examples/performance-thread/l3fwd-thread/main.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2016 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 #define _GNU_SOURCE
  35
  36 #include <stdio.h>
  37 #include <stdlib.h>
  38 #include <stdint.h>
  39 #include <inttypes.h>
  40 #include <sys/types.h>
  41 #include <string.h>
  42 #include <sys/queue.h>
  43 #include <stdarg.h>
  44 #include <errno.h>
  45 #include <getopt.h>
  46
  47 #include <rte_common.h>
  48 #include <rte_vect.h>
  49 #include <rte_byteorder.h>
  50 #include <rte_log.h>
  51 #include <rte_memory.h>
  52 #include <rte_memcpy.h>
  53 #include <rte_memzone.h>
  54 #include <rte_eal.h>
  55 #include <rte_launch.h>
  56 #include <rte_atomic.h>
  57 #include <rte_cycles.h>
  58 #include <rte_prefetch.h>
  59 #include <rte_lcore.h>
  60 #include <rte_per_lcore.h>
  61 #include <rte_branch_prediction.h>
  62 #include <rte_interrupts.h>
  63 #include <rte_pci.h>
  64 #include <rte_random.h>
  65 #include <rte_debug.h>
  66 #include <rte_ether.h>
  67 #include <rte_ethdev.h>
  68 #include <rte_ring.h>
  69 #include <rte_mempool.h>
  70 #include <rte_mbuf.h>
  71 #include <rte_ip.h>
  72 #include <rte_tcp.h>
  73 #include <rte_udp.h>
  74 #include <rte_string_fns.h>
  75 #include <rte_pause.h>
  76
  77 #include <cmdline_parse.h>
  78 #include <cmdline_parse_etheraddr.h>
  79
  80 #include <lthread_api.h>
  81
  82 #define APP_LOOKUP_EXACT_MATCH          0
  83 #define APP_LOOKUP_LPM                  1
  84 #define DO_RFC_1812_CHECKS
  85
  86 /* Enable cpu-load stats 0-off, 1-on */
  87 #define APP_CPU_LOAD                 1
  88
  89 #ifndef APP_LOOKUP_METHOD
  90 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  91 #endif
  92
  93 #ifndef __GLIBC__ /* sched_getcpu() is glibc specific */
  94 #define sched_getcpu() rte_lcore_id()
  95 #endif
  96
  97 static int
  98 check_ptype(int portid)
  99 {
 100         int i, ret;
 101         int ipv4 = 0, ipv6 = 0;
 102
 103         ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK, NULL,
 104                         0);
 105         if (ret <= 0)
 106                 return 0;
 107
 108         uint32_t ptypes[ret];
 109
 110         ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK,
 111                         ptypes, ret);
 112         for (i = 0; i < ret; ++i) {
 113                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
 114                         ipv4 = 1;
 115                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
 116                         ipv6 = 1;
 117         }
 118
 119         if (ipv4 && ipv6)
 120                 return 1;
 121
 122         return 0;
 123 }
 124
 125 static inline void
 126 parse_ptype(struct rte_mbuf *m)
 127 {
 128         struct ether_hdr *eth_hdr;
 129         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 130         uint16_t ether_type;
 131
 132         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 133         ether_type = eth_hdr->ether_type;
 134         if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
 135                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 136         else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
 137                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 138
 139         m->packet_type = packet_type;
 140 }
 141
 142 static uint16_t
 143 cb_parse_ptype(__rte_unused uint16_t port, __rte_unused uint16_t queue,
 144                 struct rte_mbuf *pkts[], uint16_t nb_pkts,
 145                 __rte_unused uint16_t max_pkts, __rte_unused void *user_param)
 146 {
 147         unsigned int i;
 148
 149         for (i = 0; i < nb_pkts; i++)
 150                 parse_ptype(pkts[i]);
 151
 152         return nb_pkts;
 153 }
 154
 155 /*
 156  *  When set to zero, simple forwaring path is eanbled.
 157  *  When set to one, optimized forwarding path is enabled.
 158  *  Note that LPM optimisation path uses SSE4.1 instructions.
 159  */
 160 #define ENABLE_MULTI_BUFFER_OPTIMIZE    1
 161
 162 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 163 #include <rte_hash.h>
 164 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 165 #include <rte_lpm.h>
 166 #include <rte_lpm6.h>
 167 #else
 168 #error "APP_LOOKUP_METHOD set to incorrect value"
 169 #endif
 170
 171 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
 172
 173 #define MAX_JUMBO_PKT_LEN  9600
 174
 175 #define IPV6_ADDR_LEN 16
 176
 177 #define MEMPOOL_CACHE_SIZE 256
 178
 179 /*
 180  * This expression is used to calculate the number of mbufs needed depending on
 181  * user input, taking into account memory for rx and tx hardware rings, cache
 182  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 183  * NB_MBUF never goes below a minimum value of 8192
 184  */
 185
 186 #define NB_MBUF RTE_MAX(\
 187                 (nb_ports*nb_rx_queue*nb_rxd +      \
 188                 nb_ports*nb_lcores*MAX_PKT_BURST +  \
 189                 nb_ports*n_tx_queue*nb_txd +        \
 190                 nb_lcores*MEMPOOL_CACHE_SIZE),      \
 191                 (unsigned)8192)
 192
 193 #define MAX_PKT_BURST     32
 194 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 195
 196 /*
 197  * Try to avoid TX buffering if we have at least MAX_TX_BURST packets to send.
 198  */
 199 #define MAX_TX_BURST  (MAX_PKT_BURST / 2)
 200 #define BURST_SIZE    MAX_TX_BURST
 201
 202 #define NB_SOCKETS 8
 203
 204 /* Configure how many packets ahead to prefetch, when reading packets */
 205 #define PREFETCH_OFFSET 3
 206
 207 /* Used to mark destination port as 'invalid'. */
 208 #define BAD_PORT        ((uint16_t)-1)
 209
 210 #define FWDSTEP 4
 211
 212 /*
 213  * Configurable number of RX/TX ring descriptors
 214  */
 215 #define RTE_TEST_RX_DESC_DEFAULT 128
 216 #define RTE_TEST_TX_DESC_DEFAULT 128
 217 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 218 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 219
 220 /* ethernet addresses of ports */
 221 static uint64_t dest_eth_addr[RTE_MAX_ETHPORTS];
 222 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 223
 224 static xmm_t val_eth[RTE_MAX_ETHPORTS];
 225
 226 /* replace first 12B of the ethernet header. */
 227 #define MASK_ETH 0x3f
 228
 229 /* mask of enabled ports */
 230 static uint32_t enabled_port_mask;
 231 static int promiscuous_on; /**< Set in promiscuous mode off by default. */
 232 static int numa_on = 1;    /**< NUMA is enabled by default. */
 233 static int parse_ptype_on;
 234
 235 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 236 static int ipv6;           /**< ipv6 is false by default. */
 237 #endif
 238
 239 #if (APP_CPU_LOAD == 1)
 240
 241 #define MAX_CPU RTE_MAX_LCORE
 242 #define CPU_LOAD_TIMEOUT_US (5 * 1000 * 1000)  /**< Timeout for collecting 5s */
 243
 244 #define CPU_PROCESS     0
 245 #define CPU_POLL        1
 246 #define MAX_CPU_COUNTER 2
 247
 248 struct cpu_load {
 249         uint16_t       n_cpu;
 250         uint64_t       counter;
 251         uint64_t       hits[MAX_CPU_COUNTER][MAX_CPU];
 252 } __rte_cache_aligned;
 253
 254 static struct cpu_load cpu_load;
 255 static int cpu_load_lcore_id = -1;
 256
 257 #define SET_CPU_BUSY(thread, counter) \
 258                 thread->conf.busy[counter] = 1
 259
 260 #define SET_CPU_IDLE(thread, counter) \
 261                 thread->conf.busy[counter] = 0
 262
 263 #define IS_CPU_BUSY(thread, counter) \
 264                 (thread->conf.busy[counter] > 0)
 265
 266 #else
 267
 268 #define SET_CPU_BUSY(thread, counter)
 269 #define SET_CPU_IDLE(thread, counter)
 270 #define IS_CPU_BUSY(thread, counter) 0
 271
 272 #endif
 273
 274 struct mbuf_table {
 275         uint16_t len;
 276         struct rte_mbuf *m_table[MAX_PKT_BURST];
 277 };
 278
 279 struct lcore_rx_queue {
 280         uint16_t port_id;
 281         uint8_t queue_id;
 282 } __rte_cache_aligned;
 283
 284 #define MAX_RX_QUEUE_PER_LCORE 16
 285 #define MAX_TX_QUEUE_PER_PORT  RTE_MAX_ETHPORTS
 286 #define MAX_RX_QUEUE_PER_PORT  128
 287
 288 #define MAX_LCORE_PARAMS       1024
 289 struct rx_thread_params {
 290         uint16_t port_id;
 291         uint8_t queue_id;
 292         uint8_t lcore_id;
 293         uint8_t thread_id;
 294 } __rte_cache_aligned;
 295
 296 static struct rx_thread_params rx_thread_params_array[MAX_LCORE_PARAMS];
 297 static struct rx_thread_params rx_thread_params_array_default[] = {
 298         {0, 0, 2, 0},
 299         {0, 1, 2, 1},
 300         {0, 2, 2, 2},
 301         {1, 0, 2, 3},
 302         {1, 1, 2, 4},
 303         {1, 2, 2, 5},
 304         {2, 0, 2, 6},
 305         {3, 0, 3, 7},
 306         {3, 1, 3, 8},
 307 };
 308
 309 static struct rx_thread_params *rx_thread_params =
 310                 rx_thread_params_array_default;
 311 static uint16_t nb_rx_thread_params = RTE_DIM(rx_thread_params_array_default);
 312
 313 struct tx_thread_params {
 314         uint8_t lcore_id;
 315         uint8_t thread_id;
 316 } __rte_cache_aligned;
 317
 318 static struct tx_thread_params tx_thread_params_array[MAX_LCORE_PARAMS];
 319 static struct tx_thread_params tx_thread_params_array_default[] = {
 320         {4, 0},
 321         {5, 1},
 322         {6, 2},
 323         {7, 3},
 324         {8, 4},
 325         {9, 5},
 326         {10, 6},
 327         {11, 7},
 328         {12, 8},
 329 };
 330
 331 static struct tx_thread_params *tx_thread_params =
 332                 tx_thread_params_array_default;
 333 static uint16_t nb_tx_thread_params = RTE_DIM(tx_thread_params_array_default);
 334
 335 static struct rte_eth_conf port_conf = {
 336         .rxmode = {
 337                 .mq_mode = ETH_MQ_RX_RSS,
 338                 .max_rx_pkt_len = ETHER_MAX_LEN,
 339                 .split_hdr_size = 0,
 340                 .header_split   = 0, /**< Header Split disabled */
 341                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 342                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 343                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 344                 .hw_strip_crc   = 1, /**< CRC stripped by hardware */
 345         },
 346         .rx_adv_conf = {
 347                 .rss_conf = {
 348                         .rss_key = NULL,
 349                         .rss_hf = ETH_RSS_TCP,
 350                 },
 351         },
 352         .txmode = {
 353                 .mq_mode = ETH_MQ_TX_NONE,
 354         },
 355 };
 356
 357 static struct rte_mempool *pktmbuf_pool[NB_SOCKETS];
 358
 359 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 360
 361 #include <rte_hash_crc.h>
 362 #define DEFAULT_HASH_FUNC       rte_hash_crc
 363
 364 struct ipv4_5tuple {
 365         uint32_t ip_dst;
 366         uint32_t ip_src;
 367         uint16_t port_dst;
 368         uint16_t port_src;
 369         uint8_t  proto;
 370 } __attribute__((__packed__));
 371
 372 union ipv4_5tuple_host {
 373         struct {
 374                 uint8_t  pad0;
 375                 uint8_t  proto;
 376                 uint16_t pad1;
 377                 uint32_t ip_src;
 378                 uint32_t ip_dst;
 379                 uint16_t port_src;
 380                 uint16_t port_dst;
 381         };
 382         __m128i xmm;
 383 };
 384
 385 #define XMM_NUM_IN_IPV6_5TUPLE 3
 386
 387 struct ipv6_5tuple {
 388         uint8_t  ip_dst[IPV6_ADDR_LEN];
 389         uint8_t  ip_src[IPV6_ADDR_LEN];
 390         uint16_t port_dst;
 391         uint16_t port_src;
 392         uint8_t  proto;
 393 } __attribute__((__packed__));
 394
 395 union ipv6_5tuple_host {
 396         struct {
 397                 uint16_t pad0;
 398                 uint8_t  proto;
 399                 uint8_t  pad1;
 400                 uint8_t  ip_src[IPV6_ADDR_LEN];
 401                 uint8_t  ip_dst[IPV6_ADDR_LEN];
 402                 uint16_t port_src;
 403                 uint16_t port_dst;
 404                 uint64_t reserve;
 405         };
 406         __m128i xmm[XMM_NUM_IN_IPV6_5TUPLE];
 407 };
 408
 409 struct ipv4_l3fwd_route {
 410         struct ipv4_5tuple key;
 411         uint8_t if_out;
 412 };
 413
 414 struct ipv6_l3fwd_route {
 415         struct ipv6_5tuple key;
 416         uint8_t if_out;
 417 };
 418
 419 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 420         {{IPv4(101, 0, 0, 0), IPv4(100, 10, 0, 1),  101, 11, IPPROTO_TCP}, 0},
 421         {{IPv4(201, 0, 0, 0), IPv4(200, 20, 0, 1),  102, 12, IPPROTO_TCP}, 1},
 422         {{IPv4(111, 0, 0, 0), IPv4(100, 30, 0, 1),  101, 11, IPPROTO_TCP}, 2},
 423         {{IPv4(211, 0, 0, 0), IPv4(200, 40, 0, 1),  102, 12, IPPROTO_TCP}, 3},
 424 };
 425
 426 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 427         {{
 428         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 429         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 430                         0x05},
 431         101, 11, IPPROTO_TCP}, 0},
 432
 433         {{
 434         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 435         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 436                         0x05},
 437         102, 12, IPPROTO_TCP}, 1},
 438
 439         {{
 440         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 441         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 442                         0x05},
 443         101, 11, IPPROTO_TCP}, 2},
 444
 445         {{
 446         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 447         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 448                         0x05},
 449         102, 12, IPPROTO_TCP}, 3},
 450 };
 451
 452 typedef struct rte_hash lookup_struct_t;
 453 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 454 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 455
 456 #ifdef RTE_ARCH_X86_64
 457 /* default to 4 million hash entries (approx) */
 458 #define L3FWD_HASH_ENTRIES (1024*1024*4)
 459 #else
 460 /* 32-bit has less address-space for hugepage memory, limit to 1M entries */
 461 #define L3FWD_HASH_ENTRIES (1024*1024*1)
 462 #endif
 463 #define HASH_ENTRY_NUMBER_DEFAULT 4
 464
 465 static uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
 466
 467 static inline uint32_t
 468 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
 469                 uint32_t init_val)
 470 {
 471         const union ipv4_5tuple_host *k;
 472         uint32_t t;
 473         const uint32_t *p;
 474
 475         k = data;
 476         t = k->proto;
 477         p = (const uint32_t *)&k->port_src;
 478
 479         init_val = rte_hash_crc_4byte(t, init_val);
 480         init_val = rte_hash_crc_4byte(k->ip_src, init_val);
 481         init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
 482         init_val = rte_hash_crc_4byte(*p, init_val);
 483         return init_val;
 484 }
 485
 486 static inline uint32_t
 487 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len,
 488                 uint32_t init_val)
 489 {
 490         const union ipv6_5tuple_host *k;
 491         uint32_t t;
 492         const uint32_t *p;
 493         const uint32_t *ip_src0, *ip_src1, *ip_src2, *ip_src3;
 494         const uint32_t *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
 495
 496         k = data;
 497         t = k->proto;
 498         p = (const uint32_t *)&k->port_src;
 499
 500         ip_src0 = (const uint32_t *) k->ip_src;
 501         ip_src1 = (const uint32_t *)(k->ip_src + 4);
 502         ip_src2 = (const uint32_t *)(k->ip_src + 8);
 503         ip_src3 = (const uint32_t *)(k->ip_src + 12);
 504         ip_dst0 = (const uint32_t *) k->ip_dst;
 505         ip_dst1 = (const uint32_t *)(k->ip_dst + 4);
 506         ip_dst2 = (const uint32_t *)(k->ip_dst + 8);
 507         ip_dst3 = (const uint32_t *)(k->ip_dst + 12);
 508         init_val = rte_hash_crc_4byte(t, init_val);
 509         init_val = rte_hash_crc_4byte(*ip_src0, init_val);
 510         init_val = rte_hash_crc_4byte(*ip_src1, init_val);
 511         init_val = rte_hash_crc_4byte(*ip_src2, init_val);
 512         init_val = rte_hash_crc_4byte(*ip_src3, init_val);
 513         init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
 514         init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
 515         init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
 516         init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
 517         init_val = rte_hash_crc_4byte(*p, init_val);
 518         return init_val;
 519 }
 520
 521 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
 522 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
 523
 524 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 525 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 526
 527 #endif
 528
 529 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 530 struct ipv4_l3fwd_route {
 531         uint32_t ip;
 532         uint8_t  depth;
 533         uint8_t  if_out;
 534 };
 535
 536 struct ipv6_l3fwd_route {
 537         uint8_t ip[16];
 538         uint8_t depth;
 539         uint8_t if_out;
 540 };
 541
 542 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 543         {IPv4(1, 1, 1, 0), 24, 0},
 544         {IPv4(2, 1, 1, 0), 24, 1},
 545         {IPv4(3, 1, 1, 0), 24, 2},
 546         {IPv4(4, 1, 1, 0), 24, 3},
 547         {IPv4(5, 1, 1, 0), 24, 4},
 548         {IPv4(6, 1, 1, 0), 24, 5},
 549         {IPv4(7, 1, 1, 0), 24, 6},
 550         {IPv4(8, 1, 1, 0), 24, 7},
 551 };
 552
 553 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 554         {{1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 0},
 555         {{2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 1},
 556         {{3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 2},
 557         {{4, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 3},
 558         {{5, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 4},
 559         {{6, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 5},
 560         {{7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 6},
 561         {{8, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 7},
 562 };
 563
 564 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
 565 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
 566
 567 #define IPV4_L3FWD_LPM_MAX_RULES         1024
 568 #define IPV6_L3FWD_LPM_MAX_RULES         1024
 569 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
 570
 571 typedef struct rte_lpm lookup_struct_t;
 572 typedef struct rte_lpm6 lookup6_struct_t;
 573 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 574 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 575 #endif
 576
 577 struct lcore_conf {
 578         lookup_struct_t *ipv4_lookup_struct;
 579 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 580         lookup6_struct_t *ipv6_lookup_struct;
 581 #else
 582         lookup_struct_t *ipv6_lookup_struct;
 583 #endif
 584         void *data;
 585 } __rte_cache_aligned;
 586
 587 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 588 RTE_DEFINE_PER_LCORE(struct lcore_conf *, lcore_conf);
 589
 590 #define MAX_RX_QUEUE_PER_THREAD 16
 591 #define MAX_TX_PORT_PER_THREAD  RTE_MAX_ETHPORTS
 592 #define MAX_TX_QUEUE_PER_PORT   RTE_MAX_ETHPORTS
 593 #define MAX_RX_QUEUE_PER_PORT   128
 594
 595 #define MAX_RX_THREAD 1024
 596 #define MAX_TX_THREAD 1024
 597 #define MAX_THREAD    (MAX_RX_THREAD + MAX_TX_THREAD)
 598
 599 /**
 600  * Producers and consumers threads configuration
 601  */
 602 static int lthreads_on = 1; /**< Use lthreads for processing*/
 603
 604 rte_atomic16_t rx_counter;  /**< Number of spawned rx threads */
 605 rte_atomic16_t tx_counter;  /**< Number of spawned tx threads */
 606
 607 struct thread_conf {
 608         uint16_t lcore_id;      /**< Initial lcore for rx thread */
 609         uint16_t cpu_id;        /**< Cpu id for cpu load stats counter */
 610         uint16_t thread_id;     /**< Thread ID */
 611
 612 #if (APP_CPU_LOAD > 0)
 613         int busy[MAX_CPU_COUNTER];
 614 #endif
 615 };
 616
 617 struct thread_rx_conf {
 618         struct thread_conf conf;
 619
 620         uint16_t n_rx_queue;
 621         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 622
 623         uint16_t n_ring;        /**< Number of output rings */
 624         struct rte_ring *ring[RTE_MAX_LCORE];
 625         struct lthread_cond *ready[RTE_MAX_LCORE];
 626
 627 #if (APP_CPU_LOAD > 0)
 628         int busy[MAX_CPU_COUNTER];
 629 #endif
 630 } __rte_cache_aligned;
 631
 632 uint16_t n_rx_thread;
 633 struct thread_rx_conf rx_thread[MAX_RX_THREAD];
 634
 635 struct thread_tx_conf {
 636         struct thread_conf conf;
 637
 638         uint16_t tx_queue_id[RTE_MAX_LCORE];
 639         struct mbuf_table tx_mbufs[RTE_MAX_LCORE];
 640
 641         struct rte_ring *ring;
 642         struct lthread_cond **ready;
 643
 644 } __rte_cache_aligned;
 645
 646 uint16_t n_tx_thread;
 647 struct thread_tx_conf tx_thread[MAX_TX_THREAD];
 648
 649 /* Send burst of packets on an output interface */
 650 static inline int
 651 send_burst(struct thread_tx_conf *qconf, uint16_t n, uint16_t port)
 652 {
 653         struct rte_mbuf **m_table;
 654         int ret;
 655         uint16_t queueid;
 656
 657         queueid = qconf->tx_queue_id[port];
 658         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 659
 660         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 661         if (unlikely(ret < n)) {
 662                 do {
 663                         rte_pktmbuf_free(m_table[ret]);
 664                 } while (++ret < n);
 665         }
 666
 667         return 0;
 668 }
 669
 670 /* Enqueue a single packet, and send burst if queue is filled */
 671 static inline int
 672 send_single_packet(struct rte_mbuf *m, uint16_t port)
 673 {
 674         uint16_t len;
 675         struct thread_tx_conf *qconf;
 676
 677         if (lthreads_on)
 678                 qconf = (struct thread_tx_conf *)lthread_get_data();
 679         else
 680                 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
 681
 682         len = qconf->tx_mbufs[port].len;
 683         qconf->tx_mbufs[port].m_table[len] = m;
 684         len++;
 685
 686         /* enough pkts to be sent */
 687         if (unlikely(len == MAX_PKT_BURST)) {
 688                 send_burst(qconf, MAX_PKT_BURST, port);
 689                 len = 0;
 690         }
 691
 692         qconf->tx_mbufs[port].len = len;
 693         return 0;
 694 }
 695
 696 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
 697         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
 698 static __rte_always_inline void
 699 send_packetsx4(uint16_t port,
 700         struct rte_mbuf *m[], uint32_t num)
 701 {
 702         uint32_t len, j, n;
 703         struct thread_tx_conf *qconf;
 704
 705         if (lthreads_on)
 706                 qconf = (struct thread_tx_conf *)lthread_get_data();
 707         else
 708                 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
 709
 710         len = qconf->tx_mbufs[port].len;
 711
 712         /*
 713          * If TX buffer for that queue is empty, and we have enough packets,
 714          * then send them straightway.
 715          */
 716         if (num >= MAX_TX_BURST && len == 0) {
 717                 n = rte_eth_tx_burst(port, qconf->tx_queue_id[port], m, num);
 718                 if (unlikely(n < num)) {
 719                         do {
 720                                 rte_pktmbuf_free(m[n]);
 721                         } while (++n < num);
 722                 }
 723                 return;
 724         }
 725
 726         /*
 727          * Put packets into TX buffer for that queue.
 728          */
 729
 730         n = len + num;
 731         n = (n > MAX_PKT_BURST) ? MAX_PKT_BURST - len : num;
 732
 733         j = 0;
 734         switch (n % FWDSTEP) {
 735         while (j < n) {
 736         case 0:
 737                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 738                 j++;
 739                 /* fall-through */
 740         case 3:
 741                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 742                 j++;
 743                 /* fall-through */
 744         case 2:
 745                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 746                 j++;
 747                 /* fall-through */
 748         case 1:
 749                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 750                 j++;
 751         }
 752         }
 753
 754         len += n;
 755
 756         /* enough pkts to be sent */
 757         if (unlikely(len == MAX_PKT_BURST)) {
 758
 759                 send_burst(qconf, MAX_PKT_BURST, port);
 760
 761                 /* copy rest of the packets into the TX buffer. */
 762                 len = num - n;
 763                 j = 0;
 764                 switch (len % FWDSTEP) {
 765                 while (j < len) {
 766                 case 0:
 767                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 768                         j++;
 769                         /* fall-through */
 770                 case 3:
 771                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 772                         j++;
 773                         /* fall-through */
 774                 case 2:
 775                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 776                         j++;
 777                         /* fall-through */
 778                 case 1:
 779                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 780                         j++;
 781                 }
 782                 }
 783         }
 784
 785         qconf->tx_mbufs[port].len = len;
 786 }
 787 #endif /* APP_LOOKUP_LPM */
 788
 789 #ifdef DO_RFC_1812_CHECKS
 790 static inline int
 791 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 792 {
 793         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 794         /*
 795          * 1. The packet length reported by the Link Layer must be large
 796          * enough to hold the minimum length legal IP datagram (20 bytes).
 797          */
 798         if (link_len < sizeof(struct ipv4_hdr))
 799                 return -1;
 800
 801         /* 2. The IP checksum must be correct. */
 802         /* this is checked in H/W */
 803
 804         /*
 805          * 3. The IP version number must be 4. If the version number is not 4
 806          * then the packet may be another version of IP, such as IPng or
 807          * ST-II.
 808          */
 809         if (((pkt->version_ihl) >> 4) != 4)
 810                 return -3;
 811         /*
 812          * 4. The IP header length field must be large enough to hold the
 813          * minimum length legal IP datagram (20 bytes = 5 words).
 814          */
 815         if ((pkt->version_ihl & 0xf) < 5)
 816                 return -4;
 817
 818         /*
 819          * 5. The IP total length field must be large enough to hold the IP
 820          * datagram header, whose length is specified in the IP header length
 821          * field.
 822          */
 823         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 824                 return -5;
 825
 826         return 0;
 827 }
 828 #endif
 829
 830 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 831
 832 static __m128i mask0;
 833 static __m128i mask1;
 834 static __m128i mask2;
 835 static inline uint16_t
 836 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid,
 837                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 838 {
 839         int ret = 0;
 840         union ipv4_5tuple_host key;
 841
 842         ipv4_hdr = (uint8_t *)ipv4_hdr + offsetof(struct ipv4_hdr, time_to_live);
 843         __m128i data = _mm_loadu_si128((__m128i *)(ipv4_hdr));
 844         /* Get 5 tuple: dst port, src port, dst IP address, src IP address and
 845            protocol */
 846         key.xmm = _mm_and_si128(data, mask0);
 847         /* Find destination port */
 848         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 849         return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
 850 }
 851
 852 static inline uint16_t
 853 get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid,
 854                 lookup_struct_t *ipv6_l3fwd_lookup_struct)
 855 {
 856         int ret = 0;
 857         union ipv6_5tuple_host key;
 858
 859         ipv6_hdr = (uint8_t *)ipv6_hdr + offsetof(struct ipv6_hdr, payload_len);
 860         __m128i data0 = _mm_loadu_si128((__m128i *)(ipv6_hdr));
 861         __m128i data1 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
 862                         sizeof(__m128i)));
 863         __m128i data2 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
 864                         sizeof(__m128i) + sizeof(__m128i)));
 865         /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
 866         key.xmm[0] = _mm_and_si128(data0, mask1);
 867         /* Get part of 5 tuple: dst IP address lower 96 bits and src IP address
 868            higher 32 bits */
 869         key.xmm[1] = data1;
 870         /* Get part of 5 tuple: dst port and src port and dst IP address higher
 871            32 bits */
 872         key.xmm[2] = _mm_and_si128(data2, mask2);
 873
 874         /* Find destination port */
 875         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 876         return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
 877 }
 878 #endif
 879
 880 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 881
 882 static inline uint16_t
 883 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid,
 884                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 885 {
 886         uint32_t next_hop;
 887
 888         return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 889                 rte_be_to_cpu_32(((struct ipv4_hdr *)ipv4_hdr)->dst_addr),
 890                 &next_hop) == 0) ? next_hop : portid);
 891 }
 892
 893 static inline uint16_t
 894 get_ipv6_dst_port(void *ipv6_hdr,  uint16_t portid,
 895                 lookup6_struct_t *ipv6_l3fwd_lookup_struct)
 896 {
 897         uint32_t next_hop;
 898
 899         return ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
 900                         ((struct ipv6_hdr *)ipv6_hdr)->dst_addr, &next_hop) == 0) ?
 901                         next_hop : portid);
 902 }
 903 #endif
 904
 905 static inline void l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid)
 906                 __attribute__((unused));
 907
 908 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) && \
 909         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
 910
 911 #define MASK_ALL_PKTS   0xff
 912 #define EXCLUDE_1ST_PKT 0xfe
 913 #define EXCLUDE_2ND_PKT 0xfd
 914 #define EXCLUDE_3RD_PKT 0xfb
 915 #define EXCLUDE_4TH_PKT 0xf7
 916 #define EXCLUDE_5TH_PKT 0xef
 917 #define EXCLUDE_6TH_PKT 0xdf
 918 #define EXCLUDE_7TH_PKT 0xbf
 919 #define EXCLUDE_8TH_PKT 0x7f
 920
 921 static inline void
 922 simple_ipv4_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid)
 923 {
 924         struct ether_hdr *eth_hdr[8];
 925         struct ipv4_hdr *ipv4_hdr[8];
 926         uint16_t dst_port[8];
 927         int32_t ret[8];
 928         union ipv4_5tuple_host key[8];
 929         __m128i data[8];
 930
 931         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
 932         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
 933         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
 934         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
 935         eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *);
 936         eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *);
 937         eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *);
 938         eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *);
 939
 940         /* Handle IPv4 headers.*/
 941         ipv4_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv4_hdr *,
 942                         sizeof(struct ether_hdr));
 943         ipv4_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv4_hdr *,
 944                         sizeof(struct ether_hdr));
 945         ipv4_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv4_hdr *,
 946                         sizeof(struct ether_hdr));
 947         ipv4_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv4_hdr *,
 948                         sizeof(struct ether_hdr));
 949         ipv4_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv4_hdr *,
 950                         sizeof(struct ether_hdr));
 951         ipv4_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv4_hdr *,
 952                         sizeof(struct ether_hdr));
 953         ipv4_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv4_hdr *,
 954                         sizeof(struct ether_hdr));
 955         ipv4_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv4_hdr *,
 956                         sizeof(struct ether_hdr));
 957
 958 #ifdef DO_RFC_1812_CHECKS
 959         /* Check to make sure the packet is valid (RFC1812) */
 960         uint8_t valid_mask = MASK_ALL_PKTS;
 961
 962         if (is_valid_ipv4_pkt(ipv4_hdr[0], m[0]->pkt_len) < 0) {
 963                 rte_pktmbuf_free(m[0]);
 964                 valid_mask &= EXCLUDE_1ST_PKT;
 965         }
 966         if (is_valid_ipv4_pkt(ipv4_hdr[1], m[1]->pkt_len) < 0) {
 967                 rte_pktmbuf_free(m[1]);
 968                 valid_mask &= EXCLUDE_2ND_PKT;
 969         }
 970         if (is_valid_ipv4_pkt(ipv4_hdr[2], m[2]->pkt_len) < 0) {
 971                 rte_pktmbuf_free(m[2]);
 972                 valid_mask &= EXCLUDE_3RD_PKT;
 973         }
 974         if (is_valid_ipv4_pkt(ipv4_hdr[3], m[3]->pkt_len) < 0) {
 975                 rte_pktmbuf_free(m[3]);
 976                 valid_mask &= EXCLUDE_4TH_PKT;
 977         }
 978         if (is_valid_ipv4_pkt(ipv4_hdr[4], m[4]->pkt_len) < 0) {
 979                 rte_pktmbuf_free(m[4]);
 980                 valid_mask &= EXCLUDE_5TH_PKT;
 981         }
 982         if (is_valid_ipv4_pkt(ipv4_hdr[5], m[5]->pkt_len) < 0) {
 983                 rte_pktmbuf_free(m[5]);
 984                 valid_mask &= EXCLUDE_6TH_PKT;
 985         }
 986         if (is_valid_ipv4_pkt(ipv4_hdr[6], m[6]->pkt_len) < 0) {
 987                 rte_pktmbuf_free(m[6]);
 988                 valid_mask &= EXCLUDE_7TH_PKT;
 989         }
 990         if (is_valid_ipv4_pkt(ipv4_hdr[7], m[7]->pkt_len) < 0) {
 991                 rte_pktmbuf_free(m[7]);
 992                 valid_mask &= EXCLUDE_8TH_PKT;
 993         }
 994         if (unlikely(valid_mask != MASK_ALL_PKTS)) {
 995                 if (valid_mask == 0)
 996                         return;
 997
 998                 uint8_t i = 0;
 999
1000                 for (i = 0; i < 8; i++)
1001                         if ((0x1 << i) & valid_mask)
1002                                 l3fwd_simple_forward(m[i], portid);
1003         }
1004 #endif /* End of #ifdef DO_RFC_1812_CHECKS */
1005
1006         data[0] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[0], __m128i *,
1007                         sizeof(struct ether_hdr) +
1008                         offsetof(struct ipv4_hdr, time_to_live)));
1009         data[1] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[1], __m128i *,
1010                         sizeof(struct ether_hdr) +
1011                         offsetof(struct ipv4_hdr, time_to_live)));
1012         data[2] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[2], __m128i *,
1013                         sizeof(struct ether_hdr) +
1014                         offsetof(struct ipv4_hdr, time_to_live)));
1015         data[3] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[3], __m128i *,
1016                         sizeof(struct ether_hdr) +
1017                         offsetof(struct ipv4_hdr, time_to_live)));
1018         data[4] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[4], __m128i *,
1019                         sizeof(struct ether_hdr) +
1020                         offsetof(struct ipv4_hdr, time_to_live)));
1021         data[5] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[5], __m128i *,
1022                         sizeof(struct ether_hdr) +
1023                         offsetof(struct ipv4_hdr, time_to_live)));
1024         data[6] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[6], __m128i *,
1025                         sizeof(struct ether_hdr) +
1026                         offsetof(struct ipv4_hdr, time_to_live)));
1027         data[7] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[7], __m128i *,
1028                         sizeof(struct ether_hdr) +
1029                         offsetof(struct ipv4_hdr, time_to_live)));
1030
1031         key[0].xmm = _mm_and_si128(data[0], mask0);
1032         key[1].xmm = _mm_and_si128(data[1], mask0);
1033         key[2].xmm = _mm_and_si128(data[2], mask0);
1034         key[3].xmm = _mm_and_si128(data[3], mask0);
1035         key[4].xmm = _mm_and_si128(data[4], mask0);
1036         key[5].xmm = _mm_and_si128(data[5], mask0);
1037         key[6].xmm = _mm_and_si128(data[6], mask0);
1038         key[7].xmm = _mm_and_si128(data[7], mask0);
1039
1040         const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1041                         &key[4], &key[5], &key[6], &key[7]};
1042
1043         rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct,
1044                         &key_array[0], 8, ret);
1045         dst_port[0] = ((ret[0] < 0) ? portid : ipv4_l3fwd_out_if[ret[0]]);
1046         dst_port[1] = ((ret[1] < 0) ? portid : ipv4_l3fwd_out_if[ret[1]]);
1047         dst_port[2] = ((ret[2] < 0) ? portid : ipv4_l3fwd_out_if[ret[2]]);
1048         dst_port[3] = ((ret[3] < 0) ? portid : ipv4_l3fwd_out_if[ret[3]]);
1049         dst_port[4] = ((ret[4] < 0) ? portid : ipv4_l3fwd_out_if[ret[4]]);
1050         dst_port[5] = ((ret[5] < 0) ? portid : ipv4_l3fwd_out_if[ret[5]]);
1051         dst_port[6] = ((ret[6] < 0) ? portid : ipv4_l3fwd_out_if[ret[6]]);
1052         dst_port[7] = ((ret[7] < 0) ? portid : ipv4_l3fwd_out_if[ret[7]]);
1053
1054         if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1055                         (enabled_port_mask & 1 << dst_port[0]) == 0)
1056                 dst_port[0] = portid;
1057         if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1058                         (enabled_port_mask & 1 << dst_port[1]) == 0)
1059                 dst_port[1] = portid;
1060         if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1061                         (enabled_port_mask & 1 << dst_port[2]) == 0)
1062                 dst_port[2] = portid;
1063         if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1064                         (enabled_port_mask & 1 << dst_port[3]) == 0)
1065                 dst_port[3] = portid;
1066         if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1067                         (enabled_port_mask & 1 << dst_port[4]) == 0)
1068                 dst_port[4] = portid;
1069         if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1070                         (enabled_port_mask & 1 << dst_port[5]) == 0)
1071                 dst_port[5] = portid;
1072         if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1073                         (enabled_port_mask & 1 << dst_port[6]) == 0)
1074                 dst_port[6] = portid;
1075         if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1076                         (enabled_port_mask & 1 << dst_port[7]) == 0)
1077                 dst_port[7] = portid;
1078
1079 #ifdef DO_RFC_1812_CHECKS
1080         /* Update time to live and header checksum */
1081         --(ipv4_hdr[0]->time_to_live);
1082         --(ipv4_hdr[1]->time_to_live);
1083         --(ipv4_hdr[2]->time_to_live);
1084         --(ipv4_hdr[3]->time_to_live);
1085         ++(ipv4_hdr[0]->hdr_checksum);
1086         ++(ipv4_hdr[1]->hdr_checksum);
1087         ++(ipv4_hdr[2]->hdr_checksum);
1088         ++(ipv4_hdr[3]->hdr_checksum);
1089         --(ipv4_hdr[4]->time_to_live);
1090         --(ipv4_hdr[5]->time_to_live);
1091         --(ipv4_hdr[6]->time_to_live);
1092         --(ipv4_hdr[7]->time_to_live);
1093         ++(ipv4_hdr[4]->hdr_checksum);
1094         ++(ipv4_hdr[5]->hdr_checksum);
1095         ++(ipv4_hdr[6]->hdr_checksum);
1096         ++(ipv4_hdr[7]->hdr_checksum);
1097 #endif
1098
1099         /* dst addr */
1100         *(uint64_t *)&eth_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1101         *(uint64_t *)&eth_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1102         *(uint64_t *)&eth_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1103         *(uint64_t *)&eth_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1104         *(uint64_t *)&eth_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1105         *(uint64_t *)&eth_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1106         *(uint64_t *)&eth_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1107         *(uint64_t *)&eth_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1108
1109         /* src addr */
1110         ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
1111         ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
1112         ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
1113         ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
1114         ether_addr_copy(&ports_eth_addr[dst_port[4]], &eth_hdr[4]->s_addr);
1115         ether_addr_copy(&ports_eth_addr[dst_port[5]], &eth_hdr[5]->s_addr);
1116         ether_addr_copy(&ports_eth_addr[dst_port[6]], &eth_hdr[6]->s_addr);
1117         ether_addr_copy(&ports_eth_addr[dst_port[7]], &eth_hdr[7]->s_addr);
1118
1119         send_single_packet(m[0], (uint8_t)dst_port[0]);
1120         send_single_packet(m[1], (uint8_t)dst_port[1]);
1121         send_single_packet(m[2], (uint8_t)dst_port[2]);
1122         send_single_packet(m[3], (uint8_t)dst_port[3]);
1123         send_single_packet(m[4], (uint8_t)dst_port[4]);
1124         send_single_packet(m[5], (uint8_t)dst_port[5]);
1125         send_single_packet(m[6], (uint8_t)dst_port[6]);
1126         send_single_packet(m[7], (uint8_t)dst_port[7]);
1127
1128 }
1129
1130 static inline void get_ipv6_5tuple(struct rte_mbuf *m0, __m128i mask0,
1131                 __m128i mask1, union ipv6_5tuple_host *key)
1132 {
1133         __m128i tmpdata0 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1134                         __m128i *, sizeof(struct ether_hdr) +
1135                         offsetof(struct ipv6_hdr, payload_len)));
1136         __m128i tmpdata1 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1137                         __m128i *, sizeof(struct ether_hdr) +
1138                         offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i)));
1139         __m128i tmpdata2 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1140                         __m128i *, sizeof(struct ether_hdr) +
1141                         offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i) +
1142                         sizeof(__m128i)));
1143         key->xmm[0] = _mm_and_si128(tmpdata0, mask0);
1144         key->xmm[1] = tmpdata1;
1145         key->xmm[2] = _mm_and_si128(tmpdata2, mask1);
1146 }
1147
1148 static inline void
1149 simple_ipv6_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid)
1150 {
1151         int32_t ret[8];
1152         uint16_t dst_port[8];
1153         struct ether_hdr *eth_hdr[8];
1154         union ipv6_5tuple_host key[8];
1155
1156         __attribute__((unused)) struct ipv6_hdr *ipv6_hdr[8];
1157
1158         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *);
1159         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *);
1160         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *);
1161         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *);
1162         eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *);
1163         eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *);
1164         eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *);
1165         eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *);
1166
1167         /* Handle IPv6 headers.*/
1168         ipv6_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv6_hdr *,
1169                         sizeof(struct ether_hdr));
1170         ipv6_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv6_hdr *,
1171                         sizeof(struct ether_hdr));
1172         ipv6_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv6_hdr *,
1173                         sizeof(struct ether_hdr));
1174         ipv6_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv6_hdr *,
1175                         sizeof(struct ether_hdr));
1176         ipv6_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv6_hdr *,
1177                         sizeof(struct ether_hdr));
1178         ipv6_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv6_hdr *,
1179                         sizeof(struct ether_hdr));
1180         ipv6_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv6_hdr *,
1181                         sizeof(struct ether_hdr));
1182         ipv6_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv6_hdr *,
1183                         sizeof(struct ether_hdr));
1184
1185         get_ipv6_5tuple(m[0], mask1, mask2, &key[0]);
1186         get_ipv6_5tuple(m[1], mask1, mask2, &key[1]);
1187         get_ipv6_5tuple(m[2], mask1, mask2, &key[2]);
1188         get_ipv6_5tuple(m[3], mask1, mask2, &key[3]);
1189         get_ipv6_5tuple(m[4], mask1, mask2, &key[4]);
1190         get_ipv6_5tuple(m[5], mask1, mask2, &key[5]);
1191         get_ipv6_5tuple(m[6], mask1, mask2, &key[6]);
1192         get_ipv6_5tuple(m[7], mask1, mask2, &key[7]);
1193
1194         const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1195                         &key[4], &key[5], &key[6], &key[7]};
1196
1197         rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1198                         &key_array[0], 4, ret);
1199         dst_port[0] = ((ret[0] < 0) ? portid : ipv6_l3fwd_out_if[ret[0]]);
1200         dst_port[1] = ((ret[1] < 0) ? portid : ipv6_l3fwd_out_if[ret[1]]);
1201         dst_port[2] = ((ret[2] < 0) ? portid : ipv6_l3fwd_out_if[ret[2]]);
1202         dst_port[3] = ((ret[3] < 0) ? portid : ipv6_l3fwd_out_if[ret[3]]);
1203         dst_port[4] = ((ret[4] < 0) ? portid : ipv6_l3fwd_out_if[ret[4]]);
1204         dst_port[5] = ((ret[5] < 0) ? portid : ipv6_l3fwd_out_if[ret[5]]);
1205         dst_port[6] = ((ret[6] < 0) ? portid : ipv6_l3fwd_out_if[ret[6]]);
1206         dst_port[7] = ((ret[7] < 0) ? portid : ipv6_l3fwd_out_if[ret[7]]);
1207
1208         if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1209                         (enabled_port_mask & 1 << dst_port[0]) == 0)
1210                 dst_port[0] = portid;
1211         if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1212                         (enabled_port_mask & 1 << dst_port[1]) == 0)
1213                 dst_port[1] = portid;
1214         if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1215                         (enabled_port_mask & 1 << dst_port[2]) == 0)
1216                 dst_port[2] = portid;
1217         if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1218                         (enabled_port_mask & 1 << dst_port[3]) == 0)
1219                 dst_port[3] = portid;
1220         if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1221                         (enabled_port_mask & 1 << dst_port[4]) == 0)
1222                 dst_port[4] = portid;
1223         if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1224                         (enabled_port_mask & 1 << dst_port[5]) == 0)
1225                 dst_port[5] = portid;
1226         if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1227                         (enabled_port_mask & 1 << dst_port[6]) == 0)
1228                 dst_port[6] = portid;
1229         if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1230                         (enabled_port_mask & 1 << dst_port[7]) == 0)
1231                 dst_port[7] = portid;
1232
1233         /* dst addr */
1234         *(uint64_t *)&eth_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1235         *(uint64_t *)&eth_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1236         *(uint64_t *)&eth_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1237         *(uint64_t *)&eth_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1238         *(uint64_t *)&eth_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1239         *(uint64_t *)&eth_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1240         *(uint64_t *)&eth_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1241         *(uint64_t *)&eth_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1242
1243         /* src addr */
1244         ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
1245         ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
1246         ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
1247         ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
1248         ether_addr_copy(&ports_eth_addr[dst_port[4]], &eth_hdr[4]->s_addr);
1249         ether_addr_copy(&ports_eth_addr[dst_port[5]], &eth_hdr[5]->s_addr);
1250         ether_addr_copy(&ports_eth_addr[dst_port[6]], &eth_hdr[6]->s_addr);
1251         ether_addr_copy(&ports_eth_addr[dst_port[7]], &eth_hdr[7]->s_addr);
1252
1253         send_single_packet(m[0], dst_port[0]);
1254         send_single_packet(m[1], dst_port[1]);
1255         send_single_packet(m[2], dst_port[2]);
1256         send_single_packet(m[3], dst_port[3]);
1257         send_single_packet(m[4], dst_port[4]);
1258         send_single_packet(m[5], dst_port[5]);
1259         send_single_packet(m[6], dst_port[6]);
1260         send_single_packet(m[7], dst_port[7]);
1261
1262 }
1263 #endif /* APP_LOOKUP_METHOD */
1264
1265 static __rte_always_inline void
1266 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid)
1267 {
1268         struct ether_hdr *eth_hdr;
1269         struct ipv4_hdr *ipv4_hdr;
1270         uint16_t dst_port;
1271
1272         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
1273
1274         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
1275                 /* Handle IPv4 headers.*/
1276                 ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
1277                                 sizeof(struct ether_hdr));
1278
1279 #ifdef DO_RFC_1812_CHECKS
1280                 /* Check to make sure the packet is valid (RFC1812) */
1281                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
1282                         rte_pktmbuf_free(m);
1283                         return;
1284                 }
1285 #endif
1286
1287                  dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
1288                         RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct);
1289                 if (dst_port >= RTE_MAX_ETHPORTS ||
1290                                 (enabled_port_mask & 1 << dst_port) == 0)
1291                         dst_port = portid;
1292
1293 #ifdef DO_RFC_1812_CHECKS
1294                 /* Update time to live and header checksum */
1295                 --(ipv4_hdr->time_to_live);
1296                 ++(ipv4_hdr->hdr_checksum);
1297 #endif
1298                 /* dst addr */
1299                 *(uint64_t *)&eth_hdr->d_addr = dest_eth_addr[dst_port];
1300
1301                 /* src addr */
1302                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
1303
1304                 send_single_packet(m, dst_port);
1305         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
1306                 /* Handle IPv6 headers.*/
1307                 struct ipv6_hdr *ipv6_hdr;
1308
1309                 ipv6_hdr = rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
1310                                 sizeof(struct ether_hdr));
1311
1312                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
1313                                 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct);
1314
1315                 if (dst_port >= RTE_MAX_ETHPORTS ||
1316                                 (enabled_port_mask & 1 << dst_port) == 0)
1317                         dst_port = portid;
1318
1319                 /* dst addr */
1320                 *(uint64_t *)&eth_hdr->d_addr = dest_eth_addr[dst_port];
1321
1322                 /* src addr */
1323                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
1324
1325                 send_single_packet(m, dst_port);
1326         } else
1327                 /* Free the mbuf that contains non-IPV4/IPV6 packet */
1328                 rte_pktmbuf_free(m);
1329 }
1330
1331 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1332         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1333 #ifdef DO_RFC_1812_CHECKS
1334
1335 #define IPV4_MIN_VER_IHL        0x45
1336 #define IPV4_MAX_VER_IHL        0x4f
1337 #define IPV4_MAX_VER_IHL_DIFF   (IPV4_MAX_VER_IHL - IPV4_MIN_VER_IHL)
1338
1339 /* Minimum value of IPV4 total length (20B) in network byte order. */
1340 #define IPV4_MIN_LEN_BE (sizeof(struct ipv4_hdr) << 8)
1341
1342 /*
1343  * From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2:
1344  * - The IP version number must be 4.
1345  * - The IP header length field must be large enough to hold the
1346  *    minimum length legal IP datagram (20 bytes = 5 words).
1347  * - The IP total length field must be large enough to hold the IP
1348  *   datagram header, whose length is specified in the IP header length
1349  *   field.
1350  * If we encounter invalid IPV4 packet, then set destination port for it
1351  * to BAD_PORT value.
1352  */
1353 static __rte_always_inline void
1354 rfc1812_process(struct ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t ptype)
1355 {
1356         uint8_t ihl;
1357
1358         if (RTE_ETH_IS_IPV4_HDR(ptype)) {
1359                 ihl = ipv4_hdr->version_ihl - IPV4_MIN_VER_IHL;
1360
1361                 ipv4_hdr->time_to_live--;
1362                 ipv4_hdr->hdr_checksum++;
1363
1364                 if (ihl > IPV4_MAX_VER_IHL_DIFF ||
1365                                 ((uint8_t)ipv4_hdr->total_length == 0 &&
1366                                 ipv4_hdr->total_length < IPV4_MIN_LEN_BE)) {
1367                         dp[0] = BAD_PORT;
1368                 }
1369         }
1370 }
1371
1372 #else
1373 #define rfc1812_process(mb, dp, ptype)  do { } while (0)
1374 #endif /* DO_RFC_1812_CHECKS */
1375 #endif /* APP_LOOKUP_LPM && ENABLE_MULTI_BUFFER_OPTIMIZE */
1376
1377
1378 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1379         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1380
1381 static __rte_always_inline uint16_t
1382 get_dst_port(struct rte_mbuf *pkt, uint32_t dst_ipv4, uint16_t portid)
1383 {
1384         uint32_t next_hop;
1385         struct ipv6_hdr *ipv6_hdr;
1386         struct ether_hdr *eth_hdr;
1387
1388         if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
1389                 return (uint16_t) ((rte_lpm_lookup(
1390                                 RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dst_ipv4,
1391                                 &next_hop) == 0) ? next_hop : portid);
1392
1393         } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
1394
1395                 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1396                 ipv6_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
1397
1398                 return (uint16_t) ((rte_lpm6_lookup(
1399                                 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1400                                 ipv6_hdr->dst_addr, &next_hop) == 0) ?
1401                                 next_hop : portid);
1402
1403         }
1404
1405         return portid;
1406 }
1407
1408 static inline void
1409 process_packet(struct rte_mbuf *pkt, uint16_t *dst_port, uint16_t portid)
1410 {
1411         struct ether_hdr *eth_hdr;
1412         struct ipv4_hdr *ipv4_hdr;
1413         uint32_t dst_ipv4;
1414         uint16_t dp;
1415         __m128i te, ve;
1416
1417         eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
1418         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1419
1420         dst_ipv4 = ipv4_hdr->dst_addr;
1421         dst_ipv4 = rte_be_to_cpu_32(dst_ipv4);
1422         dp = get_dst_port(pkt, dst_ipv4, portid);
1423
1424         te = _mm_load_si128((__m128i *)eth_hdr);
1425         ve = val_eth[dp];
1426
1427         dst_port[0] = dp;
1428         rfc1812_process(ipv4_hdr, dst_port, pkt->packet_type);
1429
1430         te =  _mm_blend_epi16(te, ve, MASK_ETH);
1431         _mm_store_si128((__m128i *)eth_hdr, te);
1432 }
1433
1434 /*
1435  * Read packet_type and destination IPV4 addresses from 4 mbufs.
1436  */
1437 static inline void
1438 processx4_step1(struct rte_mbuf *pkt[FWDSTEP],
1439                 __m128i *dip,
1440                 uint32_t *ipv4_flag)
1441 {
1442         struct ipv4_hdr *ipv4_hdr;
1443         struct ether_hdr *eth_hdr;
1444         uint32_t x0, x1, x2, x3;
1445
1446         eth_hdr = rte_pktmbuf_mtod(pkt[0], struct ether_hdr *);
1447         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1448         x0 = ipv4_hdr->dst_addr;
1449         ipv4_flag[0] = pkt[0]->packet_type & RTE_PTYPE_L3_IPV4;
1450
1451         eth_hdr = rte_pktmbuf_mtod(pkt[1], struct ether_hdr *);
1452         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1453         x1 = ipv4_hdr->dst_addr;
1454         ipv4_flag[0] &= pkt[1]->packet_type;
1455
1456         eth_hdr = rte_pktmbuf_mtod(pkt[2], struct ether_hdr *);
1457         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1458         x2 = ipv4_hdr->dst_addr;
1459         ipv4_flag[0] &= pkt[2]->packet_type;
1460
1461         eth_hdr = rte_pktmbuf_mtod(pkt[3], struct ether_hdr *);
1462         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1463         x3 = ipv4_hdr->dst_addr;
1464         ipv4_flag[0] &= pkt[3]->packet_type;
1465
1466         dip[0] = _mm_set_epi32(x3, x2, x1, x0);
1467 }
1468
1469 /*
1470  * Lookup into LPM for destination port.
1471  * If lookup fails, use incoming port (portid) as destination port.
1472  */
1473 static inline void
1474 processx4_step2(__m128i dip,
1475                 uint32_t ipv4_flag,
1476                 uint16_t portid,
1477                 struct rte_mbuf *pkt[FWDSTEP],
1478                 uint16_t dprt[FWDSTEP])
1479 {
1480         rte_xmm_t dst;
1481         const __m128i bswap_mask = _mm_set_epi8(12, 13, 14, 15, 8, 9, 10, 11,
1482                         4, 5, 6, 7, 0, 1, 2, 3);
1483
1484         /* Byte swap 4 IPV4 addresses. */
1485         dip = _mm_shuffle_epi8(dip, bswap_mask);
1486
1487         /* if all 4 packets are IPV4. */
1488         if (likely(ipv4_flag)) {
1489                 rte_lpm_lookupx4(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dip,
1490                                 dst.u32, portid);
1491
1492                 /* get rid of unused upper 16 bit for each dport. */
1493                 dst.x = _mm_packs_epi32(dst.x, dst.x);
1494                 *(uint64_t *)dprt = dst.u64[0];
1495         } else {
1496                 dst.x = dip;
1497                 dprt[0] = get_dst_port(pkt[0], dst.u32[0], portid);
1498                 dprt[1] = get_dst_port(pkt[1], dst.u32[1], portid);
1499                 dprt[2] = get_dst_port(pkt[2], dst.u32[2], portid);
1500                 dprt[3] = get_dst_port(pkt[3], dst.u32[3], portid);
1501         }
1502 }
1503
1504 /*
1505  * Update source and destination MAC addresses in the ethernet header.
1506  * Perform RFC1812 checks and updates for IPV4 packets.
1507  */
1508 static inline void
1509 processx4_step3(struct rte_mbuf *pkt[FWDSTEP], uint16_t dst_port[FWDSTEP])
1510 {
1511         __m128i te[FWDSTEP];
1512         __m128i ve[FWDSTEP];
1513         __m128i *p[FWDSTEP];
1514
1515         p[0] = rte_pktmbuf_mtod(pkt[0], __m128i *);
1516         p[1] = rte_pktmbuf_mtod(pkt[1], __m128i *);
1517         p[2] = rte_pktmbuf_mtod(pkt[2], __m128i *);
1518         p[3] = rte_pktmbuf_mtod(pkt[3], __m128i *);
1519
1520         ve[0] = val_eth[dst_port[0]];
1521         te[0] = _mm_load_si128(p[0]);
1522
1523         ve[1] = val_eth[dst_port[1]];
1524         te[1] = _mm_load_si128(p[1]);
1525
1526         ve[2] = val_eth[dst_port[2]];
1527         te[2] = _mm_load_si128(p[2]);
1528
1529         ve[3] = val_eth[dst_port[3]];
1530         te[3] = _mm_load_si128(p[3]);
1531
1532         /* Update first 12 bytes, keep rest bytes intact. */
1533         te[0] =  _mm_blend_epi16(te[0], ve[0], MASK_ETH);
1534         te[1] =  _mm_blend_epi16(te[1], ve[1], MASK_ETH);
1535         te[2] =  _mm_blend_epi16(te[2], ve[2], MASK_ETH);
1536         te[3] =  _mm_blend_epi16(te[3], ve[3], MASK_ETH);
1537
1538         _mm_store_si128(p[0], te[0]);
1539         _mm_store_si128(p[1], te[1]);
1540         _mm_store_si128(p[2], te[2]);
1541         _mm_store_si128(p[3], te[3]);
1542
1543         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[0] + 1),
1544                         &dst_port[0], pkt[0]->packet_type);
1545         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[1] + 1),
1546                         &dst_port[1], pkt[1]->packet_type);
1547         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[2] + 1),
1548                         &dst_port[2], pkt[2]->packet_type);
1549         rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[3] + 1),
1550                         &dst_port[3], pkt[3]->packet_type);
1551 }
1552
1553 /*
1554  * We group consecutive packets with the same destionation port into one burst.
1555  * To avoid extra latency this is done together with some other packet
1556  * processing, but after we made a final decision about packet's destination.
1557  * To do this we maintain:
1558  * pnum - array of number of consecutive packets with the same dest port for
1559  * each packet in the input burst.
1560  * lp - pointer to the last updated element in the pnum.
1561  * dlp - dest port value lp corresponds to.
1562  */
1563
1564 #define GRPSZ   (1 << FWDSTEP)
1565 #define GRPMSK  (GRPSZ - 1)
1566
1567 #define GROUP_PORT_STEP(dlp, dcp, lp, pn, idx)  do { \
1568         if (likely((dlp) == (dcp)[(idx)])) {         \
1569                 (lp)[0]++;                           \
1570         } else {                                     \
1571                 (dlp) = (dcp)[idx];                  \
1572                 (lp) = (pn) + (idx);                 \
1573                 (lp)[0] = 1;                         \
1574         }                                            \
1575 } while (0)
1576
1577 /*
1578  * Group consecutive packets with the same destination port in bursts of 4.
1579  * Suppose we have array of destionation ports:
1580  * dst_port[] = {a, b, c, d,, e, ... }
1581  * dp1 should contain: <a, b, c, d>, dp2: <b, c, d, e>.
1582  * We doing 4 comparisons at once and the result is 4 bit mask.
1583  * This mask is used as an index into prebuild array of pnum values.
1584  */
1585 static inline uint16_t *
1586 port_groupx4(uint16_t pn[FWDSTEP + 1], uint16_t *lp, __m128i dp1, __m128i dp2)
1587 {
1588         static const struct {
1589                 uint64_t pnum; /* prebuild 4 values for pnum[]. */
1590                 int32_t  idx;  /* index for new last updated elemnet. */
1591                 uint16_t lpv;  /* add value to the last updated element. */
1592         } gptbl[GRPSZ] = {
1593         {
1594                 /* 0: a != b, b != c, c != d, d != e */
1595                 .pnum = UINT64_C(0x0001000100010001),
1596                 .idx = 4,
1597                 .lpv = 0,
1598         },
1599         {
1600                 /* 1: a == b, b != c, c != d, d != e */
1601                 .pnum = UINT64_C(0x0001000100010002),
1602                 .idx = 4,
1603                 .lpv = 1,
1604         },
1605         {
1606                 /* 2: a != b, b == c, c != d, d != e */
1607                 .pnum = UINT64_C(0x0001000100020001),
1608                 .idx = 4,
1609                 .lpv = 0,
1610         },
1611         {
1612                 /* 3: a == b, b == c, c != d, d != e */
1613                 .pnum = UINT64_C(0x0001000100020003),
1614                 .idx = 4,
1615                 .lpv = 2,
1616         },
1617         {
1618                 /* 4: a != b, b != c, c == d, d != e */
1619                 .pnum = UINT64_C(0x0001000200010001),
1620                 .idx = 4,
1621                 .lpv = 0,
1622         },
1623         {
1624                 /* 5: a == b, b != c, c == d, d != e */
1625                 .pnum = UINT64_C(0x0001000200010002),
1626                 .idx = 4,
1627                 .lpv = 1,
1628         },
1629         {
1630                 /* 6: a != b, b == c, c == d, d != e */
1631                 .pnum = UINT64_C(0x0001000200030001),
1632                 .idx = 4,
1633                 .lpv = 0,
1634         },
1635         {
1636                 /* 7: a == b, b == c, c == d, d != e */
1637                 .pnum = UINT64_C(0x0001000200030004),
1638                 .idx = 4,
1639                 .lpv = 3,
1640         },
1641         {
1642                 /* 8: a != b, b != c, c != d, d == e */
1643                 .pnum = UINT64_C(0x0002000100010001),
1644                 .idx = 3,
1645                 .lpv = 0,
1646         },
1647         {
1648                 /* 9: a == b, b != c, c != d, d == e */
1649                 .pnum = UINT64_C(0x0002000100010002),
1650                 .idx = 3,
1651                 .lpv = 1,
1652         },
1653         {
1654                 /* 0xa: a != b, b == c, c != d, d == e */
1655                 .pnum = UINT64_C(0x0002000100020001),
1656                 .idx = 3,
1657                 .lpv = 0,
1658         },
1659         {
1660                 /* 0xb: a == b, b == c, c != d, d == e */
1661                 .pnum = UINT64_C(0x0002000100020003),
1662                 .idx = 3,
1663                 .lpv = 2,
1664         },
1665         {
1666                 /* 0xc: a != b, b != c, c == d, d == e */
1667                 .pnum = UINT64_C(0x0002000300010001),
1668                 .idx = 2,
1669                 .lpv = 0,
1670         },
1671         {
1672                 /* 0xd: a == b, b != c, c == d, d == e */
1673                 .pnum = UINT64_C(0x0002000300010002),
1674                 .idx = 2,
1675                 .lpv = 1,
1676         },
1677         {
1678                 /* 0xe: a != b, b == c, c == d, d == e */
1679                 .pnum = UINT64_C(0x0002000300040001),
1680                 .idx = 1,
1681                 .lpv = 0,
1682         },
1683         {
1684                 /* 0xf: a == b, b == c, c == d, d == e */
1685                 .pnum = UINT64_C(0x0002000300040005),
1686                 .idx = 0,
1687                 .lpv = 4,
1688         },
1689         };
1690
1691         union {
1692                 uint16_t u16[FWDSTEP + 1];
1693                 uint64_t u64;
1694         } *pnum = (void *)pn;
1695
1696         int32_t v;
1697
1698         dp1 = _mm_cmpeq_epi16(dp1, dp2);
1699         dp1 = _mm_unpacklo_epi16(dp1, dp1);
1700         v = _mm_movemask_ps((__m128)dp1);
1701
1702         /* update last port counter. */
1703         lp[0] += gptbl[v].lpv;
1704
1705         /* if dest port value has changed. */
1706         if (v != GRPMSK) {
1707                 pnum->u64 = gptbl[v].pnum;
1708                 pnum->u16[FWDSTEP] = 1;
1709                 lp = pnum->u16 + gptbl[v].idx;
1710         }
1711
1712         return lp;
1713 }
1714
1715 #endif /* APP_LOOKUP_METHOD */
1716
1717 static void
1718 process_burst(struct rte_mbuf *pkts_burst[MAX_PKT_BURST], int nb_rx,
1719                 uint16_t portid)
1720 {
1721
1722         int j;
1723
1724 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1725         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1726         int32_t k;
1727         uint16_t dlp;
1728         uint16_t *lp;
1729         uint16_t dst_port[MAX_PKT_BURST];
1730         __m128i dip[MAX_PKT_BURST / FWDSTEP];
1731         uint32_t ipv4_flag[MAX_PKT_BURST / FWDSTEP];
1732         uint16_t pnum[MAX_PKT_BURST + 1];
1733 #endif
1734
1735
1736 #if (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
1737 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1738         {
1739                 /*
1740                  * Send nb_rx - nb_rx%8 packets
1741                  * in groups of 8.
1742                  */
1743                 int32_t n = RTE_ALIGN_FLOOR(nb_rx, 8);
1744
1745                 for (j = 0; j < n; j += 8) {
1746                         uint32_t pkt_type =
1747                                 pkts_burst[j]->packet_type &
1748                                 pkts_burst[j+1]->packet_type &
1749                                 pkts_burst[j+2]->packet_type &
1750                                 pkts_burst[j+3]->packet_type &
1751                                 pkts_burst[j+4]->packet_type &
1752                                 pkts_burst[j+5]->packet_type &
1753                                 pkts_burst[j+6]->packet_type &
1754                                 pkts_burst[j+7]->packet_type;
1755                         if (pkt_type & RTE_PTYPE_L3_IPV4) {
1756                                 simple_ipv4_fwd_8pkts(&pkts_burst[j], portid);
1757                         } else if (pkt_type &
1758                                 RTE_PTYPE_L3_IPV6) {
1759                                 simple_ipv6_fwd_8pkts(&pkts_burst[j], portid);
1760                         } else {
1761                                 l3fwd_simple_forward(pkts_burst[j], portid);
1762                                 l3fwd_simple_forward(pkts_burst[j+1], portid);
1763                                 l3fwd_simple_forward(pkts_burst[j+2], portid);
1764                                 l3fwd_simple_forward(pkts_burst[j+3], portid);
1765                                 l3fwd_simple_forward(pkts_burst[j+4], portid);
1766                                 l3fwd_simple_forward(pkts_burst[j+5], portid);
1767                                 l3fwd_simple_forward(pkts_burst[j+6], portid);
1768                                 l3fwd_simple_forward(pkts_burst[j+7], portid);
1769                         }
1770                 }
1771                 for (; j < nb_rx ; j++)
1772                         l3fwd_simple_forward(pkts_burst[j], portid);
1773         }
1774 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1775
1776         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1777         for (j = 0; j != k; j += FWDSTEP)
1778                 processx4_step1(&pkts_burst[j], &dip[j / FWDSTEP],
1779                                 &ipv4_flag[j / FWDSTEP]);
1780
1781         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1782         for (j = 0; j != k; j += FWDSTEP)
1783                 processx4_step2(dip[j / FWDSTEP], ipv4_flag[j / FWDSTEP],
1784                                 portid, &pkts_burst[j], &dst_port[j]);
1785
1786         /*
1787          * Finish packet processing and group consecutive
1788          * packets with the same destination port.
1789          */
1790         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1791         if (k != 0) {
1792                 __m128i dp1, dp2;
1793
1794                 lp = pnum;
1795                 lp[0] = 1;
1796
1797                 processx4_step3(pkts_burst, dst_port);
1798
1799                 /* dp1: <d[0], d[1], d[2], d[3], ... > */
1800                 dp1 = _mm_loadu_si128((__m128i *)dst_port);
1801
1802                 for (j = FWDSTEP; j != k; j += FWDSTEP) {
1803                         processx4_step3(&pkts_burst[j], &dst_port[j]);
1804
1805                         /*
1806                          * dp2:
1807                          * <d[j-3], d[j-2], d[j-1], d[j], ... >
1808                          */
1809                         dp2 = _mm_loadu_si128(
1810                                         (__m128i *)&dst_port[j - FWDSTEP + 1]);
1811                         lp  = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1812
1813                         /*
1814                          * dp1:
1815                          * <d[j], d[j+1], d[j+2], d[j+3], ... >
1816                          */
1817                         dp1 = _mm_srli_si128(dp2, (FWDSTEP - 1) *
1818                                         sizeof(dst_port[0]));
1819                 }
1820
1821                 /*
1822                  * dp2: <d[j-3], d[j-2], d[j-1], d[j-1], ... >
1823                  */
1824                 dp2 = _mm_shufflelo_epi16(dp1, 0xf9);
1825                 lp  = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1826
1827                 /*
1828                  * remove values added by the last repeated
1829                  * dst port.
1830                  */
1831                 lp[0]--;
1832                 dlp = dst_port[j - 1];
1833         } else {
1834                 /* set dlp and lp to the never used values. */
1835                 dlp = BAD_PORT - 1;
1836                 lp = pnum + MAX_PKT_BURST;
1837         }
1838
1839         /* Process up to last 3 packets one by one. */
1840         switch (nb_rx % FWDSTEP) {
1841         case 3:
1842                 process_packet(pkts_burst[j], dst_port + j, portid);
1843                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1844                 j++;
1845                 /* fall-through */
1846         case 2:
1847                 process_packet(pkts_burst[j], dst_port + j, portid);
1848                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1849                 j++;
1850                 /* fall-through */
1851         case 1:
1852                 process_packet(pkts_burst[j], dst_port + j, portid);
1853                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1854                 j++;
1855         }
1856
1857         /*
1858          * Send packets out, through destination port.
1859          * Consecuteve pacekts with the same destination port
1860          * are already grouped together.
1861          * If destination port for the packet equals BAD_PORT,
1862          * then free the packet without sending it out.
1863          */
1864         for (j = 0; j < nb_rx; j += k) {
1865
1866                 int32_t m;
1867                 uint16_t pn;
1868
1869                 pn = dst_port[j];
1870                 k = pnum[j];
1871
1872                 if (likely(pn != BAD_PORT))
1873                         send_packetsx4(pn, pkts_burst + j, k);
1874                 else
1875                         for (m = j; m != j + k; m++)
1876                                 rte_pktmbuf_free(pkts_burst[m]);
1877
1878         }
1879
1880 #endif /* APP_LOOKUP_METHOD */
1881 #else /* ENABLE_MULTI_BUFFER_OPTIMIZE == 0 */
1882
1883         /* Prefetch first packets */
1884         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++)
1885                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[j], void *));
1886
1887         /* Prefetch and forward already prefetched packets */
1888         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1889                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1890                                 j + PREFETCH_OFFSET], void *));
1891                 l3fwd_simple_forward(pkts_burst[j], portid);
1892         }
1893
1894         /* Forward remaining prefetched packets */
1895         for (; j < nb_rx; j++)
1896                 l3fwd_simple_forward(pkts_burst[j], portid);
1897
1898 #endif /* ENABLE_MULTI_BUFFER_OPTIMIZE */
1899
1900 }
1901
1902 #if (APP_CPU_LOAD > 0)
1903
1904 /*
1905  * CPU-load stats collector
1906  */
1907 static int
1908 cpu_load_collector(__rte_unused void *arg) {
1909         unsigned i, j, k;
1910         uint64_t hits;
1911         uint64_t prev_tsc, diff_tsc, cur_tsc;
1912         uint64_t total[MAX_CPU] = { 0 };
1913         unsigned min_cpu = MAX_CPU;
1914         unsigned max_cpu = 0;
1915         unsigned cpu_id;
1916         int busy_total = 0;
1917         int busy_flag = 0;
1918
1919         unsigned int n_thread_per_cpu[MAX_CPU] = { 0 };
1920         struct thread_conf *thread_per_cpu[MAX_CPU][MAX_THREAD];
1921
1922         struct thread_conf *thread_conf;
1923
1924         const uint64_t interval_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1925                 US_PER_S * CPU_LOAD_TIMEOUT_US;
1926
1927         prev_tsc = 0;
1928         /*
1929          * Wait for all threads
1930          */
1931
1932         printf("Waiting for %d rx threads and %d tx threads\n", n_rx_thread,
1933                         n_tx_thread);
1934
1935         while (rte_atomic16_read(&rx_counter) < n_rx_thread)
1936                 rte_pause();
1937
1938         while (rte_atomic16_read(&tx_counter) < n_tx_thread)
1939                 rte_pause();
1940
1941         for (i = 0; i < n_rx_thread; i++) {
1942
1943                 thread_conf = &rx_thread[i].conf;
1944                 cpu_id = thread_conf->cpu_id;
1945                 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1946
1947                 if (cpu_id > max_cpu)
1948                         max_cpu = cpu_id;
1949                 if (cpu_id < min_cpu)
1950                         min_cpu = cpu_id;
1951         }
1952         for (i = 0; i < n_tx_thread; i++) {
1953
1954                 thread_conf = &tx_thread[i].conf;
1955                 cpu_id = thread_conf->cpu_id;
1956                 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1957
1958                 if (thread_conf->cpu_id > max_cpu)
1959                         max_cpu = thread_conf->cpu_id;
1960                 if (thread_conf->cpu_id < min_cpu)
1961                         min_cpu = thread_conf->cpu_id;
1962         }
1963
1964         while (1) {
1965
1966                 cpu_load.counter++;
1967                 for (i = min_cpu; i <= max_cpu; i++) {
1968                         for (j = 0; j < MAX_CPU_COUNTER; j++) {
1969                                 for (k = 0; k < n_thread_per_cpu[i]; k++)
1970                                         if (thread_per_cpu[i][k]->busy[j]) {
1971                                                 busy_flag = 1;
1972                                                 break;
1973                                         }
1974                                 if (busy_flag) {
1975                                         cpu_load.hits[j][i]++;
1976                                         busy_total = 1;
1977                                         busy_flag = 0;
1978                                 }
1979                         }
1980
1981                         if (busy_total) {
1982                                 total[i]++;
1983                                 busy_total = 0;
1984                         }
1985                 }
1986
1987                 cur_tsc = rte_rdtsc();
1988
1989                 diff_tsc = cur_tsc - prev_tsc;
1990                 if (unlikely(diff_tsc > interval_tsc)) {
1991
1992                         printf("\033c");
1993
1994                         printf("Cpu usage for %d rx threads and %d tx threads:\n\n",
1995                                         n_rx_thread, n_tx_thread);
1996
1997                         printf("cpu#     proc%%  poll%%  overhead%%\n\n");
1998
1999                         for (i = min_cpu; i <= max_cpu; i++) {
2000                                 hits = 0;
2001                                 printf("CPU %d:", i);
2002                                 for (j = 0; j < MAX_CPU_COUNTER; j++) {
2003                                         printf("%7" PRIu64 "",
2004                                                         cpu_load.hits[j][i] * 100 / cpu_load.counter);
2005                                         hits += cpu_load.hits[j][i];
2006                                         cpu_load.hits[j][i] = 0;
2007                                 }
2008                                 printf("%7" PRIu64 "\n",
2009                                                 100 - total[i] * 100 / cpu_load.counter);
2010                                 total[i] = 0;
2011                         }
2012                         cpu_load.counter = 0;
2013
2014                         prev_tsc = cur_tsc;
2015                 }
2016
2017         }
2018 }
2019 #endif /* APP_CPU_LOAD */
2020
2021 /*
2022  * Null processing lthread loop
2023  *
2024  * This loop is used to start empty scheduler on lcore.
2025  */
2026 static void
2027 lthread_null(__rte_unused void *args)
2028 {
2029         int lcore_id = rte_lcore_id();
2030
2031         RTE_LOG(INFO, L3FWD, "Starting scheduler on lcore %d.\n", lcore_id);
2032         lthread_exit(NULL);
2033 }
2034
2035 /* main processing loop */
2036 static void
2037 lthread_tx_per_ring(void *dummy)
2038 {
2039         int nb_rx;
2040         uint16_t portid;
2041         struct rte_ring *ring;
2042         struct thread_tx_conf *tx_conf;
2043         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2044         struct lthread_cond *ready;
2045
2046         tx_conf = (struct thread_tx_conf *)dummy;
2047         ring = tx_conf->ring;
2048         ready = *tx_conf->ready;
2049
2050         lthread_set_data((void *)tx_conf);
2051
2052         /*
2053          * Move this lthread to lcore
2054          */
2055         lthread_set_affinity(tx_conf->conf.lcore_id);
2056
2057         RTE_LOG(INFO, L3FWD, "entering main tx loop on lcore %u\n", rte_lcore_id());
2058
2059         nb_rx = 0;
2060         rte_atomic16_inc(&tx_counter);
2061         while (1) {
2062
2063                 /*
2064                  * Read packet from ring
2065                  */
2066                 SET_CPU_BUSY(tx_conf, CPU_POLL);
2067                 nb_rx = rte_ring_sc_dequeue_burst(ring, (void **)pkts_burst,
2068                                 MAX_PKT_BURST, NULL);
2069                 SET_CPU_IDLE(tx_conf, CPU_POLL);
2070
2071                 if (nb_rx > 0) {
2072                         SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2073                         portid = pkts_burst[0]->port;
2074                         process_burst(pkts_burst, nb_rx, portid);
2075                         SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2076                         lthread_yield();
2077                 } else
2078                         lthread_cond_wait(ready, 0);
2079
2080         }
2081 }
2082
2083 /*
2084  * Main tx-lthreads spawner lthread.
2085  *
2086  * This lthread is used to spawn one new lthread per ring from producers.
2087  *
2088  */
2089 static void
2090 lthread_tx(void *args)
2091 {
2092         struct lthread *lt;
2093
2094         unsigned lcore_id;
2095         uint16_t portid;
2096         struct thread_tx_conf *tx_conf;
2097
2098         tx_conf = (struct thread_tx_conf *)args;
2099         lthread_set_data((void *)tx_conf);
2100
2101         /*
2102          * Move this lthread to the selected lcore
2103          */
2104         lthread_set_affinity(tx_conf->conf.lcore_id);
2105
2106         /*
2107          * Spawn tx readers (one per input ring)
2108          */
2109         lthread_create(&lt, tx_conf->conf.lcore_id, lthread_tx_per_ring,
2110                         (void *)tx_conf);
2111
2112         lcore_id = rte_lcore_id();
2113
2114         RTE_LOG(INFO, L3FWD, "Entering Tx main loop on lcore %u\n", lcore_id);
2115
2116         tx_conf->conf.cpu_id = sched_getcpu();
2117         while (1) {
2118
2119                 lthread_sleep(BURST_TX_DRAIN_US * 1000);
2120
2121                 /*
2122                  * TX burst queue drain
2123                  */
2124                 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2125                         if (tx_conf->tx_mbufs[portid].len == 0)
2126                                 continue;
2127                         SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2128                         send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2129                         SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2130                         tx_conf->tx_mbufs[portid].len = 0;
2131                 }
2132
2133         }
2134 }
2135
2136 static void
2137 lthread_rx(void *dummy)
2138 {
2139         int ret;
2140         uint16_t nb_rx;
2141         int i;
2142         uint16_t portid;
2143         uint8_t queueid;
2144         int worker_id;
2145         int len[RTE_MAX_LCORE] = { 0 };
2146         int old_len, new_len;
2147         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2148         struct thread_rx_conf *rx_conf;
2149
2150         rx_conf = (struct thread_rx_conf *)dummy;
2151         lthread_set_data((void *)rx_conf);
2152
2153         /*
2154          * Move this lthread to lcore
2155          */
2156         lthread_set_affinity(rx_conf->conf.lcore_id);
2157
2158         if (rx_conf->n_rx_queue == 0) {
2159                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", rte_lcore_id());
2160                 return;
2161         }
2162
2163         RTE_LOG(INFO, L3FWD, "Entering main Rx loop on lcore %u\n", rte_lcore_id());
2164
2165         for (i = 0; i < rx_conf->n_rx_queue; i++) {
2166
2167                 portid = rx_conf->rx_queue_list[i].port_id;
2168                 queueid = rx_conf->rx_queue_list[i].queue_id;
2169                 RTE_LOG(INFO, L3FWD,
2170                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2171                                 rte_lcore_id(), portid, queueid);
2172         }
2173
2174         /*
2175          * Init all condition variables (one per rx thread)
2176          */
2177         for (i = 0; i < rx_conf->n_rx_queue; i++)
2178                 lthread_cond_init(NULL, &rx_conf->ready[i], NULL);
2179
2180         worker_id = 0;
2181
2182         rx_conf->conf.cpu_id = sched_getcpu();
2183         rte_atomic16_inc(&rx_counter);
2184         while (1) {
2185
2186                 /*
2187                  * Read packet from RX queues
2188                  */
2189                 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2190                         portid = rx_conf->rx_queue_list[i].port_id;
2191                         queueid = rx_conf->rx_queue_list[i].queue_id;
2192
2193                         SET_CPU_BUSY(rx_conf, CPU_POLL);
2194                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2195                                 MAX_PKT_BURST);
2196                         SET_CPU_IDLE(rx_conf, CPU_POLL);
2197
2198                         if (nb_rx != 0) {
2199                                 worker_id = (worker_id + 1) % rx_conf->n_ring;
2200                                 old_len = len[worker_id];
2201
2202                                 SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2203                                 ret = rte_ring_sp_enqueue_burst(
2204                                                 rx_conf->ring[worker_id],
2205                                                 (void **) pkts_burst,
2206                                                 nb_rx, NULL);
2207
2208                                 new_len = old_len + ret;
2209
2210                                 if (new_len >= BURST_SIZE) {
2211                                         lthread_cond_signal(rx_conf->ready[worker_id]);
2212                                         new_len = 0;
2213                                 }
2214
2215                                 len[worker_id] = new_len;
2216
2217                                 if (unlikely(ret < nb_rx)) {
2218                                         uint32_t k;
2219
2220                                         for (k = ret; k < nb_rx; k++) {
2221                                                 struct rte_mbuf *m = pkts_burst[k];
2222
2223                                                 rte_pktmbuf_free(m);
2224                                         }
2225                                 }
2226                                 SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2227                         }
2228
2229                         lthread_yield();
2230                 }
2231         }
2232 }
2233
2234 /*
2235  * Start scheduler with initial lthread on lcore
2236  *
2237  * This lthread loop spawns all rx and tx lthreads on master lcore
2238  */
2239
2240 static void
2241 lthread_spawner(__rte_unused void *arg) {
2242         struct lthread *lt[MAX_THREAD];
2243         int i;
2244         int n_thread = 0;
2245
2246         printf("Entering lthread_spawner\n");
2247
2248         /*
2249          * Create producers (rx threads) on default lcore
2250          */
2251         for (i = 0; i < n_rx_thread; i++) {
2252                 rx_thread[i].conf.thread_id = i;
2253                 lthread_create(&lt[n_thread], -1, lthread_rx,
2254                                 (void *)&rx_thread[i]);
2255                 n_thread++;
2256         }
2257
2258         /*
2259          * Wait for all producers. Until some producers can be started on the same
2260          * scheduler as this lthread, yielding is required to let them to run and
2261          * prevent deadlock here.
2262          */
2263         while (rte_atomic16_read(&rx_counter) < n_rx_thread)
2264                 lthread_sleep(100000);
2265
2266         /*
2267          * Create consumers (tx threads) on default lcore_id
2268          */
2269         for (i = 0; i < n_tx_thread; i++) {
2270                 tx_thread[i].conf.thread_id = i;
2271                 lthread_create(&lt[n_thread], -1, lthread_tx,
2272                                 (void *)&tx_thread[i]);
2273                 n_thread++;
2274         }
2275
2276         /*
2277          * Wait for all threads finished
2278          */
2279         for (i = 0; i < n_thread; i++)
2280                 lthread_join(lt[i], NULL);
2281
2282 }
2283
2284 /*
2285  * Start master scheduler with initial lthread spawning rx and tx lthreads
2286  * (main_lthread_master).
2287  */
2288 static int
2289 lthread_master_spawner(__rte_unused void *arg) {
2290         struct lthread *lt;
2291         int lcore_id = rte_lcore_id();
2292
2293         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2294         lthread_create(&lt, -1, lthread_spawner, NULL);
2295         lthread_run();
2296
2297         return 0;
2298 }
2299
2300 /*
2301  * Start scheduler on lcore.
2302  */
2303 static int
2304 sched_spawner(__rte_unused void *arg) {
2305         struct lthread *lt;
2306         int lcore_id = rte_lcore_id();
2307
2308 #if (APP_CPU_LOAD)
2309         if (lcore_id == cpu_load_lcore_id) {
2310                 cpu_load_collector(arg);
2311                 return 0;
2312         }
2313 #endif /* APP_CPU_LOAD */
2314
2315         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2316         lthread_create(&lt, -1, lthread_null, NULL);
2317         lthread_run();
2318
2319         return 0;
2320 }
2321
2322 /* main processing loop */
2323 static int
2324 pthread_tx(void *dummy)
2325 {
2326         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2327         uint64_t prev_tsc, diff_tsc, cur_tsc;
2328         int nb_rx;
2329         uint16_t portid;
2330         struct thread_tx_conf *tx_conf;
2331
2332         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
2333                 US_PER_S * BURST_TX_DRAIN_US;
2334
2335         prev_tsc = 0;
2336
2337         tx_conf = (struct thread_tx_conf *)dummy;
2338
2339         RTE_LOG(INFO, L3FWD, "Entering main Tx loop on lcore %u\n", rte_lcore_id());
2340
2341         tx_conf->conf.cpu_id = sched_getcpu();
2342         rte_atomic16_inc(&tx_counter);
2343         while (1) {
2344
2345                 cur_tsc = rte_rdtsc();
2346
2347                 /*
2348                  * TX burst queue drain
2349                  */
2350                 diff_tsc = cur_tsc - prev_tsc;
2351                 if (unlikely(diff_tsc > drain_tsc)) {
2352
2353                         /*
2354                          * This could be optimized (use queueid instead of
2355                          * portid), but it is not called so often
2356                          */
2357                         SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2358                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2359                                 if (tx_conf->tx_mbufs[portid].len == 0)
2360                                         continue;
2361                                 send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2362                                 tx_conf->tx_mbufs[portid].len = 0;
2363                         }
2364                         SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2365
2366                         prev_tsc = cur_tsc;
2367                 }
2368
2369                 /*
2370                  * Read packet from ring
2371                  */
2372                 SET_CPU_BUSY(tx_conf, CPU_POLL);
2373                 nb_rx = rte_ring_sc_dequeue_burst(tx_conf->ring,
2374                                 (void **)pkts_burst, MAX_PKT_BURST, NULL);
2375                 SET_CPU_IDLE(tx_conf, CPU_POLL);
2376
2377                 if (unlikely(nb_rx == 0)) {
2378                         sched_yield();
2379                         continue;
2380                 }
2381
2382                 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2383                 portid = pkts_burst[0]->port;
2384                 process_burst(pkts_burst, nb_rx, portid);
2385                 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2386
2387         }
2388 }
2389
2390 static int
2391 pthread_rx(void *dummy)
2392 {
2393         int i;
2394         int worker_id;
2395         uint32_t n;
2396         uint32_t nb_rx;
2397         unsigned lcore_id;
2398         uint8_t queueid;
2399         uint16_t portid;
2400         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2401
2402         struct thread_rx_conf *rx_conf;
2403
2404         lcore_id = rte_lcore_id();
2405         rx_conf = (struct thread_rx_conf *)dummy;
2406
2407         if (rx_conf->n_rx_queue == 0) {
2408                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
2409                 return 0;
2410         }
2411
2412         RTE_LOG(INFO, L3FWD, "entering main rx loop on lcore %u\n", lcore_id);
2413
2414         for (i = 0; i < rx_conf->n_rx_queue; i++) {
2415
2416                 portid = rx_conf->rx_queue_list[i].port_id;
2417                 queueid = rx_conf->rx_queue_list[i].queue_id;
2418                 RTE_LOG(INFO, L3FWD,
2419                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2420                                 lcore_id, portid, queueid);
2421         }
2422
2423         worker_id = 0;
2424         rx_conf->conf.cpu_id = sched_getcpu();
2425         rte_atomic16_inc(&rx_counter);
2426         while (1) {
2427
2428                 /*
2429                  * Read packet from RX queues
2430                  */
2431                 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2432                         portid = rx_conf->rx_queue_list[i].port_id;
2433                         queueid = rx_conf->rx_queue_list[i].queue_id;
2434
2435                         SET_CPU_BUSY(rx_conf, CPU_POLL);
2436                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2437                                 MAX_PKT_BURST);
2438                         SET_CPU_IDLE(rx_conf, CPU_POLL);
2439
2440                         if (nb_rx == 0) {
2441                                 sched_yield();
2442                                 continue;
2443                         }
2444
2445                         SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2446                         worker_id = (worker_id + 1) % rx_conf->n_ring;
2447                         n = rte_ring_sp_enqueue_burst(rx_conf->ring[worker_id],
2448                                         (void **)pkts_burst, nb_rx, NULL);
2449
2450                         if (unlikely(n != nb_rx)) {
2451                                 uint32_t k;
2452
2453                                 for (k = n; k < nb_rx; k++) {
2454                                         struct rte_mbuf *m = pkts_burst[k];
2455
2456                                         rte_pktmbuf_free(m);
2457                                 }
2458                         }
2459
2460                         SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2461
2462                 }
2463         }
2464 }
2465
2466 /*
2467  * P-Thread spawner.
2468  */
2469 static int
2470 pthread_run(__rte_unused void *arg) {
2471         int lcore_id = rte_lcore_id();
2472         int i;
2473
2474         for (i = 0; i < n_rx_thread; i++)
2475                 if (rx_thread[i].conf.lcore_id == lcore_id) {
2476                         printf("Start rx thread on %d...\n", lcore_id);
2477                         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2478                         RTE_PER_LCORE(lcore_conf)->data = (void *)&rx_thread[i];
2479                         pthread_rx((void *)&rx_thread[i]);
2480                         return 0;
2481                 }
2482
2483         for (i = 0; i < n_tx_thread; i++)
2484                 if (tx_thread[i].conf.lcore_id == lcore_id) {
2485                         printf("Start tx thread on %d...\n", lcore_id);
2486                         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2487                         RTE_PER_LCORE(lcore_conf)->data = (void *)&tx_thread[i];
2488                         pthread_tx((void *)&tx_thread[i]);
2489                         return 0;
2490                 }
2491
2492 #if (APP_CPU_LOAD)
2493         if (lcore_id == cpu_load_lcore_id)
2494                 cpu_load_collector(arg);
2495 #endif /* APP_CPU_LOAD */
2496
2497         return 0;
2498 }
2499
2500 static int
2501 check_lcore_params(void)
2502 {
2503         uint8_t queue, lcore;
2504         uint16_t i;
2505         int socketid;
2506
2507         for (i = 0; i < nb_rx_thread_params; ++i) {
2508                 queue = rx_thread_params[i].queue_id;
2509                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
2510                         printf("invalid queue number: %hhu\n", queue);
2511                         return -1;
2512                 }
2513                 lcore = rx_thread_params[i].lcore_id;
2514                 if (!rte_lcore_is_enabled(lcore)) {
2515                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
2516                         return -1;
2517                 }
2518                 socketid = rte_lcore_to_socket_id(lcore);
2519                 if ((socketid != 0) && (numa_on == 0))
2520                         printf("warning: lcore %hhu is on socket %d with numa off\n",
2521                                 lcore, socketid);
2522         }
2523         return 0;
2524 }
2525
2526 static int
2527 check_port_config(const unsigned nb_ports)
2528 {
2529         unsigned portid;
2530         uint16_t i;
2531
2532         for (i = 0; i < nb_rx_thread_params; ++i) {
2533                 portid = rx_thread_params[i].port_id;
2534                 if ((enabled_port_mask & (1 << portid)) == 0) {
2535                         printf("port %u is not enabled in port mask\n", portid);
2536                         return -1;
2537                 }
2538                 if (portid >= nb_ports) {
2539                         printf("port %u is not present on the board\n", portid);
2540                         return -1;
2541                 }
2542         }
2543         return 0;
2544 }
2545
2546 static uint8_t
2547 get_port_n_rx_queues(const uint16_t port)
2548 {
2549         int queue = -1;
2550         uint16_t i;
2551
2552         for (i = 0; i < nb_rx_thread_params; ++i)
2553                 if (rx_thread_params[i].port_id == port &&
2554                                 rx_thread_params[i].queue_id > queue)
2555                         queue = rx_thread_params[i].queue_id;
2556
2557         return (uint8_t)(++queue);
2558 }
2559
2560 static int
2561 init_rx_rings(void)
2562 {
2563         unsigned socket_io;
2564         struct thread_rx_conf *rx_conf;
2565         struct thread_tx_conf *tx_conf;
2566         unsigned rx_thread_id, tx_thread_id;
2567         char name[256];
2568         struct rte_ring *ring = NULL;
2569
2570         for (tx_thread_id = 0; tx_thread_id < n_tx_thread; tx_thread_id++) {
2571
2572                 tx_conf = &tx_thread[tx_thread_id];
2573
2574                 printf("Connecting tx-thread %d with rx-thread %d\n", tx_thread_id,
2575                                 tx_conf->conf.thread_id);
2576
2577                 rx_thread_id = tx_conf->conf.thread_id;
2578                 if (rx_thread_id > n_tx_thread) {
2579                         printf("connection from tx-thread %u to rx-thread %u fails "
2580                                         "(rx-thread not defined)\n", tx_thread_id, rx_thread_id);
2581                         return -1;
2582                 }
2583
2584                 rx_conf = &rx_thread[rx_thread_id];
2585                 socket_io = rte_lcore_to_socket_id(rx_conf->conf.lcore_id);
2586
2587                 snprintf(name, sizeof(name), "app_ring_s%u_rx%u_tx%u",
2588                                 socket_io, rx_thread_id, tx_thread_id);
2589
2590                 ring = rte_ring_create(name, 1024 * 4, socket_io,
2591                                 RING_F_SP_ENQ | RING_F_SC_DEQ);
2592
2593                 if (ring == NULL) {
2594                         rte_panic("Cannot create ring to connect rx-thread %u "
2595                                         "with tx-thread %u\n", rx_thread_id, tx_thread_id);
2596                 }
2597
2598                 rx_conf->ring[rx_conf->n_ring] = ring;
2599
2600                 tx_conf->ring = ring;
2601                 tx_conf->ready = &rx_conf->ready[rx_conf->n_ring];
2602
2603                 rx_conf->n_ring++;
2604         }
2605         return 0;
2606 }
2607
2608 static int
2609 init_rx_queues(void)
2610 {
2611         uint16_t i, nb_rx_queue;
2612         uint8_t thread;
2613
2614         n_rx_thread = 0;
2615
2616         for (i = 0; i < nb_rx_thread_params; ++i) {
2617                 thread = rx_thread_params[i].thread_id;
2618                 nb_rx_queue = rx_thread[thread].n_rx_queue;
2619
2620                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
2621                         printf("error: too many queues (%u) for thread: %u\n",
2622                                 (unsigned)nb_rx_queue + 1, (unsigned)thread);
2623                         return -1;
2624                 }
2625
2626                 rx_thread[thread].conf.thread_id = thread;
2627                 rx_thread[thread].conf.lcore_id = rx_thread_params[i].lcore_id;
2628                 rx_thread[thread].rx_queue_list[nb_rx_queue].port_id =
2629                         rx_thread_params[i].port_id;
2630                 rx_thread[thread].rx_queue_list[nb_rx_queue].queue_id =
2631                         rx_thread_params[i].queue_id;
2632                 rx_thread[thread].n_rx_queue++;
2633
2634                 if (thread >= n_rx_thread)
2635                         n_rx_thread = thread + 1;
2636
2637         }
2638         return 0;
2639 }
2640
2641 static int
2642 init_tx_threads(void)
2643 {
2644         int i;
2645
2646         n_tx_thread = 0;
2647         for (i = 0; i < nb_tx_thread_params; ++i) {
2648                 tx_thread[n_tx_thread].conf.thread_id = tx_thread_params[i].thread_id;
2649                 tx_thread[n_tx_thread].conf.lcore_id = tx_thread_params[i].lcore_id;
2650                 n_tx_thread++;
2651         }
2652         return 0;
2653 }
2654
2655 /* display usage */
2656 static void
2657 print_usage(const char *prgname)
2658 {
2659         printf("%s [EAL options] -- -p PORTMASK -P"
2660                 "  [--rx (port,queue,lcore,thread)[,(port,queue,lcore,thread]]"
2661                 "  [--tx (lcore,thread)[,(lcore,thread]]"
2662                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
2663                 "  [--parse-ptype]\n\n"
2664                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
2665                 "  -P : enable promiscuous mode\n"
2666                 "  --rx (port,queue,lcore,thread): rx queues configuration\n"
2667                 "  --tx (lcore,thread): tx threads configuration\n"
2668                 "  --stat-lcore LCORE: use lcore for stat collector\n"
2669                 "  --eth-dest=X,MM:MM:MM:MM:MM:MM: optional, ethernet destination for port X\n"
2670                 "  --no-numa: optional, disable numa awareness\n"
2671                 "  --ipv6: optional, specify it if running ipv6 packets\n"
2672                 "  --enable-jumbo: enable jumbo frame"
2673                 " which max packet len is PKTLEN in decimal (64-9600)\n"
2674                 "  --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n"
2675                 "  --no-lthreads: turn off lthread model\n"
2676                 "  --parse-ptype: set to use software to analyze packet type\n\n",
2677                 prgname);
2678 }
2679
2680 static int parse_max_pkt_len(const char *pktlen)
2681 {
2682         char *end = NULL;
2683         unsigned long len;
2684
2685         /* parse decimal string */
2686         len = strtoul(pktlen, &end, 10);
2687         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
2688                 return -1;
2689
2690         if (len == 0)
2691                 return -1;
2692
2693         return len;
2694 }
2695
2696 static int
2697 parse_portmask(const char *portmask)
2698 {
2699         char *end = NULL;
2700         unsigned long pm;
2701
2702         /* parse hexadecimal string */
2703         pm = strtoul(portmask, &end, 16);
2704         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
2705                 return -1;
2706
2707         if (pm == 0)
2708                 return -1;
2709
2710         return pm;
2711 }
2712
2713 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2714 static int
2715 parse_hash_entry_number(const char *hash_entry_num)
2716 {
2717         char *end = NULL;
2718         unsigned long hash_en;
2719
2720         /* parse hexadecimal string */
2721         hash_en = strtoul(hash_entry_num, &end, 16);
2722         if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
2723                 return -1;
2724
2725         if (hash_en == 0)
2726                 return -1;
2727
2728         return hash_en;
2729 }
2730 #endif
2731
2732 static int
2733 parse_rx_config(const char *q_arg)
2734 {
2735         char s[256];
2736         const char *p, *p0 = q_arg;
2737         char *end;
2738         enum fieldnames {
2739                 FLD_PORT = 0,
2740                 FLD_QUEUE,
2741                 FLD_LCORE,
2742                 FLD_THREAD,
2743                 _NUM_FLD
2744         };
2745         unsigned long int_fld[_NUM_FLD];
2746         char *str_fld[_NUM_FLD];
2747         int i;
2748         unsigned size;
2749
2750         nb_rx_thread_params = 0;
2751
2752         while ((p = strchr(p0, '(')) != NULL) {
2753                 ++p;
2754                 p0 = strchr(p, ')');
2755                 if (p0 == NULL)
2756                         return -1;
2757
2758                 size = p0 - p;
2759                 if (size >= sizeof(s))
2760                         return -1;
2761
2762                 snprintf(s, sizeof(s), "%.*s", size, p);
2763                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2764                         return -1;
2765                 for (i = 0; i < _NUM_FLD; i++) {
2766                         errno = 0;
2767                         int_fld[i] = strtoul(str_fld[i], &end, 0);
2768                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2769                                 return -1;
2770                 }
2771                 if (nb_rx_thread_params >= MAX_LCORE_PARAMS) {
2772                         printf("exceeded max number of rx params: %hu\n",
2773                                         nb_rx_thread_params);
2774                         return -1;
2775                 }
2776                 rx_thread_params_array[nb_rx_thread_params].port_id =
2777                                 int_fld[FLD_PORT];
2778                 rx_thread_params_array[nb_rx_thread_params].queue_id =
2779                                 (uint8_t)int_fld[FLD_QUEUE];
2780                 rx_thread_params_array[nb_rx_thread_params].lcore_id =
2781                                 (uint8_t)int_fld[FLD_LCORE];
2782                 rx_thread_params_array[nb_rx_thread_params].thread_id =
2783                                 (uint8_t)int_fld[FLD_THREAD];
2784                 ++nb_rx_thread_params;
2785         }
2786         rx_thread_params = rx_thread_params_array;
2787         return 0;
2788 }
2789
2790 static int
2791 parse_tx_config(const char *q_arg)
2792 {
2793         char s[256];
2794         const char *p, *p0 = q_arg;
2795         char *end;
2796         enum fieldnames {
2797                 FLD_LCORE = 0,
2798                 FLD_THREAD,
2799                 _NUM_FLD
2800         };
2801         unsigned long int_fld[_NUM_FLD];
2802         char *str_fld[_NUM_FLD];
2803         int i;
2804         unsigned size;
2805
2806         nb_tx_thread_params = 0;
2807
2808         while ((p = strchr(p0, '(')) != NULL) {
2809                 ++p;
2810                 p0 = strchr(p, ')');
2811                 if (p0 == NULL)
2812                         return -1;
2813
2814                 size = p0 - p;
2815                 if (size >= sizeof(s))
2816                         return -1;
2817
2818                 snprintf(s, sizeof(s), "%.*s", size, p);
2819                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2820                         return -1;
2821                 for (i = 0; i < _NUM_FLD; i++) {
2822                         errno = 0;
2823                         int_fld[i] = strtoul(str_fld[i], &end, 0);
2824                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2825                                 return -1;
2826                 }
2827                 if (nb_tx_thread_params >= MAX_LCORE_PARAMS) {
2828                         printf("exceeded max number of tx params: %hu\n",
2829                                 nb_tx_thread_params);
2830                         return -1;
2831                 }
2832                 tx_thread_params_array[nb_tx_thread_params].lcore_id =
2833                                 (uint8_t)int_fld[FLD_LCORE];
2834                 tx_thread_params_array[nb_tx_thread_params].thread_id =
2835                                 (uint8_t)int_fld[FLD_THREAD];
2836                 ++nb_tx_thread_params;
2837         }
2838         tx_thread_params = tx_thread_params_array;
2839
2840         return 0;
2841 }
2842
2843 #if (APP_CPU_LOAD > 0)
2844 static int
2845 parse_stat_lcore(const char *stat_lcore)
2846 {
2847         char *end = NULL;
2848         unsigned long lcore_id;
2849
2850         lcore_id = strtoul(stat_lcore, &end, 10);
2851         if ((stat_lcore[0] == '\0') || (end == NULL) || (*end != '\0'))
2852                 return -1;
2853
2854         return lcore_id;
2855 }
2856 #endif
2857
2858 static void
2859 parse_eth_dest(const char *optarg)
2860 {
2861         uint16_t portid;
2862         char *port_end;
2863         uint8_t c, *dest, peer_addr[6];
2864
2865         errno = 0;
2866         portid = strtoul(optarg, &port_end, 10);
2867         if (errno != 0 || port_end == optarg || *port_end++ != ',')
2868                 rte_exit(EXIT_FAILURE,
2869                 "Invalid eth-dest: %s", optarg);
2870         if (portid >= RTE_MAX_ETHPORTS)
2871                 rte_exit(EXIT_FAILURE,
2872                 "eth-dest: port %d >= RTE_MAX_ETHPORTS(%d)\n",
2873                 portid, RTE_MAX_ETHPORTS);
2874
2875         if (cmdline_parse_etheraddr(NULL, port_end,
2876                 &peer_addr, sizeof(peer_addr)) < 0)
2877                 rte_exit(EXIT_FAILURE,
2878                 "Invalid ethernet address: %s\n",
2879                 port_end);
2880         dest = (uint8_t *)&dest_eth_addr[portid];
2881         for (c = 0; c < 6; c++)
2882                 dest[c] = peer_addr[c];
2883         *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
2884 }
2885
2886 #define CMD_LINE_OPT_RX_CONFIG "rx"
2887 #define CMD_LINE_OPT_TX_CONFIG "tx"
2888 #define CMD_LINE_OPT_STAT_LCORE "stat-lcore"
2889 #define CMD_LINE_OPT_ETH_DEST "eth-dest"
2890 #define CMD_LINE_OPT_NO_NUMA "no-numa"
2891 #define CMD_LINE_OPT_IPV6 "ipv6"
2892 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
2893 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
2894 #define CMD_LINE_OPT_NO_LTHREADS "no-lthreads"
2895 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
2896
2897 /* Parse the argument given in the command line of the application */
2898 static int
2899 parse_args(int argc, char **argv)
2900 {
2901         int opt, ret;
2902         char **argvopt;
2903         int option_index;
2904         char *prgname = argv[0];
2905         static struct option lgopts[] = {
2906                 {CMD_LINE_OPT_RX_CONFIG, 1, 0, 0},
2907                 {CMD_LINE_OPT_TX_CONFIG, 1, 0, 0},
2908                 {CMD_LINE_OPT_STAT_LCORE, 1, 0, 0},
2909                 {CMD_LINE_OPT_ETH_DEST, 1, 0, 0},
2910                 {CMD_LINE_OPT_NO_NUMA, 0, 0, 0},
2911                 {CMD_LINE_OPT_IPV6, 0, 0, 0},
2912                 {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, 0},
2913                 {CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, 0},
2914                 {CMD_LINE_OPT_NO_LTHREADS, 0, 0, 0},
2915                 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
2916                 {NULL, 0, 0, 0}
2917         };
2918
2919         argvopt = argv;
2920
2921         while ((opt = getopt_long(argc, argvopt, "p:P",
2922                                 lgopts, &option_index)) != EOF) {
2923
2924                 switch (opt) {
2925                 /* portmask */
2926                 case 'p':
2927                         enabled_port_mask = parse_portmask(optarg);
2928                         if (enabled_port_mask == 0) {
2929                                 printf("invalid portmask\n");
2930                                 print_usage(prgname);
2931                                 return -1;
2932                         }
2933                         break;
2934                 case 'P':
2935                         printf("Promiscuous mode selected\n");
2936                         promiscuous_on = 1;
2937                         break;
2938
2939                 /* long options */
2940                 case 0:
2941                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_RX_CONFIG,
2942                                 sizeof(CMD_LINE_OPT_RX_CONFIG))) {
2943                                 ret = parse_rx_config(optarg);
2944                                 if (ret) {
2945                                         printf("invalid rx-config\n");
2946                                         print_usage(prgname);
2947                                         return -1;
2948                                 }
2949                         }
2950
2951                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_TX_CONFIG,
2952                                 sizeof(CMD_LINE_OPT_TX_CONFIG))) {
2953                                 ret = parse_tx_config(optarg);
2954                                 if (ret) {
2955                                         printf("invalid tx-config\n");
2956                                         print_usage(prgname);
2957                                         return -1;
2958                                 }
2959                         }
2960
2961 #if (APP_CPU_LOAD > 0)
2962                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_STAT_LCORE,
2963                                         sizeof(CMD_LINE_OPT_STAT_LCORE))) {
2964                                 cpu_load_lcore_id = parse_stat_lcore(optarg);
2965                         }
2966 #endif
2967
2968                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ETH_DEST,
2969                                 sizeof(CMD_LINE_OPT_ETH_DEST)))
2970                                         parse_eth_dest(optarg);
2971
2972                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_NUMA,
2973                                 sizeof(CMD_LINE_OPT_NO_NUMA))) {
2974                                 printf("numa is disabled\n");
2975                                 numa_on = 0;
2976                         }
2977
2978 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2979                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_IPV6,
2980                                 sizeof(CMD_LINE_OPT_IPV6))) {
2981                                 printf("ipv6 is specified\n");
2982                                 ipv6 = 1;
2983                         }
2984 #endif
2985
2986                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_LTHREADS,
2987                                         sizeof(CMD_LINE_OPT_NO_LTHREADS))) {
2988                                 printf("l-threads model is disabled\n");
2989                                 lthreads_on = 0;
2990                         }
2991
2992                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_PARSE_PTYPE,
2993                                         sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
2994                                 printf("software packet type parsing enabled\n");
2995                                 parse_ptype_on = 1;
2996                         }
2997
2998                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ENABLE_JUMBO,
2999                                 sizeof(CMD_LINE_OPT_ENABLE_JUMBO))) {
3000                                 struct option lenopts = {"max-pkt-len", required_argument, 0,
3001                                                 0};
3002
3003                                 printf("jumbo frame is enabled - disabling simple TX path\n");
3004                                 port_conf.rxmode.jumbo_frame = 1;
3005
3006                                 /* if no max-pkt-len set, use the default value ETHER_MAX_LEN */
3007                                 if (0 == getopt_long(argc, argvopt, "", &lenopts,
3008                                                 &option_index)) {
3009
3010                                         ret = parse_max_pkt_len(optarg);
3011                                         if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)) {
3012                                                 printf("invalid packet length\n");
3013                                                 print_usage(prgname);
3014                                                 return -1;
3015                                         }
3016                                         port_conf.rxmode.max_rx_pkt_len = ret;
3017                                 }
3018                                 printf("set jumbo frame max packet length to %u\n",
3019                                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
3020                         }
3021 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
3022                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_HASH_ENTRY_NUM,
3023                                 sizeof(CMD_LINE_OPT_HASH_ENTRY_NUM))) {
3024                                 ret = parse_hash_entry_number(optarg);
3025                                 if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
3026                                         hash_entry_number = ret;
3027                                 } else {
3028                                         printf("invalid hash entry number\n");
3029                                         print_usage(prgname);
3030                                         return -1;
3031                                 }
3032                         }
3033 #endif
3034                         break;
3035
3036                 default:
3037                         print_usage(prgname);
3038                         return -1;
3039                 }
3040         }
3041
3042         if (optind >= 0)
3043                 argv[optind-1] = prgname;
3044
3045         ret = optind-1;
3046         optind = 1; /* reset getopt lib */
3047         return ret;
3048 }
3049
3050 static void
3051 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
3052 {
3053         char buf[ETHER_ADDR_FMT_SIZE];
3054
3055         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
3056         printf("%s%s", name, buf);
3057 }
3058
3059 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
3060
3061 static void convert_ipv4_5tuple(struct ipv4_5tuple *key1,
3062                 union ipv4_5tuple_host *key2)
3063 {
3064         key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
3065         key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
3066         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3067         key2->port_src = rte_cpu_to_be_16(key1->port_src);
3068         key2->proto = key1->proto;
3069         key2->pad0 = 0;
3070         key2->pad1 = 0;
3071 }
3072
3073 static void convert_ipv6_5tuple(struct ipv6_5tuple *key1,
3074                 union ipv6_5tuple_host *key2)
3075 {
3076         uint32_t i;
3077
3078         for (i = 0; i < 16; i++) {
3079                 key2->ip_dst[i] = key1->ip_dst[i];
3080                 key2->ip_src[i] = key1->ip_src[i];
3081         }
3082         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3083         key2->port_src = rte_cpu_to_be_16(key1->port_src);
3084         key2->proto = key1->proto;
3085         key2->pad0 = 0;
3086         key2->pad1 = 0;
3087         key2->reserve = 0;
3088 }
3089
3090 #define BYTE_VALUE_MAX 256
3091 #define ALL_32_BITS 0xffffffff
3092 #define BIT_8_TO_15 0x0000ff00
3093 static inline void
3094 populate_ipv4_few_flow_into_table(const struct rte_hash *h)
3095 {
3096         uint32_t i;
3097         int32_t ret;
3098         uint32_t array_len = RTE_DIM(ipv4_l3fwd_route_array);
3099
3100         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3101         for (i = 0; i < array_len; i++) {
3102                 struct ipv4_l3fwd_route  entry;
3103                 union ipv4_5tuple_host newkey;
3104
3105                 entry = ipv4_l3fwd_route_array[i];
3106                 convert_ipv4_5tuple(&entry.key, &newkey);
3107                 ret = rte_hash_add_key(h, (void *)&newkey);
3108                 if (ret < 0) {
3109                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3110                                 " to the l3fwd hash.\n", i);
3111                 }
3112                 ipv4_l3fwd_out_if[ret] = entry.if_out;
3113         }
3114         printf("Hash: Adding 0x%" PRIx32 " keys\n", array_len);
3115 }
3116
3117 #define BIT_16_TO_23 0x00ff0000
3118 static inline void
3119 populate_ipv6_few_flow_into_table(const struct rte_hash *h)
3120 {
3121         uint32_t i;
3122         int32_t ret;
3123         uint32_t array_len = RTE_DIM(ipv6_l3fwd_route_array);
3124
3125         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3126         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3127         for (i = 0; i < array_len; i++) {
3128                 struct ipv6_l3fwd_route entry;
3129                 union ipv6_5tuple_host newkey;
3130
3131                 entry = ipv6_l3fwd_route_array[i];
3132                 convert_ipv6_5tuple(&entry.key, &newkey);
3133                 ret = rte_hash_add_key(h, (void *)&newkey);
3134                 if (ret < 0) {
3135                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3136                                 " to the l3fwd hash.\n", i);
3137                 }
3138                 ipv6_l3fwd_out_if[ret] = entry.if_out;
3139         }
3140         printf("Hash: Adding 0x%" PRIx32 "keys\n", array_len);
3141 }
3142
3143 #define NUMBER_PORT_USED 4
3144 static inline void
3145 populate_ipv4_many_flow_into_table(const struct rte_hash *h,
3146                 unsigned int nr_flow)
3147 {
3148         unsigned i;
3149
3150         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3151
3152         for (i = 0; i < nr_flow; i++) {
3153                 struct ipv4_l3fwd_route entry;
3154                 union ipv4_5tuple_host newkey;
3155                 uint8_t a = (uint8_t)((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3156                 uint8_t b = (uint8_t)(((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3157                                 BYTE_VALUE_MAX);
3158                 uint8_t c = (uint8_t)((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3159                                 BYTE_VALUE_MAX));
3160                 /* Create the ipv4 exact match flow */
3161                 memset(&entry, 0, sizeof(entry));
3162                 switch (i & (NUMBER_PORT_USED - 1)) {
3163                 case 0:
3164                         entry = ipv4_l3fwd_route_array[0];
3165                         entry.key.ip_dst = IPv4(101, c, b, a);
3166                         break;
3167                 case 1:
3168                         entry = ipv4_l3fwd_route_array[1];
3169                         entry.key.ip_dst = IPv4(201, c, b, a);
3170                         break;
3171                 case 2:
3172                         entry = ipv4_l3fwd_route_array[2];
3173                         entry.key.ip_dst = IPv4(111, c, b, a);
3174                         break;
3175                 case 3:
3176                         entry = ipv4_l3fwd_route_array[3];
3177                         entry.key.ip_dst = IPv4(211, c, b, a);
3178                         break;
3179                 };
3180                 convert_ipv4_5tuple(&entry.key, &newkey);
3181                 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3182
3183                 if (ret < 0)
3184                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3185
3186                 ipv4_l3fwd_out_if[ret] = (uint8_t)entry.if_out;
3187
3188         }
3189         printf("Hash: Adding 0x%x keys\n", nr_flow);
3190 }
3191
3192 static inline void
3193 populate_ipv6_many_flow_into_table(const struct rte_hash *h,
3194                 unsigned int nr_flow)
3195 {
3196         unsigned i;
3197
3198         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3199         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3200         for (i = 0; i < nr_flow; i++) {
3201                 struct ipv6_l3fwd_route entry;
3202                 union ipv6_5tuple_host newkey;
3203
3204                 uint8_t a = (uint8_t) ((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3205                 uint8_t b = (uint8_t) (((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3206                                 BYTE_VALUE_MAX);
3207                 uint8_t c = (uint8_t) ((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3208                                 BYTE_VALUE_MAX));
3209
3210                 /* Create the ipv6 exact match flow */
3211                 memset(&entry, 0, sizeof(entry));
3212                 switch (i & (NUMBER_PORT_USED - 1)) {
3213                 case 0:
3214                         entry = ipv6_l3fwd_route_array[0];
3215                         break;
3216                 case 1:
3217                         entry = ipv6_l3fwd_route_array[1];
3218                         break;
3219                 case 2:
3220                         entry = ipv6_l3fwd_route_array[2];
3221                         break;
3222                 case 3:
3223                         entry = ipv6_l3fwd_route_array[3];
3224                         break;
3225                 };
3226                 entry.key.ip_dst[13] = c;
3227                 entry.key.ip_dst[14] = b;
3228                 entry.key.ip_dst[15] = a;
3229                 convert_ipv6_5tuple(&entry.key, &newkey);
3230                 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3231
3232                 if (ret < 0)
3233                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3234
3235                 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
3236
3237         }
3238         printf("Hash: Adding 0x%x keys\n", nr_flow);
3239 }
3240
3241 static void
3242 setup_hash(int socketid)
3243 {
3244         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
3245                 .name = NULL,
3246                 .entries = L3FWD_HASH_ENTRIES,
3247                 .key_len = sizeof(union ipv4_5tuple_host),
3248                 .hash_func = ipv4_hash_crc,
3249                 .hash_func_init_val = 0,
3250         };
3251
3252         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
3253                 .name = NULL,
3254                 .entries = L3FWD_HASH_ENTRIES,
3255                 .key_len = sizeof(union ipv6_5tuple_host),
3256                 .hash_func = ipv6_hash_crc,
3257                 .hash_func_init_val = 0,
3258         };
3259
3260         char s[64];
3261
3262         /* create ipv4 hash */
3263         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
3264         ipv4_l3fwd_hash_params.name = s;
3265         ipv4_l3fwd_hash_params.socket_id = socketid;
3266         ipv4_l3fwd_lookup_struct[socketid] =
3267                         rte_hash_create(&ipv4_l3fwd_hash_params);
3268         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3269                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3270                                 "socket %d\n", socketid);
3271
3272         /* create ipv6 hash */
3273         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
3274         ipv6_l3fwd_hash_params.name = s;
3275         ipv6_l3fwd_hash_params.socket_id = socketid;
3276         ipv6_l3fwd_lookup_struct[socketid] =
3277                         rte_hash_create(&ipv6_l3fwd_hash_params);
3278         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3279                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3280                                 "socket %d\n", socketid);
3281
3282         if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
3283                 /* For testing hash matching with a large number of flows we
3284                  * generate millions of IP 5-tuples with an incremented dst
3285                  * address to initialize the hash table. */
3286                 if (ipv6 == 0) {
3287                         /* populate the ipv4 hash */
3288                         populate_ipv4_many_flow_into_table(
3289                                 ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
3290                 } else {
3291                         /* populate the ipv6 hash */
3292                         populate_ipv6_many_flow_into_table(
3293                                 ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
3294                 }
3295         } else {
3296                 /* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize
3297                  * the hash table */
3298                 if (ipv6 == 0) {
3299                         /* populate the ipv4 hash */
3300                         populate_ipv4_few_flow_into_table(
3301                                         ipv4_l3fwd_lookup_struct[socketid]);
3302                 } else {
3303                         /* populate the ipv6 hash */
3304                         populate_ipv6_few_flow_into_table(
3305                                         ipv6_l3fwd_lookup_struct[socketid]);
3306                 }
3307         }
3308 }
3309 #endif
3310
3311 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3312 static void
3313 setup_lpm(int socketid)
3314 {
3315         struct rte_lpm6_config config;
3316         struct rte_lpm_config lpm_ipv4_config;
3317         unsigned i;
3318         int ret;
3319         char s[64];
3320
3321         /* create the LPM table */
3322         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
3323         lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
3324         lpm_ipv4_config.number_tbl8s = 256;
3325         lpm_ipv4_config.flags = 0;
3326         ipv4_l3fwd_lookup_struct[socketid] =
3327                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
3328         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3329                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3330                                 " on socket %d\n", socketid);
3331
3332         /* populate the LPM table */
3333         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
3334
3335                 /* skip unused ports */
3336                 if ((1 << ipv4_l3fwd_route_array[i].if_out &
3337                                 enabled_port_mask) == 0)
3338                         continue;
3339
3340                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
3341                         ipv4_l3fwd_route_array[i].ip,
3342                         ipv4_l3fwd_route_array[i].depth,
3343                         ipv4_l3fwd_route_array[i].if_out);
3344
3345                 if (ret < 0) {
3346                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3347                                 "l3fwd LPM table on socket %d\n",
3348                                 i, socketid);
3349                 }
3350
3351                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
3352                         (unsigned)ipv4_l3fwd_route_array[i].ip,
3353                         ipv4_l3fwd_route_array[i].depth,
3354                         ipv4_l3fwd_route_array[i].if_out);
3355         }
3356
3357         /* create the LPM6 table */
3358         snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
3359
3360         config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
3361         config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
3362         config.flags = 0;
3363         ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
3364                                 &config);
3365         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3366                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3367                                 " on socket %d\n", socketid);
3368
3369         /* populate the LPM table */
3370         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
3371
3372                 /* skip unused ports */
3373                 if ((1 << ipv6_l3fwd_route_array[i].if_out &
3374                                 enabled_port_mask) == 0)
3375                         continue;
3376
3377                 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
3378                         ipv6_l3fwd_route_array[i].ip,
3379                         ipv6_l3fwd_route_array[i].depth,
3380                         ipv6_l3fwd_route_array[i].if_out);
3381
3382                 if (ret < 0) {
3383                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3384                                 "l3fwd LPM table on socket %d\n",
3385                                 i, socketid);
3386                 }
3387
3388                 printf("LPM: Adding route %s / %d (%d)\n",
3389                         "IPV6",
3390                         ipv6_l3fwd_route_array[i].depth,
3391                         ipv6_l3fwd_route_array[i].if_out);
3392         }
3393 }
3394 #endif
3395
3396 static int
3397 init_mem(unsigned nb_mbuf)
3398 {
3399         struct lcore_conf *qconf;
3400         int socketid;
3401         unsigned lcore_id;
3402         char s[64];
3403
3404         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3405                 if (rte_lcore_is_enabled(lcore_id) == 0)
3406                         continue;
3407
3408                 if (numa_on)
3409                         socketid = rte_lcore_to_socket_id(lcore_id);
3410                 else
3411                         socketid = 0;
3412
3413                 if (socketid >= NB_SOCKETS) {
3414                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
3415                                 socketid, lcore_id, NB_SOCKETS);
3416                 }
3417                 if (pktmbuf_pool[socketid] == NULL) {
3418                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
3419                         pktmbuf_pool[socketid] =
3420                                 rte_pktmbuf_pool_create(s, nb_mbuf,
3421                                         MEMPOOL_CACHE_SIZE, 0,
3422                                         RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
3423                         if (pktmbuf_pool[socketid] == NULL)
3424                                 rte_exit(EXIT_FAILURE,
3425                                                 "Cannot init mbuf pool on socket %d\n", socketid);
3426                         else
3427                                 printf("Allocated mbuf pool on socket %d\n", socketid);
3428
3429 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3430                         setup_lpm(socketid);
3431 #else
3432                         setup_hash(socketid);
3433 #endif
3434                 }
3435                 qconf = &lcore_conf[lcore_id];
3436                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
3437                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
3438         }
3439         return 0;
3440 }
3441
3442 /* Check the link status of all ports in up to 9s, and print them finally */
3443 static void
3444 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask)
3445 {
3446 #define CHECK_INTERVAL 100 /* 100ms */
3447 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
3448         uint16_t portid;
3449         uint8_t count, all_ports_up, print_flag = 0;
3450         struct rte_eth_link link;
3451
3452         printf("\nChecking link status");
3453         fflush(stdout);
3454         for (count = 0; count <= MAX_CHECK_TIME; count++) {
3455                 all_ports_up = 1;
3456                 for (portid = 0; portid < port_num; portid++) {
3457                         if ((port_mask & (1 << portid)) == 0)
3458                                 continue;
3459                         memset(&link, 0, sizeof(link));
3460                         rte_eth_link_get_nowait(portid, &link);
3461                         /* print link status if flag set */
3462                         if (print_flag == 1) {
3463                                 if (link.link_status)
3464                                         printf(
3465                                         "Port%d Link Up. Speed %u Mbps - %s\n",
3466                                                 portid, link.link_speed,
3467                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
3468                                         ("full-duplex") : ("half-duplex\n"));
3469                                 else
3470                                         printf("Port %d Link Down\n", portid);
3471                                 continue;
3472                         }
3473                         /* clear all_ports_up flag if any link down */
3474                         if (link.link_status == ETH_LINK_DOWN) {
3475                                 all_ports_up = 0;
3476                                 break;
3477                         }
3478                 }
3479                 /* after finally printing all link status, get out */
3480                 if (print_flag == 1)
3481                         break;
3482
3483                 if (all_ports_up == 0) {
3484                         printf(".");
3485                         fflush(stdout);
3486                         rte_delay_ms(CHECK_INTERVAL);
3487                 }
3488
3489                 /* set the print_flag if all ports up or timeout */
3490                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
3491                         print_flag = 1;
3492                         printf("done\n");
3493                 }
3494         }
3495 }
3496
3497 int
3498 main(int argc, char **argv)
3499 {
3500         struct rte_eth_dev_info dev_info;
3501         struct rte_eth_txconf *txconf;
3502         int ret;
3503         int i;
3504         unsigned nb_ports;
3505         uint16_t queueid, portid;
3506         unsigned lcore_id;
3507         uint32_t n_tx_queue, nb_lcores;
3508         uint8_t nb_rx_queue, queue, socketid;
3509
3510         /* init EAL */
3511         ret = rte_eal_init(argc, argv);
3512         if (ret < 0)
3513                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
3514         argc -= ret;
3515         argv += ret;
3516
3517         /* pre-init dst MACs for all ports to 02:00:00:00:00:xx */
3518         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
3519                 dest_eth_addr[portid] = ETHER_LOCAL_ADMIN_ADDR +
3520                                 ((uint64_t)portid << 40);
3521                 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
3522         }
3523
3524         /* parse application arguments (after the EAL ones) */
3525         ret = parse_args(argc, argv);
3526         if (ret < 0)
3527                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
3528
3529         if (check_lcore_params() < 0)
3530                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
3531
3532         printf("Initializing rx-queues...\n");
3533         ret = init_rx_queues();
3534         if (ret < 0)
3535                 rte_exit(EXIT_FAILURE, "init_rx_queues failed\n");
3536
3537         printf("Initializing tx-threads...\n");
3538         ret = init_tx_threads();
3539         if (ret < 0)
3540                 rte_exit(EXIT_FAILURE, "init_tx_threads failed\n");
3541
3542         printf("Initializing rings...\n");
3543         ret = init_rx_rings();
3544         if (ret < 0)
3545                 rte_exit(EXIT_FAILURE, "init_rx_rings failed\n");
3546
3547         nb_ports = rte_eth_dev_count();
3548
3549         if (check_port_config(nb_ports) < 0)
3550                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
3551
3552         nb_lcores = rte_lcore_count();
3553
3554         /* initialize all ports */
3555         for (portid = 0; portid < nb_ports; portid++) {
3556                 /* skip ports that are not enabled */
3557                 if ((enabled_port_mask & (1 << portid)) == 0) {
3558                         printf("\nSkipping disabled port %d\n", portid);
3559                         continue;
3560                 }
3561
3562                 /* init port */
3563                 printf("Initializing port %d ... ", portid);
3564                 fflush(stdout);
3565
3566                 nb_rx_queue = get_port_n_rx_queues(portid);
3567                 n_tx_queue = nb_lcores;
3568                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
3569                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
3570                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
3571                         nb_rx_queue, (unsigned)n_tx_queue);
3572                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
3573                                         (uint16_t)n_tx_queue, &port_conf);
3574                 if (ret < 0)
3575                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
3576                                 ret, portid);
3577
3578                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
3579                                                        &nb_txd);
3580                 if (ret < 0)
3581                         rte_exit(EXIT_FAILURE,
3582                                  "rte_eth_dev_adjust_nb_rx_tx_desc: err=%d, port=%d\n",
3583                                  ret, portid);
3584
3585                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
3586                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
3587                 printf(", ");
3588                 print_ethaddr("Destination:",
3589                         (const struct ether_addr *)&dest_eth_addr[portid]);
3590                 printf(", ");
3591
3592                 /*
3593                  * prepare src MACs for each port.
3594                  */
3595                 ether_addr_copy(&ports_eth_addr[portid],
3596                         (struct ether_addr *)(val_eth + portid) + 1);
3597
3598                 /* init memory */
3599                 ret = init_mem(NB_MBUF);
3600                 if (ret < 0)
3601                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
3602
3603                 /* init one TX queue per couple (lcore,port) */
3604                 queueid = 0;
3605                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3606                         if (rte_lcore_is_enabled(lcore_id) == 0)
3607                                 continue;
3608
3609                         if (numa_on)
3610                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3611                         else
3612                                 socketid = 0;
3613
3614                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
3615                         fflush(stdout);
3616
3617                         rte_eth_dev_info_get(portid, &dev_info);
3618                         txconf = &dev_info.default_txconf;
3619                         if (port_conf.rxmode.jumbo_frame)
3620                                 txconf->txq_flags = 0;
3621                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
3622                                                      socketid, txconf);
3623                         if (ret < 0)
3624                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
3625                                         "port=%d\n", ret, portid);
3626
3627                         tx_thread[lcore_id].tx_queue_id[portid] = queueid;
3628                         queueid++;
3629                 }
3630                 printf("\n");
3631         }
3632
3633         for (i = 0; i < n_rx_thread; i++) {
3634                 lcore_id = rx_thread[i].conf.lcore_id;
3635
3636                 if (rte_lcore_is_enabled(lcore_id) == 0) {
3637                         rte_exit(EXIT_FAILURE,
3638                                         "Cannot start Rx thread on lcore %u: lcore disabled\n",
3639                                         lcore_id
3640                                 );
3641                 }
3642
3643                 printf("\nInitializing rx queues for Rx thread %d on lcore %u ... ",
3644                                 i, lcore_id);
3645                 fflush(stdout);
3646
3647                 /* init RX queues */
3648                 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3649                         portid = rx_thread[i].rx_queue_list[queue].port_id;
3650                         queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3651
3652                         if (numa_on)
3653                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3654                         else
3655                                 socketid = 0;
3656
3657                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
3658                         fflush(stdout);
3659
3660                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
3661                                         socketid,
3662                                         NULL,
3663                                         pktmbuf_pool[socketid]);
3664                         if (ret < 0)
3665                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, "
3666                                                 "port=%d\n", ret, portid);
3667                 }
3668         }
3669
3670         printf("\n");
3671
3672         /* start ports */
3673         for (portid = 0; portid < nb_ports; portid++) {
3674                 if ((enabled_port_mask & (1 << portid)) == 0)
3675                         continue;
3676
3677                 /* Start device */
3678                 ret = rte_eth_dev_start(portid);
3679                 if (ret < 0)
3680                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
3681                                 ret, portid);
3682
3683                 /*
3684                  * If enabled, put device in promiscuous mode.
3685                  * This allows IO forwarding mode to forward packets
3686                  * to itself through 2 cross-connected  ports of the
3687                  * target machine.
3688                  */
3689                 if (promiscuous_on)
3690                         rte_eth_promiscuous_enable(portid);
3691         }
3692
3693         for (i = 0; i < n_rx_thread; i++) {
3694                 lcore_id = rx_thread[i].conf.lcore_id;
3695                 if (rte_lcore_is_enabled(lcore_id) == 0)
3696                         continue;
3697
3698                 /* check if hw packet type is supported */
3699                 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3700                         portid = rx_thread[i].rx_queue_list[queue].port_id;
3701                         queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3702
3703                         if (parse_ptype_on) {
3704                                 if (!rte_eth_add_rx_callback(portid, queueid,
3705                                                 cb_parse_ptype, NULL))
3706                                         rte_exit(EXIT_FAILURE,
3707                                                 "Failed to add rx callback: "
3708                                                 "port=%d\n", portid);
3709                         } else if (!check_ptype(portid))
3710                                 rte_exit(EXIT_FAILURE,
3711                                         "Port %d cannot parse packet type.\n\n"
3712                                         "Please add --parse-ptype to use sw "
3713                                         "packet type analyzer.\n\n",
3714                                         portid);
3715                 }
3716         }
3717
3718         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
3719
3720         if (lthreads_on) {
3721                 printf("Starting L-Threading Model\n");
3722
3723 #if (APP_CPU_LOAD > 0)
3724                 if (cpu_load_lcore_id > 0)
3725                         /* Use one lcore for cpu load collector */
3726                         nb_lcores--;
3727 #endif
3728
3729                 lthread_num_schedulers_set(nb_lcores);
3730                 rte_eal_mp_remote_launch(sched_spawner, NULL, SKIP_MASTER);
3731                 lthread_master_spawner(NULL);
3732
3733         } else {
3734                 printf("Starting P-Threading Model\n");
3735                 /* launch per-lcore init on every lcore */
3736                 rte_eal_mp_remote_launch(pthread_run, NULL, CALL_MASTER);
3737                 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
3738                         if (rte_eal_wait_lcore(lcore_id) < 0)
3739                                 return -1;
3740                 }
3741         }
3742
3743         return 0;
3744 }