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

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2016 Intel Corporation
   3  */
   4
   5 #include <stdio.h>
   6 #include <stdlib.h>
   7 #include <stdint.h>
   8 #include <inttypes.h>
   9 #include <sys/types.h>
  10 #include <string.h>
  11 #include <sys/queue.h>
  12 #include <stdarg.h>
  13 #include <errno.h>
  14 #include <getopt.h>
  15
  16 #include <rte_common.h>
  17 #include <rte_vect.h>
  18 #include <rte_byteorder.h>
  19 #include <rte_log.h>
  20 #include <rte_memory.h>
  21 #include <rte_memcpy.h>
  22 #include <rte_eal.h>
  23 #include <rte_launch.h>
  24 #include <rte_atomic.h>
  25 #include <rte_cycles.h>
  26 #include <rte_prefetch.h>
  27 #include <rte_lcore.h>
  28 #include <rte_per_lcore.h>
  29 #include <rte_branch_prediction.h>
  30 #include <rte_interrupts.h>
  31 #include <rte_random.h>
  32 #include <rte_debug.h>
  33 #include <rte_ether.h>
  34 #include <rte_ethdev.h>
  35 #include <rte_ring.h>
  36 #include <rte_mempool.h>
  37 #include <rte_mbuf.h>
  38 #include <rte_ip.h>
  39 #include <rte_tcp.h>
  40 #include <rte_udp.h>
  41 #include <rte_string_fns.h>
  42 #include <rte_pause.h>
  43 #include <rte_timer.h>
  44
  45 #include <cmdline_parse.h>
  46 #include <cmdline_parse_etheraddr.h>
  47
  48 #include <lthread_api.h>
  49
  50 #define APP_LOOKUP_EXACT_MATCH          0
  51 #define APP_LOOKUP_LPM                  1
  52 #define DO_RFC_1812_CHECKS
  53
  54 /* Enable cpu-load stats 0-off, 1-on */
  55 #define APP_CPU_LOAD                 1
  56
  57 #ifndef APP_LOOKUP_METHOD
  58 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  59 #endif
  60
  61 #ifndef __GLIBC__ /* sched_getcpu() is glibc specific */
  62 #define sched_getcpu() rte_lcore_id()
  63 #endif
  64
  65 static int
  66 check_ptype(int portid)
  67 {
  68         int i, ret;
  69         int ipv4 = 0, ipv6 = 0;
  70
  71         ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK, NULL,
  72                         0);
  73         if (ret <= 0)
  74                 return 0;
  75
  76         uint32_t ptypes[ret];
  77
  78         ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK,
  79                         ptypes, ret);
  80         for (i = 0; i < ret; ++i) {
  81                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
  82                         ipv4 = 1;
  83                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
  84                         ipv6 = 1;
  85         }
  86
  87         if (ipv4 && ipv6)
  88                 return 1;
  89
  90         return 0;
  91 }
  92
  93 static inline void
  94 parse_ptype(struct rte_mbuf *m)
  95 {
  96         struct rte_ether_hdr *eth_hdr;
  97         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
  98         uint16_t ether_type;
  99
 100         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 101         ether_type = eth_hdr->ether_type;
 102         if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
 103                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 104         else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6))
 105                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 106
 107         m->packet_type = packet_type;
 108 }
 109
 110 static uint16_t
 111 cb_parse_ptype(__rte_unused uint16_t port, __rte_unused uint16_t queue,
 112                 struct rte_mbuf *pkts[], uint16_t nb_pkts,
 113                 __rte_unused uint16_t max_pkts, __rte_unused void *user_param)
 114 {
 115         unsigned int i;
 116
 117         for (i = 0; i < nb_pkts; i++)
 118                 parse_ptype(pkts[i]);
 119
 120         return nb_pkts;
 121 }
 122
 123 /*
 124  *  When set to zero, simple forwaring path is eanbled.
 125  *  When set to one, optimized forwarding path is enabled.
 126  *  Note that LPM optimisation path uses SSE4.1 instructions.
 127  */
 128 #define ENABLE_MULTI_BUFFER_OPTIMIZE    1
 129
 130 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 131 #include <rte_hash.h>
 132 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 133 #include <rte_lpm.h>
 134 #include <rte_lpm6.h>
 135 #else
 136 #error "APP_LOOKUP_METHOD set to incorrect value"
 137 #endif
 138
 139 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
 140
 141 #define MAX_JUMBO_PKT_LEN  9600
 142
 143 #define IPV6_ADDR_LEN 16
 144
 145 #define MEMPOOL_CACHE_SIZE 256
 146
 147 /*
 148  * This expression is used to calculate the number of mbufs needed depending on
 149  * user input, taking into account memory for rx and tx hardware rings, cache
 150  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 151  * NB_MBUF never goes below a minimum value of 8192
 152  */
 153
 154 #define NB_MBUF RTE_MAX(\
 155                 (nb_ports*nb_rx_queue*nb_rxd +      \
 156                 nb_ports*nb_lcores*MAX_PKT_BURST +  \
 157                 nb_ports*n_tx_queue*nb_txd +        \
 158                 nb_lcores*MEMPOOL_CACHE_SIZE),      \
 159                 (unsigned)8192)
 160
 161 #define MAX_PKT_BURST     32
 162 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 163
 164 /*
 165  * Try to avoid TX buffering if we have at least MAX_TX_BURST packets to send.
 166  */
 167 #define MAX_TX_BURST  (MAX_PKT_BURST / 2)
 168 #define BURST_SIZE    MAX_TX_BURST
 169
 170 #define NB_SOCKETS 8
 171
 172 /* Configure how many packets ahead to prefetch, when reading packets */
 173 #define PREFETCH_OFFSET 3
 174
 175 /* Used to mark destination port as 'invalid'. */
 176 #define BAD_PORT        ((uint16_t)-1)
 177
 178 #define FWDSTEP 4
 179
 180 /*
 181  * Configurable number of RX/TX ring descriptors
 182  */
 183 #define RTE_TEST_RX_DESC_DEFAULT 1024
 184 #define RTE_TEST_TX_DESC_DEFAULT 1024
 185 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 186 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 187
 188 /* ethernet addresses of ports */
 189 static uint64_t dest_eth_addr[RTE_MAX_ETHPORTS];
 190 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 191
 192 static xmm_t val_eth[RTE_MAX_ETHPORTS];
 193
 194 /* replace first 12B of the ethernet header. */
 195 #define MASK_ETH 0x3f
 196
 197 /* mask of enabled ports */
 198 static uint32_t enabled_port_mask;
 199 static int promiscuous_on; /**< Set in promiscuous mode off by default. */
 200 static int numa_on = 1;    /**< NUMA is enabled by default. */
 201 static int parse_ptype_on;
 202
 203 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 204 static int ipv6;           /**< ipv6 is false by default. */
 205 #endif
 206
 207 #if (APP_CPU_LOAD == 1)
 208
 209 #define MAX_CPU RTE_MAX_LCORE
 210 #define CPU_LOAD_TIMEOUT_US (5 * 1000 * 1000)  /**< Timeout for collecting 5s */
 211
 212 #define CPU_PROCESS     0
 213 #define CPU_POLL        1
 214 #define MAX_CPU_COUNTER 2
 215
 216 struct cpu_load {
 217         uint16_t       n_cpu;
 218         uint64_t       counter;
 219         uint64_t       hits[MAX_CPU_COUNTER][MAX_CPU];
 220 } __rte_cache_aligned;
 221
 222 static struct cpu_load cpu_load;
 223 static int cpu_load_lcore_id = -1;
 224
 225 #define SET_CPU_BUSY(thread, counter) \
 226                 thread->conf.busy[counter] = 1
 227
 228 #define SET_CPU_IDLE(thread, counter) \
 229                 thread->conf.busy[counter] = 0
 230
 231 #define IS_CPU_BUSY(thread, counter) \
 232                 (thread->conf.busy[counter] > 0)
 233
 234 #else
 235
 236 #define SET_CPU_BUSY(thread, counter)
 237 #define SET_CPU_IDLE(thread, counter)
 238 #define IS_CPU_BUSY(thread, counter) 0
 239
 240 #endif
 241
 242 struct mbuf_table {
 243         uint16_t len;
 244         struct rte_mbuf *m_table[MAX_PKT_BURST];
 245 };
 246
 247 struct lcore_rx_queue {
 248         uint16_t port_id;
 249         uint8_t queue_id;
 250 } __rte_cache_aligned;
 251
 252 #define MAX_RX_QUEUE_PER_LCORE 16
 253 #define MAX_TX_QUEUE_PER_PORT  RTE_MAX_ETHPORTS
 254 #define MAX_RX_QUEUE_PER_PORT  128
 255
 256 #define MAX_LCORE_PARAMS       1024
 257 struct rx_thread_params {
 258         uint16_t port_id;
 259         uint8_t queue_id;
 260         uint8_t lcore_id;
 261         uint8_t thread_id;
 262 } __rte_cache_aligned;
 263
 264 static struct rx_thread_params rx_thread_params_array[MAX_LCORE_PARAMS];
 265 static struct rx_thread_params rx_thread_params_array_default[] = {
 266         {0, 0, 2, 0},
 267         {0, 1, 2, 1},
 268         {0, 2, 2, 2},
 269         {1, 0, 2, 3},
 270         {1, 1, 2, 4},
 271         {1, 2, 2, 5},
 272         {2, 0, 2, 6},
 273         {3, 0, 3, 7},
 274         {3, 1, 3, 8},
 275 };
 276
 277 static struct rx_thread_params *rx_thread_params =
 278                 rx_thread_params_array_default;
 279 static uint16_t nb_rx_thread_params = RTE_DIM(rx_thread_params_array_default);
 280
 281 struct tx_thread_params {
 282         uint8_t lcore_id;
 283         uint8_t thread_id;
 284 } __rte_cache_aligned;
 285
 286 static struct tx_thread_params tx_thread_params_array[MAX_LCORE_PARAMS];
 287 static struct tx_thread_params tx_thread_params_array_default[] = {
 288         {4, 0},
 289         {5, 1},
 290         {6, 2},
 291         {7, 3},
 292         {8, 4},
 293         {9, 5},
 294         {10, 6},
 295         {11, 7},
 296         {12, 8},
 297 };
 298
 299 static struct tx_thread_params *tx_thread_params =
 300                 tx_thread_params_array_default;
 301 static uint16_t nb_tx_thread_params = RTE_DIM(tx_thread_params_array_default);
 302
 303 static struct rte_eth_conf port_conf = {
 304         .rxmode = {
 305                 .mq_mode = ETH_MQ_RX_RSS,
 306                 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
 307                 .split_hdr_size = 0,
 308                 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
 309         },
 310         .rx_adv_conf = {
 311                 .rss_conf = {
 312                         .rss_key = NULL,
 313                         .rss_hf = ETH_RSS_TCP,
 314                 },
 315         },
 316         .txmode = {
 317                 .mq_mode = ETH_MQ_TX_NONE,
 318         },
 319 };
 320
 321 static struct rte_mempool *pktmbuf_pool[NB_SOCKETS];
 322
 323 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 324
 325 #include <rte_hash_crc.h>
 326 #define DEFAULT_HASH_FUNC       rte_hash_crc
 327
 328 struct ipv4_5tuple {
 329         uint32_t ip_dst;
 330         uint32_t ip_src;
 331         uint16_t port_dst;
 332         uint16_t port_src;
 333         uint8_t  proto;
 334 } __rte_packed;
 335
 336 union ipv4_5tuple_host {
 337         struct {
 338                 uint8_t  pad0;
 339                 uint8_t  proto;
 340                 uint16_t pad1;
 341                 uint32_t ip_src;
 342                 uint32_t ip_dst;
 343                 uint16_t port_src;
 344                 uint16_t port_dst;
 345         };
 346         __m128i xmm;
 347 };
 348
 349 #define XMM_NUM_IN_IPV6_5TUPLE 3
 350
 351 struct ipv6_5tuple {
 352         uint8_t  ip_dst[IPV6_ADDR_LEN];
 353         uint8_t  ip_src[IPV6_ADDR_LEN];
 354         uint16_t port_dst;
 355         uint16_t port_src;
 356         uint8_t  proto;
 357 } __rte_packed;
 358
 359 union ipv6_5tuple_host {
 360         struct {
 361                 uint16_t pad0;
 362                 uint8_t  proto;
 363                 uint8_t  pad1;
 364                 uint8_t  ip_src[IPV6_ADDR_LEN];
 365                 uint8_t  ip_dst[IPV6_ADDR_LEN];
 366                 uint16_t port_src;
 367                 uint16_t port_dst;
 368                 uint64_t reserve;
 369         };
 370         __m128i xmm[XMM_NUM_IN_IPV6_5TUPLE];
 371 };
 372
 373 struct ipv4_l3fwd_route {
 374         struct ipv4_5tuple key;
 375         uint8_t if_out;
 376 };
 377
 378 struct ipv6_l3fwd_route {
 379         struct ipv6_5tuple key;
 380         uint8_t if_out;
 381 };
 382
 383 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 384         {{RTE_IPV4(101, 0, 0, 0), RTE_IPV4(100, 10, 0, 1),  101, 11, IPPROTO_TCP}, 0},
 385         {{RTE_IPV4(201, 0, 0, 0), RTE_IPV4(200, 20, 0, 1),  102, 12, IPPROTO_TCP}, 1},
 386         {{RTE_IPV4(111, 0, 0, 0), RTE_IPV4(100, 30, 0, 1),  101, 11, IPPROTO_TCP}, 2},
 387         {{RTE_IPV4(211, 0, 0, 0), RTE_IPV4(200, 40, 0, 1),  102, 12, IPPROTO_TCP}, 3},
 388 };
 389
 390 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 391         {{
 392         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 393         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 394                         0x05},
 395         101, 11, IPPROTO_TCP}, 0},
 396
 397         {{
 398         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 399         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 400                         0x05},
 401         102, 12, IPPROTO_TCP}, 1},
 402
 403         {{
 404         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 405         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 406                         0x05},
 407         101, 11, IPPROTO_TCP}, 2},
 408
 409         {{
 410         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 411         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 412                         0x05},
 413         102, 12, IPPROTO_TCP}, 3},
 414 };
 415
 416 typedef struct rte_hash lookup_struct_t;
 417 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 418 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 419
 420 #ifdef RTE_ARCH_X86_64
 421 /* default to 4 million hash entries (approx) */
 422 #define L3FWD_HASH_ENTRIES (1024*1024*4)
 423 #else
 424 /* 32-bit has less address-space for hugepage memory, limit to 1M entries */
 425 #define L3FWD_HASH_ENTRIES (1024*1024*1)
 426 #endif
 427 #define HASH_ENTRY_NUMBER_DEFAULT 4
 428
 429 static uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
 430
 431 static inline uint32_t
 432 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
 433                 uint32_t init_val)
 434 {
 435         const union ipv4_5tuple_host *k;
 436         uint32_t t;
 437         const uint32_t *p;
 438
 439         k = data;
 440         t = k->proto;
 441         p = (const uint32_t *)&k->port_src;
 442
 443         init_val = rte_hash_crc_4byte(t, init_val);
 444         init_val = rte_hash_crc_4byte(k->ip_src, init_val);
 445         init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
 446         init_val = rte_hash_crc_4byte(*p, init_val);
 447         return init_val;
 448 }
 449
 450 static inline uint32_t
 451 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len,
 452                 uint32_t init_val)
 453 {
 454         const union ipv6_5tuple_host *k;
 455         uint32_t t;
 456         const uint32_t *p;
 457         const uint32_t *ip_src0, *ip_src1, *ip_src2, *ip_src3;
 458         const uint32_t *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
 459
 460         k = data;
 461         t = k->proto;
 462         p = (const uint32_t *)&k->port_src;
 463
 464         ip_src0 = (const uint32_t *) k->ip_src;
 465         ip_src1 = (const uint32_t *)(k->ip_src + 4);
 466         ip_src2 = (const uint32_t *)(k->ip_src + 8);
 467         ip_src3 = (const uint32_t *)(k->ip_src + 12);
 468         ip_dst0 = (const uint32_t *) k->ip_dst;
 469         ip_dst1 = (const uint32_t *)(k->ip_dst + 4);
 470         ip_dst2 = (const uint32_t *)(k->ip_dst + 8);
 471         ip_dst3 = (const uint32_t *)(k->ip_dst + 12);
 472         init_val = rte_hash_crc_4byte(t, init_val);
 473         init_val = rte_hash_crc_4byte(*ip_src0, init_val);
 474         init_val = rte_hash_crc_4byte(*ip_src1, init_val);
 475         init_val = rte_hash_crc_4byte(*ip_src2, init_val);
 476         init_val = rte_hash_crc_4byte(*ip_src3, init_val);
 477         init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
 478         init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
 479         init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
 480         init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
 481         init_val = rte_hash_crc_4byte(*p, init_val);
 482         return init_val;
 483 }
 484
 485 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
 486 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
 487
 488 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 489 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 490
 491 #endif
 492
 493 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 494 struct ipv4_l3fwd_route {
 495         uint32_t ip;
 496         uint8_t  depth;
 497         uint8_t  if_out;
 498 };
 499
 500 struct ipv6_l3fwd_route {
 501         uint8_t ip[16];
 502         uint8_t depth;
 503         uint8_t if_out;
 504 };
 505
 506 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 507         {RTE_IPV4(1, 1, 1, 0), 24, 0},
 508         {RTE_IPV4(2, 1, 1, 0), 24, 1},
 509         {RTE_IPV4(3, 1, 1, 0), 24, 2},
 510         {RTE_IPV4(4, 1, 1, 0), 24, 3},
 511         {RTE_IPV4(5, 1, 1, 0), 24, 4},
 512         {RTE_IPV4(6, 1, 1, 0), 24, 5},
 513         {RTE_IPV4(7, 1, 1, 0), 24, 6},
 514         {RTE_IPV4(8, 1, 1, 0), 24, 7},
 515 };
 516
 517 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 518         {{1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 0},
 519         {{2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 1},
 520         {{3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 2},
 521         {{4, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 3},
 522         {{5, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 4},
 523         {{6, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 5},
 524         {{7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 6},
 525         {{8, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 7},
 526 };
 527
 528 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
 529 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
 530
 531 #define IPV4_L3FWD_LPM_MAX_RULES         1024
 532 #define IPV6_L3FWD_LPM_MAX_RULES         1024
 533 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
 534
 535 typedef struct rte_lpm lookup_struct_t;
 536 typedef struct rte_lpm6 lookup6_struct_t;
 537 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 538 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 539 #endif
 540
 541 struct lcore_conf {
 542         lookup_struct_t *ipv4_lookup_struct;
 543 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 544         lookup6_struct_t *ipv6_lookup_struct;
 545 #else
 546         lookup_struct_t *ipv6_lookup_struct;
 547 #endif
 548         void *data;
 549 } __rte_cache_aligned;
 550
 551 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 552 RTE_DEFINE_PER_LCORE(struct lcore_conf *, lcore_conf);
 553
 554 #define MAX_RX_QUEUE_PER_THREAD 16
 555 #define MAX_TX_PORT_PER_THREAD  RTE_MAX_ETHPORTS
 556 #define MAX_TX_QUEUE_PER_PORT   RTE_MAX_ETHPORTS
 557 #define MAX_RX_QUEUE_PER_PORT   128
 558
 559 #define MAX_RX_THREAD 1024
 560 #define MAX_TX_THREAD 1024
 561 #define MAX_THREAD    (MAX_RX_THREAD + MAX_TX_THREAD)
 562
 563 /**
 564  * Producers and consumers threads configuration
 565  */
 566 static int lthreads_on = 1; /**< Use lthreads for processing*/
 567
 568 rte_atomic16_t rx_counter;  /**< Number of spawned rx threads */
 569 rte_atomic16_t tx_counter;  /**< Number of spawned tx threads */
 570
 571 struct thread_conf {
 572         uint16_t lcore_id;      /**< Initial lcore for rx thread */
 573         uint16_t cpu_id;        /**< Cpu id for cpu load stats counter */
 574         uint16_t thread_id;     /**< Thread ID */
 575
 576 #if (APP_CPU_LOAD > 0)
 577         int busy[MAX_CPU_COUNTER];
 578 #endif
 579 };
 580
 581 struct thread_rx_conf {
 582         struct thread_conf conf;
 583
 584         uint16_t n_rx_queue;
 585         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 586
 587         uint16_t n_ring;        /**< Number of output rings */
 588         struct rte_ring *ring[RTE_MAX_LCORE];
 589         struct lthread_cond *ready[RTE_MAX_LCORE];
 590
 591 #if (APP_CPU_LOAD > 0)
 592         int busy[MAX_CPU_COUNTER];
 593 #endif
 594 } __rte_cache_aligned;
 595
 596 uint16_t n_rx_thread;
 597 struct thread_rx_conf rx_thread[MAX_RX_THREAD];
 598
 599 struct thread_tx_conf {
 600         struct thread_conf conf;
 601
 602         uint16_t tx_queue_id[RTE_MAX_LCORE];
 603         struct mbuf_table tx_mbufs[RTE_MAX_LCORE];
 604
 605         struct rte_ring *ring;
 606         struct lthread_cond **ready;
 607
 608 } __rte_cache_aligned;
 609
 610 uint16_t n_tx_thread;
 611 struct thread_tx_conf tx_thread[MAX_TX_THREAD];
 612
 613 /* Send burst of packets on an output interface */
 614 static inline int
 615 send_burst(struct thread_tx_conf *qconf, uint16_t n, uint16_t port)
 616 {
 617         struct rte_mbuf **m_table;
 618         int ret;
 619         uint16_t queueid;
 620
 621         queueid = qconf->tx_queue_id[port];
 622         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 623
 624         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 625         if (unlikely(ret < n)) {
 626                 do {
 627                         rte_pktmbuf_free(m_table[ret]);
 628                 } while (++ret < n);
 629         }
 630
 631         return 0;
 632 }
 633
 634 /* Enqueue a single packet, and send burst if queue is filled */
 635 static inline int
 636 send_single_packet(struct rte_mbuf *m, uint16_t port)
 637 {
 638         uint16_t len;
 639         struct thread_tx_conf *qconf;
 640
 641         if (lthreads_on)
 642                 qconf = (struct thread_tx_conf *)lthread_get_data();
 643         else
 644                 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
 645
 646         len = qconf->tx_mbufs[port].len;
 647         qconf->tx_mbufs[port].m_table[len] = m;
 648         len++;
 649
 650         /* enough pkts to be sent */
 651         if (unlikely(len == MAX_PKT_BURST)) {
 652                 send_burst(qconf, MAX_PKT_BURST, port);
 653                 len = 0;
 654         }
 655
 656         qconf->tx_mbufs[port].len = len;
 657         return 0;
 658 }
 659
 660 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
 661         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
 662 static __rte_always_inline void
 663 send_packetsx4(uint16_t port,
 664         struct rte_mbuf *m[], uint32_t num)
 665 {
 666         uint32_t len, j, n;
 667         struct thread_tx_conf *qconf;
 668
 669         if (lthreads_on)
 670                 qconf = (struct thread_tx_conf *)lthread_get_data();
 671         else
 672                 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
 673
 674         len = qconf->tx_mbufs[port].len;
 675
 676         /*
 677          * If TX buffer for that queue is empty, and we have enough packets,
 678          * then send them straightway.
 679          */
 680         if (num >= MAX_TX_BURST && len == 0) {
 681                 n = rte_eth_tx_burst(port, qconf->tx_queue_id[port], m, num);
 682                 if (unlikely(n < num)) {
 683                         do {
 684                                 rte_pktmbuf_free(m[n]);
 685                         } while (++n < num);
 686                 }
 687                 return;
 688         }
 689
 690         /*
 691          * Put packets into TX buffer for that queue.
 692          */
 693
 694         n = len + num;
 695         n = (n > MAX_PKT_BURST) ? MAX_PKT_BURST - len : num;
 696
 697         j = 0;
 698         switch (n % FWDSTEP) {
 699         while (j < n) {
 700         case 0:
 701                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 702                 j++;
 703                 /* fall-through */
 704         case 3:
 705                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 706                 j++;
 707                 /* fall-through */
 708         case 2:
 709                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 710                 j++;
 711                 /* fall-through */
 712         case 1:
 713                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 714                 j++;
 715         }
 716         }
 717
 718         len += n;
 719
 720         /* enough pkts to be sent */
 721         if (unlikely(len == MAX_PKT_BURST)) {
 722
 723                 send_burst(qconf, MAX_PKT_BURST, port);
 724
 725                 /* copy rest of the packets into the TX buffer. */
 726                 len = num - n;
 727                 j = 0;
 728                 switch (len % FWDSTEP) {
 729                 while (j < len) {
 730                 case 0:
 731                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 732                         j++;
 733                         /* fall-through */
 734                 case 3:
 735                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 736                         j++;
 737                         /* fall-through */
 738                 case 2:
 739                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 740                         j++;
 741                         /* fall-through */
 742                 case 1:
 743                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 744                         j++;
 745                 }
 746                 }
 747         }
 748
 749         qconf->tx_mbufs[port].len = len;
 750 }
 751 #endif /* APP_LOOKUP_LPM */
 752
 753 #ifdef DO_RFC_1812_CHECKS
 754 static inline int
 755 is_valid_ipv4_pkt(struct rte_ipv4_hdr *pkt, uint32_t link_len)
 756 {
 757         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 758         /*
 759          * 1. The packet length reported by the Link Layer must be large
 760          * enough to hold the minimum length legal IP datagram (20 bytes).
 761          */
 762         if (link_len < sizeof(struct rte_ipv4_hdr))
 763                 return -1;
 764
 765         /* 2. The IP checksum must be correct. */
 766         /* this is checked in H/W */
 767
 768         /*
 769          * 3. The IP version number must be 4. If the version number is not 4
 770          * then the packet may be another version of IP, such as IPng or
 771          * ST-II.
 772          */
 773         if (((pkt->version_ihl) >> 4) != 4)
 774                 return -3;
 775         /*
 776          * 4. The IP header length field must be large enough to hold the
 777          * minimum length legal IP datagram (20 bytes = 5 words).
 778          */
 779         if ((pkt->version_ihl & 0xf) < 5)
 780                 return -4;
 781
 782         /*
 783          * 5. The IP total length field must be large enough to hold the IP
 784          * datagram header, whose length is specified in the IP header length
 785          * field.
 786          */
 787         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct rte_ipv4_hdr))
 788                 return -5;
 789
 790         return 0;
 791 }
 792 #endif
 793
 794 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 795
 796 static __m128i mask0;
 797 static __m128i mask1;
 798 static __m128i mask2;
 799 static inline uint16_t
 800 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid,
 801                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 802 {
 803         int ret = 0;
 804         union ipv4_5tuple_host key;
 805
 806         ipv4_hdr = (uint8_t *)ipv4_hdr +
 807                 offsetof(struct rte_ipv4_hdr, time_to_live);
 808         __m128i data = _mm_loadu_si128((__m128i *)(ipv4_hdr));
 809         /* Get 5 tuple: dst port, src port, dst IP address, src IP address and
 810            protocol */
 811         key.xmm = _mm_and_si128(data, mask0);
 812         /* Find destination port */
 813         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 814         return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
 815 }
 816
 817 static inline uint16_t
 818 get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid,
 819                 lookup_struct_t *ipv6_l3fwd_lookup_struct)
 820 {
 821         int ret = 0;
 822         union ipv6_5tuple_host key;
 823
 824         ipv6_hdr = (uint8_t *)ipv6_hdr +
 825                 offsetof(struct rte_ipv6_hdr, payload_len);
 826         __m128i data0 = _mm_loadu_si128((__m128i *)(ipv6_hdr));
 827         __m128i data1 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
 828                         sizeof(__m128i)));
 829         __m128i data2 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
 830                         sizeof(__m128i) + sizeof(__m128i)));
 831         /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
 832         key.xmm[0] = _mm_and_si128(data0, mask1);
 833         /* Get part of 5 tuple: dst IP address lower 96 bits and src IP address
 834            higher 32 bits */
 835         key.xmm[1] = data1;
 836         /* Get part of 5 tuple: dst port and src port and dst IP address higher
 837            32 bits */
 838         key.xmm[2] = _mm_and_si128(data2, mask2);
 839
 840         /* Find destination port */
 841         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 842         return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
 843 }
 844 #endif
 845
 846 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 847
 848 static inline uint16_t
 849 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid,
 850                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 851 {
 852         uint32_t next_hop;
 853
 854         return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 855                 rte_be_to_cpu_32(((struct rte_ipv4_hdr *)ipv4_hdr)->dst_addr),
 856                 &next_hop) == 0) ? next_hop : portid);
 857 }
 858
 859 static inline uint16_t
 860 get_ipv6_dst_port(void *ipv6_hdr,  uint16_t portid,
 861                 lookup6_struct_t *ipv6_l3fwd_lookup_struct)
 862 {
 863         uint32_t next_hop;
 864
 865         return ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
 866                 ((struct rte_ipv6_hdr *)ipv6_hdr)->dst_addr, &next_hop) == 0) ?
 867                 next_hop : portid);
 868 }
 869 #endif
 870
 871 static inline void l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid)
 872                 __rte_unused;
 873
 874 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) && \
 875         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
 876
 877 #define MASK_ALL_PKTS   0xff
 878 #define EXCLUDE_1ST_PKT 0xfe
 879 #define EXCLUDE_2ND_PKT 0xfd
 880 #define EXCLUDE_3RD_PKT 0xfb
 881 #define EXCLUDE_4TH_PKT 0xf7
 882 #define EXCLUDE_5TH_PKT 0xef
 883 #define EXCLUDE_6TH_PKT 0xdf
 884 #define EXCLUDE_7TH_PKT 0xbf
 885 #define EXCLUDE_8TH_PKT 0x7f
 886
 887 static inline void
 888 simple_ipv4_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid)
 889 {
 890         struct rte_ether_hdr *eth_hdr[8];
 891         struct rte_ipv4_hdr *ipv4_hdr[8];
 892         uint16_t dst_port[8];
 893         int32_t ret[8];
 894         union ipv4_5tuple_host key[8];
 895         __m128i data[8];
 896
 897         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct rte_ether_hdr *);
 898         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct rte_ether_hdr *);
 899         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct rte_ether_hdr *);
 900         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct rte_ether_hdr *);
 901         eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct rte_ether_hdr *);
 902         eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct rte_ether_hdr *);
 903         eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct rte_ether_hdr *);
 904         eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct rte_ether_hdr *);
 905
 906         /* Handle IPv4 headers.*/
 907         ipv4_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct rte_ipv4_hdr *,
 908                         sizeof(struct rte_ether_hdr));
 909         ipv4_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct rte_ipv4_hdr *,
 910                         sizeof(struct rte_ether_hdr));
 911         ipv4_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct rte_ipv4_hdr *,
 912                         sizeof(struct rte_ether_hdr));
 913         ipv4_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct rte_ipv4_hdr *,
 914                         sizeof(struct rte_ether_hdr));
 915         ipv4_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct rte_ipv4_hdr *,
 916                         sizeof(struct rte_ether_hdr));
 917         ipv4_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct rte_ipv4_hdr *,
 918                         sizeof(struct rte_ether_hdr));
 919         ipv4_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct rte_ipv4_hdr *,
 920                         sizeof(struct rte_ether_hdr));
 921         ipv4_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct rte_ipv4_hdr *,
 922                         sizeof(struct rte_ether_hdr));
 923
 924 #ifdef DO_RFC_1812_CHECKS
 925         /* Check to make sure the packet is valid (RFC1812) */
 926         uint8_t valid_mask = MASK_ALL_PKTS;
 927
 928         if (is_valid_ipv4_pkt(ipv4_hdr[0], m[0]->pkt_len) < 0) {
 929                 rte_pktmbuf_free(m[0]);
 930                 valid_mask &= EXCLUDE_1ST_PKT;
 931         }
 932         if (is_valid_ipv4_pkt(ipv4_hdr[1], m[1]->pkt_len) < 0) {
 933                 rte_pktmbuf_free(m[1]);
 934                 valid_mask &= EXCLUDE_2ND_PKT;
 935         }
 936         if (is_valid_ipv4_pkt(ipv4_hdr[2], m[2]->pkt_len) < 0) {
 937                 rte_pktmbuf_free(m[2]);
 938                 valid_mask &= EXCLUDE_3RD_PKT;
 939         }
 940         if (is_valid_ipv4_pkt(ipv4_hdr[3], m[3]->pkt_len) < 0) {
 941                 rte_pktmbuf_free(m[3]);
 942                 valid_mask &= EXCLUDE_4TH_PKT;
 943         }
 944         if (is_valid_ipv4_pkt(ipv4_hdr[4], m[4]->pkt_len) < 0) {
 945                 rte_pktmbuf_free(m[4]);
 946                 valid_mask &= EXCLUDE_5TH_PKT;
 947         }
 948         if (is_valid_ipv4_pkt(ipv4_hdr[5], m[5]->pkt_len) < 0) {
 949                 rte_pktmbuf_free(m[5]);
 950                 valid_mask &= EXCLUDE_6TH_PKT;
 951         }
 952         if (is_valid_ipv4_pkt(ipv4_hdr[6], m[6]->pkt_len) < 0) {
 953                 rte_pktmbuf_free(m[6]);
 954                 valid_mask &= EXCLUDE_7TH_PKT;
 955         }
 956         if (is_valid_ipv4_pkt(ipv4_hdr[7], m[7]->pkt_len) < 0) {
 957                 rte_pktmbuf_free(m[7]);
 958                 valid_mask &= EXCLUDE_8TH_PKT;
 959         }
 960         if (unlikely(valid_mask != MASK_ALL_PKTS)) {
 961                 if (valid_mask == 0)
 962                         return;
 963
 964                 uint8_t i = 0;
 965
 966                 for (i = 0; i < 8; i++)
 967                         if ((0x1 << i) & valid_mask)
 968                                 l3fwd_simple_forward(m[i], portid);
 969         }
 970 #endif /* End of #ifdef DO_RFC_1812_CHECKS */
 971
 972         data[0] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[0], __m128i *,
 973                         sizeof(struct rte_ether_hdr) +
 974                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 975         data[1] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[1], __m128i *,
 976                         sizeof(struct rte_ether_hdr) +
 977                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 978         data[2] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[2], __m128i *,
 979                         sizeof(struct rte_ether_hdr) +
 980                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 981         data[3] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[3], __m128i *,
 982                         sizeof(struct rte_ether_hdr) +
 983                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 984         data[4] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[4], __m128i *,
 985                         sizeof(struct rte_ether_hdr) +
 986                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 987         data[5] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[5], __m128i *,
 988                         sizeof(struct rte_ether_hdr) +
 989                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 990         data[6] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[6], __m128i *,
 991                         sizeof(struct rte_ether_hdr) +
 992                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 993         data[7] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[7], __m128i *,
 994                         sizeof(struct rte_ether_hdr) +
 995                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 996
 997         key[0].xmm = _mm_and_si128(data[0], mask0);
 998         key[1].xmm = _mm_and_si128(data[1], mask0);
 999         key[2].xmm = _mm_and_si128(data[2], mask0);
1000         key[3].xmm = _mm_and_si128(data[3], mask0);
1001         key[4].xmm = _mm_and_si128(data[4], mask0);
1002         key[5].xmm = _mm_and_si128(data[5], mask0);
1003         key[6].xmm = _mm_and_si128(data[6], mask0);
1004         key[7].xmm = _mm_and_si128(data[7], mask0);
1005
1006         const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1007                         &key[4], &key[5], &key[6], &key[7]};
1008
1009         rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct,
1010                         &key_array[0], 8, ret);
1011         dst_port[0] = ((ret[0] < 0) ? portid : ipv4_l3fwd_out_if[ret[0]]);
1012         dst_port[1] = ((ret[1] < 0) ? portid : ipv4_l3fwd_out_if[ret[1]]);
1013         dst_port[2] = ((ret[2] < 0) ? portid : ipv4_l3fwd_out_if[ret[2]]);
1014         dst_port[3] = ((ret[3] < 0) ? portid : ipv4_l3fwd_out_if[ret[3]]);
1015         dst_port[4] = ((ret[4] < 0) ? portid : ipv4_l3fwd_out_if[ret[4]]);
1016         dst_port[5] = ((ret[5] < 0) ? portid : ipv4_l3fwd_out_if[ret[5]]);
1017         dst_port[6] = ((ret[6] < 0) ? portid : ipv4_l3fwd_out_if[ret[6]]);
1018         dst_port[7] = ((ret[7] < 0) ? portid : ipv4_l3fwd_out_if[ret[7]]);
1019
1020         if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1021                         (enabled_port_mask & 1 << dst_port[0]) == 0)
1022                 dst_port[0] = portid;
1023         if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1024                         (enabled_port_mask & 1 << dst_port[1]) == 0)
1025                 dst_port[1] = portid;
1026         if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1027                         (enabled_port_mask & 1 << dst_port[2]) == 0)
1028                 dst_port[2] = portid;
1029         if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1030                         (enabled_port_mask & 1 << dst_port[3]) == 0)
1031                 dst_port[3] = portid;
1032         if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1033                         (enabled_port_mask & 1 << dst_port[4]) == 0)
1034                 dst_port[4] = portid;
1035         if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1036                         (enabled_port_mask & 1 << dst_port[5]) == 0)
1037                 dst_port[5] = portid;
1038         if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1039                         (enabled_port_mask & 1 << dst_port[6]) == 0)
1040                 dst_port[6] = portid;
1041         if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1042                         (enabled_port_mask & 1 << dst_port[7]) == 0)
1043                 dst_port[7] = portid;
1044
1045 #ifdef DO_RFC_1812_CHECKS
1046         /* Update time to live and header checksum */
1047         --(ipv4_hdr[0]->time_to_live);
1048         --(ipv4_hdr[1]->time_to_live);
1049         --(ipv4_hdr[2]->time_to_live);
1050         --(ipv4_hdr[3]->time_to_live);
1051         ++(ipv4_hdr[0]->hdr_checksum);
1052         ++(ipv4_hdr[1]->hdr_checksum);
1053         ++(ipv4_hdr[2]->hdr_checksum);
1054         ++(ipv4_hdr[3]->hdr_checksum);
1055         --(ipv4_hdr[4]->time_to_live);
1056         --(ipv4_hdr[5]->time_to_live);
1057         --(ipv4_hdr[6]->time_to_live);
1058         --(ipv4_hdr[7]->time_to_live);
1059         ++(ipv4_hdr[4]->hdr_checksum);
1060         ++(ipv4_hdr[5]->hdr_checksum);
1061         ++(ipv4_hdr[6]->hdr_checksum);
1062         ++(ipv4_hdr[7]->hdr_checksum);
1063 #endif
1064
1065         /* dst addr */
1066         *(uint64_t *)&eth_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1067         *(uint64_t *)&eth_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1068         *(uint64_t *)&eth_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1069         *(uint64_t *)&eth_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1070         *(uint64_t *)&eth_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1071         *(uint64_t *)&eth_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1072         *(uint64_t *)&eth_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1073         *(uint64_t *)&eth_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1074
1075         /* src addr */
1076         rte_ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
1077         rte_ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
1078         rte_ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
1079         rte_ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
1080         rte_ether_addr_copy(&ports_eth_addr[dst_port[4]], &eth_hdr[4]->s_addr);
1081         rte_ether_addr_copy(&ports_eth_addr[dst_port[5]], &eth_hdr[5]->s_addr);
1082         rte_ether_addr_copy(&ports_eth_addr[dst_port[6]], &eth_hdr[6]->s_addr);
1083         rte_ether_addr_copy(&ports_eth_addr[dst_port[7]], &eth_hdr[7]->s_addr);
1084
1085         send_single_packet(m[0], (uint8_t)dst_port[0]);
1086         send_single_packet(m[1], (uint8_t)dst_port[1]);
1087         send_single_packet(m[2], (uint8_t)dst_port[2]);
1088         send_single_packet(m[3], (uint8_t)dst_port[3]);
1089         send_single_packet(m[4], (uint8_t)dst_port[4]);
1090         send_single_packet(m[5], (uint8_t)dst_port[5]);
1091         send_single_packet(m[6], (uint8_t)dst_port[6]);
1092         send_single_packet(m[7], (uint8_t)dst_port[7]);
1093
1094 }
1095
1096 static inline void get_ipv6_5tuple(struct rte_mbuf *m0, __m128i mask0,
1097                 __m128i mask1, union ipv6_5tuple_host *key)
1098 {
1099         __m128i tmpdata0 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1100                         __m128i *, sizeof(struct rte_ether_hdr) +
1101                         offsetof(struct rte_ipv6_hdr, payload_len)));
1102         __m128i tmpdata1 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1103                         __m128i *, sizeof(struct rte_ether_hdr) +
1104                         offsetof(struct rte_ipv6_hdr, payload_len) +
1105                         sizeof(__m128i)));
1106         __m128i tmpdata2 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1107                         __m128i *, sizeof(struct rte_ether_hdr) +
1108                         offsetof(struct rte_ipv6_hdr, payload_len) +
1109                         sizeof(__m128i) + sizeof(__m128i)));
1110         key->xmm[0] = _mm_and_si128(tmpdata0, mask0);
1111         key->xmm[1] = tmpdata1;
1112         key->xmm[2] = _mm_and_si128(tmpdata2, mask1);
1113 }
1114
1115 static inline void
1116 simple_ipv6_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid)
1117 {
1118         int32_t ret[8];
1119         uint16_t dst_port[8];
1120         struct rte_ether_hdr *eth_hdr[8];
1121         union ipv6_5tuple_host key[8];
1122
1123         __rte_unused struct rte_ipv6_hdr *ipv6_hdr[8];
1124
1125         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct rte_ether_hdr *);
1126         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct rte_ether_hdr *);
1127         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct rte_ether_hdr *);
1128         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct rte_ether_hdr *);
1129         eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct rte_ether_hdr *);
1130         eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct rte_ether_hdr *);
1131         eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct rte_ether_hdr *);
1132         eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct rte_ether_hdr *);
1133
1134         /* Handle IPv6 headers.*/
1135         ipv6_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct rte_ipv6_hdr *,
1136                         sizeof(struct rte_ether_hdr));
1137         ipv6_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct rte_ipv6_hdr *,
1138                         sizeof(struct rte_ether_hdr));
1139         ipv6_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct rte_ipv6_hdr *,
1140                         sizeof(struct rte_ether_hdr));
1141         ipv6_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct rte_ipv6_hdr *,
1142                         sizeof(struct rte_ether_hdr));
1143         ipv6_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct rte_ipv6_hdr *,
1144                         sizeof(struct rte_ether_hdr));
1145         ipv6_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct rte_ipv6_hdr *,
1146                         sizeof(struct rte_ether_hdr));
1147         ipv6_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct rte_ipv6_hdr *,
1148                         sizeof(struct rte_ether_hdr));
1149         ipv6_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct rte_ipv6_hdr *,
1150                         sizeof(struct rte_ether_hdr));
1151
1152         get_ipv6_5tuple(m[0], mask1, mask2, &key[0]);
1153         get_ipv6_5tuple(m[1], mask1, mask2, &key[1]);
1154         get_ipv6_5tuple(m[2], mask1, mask2, &key[2]);
1155         get_ipv6_5tuple(m[3], mask1, mask2, &key[3]);
1156         get_ipv6_5tuple(m[4], mask1, mask2, &key[4]);
1157         get_ipv6_5tuple(m[5], mask1, mask2, &key[5]);
1158         get_ipv6_5tuple(m[6], mask1, mask2, &key[6]);
1159         get_ipv6_5tuple(m[7], mask1, mask2, &key[7]);
1160
1161         const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1162                         &key[4], &key[5], &key[6], &key[7]};
1163
1164         rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1165                         &key_array[0], 4, ret);
1166         dst_port[0] = ((ret[0] < 0) ? portid : ipv6_l3fwd_out_if[ret[0]]);
1167         dst_port[1] = ((ret[1] < 0) ? portid : ipv6_l3fwd_out_if[ret[1]]);
1168         dst_port[2] = ((ret[2] < 0) ? portid : ipv6_l3fwd_out_if[ret[2]]);
1169         dst_port[3] = ((ret[3] < 0) ? portid : ipv6_l3fwd_out_if[ret[3]]);
1170         dst_port[4] = ((ret[4] < 0) ? portid : ipv6_l3fwd_out_if[ret[4]]);
1171         dst_port[5] = ((ret[5] < 0) ? portid : ipv6_l3fwd_out_if[ret[5]]);
1172         dst_port[6] = ((ret[6] < 0) ? portid : ipv6_l3fwd_out_if[ret[6]]);
1173         dst_port[7] = ((ret[7] < 0) ? portid : ipv6_l3fwd_out_if[ret[7]]);
1174
1175         if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1176                         (enabled_port_mask & 1 << dst_port[0]) == 0)
1177                 dst_port[0] = portid;
1178         if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1179                         (enabled_port_mask & 1 << dst_port[1]) == 0)
1180                 dst_port[1] = portid;
1181         if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1182                         (enabled_port_mask & 1 << dst_port[2]) == 0)
1183                 dst_port[2] = portid;
1184         if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1185                         (enabled_port_mask & 1 << dst_port[3]) == 0)
1186                 dst_port[3] = portid;
1187         if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1188                         (enabled_port_mask & 1 << dst_port[4]) == 0)
1189                 dst_port[4] = portid;
1190         if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1191                         (enabled_port_mask & 1 << dst_port[5]) == 0)
1192                 dst_port[5] = portid;
1193         if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1194                         (enabled_port_mask & 1 << dst_port[6]) == 0)
1195                 dst_port[6] = portid;
1196         if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1197                         (enabled_port_mask & 1 << dst_port[7]) == 0)
1198                 dst_port[7] = portid;
1199
1200         /* dst addr */
1201         *(uint64_t *)&eth_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1202         *(uint64_t *)&eth_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1203         *(uint64_t *)&eth_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1204         *(uint64_t *)&eth_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1205         *(uint64_t *)&eth_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1206         *(uint64_t *)&eth_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1207         *(uint64_t *)&eth_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1208         *(uint64_t *)&eth_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1209
1210         /* src addr */
1211         rte_ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
1212         rte_ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
1213         rte_ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
1214         rte_ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
1215         rte_ether_addr_copy(&ports_eth_addr[dst_port[4]], &eth_hdr[4]->s_addr);
1216         rte_ether_addr_copy(&ports_eth_addr[dst_port[5]], &eth_hdr[5]->s_addr);
1217         rte_ether_addr_copy(&ports_eth_addr[dst_port[6]], &eth_hdr[6]->s_addr);
1218         rte_ether_addr_copy(&ports_eth_addr[dst_port[7]], &eth_hdr[7]->s_addr);
1219
1220         send_single_packet(m[0], dst_port[0]);
1221         send_single_packet(m[1], dst_port[1]);
1222         send_single_packet(m[2], dst_port[2]);
1223         send_single_packet(m[3], dst_port[3]);
1224         send_single_packet(m[4], dst_port[4]);
1225         send_single_packet(m[5], dst_port[5]);
1226         send_single_packet(m[6], dst_port[6]);
1227         send_single_packet(m[7], dst_port[7]);
1228
1229 }
1230 #endif /* APP_LOOKUP_METHOD */
1231
1232 static __rte_always_inline void
1233 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid)
1234 {
1235         struct rte_ether_hdr *eth_hdr;
1236         struct rte_ipv4_hdr *ipv4_hdr;
1237         uint16_t dst_port;
1238
1239         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
1240
1241         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
1242                 /* Handle IPv4 headers.*/
1243                 ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
1244                                 sizeof(struct rte_ether_hdr));
1245
1246 #ifdef DO_RFC_1812_CHECKS
1247                 /* Check to make sure the packet is valid (RFC1812) */
1248                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
1249                         rte_pktmbuf_free(m);
1250                         return;
1251                 }
1252 #endif
1253
1254                  dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
1255                         RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct);
1256                 if (dst_port >= RTE_MAX_ETHPORTS ||
1257                                 (enabled_port_mask & 1 << dst_port) == 0)
1258                         dst_port = portid;
1259
1260 #ifdef DO_RFC_1812_CHECKS
1261                 /* Update time to live and header checksum */
1262                 --(ipv4_hdr->time_to_live);
1263                 ++(ipv4_hdr->hdr_checksum);
1264 #endif
1265                 /* dst addr */
1266                 *(uint64_t *)&eth_hdr->d_addr = dest_eth_addr[dst_port];
1267
1268                 /* src addr */
1269                 rte_ether_addr_copy(&ports_eth_addr[dst_port],
1270                                 &eth_hdr->s_addr);
1271
1272                 send_single_packet(m, dst_port);
1273         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
1274                 /* Handle IPv6 headers.*/
1275                 struct rte_ipv6_hdr *ipv6_hdr;
1276
1277                 ipv6_hdr = rte_pktmbuf_mtod_offset(m, struct rte_ipv6_hdr *,
1278                                 sizeof(struct rte_ether_hdr));
1279
1280                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
1281                                 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct);
1282
1283                 if (dst_port >= RTE_MAX_ETHPORTS ||
1284                                 (enabled_port_mask & 1 << dst_port) == 0)
1285                         dst_port = portid;
1286
1287                 /* dst addr */
1288                 *(uint64_t *)&eth_hdr->d_addr = dest_eth_addr[dst_port];
1289
1290                 /* src addr */
1291                 rte_ether_addr_copy(&ports_eth_addr[dst_port],
1292                                 &eth_hdr->s_addr);
1293
1294                 send_single_packet(m, dst_port);
1295         } else
1296                 /* Free the mbuf that contains non-IPV4/IPV6 packet */
1297                 rte_pktmbuf_free(m);
1298 }
1299
1300 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1301         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1302 #ifdef DO_RFC_1812_CHECKS
1303
1304 #define IPV4_MIN_VER_IHL        0x45
1305 #define IPV4_MAX_VER_IHL        0x4f
1306 #define IPV4_MAX_VER_IHL_DIFF   (IPV4_MAX_VER_IHL - IPV4_MIN_VER_IHL)
1307
1308 /* Minimum value of IPV4 total length (20B) in network byte order. */
1309 #define IPV4_MIN_LEN_BE (sizeof(struct rte_ipv4_hdr) << 8)
1310
1311 /*
1312  * From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2:
1313  * - The IP version number must be 4.
1314  * - The IP header length field must be large enough to hold the
1315  *    minimum length legal IP datagram (20 bytes = 5 words).
1316  * - The IP total length field must be large enough to hold the IP
1317  *   datagram header, whose length is specified in the IP header length
1318  *   field.
1319  * If we encounter invalid IPV4 packet, then set destination port for it
1320  * to BAD_PORT value.
1321  */
1322 static __rte_always_inline void
1323 rfc1812_process(struct rte_ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t ptype)
1324 {
1325         uint8_t ihl;
1326
1327         if (RTE_ETH_IS_IPV4_HDR(ptype)) {
1328                 ihl = ipv4_hdr->version_ihl - IPV4_MIN_VER_IHL;
1329
1330                 ipv4_hdr->time_to_live--;
1331                 ipv4_hdr->hdr_checksum++;
1332
1333                 if (ihl > IPV4_MAX_VER_IHL_DIFF ||
1334                                 ((uint8_t)ipv4_hdr->total_length == 0 &&
1335                                 ipv4_hdr->total_length < IPV4_MIN_LEN_BE)) {
1336                         dp[0] = BAD_PORT;
1337                 }
1338         }
1339 }
1340
1341 #else
1342 #define rfc1812_process(mb, dp, ptype)  do { } while (0)
1343 #endif /* DO_RFC_1812_CHECKS */
1344 #endif /* APP_LOOKUP_LPM && ENABLE_MULTI_BUFFER_OPTIMIZE */
1345
1346
1347 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1348         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1349
1350 static __rte_always_inline uint16_t
1351 get_dst_port(struct rte_mbuf *pkt, uint32_t dst_ipv4, uint16_t portid)
1352 {
1353         uint32_t next_hop;
1354         struct rte_ipv6_hdr *ipv6_hdr;
1355         struct rte_ether_hdr *eth_hdr;
1356
1357         if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
1358                 return (uint16_t) ((rte_lpm_lookup(
1359                                 RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dst_ipv4,
1360                                 &next_hop) == 0) ? next_hop : portid);
1361
1362         } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
1363
1364                 eth_hdr = rte_pktmbuf_mtod(pkt, struct rte_ether_hdr *);
1365                 ipv6_hdr = (struct rte_ipv6_hdr *)(eth_hdr + 1);
1366
1367                 return (uint16_t) ((rte_lpm6_lookup(
1368                                 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1369                                 ipv6_hdr->dst_addr, &next_hop) == 0) ?
1370                                 next_hop : portid);
1371
1372         }
1373
1374         return portid;
1375 }
1376
1377 static inline void
1378 process_packet(struct rte_mbuf *pkt, uint16_t *dst_port, uint16_t portid)
1379 {
1380         struct rte_ether_hdr *eth_hdr;
1381         struct rte_ipv4_hdr *ipv4_hdr;
1382         uint32_t dst_ipv4;
1383         uint16_t dp;
1384         __m128i te, ve;
1385
1386         eth_hdr = rte_pktmbuf_mtod(pkt, struct rte_ether_hdr *);
1387         ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
1388
1389         dst_ipv4 = ipv4_hdr->dst_addr;
1390         dst_ipv4 = rte_be_to_cpu_32(dst_ipv4);
1391         dp = get_dst_port(pkt, dst_ipv4, portid);
1392
1393         te = _mm_load_si128((__m128i *)eth_hdr);
1394         ve = val_eth[dp];
1395
1396         dst_port[0] = dp;
1397         rfc1812_process(ipv4_hdr, dst_port, pkt->packet_type);
1398
1399         te =  _mm_blend_epi16(te, ve, MASK_ETH);
1400         _mm_store_si128((__m128i *)eth_hdr, te);
1401 }
1402
1403 /*
1404  * Read packet_type and destination IPV4 addresses from 4 mbufs.
1405  */
1406 static inline void
1407 processx4_step1(struct rte_mbuf *pkt[FWDSTEP],
1408                 __m128i *dip,
1409                 uint32_t *ipv4_flag)
1410 {
1411         struct rte_ipv4_hdr *ipv4_hdr;
1412         struct rte_ether_hdr *eth_hdr;
1413         uint32_t x0, x1, x2, x3;
1414
1415         eth_hdr = rte_pktmbuf_mtod(pkt[0], struct rte_ether_hdr *);
1416         ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
1417         x0 = ipv4_hdr->dst_addr;
1418         ipv4_flag[0] = pkt[0]->packet_type & RTE_PTYPE_L3_IPV4;
1419
1420         eth_hdr = rte_pktmbuf_mtod(pkt[1], struct rte_ether_hdr *);
1421         ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
1422         x1 = ipv4_hdr->dst_addr;
1423         ipv4_flag[0] &= pkt[1]->packet_type;
1424
1425         eth_hdr = rte_pktmbuf_mtod(pkt[2], struct rte_ether_hdr *);
1426         ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
1427         x2 = ipv4_hdr->dst_addr;
1428         ipv4_flag[0] &= pkt[2]->packet_type;
1429
1430         eth_hdr = rte_pktmbuf_mtod(pkt[3], struct rte_ether_hdr *);
1431         ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
1432         x3 = ipv4_hdr->dst_addr;
1433         ipv4_flag[0] &= pkt[3]->packet_type;
1434
1435         dip[0] = _mm_set_epi32(x3, x2, x1, x0);
1436 }
1437
1438 /*
1439  * Lookup into LPM for destination port.
1440  * If lookup fails, use incoming port (portid) as destination port.
1441  */
1442 static inline void
1443 processx4_step2(__m128i dip,
1444                 uint32_t ipv4_flag,
1445                 uint16_t portid,
1446                 struct rte_mbuf *pkt[FWDSTEP],
1447                 uint16_t dprt[FWDSTEP])
1448 {
1449         rte_xmm_t dst;
1450         const __m128i bswap_mask = _mm_set_epi8(12, 13, 14, 15, 8, 9, 10, 11,
1451                         4, 5, 6, 7, 0, 1, 2, 3);
1452
1453         /* Byte swap 4 IPV4 addresses. */
1454         dip = _mm_shuffle_epi8(dip, bswap_mask);
1455
1456         /* if all 4 packets are IPV4. */
1457         if (likely(ipv4_flag)) {
1458                 rte_lpm_lookupx4(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dip,
1459                                 dst.u32, portid);
1460
1461                 /* get rid of unused upper 16 bit for each dport. */
1462                 dst.x = _mm_packs_epi32(dst.x, dst.x);
1463                 *(uint64_t *)dprt = dst.u64[0];
1464         } else {
1465                 dst.x = dip;
1466                 dprt[0] = get_dst_port(pkt[0], dst.u32[0], portid);
1467                 dprt[1] = get_dst_port(pkt[1], dst.u32[1], portid);
1468                 dprt[2] = get_dst_port(pkt[2], dst.u32[2], portid);
1469                 dprt[3] = get_dst_port(pkt[3], dst.u32[3], portid);
1470         }
1471 }
1472
1473 /*
1474  * Update source and destination MAC addresses in the ethernet header.
1475  * Perform RFC1812 checks and updates for IPV4 packets.
1476  */
1477 static inline void
1478 processx4_step3(struct rte_mbuf *pkt[FWDSTEP], uint16_t dst_port[FWDSTEP])
1479 {
1480         __m128i te[FWDSTEP];
1481         __m128i ve[FWDSTEP];
1482         __m128i *p[FWDSTEP];
1483
1484         p[0] = rte_pktmbuf_mtod(pkt[0], __m128i *);
1485         p[1] = rte_pktmbuf_mtod(pkt[1], __m128i *);
1486         p[2] = rte_pktmbuf_mtod(pkt[2], __m128i *);
1487         p[3] = rte_pktmbuf_mtod(pkt[3], __m128i *);
1488
1489         ve[0] = val_eth[dst_port[0]];
1490         te[0] = _mm_load_si128(p[0]);
1491
1492         ve[1] = val_eth[dst_port[1]];
1493         te[1] = _mm_load_si128(p[1]);
1494
1495         ve[2] = val_eth[dst_port[2]];
1496         te[2] = _mm_load_si128(p[2]);
1497
1498         ve[3] = val_eth[dst_port[3]];
1499         te[3] = _mm_load_si128(p[3]);
1500
1501         /* Update first 12 bytes, keep rest bytes intact. */
1502         te[0] =  _mm_blend_epi16(te[0], ve[0], MASK_ETH);
1503         te[1] =  _mm_blend_epi16(te[1], ve[1], MASK_ETH);
1504         te[2] =  _mm_blend_epi16(te[2], ve[2], MASK_ETH);
1505         te[3] =  _mm_blend_epi16(te[3], ve[3], MASK_ETH);
1506
1507         _mm_store_si128(p[0], te[0]);
1508         _mm_store_si128(p[1], te[1]);
1509         _mm_store_si128(p[2], te[2]);
1510         _mm_store_si128(p[3], te[3]);
1511
1512         rfc1812_process((struct rte_ipv4_hdr *)
1513                         ((struct rte_ether_hdr *)p[0] + 1),
1514                         &dst_port[0], pkt[0]->packet_type);
1515         rfc1812_process((struct rte_ipv4_hdr *)
1516                         ((struct rte_ether_hdr *)p[1] + 1),
1517                         &dst_port[1], pkt[1]->packet_type);
1518         rfc1812_process((struct rte_ipv4_hdr *)
1519                         ((struct rte_ether_hdr *)p[2] + 1),
1520                         &dst_port[2], pkt[2]->packet_type);
1521         rfc1812_process((struct rte_ipv4_hdr *)
1522                         ((struct rte_ether_hdr *)p[3] + 1),
1523                         &dst_port[3], pkt[3]->packet_type);
1524 }
1525
1526 /*
1527  * We group consecutive packets with the same destionation port into one burst.
1528  * To avoid extra latency this is done together with some other packet
1529  * processing, but after we made a final decision about packet's destination.
1530  * To do this we maintain:
1531  * pnum - array of number of consecutive packets with the same dest port for
1532  * each packet in the input burst.
1533  * lp - pointer to the last updated element in the pnum.
1534  * dlp - dest port value lp corresponds to.
1535  */
1536
1537 #define GRPSZ   (1 << FWDSTEP)
1538 #define GRPMSK  (GRPSZ - 1)
1539
1540 #define GROUP_PORT_STEP(dlp, dcp, lp, pn, idx)  do { \
1541         if (likely((dlp) == (dcp)[(idx)])) {         \
1542                 (lp)[0]++;                           \
1543         } else {                                     \
1544                 (dlp) = (dcp)[idx];                  \
1545                 (lp) = (pn) + (idx);                 \
1546                 (lp)[0] = 1;                         \
1547         }                                            \
1548 } while (0)
1549
1550 /*
1551  * Group consecutive packets with the same destination port in bursts of 4.
1552  * Suppose we have array of destionation ports:
1553  * dst_port[] = {a, b, c, d,, e, ... }
1554  * dp1 should contain: <a, b, c, d>, dp2: <b, c, d, e>.
1555  * We doing 4 comparisons at once and the result is 4 bit mask.
1556  * This mask is used as an index into prebuild array of pnum values.
1557  */
1558 static inline uint16_t *
1559 port_groupx4(uint16_t pn[FWDSTEP + 1], uint16_t *lp, __m128i dp1, __m128i dp2)
1560 {
1561         static const struct {
1562                 uint64_t pnum; /* prebuild 4 values for pnum[]. */
1563                 int32_t  idx;  /* index for new last updated elemnet. */
1564                 uint16_t lpv;  /* add value to the last updated element. */
1565         } gptbl[GRPSZ] = {
1566         {
1567                 /* 0: a != b, b != c, c != d, d != e */
1568                 .pnum = UINT64_C(0x0001000100010001),
1569                 .idx = 4,
1570                 .lpv = 0,
1571         },
1572         {
1573                 /* 1: a == b, b != c, c != d, d != e */
1574                 .pnum = UINT64_C(0x0001000100010002),
1575                 .idx = 4,
1576                 .lpv = 1,
1577         },
1578         {
1579                 /* 2: a != b, b == c, c != d, d != e */
1580                 .pnum = UINT64_C(0x0001000100020001),
1581                 .idx = 4,
1582                 .lpv = 0,
1583         },
1584         {
1585                 /* 3: a == b, b == c, c != d, d != e */
1586                 .pnum = UINT64_C(0x0001000100020003),
1587                 .idx = 4,
1588                 .lpv = 2,
1589         },
1590         {
1591                 /* 4: a != b, b != c, c == d, d != e */
1592                 .pnum = UINT64_C(0x0001000200010001),
1593                 .idx = 4,
1594                 .lpv = 0,
1595         },
1596         {
1597                 /* 5: a == b, b != c, c == d, d != e */
1598                 .pnum = UINT64_C(0x0001000200010002),
1599                 .idx = 4,
1600                 .lpv = 1,
1601         },
1602         {
1603                 /* 6: a != b, b == c, c == d, d != e */
1604                 .pnum = UINT64_C(0x0001000200030001),
1605                 .idx = 4,
1606                 .lpv = 0,
1607         },
1608         {
1609                 /* 7: a == b, b == c, c == d, d != e */
1610                 .pnum = UINT64_C(0x0001000200030004),
1611                 .idx = 4,
1612                 .lpv = 3,
1613         },
1614         {
1615                 /* 8: a != b, b != c, c != d, d == e */
1616                 .pnum = UINT64_C(0x0002000100010001),
1617                 .idx = 3,
1618                 .lpv = 0,
1619         },
1620         {
1621                 /* 9: a == b, b != c, c != d, d == e */
1622                 .pnum = UINT64_C(0x0002000100010002),
1623                 .idx = 3,
1624                 .lpv = 1,
1625         },
1626         {
1627                 /* 0xa: a != b, b == c, c != d, d == e */
1628                 .pnum = UINT64_C(0x0002000100020001),
1629                 .idx = 3,
1630                 .lpv = 0,
1631         },
1632         {
1633                 /* 0xb: a == b, b == c, c != d, d == e */
1634                 .pnum = UINT64_C(0x0002000100020003),
1635                 .idx = 3,
1636                 .lpv = 2,
1637         },
1638         {
1639                 /* 0xc: a != b, b != c, c == d, d == e */
1640                 .pnum = UINT64_C(0x0002000300010001),
1641                 .idx = 2,
1642                 .lpv = 0,
1643         },
1644         {
1645                 /* 0xd: a == b, b != c, c == d, d == e */
1646                 .pnum = UINT64_C(0x0002000300010002),
1647                 .idx = 2,
1648                 .lpv = 1,
1649         },
1650         {
1651                 /* 0xe: a != b, b == c, c == d, d == e */
1652                 .pnum = UINT64_C(0x0002000300040001),
1653                 .idx = 1,
1654                 .lpv = 0,
1655         },
1656         {
1657                 /* 0xf: a == b, b == c, c == d, d == e */
1658                 .pnum = UINT64_C(0x0002000300040005),
1659                 .idx = 0,
1660                 .lpv = 4,
1661         },
1662         };
1663
1664         union {
1665                 uint16_t u16[FWDSTEP + 1];
1666                 uint64_t u64;
1667         } *pnum = (void *)pn;
1668
1669         int32_t v;
1670
1671         dp1 = _mm_cmpeq_epi16(dp1, dp2);
1672         dp1 = _mm_unpacklo_epi16(dp1, dp1);
1673         v = _mm_movemask_ps((__m128)dp1);
1674
1675         /* update last port counter. */
1676         lp[0] += gptbl[v].lpv;
1677
1678         /* if dest port value has changed. */
1679         if (v != GRPMSK) {
1680                 pnum->u64 = gptbl[v].pnum;
1681                 pnum->u16[FWDSTEP] = 1;
1682                 lp = pnum->u16 + gptbl[v].idx;
1683         }
1684
1685         return lp;
1686 }
1687
1688 #endif /* APP_LOOKUP_METHOD */
1689
1690 static void
1691 process_burst(struct rte_mbuf *pkts_burst[MAX_PKT_BURST], int nb_rx,
1692                 uint16_t portid)
1693 {
1694
1695         int j;
1696
1697 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1698         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1699         int32_t k;
1700         uint16_t dlp;
1701         uint16_t *lp;
1702         uint16_t dst_port[MAX_PKT_BURST];
1703         __m128i dip[MAX_PKT_BURST / FWDSTEP];
1704         uint32_t ipv4_flag[MAX_PKT_BURST / FWDSTEP];
1705         uint16_t pnum[MAX_PKT_BURST + 1];
1706 #endif
1707
1708
1709 #if (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
1710 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1711         {
1712                 /*
1713                  * Send nb_rx - nb_rx%8 packets
1714                  * in groups of 8.
1715                  */
1716                 int32_t n = RTE_ALIGN_FLOOR(nb_rx, 8);
1717
1718                 for (j = 0; j < n; j += 8) {
1719                         uint32_t pkt_type =
1720                                 pkts_burst[j]->packet_type &
1721                                 pkts_burst[j+1]->packet_type &
1722                                 pkts_burst[j+2]->packet_type &
1723                                 pkts_burst[j+3]->packet_type &
1724                                 pkts_burst[j+4]->packet_type &
1725                                 pkts_burst[j+5]->packet_type &
1726                                 pkts_burst[j+6]->packet_type &
1727                                 pkts_burst[j+7]->packet_type;
1728                         if (pkt_type & RTE_PTYPE_L3_IPV4) {
1729                                 simple_ipv4_fwd_8pkts(&pkts_burst[j], portid);
1730                         } else if (pkt_type &
1731                                 RTE_PTYPE_L3_IPV6) {
1732                                 simple_ipv6_fwd_8pkts(&pkts_burst[j], portid);
1733                         } else {
1734                                 l3fwd_simple_forward(pkts_burst[j], portid);
1735                                 l3fwd_simple_forward(pkts_burst[j+1], portid);
1736                                 l3fwd_simple_forward(pkts_burst[j+2], portid);
1737                                 l3fwd_simple_forward(pkts_burst[j+3], portid);
1738                                 l3fwd_simple_forward(pkts_burst[j+4], portid);
1739                                 l3fwd_simple_forward(pkts_burst[j+5], portid);
1740                                 l3fwd_simple_forward(pkts_burst[j+6], portid);
1741                                 l3fwd_simple_forward(pkts_burst[j+7], portid);
1742                         }
1743                 }
1744                 for (; j < nb_rx ; j++)
1745                         l3fwd_simple_forward(pkts_burst[j], portid);
1746         }
1747 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1748
1749         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1750         for (j = 0; j != k; j += FWDSTEP)
1751                 processx4_step1(&pkts_burst[j], &dip[j / FWDSTEP],
1752                                 &ipv4_flag[j / FWDSTEP]);
1753
1754         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1755         for (j = 0; j != k; j += FWDSTEP)
1756                 processx4_step2(dip[j / FWDSTEP], ipv4_flag[j / FWDSTEP],
1757                                 portid, &pkts_burst[j], &dst_port[j]);
1758
1759         /*
1760          * Finish packet processing and group consecutive
1761          * packets with the same destination port.
1762          */
1763         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1764         if (k != 0) {
1765                 __m128i dp1, dp2;
1766
1767                 lp = pnum;
1768                 lp[0] = 1;
1769
1770                 processx4_step3(pkts_burst, dst_port);
1771
1772                 /* dp1: <d[0], d[1], d[2], d[3], ... > */
1773                 dp1 = _mm_loadu_si128((__m128i *)dst_port);
1774
1775                 for (j = FWDSTEP; j != k; j += FWDSTEP) {
1776                         processx4_step3(&pkts_burst[j], &dst_port[j]);
1777
1778                         /*
1779                          * dp2:
1780                          * <d[j-3], d[j-2], d[j-1], d[j], ... >
1781                          */
1782                         dp2 = _mm_loadu_si128(
1783                                         (__m128i *)&dst_port[j - FWDSTEP + 1]);
1784                         lp  = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1785
1786                         /*
1787                          * dp1:
1788                          * <d[j], d[j+1], d[j+2], d[j+3], ... >
1789                          */
1790                         dp1 = _mm_srli_si128(dp2, (FWDSTEP - 1) *
1791                                         sizeof(dst_port[0]));
1792                 }
1793
1794                 /*
1795                  * dp2: <d[j-3], d[j-2], d[j-1], d[j-1], ... >
1796                  */
1797                 dp2 = _mm_shufflelo_epi16(dp1, 0xf9);
1798                 lp  = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1799
1800                 /*
1801                  * remove values added by the last repeated
1802                  * dst port.
1803                  */
1804                 lp[0]--;
1805                 dlp = dst_port[j - 1];
1806         } else {
1807                 /* set dlp and lp to the never used values. */
1808                 dlp = BAD_PORT - 1;
1809                 lp = pnum + MAX_PKT_BURST;
1810         }
1811
1812         /* Process up to last 3 packets one by one. */
1813         switch (nb_rx % FWDSTEP) {
1814         case 3:
1815                 process_packet(pkts_burst[j], dst_port + j, portid);
1816                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1817                 j++;
1818                 /* fall-through */
1819         case 2:
1820                 process_packet(pkts_burst[j], dst_port + j, portid);
1821                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1822                 j++;
1823                 /* fall-through */
1824         case 1:
1825                 process_packet(pkts_burst[j], dst_port + j, portid);
1826                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1827                 j++;
1828         }
1829
1830         /*
1831          * Send packets out, through destination port.
1832          * Consecuteve pacekts with the same destination port
1833          * are already grouped together.
1834          * If destination port for the packet equals BAD_PORT,
1835          * then free the packet without sending it out.
1836          */
1837         for (j = 0; j < nb_rx; j += k) {
1838
1839                 int32_t m;
1840                 uint16_t pn;
1841
1842                 pn = dst_port[j];
1843                 k = pnum[j];
1844
1845                 if (likely(pn != BAD_PORT))
1846                         send_packetsx4(pn, pkts_burst + j, k);
1847                 else
1848                         for (m = j; m != j + k; m++)
1849                                 rte_pktmbuf_free(pkts_burst[m]);
1850
1851         }
1852
1853 #endif /* APP_LOOKUP_METHOD */
1854 #else /* ENABLE_MULTI_BUFFER_OPTIMIZE == 0 */
1855
1856         /* Prefetch first packets */
1857         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++)
1858                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[j], void *));
1859
1860         /* Prefetch and forward already prefetched packets */
1861         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1862                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1863                                 j + PREFETCH_OFFSET], void *));
1864                 l3fwd_simple_forward(pkts_burst[j], portid);
1865         }
1866
1867         /* Forward remaining prefetched packets */
1868         for (; j < nb_rx; j++)
1869                 l3fwd_simple_forward(pkts_burst[j], portid);
1870
1871 #endif /* ENABLE_MULTI_BUFFER_OPTIMIZE */
1872
1873 }
1874
1875 #if (APP_CPU_LOAD > 0)
1876
1877 /*
1878  * CPU-load stats collector
1879  */
1880 static int __rte_noreturn
1881 cpu_load_collector(__rte_unused void *arg) {
1882         unsigned i, j, k;
1883         uint64_t hits;
1884         uint64_t prev_tsc, diff_tsc, cur_tsc;
1885         uint64_t total[MAX_CPU] = { 0 };
1886         unsigned min_cpu = MAX_CPU;
1887         unsigned max_cpu = 0;
1888         unsigned cpu_id;
1889         int busy_total = 0;
1890         int busy_flag = 0;
1891
1892         unsigned int n_thread_per_cpu[MAX_CPU] = { 0 };
1893         struct thread_conf *thread_per_cpu[MAX_CPU][MAX_THREAD];
1894
1895         struct thread_conf *thread_conf;
1896
1897         const uint64_t interval_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1898                 US_PER_S * CPU_LOAD_TIMEOUT_US;
1899
1900         prev_tsc = 0;
1901         /*
1902          * Wait for all threads
1903          */
1904
1905         printf("Waiting for %d rx threads and %d tx threads\n", n_rx_thread,
1906                         n_tx_thread);
1907
1908         while (rte_atomic16_read(&rx_counter) < n_rx_thread)
1909                 rte_pause();
1910
1911         while (rte_atomic16_read(&tx_counter) < n_tx_thread)
1912                 rte_pause();
1913
1914         for (i = 0; i < n_rx_thread; i++) {
1915
1916                 thread_conf = &rx_thread[i].conf;
1917                 cpu_id = thread_conf->cpu_id;
1918                 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1919
1920                 if (cpu_id > max_cpu)
1921                         max_cpu = cpu_id;
1922                 if (cpu_id < min_cpu)
1923                         min_cpu = cpu_id;
1924         }
1925         for (i = 0; i < n_tx_thread; i++) {
1926
1927                 thread_conf = &tx_thread[i].conf;
1928                 cpu_id = thread_conf->cpu_id;
1929                 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1930
1931                 if (thread_conf->cpu_id > max_cpu)
1932                         max_cpu = thread_conf->cpu_id;
1933                 if (thread_conf->cpu_id < min_cpu)
1934                         min_cpu = thread_conf->cpu_id;
1935         }
1936
1937         while (1) {
1938
1939                 cpu_load.counter++;
1940                 for (i = min_cpu; i <= max_cpu; i++) {
1941                         for (j = 0; j < MAX_CPU_COUNTER; j++) {
1942                                 for (k = 0; k < n_thread_per_cpu[i]; k++)
1943                                         if (thread_per_cpu[i][k]->busy[j]) {
1944                                                 busy_flag = 1;
1945                                                 break;
1946                                         }
1947                                 if (busy_flag) {
1948                                         cpu_load.hits[j][i]++;
1949                                         busy_total = 1;
1950                                         busy_flag = 0;
1951                                 }
1952                         }
1953
1954                         if (busy_total) {
1955                                 total[i]++;
1956                                 busy_total = 0;
1957                         }
1958                 }
1959
1960                 cur_tsc = rte_rdtsc();
1961
1962                 diff_tsc = cur_tsc - prev_tsc;
1963                 if (unlikely(diff_tsc > interval_tsc)) {
1964
1965                         printf("\033c");
1966
1967                         printf("Cpu usage for %d rx threads and %d tx threads:\n\n",
1968                                         n_rx_thread, n_tx_thread);
1969
1970                         printf("cpu#     proc%%  poll%%  overhead%%\n\n");
1971
1972                         for (i = min_cpu; i <= max_cpu; i++) {
1973                                 hits = 0;
1974                                 printf("CPU %d:", i);
1975                                 for (j = 0; j < MAX_CPU_COUNTER; j++) {
1976                                         printf("%7" PRIu64 "",
1977                                                         cpu_load.hits[j][i] * 100 / cpu_load.counter);
1978                                         hits += cpu_load.hits[j][i];
1979                                         cpu_load.hits[j][i] = 0;
1980                                 }
1981                                 printf("%7" PRIu64 "\n",
1982                                                 100 - total[i] * 100 / cpu_load.counter);
1983                                 total[i] = 0;
1984                         }
1985                         cpu_load.counter = 0;
1986
1987                         prev_tsc = cur_tsc;
1988                 }
1989
1990         }
1991 }
1992 #endif /* APP_CPU_LOAD */
1993
1994 /*
1995  * Null processing lthread loop
1996  *
1997  * This loop is used to start empty scheduler on lcore.
1998  */
1999 static void *
2000 lthread_null(__rte_unused void *args)
2001 {
2002         int lcore_id = rte_lcore_id();
2003
2004         RTE_LOG(INFO, L3FWD, "Starting scheduler on lcore %d.\n", lcore_id);
2005         lthread_exit(NULL);
2006         return NULL;
2007 }
2008
2009 /* main processing loop */
2010 static void *
2011 lthread_tx_per_ring(void *dummy)
2012 {
2013         int nb_rx;
2014         uint16_t portid;
2015         struct rte_ring *ring;
2016         struct thread_tx_conf *tx_conf;
2017         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2018         struct lthread_cond *ready;
2019
2020         tx_conf = (struct thread_tx_conf *)dummy;
2021         ring = tx_conf->ring;
2022         ready = *tx_conf->ready;
2023
2024         lthread_set_data((void *)tx_conf);
2025
2026         /*
2027          * Move this lthread to lcore
2028          */
2029         lthread_set_affinity(tx_conf->conf.lcore_id);
2030
2031         RTE_LOG(INFO, L3FWD, "entering main tx loop on lcore %u\n", rte_lcore_id());
2032
2033         nb_rx = 0;
2034         rte_atomic16_inc(&tx_counter);
2035         while (1) {
2036
2037                 /*
2038                  * Read packet from ring
2039                  */
2040                 SET_CPU_BUSY(tx_conf, CPU_POLL);
2041                 nb_rx = rte_ring_sc_dequeue_burst(ring, (void **)pkts_burst,
2042                                 MAX_PKT_BURST, NULL);
2043                 SET_CPU_IDLE(tx_conf, CPU_POLL);
2044
2045                 if (nb_rx > 0) {
2046                         SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2047                         portid = pkts_burst[0]->port;
2048                         process_burst(pkts_burst, nb_rx, portid);
2049                         SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2050                         lthread_yield();
2051                 } else
2052                         lthread_cond_wait(ready, 0);
2053
2054         }
2055         return NULL;
2056 }
2057
2058 /*
2059  * Main tx-lthreads spawner lthread.
2060  *
2061  * This lthread is used to spawn one new lthread per ring from producers.
2062  *
2063  */
2064 static void *
2065 lthread_tx(void *args)
2066 {
2067         struct lthread *lt;
2068
2069         unsigned lcore_id;
2070         uint16_t portid;
2071         struct thread_tx_conf *tx_conf;
2072
2073         tx_conf = (struct thread_tx_conf *)args;
2074         lthread_set_data((void *)tx_conf);
2075
2076         /*
2077          * Move this lthread to the selected lcore
2078          */
2079         lthread_set_affinity(tx_conf->conf.lcore_id);
2080
2081         /*
2082          * Spawn tx readers (one per input ring)
2083          */
2084         lthread_create(&lt, tx_conf->conf.lcore_id, lthread_tx_per_ring,
2085                         (void *)tx_conf);
2086
2087         lcore_id = rte_lcore_id();
2088
2089         RTE_LOG(INFO, L3FWD, "Entering Tx main loop on lcore %u\n", lcore_id);
2090
2091         tx_conf->conf.cpu_id = sched_getcpu();
2092         while (1) {
2093
2094                 lthread_sleep(BURST_TX_DRAIN_US * 1000);
2095
2096                 /*
2097                  * TX burst queue drain
2098                  */
2099                 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2100                         if (tx_conf->tx_mbufs[portid].len == 0)
2101                                 continue;
2102                         SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2103                         send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2104                         SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2105                         tx_conf->tx_mbufs[portid].len = 0;
2106                 }
2107
2108         }
2109         return NULL;
2110 }
2111
2112 static void *
2113 lthread_rx(void *dummy)
2114 {
2115         int ret;
2116         uint16_t nb_rx;
2117         int i;
2118         uint16_t portid;
2119         uint8_t queueid;
2120         int worker_id;
2121         int len[RTE_MAX_LCORE] = { 0 };
2122         int old_len, new_len;
2123         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2124         struct thread_rx_conf *rx_conf;
2125
2126         rx_conf = (struct thread_rx_conf *)dummy;
2127         lthread_set_data((void *)rx_conf);
2128
2129         /*
2130          * Move this lthread to lcore
2131          */
2132         lthread_set_affinity(rx_conf->conf.lcore_id);
2133
2134         if (rx_conf->n_rx_queue == 0) {
2135                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", rte_lcore_id());
2136                 return NULL;
2137         }
2138
2139         RTE_LOG(INFO, L3FWD, "Entering main Rx loop on lcore %u\n", rte_lcore_id());
2140
2141         for (i = 0; i < rx_conf->n_rx_queue; i++) {
2142
2143                 portid = rx_conf->rx_queue_list[i].port_id;
2144                 queueid = rx_conf->rx_queue_list[i].queue_id;
2145                 RTE_LOG(INFO, L3FWD,
2146                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2147                                 rte_lcore_id(), portid, queueid);
2148         }
2149
2150         /*
2151          * Init all condition variables (one per rx thread)
2152          */
2153         for (i = 0; i < rx_conf->n_rx_queue; i++)
2154                 lthread_cond_init(NULL, &rx_conf->ready[i], NULL);
2155
2156         worker_id = 0;
2157
2158         rx_conf->conf.cpu_id = sched_getcpu();
2159         rte_atomic16_inc(&rx_counter);
2160         while (1) {
2161
2162                 /*
2163                  * Read packet from RX queues
2164                  */
2165                 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2166                         portid = rx_conf->rx_queue_list[i].port_id;
2167                         queueid = rx_conf->rx_queue_list[i].queue_id;
2168
2169                         SET_CPU_BUSY(rx_conf, CPU_POLL);
2170                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2171                                 MAX_PKT_BURST);
2172                         SET_CPU_IDLE(rx_conf, CPU_POLL);
2173
2174                         if (nb_rx != 0) {
2175                                 worker_id = (worker_id + 1) % rx_conf->n_ring;
2176                                 old_len = len[worker_id];
2177
2178                                 SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2179                                 ret = rte_ring_sp_enqueue_burst(
2180                                                 rx_conf->ring[worker_id],
2181                                                 (void **) pkts_burst,
2182                                                 nb_rx, NULL);
2183
2184                                 new_len = old_len + ret;
2185
2186                                 if (new_len >= BURST_SIZE) {
2187                                         lthread_cond_signal(rx_conf->ready[worker_id]);
2188                                         new_len = 0;
2189                                 }
2190
2191                                 len[worker_id] = new_len;
2192
2193                                 if (unlikely(ret < nb_rx)) {
2194                                         uint32_t k;
2195
2196                                         for (k = ret; k < nb_rx; k++) {
2197                                                 struct rte_mbuf *m = pkts_burst[k];
2198
2199                                                 rte_pktmbuf_free(m);
2200                                         }
2201                                 }
2202                                 SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2203                         }
2204
2205                         lthread_yield();
2206                 }
2207         }
2208         return NULL;
2209 }
2210
2211 /*
2212  * Start scheduler with initial lthread on lcore
2213  *
2214  * This lthread loop spawns all rx and tx lthreads on master lcore
2215  */
2216
2217 static void *
2218 lthread_spawner(__rte_unused void *arg)
2219 {
2220         struct lthread *lt[MAX_THREAD];
2221         int i;
2222         int n_thread = 0;
2223
2224         printf("Entering lthread_spawner\n");
2225
2226         /*
2227          * Create producers (rx threads) on default lcore
2228          */
2229         for (i = 0; i < n_rx_thread; i++) {
2230                 rx_thread[i].conf.thread_id = i;
2231                 lthread_create(&lt[n_thread], -1, lthread_rx,
2232                                 (void *)&rx_thread[i]);
2233                 n_thread++;
2234         }
2235
2236         /*
2237          * Wait for all producers. Until some producers can be started on the same
2238          * scheduler as this lthread, yielding is required to let them to run and
2239          * prevent deadlock here.
2240          */
2241         while (rte_atomic16_read(&rx_counter) < n_rx_thread)
2242                 lthread_sleep(100000);
2243
2244         /*
2245          * Create consumers (tx threads) on default lcore_id
2246          */
2247         for (i = 0; i < n_tx_thread; i++) {
2248                 tx_thread[i].conf.thread_id = i;
2249                 lthread_create(&lt[n_thread], -1, lthread_tx,
2250                                 (void *)&tx_thread[i]);
2251                 n_thread++;
2252         }
2253
2254         /*
2255          * Wait for all threads finished
2256          */
2257         for (i = 0; i < n_thread; i++)
2258                 lthread_join(lt[i], NULL);
2259
2260         return NULL;
2261 }
2262
2263 /*
2264  * Start master scheduler with initial lthread spawning rx and tx lthreads
2265  * (main_lthread_master).
2266  */
2267 static int
2268 lthread_master_spawner(__rte_unused void *arg) {
2269         struct lthread *lt;
2270         int lcore_id = rte_lcore_id();
2271
2272         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2273         lthread_create(&lt, -1, lthread_spawner, NULL);
2274         lthread_run();
2275
2276         return 0;
2277 }
2278
2279 /*
2280  * Start scheduler on lcore.
2281  */
2282 static int
2283 sched_spawner(__rte_unused void *arg) {
2284         struct lthread *lt;
2285         int lcore_id = rte_lcore_id();
2286
2287 #if (APP_CPU_LOAD)
2288         if (lcore_id == cpu_load_lcore_id) {
2289                 cpu_load_collector(arg);
2290                 return 0;
2291         }
2292 #endif /* APP_CPU_LOAD */
2293
2294         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2295         lthread_create(&lt, -1, lthread_null, NULL);
2296         lthread_run();
2297
2298         return 0;
2299 }
2300
2301 /* main processing loop */
2302 static int __rte_noreturn
2303 pthread_tx(void *dummy)
2304 {
2305         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2306         uint64_t prev_tsc, diff_tsc, cur_tsc;
2307         int nb_rx;
2308         uint16_t portid;
2309         struct thread_tx_conf *tx_conf;
2310
2311         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
2312                 US_PER_S * BURST_TX_DRAIN_US;
2313
2314         prev_tsc = 0;
2315
2316         tx_conf = (struct thread_tx_conf *)dummy;
2317
2318         RTE_LOG(INFO, L3FWD, "Entering main Tx loop on lcore %u\n", rte_lcore_id());
2319
2320         tx_conf->conf.cpu_id = sched_getcpu();
2321         rte_atomic16_inc(&tx_counter);
2322         while (1) {
2323
2324                 cur_tsc = rte_rdtsc();
2325
2326                 /*
2327                  * TX burst queue drain
2328                  */
2329                 diff_tsc = cur_tsc - prev_tsc;
2330                 if (unlikely(diff_tsc > drain_tsc)) {
2331
2332                         /*
2333                          * This could be optimized (use queueid instead of
2334                          * portid), but it is not called so often
2335                          */
2336                         SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2337                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2338                                 if (tx_conf->tx_mbufs[portid].len == 0)
2339                                         continue;
2340                                 send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2341                                 tx_conf->tx_mbufs[portid].len = 0;
2342                         }
2343                         SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2344
2345                         prev_tsc = cur_tsc;
2346                 }
2347
2348                 /*
2349                  * Read packet from ring
2350                  */
2351                 SET_CPU_BUSY(tx_conf, CPU_POLL);
2352                 nb_rx = rte_ring_sc_dequeue_burst(tx_conf->ring,
2353                                 (void **)pkts_burst, MAX_PKT_BURST, NULL);
2354                 SET_CPU_IDLE(tx_conf, CPU_POLL);
2355
2356                 if (unlikely(nb_rx == 0)) {
2357                         sched_yield();
2358                         continue;
2359                 }
2360
2361                 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2362                 portid = pkts_burst[0]->port;
2363                 process_burst(pkts_burst, nb_rx, portid);
2364                 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2365
2366         }
2367 }
2368
2369 static int
2370 pthread_rx(void *dummy)
2371 {
2372         int i;
2373         int worker_id;
2374         uint32_t n;
2375         uint32_t nb_rx;
2376         unsigned lcore_id;
2377         uint8_t queueid;
2378         uint16_t portid;
2379         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2380
2381         struct thread_rx_conf *rx_conf;
2382
2383         lcore_id = rte_lcore_id();
2384         rx_conf = (struct thread_rx_conf *)dummy;
2385
2386         if (rx_conf->n_rx_queue == 0) {
2387                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
2388                 return 0;
2389         }
2390
2391         RTE_LOG(INFO, L3FWD, "entering main rx loop on lcore %u\n", lcore_id);
2392
2393         for (i = 0; i < rx_conf->n_rx_queue; i++) {
2394
2395                 portid = rx_conf->rx_queue_list[i].port_id;
2396                 queueid = rx_conf->rx_queue_list[i].queue_id;
2397                 RTE_LOG(INFO, L3FWD,
2398                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2399                                 lcore_id, portid, queueid);
2400         }
2401
2402         worker_id = 0;
2403         rx_conf->conf.cpu_id = sched_getcpu();
2404         rte_atomic16_inc(&rx_counter);
2405         while (1) {
2406
2407                 /*
2408                  * Read packet from RX queues
2409                  */
2410                 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2411                         portid = rx_conf->rx_queue_list[i].port_id;
2412                         queueid = rx_conf->rx_queue_list[i].queue_id;
2413
2414                         SET_CPU_BUSY(rx_conf, CPU_POLL);
2415                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2416                                 MAX_PKT_BURST);
2417                         SET_CPU_IDLE(rx_conf, CPU_POLL);
2418
2419                         if (nb_rx == 0) {
2420                                 sched_yield();
2421                                 continue;
2422                         }
2423
2424                         SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2425                         worker_id = (worker_id + 1) % rx_conf->n_ring;
2426                         n = rte_ring_sp_enqueue_burst(rx_conf->ring[worker_id],
2427                                         (void **)pkts_burst, nb_rx, NULL);
2428
2429                         if (unlikely(n != nb_rx)) {
2430                                 uint32_t k;
2431
2432                                 for (k = n; k < nb_rx; k++) {
2433                                         struct rte_mbuf *m = pkts_burst[k];
2434
2435                                         rte_pktmbuf_free(m);
2436                                 }
2437                         }
2438
2439                         SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2440
2441                 }
2442         }
2443 }
2444
2445 /*
2446  * P-Thread spawner.
2447  */
2448 static int
2449 pthread_run(__rte_unused void *arg) {
2450         int lcore_id = rte_lcore_id();
2451         int i;
2452
2453         for (i = 0; i < n_rx_thread; i++)
2454                 if (rx_thread[i].conf.lcore_id == lcore_id) {
2455                         printf("Start rx thread on %d...\n", lcore_id);
2456                         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2457                         RTE_PER_LCORE(lcore_conf)->data = (void *)&rx_thread[i];
2458                         pthread_rx((void *)&rx_thread[i]);
2459                         return 0;
2460                 }
2461
2462         for (i = 0; i < n_tx_thread; i++)
2463                 if (tx_thread[i].conf.lcore_id == lcore_id) {
2464                         printf("Start tx thread on %d...\n", lcore_id);
2465                         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2466                         RTE_PER_LCORE(lcore_conf)->data = (void *)&tx_thread[i];
2467                         pthread_tx((void *)&tx_thread[i]);
2468                         return 0;
2469                 }
2470
2471 #if (APP_CPU_LOAD)
2472         if (lcore_id == cpu_load_lcore_id)
2473                 cpu_load_collector(arg);
2474 #endif /* APP_CPU_LOAD */
2475
2476         return 0;
2477 }
2478
2479 static int
2480 check_lcore_params(void)
2481 {
2482         uint8_t queue, lcore;
2483         uint16_t i;
2484         int socketid;
2485
2486         for (i = 0; i < nb_rx_thread_params; ++i) {
2487                 queue = rx_thread_params[i].queue_id;
2488                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
2489                         printf("invalid queue number: %hhu\n", queue);
2490                         return -1;
2491                 }
2492                 lcore = rx_thread_params[i].lcore_id;
2493                 if (!rte_lcore_is_enabled(lcore)) {
2494                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
2495                         return -1;
2496                 }
2497                 socketid = rte_lcore_to_socket_id(lcore);
2498                 if ((socketid != 0) && (numa_on == 0))
2499                         printf("warning: lcore %hhu is on socket %d with numa off\n",
2500                                 lcore, socketid);
2501         }
2502         return 0;
2503 }
2504
2505 static int
2506 check_port_config(void)
2507 {
2508         unsigned portid;
2509         uint16_t i;
2510
2511         for (i = 0; i < nb_rx_thread_params; ++i) {
2512                 portid = rx_thread_params[i].port_id;
2513                 if ((enabled_port_mask & (1 << portid)) == 0) {
2514                         printf("port %u is not enabled in port mask\n", portid);
2515                         return -1;
2516                 }
2517                 if (!rte_eth_dev_is_valid_port(portid)) {
2518                         printf("port %u is not present on the board\n", portid);
2519                         return -1;
2520                 }
2521         }
2522         return 0;
2523 }
2524
2525 static uint8_t
2526 get_port_n_rx_queues(const uint16_t port)
2527 {
2528         int queue = -1;
2529         uint16_t i;
2530
2531         for (i = 0; i < nb_rx_thread_params; ++i)
2532                 if (rx_thread_params[i].port_id == port &&
2533                                 rx_thread_params[i].queue_id > queue)
2534                         queue = rx_thread_params[i].queue_id;
2535
2536         return (uint8_t)(++queue);
2537 }
2538
2539 static int
2540 init_rx_rings(void)
2541 {
2542         unsigned socket_io;
2543         struct thread_rx_conf *rx_conf;
2544         struct thread_tx_conf *tx_conf;
2545         unsigned rx_thread_id, tx_thread_id;
2546         char name[256];
2547         struct rte_ring *ring = NULL;
2548
2549         for (tx_thread_id = 0; tx_thread_id < n_tx_thread; tx_thread_id++) {
2550
2551                 tx_conf = &tx_thread[tx_thread_id];
2552
2553                 printf("Connecting tx-thread %d with rx-thread %d\n", tx_thread_id,
2554                                 tx_conf->conf.thread_id);
2555
2556                 rx_thread_id = tx_conf->conf.thread_id;
2557                 if (rx_thread_id > n_tx_thread) {
2558                         printf("connection from tx-thread %u to rx-thread %u fails "
2559                                         "(rx-thread not defined)\n", tx_thread_id, rx_thread_id);
2560                         return -1;
2561                 }
2562
2563                 rx_conf = &rx_thread[rx_thread_id];
2564                 socket_io = rte_lcore_to_socket_id(rx_conf->conf.lcore_id);
2565
2566                 snprintf(name, sizeof(name), "app_ring_s%u_rx%u_tx%u",
2567                                 socket_io, rx_thread_id, tx_thread_id);
2568
2569                 ring = rte_ring_create(name, 1024 * 4, socket_io,
2570                                 RING_F_SP_ENQ | RING_F_SC_DEQ);
2571
2572                 if (ring == NULL) {
2573                         rte_panic("Cannot create ring to connect rx-thread %u "
2574                                         "with tx-thread %u\n", rx_thread_id, tx_thread_id);
2575                 }
2576
2577                 rx_conf->ring[rx_conf->n_ring] = ring;
2578
2579                 tx_conf->ring = ring;
2580                 tx_conf->ready = &rx_conf->ready[rx_conf->n_ring];
2581
2582                 rx_conf->n_ring++;
2583         }
2584         return 0;
2585 }
2586
2587 static int
2588 init_rx_queues(void)
2589 {
2590         uint16_t i, nb_rx_queue;
2591         uint8_t thread;
2592
2593         n_rx_thread = 0;
2594
2595         for (i = 0; i < nb_rx_thread_params; ++i) {
2596                 thread = rx_thread_params[i].thread_id;
2597                 nb_rx_queue = rx_thread[thread].n_rx_queue;
2598
2599                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
2600                         printf("error: too many queues (%u) for thread: %u\n",
2601                                 (unsigned)nb_rx_queue + 1, (unsigned)thread);
2602                         return -1;
2603                 }
2604
2605                 rx_thread[thread].conf.thread_id = thread;
2606                 rx_thread[thread].conf.lcore_id = rx_thread_params[i].lcore_id;
2607                 rx_thread[thread].rx_queue_list[nb_rx_queue].port_id =
2608                         rx_thread_params[i].port_id;
2609                 rx_thread[thread].rx_queue_list[nb_rx_queue].queue_id =
2610                         rx_thread_params[i].queue_id;
2611                 rx_thread[thread].n_rx_queue++;
2612
2613                 if (thread >= n_rx_thread)
2614                         n_rx_thread = thread + 1;
2615
2616         }
2617         return 0;
2618 }
2619
2620 static int
2621 init_tx_threads(void)
2622 {
2623         int i;
2624
2625         n_tx_thread = 0;
2626         for (i = 0; i < nb_tx_thread_params; ++i) {
2627                 tx_thread[n_tx_thread].conf.thread_id = tx_thread_params[i].thread_id;
2628                 tx_thread[n_tx_thread].conf.lcore_id = tx_thread_params[i].lcore_id;
2629                 n_tx_thread++;
2630         }
2631         return 0;
2632 }
2633
2634 /* display usage */
2635 static void
2636 print_usage(const char *prgname)
2637 {
2638         printf("%s [EAL options] -- -p PORTMASK -P"
2639                 "  [--rx (port,queue,lcore,thread)[,(port,queue,lcore,thread]]"
2640                 "  [--tx (lcore,thread)[,(lcore,thread]]"
2641                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
2642                 "  [--parse-ptype]\n\n"
2643                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
2644                 "  -P : enable promiscuous mode\n"
2645                 "  --rx (port,queue,lcore,thread): rx queues configuration\n"
2646                 "  --tx (lcore,thread): tx threads configuration\n"
2647                 "  --stat-lcore LCORE: use lcore for stat collector\n"
2648                 "  --eth-dest=X,MM:MM:MM:MM:MM:MM: optional, ethernet destination for port X\n"
2649                 "  --no-numa: optional, disable numa awareness\n"
2650                 "  --ipv6: optional, specify it if running ipv6 packets\n"
2651                 "  --enable-jumbo: enable jumbo frame"
2652                 " which max packet len is PKTLEN in decimal (64-9600)\n"
2653                 "  --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n"
2654                 "  --no-lthreads: turn off lthread model\n"
2655                 "  --parse-ptype: set to use software to analyze packet type\n\n",
2656                 prgname);
2657 }
2658
2659 static int parse_max_pkt_len(const char *pktlen)
2660 {
2661         char *end = NULL;
2662         unsigned long len;
2663
2664         /* parse decimal string */
2665         len = strtoul(pktlen, &end, 10);
2666         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
2667                 return -1;
2668
2669         if (len == 0)
2670                 return -1;
2671
2672         return len;
2673 }
2674
2675 static int
2676 parse_portmask(const char *portmask)
2677 {
2678         char *end = NULL;
2679         unsigned long pm;
2680
2681         /* parse hexadecimal string */
2682         pm = strtoul(portmask, &end, 16);
2683         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
2684                 return 0;
2685
2686         return pm;
2687 }
2688
2689 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2690 static int
2691 parse_hash_entry_number(const char *hash_entry_num)
2692 {
2693         char *end = NULL;
2694         unsigned long hash_en;
2695
2696         /* parse hexadecimal string */
2697         hash_en = strtoul(hash_entry_num, &end, 16);
2698         if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
2699                 return -1;
2700
2701         if (hash_en == 0)
2702                 return -1;
2703
2704         return hash_en;
2705 }
2706 #endif
2707
2708 static int
2709 parse_rx_config(const char *q_arg)
2710 {
2711         char s[256];
2712         const char *p, *p0 = q_arg;
2713         char *end;
2714         enum fieldnames {
2715                 FLD_PORT = 0,
2716                 FLD_QUEUE,
2717                 FLD_LCORE,
2718                 FLD_THREAD,
2719                 _NUM_FLD
2720         };
2721         unsigned long int_fld[_NUM_FLD];
2722         char *str_fld[_NUM_FLD];
2723         int i;
2724         unsigned size;
2725
2726         nb_rx_thread_params = 0;
2727
2728         while ((p = strchr(p0, '(')) != NULL) {
2729                 ++p;
2730                 p0 = strchr(p, ')');
2731                 if (p0 == NULL)
2732                         return -1;
2733
2734                 size = p0 - p;
2735                 if (size >= sizeof(s))
2736                         return -1;
2737
2738                 snprintf(s, sizeof(s), "%.*s", size, p);
2739                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2740                         return -1;
2741                 for (i = 0; i < _NUM_FLD; i++) {
2742                         errno = 0;
2743                         int_fld[i] = strtoul(str_fld[i], &end, 0);
2744                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2745                                 return -1;
2746                 }
2747                 if (nb_rx_thread_params >= MAX_LCORE_PARAMS) {
2748                         printf("exceeded max number of rx params: %hu\n",
2749                                         nb_rx_thread_params);
2750                         return -1;
2751                 }
2752                 rx_thread_params_array[nb_rx_thread_params].port_id =
2753                                 int_fld[FLD_PORT];
2754                 rx_thread_params_array[nb_rx_thread_params].queue_id =
2755                                 (uint8_t)int_fld[FLD_QUEUE];
2756                 rx_thread_params_array[nb_rx_thread_params].lcore_id =
2757                                 (uint8_t)int_fld[FLD_LCORE];
2758                 rx_thread_params_array[nb_rx_thread_params].thread_id =
2759                                 (uint8_t)int_fld[FLD_THREAD];
2760                 ++nb_rx_thread_params;
2761         }
2762         rx_thread_params = rx_thread_params_array;
2763         return 0;
2764 }
2765
2766 static int
2767 parse_tx_config(const char *q_arg)
2768 {
2769         char s[256];
2770         const char *p, *p0 = q_arg;
2771         char *end;
2772         enum fieldnames {
2773                 FLD_LCORE = 0,
2774                 FLD_THREAD,
2775                 _NUM_FLD
2776         };
2777         unsigned long int_fld[_NUM_FLD];
2778         char *str_fld[_NUM_FLD];
2779         int i;
2780         unsigned size;
2781
2782         nb_tx_thread_params = 0;
2783
2784         while ((p = strchr(p0, '(')) != NULL) {
2785                 ++p;
2786                 p0 = strchr(p, ')');
2787                 if (p0 == NULL)
2788                         return -1;
2789
2790                 size = p0 - p;
2791                 if (size >= sizeof(s))
2792                         return -1;
2793
2794                 snprintf(s, sizeof(s), "%.*s", size, p);
2795                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2796                         return -1;
2797                 for (i = 0; i < _NUM_FLD; i++) {
2798                         errno = 0;
2799                         int_fld[i] = strtoul(str_fld[i], &end, 0);
2800                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2801                                 return -1;
2802                 }
2803                 if (nb_tx_thread_params >= MAX_LCORE_PARAMS) {
2804                         printf("exceeded max number of tx params: %hu\n",
2805                                 nb_tx_thread_params);
2806                         return -1;
2807                 }
2808                 tx_thread_params_array[nb_tx_thread_params].lcore_id =
2809                                 (uint8_t)int_fld[FLD_LCORE];
2810                 tx_thread_params_array[nb_tx_thread_params].thread_id =
2811                                 (uint8_t)int_fld[FLD_THREAD];
2812                 ++nb_tx_thread_params;
2813         }
2814         tx_thread_params = tx_thread_params_array;
2815
2816         return 0;
2817 }
2818
2819 #if (APP_CPU_LOAD > 0)
2820 static int
2821 parse_stat_lcore(const char *stat_lcore)
2822 {
2823         char *end = NULL;
2824         unsigned long lcore_id;
2825
2826         lcore_id = strtoul(stat_lcore, &end, 10);
2827         if ((stat_lcore[0] == '\0') || (end == NULL) || (*end != '\0'))
2828                 return -1;
2829
2830         return lcore_id;
2831 }
2832 #endif
2833
2834 static void
2835 parse_eth_dest(const char *optarg)
2836 {
2837         uint16_t portid;
2838         char *port_end;
2839         uint8_t c, *dest, peer_addr[6];
2840
2841         errno = 0;
2842         portid = strtoul(optarg, &port_end, 10);
2843         if (errno != 0 || port_end == optarg || *port_end++ != ',')
2844                 rte_exit(EXIT_FAILURE,
2845                 "Invalid eth-dest: %s", optarg);
2846         if (portid >= RTE_MAX_ETHPORTS)
2847                 rte_exit(EXIT_FAILURE,
2848                 "eth-dest: port %d >= RTE_MAX_ETHPORTS(%d)\n",
2849                 portid, RTE_MAX_ETHPORTS);
2850
2851         if (cmdline_parse_etheraddr(NULL, port_end,
2852                 &peer_addr, sizeof(peer_addr)) < 0)
2853                 rte_exit(EXIT_FAILURE,
2854                 "Invalid ethernet address: %s\n",
2855                 port_end);
2856         dest = (uint8_t *)&dest_eth_addr[portid];
2857         for (c = 0; c < 6; c++)
2858                 dest[c] = peer_addr[c];
2859         *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
2860 }
2861
2862 #define CMD_LINE_OPT_RX_CONFIG "rx"
2863 #define CMD_LINE_OPT_TX_CONFIG "tx"
2864 #define CMD_LINE_OPT_STAT_LCORE "stat-lcore"
2865 #define CMD_LINE_OPT_ETH_DEST "eth-dest"
2866 #define CMD_LINE_OPT_NO_NUMA "no-numa"
2867 #define CMD_LINE_OPT_IPV6 "ipv6"
2868 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
2869 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
2870 #define CMD_LINE_OPT_NO_LTHREADS "no-lthreads"
2871 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
2872
2873 /* Parse the argument given in the command line of the application */
2874 static int
2875 parse_args(int argc, char **argv)
2876 {
2877         int opt, ret;
2878         char **argvopt;
2879         int option_index;
2880         char *prgname = argv[0];
2881         static struct option lgopts[] = {
2882                 {CMD_LINE_OPT_RX_CONFIG, 1, 0, 0},
2883                 {CMD_LINE_OPT_TX_CONFIG, 1, 0, 0},
2884                 {CMD_LINE_OPT_STAT_LCORE, 1, 0, 0},
2885                 {CMD_LINE_OPT_ETH_DEST, 1, 0, 0},
2886                 {CMD_LINE_OPT_NO_NUMA, 0, 0, 0},
2887                 {CMD_LINE_OPT_IPV6, 0, 0, 0},
2888                 {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, 0},
2889                 {CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, 0},
2890                 {CMD_LINE_OPT_NO_LTHREADS, 0, 0, 0},
2891                 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
2892                 {NULL, 0, 0, 0}
2893         };
2894
2895         argvopt = argv;
2896
2897         while ((opt = getopt_long(argc, argvopt, "p:P",
2898                                 lgopts, &option_index)) != EOF) {
2899
2900                 switch (opt) {
2901                 /* portmask */
2902                 case 'p':
2903                         enabled_port_mask = parse_portmask(optarg);
2904                         if (enabled_port_mask == 0) {
2905                                 printf("invalid portmask\n");
2906                                 print_usage(prgname);
2907                                 return -1;
2908                         }
2909                         break;
2910                 case 'P':
2911                         printf("Promiscuous mode selected\n");
2912                         promiscuous_on = 1;
2913                         break;
2914
2915                 /* long options */
2916                 case 0:
2917                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_RX_CONFIG,
2918                                 sizeof(CMD_LINE_OPT_RX_CONFIG))) {
2919                                 ret = parse_rx_config(optarg);
2920                                 if (ret) {
2921                                         printf("invalid rx-config\n");
2922                                         print_usage(prgname);
2923                                         return -1;
2924                                 }
2925                         }
2926
2927                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_TX_CONFIG,
2928                                 sizeof(CMD_LINE_OPT_TX_CONFIG))) {
2929                                 ret = parse_tx_config(optarg);
2930                                 if (ret) {
2931                                         printf("invalid tx-config\n");
2932                                         print_usage(prgname);
2933                                         return -1;
2934                                 }
2935                         }
2936
2937 #if (APP_CPU_LOAD > 0)
2938                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_STAT_LCORE,
2939                                         sizeof(CMD_LINE_OPT_STAT_LCORE))) {
2940                                 cpu_load_lcore_id = parse_stat_lcore(optarg);
2941                         }
2942 #endif
2943
2944                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ETH_DEST,
2945                                 sizeof(CMD_LINE_OPT_ETH_DEST)))
2946                                         parse_eth_dest(optarg);
2947
2948                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_NUMA,
2949                                 sizeof(CMD_LINE_OPT_NO_NUMA))) {
2950                                 printf("numa is disabled\n");
2951                                 numa_on = 0;
2952                         }
2953
2954 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2955                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_IPV6,
2956                                 sizeof(CMD_LINE_OPT_IPV6))) {
2957                                 printf("ipv6 is specified\n");
2958                                 ipv6 = 1;
2959                         }
2960 #endif
2961
2962                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_LTHREADS,
2963                                         sizeof(CMD_LINE_OPT_NO_LTHREADS))) {
2964                                 printf("l-threads model is disabled\n");
2965                                 lthreads_on = 0;
2966                         }
2967
2968                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_PARSE_PTYPE,
2969                                         sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
2970                                 printf("software packet type parsing enabled\n");
2971                                 parse_ptype_on = 1;
2972                         }
2973
2974                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ENABLE_JUMBO,
2975                                 sizeof(CMD_LINE_OPT_ENABLE_JUMBO))) {
2976                                 struct option lenopts = {"max-pkt-len", required_argument, 0,
2977                                                 0};
2978
2979                                 printf("jumbo frame is enabled - disabling simple TX path\n");
2980                                 port_conf.rxmode.offloads |=
2981                                                 DEV_RX_OFFLOAD_JUMBO_FRAME;
2982                                 port_conf.txmode.offloads |=
2983                                                 DEV_TX_OFFLOAD_MULTI_SEGS;
2984
2985                                 /* if no max-pkt-len set, use the default value
2986                                  * RTE_ETHER_MAX_LEN
2987                                  */
2988                                 if (0 == getopt_long(argc, argvopt, "", &lenopts,
2989                                                 &option_index)) {
2990
2991                                         ret = parse_max_pkt_len(optarg);
2992                                         if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)) {
2993                                                 printf("invalid packet length\n");
2994                                                 print_usage(prgname);
2995                                                 return -1;
2996                                         }
2997                                         port_conf.rxmode.max_rx_pkt_len = ret;
2998                                 }
2999                                 printf("set jumbo frame max packet length to %u\n",
3000                                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
3001                         }
3002 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
3003                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_HASH_ENTRY_NUM,
3004                                 sizeof(CMD_LINE_OPT_HASH_ENTRY_NUM))) {
3005                                 ret = parse_hash_entry_number(optarg);
3006                                 if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
3007                                         hash_entry_number = ret;
3008                                 } else {
3009                                         printf("invalid hash entry number\n");
3010                                         print_usage(prgname);
3011                                         return -1;
3012                                 }
3013                         }
3014 #endif
3015                         break;
3016
3017                 default:
3018                         print_usage(prgname);
3019                         return -1;
3020                 }
3021         }
3022
3023         if (optind >= 0)
3024                 argv[optind-1] = prgname;
3025
3026         ret = optind-1;
3027         optind = 1; /* reset getopt lib */
3028         return ret;
3029 }
3030
3031 static void
3032 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
3033 {
3034         char buf[RTE_ETHER_ADDR_FMT_SIZE];
3035
3036         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
3037         printf("%s%s", name, buf);
3038 }
3039
3040 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
3041
3042 static void convert_ipv4_5tuple(struct ipv4_5tuple *key1,
3043                 union ipv4_5tuple_host *key2)
3044 {
3045         key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
3046         key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
3047         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3048         key2->port_src = rte_cpu_to_be_16(key1->port_src);
3049         key2->proto = key1->proto;
3050         key2->pad0 = 0;
3051         key2->pad1 = 0;
3052 }
3053
3054 static void convert_ipv6_5tuple(struct ipv6_5tuple *key1,
3055                 union ipv6_5tuple_host *key2)
3056 {
3057         uint32_t i;
3058
3059         for (i = 0; i < 16; i++) {
3060                 key2->ip_dst[i] = key1->ip_dst[i];
3061                 key2->ip_src[i] = key1->ip_src[i];
3062         }
3063         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3064         key2->port_src = rte_cpu_to_be_16(key1->port_src);
3065         key2->proto = key1->proto;
3066         key2->pad0 = 0;
3067         key2->pad1 = 0;
3068         key2->reserve = 0;
3069 }
3070
3071 #define BYTE_VALUE_MAX 256
3072 #define ALL_32_BITS 0xffffffff
3073 #define BIT_8_TO_15 0x0000ff00
3074 static inline void
3075 populate_ipv4_few_flow_into_table(const struct rte_hash *h)
3076 {
3077         uint32_t i;
3078         int32_t ret;
3079         uint32_t array_len = RTE_DIM(ipv4_l3fwd_route_array);
3080
3081         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3082         for (i = 0; i < array_len; i++) {
3083                 struct ipv4_l3fwd_route  entry;
3084                 union ipv4_5tuple_host newkey;
3085
3086                 entry = ipv4_l3fwd_route_array[i];
3087                 convert_ipv4_5tuple(&entry.key, &newkey);
3088                 ret = rte_hash_add_key(h, (void *)&newkey);
3089                 if (ret < 0) {
3090                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3091                                 " to the l3fwd hash.\n", i);
3092                 }
3093                 ipv4_l3fwd_out_if[ret] = entry.if_out;
3094         }
3095         printf("Hash: Adding 0x%" PRIx32 " keys\n", array_len);
3096 }
3097
3098 #define BIT_16_TO_23 0x00ff0000
3099 static inline void
3100 populate_ipv6_few_flow_into_table(const struct rte_hash *h)
3101 {
3102         uint32_t i;
3103         int32_t ret;
3104         uint32_t array_len = RTE_DIM(ipv6_l3fwd_route_array);
3105
3106         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3107         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3108         for (i = 0; i < array_len; i++) {
3109                 struct ipv6_l3fwd_route entry;
3110                 union ipv6_5tuple_host newkey;
3111
3112                 entry = ipv6_l3fwd_route_array[i];
3113                 convert_ipv6_5tuple(&entry.key, &newkey);
3114                 ret = rte_hash_add_key(h, (void *)&newkey);
3115                 if (ret < 0) {
3116                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3117                                 " to the l3fwd hash.\n", i);
3118                 }
3119                 ipv6_l3fwd_out_if[ret] = entry.if_out;
3120         }
3121         printf("Hash: Adding 0x%" PRIx32 "keys\n", array_len);
3122 }
3123
3124 #define NUMBER_PORT_USED 4
3125 static inline void
3126 populate_ipv4_many_flow_into_table(const struct rte_hash *h,
3127                 unsigned int nr_flow)
3128 {
3129         unsigned i;
3130
3131         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3132
3133         for (i = 0; i < nr_flow; i++) {
3134                 struct ipv4_l3fwd_route entry;
3135                 union ipv4_5tuple_host newkey;
3136                 uint8_t a = (uint8_t)((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3137                 uint8_t b = (uint8_t)(((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3138                                 BYTE_VALUE_MAX);
3139                 uint8_t c = (uint8_t)((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3140                                 BYTE_VALUE_MAX));
3141                 /* Create the ipv4 exact match flow */
3142                 memset(&entry, 0, sizeof(entry));
3143                 switch (i & (NUMBER_PORT_USED - 1)) {
3144                 case 0:
3145                         entry = ipv4_l3fwd_route_array[0];
3146                         entry.key.ip_dst = RTE_IPV4(101, c, b, a);
3147                         break;
3148                 case 1:
3149                         entry = ipv4_l3fwd_route_array[1];
3150                         entry.key.ip_dst = RTE_IPV4(201, c, b, a);
3151                         break;
3152                 case 2:
3153                         entry = ipv4_l3fwd_route_array[2];
3154                         entry.key.ip_dst = RTE_IPV4(111, c, b, a);
3155                         break;
3156                 case 3:
3157                         entry = ipv4_l3fwd_route_array[3];
3158                         entry.key.ip_dst = RTE_IPV4(211, c, b, a);
3159                         break;
3160                 };
3161                 convert_ipv4_5tuple(&entry.key, &newkey);
3162                 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3163
3164                 if (ret < 0)
3165                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3166
3167                 ipv4_l3fwd_out_if[ret] = (uint8_t)entry.if_out;
3168
3169         }
3170         printf("Hash: Adding 0x%x keys\n", nr_flow);
3171 }
3172
3173 static inline void
3174 populate_ipv6_many_flow_into_table(const struct rte_hash *h,
3175                 unsigned int nr_flow)
3176 {
3177         unsigned i;
3178
3179         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3180         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3181         for (i = 0; i < nr_flow; i++) {
3182                 struct ipv6_l3fwd_route entry;
3183                 union ipv6_5tuple_host newkey;
3184
3185                 uint8_t a = (uint8_t) ((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3186                 uint8_t b = (uint8_t) (((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3187                                 BYTE_VALUE_MAX);
3188                 uint8_t c = (uint8_t) ((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3189                                 BYTE_VALUE_MAX));
3190
3191                 /* Create the ipv6 exact match flow */
3192                 memset(&entry, 0, sizeof(entry));
3193                 switch (i & (NUMBER_PORT_USED - 1)) {
3194                 case 0:
3195                         entry = ipv6_l3fwd_route_array[0];
3196                         break;
3197                 case 1:
3198                         entry = ipv6_l3fwd_route_array[1];
3199                         break;
3200                 case 2:
3201                         entry = ipv6_l3fwd_route_array[2];
3202                         break;
3203                 case 3:
3204                         entry = ipv6_l3fwd_route_array[3];
3205                         break;
3206                 };
3207                 entry.key.ip_dst[13] = c;
3208                 entry.key.ip_dst[14] = b;
3209                 entry.key.ip_dst[15] = a;
3210                 convert_ipv6_5tuple(&entry.key, &newkey);
3211                 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3212
3213                 if (ret < 0)
3214                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3215
3216                 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
3217
3218         }
3219         printf("Hash: Adding 0x%x keys\n", nr_flow);
3220 }
3221
3222 static void
3223 setup_hash(int socketid)
3224 {
3225         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
3226                 .name = NULL,
3227                 .entries = L3FWD_HASH_ENTRIES,
3228                 .key_len = sizeof(union ipv4_5tuple_host),
3229                 .hash_func = ipv4_hash_crc,
3230                 .hash_func_init_val = 0,
3231         };
3232
3233         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
3234                 .name = NULL,
3235                 .entries = L3FWD_HASH_ENTRIES,
3236                 .key_len = sizeof(union ipv6_5tuple_host),
3237                 .hash_func = ipv6_hash_crc,
3238                 .hash_func_init_val = 0,
3239         };
3240
3241         char s[64];
3242
3243         /* create ipv4 hash */
3244         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
3245         ipv4_l3fwd_hash_params.name = s;
3246         ipv4_l3fwd_hash_params.socket_id = socketid;
3247         ipv4_l3fwd_lookup_struct[socketid] =
3248                         rte_hash_create(&ipv4_l3fwd_hash_params);
3249         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3250                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3251                                 "socket %d\n", socketid);
3252
3253         /* create ipv6 hash */
3254         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
3255         ipv6_l3fwd_hash_params.name = s;
3256         ipv6_l3fwd_hash_params.socket_id = socketid;
3257         ipv6_l3fwd_lookup_struct[socketid] =
3258                         rte_hash_create(&ipv6_l3fwd_hash_params);
3259         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3260                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3261                                 "socket %d\n", socketid);
3262
3263         if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
3264                 /* For testing hash matching with a large number of flows we
3265                  * generate millions of IP 5-tuples with an incremented dst
3266                  * address to initialize the hash table. */
3267                 if (ipv6 == 0) {
3268                         /* populate the ipv4 hash */
3269                         populate_ipv4_many_flow_into_table(
3270                                 ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
3271                 } else {
3272                         /* populate the ipv6 hash */
3273                         populate_ipv6_many_flow_into_table(
3274                                 ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
3275                 }
3276         } else {
3277                 /* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize
3278                  * the hash table */
3279                 if (ipv6 == 0) {
3280                         /* populate the ipv4 hash */
3281                         populate_ipv4_few_flow_into_table(
3282                                         ipv4_l3fwd_lookup_struct[socketid]);
3283                 } else {
3284                         /* populate the ipv6 hash */
3285                         populate_ipv6_few_flow_into_table(
3286                                         ipv6_l3fwd_lookup_struct[socketid]);
3287                 }
3288         }
3289 }
3290 #endif
3291
3292 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3293 static void
3294 setup_lpm(int socketid)
3295 {
3296         struct rte_lpm6_config config;
3297         struct rte_lpm_config lpm_ipv4_config;
3298         unsigned i;
3299         int ret;
3300         char s[64];
3301
3302         /* create the LPM table */
3303         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
3304         lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
3305         lpm_ipv4_config.number_tbl8s = 256;
3306         lpm_ipv4_config.flags = 0;
3307         ipv4_l3fwd_lookup_struct[socketid] =
3308                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
3309         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3310                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3311                                 " on socket %d\n", socketid);
3312
3313         /* populate the LPM table */
3314         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
3315
3316                 /* skip unused ports */
3317                 if ((1 << ipv4_l3fwd_route_array[i].if_out &
3318                                 enabled_port_mask) == 0)
3319                         continue;
3320
3321                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
3322                         ipv4_l3fwd_route_array[i].ip,
3323                         ipv4_l3fwd_route_array[i].depth,
3324                         ipv4_l3fwd_route_array[i].if_out);
3325
3326                 if (ret < 0) {
3327                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3328                                 "l3fwd LPM table on socket %d\n",
3329                                 i, socketid);
3330                 }
3331
3332                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
3333                         (unsigned)ipv4_l3fwd_route_array[i].ip,
3334                         ipv4_l3fwd_route_array[i].depth,
3335                         ipv4_l3fwd_route_array[i].if_out);
3336         }
3337
3338         /* create the LPM6 table */
3339         snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
3340
3341         config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
3342         config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
3343         config.flags = 0;
3344         ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
3345                                 &config);
3346         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3347                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3348                                 " on socket %d\n", socketid);
3349
3350         /* populate the LPM table */
3351         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
3352
3353                 /* skip unused ports */
3354                 if ((1 << ipv6_l3fwd_route_array[i].if_out &
3355                                 enabled_port_mask) == 0)
3356                         continue;
3357
3358                 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
3359                         ipv6_l3fwd_route_array[i].ip,
3360                         ipv6_l3fwd_route_array[i].depth,
3361                         ipv6_l3fwd_route_array[i].if_out);
3362
3363                 if (ret < 0) {
3364                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3365                                 "l3fwd LPM table on socket %d\n",
3366                                 i, socketid);
3367                 }
3368
3369                 printf("LPM: Adding route %s / %d (%d)\n",
3370                         "IPV6",
3371                         ipv6_l3fwd_route_array[i].depth,
3372                         ipv6_l3fwd_route_array[i].if_out);
3373         }
3374 }
3375 #endif
3376
3377 static int
3378 init_mem(unsigned nb_mbuf)
3379 {
3380         struct lcore_conf *qconf;
3381         int socketid;
3382         unsigned lcore_id;
3383         char s[64];
3384
3385         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3386                 if (rte_lcore_is_enabled(lcore_id) == 0)
3387                         continue;
3388
3389                 if (numa_on)
3390                         socketid = rte_lcore_to_socket_id(lcore_id);
3391                 else
3392                         socketid = 0;
3393
3394                 if (socketid >= NB_SOCKETS) {
3395                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
3396                                 socketid, lcore_id, NB_SOCKETS);
3397                 }
3398                 if (pktmbuf_pool[socketid] == NULL) {
3399                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
3400                         pktmbuf_pool[socketid] =
3401                                 rte_pktmbuf_pool_create(s, nb_mbuf,
3402                                         MEMPOOL_CACHE_SIZE, 0,
3403                                         RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
3404                         if (pktmbuf_pool[socketid] == NULL)
3405                                 rte_exit(EXIT_FAILURE,
3406                                                 "Cannot init mbuf pool on socket %d\n", socketid);
3407                         else
3408                                 printf("Allocated mbuf pool on socket %d\n", socketid);
3409
3410 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3411                         setup_lpm(socketid);
3412 #else
3413                         setup_hash(socketid);
3414 #endif
3415                 }
3416                 qconf = &lcore_conf[lcore_id];
3417                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
3418                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
3419         }
3420         return 0;
3421 }
3422
3423 /* Check the link status of all ports in up to 9s, and print them finally */
3424 static void
3425 check_all_ports_link_status(uint32_t port_mask)
3426 {
3427 #define CHECK_INTERVAL 100 /* 100ms */
3428 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
3429         uint16_t portid;
3430         uint8_t count, all_ports_up, print_flag = 0;
3431         struct rte_eth_link link;
3432         int ret;
3433
3434         printf("\nChecking link status");
3435         fflush(stdout);
3436         for (count = 0; count <= MAX_CHECK_TIME; count++) {
3437                 all_ports_up = 1;
3438                 RTE_ETH_FOREACH_DEV(portid) {
3439                         if ((port_mask & (1 << portid)) == 0)
3440                                 continue;
3441                         memset(&link, 0, sizeof(link));
3442                         ret = rte_eth_link_get_nowait(portid, &link);
3443                         if (ret < 0) {
3444                                 all_ports_up = 0;
3445                                 if (print_flag == 1)
3446                                         printf("Port %u link get failed: %s\n",
3447                                                 portid, rte_strerror(-ret));
3448                                 continue;
3449                         }
3450                         /* print link status if flag set */
3451                         if (print_flag == 1) {
3452                                 if (link.link_status)
3453                                         printf(
3454                                         "Port%d Link Up. Speed %u Mbps - %s\n",
3455                                                 portid, link.link_speed,
3456                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
3457                                         ("full-duplex") : ("half-duplex"));
3458                                 else
3459                                         printf("Port %d Link Down\n", portid);
3460                                 continue;
3461                         }
3462                         /* clear all_ports_up flag if any link down */
3463                         if (link.link_status == ETH_LINK_DOWN) {
3464                                 all_ports_up = 0;
3465                                 break;
3466                         }
3467                 }
3468                 /* after finally printing all link status, get out */
3469                 if (print_flag == 1)
3470                         break;
3471
3472                 if (all_ports_up == 0) {
3473                         printf(".");
3474                         fflush(stdout);
3475                         rte_delay_ms(CHECK_INTERVAL);
3476                 }
3477
3478                 /* set the print_flag if all ports up or timeout */
3479                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
3480                         print_flag = 1;
3481                         printf("done\n");
3482                 }
3483         }
3484 }
3485
3486 int
3487 main(int argc, char **argv)
3488 {
3489         struct rte_eth_dev_info dev_info;
3490         struct rte_eth_txconf *txconf;
3491         int ret;
3492         int i;
3493         unsigned nb_ports;
3494         uint16_t queueid, portid;
3495         unsigned lcore_id;
3496         uint32_t n_tx_queue, nb_lcores;
3497         uint8_t nb_rx_queue, queue, socketid;
3498
3499         /* init EAL */
3500         ret = rte_eal_init(argc, argv);
3501         if (ret < 0)
3502                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
3503         argc -= ret;
3504         argv += ret;
3505
3506         ret = rte_timer_subsystem_init();
3507         if (ret < 0)
3508                 rte_exit(EXIT_FAILURE, "Failed to initialize timer subystem\n");
3509
3510         /* pre-init dst MACs for all ports to 02:00:00:00:00:xx */
3511         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
3512                 dest_eth_addr[portid] = RTE_ETHER_LOCAL_ADMIN_ADDR +
3513                                 ((uint64_t)portid << 40);
3514                 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
3515         }
3516
3517         /* parse application arguments (after the EAL ones) */
3518         ret = parse_args(argc, argv);
3519         if (ret < 0)
3520                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
3521
3522         if (check_lcore_params() < 0)
3523                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
3524
3525         printf("Initializing rx-queues...\n");
3526         ret = init_rx_queues();
3527         if (ret < 0)
3528                 rte_exit(EXIT_FAILURE, "init_rx_queues failed\n");
3529
3530         printf("Initializing tx-threads...\n");
3531         ret = init_tx_threads();
3532         if (ret < 0)
3533                 rte_exit(EXIT_FAILURE, "init_tx_threads failed\n");
3534
3535         printf("Initializing rings...\n");
3536         ret = init_rx_rings();
3537         if (ret < 0)
3538                 rte_exit(EXIT_FAILURE, "init_rx_rings failed\n");
3539
3540         nb_ports = rte_eth_dev_count_avail();
3541
3542         if (check_port_config() < 0)
3543                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
3544
3545         nb_lcores = rte_lcore_count();
3546
3547         /* initialize all ports */
3548         RTE_ETH_FOREACH_DEV(portid) {
3549                 struct rte_eth_conf local_port_conf = port_conf;
3550
3551                 /* skip ports that are not enabled */
3552                 if ((enabled_port_mask & (1 << portid)) == 0) {
3553                         printf("\nSkipping disabled port %d\n", portid);
3554                         continue;
3555                 }
3556
3557                 /* init port */
3558                 printf("Initializing port %d ... ", portid);
3559                 fflush(stdout);
3560
3561                 nb_rx_queue = get_port_n_rx_queues(portid);
3562                 n_tx_queue = nb_lcores;
3563                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
3564                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
3565                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
3566                         nb_rx_queue, (unsigned)n_tx_queue);
3567
3568                 ret = rte_eth_dev_info_get(portid, &dev_info);
3569                 if (ret != 0)
3570                         rte_exit(EXIT_FAILURE,
3571                                 "Error during getting device (port %u) info: %s\n",
3572                                 portid, strerror(-ret));
3573
3574                 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
3575                         local_port_conf.txmode.offloads |=
3576                                 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
3577
3578                 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
3579                         dev_info.flow_type_rss_offloads;
3580                 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
3581                                 port_conf.rx_adv_conf.rss_conf.rss_hf) {
3582                         printf("Port %u modified RSS hash function based on hardware support,"
3583                                 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
3584                                 portid,
3585                                 port_conf.rx_adv_conf.rss_conf.rss_hf,
3586                                 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
3587                 }
3588
3589                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
3590                                         (uint16_t)n_tx_queue, &local_port_conf);
3591                 if (ret < 0)
3592                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
3593                                 ret, portid);
3594
3595                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
3596                                                        &nb_txd);
3597                 if (ret < 0)
3598                         rte_exit(EXIT_FAILURE,
3599                                  "rte_eth_dev_adjust_nb_rx_tx_desc: err=%d, port=%d\n",
3600                                  ret, portid);
3601
3602                 ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
3603                 if (ret < 0)
3604                         rte_exit(EXIT_FAILURE,
3605                                  "rte_eth_macaddr_get: err=%d, port=%d\n",
3606                                  ret, portid);
3607
3608                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
3609                 printf(", ");
3610                 print_ethaddr("Destination:",
3611                         (const struct rte_ether_addr *)&dest_eth_addr[portid]);
3612                 printf(", ");
3613
3614                 /*
3615                  * prepare src MACs for each port.
3616                  */
3617                 rte_ether_addr_copy(&ports_eth_addr[portid],
3618                         (struct rte_ether_addr *)(val_eth + portid) + 1);
3619
3620                 /* init memory */
3621                 ret = init_mem(NB_MBUF);
3622                 if (ret < 0)
3623                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
3624
3625                 /* init one TX queue per couple (lcore,port) */
3626                 queueid = 0;
3627                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3628                         if (rte_lcore_is_enabled(lcore_id) == 0)
3629                                 continue;
3630
3631                         if (numa_on)
3632                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3633                         else
3634                                 socketid = 0;
3635
3636                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
3637                         fflush(stdout);
3638
3639                         txconf = &dev_info.default_txconf;
3640                         txconf->offloads = local_port_conf.txmode.offloads;
3641                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
3642                                                      socketid, txconf);
3643                         if (ret < 0)
3644                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
3645                                         "port=%d\n", ret, portid);
3646
3647                         tx_thread[lcore_id].tx_queue_id[portid] = queueid;
3648                         queueid++;
3649                 }
3650                 printf("\n");
3651         }
3652
3653         for (i = 0; i < n_rx_thread; i++) {
3654                 lcore_id = rx_thread[i].conf.lcore_id;
3655
3656                 if (rte_lcore_is_enabled(lcore_id) == 0) {
3657                         rte_exit(EXIT_FAILURE,
3658                                         "Cannot start Rx thread on lcore %u: lcore disabled\n",
3659                                         lcore_id
3660                                 );
3661                 }
3662
3663                 printf("\nInitializing rx queues for Rx thread %d on lcore %u ... ",
3664                                 i, lcore_id);
3665                 fflush(stdout);
3666
3667                 /* init RX queues */
3668                 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3669                         struct rte_eth_rxconf rxq_conf;
3670
3671                         portid = rx_thread[i].rx_queue_list[queue].port_id;
3672                         queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3673
3674                         if (numa_on)
3675                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3676                         else
3677                                 socketid = 0;
3678
3679                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
3680                         fflush(stdout);
3681
3682                         ret = rte_eth_dev_info_get(portid, &dev_info);
3683                         if (ret != 0)
3684                                 rte_exit(EXIT_FAILURE,
3685                                         "Error during getting device (port %u) info: %s\n",
3686                                         portid, strerror(-ret));
3687
3688                         rxq_conf = dev_info.default_rxconf;
3689                         rxq_conf.offloads = port_conf.rxmode.offloads;
3690                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
3691                                         socketid,
3692                                         &rxq_conf,
3693                                         pktmbuf_pool[socketid]);
3694                         if (ret < 0)
3695                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, "
3696                                                 "port=%d\n", ret, portid);
3697                 }
3698         }
3699
3700         printf("\n");
3701
3702         /* start ports */
3703         RTE_ETH_FOREACH_DEV(portid) {
3704                 if ((enabled_port_mask & (1 << portid)) == 0)
3705                         continue;
3706
3707                 /* Start device */
3708                 ret = rte_eth_dev_start(portid);
3709                 if (ret < 0)
3710                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
3711                                 ret, portid);
3712
3713                 /*
3714                  * If enabled, put device in promiscuous mode.
3715                  * This allows IO forwarding mode to forward packets
3716                  * to itself through 2 cross-connected  ports of the
3717                  * target machine.
3718                  */
3719                 if (promiscuous_on) {
3720                         ret = rte_eth_promiscuous_enable(portid);
3721                         if (ret != 0)
3722                                 rte_exit(EXIT_FAILURE,
3723                                         "rte_eth_promiscuous_enable: err=%s, port=%u\n",
3724                                         rte_strerror(-ret), portid);
3725                 }
3726         }
3727
3728         for (i = 0; i < n_rx_thread; i++) {
3729                 lcore_id = rx_thread[i].conf.lcore_id;
3730                 if (rte_lcore_is_enabled(lcore_id) == 0)
3731                         continue;
3732
3733                 /* check if hw packet type is supported */
3734                 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3735                         portid = rx_thread[i].rx_queue_list[queue].port_id;
3736                         queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3737
3738                         if (parse_ptype_on) {
3739                                 if (!rte_eth_add_rx_callback(portid, queueid,
3740                                                 cb_parse_ptype, NULL))
3741                                         rte_exit(EXIT_FAILURE,
3742                                                 "Failed to add rx callback: "
3743                                                 "port=%d\n", portid);
3744                         } else if (!check_ptype(portid))
3745                                 rte_exit(EXIT_FAILURE,
3746                                         "Port %d cannot parse packet type.\n\n"
3747                                         "Please add --parse-ptype to use sw "
3748                                         "packet type analyzer.\n\n",
3749                                         portid);
3750                 }
3751         }
3752
3753         check_all_ports_link_status(enabled_port_mask);
3754
3755         if (lthreads_on) {
3756                 printf("Starting L-Threading Model\n");
3757
3758 #if (APP_CPU_LOAD > 0)
3759                 if (cpu_load_lcore_id > 0)
3760                         /* Use one lcore for cpu load collector */
3761                         nb_lcores--;
3762 #endif
3763
3764                 lthread_num_schedulers_set(nb_lcores);
3765                 rte_eal_mp_remote_launch(sched_spawner, NULL, SKIP_MASTER);
3766                 lthread_master_spawner(NULL);
3767
3768         } else {
3769                 printf("Starting P-Threading Model\n");
3770                 /* launch per-lcore init on every lcore */
3771                 rte_eal_mp_remote_launch(pthread_run, NULL, CALL_MASTER);
3772                 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
3773                         if (rte_eal_wait_lcore(lcore_id) < 0)
3774                                 return -1;
3775                 }
3776         }
3777
3778         return 0;
3779 }