 16
 122
 123 #define MEMPOOL_CACHE_SIZE 256
 124
 125 /*
 126  * This expression is used to calculate the number of mbufs needed depending on user input, taking
 127  *  into account memory for rx and tx hardware rings, cache per lcore and mtable per port per lcore.
 128  *  RTE_MAX is used to ensure that NB_MBUF never goes below a minimum value of 8192
 129  */
 130
 131 #define NB_MBUF RTE_MAX (                                                                                                                                       \
 132                                 (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT +                                                        \
 133                                 nb_ports*nb_lcores*MAX_PKT_BURST +                                                                                      \
 134                                 nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT +                                                          \
 135                                 nb_lcores*MEMPOOL_CACHE_SIZE),                                                                                          \
 136                                 (unsigned)8192)
 137
 138 #define MAX_PKT_BURST     32
 139 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 140
 141 /*
 142  * Try to avoid TX buffering if we have at least MAX_TX_BURST packets to send.
 143  */
 144 #define MAX_TX_BURST    (MAX_PKT_BURST / 2)
 145
 146 #define NB_SOCKETS 8
 147
 148 /* Configure how many packets ahead to prefetch, when reading packets */
 149 #define PREFETCH_OFFSET 3
 150
 151 /* Used to mark destination port as 'invalid'. */
 152 #define BAD_PORT        ((uint16_t)-1)
 153
 154 #define FWDSTEP 4
 155
 156 /*
 157  * Configurable number of RX/TX ring descriptors
 158  */
 159 #define RTE_TEST_RX_DESC_DEFAULT 128
 160 #define RTE_TEST_TX_DESC_DEFAULT 512
 161 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 162 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 163
 164 /* ethernet addresses of ports */
 165 static uint64_t dest_eth_addr[RTE_MAX_ETHPORTS];
 166 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 167
 168 static __m128i val_eth[RTE_MAX_ETHPORTS];
 169
 170 /* replace first 12B of the ethernet header. */
 171 #define MASK_ETH        0x3f
 172
 173 /* mask of enabled ports */
 174 static uint32_t enabled_port_mask = 0;
 175 static int promiscuous_on = 0; /**< Ports set in promiscuous mode off by default. */
 176 static int numa_on = 1; /**< NUMA is enabled by default. */
 177
 178 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 179 static int ipv6 = 0; /**< ipv6 is false by default. */
 180 #endif
 181
 182 struct mbuf_table {
 183         uint16_t len;
 184         struct rte_mbuf *m_table[MAX_PKT_BURST];
 185 };
 186
 187 struct lcore_rx_queue {
 188         uint8_t port_id;
 189         uint8_t queue_id;
 190 } __rte_cache_aligned;
 191
 192 #define MAX_RX_QUEUE_PER_LCORE 16
 193 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 194 #define MAX_RX_QUEUE_PER_PORT 128
 195
 196 #define MAX_LCORE_PARAMS 1024
 197 struct lcore_params {
 198         uint8_t port_id;
 199         uint8_t queue_id;
 200         uint8_t lcore_id;
 201 } __rte_cache_aligned;
 202
 203 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 204 static struct lcore_params lcore_params_array_default[] = {
 205         {0, 0, 2},
 206         {0, 1, 2},
 207         {0, 2, 2},
 208         {1, 0, 2},
 209         {1, 1, 2},
 210         {1, 2, 2},
 211         {2, 0, 2},
 212         {3, 0, 3},
 213         {3, 1, 3},
 214 };
 215
 216 static struct lcore_params * lcore_params = lcore_params_array_default;
 217 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 218                                 sizeof(lcore_params_array_default[0]);
 219
 220 static struct rte_eth_conf port_conf = {
 221         .rxmode = {
 222                 .mq_mode = ETH_MQ_RX_RSS,
 223                 .max_rx_pkt_len = ETHER_MAX_LEN,
 224                 .split_hdr_size = 0,
 225                 .header_split   = 0, /**< Header Split disabled */
 226                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 227                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 228                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 229                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 230         },
 231         .rx_adv_conf = {
 232                 .rss_conf = {
 233                         .rss_key = NULL,
 234                         .rss_hf = ETH_RSS_IP,
 235                 },
 236         },
 237         .txmode = {
 238                 .mq_mode = ETH_MQ_TX_NONE,
 239         },
 240 };
 241
 242 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 243
 244 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 245
 246 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 247 #include <rte_hash_crc.h>
 248 #define DEFAULT_HASH_FUNC       rte_hash_crc
 249 #else
 250 #include <rte_jhash.h>
 251 #define DEFAULT_HASH_FUNC       rte_jhash
 252 #endif
 253
 254 struct ipv4_5tuple {
 255         uint32_t ip_dst;
 256         uint32_t ip_src;
 257         uint16_t port_dst;
 258         uint16_t port_src;
 259         uint8_t  proto;
 260 } __attribute__((__packed__));
 261
 262 union ipv4_5tuple_host {
 263         struct {
 264                 uint8_t  pad0;
 265                 uint8_t  proto;
 266                 uint16_t pad1;
 267                 uint32_t ip_src;
 268                 uint32_t ip_dst;
 269                 uint16_t port_src;
 270                 uint16_t port_dst;
 271         };
 272         __m128i xmm;
 273 };
 274
 275 #define XMM_NUM_IN_IPV6_5TUPLE 3
 276
 277 struct ipv6_5tuple {
 278         uint8_t  ip_dst[IPV6_ADDR_LEN];
 279         uint8_t  ip_src[IPV6_ADDR_LEN];
 280         uint16_t port_dst;
 281         uint16_t port_src;
 282         uint8_t  proto;
 283 } __attribute__((__packed__));
 284
 285 union ipv6_5tuple_host {
 286         struct {
 287                 uint16_t pad0;
 288                 uint8_t  proto;
 289                 uint8_t  pad1;
 290                 uint8_t  ip_src[IPV6_ADDR_LEN];
 291                 uint8_t  ip_dst[IPV6_ADDR_LEN];
 292                 uint16_t port_src;
 293                 uint16_t port_dst;
 294                 uint64_t reserve;
 295         };
 296         __m128i xmm[XMM_NUM_IN_IPV6_5TUPLE];
 297 };
 298
 299 struct ipv4_l3fwd_route {
 300         struct ipv4_5tuple key;
 301         uint8_t if_out;
 302 };
 303
 304 struct ipv6_l3fwd_route {
 305         struct ipv6_5tuple key;
 306         uint8_t if_out;
 307 };
 308
 309 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 310         {{IPv4(101,0,0,0), IPv4(100,10,0,1),  101, 11, IPPROTO_TCP}, 0},
 311         {{IPv4(201,0,0,0), IPv4(200,20,0,1),  102, 12, IPPROTO_TCP}, 1},
 312         {{IPv4(111,0,0,0), IPv4(100,30,0,1),  101, 11, IPPROTO_TCP}, 2},
 313         {{IPv4(211,0,0,0), IPv4(200,40,0,1),  102, 12, IPPROTO_TCP}, 3},
 314 };
 315
 316 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 317         {{
 318         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 319         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 320         101, 11, IPPROTO_TCP}, 0},
 321
 322         {{
 323         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 324         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 325         102, 12, IPPROTO_TCP}, 1},
 326
 327         {{
 328         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 329         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 330         101, 11, IPPROTO_TCP}, 2},
 331
 332         {{
 333         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 334         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 335         102, 12, IPPROTO_TCP}, 3},
 336 };
 337
 338 typedef struct rte_hash lookup_struct_t;
 339 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 340 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 341
 342 #ifdef RTE_ARCH_X86_64
 343 /* default to 4 million hash entries (approx) */
 344 #define L3FWD_HASH_ENTRIES              1024*1024*4
 345 #else
 346 /* 32-bit has less address-space for hugepage memory, limit to 1M entries */
 347 #define L3FWD_HASH_ENTRIES              1024*1024*1
 348 #endif
 349 #define HASH_ENTRY_NUMBER_DEFAULT       4
 350
 351 static uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
 352
 353 static inline uint32_t
 354 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
 355         uint32_t init_val)
 356 {
 357         const union ipv4_5tuple_host *k;
 358         uint32_t t;
 359         const uint32_t *p;
 360
 361         k = data;
 362         t = k->proto;
 363         p = (const uint32_t *)&k->port_src;
 364
 365 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 366         init_val = rte_hash_crc_4byte(t, init_val);
 367         init_val = rte_hash_crc_4byte(k->ip_src, init_val);
 368         init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
 369         init_val = rte_hash_crc_4byte(*p, init_val);
 370 #else /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 371         init_val = rte_jhash_1word(t, init_val);
 372         init_val = rte_jhash_1word(k->ip_src, init_val);
 373         init_val = rte_jhash_1word(k->ip_dst, init_val);
 374         init_val = rte_jhash_1word(*p, init_val);
 375 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 376         return (init_val);
 377 }
 378
 379 static inline uint32_t
 380 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len, uint32_t init_val)
 381 {
 382         const union ipv6_5tuple_host *k;
 383         uint32_t t;
 384         const uint32_t *p;
 385 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 386         const uint32_t  *ip_src0, *ip_src1, *ip_src2, *ip_src3;
 387         const uint32_t  *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
 388 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 389
 390         k = data;
 391         t = k->proto;
 392         p = (const uint32_t *)&k->port_src;
 393
 394 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 395         ip_src0 = (const uint32_t *) k->ip_src;
 396         ip_src1 = (const uint32_t *)(k->ip_src+4);
 397         ip_src2 = (const uint32_t *)(k->ip_src+8);
 398         ip_src3 = (const uint32_t *)(k->ip_src+12);
 399         ip_dst0 = (const uint32_t *) k->ip_dst;
 400         ip_dst1 = (const uint32_t *)(k->ip_dst+4);
 401         ip_dst2 = (const uint32_t *)(k->ip_dst+8);
 402         ip_dst3 = (const uint32_t *)(k->ip_dst+12);
 403         init_val = rte_hash_crc_4byte(t, init_val);
 404         init_val = rte_hash_crc_4byte(*ip_src0, init_val);
 405         init_val = rte_hash_crc_4byte(*ip_src1, init_val);
 406         init_val = rte_hash_crc_4byte(*ip_src2, init_val);
 407         init_val = rte_hash_crc_4byte(*ip_src3, init_val);
 408         init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
 409         init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
 410         init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
 411         init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
 412         init_val = rte_hash_crc_4byte(*p, init_val);
 413 #else /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 414         init_val = rte_jhash_1word(t, init_val);
 415         init_val = rte_jhash(k->ip_src, sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
 416         init_val = rte_jhash(k->ip_dst, sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
 417         init_val = rte_jhash_1word(*p, init_val);
 418 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 419         return (init_val);
 420 }
 421
 422 #define IPV4_L3FWD_NUM_ROUTES \
 423         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 424
 425 #define IPV6_L3FWD_NUM_ROUTES \
 426         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 427
 428 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 429 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 430
 431 #endif
 432
 433 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 434 struct ipv4_l3fwd_route {
 435         uint32_t ip;
 436         uint8_t  depth;
 437         uint8_t  if_out;
 438 };
 439
 440 struct ipv6_l3fwd_route {
 441         uint8_t ip[16];
 442         uint8_t  depth;
 443         uint8_t  if_out;
 444 };
 445
 446 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 447         {IPv4(1,1,1,0), 24, 0},
 448         {IPv4(2,1,1,0), 24, 1},
 449         {IPv4(3,1,1,0), 24, 2},
 450         {IPv4(4,1,1,0), 24, 3},
 451         {IPv4(5,1,1,0), 24, 4},
 452         {IPv4(6,1,1,0), 24, 5},
 453         {IPv4(7,1,1,0), 24, 6},
 454         {IPv4(8,1,1,0), 24, 7},
 455 };
 456
 457 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 458         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
 459         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
 460         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
 461         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
 462         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
 463         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
 464         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
 465         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
 466 };
 467
 468 #define IPV4_L3FWD_NUM_ROUTES \
 469         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 470 #define IPV6_L3FWD_NUM_ROUTES \
 471         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 472
 473 #define IPV4_L3FWD_LPM_MAX_RULES         1024
 474 #define IPV6_L3FWD_LPM_MAX_RULES         1024
 475 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
 476
 477 typedef struct rte_lpm lookup_struct_t;
 478 typedef struct rte_lpm6 lookup6_struct_t;
 479 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 480 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 481 #endif
 482
 483 struct lcore_conf {
 484         uint16_t n_rx_queue;
 485         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 486         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 487         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 488         lookup_struct_t * ipv4_lookup_struct;
 489 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 490         lookup6_struct_t * ipv6_lookup_struct;
 491 #else
 492         lookup_struct_t * ipv6_lookup_struct;
 493 #endif
 494 } __rte_cache_aligned;
 495
 496 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 497
 498 /* Send burst of packets on an output interface */
 499 static inline int
 500 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 501 {
 502         struct rte_mbuf **m_table;
 503         int ret;
 504         uint16_t queueid;
 505
 506         queueid = qconf->tx_queue_id[port];
 507         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 508
 509         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 510         if (unlikely(ret < n)) {
 511                 do {
 512                         rte_pktmbuf_free(m_table[ret]);
 513                 } while (++ret < n);
 514         }
 515
 516         return 0;
 517 }
 518
 519 /* Enqueue a single packet, and send burst if queue is filled */
 520 static inline int
 521 send_single_packet(struct rte_mbuf *m, uint8_t port)
 522 {
 523         uint32_t lcore_id;
 524         uint16_t len;
 525         struct lcore_conf *qconf;
 526
 527         lcore_id = rte_lcore_id();
 528
 529         qconf = &lcore_conf[lcore_id];
 530         len = qconf->tx_mbufs[port].len;
 531         qconf->tx_mbufs[port].m_table[len] = m;
 532         len++;
 533
 534         /* enough pkts to be sent */
 535         if (unlikely(len == MAX_PKT_BURST)) {
 536                 send_burst(qconf, MAX_PKT_BURST, port);
 537                 len = 0;
 538         }
 539
 540         qconf->tx_mbufs[port].len = len;
 541         return 0;
 542 }
 543
 544 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 545 static inline __attribute__((always_inline)) void
 546 send_packetsx4(struct lcore_conf *qconf, uint8_t port,
 547         struct rte_mbuf *m[], uint32_t num)
 548 {
 549         uint32_t len, j, n;
 550
 551         len = qconf->tx_mbufs[port].len;
 552
 553         /*
 554          * If TX buffer for that queue is empty, and we have enough packets,
 555          * then send them straightway.
 556          */
 557         if (num >= MAX_TX_BURST && len == 0) {
 558                 n = rte_eth_tx_burst(port, qconf->tx_queue_id[port], m, num);
 559                 if (unlikely(n < num)) {
 560                         do {
 561                                 rte_pktmbuf_free(m[n]);
 562                         } while (++n < num);
 563                 }
 564                 return;
 565         }
 566
 567         /*
 568          * Put packets into TX buffer for that queue.
 569          */
 570
 571         n = len + num;
 572         n = (n > MAX_PKT_BURST) ? MAX_PKT_BURST - len : num;
 573
 574         j = 0;
 575         switch (n % FWDSTEP) {
 576         while (j < n) {
 577         case 0:
 578                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 579                 j++;
 580         case 3:
 581                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 582                 j++;
 583         case 2:
 584                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 585                 j++;
 586         case 1:
 587                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 588                 j++;
 589         }
 590         }
 591
 592         len += n;
 593
 594         /* enough pkts to be sent */
 595         if (unlikely(len == MAX_PKT_BURST)) {
 596
 597                 send_burst(qconf, MAX_PKT_BURST, port);
 598
 599                 /* copy rest of the packets into the TX buffer. */
 600                 len = num - n;
 601                 j = 0;
 602                 switch (len % FWDSTEP) {
 603                 while (j < len) {
 604                 case 0:
 605                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 606                         j++;
 607                 case 3:
 608                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 609                         j++;
 610                 case 2:
 611                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 612                         j++;
 613                 case 1:
 614                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 615                         j++;
 616                 }
 617                 }
 618         }
 619
 620         qconf->tx_mbufs[port].len = len;
 621 }
 622 #endif /* APP_LOOKUP_LPM */
 623
 624 #ifdef DO_RFC_1812_CHECKS
 625 static inline int
 626 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 627 {
 628         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 629         /*
 630          * 1. The packet length reported by the Link Layer must be large
 631          * enough to hold the minimum length legal IP datagram (20 bytes).
 632          */
 633         if (link_len < sizeof(struct ipv4_hdr))
 634                 return -1;
 635
 636         /* 2. The IP checksum must be correct. */
 637         /* this is checked in H/W */
 638
 639         /*
 640          * 3. The IP version number must be 4. If the version number is not 4
 641          * then the packet may be another version of IP, such as IPng or
 642          * ST-II.
 643          */
 644         if (((pkt->version_ihl) >> 4) != 4)
 645                 return -3;
 646         /*
 647          * 4. The IP header length field must be large enough to hold the
 648          * minimum length legal IP datagram (20 bytes = 5 words).
 649          */
 650         if ((pkt->version_ihl & 0xf) < 5)
 651                 return -4;
 652
 653         /*
 654          * 5. The IP total length field must be large enough to hold the IP
 655          * datagram header, whose length is specified in the IP header length
 656          * field.
 657          */
 658         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 659                 return -5;
 660
 661         return 0;
 662 }
 663 #endif
 664
 665 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 666
 667 static __m128i mask0;
 668 static __m128i mask1;
 669 static __m128i mask2;
 670 static inline uint8_t
 671 get_ipv4_dst_port(void *ipv4_hdr, uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
 672 {
 673         int ret = 0;
 674         union ipv4_5tuple_host key;
 675
 676         ipv4_hdr = (uint8_t *)ipv4_hdr + offsetof(struct ipv4_hdr, time_to_live);
 677         __m128i data = _mm_loadu_si128((__m128i*)(ipv4_hdr));
 678         /* Get 5 tuple: dst port, src port, dst IP address, src IP address and protocol */
 679         key.xmm = _mm_and_si128(data, mask0);
 680         /* Find destination port */
 681         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 682         return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
 683 }
 684
 685 static inline uint8_t
 686 get_ipv6_dst_port(void *ipv6_hdr,  uint8_t portid, lookup_struct_t * ipv6_l3fwd_lookup_struct)
 687 {
 688         int ret = 0;
 689         union ipv6_5tuple_host key;
 690
 691         ipv6_hdr = (uint8_t *)ipv6_hdr + offsetof(struct ipv6_hdr, payload_len);
 692         __m128i data0 = _mm_loadu_si128((__m128i*)(ipv6_hdr));
 693         __m128i data1 = _mm_loadu_si128((__m128i*)(((uint8_t*)ipv6_hdr)+sizeof(__m128i)));
 694         __m128i data2 = _mm_loadu_si128((__m128i*)(((uint8_t*)ipv6_hdr)+sizeof(__m128i)+sizeof(__m128i)));
 695         /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
 696         key.xmm[0] = _mm_and_si128(data0, mask1);
 697         /* Get part of 5 tuple: dst IP address lower 96 bits and src IP address higher 32 bits */
 698         key.xmm[1] = data1;
 699         /* Get part of 5 tuple: dst port and src port and dst IP address higher 32 bits */
 700         key.xmm[2] = _mm_and_si128(data2, mask2);
 701
 702         /* Find destination port */
 703         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 704         return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
 705 }
 706 #endif
 707
 708 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 709
 710 static inline uint8_t
 711 get_ipv4_dst_port(void *ipv4_hdr,  uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
 712 {
 713         uint8_t next_hop;
 714
 715         return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 716                 rte_be_to_cpu_32(((struct ipv4_hdr *)ipv4_hdr)->dst_addr),
 717                 &next_hop) == 0) ? next_hop : portid);
 718 }
 719
 720 static inline uint8_t
 721 get_ipv6_dst_port(void *ipv6_hdr,  uint8_t portid, lookup6_struct_t * ipv6_l3fwd_lookup_struct)
 722 {
 723         uint8_t next_hop;
 724         return (uint8_t) ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
 725                         ((struct ipv6_hdr*)ipv6_hdr)->dst_addr, &next_hop) == 0)?
 726                         next_hop : portid);
 727 }
 728 #endif
 729
 730 static inline void l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
 731         struct lcore_conf *qconf)  __attribute__((unused));
 732
 733 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) && \
 734         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
 735
 736 #define MASK_ALL_PKTS    0xff
 737 #define EXCLUDE_1ST_PKT 0xfe
 738 #define EXCLUDE_2ND_PKT 0xfd
 739 #define EXCLUDE_3RD_PKT 0xfb
 740 #define EXCLUDE_4TH_PKT 0xf7
 741 #define EXCLUDE_5TH_PKT 0xef
 742 #define EXCLUDE_6TH_PKT 0xdf
 743 #define EXCLUDE_7TH_PKT 0xbf
 744 #define EXCLUDE_8TH_PKT 0x7f
 745
 746 static inline void
 747 simple_ipv4_fwd_8pkts(struct rte_mbuf *m[8], uint8_t portid, struct lcore_conf *qconf)
 748 {
 749         struct ether_hdr *eth_hdr[8];
 750         struct ipv4_hdr *ipv4_hdr[8];
 751         uint8_t dst_port[8];
 752         int32_t ret[8];
 753         union ipv4_5tuple_host key[8];
 754         __m128i data[8];
 755
 756         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
 757         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
 758         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
 759         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
 760         eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *);
 761         eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *);
 762         eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *);
 763         eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *);
 764
 765         /* Handle IPv4 headers.*/
 766         ipv4_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv4_hdr *,
 767                                               sizeof(struct ether_hdr));
 768         ipv4_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv4_hdr *,
 769                                               sizeof(struct ether_hdr));
 770         ipv4_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv4_hdr *,
 771                                               sizeof(struct ether_hdr));
 772         ipv4_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv4_hdr *,
 773                                               sizeof(struct ether_hdr));
 774         ipv4_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv4_hdr *,
 775                                               sizeof(struct ether_hdr));
 776         ipv4_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv4_hdr *,
 777                                               sizeof(struct ether_hdr));
 778         ipv4_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv4_hdr *,
 779                                               sizeof(struct ether_hdr));
 780         ipv4_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv4_hdr *,
 781                                               sizeof(struct ether_hdr));
 782
 783 #ifdef DO_RFC_1812_CHECKS
 784         /* Check to make sure the packet is valid (RFC1812) */
 785         uint8_t valid_mask = MASK_ALL_PKTS;
 786         if (is_valid_ipv4_pkt(ipv4_hdr[0], m[0]->pkt_len) < 0) {
 787                 rte_pktmbuf_free(m[0]);
 788                 valid_mask &= EXCLUDE_1ST_PKT;
 789         }
 790         if (is_valid_ipv4_pkt(ipv4_hdr[1], m[1]->pkt_len) < 0) {
 791                 rte_pktmbuf_free(m[1]);
 792                 valid_mask &= EXCLUDE_2ND_PKT;
 793         }
 794         if (is_valid_ipv4_pkt(ipv4_hdr[2], m[2]->pkt_len) < 0) {
 795                 rte_pktmbuf_free(m[2]);
 796                 valid_mask &= EXCLUDE_3RD_PKT;
 797         }
 798         if (is_valid_ipv4_pkt(ipv4_hdr[3], m[3]->pkt_len) < 0) {
 799                 rte_pktmbuf_free(m[3]);
 800                 valid_mask &= EXCLUDE_4TH_PKT;
 801         }
 802         if (is_valid_ipv4_pkt(ipv4_hdr[4], m[4]->pkt_len) < 0) {
 803                 rte_pktmbuf_free(m[4]);
 804                 valid_mask &= EXCLUDE_5TH_PKT;
 805         }
 806         if (is_valid_ipv4_pkt(ipv4_hdr[5], m[5]->pkt_len) < 0) {
 807                 rte_pktmbuf_free(m[5]);
 808                 valid_mask &= EXCLUDE_6TH_PKT;
 809         }
 810         if (is_valid_ipv4_pkt(ipv4_hdr[6], m[6]->pkt_len) < 0) {
 811                 rte_pktmbuf_free(m[6]);
 812                 valid_mask &= EXCLUDE_7TH_PKT;
 813         }
 814         if (is_valid_ipv4_pkt(ipv4_hdr[7], m[7]->pkt_len) < 0) {
 815                 rte_pktmbuf_free(m[7]);
 816                 valid_mask &= EXCLUDE_8TH_PKT;
 817         }
 818         if (unlikely(valid_mask != MASK_ALL_PKTS)) {
 819                 if (valid_mask == 0){
 820                         return;
 821                 } else {
 822                         uint8_t i = 0;
 823                         for (i = 0; i < 8; i++) {
 824                                 if ((0x1 << i) & valid_mask) {
 825                                         l3fwd_simple_forward(m[i], portid, qconf);
 826                                 }
 827                         }
 828                         return;
 829                 }
 830         }
 831 #endif // End of #ifdef DO_RFC_1812_CHECKS
 832
 833         data[0] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[0], __m128i *,
 834                                         sizeof(struct ether_hdr) +
 835                                         offsetof(struct ipv4_hdr, time_to_live)));
 836         data[1] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[1], __m128i *,
 837                                         sizeof(struct ether_hdr) +
 838                                         offsetof(struct ipv4_hdr, time_to_live)));
 839         data[2] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[2], __m128i *,
 840                                         sizeof(struct ether_hdr) +
 841                                         offsetof(struct ipv4_hdr, time_to_live)));
 842         data[3] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[3], __m128i *,
 843                                         sizeof(struct ether_hdr) +
 844                                         offsetof(struct ipv4_hdr, time_to_live)));
 845         data[4] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[4], __m128i *,
 846                                         sizeof(struct ether_hdr) +
 847                                         offsetof(struct ipv4_hdr, time_to_live)));
 848         data[5] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[5], __m128i *,
 849                                         sizeof(struct ether_hdr) +
 850                                         offsetof(struct ipv4_hdr, time_to_live)));
 851         data[6] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[6], __m128i *,
 852                                         sizeof(struct ether_hdr) +
 853                                         offsetof(struct ipv4_hdr, time_to_live)));
 854         data[7] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[7], __m128i *,
 855                                         sizeof(struct ether_hdr) +
 856                                         offsetof(struct ipv4_hdr, time_to_live)));
 857
 858         key[0].xmm = _mm_and_si128(data[0], mask0);
 859         key[1].xmm = _mm_and_si128(data[1], mask0);
 860         key[2].xmm = _mm_and_si128(data[2], mask0);
 861         key[3].xmm = _mm_and_si128(data[3], mask0);
 862         key[4].xmm = _mm_and_si128(data[4], mask0);
 863         key[5].xmm = _mm_and_si128(data[5], mask0);
 864         key[6].xmm = _mm_and_si128(data[6], mask0);
 865         key[7].xmm = _mm_and_si128(data[7], mask0);
 866
 867         const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
 868                                 &key[4], &key[5], &key[6], &key[7]};
 869
 870         rte_hash_lookup_multi(qconf->ipv4_lookup_struct, &key_array[0], 8, ret);
 871         dst_port[0] = (uint8_t) ((ret[0] < 0) ? portid : ipv4_l3fwd_out_if[ret[0]]);
 872         dst_port[1] = (uint8_t) ((ret[1] < 0) ? portid : ipv4_l3fwd_out_if[ret[1]]);
 873         dst_port[2] = (uint8_t) ((ret[2] < 0) ? portid : ipv4_l3fwd_out_if[ret[2]]);
 874         dst_port[3] = (uint8_t) ((ret[3] < 0) ? portid : ipv4_l3fwd_out_if[ret[3]]);
 875         dst_port[4] = (uint8_t) ((ret[4] < 0) ? portid : ipv4_l3fwd_out_if[ret[4]]);
 876         dst_port[5] = (uint8_t) ((ret[5] < 0) ? portid : ipv4_l3fwd_out_if[ret[5]]);
 877         dst_port[6] = (uint8_t) ((ret[6] < 0) ? portid : ipv4_l3fwd_out_if[ret[6]]);
 878         dst_port[7] = (uint8_t) ((ret[7] < 0) ? portid : ipv4_l3fwd_out_if[ret[7]]);
 879
 880         if (dst_port[0] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[0]) == 0)
 881                 dst_port[0] = portid;
 882         if (dst_port[1] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[1]) == 0)
 883                 dst_port[1] = portid;
 884         if (dst_port[2] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[2]) == 0)
 885                 dst_port[2] = portid;
 886         if (dst_port[3] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[3]) == 0)
 887                 dst_port[3] = portid;
 888         if (dst_port[4] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[4]) == 0)
 889                 dst_port[4] = portid;
 890         if (dst_port[5] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[5]) == 0)
 891                 dst_port[5] = portid;
 892         if (dst_port[6] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[6]) == 0)
 893                 dst_port[6] = portid;
 894         if (dst_port[7] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[7]) == 0)
 895                 dst_port[7] = portid;
 896
 897 #ifdef DO_RFC_1812_CHECKS
 898         /* Update time to live and header checksum */
 899         --(ipv4_hdr[0]->time_to_live);
 900         --(ipv4_hdr[1]->time_to_live);
 901         --(ipv4_hdr[2]->time_to_live);
 902         --(ipv4_hdr[3]->time_to_live);
 903         ++(ipv4_hdr[0]->hdr_checksum);
 904         ++(ipv4_hdr[1]->hdr_checksum);
 905         ++(ipv4_hdr[2]->hdr_checksum);
 906         ++(ipv4_hdr[3]->hdr_checksum);
 907         --(ipv4_hdr[4]->time_to_live);
 908         --(ipv4_hdr[5]->time_to_live);
 909         --(ipv4_hdr[6]->time_to_live);
 910         --(ipv4_hdr[7]->time_to_live);
 911         ++(ipv4_hdr[4]->hdr_checksum);
 912         ++(ipv4_hdr[5]->hdr_checksum);
 913         ++(ipv4_hdr[6]->hdr_checksum);
 914         ++(ipv4_hdr[7]->hdr_checksum);
 915 #endif
 916
 917         /* dst addr */
 918         *(uint64_t *)&eth_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
 919         *(uint64_t *)&eth_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
 920         *(uint64_t *)&eth_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
 921         *(uint64_t *)&eth_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
 922         *(uint64_t *)&eth_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
 923         *(uint64_t *)&eth_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
 924         *(uint64_t *)&eth_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
 925         *(uint64_t *)&eth_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
 926
 927         /* src addr */
 928         ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
 929         ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
 930         ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
 931         ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
 932         ether_addr_copy(&ports_eth_addr[dst_port[4]], &eth_hdr[4]->s_addr);
 933         ether_addr_copy(&ports_eth_addr[dst_port[5]], &eth_hdr[5]->s_addr);
 934         ether_addr_copy(&ports_eth_addr[dst_port[6]], &eth_hdr[6]->s_addr);
 935         ether_addr_copy(&ports_eth_addr[dst_port[7]], &eth_hdr[7]->s_addr);
 936
 937         send_single_packet(m[0], (uint8_t)dst_port[0]);
 938         send_single_packet(m[1], (uint8_t)dst_port[1]);
 939         send_single_packet(m[2], (uint8_t)dst_port[2]);
 940         send_single_packet(m[3], (uint8_t)dst_port[3]);
 941         send_single_packet(m[4], (uint8_t)dst_port[4]);
 942         send_single_packet(m[5], (uint8_t)dst_port[5]);
 943         send_single_packet(m[6], (uint8_t)dst_port[6]);
 944         send_single_packet(m[7], (uint8_t)dst_port[7]);
 945
 946 }
 947
 948 static inline void get_ipv6_5tuple(struct rte_mbuf* m0, __m128i mask0, __m128i mask1,
 949                                  union ipv6_5tuple_host * key)
 950 {
 951         __m128i tmpdata0 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0, __m128i *, sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len)));
 952         __m128i tmpdata1 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0, __m128i *, sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i)));
 953         __m128i tmpdata2 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0, __m128i *, sizeof(struct ether_hdr) + offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i) + sizeof(__m128i)));
 954         key->xmm[0] = _mm_and_si128(tmpdata0, mask0);
 955         key->xmm[1] = tmpdata1;
 956         key->xmm[2] = _mm_and_si128(tmpdata2, mask1);
 957         return;
 958 }
 959
 960 static inline void
 961 simple_ipv6_fwd_8pkts(struct rte_mbuf *m[8], uint8_t portid, struct lcore_conf *qconf)
 962 {
 963         struct ether_hdr *eth_hdr[8];
 964         __attribute__((unused)) struct ipv6_hdr *ipv6_hdr[8];
 965         uint8_t dst_port[8];
 966         int32_t ret[8];
 967         union ipv6_5tuple_host key[8];
 968
 969         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
 970         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
 971         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
 972         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
 973         eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *);
 974         eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *);
 975         eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *);
 976         eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *);
 977
 978         /* Handle IPv6 headers.*/
 979         ipv6_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv6_hdr *,
 980                                               sizeof(struct ether_hdr));
 981         ipv6_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv6_hdr *,
 982                                               sizeof(struct ether_hdr));
 983         ipv6_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv6_hdr *,
 984                                               sizeof(struct ether_hdr));
 985         ipv6_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv6_hdr *,
 986                                               sizeof(struct ether_hdr));
 987         ipv6_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv6_hdr *,
 988                                               sizeof(struct ether_hdr));
 989         ipv6_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv6_hdr *,
 990                                               sizeof(struct ether_hdr));
 991         ipv6_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv6_hdr *,
 992                                               sizeof(struct ether_hdr));
 993         ipv6_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv6_hdr *,
 994                                               sizeof(struct ether_hdr));
 995
 996         get_ipv6_5tuple(m[0], mask1, mask2, &key[0]);
 997         get_ipv6_5tuple(m[1], mask1, mask2, &key[1]);
 998         get_ipv6_5tuple(m[2], mask1, mask2, &key[2]);
 999         get_ipv6_5tuple(m[3], mask1, mask2, &key[3]);
1000         get_ipv6_5tuple(m[4], mask1, mask2, &key[4]);
1001         get_ipv6_5tuple(m[5], mask1, mask2, &key[5]);
1002         get_ipv6_5tuple(m[6], mask1, mask2, &key[6]);
1003         get_ipv6_5tuple(m[7], mask1, mask2, &key[7]);
1004
1005         const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1006                                 &key[4], &key[5], &key[6], &key[7]};
1007
1008         rte_hash_lookup_multi(qconf->ipv6_lookup_struct, &key_array[0], 4, ret);
1009         dst_port[0] = (uint8_t) ((ret[0] < 0) ? portid:ipv6_l3fwd_out_if[ret[0]]);
1010         dst_port[1] = (uint8_t) ((ret[1] < 0) ? portid:ipv6_l3fwd_out_if[ret[1]]);
1011         dst_port[2] = (uint8_t) ((ret[2] < 0) ? portid:ipv6_l3fwd_out_if[ret[2]]);
1012         dst_port[3] = (uint8_t) ((ret[3] < 0) ? portid:ipv6_l3fwd_out_if[ret[3]]);
1013         dst_port[4] = (uint8_t) ((ret[4] < 0) ? portid:ipv6_l3fwd_out_if[ret[4]]);
1014         dst_port[5] = (uint8_t) ((ret[5] < 0) ? portid:ipv6_l3fwd_out_if[ret[5]]);
1015         dst_port[6] = (uint8_t) ((ret[6] < 0) ? portid:ipv6_l3fwd_out_if[ret[6]]);
1016         dst_port[7] = (uint8_t) ((ret[7] < 0) ? portid:ipv6_l3fwd_out_if[ret[7]]);
1017
1018         if (dst_port[0] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[0]) == 0)
1019                 dst_port[0] = portid;
1020         if (dst_port[1] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[1]) == 0)
1021                 dst_port[1] = portid;
1022         if (dst_port[2] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[2]) == 0)
1023                 dst_port[2] = portid;
1024         if (dst_port[3] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[3]) == 0)
1025                 dst_port[3] = portid;
1026         if (dst_port[4] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[4]) == 0)
1027                 dst_port[4] = portid;
1028         if (dst_port[5] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[5]) == 0)
1029                 dst_port[5] = portid;
1030         if (dst_port[6] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[6]) == 0)
1031                 dst_port[6] = portid;
1032         if (dst_port[7] >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port[7]) == 0)
1033                 dst_port[7] = portid;
1034
1035         /* dst addr */
1036         *(uint64_t *)&eth_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1037         *(uint64_t *)&eth_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1038         *(uint64_t *)&eth_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1039         *(uint64_t *)&eth_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1040         *(uint64_t *)&eth_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1041         *(uint64_t *)&eth_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1042         *(uint64_t *)&eth_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1043         *(uint64_t *)&eth_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1044
1045         /* src addr */
1046         ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
1047         ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
1048         ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
1049         ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
1050         ether_addr_copy(&ports_eth_addr[dst_port[4]], &eth_hdr[4]->s_addr);
1051         ether_addr_copy(&ports_eth_addr[dst_port[5]], &eth_hdr[5]->s_addr);
1052         ether_addr_copy(&ports_eth_addr[dst_port[6]], &eth_hdr[6]->s_addr);
1053         ether_addr_copy(&ports_eth_addr[dst_port[7]], &eth_hdr[7]->s_addr);
1054
1055         send_single_packet(m[0], (uint8_t)dst_port[0]);
1056         send_single_packet(m[1], (uint8_t)dst_port[1]);
1057         send_single_packet(m[2], (uint8_t)dst_port[2]);
1058         send_single_packet(m[3], (uint8_t)dst_port[3]);
1059         send_single_packet(m[4], (uint8_t)dst_port[4]);
1060         send_single_packet(m[5], (uint8_t)dst_port[5]);
1061         send_single_packet(m[6], (uint8_t)dst_port[6]);
1062         send_single_packet(m[7], (uint8_t)dst_port[7]);
1063
1064 }
1065 #endif /* APP_LOOKUP_METHOD */
1066
1067 static inline __attribute__((always_inline)) void
1068 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, struct lcore_conf *qconf)
1069 {
1070         struct ether_hdr *eth_hdr;
1071         struct ipv4_hdr *ipv4_hdr;
1072         uint8_t dst_port;
1073
1074         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
1075
1076 #ifdef RTE_NEXT_ABI
1077         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
1078 #else
1079         if (m->ol_flags & PKT_RX_IPV4_HDR) {
1080 #endif
1081                 /* Handle IPv4 headers.*/
1082                 ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
1083                                                    sizeof(struct ether_hdr));
1084
1085 #ifdef DO_RFC_1812_CHECKS
1086                 /* Check to make sure the packet is valid (RFC1812) */
1087                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
1088                         rte_pktmbuf_free(m);
1089                         return;
1090                 }
1091 #endif
1092
1093                  dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
1094                         qconf->ipv4_lookup_struct);
1095                 if (dst_port >= RTE_MAX_ETHPORTS ||
1096                                 (enabled_port_mask & 1 << dst_port) == 0)
1097                         dst_port = portid;
1098
1099 #ifdef DO_RFC_1812_CHECKS
1100                 /* Update time to live and header checksum */
1101                 --(ipv4_hdr->time_to_live);
1102                 ++(ipv4_hdr->hdr_checksum);
1103 #endif
1104                 /* dst addr */
1105                 *(uint64_t *)&eth_hdr->d_addr = dest_eth_addr[dst_port];
1106
1107                 /* src addr */
1108                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
1109
1110                 send_single_packet(m, dst_port);
1111 #ifdef RTE_NEXT_ABI
1112         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
1113 #else
1114         } else {
1115 #endif
1116                 /* Handle IPv6 headers.*/
1117                 struct ipv6_hdr *ipv6_hdr;
1118
1119                 ipv6_hdr = rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
1120                                                    sizeof(struct ether_hdr));
1121
1122                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid, qconf->ipv6_lookup_struct);
1123
1124                 if (dst_port >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port) == 0)
1125                         dst_port = portid;
1126
1127                 /* dst addr */
1128                 *(uint64_t *)&eth_hdr->d_addr = dest_eth_addr[dst_port];
1129
1130                 /* src addr */
1131                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
1132
1133                 send_single_packet(m, dst_port);
1134 #ifdef RTE_NEXT_ABI
1135         } else
1136                 /* Free the mbuf that contains non-IPV4/IPV6 packet */
1137                 rte_pktmbuf_free(m);
1138 #else
1139         }
1140 #endif
1141 }
1142
1143 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1144         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1145 #ifdef DO_RFC_1812_CHECKS
1146
1147 #define IPV4_MIN_VER_IHL        0x45
1148 #define IPV4_MAX_VER_IHL        0x4f
1149 #define IPV4_MAX_VER_IHL_DIFF   (IPV4_MAX_VER_IHL - IPV4_MIN_VER_IHL)
1150
1151 /* Minimum value of IPV4 total length (20B) in network byte order. */
1152 #define IPV4_MIN_LEN_BE (sizeof(struct ipv4_hdr) << 8)
1153
1154 /*
1155  * From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2:
1156  * - The IP version number must be 4.
1157  * - The IP header length field must be large enough to hold the
1158  *    minimum length legal IP datagram (20 bytes = 5 words).
1159  * - The IP total length field must be large enough to hold the IP
1160  *   datagram header, whose length is specified in the IP header length
1161  *   field.
1162  * If we encounter invalid IPV4 packet, then set destination port for it
1163  * to BAD_PORT value.
1164  */
1165 static inline __attribute__((always_inline)) void
1166 #ifdef RTE_NEXT_ABI
1167 rfc1812_process(struct ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t ptype)
1168 #else
1169 rfc1812_process(struct ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t flags)
1170 #endif
1171 {
1172         uint8_t ihl;
1173
1174 #ifdef RTE_NEXT_ABI
1175         if (RTE_ETH_IS_IPV4_HDR(ptype)) {
1176 #else
1177         if ((flags & PKT_RX_IPV4_HDR) != 0) {
1178 #endif
1179                 ihl = ipv4_hdr->version_ihl - IPV4_MIN_VER_IHL;
1180
1181                 ipv4_hdr->time_to_live--;
1182                 ipv4_hdr->hdr_checksum++;
1183
1184                 if (ihl > IPV4_MAX_VER_IHL_DIFF ||
1185                                 ((uint8_t)ipv4_hdr->total_length == 0 &&
1186                                 ipv4_hdr->total_length < IPV4_MIN_LEN_BE)) {
1187                         dp[0] = BAD_PORT;
1188                 }
1189         }
1190 }
1191
1192 #else
1193 #define rfc1812_process(mb, dp) do { } while (0)
1194 #endif /* DO_RFC_1812_CHECKS */
1195 #endif /* APP_LOOKUP_LPM && ENABLE_MULTI_BUFFER_OPTIMIZE */
1196
1197
1198 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1199         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1200
1201 static inline __attribute__((always_inline)) uint16_t
1202 get_dst_port(const struct lcore_conf *qconf, struct rte_mbuf *pkt,
1203         uint32_t dst_ipv4, uint8_t portid)
1204 {
1205         uint8_t next_hop;
1206         struct ipv6_hdr *ipv6_hdr;
1207         struct ether_hdr *eth_hdr;
1208
1209 #ifdef RTE_NEXT_ABI
1210         if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
1211 #else
1212         if (pkt->ol_flags & PKT_RX_IPV4_HDR) {
1213 #endif
1214                 if (rte_lpm_lookup(qconf->ipv4_lookup_struct, dst_ipv4,
1215                                 &next_hop) != 0)
1216                         next_hop = portid;
1217 #ifdef RTE_NEXT_ABI
1218         } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
1219 #else
1220         } else if (pkt->ol_flags & PKT_RX_IPV6_HDR) {
1221 #endif
1222                 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1223                 ipv6_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
1224                 if (rte_lpm6_lookup(qconf->ipv6_lookup_struct,
1225                                 ipv6_hdr->dst_addr, &next_hop) != 0)
1226                         next_hop = portid;
1227         } else {
1228                 next_hop = portid;
1229         }
1230
1231         return next_hop;
1232 }
1233
1234 static inline void
1235 process_packet(struct lcore_conf *qconf, struct rte_mbuf *pkt,
1236         uint16_t *dst_port, uint8_t portid)
1237 {
1238         struct ether_hdr *eth_hdr;
1239         struct ipv4_hdr *ipv4_hdr;
1240         uint32_t dst_ipv4;
1241         uint16_t dp;
1242         __m128i te, ve;
1243
1244         eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1245         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1246
1247         dst_ipv4 = ipv4_hdr->dst_addr;
1248         dst_ipv4 = rte_be_to_cpu_32(dst_ipv4);
1249         dp = get_dst_port(qconf, pkt, dst_ipv4, portid);
1250
1251         te = _mm_load_si128((__m128i *)eth_hdr);
1252         ve = val_eth[dp];
1253
1254         dst_port[0] = dp;
1255 #ifdef RTE_NEXT_ABI
1256         rfc1812_process(ipv4_hdr, dst_port, pkt->packet_type);
1257 #else
1258         rfc1812_process(ipv4_hdr, dst_port, pkt->ol_flags);
1259 #endif
1260
1261         te =  _mm_blend_epi16(te, ve, MASK_ETH);
1262         _mm_store_si128((__m128i *)eth_hdr, te);
1263 }
1264
1265 #ifdef RTE_NEXT_ABI
1266 /*
1267  * Read packet_type and destination IPV4 addresses from 4 mbufs.
1268  */
1269 static inline void
1270 processx4_step1(struct rte_mbuf *pkt[FWDSTEP],
1271                 __m128i *dip,
1272                 uint32_t *ipv4_flag)
1273 {
1274         struct ipv4_hdr *ipv4_hdr;
1275         struct ether_hdr *eth_hdr;
1276         uint32_t x0, x1, x2, x3;
1277
1278         eth_hdr = rte_pktmbuf_mtod(pkt[0], struct ether_hdr *);
1279         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1280         x0 = ipv4_hdr->dst_addr;
1281         ipv4_flag[0] = pkt[0]->packet_type & RTE_PTYPE_L3_IPV4;
1282
1283         eth_hdr = rte_pktmbuf_mtod(pkt[1], struct ether_hdr *);
1284         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1285         x1 = ipv4_hdr->dst_addr;
1286         ipv4_flag[0] &= pkt[1]->packet_type;
1287
1288         eth_hdr = rte_pktmbuf_mtod(pkt[2], struct ether_hdr *);
1289         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1290         x2 = ipv4_hdr->dst_addr;
1291         ipv4_flag[0] &= pkt[2]->packet_type;
1292
1293         eth_hdr = rte_pktmbuf_mtod(pkt[3], struct ether_hdr *);
1294         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1295         x3 = ipv4_hdr->dst_addr;
1296         ipv4_flag[0] &= pkt[3]->packet_type;
1297
1298         dip[0] = _mm_set_epi32(x3, x2, x1, x0);
1299 }
1300 #else /* RTE_NEXT_ABI */
1301 /*
1302  * Read ol_flags and destination IPV4 addresses from 4 mbufs.
1303  */
1304 static inline void
1305 processx4_step1(struct rte_mbuf *pkt[FWDSTEP], __m128i *dip, uint32_t *flag)
1306 {
1307         struct ipv4_hdr *ipv4_hdr;
1308         struct ether_hdr *eth_hdr;
1309         uint32_t x0, x1, x2, x3;
1310
1311         eth_hdr = rte_pktmbuf_mtod(pkt[0], struct ether_hdr *);
1312         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1313         x0 = ipv4_hdr->dst_addr;
1314         flag[0] = pkt[0]->ol_flags & PKT_RX_IPV4_HDR;
1315
1316         eth_hdr = rte_pktmbuf_mtod(pkt[1], struct ether_hdr *);
1317         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1318         x1 = ipv4_hdr->dst_addr;
1319         flag[0] &= pkt[1]->ol_flags;
1320
1321         eth_hdr = rte_pktmbuf_mtod(pkt[2], struct ether_hdr *);
1322         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1323         x2 = ipv4_hdr->dst_addr;
1324         flag[0] &= pkt[2]->ol_flags;
1325
1326         eth_hdr = rte_pktmbuf_mtod(pkt[3], struct ether_hdr *);
1327         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1328         x3 = ipv4_hdr->dst_addr;
1329         flag[0] &= pkt[3]->ol_flags;
1330
1331         dip[0] = _mm_set_epi32(x3, x2, x1, x0);
1332 }
1333 #endif /* RTE_NEXT_ABI */
1334
1335 /*
1336  * Lookup into LPM for destination port.
1337  * If lookup fails, use incoming port (portid) as destination port.
1338  */
1339 static inline void
1340 #ifdef RTE_NEXT_ABI
1341 processx4_step2(const struct lcore_conf *qconf,
1342                 __m128i dip,
1343                 uint32_t ipv4_flag,
1344                 uint8_t portid,
1345                 struct rte_mbuf *pkt[FWDSTEP],
1346                 uint16_t dprt[FWDSTEP])
1347 #else
1348 processx4_step2(const struct lcore_conf *qconf, __m128i dip, uint32_t flag,
1349         uint8_t portid, struct rte_mbuf *pkt[FWDSTEP], uint16_t dprt[FWDSTEP])
1350 #endif /* RTE_NEXT_ABI */
1351 {
1352         rte_xmm_t dst;
1353         const  __m128i bswap_mask = _mm_set_epi8(12, 13, 14, 15, 8, 9, 10, 11,
1354                                                 4, 5, 6, 7, 0, 1, 2, 3);
1355
1356         /* Byte swap 4 IPV4 addresses. */
1357         dip = _mm_shuffle_epi8(dip, bswap_mask);
1358
1359         /* if all 4 packets are IPV4. */
1360 #ifdef RTE_NEXT_ABI
1361         if (likely(ipv4_flag)) {
1362 #else
1363         if (likely(flag != 0)) {
1364 #endif
1365                 rte_lpm_lookupx4(qconf->ipv4_lookup_struct, dip, dprt, portid);
1366         } else {
1367                 dst.x = dip;
1368                 dprt[0] = get_dst_port(qconf, pkt[0], dst.u32[0], portid);
1369                 dprt[1] = get_dst_port(qconf, pkt[1], dst.u32[1], portid);
1370                 dprt[2] = get_dst_port(qconf, pkt[2], dst.u32[2], portid);
1371                 dprt[3] = get_dst_port(qconf, pkt[3], dst.u32[3], portid);
1372         }
1373 }
1374
1375 /*
1376  * Update source and destination MAC addresses in the ethernet header.
1377  * Perform RFC1812 checks and updates for IPV4 packets.
1378  */
1379 static inline void
1380 processx4_step3(struct rte_mbuf *pkt[FWDSTEP], uint16_t dst_port[FWDSTEP])
1381 {
1382         __m128i te[FWDSTEP];
1383         __m128i ve[FWDSTEP];
1384         __m128i *p[FWDSTEP];
1385
1386         p[0] = rte_pktmbuf_mtod(pkt[0], __m128i *);
1387         p[1] = rte_pktmbuf_mtod(pkt[1], __m128i *);
1388         p[2] = rte_pktmbuf_mtod(pkt[2], __m128i *);
1389         p[3] = rte_pktmbuf_mtod(pkt[3], __m128i *);
1390
1391         ve[0] = val_eth[dst_port[0]];
1392         te[0] = _mm_load_si128(p[0]);
1393
1394         ve[1] = val_eth[dst_port[1]];
1395         te[1] = _mm_load_si128(p[1]);
1396
1397         ve[2] = val_eth[dst_port[2]];
1398         te[2] = _mm_load_si128(p[2]);
1399
1400         ve[3] = val_eth[dst_port[3]];
1401         te[3] = _mm_load_si128(p[3]);
1402
1403         /* Update first 12 bytes, keep rest bytes intact. */
1404         te[0] =  _mm_blend_epi16(te[0], ve[0], MASK_ETH);
1405         te[1] =  _mm_blend_epi16(te[1], ve[1], MASK_ETH);
1406         te[2] =  _mm_blend_epi16(te[2], ve[2], MASK_ETH);
1407         te[3] =  _mm_blend_epi16(te[3], ve[3], MASK_ETH);
1408
1409         _mm_store_si128(p[0], te[0]);
1410         _mm_store_si128(p[1], te[1]);
1411         _mm_store_si128(p[2], te[2]);
1412         _mm_store_si128(p[3], te[3]);
1413
1414 #ifdef RTE_NEXT_ABI
1415         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[0] + 1),
1416                 &dst_port[0], pkt[0]->packet_type);
1417         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[1] + 1),
1418                 &dst_port[1], pkt[1]->packet_type);
1419         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[2] + 1),
1420                 &dst_port[2], pkt[2]->packet_type);
1421         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[3] + 1),
1422                 &dst_port[3], pkt[3]->packet_type);
1423 #else /* RTE_NEXT_ABI */
1424         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[0] + 1),
1425                 &dst_port[0], pkt[0]->ol_flags);
1426         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[1] + 1),
1427                 &dst_port[1], pkt[1]->ol_flags);
1428         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[2] + 1),
1429                 &dst_port[2], pkt[2]->ol_flags);
1430         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[3] + 1),
1431                 &dst_port[3], pkt[3]->ol_flags);
1432 #endif /* RTE_NEXT_ABI */
1433 }
1434
1435 /*
1436  * We group consecutive packets with the same destionation port into one burst.
1437  * To avoid extra latency this is done together with some other packet
1438  * processing, but after we made a final decision about packet's destination.
1439  * To do this we maintain:
1440  * pnum - array of number of consecutive packets with the same dest port for
1441  * each packet in the input burst.
1442  * lp - pointer to the last updated element in the pnum.
1443  * dlp - dest port value lp corresponds to.
1444  */
1445
1446 #define GRPSZ   (1 << FWDSTEP)
1447 #define GRPMSK  (GRPSZ - 1)
1448
1449 #define GROUP_PORT_STEP(dlp, dcp, lp, pn, idx)  do { \
1450         if (likely((dlp) == (dcp)[(idx)])) {         \
1451                 (lp)[0]++;                           \
1452         } else {                                     \
1453                 (dlp) = (dcp)[idx];                  \
1454                 (lp) = (pn) + (idx);                 \
1455                 (lp)[0] = 1;                         \
1456         }                                            \
1457 } while (0)
1458
1459 /*
1460  * Group consecutive packets with the same destination port in bursts of 4.
1461  * Suppose we have array of destionation ports:
1462  * dst_port[] = {a, b, c, d,, e, ... }
1463  * dp1 should contain: <a, b, c, d>, dp2: <b, c, d, e>.
1464  * We doing 4 comparisions at once and the result is 4 bit mask.
1465  * This mask is used as an index into prebuild array of pnum values.
1466  */
1467 static inline uint16_t *
1468 port_groupx4(uint16_t pn[FWDSTEP + 1], uint16_t *lp, __m128i dp1, __m128i dp2)
1469 {
1470         static const struct {
1471                 uint64_t pnum; /* prebuild 4 values for pnum[]. */
1472                 int32_t  idx;  /* index for new last updated elemnet. */
1473                 uint16_t lpv;  /* add value to the last updated element. */
1474         } gptbl[GRPSZ] = {
1475         {
1476                 /* 0: a != b, b != c, c != d, d != e */
1477                 .pnum = UINT64_C(0x0001000100010001),
1478                 .idx = 4,
1479                 .lpv = 0,
1480         },
1481         {
1482                 /* 1: a == b, b != c, c != d, d != e */
1483                 .pnum = UINT64_C(0x0001000100010002),
1484                 .idx = 4,
1485                 .lpv = 1,
1486         },
1487         {
1488                 /* 2: a != b, b == c, c != d, d != e */
1489                 .pnum = UINT64_C(0x0001000100020001),
1490                 .idx = 4,
1491                 .lpv = 0,
1492         },
1493         {
1494                 /* 3: a == b, b == c, c != d, d != e */
1495                 .pnum = UINT64_C(0x0001000100020003),
1496                 .idx = 4,
1497                 .lpv = 2,
1498         },
1499         {
1500                 /* 4: a != b, b != c, c == d, d != e */
1501                 .pnum = UINT64_C(0x0001000200010001),
1502                 .idx = 4,
1503                 .lpv = 0,
1504         },
1505         {
1506                 /* 5: a == b, b != c, c == d, d != e */
1507                 .pnum = UINT64_C(0x0001000200010002),
1508                 .idx = 4,
1509                 .lpv = 1,
1510         },
1511         {
1512                 /* 6: a != b, b == c, c == d, d != e */
1513                 .pnum = UINT64_C(0x0001000200030001),
1514                 .idx = 4,
1515                 .lpv = 0,
1516         },
1517         {
1518                 /* 7: a == b, b == c, c == d, d != e */
1519                 .pnum = UINT64_C(0x0001000200030004),
1520                 .idx = 4,
1521                 .lpv = 3,
1522         },
1523         {
1524                 /* 8: a != b, b != c, c != d, d == e */
1525                 .pnum = UINT64_C(0x0002000100010001),
1526                 .idx = 3,
1527                 .lpv = 0,
1528         },
1529         {
1530                 /* 9: a == b, b != c, c != d, d == e */
1531                 .pnum = UINT64_C(0x0002000100010002),
1532                 .idx = 3,
1533                 .lpv = 1,
1534         },
1535         {
1536                 /* 0xa: a != b, b == c, c != d, d == e */
1537                 .pnum = UINT64_C(0x0002000100020001),
1538                 .idx = 3,
1539                 .lpv = 0,
1540         },
1541         {
1542                 /* 0xb: a == b, b == c, c != d, d == e */
1543                 .pnum = UINT64_C(0x0002000100020003),
1544                 .idx = 3,
1545                 .lpv = 2,
1546         },
1547         {
1548                 /* 0xc: a != b, b != c, c == d, d == e */
1549                 .pnum = UINT64_C(0x0002000300010001),
1550                 .idx = 2,
1551                 .lpv = 0,
1552         },
1553         {
1554                 /* 0xd: a == b, b != c, c == d, d == e */
1555                 .pnum = UINT64_C(0x0002000300010002),
1556                 .idx = 2,
1557                 .lpv = 1,
1558         },
1559         {
1560                 /* 0xe: a != b, b == c, c == d, d == e */
1561                 .pnum = UINT64_C(0x0002000300040001),
1562                 .idx = 1,
1563                 .lpv = 0,
1564         },
1565         {
1566                 /* 0xf: a == b, b == c, c == d, d == e */
1567                 .pnum = UINT64_C(0x0002000300040005),
1568                 .idx = 0,
1569                 .lpv = 4,
1570         },
1571         };
1572
1573         union {
1574                 uint16_t u16[FWDSTEP + 1];
1575                 uint64_t u64;
1576         } *pnum = (void *)pn;
1577
1578         int32_t v;
1579
1580         dp1 = _mm_cmpeq_epi16(dp1, dp2);
1581         dp1 = _mm_unpacklo_epi16(dp1, dp1);
1582         v = _mm_movemask_ps((__m128)dp1);
1583
1584         /* update last port counter. */
1585         lp[0] += gptbl[v].lpv;
1586
1587         /* if dest port value has changed. */
1588         if (v != GRPMSK) {
1589                 lp = pnum->u16 + gptbl[v].idx;
1590                 lp[0] = 1;
1591                 pnum->u64 = gptbl[v].pnum;
1592         }
1593
1594         return lp;
1595 }
1596
1597 #endif /* APP_LOOKUP_METHOD */
1598
1599 /* main processing loop */
1600 static int
1601 main_loop(__attribute__((unused)) void *dummy)
1602 {
1603         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1604         unsigned lcore_id;
1605         uint64_t prev_tsc, diff_tsc, cur_tsc;
1606         int i, j, nb_rx;
1607         uint8_t portid, queueid;
1608         struct lcore_conf *qconf;
1609         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1610                 US_PER_S * BURST_TX_DRAIN_US;
1611
1612 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1613         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1614         int32_t k;
1615         uint16_t dlp;
1616         uint16_t *lp;
1617         uint16_t dst_port[MAX_PKT_BURST];
1618         __m128i dip[MAX_PKT_BURST / FWDSTEP];
1619 #ifdef RTE_NEXT_ABI
1620         uint32_t ipv4_flag[MAX_PKT_BURST / FWDSTEP];
1621 #else
1622         uint32_t flag[MAX_PKT_BURST / FWDSTEP];
1623 #endif
1624         uint16_t pnum[MAX_PKT_BURST + 1];
1625 #endif
1626
1627         prev_tsc = 0;
1628
1629         lcore_id = rte_lcore_id();
1630         qconf = &lcore_conf[lcore_id];
1631
1632         if (qconf->n_rx_queue == 0) {
1633                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
1634                 return 0;
1635         }
1636
1637         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
1638
1639         for (i = 0; i < qconf->n_rx_queue; i++) {
1640
1641                 portid = qconf->rx_queue_list[i].port_id;
1642                 queueid = qconf->rx_queue_list[i].queue_id;
1643                 RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n", lcore_id,
1644                         portid, queueid);
1645         }
1646
1647         while (1) {
1648
1649                 cur_tsc = rte_rdtsc();
1650
1651                 /*
1652                  * TX burst queue drain
1653                  */
1654                 diff_tsc = cur_tsc - prev_tsc;
1655                 if (unlikely(diff_tsc > drain_tsc)) {
1656
1657                         /*
1658                          * This could be optimized (use queueid instead of
1659                          * portid), but it is not called so often
1660                          */
1661                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
1662                                 if (qconf->tx_mbufs[portid].len == 0)
1663                                         continue;
1664                                 send_burst(qconf,
1665                                         qconf->tx_mbufs[portid].len,
1666                                         portid);
1667                                 qconf->tx_mbufs[portid].len = 0;
1668                         }
1669
1670                         prev_tsc = cur_tsc;
1671                 }
1672
1673                 /*
1674                  * Read packet from RX queues
1675                  */
1676                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1677                         portid = qconf->rx_queue_list[i].port_id;
1678                         queueid = qconf->rx_queue_list[i].queue_id;
1679                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1680                                 MAX_PKT_BURST);
1681                         if (nb_rx == 0)
1682                                 continue;
1683
1684 #if (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
1685 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1686                         {
1687                                 /*
1688                                  * Send nb_rx - nb_rx%8 packets
1689                                  * in groups of 8.
1690                                  */
1691                                 int32_t n = RTE_ALIGN_FLOOR(nb_rx, 8);
1692                                 for (j = 0; j < n; j += 8) {
1693 #ifdef RTE_NEXT_ABI
1694                                         uint32_t pkt_type =
1695                                                 pkts_burst[j]->packet_type &
1696                                                 pkts_burst[j+1]->packet_type &
1697                                                 pkts_burst[j+2]->packet_type &
1698                                                 pkts_burst[j+3]->packet_type &
1699                                                 pkts_burst[j+4]->packet_type &
1700                                                 pkts_burst[j+5]->packet_type &
1701                                                 pkts_burst[j+6]->packet_type &
1702                                                 pkts_burst[j+7]->packet_type;
1703                                         if (pkt_type & RTE_PTYPE_L3_IPV4) {
1704                                                 simple_ipv4_fwd_8pkts(
1705                                                 &pkts_burst[j], portid, qconf);
1706                                         } else if (pkt_type &
1707                                                 RTE_PTYPE_L3_IPV6) {
1708 #else /* RTE_NEXT_ABI */
1709                                         uint32_t ol_flag = pkts_burst[j]->ol_flags
1710                                                         & pkts_burst[j+1]->ol_flags
1711                                                         & pkts_burst[j+2]->ol_flags
1712                                                         & pkts_burst[j+3]->ol_flags
1713                                                         & pkts_burst[j+4]->ol_flags
1714                                                         & pkts_burst[j+5]->ol_flags
1715                                                         & pkts_burst[j+6]->ol_flags
1716                                                         & pkts_burst[j+7]->ol_flags;
1717                                         if (ol_flag & PKT_RX_IPV4_HDR ) {
1718                                                 simple_ipv4_fwd_8pkts(&pkts_burst[j],
1719                                                                         portid, qconf);
1720                                         } else if (ol_flag & PKT_RX_IPV6_HDR) {
1721 #endif /* RTE_NEXT_ABI */
1722                                                 simple_ipv6_fwd_8pkts(&pkts_burst[j],
1723                                                                         portid, qconf);
1724                                         } else {
1725                                                 l3fwd_simple_forward(pkts_burst[j],
1726                                                                         portid, qconf);
1727                                                 l3fwd_simple_forward(pkts_burst[j+1],
1728                                                                         portid, qconf);
1729                                                 l3fwd_simple_forward(pkts_burst[j+2],
1730                                                                         portid, qconf);
1731                                                 l3fwd_simple_forward(pkts_burst[j+3],
1732                                                                         portid, qconf);
1733                                                 l3fwd_simple_forward(pkts_burst[j+4],
1734                                                                         portid, qconf);
1735                                                 l3fwd_simple_forward(pkts_burst[j+5],
1736                                                                         portid, qconf);
1737                                                 l3fwd_simple_forward(pkts_burst[j+6],
1738                                                                         portid, qconf);
1739                                                 l3fwd_simple_forward(pkts_burst[j+7],
1740                                                                         portid, qconf);
1741                                         }
1742                                 }
1743                                 for (; j < nb_rx ; j++) {
1744                                         l3fwd_simple_forward(pkts_burst[j],
1745                                                                 portid, qconf);
1746                                 }
1747                         }
1748 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1749
1750                         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1751                         for (j = 0; j != k; j += FWDSTEP) {
1752                                 processx4_step1(&pkts_burst[j],
1753                                         &dip[j / FWDSTEP],
1754 #ifdef RTE_NEXT_ABI
1755                                         &ipv4_flag[j / FWDSTEP]);
1756 #else
1757                                         &flag[j / FWDSTEP]);
1758 #endif
1759                         }
1760
1761                         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1762                         for (j = 0; j != k; j += FWDSTEP) {
1763                                 processx4_step2(qconf, dip[j / FWDSTEP],
1764 #ifdef RTE_NEXT_ABI
1765                                         ipv4_flag[j / FWDSTEP], portid,
1766 #else
1767                                         flag[j / FWDSTEP], portid,
1768 #endif
1769                                         &pkts_burst[j], &dst_port[j]);
1770                         }
1771
1772                         /*
1773                          * Finish packet processing and group consecutive
1774                          * packets with the same destination port.
1775                          */
1776                         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1777                         if (k != 0) {
1778                                 __m128i dp1, dp2;
1779
1780                                 lp = pnum;
1781                                 lp[0] = 1;
1782
1783                                 processx4_step3(pkts_burst, dst_port);
1784
1785                                 /* dp1: <d[0], d[1], d[2], d[3], ... > */
1786                                 dp1 = _mm_loadu_si128((__m128i *)dst_port);
1787
1788                                 for (j = FWDSTEP; j != k; j += FWDSTEP) {
1789                                         processx4_step3(&pkts_burst[j],
1790                                                 &dst_port[j]);
1791
1792                                         /*
1793                                          * dp2:
1794                                          * <d[j-3], d[j-2], d[j-1], d[j], ... >
1795                                          */
1796                                         dp2 = _mm_loadu_si128((__m128i *)
1797                                                 &dst_port[j - FWDSTEP + 1]);
1798                                         lp  = port_groupx4(&pnum[j - FWDSTEP],
1799                                                 lp, dp1, dp2);
1800
1801                                         /*
1802                                          * dp1:
1803                                          * <d[j], d[j+1], d[j+2], d[j+3], ... >
1804                                          */
1805                                         dp1 = _mm_srli_si128(dp2,
1806                                                 (FWDSTEP - 1) *
1807                                                 sizeof(dst_port[0]));
1808                                 }
1809
1810                                 /*
1811                                  * dp2: <d[j-3], d[j-2], d[j-1], d[j-1], ... >
1812                                  */
1813                                 dp2 = _mm_shufflelo_epi16(dp1, 0xf9);
1814                                 lp  = port_groupx4(&pnum[j - FWDSTEP], lp,
1815                                         dp1, dp2);
1816
1817                                 /*
1818                                  * remove values added by the last repeated
1819                                  * dst port.
1820                                  */
1821                                 lp[0]--;
1822                                 dlp = dst_port[j - 1];
1823                         } else {
1824                                 /* set dlp and lp to the never used values. */
1825                                 dlp = BAD_PORT - 1;
1826                                 lp = pnum + MAX_PKT_BURST;
1827                         }
1828
1829                         /* Process up to last 3 packets one by one. */
1830                         switch (nb_rx % FWDSTEP) {
1831                         case 3:
1832                                 process_packet(qconf, pkts_burst[j],
1833                                         dst_port + j, portid);
1834                                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1835                                 j++;
1836                         case 2:
1837                                 process_packet(qconf, pkts_burst[j],
1838                                         dst_port + j, portid);
1839                                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1840                                 j++;
1841                         case 1:
1842                                 process_packet(qconf, pkts_burst[j],
1843                                         dst_port + j, portid);
1844                                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1845                                 j++;
1846                         }
1847
1848                         /*
1849                          * Send packets out, through destination port.
1850                          * Consecuteve pacekts with the same destination port
1851                          * are already grouped together.
1852                          * If destination port for the packet equals BAD_PORT,
1853                          * then free the packet without sending it out.
1854                          */
1855                         for (j = 0; j < nb_rx; j += k) {
1856
1857                                 int32_t m;
1858                                 uint16_t pn;
1859
1860                                 pn = dst_port[j];
1861                                 k = pnum[j];
1862
1863                                 if (likely(pn != BAD_PORT)) {
1864                                         send_packetsx4(qconf, pn,
1865                                                 pkts_burst + j, k);
1866                                 } else {
1867                                         for (m = j; m != j + k; m++)
1868                                                 rte_pktmbuf_free(pkts_burst[m]);
1869                                 }
1870                         }
1871
1872 #endif /* APP_LOOKUP_METHOD */
1873 #else /* ENABLE_MULTI_BUFFER_OPTIMIZE == 0 */
1874
1875                         /* Prefetch first packets */
1876                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1877                                 rte_prefetch0(rte_pktmbuf_mtod(
1878                                                 pkts_burst[j], void *));
1879                         }
1880
1881                         /* Prefetch and forward already prefetched packets */
1882                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1883                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1884                                                 j + PREFETCH_OFFSET], void *));
1885                                 l3fwd_simple_forward(pkts_burst[j], portid,
1886                                         qconf);
1887                         }
1888
1889                         /* Forward remaining prefetched packets */
1890                         for (; j < nb_rx; j++) {
1891                                 l3fwd_simple_forward(pkts_burst[j], portid,
1892                                         qconf);
1893                         }
1894 #endif /* ENABLE_MULTI_BUFFER_OPTIMIZE */
1895
1896                 }
1897         }
1898 }
1899
1900 static int
1901 check_lcore_params(void)
1902 {
1903         uint8_t queue, lcore;
1904         uint16_t i;
1905         int socketid;
1906
1907         for (i = 0; i < nb_lcore_params; ++i) {
1908                 queue = lcore_params[i].queue_id;
1909                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1910                         printf("invalid queue number: %hhu\n", queue);
1911                         return -1;
1912                 }
1913                 lcore = lcore_params[i].lcore_id;
1914                 if (!rte_lcore_is_enabled(lcore)) {
1915                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
1916                         return -1;
1917                 }
1918                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1919                         (numa_on == 0)) {
1920                         printf("warning: lcore %hhu is on socket %d with numa off \n",
1921                                 lcore, socketid);
1922                 }
1923         }
1924         return 0;
1925 }
1926
1927 static int
1928 check_port_config(const unsigned nb_ports)
1929 {
1930         unsigned portid;
1931         uint16_t i;
1932
1933         for (i = 0; i < nb_lcore_params; ++i) {
1934                 portid = lcore_params[i].port_id;
1935                 if ((enabled_port_mask & (1 << portid)) == 0) {
1936                         printf("port %u is not enabled in port mask\n", portid);
1937                         return -1;
1938                 }
1939                 if (portid >= nb_ports) {
1940                         printf("port %u is not present on the board\n", portid);
1941                         return -1;
1942                 }
1943         }
1944         return 0;
1945 }
1946
1947 static uint8_t
1948 get_port_n_rx_queues(const uint8_t port)
1949 {
1950         int queue = -1;
1951         uint16_t i;
1952
1953         for (i = 0; i < nb_lcore_params; ++i) {
1954                 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
1955                         queue = lcore_params[i].queue_id;
1956         }
1957         return (uint8_t)(++queue);
1958 }
1959
1960 static int
1961 init_lcore_rx_queues(void)
1962 {
1963         uint16_t i, nb_rx_queue;
1964         uint8_t lcore;
1965
1966         for (i = 0; i < nb_lcore_params; ++i) {
1967                 lcore = lcore_params[i].lcore_id;
1968                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1969                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1970                         printf("error: too many queues (%u) for lcore: %u\n",
1971                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1972                         return -1;
1973                 } else {
1974                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1975                                 lcore_params[i].port_id;
1976                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1977                                 lcore_params[i].queue_id;
1978                         lcore_conf[lcore].n_rx_queue++;
1979                 }
1980         }
1981         return 0;
1982 }
1983
1984 /* display usage */
1985 static void
1986 print_usage(const char *prgname)
1987 {
1988         printf ("%s [EAL options] -- -p PORTMASK -P"
1989                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1990                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1991                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1992                 "  -P : enable promiscuous mode\n"
1993                 "  --config (port,queue,lcore): rx queues configuration\n"
1994                 "  --eth-dest=X,MM:MM:MM:MM:MM:MM: optional, ethernet destination for port X\n"
1995                 "  --no-numa: optional, disable numa awareness\n"
1996                 "  --ipv6: optional, specify it if running ipv6 packets\n"
1997                 "  --enable-jumbo: enable jumbo frame"
1998                 " which max packet len is PKTLEN in decimal (64-9600)\n"
1999                 "  --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n",
2000                 prgname);
2001 }
2002
2003 static int parse_max_pkt_len(const char *pktlen)
2004 {
2005         char *end = NULL;
2006         unsigned long len;
2007
2008         /* parse decimal string */
2009         len = strtoul(pktlen, &end, 10);
2010         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
2011                 return -1;
2012
2013         if (len == 0)
2014                 return -1;
2015
2016         return len;
2017 }
2018
2019 static int
2020 parse_portmask(const char *portmask)
2021 {
2022         char *end = NULL;
2023         unsigned long pm;
2024
2025         /* parse hexadecimal string */
2026         pm = strtoul(portmask, &end, 16);
2027         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
2028                 return -1;
2029
2030         if (pm == 0)
2031                 return -1;
2032
2033         return pm;
2034 }
2035
2036 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2037 static int
2038 parse_hash_entry_number(const char *hash_entry_num)
2039 {
2040         char *end = NULL;
2041         unsigned long hash_en;
2042         /* parse hexadecimal string */
2043         hash_en = strtoul(hash_entry_num, &end, 16);
2044         if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
2045                 return -1;
2046
2047         if (hash_en == 0)
2048                 return -1;
2049
2050         return hash_en;
2051 }
2052 #endif
2053
2054 static int
2055 parse_config(const char *q_arg)
2056 {
2057         char s[256];
2058         const char *p, *p0 = q_arg;
2059         char *end;
2060         enum fieldnames {
2061                 FLD_PORT = 0,
2062                 FLD_QUEUE,
2063                 FLD_LCORE,
2064                 _NUM_FLD
2065         };
2066         unsigned long int_fld[_NUM_FLD];
2067         char *str_fld[_NUM_FLD];
2068         int i;
2069         unsigned size;
2070
2071         nb_lcore_params = 0;
2072
2073         while ((p = strchr(p0,'(')) != NULL) {
2074                 ++p;
2075                 if((p0 = strchr(p,')')) == NULL)
2076                         return -1;
2077
2078                 size = p0 - p;
2079                 if(size >= sizeof(s))
2080                         return -1;
2081
2082                 snprintf(s, sizeof(s), "%.*s", size, p);
2083                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2084                         return -1;
2085                 for (i = 0; i < _NUM_FLD; i++){
2086                         errno = 0;
2087                         int_fld[i] = strtoul(str_fld[i], &end, 0);
2088                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2089                                 return -1;
2090                 }
2091                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
2092                         printf("exceeded max number of lcore params: %hu\n",
2093                                 nb_lcore_params);
2094                         return -1;
2095                 }
2096                 lcore_params_array[nb_lcore_params].port_id = (uint8_t)int_fld[FLD_PORT];
2097                 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
2098                 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
2099                 ++nb_lcore_params;
2100         }
2101         lcore_params = lcore_params_array;
2102         return 0;
2103 }
2104
2105 static void
2106 parse_eth_dest(const char *optarg)
2107 {
2108         uint8_t portid;
2109         char *port_end;
2110         uint8_t c, *dest, peer_addr[6];
2111
2112         errno = 0;
2113         portid = strtoul(optarg, &port_end, 10);
2114         if (errno != 0 || port_end == optarg || *port_end++ != ',')
2115                 rte_exit(EXIT_FAILURE,
2116                 "Invalid eth-dest: %s", optarg);
2117         if (portid >= RTE_MAX_ETHPORTS)
2118                 rte_exit(EXIT_FAILURE,
2119                 "eth-dest: port %d >= RTE_MAX_ETHPORTS(%d)\n",
2120                 portid, RTE_MAX_ETHPORTS);
2121
2122         if (cmdline_parse_etheraddr(NULL, port_end,
2123                 &peer_addr, sizeof(peer_addr)) < 0)
2124                 rte_exit(EXIT_FAILURE,
2125                 "Invalid ethernet address: %s\n",
2126                 port_end);
2127         dest = (uint8_t *)&dest_eth_addr[portid];
2128         for (c = 0; c < 6; c++)
2129                 dest[c] = peer_addr[c];
2130         *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
2131 }
2132
2133 #define CMD_LINE_OPT_CONFIG "config"
2134 #define CMD_LINE_OPT_ETH_DEST "eth-dest"
2135 #define CMD_LINE_OPT_NO_NUMA "no-numa"
2136 #define CMD_LINE_OPT_IPV6 "ipv6"
2137 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
2138 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
2139
2140 /* Parse the argument given in the command line of the application */
2141 static int
2142 parse_args(int argc, char **argv)
2143 {
2144         int opt, ret;
2145         char **argvopt;
2146         int option_index;
2147         char *prgname = argv[0];
2148         static struct option lgopts[] = {
2149                 {CMD_LINE_OPT_CONFIG, 1, 0, 0},
2150                 {CMD_LINE_OPT_ETH_DEST, 1, 0, 0},
2151                 {CMD_LINE_OPT_NO_NUMA, 0, 0, 0},
2152                 {CMD_LINE_OPT_IPV6, 0, 0, 0},
2153                 {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, 0},
2154                 {CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, 0},
2155                 {NULL, 0, 0, 0}
2156         };
2157
2158         argvopt = argv;
2159
2160         while ((opt = getopt_long(argc, argvopt, "p:P",
2161                                 lgopts, &option_index)) != EOF) {
2162
2163                 switch (opt) {
2164                 /* portmask */
2165                 case 'p':
2166                         enabled_port_mask = parse_portmask(optarg);
2167                         if (enabled_port_mask == 0) {
2168                                 printf("invalid portmask\n");
2169                                 print_usage(prgname);
2170                                 return -1;
2171                         }
2172                         break;
2173                 case 'P':
2174                         printf("Promiscuous mode selected\n");
2175                         promiscuous_on = 1;
2176                         break;
2177
2178                 /* long options */
2179                 case 0:
2180                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_CONFIG,
2181                                 sizeof (CMD_LINE_OPT_CONFIG))) {
2182                                 ret = parse_config(optarg);
2183                                 if (ret) {
2184                                         printf("invalid config\n");
2185                                         print_usage(prgname);
2186                                         return -1;
2187                                 }
2188                         }
2189
2190                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ETH_DEST,
2191                                 sizeof(CMD_LINE_OPT_CONFIG))) {
2192                                         parse_eth_dest(optarg);
2193                         }
2194
2195                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_NUMA,
2196                                 sizeof(CMD_LINE_OPT_NO_NUMA))) {
2197                                 printf("numa is disabled \n");
2198                                 numa_on = 0;
2199                         }
2200
2201 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2202                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_IPV6,
2203                                 sizeof(CMD_LINE_OPT_IPV6))) {
2204                                 printf("ipv6 is specified \n");
2205                                 ipv6 = 1;
2206                         }
2207 #endif
2208
2209                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ENABLE_JUMBO,
2210                                 sizeof (CMD_LINE_OPT_ENABLE_JUMBO))) {
2211                                 struct option lenopts = {"max-pkt-len", required_argument, 0, 0};
2212
2213                                 printf("jumbo frame is enabled - disabling simple TX path\n");
2214                                 port_conf.rxmode.jumbo_frame = 1;
2215
2216                                 /* if no max-pkt-len set, use the default value ETHER_MAX_LEN */
2217                                 if (0 == getopt_long(argc, argvopt, "", &lenopts, &option_index)) {
2218                                         ret = parse_max_pkt_len(optarg);
2219                                         if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)){
2220                                                 printf("invalid packet length\n");
2221                                                 print_usage(prgname);
2222                                                 return -1;
2223                                         }
2224                                         port_conf.rxmode.max_rx_pkt_len = ret;
2225                                 }
2226                                 printf("set jumbo frame max packet length to %u\n",
2227                                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
2228                         }
2229 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2230                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_HASH_ENTRY_NUM,
2231                                 sizeof(CMD_LINE_OPT_HASH_ENTRY_NUM))) {
2232                                 ret = parse_hash_entry_number(optarg);
2233                                 if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
2234                                         hash_entry_number = ret;
2235                                 } else {
2236                                         printf("invalid hash entry number\n");
2237                                         print_usage(prgname);
2238                                         return -1;
2239                                 }
2240                         }
2241 #endif
2242                         break;
2243
2244                 default:
2245                         print_usage(prgname);
2246                         return -1;
2247                 }
2248         }
2249
2250         if (optind >= 0)
2251                 argv[optind-1] = prgname;
2252
2253         ret = optind-1;
2254         optind = 0; /* reset getopt lib */
2255         return ret;
2256 }
2257
2258 static void
2259 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
2260 {
2261         char buf[ETHER_ADDR_FMT_SIZE];
2262         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
2263         printf("%s%s", name, buf);
2264 }
2265
2266 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2267
2268 static void convert_ipv4_5tuple(struct ipv4_5tuple* key1,
2269                 union ipv4_5tuple_host* key2)
2270 {
2271         key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
2272         key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
2273         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
2274         key2->port_src = rte_cpu_to_be_16(key1->port_src);
2275         key2->proto = key1->proto;
2276         key2->pad0 = 0;
2277         key2->pad1 = 0;
2278         return;
2279 }
2280
2281 static void convert_ipv6_5tuple(struct ipv6_5tuple* key1,
2282                 union ipv6_5tuple_host* key2)
2283 {
2284         uint32_t i;
2285         for (i = 0; i < 16; i++)
2286         {
2287                 key2->ip_dst[i] = key1->ip_dst[i];
2288                 key2->ip_src[i] = key1->ip_src[i];
2289         }
2290         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
2291         key2->port_src = rte_cpu_to_be_16(key1->port_src);
2292         key2->proto = key1->proto;
2293         key2->pad0 = 0;
2294         key2->pad1 = 0;
2295         key2->reserve = 0;
2296         return;
2297 }
2298
2299 #define BYTE_VALUE_MAX 256
2300 #define ALL_32_BITS 0xffffffff
2301 #define BIT_8_TO_15 0x0000ff00
2302 static inline void
2303 populate_ipv4_few_flow_into_table(const struct rte_hash* h)
2304 {
2305         uint32_t i;
2306         int32_t ret;
2307         uint32_t array_len = sizeof(ipv4_l3fwd_route_array)/sizeof(ipv4_l3fwd_route_array[0]);
2308
2309         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
2310         for (i = 0; i < array_len; i++) {
2311                 struct ipv4_l3fwd_route  entry;
2312                 union ipv4_5tuple_host newkey;
2313                 entry = ipv4_l3fwd_route_array[i];
2314                 convert_ipv4_5tuple(&entry.key, &newkey);
2315                 ret = rte_hash_add_key (h,(void *) &newkey);
2316                 if (ret < 0) {
2317                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
2318                                 " to the l3fwd hash.\n", i);
2319                 }
2320                 ipv4_l3fwd_out_if[ret] = entry.if_out;
2321         }
2322         printf("Hash: Adding 0x%" PRIx32 " keys\n", array_len);
2323 }
2324
2325 #define BIT_16_TO_23 0x00ff0000
2326 static inline void
2327 populate_ipv6_few_flow_into_table(const struct rte_hash* h)
2328 {
2329         uint32_t i;
2330         int32_t ret;
2331         uint32_t array_len = sizeof(ipv6_l3fwd_route_array)/sizeof(ipv6_l3fwd_route_array[0]);
2332
2333         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
2334         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
2335         for (i = 0; i < array_len; i++) {
2336                 struct ipv6_l3fwd_route entry;
2337                 union ipv6_5tuple_host newkey;
2338                 entry = ipv6_l3fwd_route_array[i];
2339                 convert_ipv6_5tuple(&entry.key, &newkey);
2340                 ret = rte_hash_add_key (h, (void *) &newkey);
2341                 if (ret < 0) {
2342                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
2343                                 " to the l3fwd hash.\n", i);
2344                 }
2345                 ipv6_l3fwd_out_if[ret] = entry.if_out;
2346         }
2347         printf("Hash: Adding 0x%" PRIx32 "keys\n", array_len);
2348 }
2349
2350 #define NUMBER_PORT_USED 4
2351 static inline void
2352 populate_ipv4_many_flow_into_table(const struct rte_hash* h,
2353                 unsigned int nr_flow)
2354 {
2355         unsigned i;
2356         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
2357         for (i = 0; i < nr_flow; i++) {
2358                 struct ipv4_l3fwd_route entry;
2359                 union ipv4_5tuple_host newkey;
2360                 uint8_t a = (uint8_t) ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
2361                 uint8_t b = (uint8_t) (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
2362                 uint8_t c = (uint8_t) ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
2363                 /* Create the ipv4 exact match flow */
2364                 memset(&entry, 0, sizeof(entry));
2365                 switch (i & (NUMBER_PORT_USED -1)) {
2366                 case 0:
2367                         entry = ipv4_l3fwd_route_array[0];
2368                         entry.key.ip_dst = IPv4(101,c,b,a);
2369                         break;
2370                 case 1:
2371                         entry = ipv4_l3fwd_route_array[1];
2372                         entry.key.ip_dst = IPv4(201,c,b,a);
2373                         break;
2374                 case 2:
2375                         entry = ipv4_l3fwd_route_array[2];
2376                         entry.key.ip_dst = IPv4(111,c,b,a);
2377                         break;
2378                 case 3:
2379                         entry = ipv4_l3fwd_route_array[3];
2380                         entry.key.ip_dst = IPv4(211,c,b,a);
2381                         break;
2382                 };
2383                 convert_ipv4_5tuple(&entry.key, &newkey);
2384                 int32_t ret = rte_hash_add_key(h,(void *) &newkey);
2385                 if (ret < 0) {
2386                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
2387                 }
2388                 ipv4_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
2389
2390         }
2391         printf("Hash: Adding 0x%x keys\n", nr_flow);
2392 }
2393
2394 static inline void
2395 populate_ipv6_many_flow_into_table(const struct rte_hash* h,
2396                 unsigned int nr_flow)
2397 {
2398         unsigned i;
2399         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
2400         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
2401         for (i = 0; i < nr_flow; i++) {
2402                 struct ipv6_l3fwd_route entry;
2403                 union ipv6_5tuple_host newkey;
2404                 uint8_t a = (uint8_t) ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
2405                 uint8_t b = (uint8_t) (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
2406                 uint8_t c = (uint8_t) ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
2407                 /* Create the ipv6 exact match flow */
2408                 memset(&entry, 0, sizeof(entry));
2409                 switch (i & (NUMBER_PORT_USED - 1)) {
2410                 case 0: entry = ipv6_l3fwd_route_array[0]; break;
2411                 case 1: entry = ipv6_l3fwd_route_array[1]; break;
2412                 case 2: entry = ipv6_l3fwd_route_array[2]; break;
2413                 case 3: entry = ipv6_l3fwd_route_array[3]; break;
2414                 };
2415                 entry.key.ip_dst[13] = c;
2416                 entry.key.ip_dst[14] = b;
2417                 entry.key.ip_dst[15] = a;
2418                 convert_ipv6_5tuple(&entry.key, &newkey);
2419                 int32_t ret = rte_hash_add_key(h,(void *) &newkey);
2420                 if (ret < 0) {
2421                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
2422                 }
2423                 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
2424
2425         }
2426         printf("Hash: Adding 0x%x keys\n", nr_flow);
2427 }
2428
2429 static void
2430 setup_hash(int socketid)
2431 {
2432     struct rte_hash_parameters ipv4_l3fwd_hash_params = {
2433         .name = NULL,
2434         .entries = L3FWD_HASH_ENTRIES,
2435         .key_len = sizeof(union ipv4_5tuple_host),
2436         .hash_func = ipv4_hash_crc,
2437         .hash_func_init_val = 0,
2438     };
2439
2440     struct rte_hash_parameters ipv6_l3fwd_hash_params = {
2441         .name = NULL,
2442         .entries = L3FWD_HASH_ENTRIES,
2443         .key_len = sizeof(union ipv6_5tuple_host),
2444         .hash_func = ipv6_hash_crc,
2445         .hash_func_init_val = 0,
2446     };
2447
2448     char s[64];
2449
2450         /* create ipv4 hash */
2451         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
2452         ipv4_l3fwd_hash_params.name = s;
2453         ipv4_l3fwd_hash_params.socket_id = socketid;
2454         ipv4_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv4_l3fwd_hash_params);
2455         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2456                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2457                                 "socket %d\n", socketid);
2458
2459         /* create ipv6 hash */
2460         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
2461         ipv6_l3fwd_hash_params.name = s;
2462         ipv6_l3fwd_hash_params.socket_id = socketid;
2463         ipv6_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv6_l3fwd_hash_params);
2464         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
2465                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2466                                 "socket %d\n", socketid);
2467
2468         if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
2469                 /* For testing hash matching with a large number of flows we
2470                  * generate millions of IP 5-tuples with an incremented dst
2471                  * address to initialize the hash table. */
2472                 if (ipv6 == 0) {
2473                         /* populate the ipv4 hash */
2474                         populate_ipv4_many_flow_into_table(
2475                                 ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
2476                 } else {
2477                         /* populate the ipv6 hash */
2478                         populate_ipv6_many_flow_into_table(
2479                                 ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
2480                 }
2481         } else {
2482                 /* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize the hash table */
2483                 if (ipv6 == 0) {
2484                         /* populate the ipv4 hash */
2485                         populate_ipv4_few_flow_into_table(ipv4_l3fwd_lookup_struct[socketid]);
2486                 } else {
2487                         /* populate the ipv6 hash */
2488                         populate_ipv6_few_flow_into_table(ipv6_l3fwd_lookup_struct[socketid]);
2489                 }
2490         }
2491 }
2492 #endif
2493
2494 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2495 static void
2496 setup_lpm(int socketid)
2497 {
2498         struct rte_lpm6_config config;
2499         unsigned i;
2500         int ret;
2501         char s[64];
2502
2503         /* create the LPM table */
2504         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
2505         ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
2506                                 IPV4_L3FWD_LPM_MAX_RULES, 0);
2507         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2508                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
2509                                 " on socket %d\n", socketid);
2510
2511         /* populate the LPM table */
2512         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
2513
2514                 /* skip unused ports */
2515                 if ((1 << ipv4_l3fwd_route_array[i].if_out &
2516                                 enabled_port_mask) == 0)
2517                         continue;
2518
2519                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
2520                         ipv4_l3fwd_route_array[i].ip,
2521                         ipv4_l3fwd_route_array[i].depth,
2522                         ipv4_l3fwd_route_array[i].if_out);
2523
2524                 if (ret < 0) {
2525                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
2526                                 "l3fwd LPM table on socket %d\n",
2527                                 i, socketid);
2528                 }
2529
2530                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
2531                         (unsigned)ipv4_l3fwd_route_array[i].ip,
2532                         ipv4_l3fwd_route_array[i].depth,
2533                         ipv4_l3fwd_route_array[i].if_out);
2534         }
2535
2536         /* create the LPM6 table */
2537         snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
2538
2539         config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
2540         config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
2541         config.flags = 0;
2542         ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
2543                                 &config);
2544         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
2545                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
2546                                 " on socket %d\n", socketid);
2547
2548         /* populate the LPM table */
2549         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
2550
2551                 /* skip unused ports */
2552                 if ((1 << ipv6_l3fwd_route_array[i].if_out &
2553                                 enabled_port_mask) == 0)
2554                         continue;
2555
2556                 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
2557                         ipv6_l3fwd_route_array[i].ip,
2558                         ipv6_l3fwd_route_array[i].depth,
2559                         ipv6_l3fwd_route_array[i].if_out);
2560
2561                 if (ret < 0) {
2562                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
2563                                 "l3fwd LPM table on socket %d\n",
2564                                 i, socketid);
2565                 }
2566
2567                 printf("LPM: Adding route %s / %d (%d)\n",
2568                         "IPV6",
2569                         ipv6_l3fwd_route_array[i].depth,
2570                         ipv6_l3fwd_route_array[i].if_out);
2571         }
2572 }
2573 #endif
2574
2575 static int
2576 init_mem(unsigned nb_mbuf)
2577 {
2578         struct lcore_conf *qconf;
2579         int socketid;
2580         unsigned lcore_id;
2581         char s[64];
2582
2583         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2584                 if (rte_lcore_is_enabled(lcore_id) == 0)
2585                         continue;
2586
2587                 if (numa_on)
2588                         socketid = rte_lcore_to_socket_id(lcore_id);
2589                 else
2590                         socketid = 0;
2591
2592                 if (socketid >= NB_SOCKETS) {
2593                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
2594                                 socketid, lcore_id, NB_SOCKETS);
2595                 }
2596                 if (pktmbuf_pool[socketid] == NULL) {
2597                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
2598                         pktmbuf_pool[socketid] =
2599                                 rte_pktmbuf_pool_create(s, nb_mbuf,
2600                                         MEMPOOL_CACHE_SIZE, 0,
2601                                         RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
2602                         if (pktmbuf_pool[socketid] == NULL)
2603                                 rte_exit(EXIT_FAILURE,
2604                                                 "Cannot init mbuf pool on socket %d\n", socketid);
2605                         else
2606                                 printf("Allocated mbuf pool on socket %d\n", socketid);
2607
2608 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2609                         setup_lpm(socketid);
2610 #else
2611                         setup_hash(socketid);
2612 #endif
2613                 }
2614                 qconf = &lcore_conf[lcore_id];
2615                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
2616                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
2617         }
2618         return 0;
2619 }
2620
2621 /* Check the link status of all ports in up to 9s, and print them finally */
2622 static void
2623 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
2624 {
2625 #define CHECK_INTERVAL 100 /* 100ms */
2626 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
2627         uint8_t portid, count, all_ports_up, print_flag = 0;
2628         struct rte_eth_link link;
2629
2630         printf("\nChecking link status");
2631         fflush(stdout);
2632         for (count = 0; count <= MAX_CHECK_TIME; count++) {
2633                 all_ports_up = 1;
2634                 for (portid = 0; portid < port_num; portid++) {
2635                         if ((port_mask & (1 << portid)) == 0)
2636                                 continue;
2637                         memset(&link, 0, sizeof(link));
2638                         rte_eth_link_get_nowait(portid, &link);
2639                         /* print link status if flag set */
2640                         if (print_flag == 1) {
2641                                 if (link.link_status)
2642                                         printf("Port %d Link Up - speed %u "
2643                                                 "Mbps - %s\n", (uint8_t)portid,
2644                                                 (unsigned)link.link_speed,
2645                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
2646                                         ("full-duplex") : ("half-duplex\n"));
2647                                 else
2648                                         printf("Port %d Link Down\n",
2649                                                 (uint8_t)portid);
2650                                 continue;
2651                         }
2652                         /* clear all_ports_up flag if any link down */
2653                         if (link.link_status == 0) {
2654                                 all_ports_up = 0;
2655                                 break;
2656                         }
2657                 }
2658                 /* after finally printing all link status, get out */
2659                 if (print_flag == 1)
2660                         break;
2661
2662                 if (all_ports_up == 0) {
2663                         printf(".");
2664                         fflush(stdout);
2665                         rte_delay_ms(CHECK_INTERVAL);
2666                 }
2667
2668                 /* set the print_flag if all ports up or timeout */
2669                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
2670                         print_flag = 1;
2671                         printf("done\n");
2672                 }
2673         }
2674 }
2675
2676 int
2677 main(int argc, char **argv)
2678 {
2679         struct lcore_conf *qconf;
2680         struct rte_eth_dev_info dev_info;
2681         struct rte_eth_txconf *txconf;
2682         int ret;
2683         unsigned nb_ports;
2684         uint16_t queueid;
2685         unsigned lcore_id;
2686         uint32_t n_tx_queue, nb_lcores;
2687         uint8_t portid, nb_rx_queue, queue, socketid;
2688
2689         /* init EAL */
2690         ret = rte_eal_init(argc, argv);
2691         if (ret < 0)
2692                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2693         argc -= ret;
2694         argv += ret;
2695
2696         /* pre-init dst MACs for all ports to 02:00:00:00:00:xx */
2697         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2698                 dest_eth_addr[portid] = ETHER_LOCAL_ADMIN_ADDR + ((uint64_t)portid << 40);
2699                 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
2700         }
2701
2702         /* parse application arguments (after the EAL ones) */
2703         ret = parse_args(argc, argv);
2704         if (ret < 0)
2705                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
2706
2707         if (check_lcore_params() < 0)
2708                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
2709
2710         ret = init_lcore_rx_queues();
2711         if (ret < 0)
2712                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2713
2714         nb_ports = rte_eth_dev_count();
2715         if (nb_ports > RTE_MAX_ETHPORTS)
2716                 nb_ports = RTE_MAX_ETHPORTS;
2717
2718         if (check_port_config(nb_ports) < 0)
2719                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
2720
2721         nb_lcores = rte_lcore_count();
2722
2723         /* initialize all ports */
2724         for (portid = 0; portid < nb_ports; portid++) {
2725                 /* skip ports that are not enabled */
2726                 if ((enabled_port_mask & (1 << portid)) == 0) {
2727                         printf("\nSkipping disabled port %d\n", portid);
2728                         continue;
2729                 }
2730
2731                 /* init port */
2732                 printf("Initializing port %d ... ", portid );
2733                 fflush(stdout);
2734
2735                 nb_rx_queue = get_port_n_rx_queues(portid);
2736                 n_tx_queue = nb_lcores;
2737                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
2738                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
2739                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
2740                         nb_rx_queue, (unsigned)n_tx_queue );
2741                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
2742                                         (uint16_t)n_tx_queue, &port_conf);
2743                 if (ret < 0)
2744                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
2745                                 ret, portid);
2746
2747                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
2748                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
2749                 printf(", ");
2750                 print_ethaddr("Destination:",
2751                         (const struct ether_addr *)&dest_eth_addr[portid]);
2752                 printf(", ");
2753
2754                 /*
2755                  * prepare src MACs for each port.
2756                  */
2757                 ether_addr_copy(&ports_eth_addr[portid],
2758                         (struct ether_addr *)(val_eth + portid) + 1);
2759
2760                 /* init memory */
2761                 ret = init_mem(NB_MBUF);
2762                 if (ret < 0)
2763                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
2764
2765                 /* init one TX queue per couple (lcore,port) */
2766                 queueid = 0;
2767                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2768                         if (rte_lcore_is_enabled(lcore_id) == 0)
2769                                 continue;
2770
2771                         if (numa_on)
2772                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2773                         else
2774                                 socketid = 0;
2775
2776                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
2777                         fflush(stdout);
2778
2779                         rte_eth_dev_info_get(portid, &dev_info);
2780                         txconf = &dev_info.default_txconf;
2781                         if (port_conf.rxmode.jumbo_frame)
2782                                 txconf->txq_flags = 0;
2783                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
2784                                                      socketid, txconf);
2785                         if (ret < 0)
2786                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
2787                                         "port=%d\n", ret, portid);
2788
2789                         qconf = &lcore_conf[lcore_id];
2790                         qconf->tx_queue_id[portid] = queueid;
2791                         queueid++;
2792                 }
2793                 printf("\n");
2794         }
2795
2796         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2797                 if (rte_lcore_is_enabled(lcore_id) == 0)
2798                         continue;
2799                 qconf = &lcore_conf[lcore_id];
2800                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
2801                 fflush(stdout);
2802                 /* init RX queues */
2803                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
2804                         portid = qconf->rx_queue_list[queue].port_id;
2805                         queueid = qconf->rx_queue_list[queue].queue_id;
2806
2807                         if (numa_on)
2808                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2809                         else
2810                                 socketid = 0;
2811
2812                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
2813                         fflush(stdout);
2814
2815                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
2816                                         socketid,
2817                                         NULL,
2818                                         pktmbuf_pool[socketid]);
2819                         if (ret < 0)
2820                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
2821                                                 "port=%d\n", ret, portid);
2822                 }
2823         }
2824
2825         printf("\n");
2826
2827         /* start ports */
2828         for (portid = 0; portid < nb_ports; portid++) {
2829                 if ((enabled_port_mask & (1 << portid)) == 0) {
2830                         continue;
2831                 }
2832                 /* Start device */
2833                 ret = rte_eth_dev_start(portid);
2834                 if (ret < 0)
2835                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
2836                                 ret, portid);
2837
2838                 /*
2839                  * If enabled, put device in promiscuous mode.
2840                  * This allows IO forwarding mode to forward packets
2841                  * to itself through 2 cross-connected  ports of the
2842                  * target machine.
2843                  */
2844                 if (promiscuous_on)
2845                         rte_eth_promiscuous_enable(portid);
2846         }
2847
2848         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
2849
2850         /* launch per-lcore init on every lcore */
2851         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2852         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2853                 if (rte_eal_wait_lcore(lcore_id) < 0)
2854                         return -1;
2855         }
2856
2857         return 0;
2858 }