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
  44 #include <cmdline_parse.h>
  45 #include <cmdline_parse_etheraddr.h>
  46
  47 #include <lthread_api.h>
  48
  49 #define APP_LOOKUP_EXACT_MATCH          0
  50 #define APP_LOOKUP_LPM                  1
  51 #define DO_RFC_1812_CHECKS
  52
  53 /* Enable cpu-load stats 0-off, 1-on */
  54 #define APP_CPU_LOAD                 1
  55
  56 #ifndef APP_LOOKUP_METHOD
  57 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  58 #endif
  59
  60 #ifndef __GLIBC__ /* sched_getcpu() is glibc specific */
  61 #define sched_getcpu() rte_lcore_id()
  62 #endif
  63
  64 static int
  65 check_ptype(int portid)
  66 {
  67         int i, ret;
  68         int ipv4 = 0, ipv6 = 0;
  69
  70         ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK, NULL,
  71                         0);
  72         if (ret <= 0)
  73                 return 0;
  74
  75         uint32_t ptypes[ret];
  76
  77         ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK,
  78                         ptypes, ret);
  79         for (i = 0; i < ret; ++i) {
  80                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
  81                         ipv4 = 1;
  82                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
  83                         ipv6 = 1;
  84         }
  85
  86         if (ipv4 && ipv6)
  87                 return 1;
  88
  89         return 0;
  90 }
  91
  92 static inline void
  93 parse_ptype(struct rte_mbuf *m)
  94 {
  95         struct rte_ether_hdr *eth_hdr;
  96         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
  97         uint16_t ether_type;
  98
  99         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 100         ether_type = eth_hdr->ether_type;
 101         if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPv4))
 102                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 103         else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPv6))
 104                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 105
 106         m->packet_type = packet_type;
 107 }
 108
 109 static uint16_t
 110 cb_parse_ptype(__rte_unused uint16_t port, __rte_unused uint16_t queue,
 111                 struct rte_mbuf *pkts[], uint16_t nb_pkts,
 112                 __rte_unused uint16_t max_pkts, __rte_unused void *user_param)
 113 {
 114         unsigned int i;
 115
 116         for (i = 0; i < nb_pkts; i++)
 117                 parse_ptype(pkts[i]);
 118
 119         return nb_pkts;
 120 }
 121
 122 /*
 123  *  When set to zero, simple forwaring path is eanbled.
 124  *  When set to one, optimized forwarding path is enabled.
 125  *  Note that LPM optimisation path uses SSE4.1 instructions.
 126  */
 127 #define ENABLE_MULTI_BUFFER_OPTIMIZE    1
 128
 129 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 130 #include <rte_hash.h>
 131 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 132 #include <rte_lpm.h>
 133 #include <rte_lpm6.h>
 134 #else
 135 #error "APP_LOOKUP_METHOD set to incorrect value"
 136 #endif
 137
 138 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
 139
 140 #define MAX_JUMBO_PKT_LEN  9600
 141
 142 #define IPV6_ADDR_LEN 16
 143
 144 #define MEMPOOL_CACHE_SIZE 256
 145
 146 /*
 147  * This expression is used to calculate the number of mbufs needed depending on
 148  * user input, taking into account memory for rx and tx hardware rings, cache
 149  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 150  * NB_MBUF never goes below a minimum value of 8192
 151  */
 152
 153 #define NB_MBUF RTE_MAX(\
 154                 (nb_ports*nb_rx_queue*nb_rxd +      \
 155                 nb_ports*nb_lcores*MAX_PKT_BURST +  \
 156                 nb_ports*n_tx_queue*nb_txd +        \
 157                 nb_lcores*MEMPOOL_CACHE_SIZE),      \
 158                 (unsigned)8192)
 159
 160 #define MAX_PKT_BURST     32
 161 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 162
 163 /*
 164  * Try to avoid TX buffering if we have at least MAX_TX_BURST packets to send.
 165  */
 166 #define MAX_TX_BURST  (MAX_PKT_BURST / 2)
 167 #define BURST_SIZE    MAX_TX_BURST
 168
 169 #define NB_SOCKETS 8
 170
 171 /* Configure how many packets ahead to prefetch, when reading packets */
 172 #define PREFETCH_OFFSET 3
 173
 174 /* Used to mark destination port as 'invalid'. */
 175 #define BAD_PORT        ((uint16_t)-1)
 176
 177 #define FWDSTEP 4
 178
 179 /*
 180  * Configurable number of RX/TX ring descriptors
 181  */
 182 #define RTE_TEST_RX_DESC_DEFAULT 1024
 183 #define RTE_TEST_TX_DESC_DEFAULT 1024
 184 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 185 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 186
 187 /* ethernet addresses of ports */
 188 static uint64_t dest_eth_addr[RTE_MAX_ETHPORTS];
 189 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 190
 191 static xmm_t val_eth[RTE_MAX_ETHPORTS];
 192
 193 /* replace first 12B of the ethernet header. */
 194 #define MASK_ETH 0x3f
 195
 196 /* mask of enabled ports */
 197 static uint32_t enabled_port_mask;
 198 static int promiscuous_on; /**< Set in promiscuous mode off by default. */
 199 static int numa_on = 1;    /**< NUMA is enabled by default. */
 200 static int parse_ptype_on;
 201
 202 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 203 static int ipv6;           /**< ipv6 is false by default. */
 204 #endif
 205
 206 #if (APP_CPU_LOAD == 1)
 207
 208 #define MAX_CPU RTE_MAX_LCORE
 209 #define CPU_LOAD_TIMEOUT_US (5 * 1000 * 1000)  /**< Timeout for collecting 5s */
 210
 211 #define CPU_PROCESS     0
 212 #define CPU_POLL        1
 213 #define MAX_CPU_COUNTER 2
 214
 215 struct cpu_load {
 216         uint16_t       n_cpu;
 217         uint64_t       counter;
 218         uint64_t       hits[MAX_CPU_COUNTER][MAX_CPU];
 219 } __rte_cache_aligned;
 220
 221 static struct cpu_load cpu_load;
 222 static int cpu_load_lcore_id = -1;
 223
 224 #define SET_CPU_BUSY(thread, counter) \
 225                 thread->conf.busy[counter] = 1
 226
 227 #define SET_CPU_IDLE(thread, counter) \
 228                 thread->conf.busy[counter] = 0
 229
 230 #define IS_CPU_BUSY(thread, counter) \
 231                 (thread->conf.busy[counter] > 0)
 232
 233 #else
 234
 235 #define SET_CPU_BUSY(thread, counter)
 236 #define SET_CPU_IDLE(thread, counter)
 237 #define IS_CPU_BUSY(thread, counter) 0
 238
 239 #endif
 240
 241 struct mbuf_table {
 242         uint16_t len;
 243         struct rte_mbuf *m_table[MAX_PKT_BURST];
 244 };
 245
 246 struct lcore_rx_queue {
 247         uint16_t port_id;
 248         uint8_t queue_id;
 249 } __rte_cache_aligned;
 250
 251 #define MAX_RX_QUEUE_PER_LCORE 16
 252 #define MAX_TX_QUEUE_PER_PORT  RTE_MAX_ETHPORTS
 253 #define MAX_RX_QUEUE_PER_PORT  128
 254
 255 #define MAX_LCORE_PARAMS       1024
 256 struct rx_thread_params {
 257         uint16_t port_id;
 258         uint8_t queue_id;
 259         uint8_t lcore_id;
 260         uint8_t thread_id;
 261 } __rte_cache_aligned;
 262
 263 static struct rx_thread_params rx_thread_params_array[MAX_LCORE_PARAMS];
 264 static struct rx_thread_params rx_thread_params_array_default[] = {
 265         {0, 0, 2, 0},
 266         {0, 1, 2, 1},
 267         {0, 2, 2, 2},
 268         {1, 0, 2, 3},
 269         {1, 1, 2, 4},
 270         {1, 2, 2, 5},
 271         {2, 0, 2, 6},
 272         {3, 0, 3, 7},
 273         {3, 1, 3, 8},
 274 };
 275
 276 static struct rx_thread_params *rx_thread_params =
 277                 rx_thread_params_array_default;
 278 static uint16_t nb_rx_thread_params = RTE_DIM(rx_thread_params_array_default);
 279
 280 struct tx_thread_params {
 281         uint8_t lcore_id;
 282         uint8_t thread_id;
 283 } __rte_cache_aligned;
 284
 285 static struct tx_thread_params tx_thread_params_array[MAX_LCORE_PARAMS];
 286 static struct tx_thread_params tx_thread_params_array_default[] = {
 287         {4, 0},
 288         {5, 1},
 289         {6, 2},
 290         {7, 3},
 291         {8, 4},
 292         {9, 5},
 293         {10, 6},
 294         {11, 7},
 295         {12, 8},
 296 };
 297
 298 static struct tx_thread_params *tx_thread_params =
 299                 tx_thread_params_array_default;
 300 static uint16_t nb_tx_thread_params = RTE_DIM(tx_thread_params_array_default);
 301
 302 static struct rte_eth_conf port_conf = {
 303         .rxmode = {
 304                 .mq_mode = ETH_MQ_RX_RSS,
 305                 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
 306                 .split_hdr_size = 0,
 307                 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
 308         },
 309         .rx_adv_conf = {
 310                 .rss_conf = {
 311                         .rss_key = NULL,
 312                         .rss_hf = ETH_RSS_TCP,
 313                 },
 314         },
 315         .txmode = {
 316                 .mq_mode = ETH_MQ_TX_NONE,
 317         },
 318 };
 319
 320 static struct rte_mempool *pktmbuf_pool[NB_SOCKETS];
 321
 322 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 323
 324 #include <rte_hash_crc.h>
 325 #define DEFAULT_HASH_FUNC       rte_hash_crc
 326
 327 struct ipv4_5tuple {
 328         uint32_t ip_dst;
 329         uint32_t ip_src;
 330         uint16_t port_dst;
 331         uint16_t port_src;
 332         uint8_t  proto;
 333 } __attribute__((__packed__));
 334
 335 union ipv4_5tuple_host {
 336         struct {
 337                 uint8_t  pad0;
 338                 uint8_t  proto;
 339                 uint16_t pad1;
 340                 uint32_t ip_src;
 341                 uint32_t ip_dst;
 342                 uint16_t port_src;
 343                 uint16_t port_dst;
 344         };
 345         __m128i xmm;
 346 };
 347
 348 #define XMM_NUM_IN_IPV6_5TUPLE 3
 349
 350 struct ipv6_5tuple {
 351         uint8_t  ip_dst[IPV6_ADDR_LEN];
 352         uint8_t  ip_src[IPV6_ADDR_LEN];
 353         uint16_t port_dst;
 354         uint16_t port_src;
 355         uint8_t  proto;
 356 } __attribute__((__packed__));
 357
 358 union ipv6_5tuple_host {
 359         struct {
 360                 uint16_t pad0;
 361                 uint8_t  proto;
 362                 uint8_t  pad1;
 363                 uint8_t  ip_src[IPV6_ADDR_LEN];
 364                 uint8_t  ip_dst[IPV6_ADDR_LEN];
 365                 uint16_t port_src;
 366                 uint16_t port_dst;
 367                 uint64_t reserve;
 368         };
 369         __m128i xmm[XMM_NUM_IN_IPV6_5TUPLE];
 370 };
 371
 372 struct ipv4_l3fwd_route {
 373         struct ipv4_5tuple key;
 374         uint8_t if_out;
 375 };
 376
 377 struct ipv6_l3fwd_route {
 378         struct ipv6_5tuple key;
 379         uint8_t if_out;
 380 };
 381
 382 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 383         {{IPv4(101, 0, 0, 0), IPv4(100, 10, 0, 1),  101, 11, IPPROTO_TCP}, 0},
 384         {{IPv4(201, 0, 0, 0), IPv4(200, 20, 0, 1),  102, 12, IPPROTO_TCP}, 1},
 385         {{IPv4(111, 0, 0, 0), IPv4(100, 30, 0, 1),  101, 11, IPPROTO_TCP}, 2},
 386         {{IPv4(211, 0, 0, 0), IPv4(200, 40, 0, 1),  102, 12, IPPROTO_TCP}, 3},
 387 };
 388
 389 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 390         {{
 391         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 392         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 393                         0x05},
 394         101, 11, IPPROTO_TCP}, 0},
 395
 396         {{
 397         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 398         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 399                         0x05},
 400         102, 12, IPPROTO_TCP}, 1},
 401
 402         {{
 403         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 404         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 405                         0x05},
 406         101, 11, IPPROTO_TCP}, 2},
 407
 408         {{
 409         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 410         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 411                         0x05},
 412         102, 12, IPPROTO_TCP}, 3},
 413 };
 414
 415 typedef struct rte_hash lookup_struct_t;
 416 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 417 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 418
 419 #ifdef RTE_ARCH_X86_64
 420 /* default to 4 million hash entries (approx) */
 421 #define L3FWD_HASH_ENTRIES (1024*1024*4)
 422 #else
 423 /* 32-bit has less address-space for hugepage memory, limit to 1M entries */
 424 #define L3FWD_HASH_ENTRIES (1024*1024*1)
 425 #endif
 426 #define HASH_ENTRY_NUMBER_DEFAULT 4
 427
 428 static uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
 429
 430 static inline uint32_t
 431 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
 432                 uint32_t init_val)
 433 {
 434         const union ipv4_5tuple_host *k;
 435         uint32_t t;
 436         const uint32_t *p;
 437
 438         k = data;
 439         t = k->proto;
 440         p = (const uint32_t *)&k->port_src;
 441
 442         init_val = rte_hash_crc_4byte(t, init_val);
 443         init_val = rte_hash_crc_4byte(k->ip_src, init_val);
 444         init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
 445         init_val = rte_hash_crc_4byte(*p, init_val);
 446         return init_val;
 447 }
 448
 449 static inline uint32_t
 450 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len,
 451                 uint32_t init_val)
 452 {
 453         const union ipv6_5tuple_host *k;
 454         uint32_t t;
 455         const uint32_t *p;
 456         const uint32_t *ip_src0, *ip_src1, *ip_src2, *ip_src3;
 457         const uint32_t *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
 458
 459         k = data;
 460         t = k->proto;
 461         p = (const uint32_t *)&k->port_src;
 462
 463         ip_src0 = (const uint32_t *) k->ip_src;
 464         ip_src1 = (const uint32_t *)(k->ip_src + 4);
 465         ip_src2 = (const uint32_t *)(k->ip_src + 8);
 466         ip_src3 = (const uint32_t *)(k->ip_src + 12);
 467         ip_dst0 = (const uint32_t *) k->ip_dst;
 468         ip_dst1 = (const uint32_t *)(k->ip_dst + 4);
 469         ip_dst2 = (const uint32_t *)(k->ip_dst + 8);
 470         ip_dst3 = (const uint32_t *)(k->ip_dst + 12);
 471         init_val = rte_hash_crc_4byte(t, init_val);
 472         init_val = rte_hash_crc_4byte(*ip_src0, init_val);
 473         init_val = rte_hash_crc_4byte(*ip_src1, init_val);
 474         init_val = rte_hash_crc_4byte(*ip_src2, init_val);
 475         init_val = rte_hash_crc_4byte(*ip_src3, init_val);
 476         init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
 477         init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
 478         init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
 479         init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
 480         init_val = rte_hash_crc_4byte(*p, init_val);
 481         return init_val;
 482 }
 483
 484 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
 485 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
 486
 487 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 488 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 489
 490 #endif
 491
 492 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 493 struct ipv4_l3fwd_route {
 494         uint32_t ip;
 495         uint8_t  depth;
 496         uint8_t  if_out;
 497 };
 498
 499 struct ipv6_l3fwd_route {
 500         uint8_t ip[16];
 501         uint8_t depth;
 502         uint8_t if_out;
 503 };
 504
 505 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 506         {IPv4(1, 1, 1, 0), 24, 0},
 507         {IPv4(2, 1, 1, 0), 24, 1},
 508         {IPv4(3, 1, 1, 0), 24, 2},
 509         {IPv4(4, 1, 1, 0), 24, 3},
 510         {IPv4(5, 1, 1, 0), 24, 4},
 511         {IPv4(6, 1, 1, 0), 24, 5},
 512         {IPv4(7, 1, 1, 0), 24, 6},
 513         {IPv4(8, 1, 1, 0), 24, 7},
 514 };
 515
 516 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 517         {{1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 0},
 518         {{2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 1},
 519         {{3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 2},
 520         {{4, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 3},
 521         {{5, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 4},
 522         {{6, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 5},
 523         {{7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 6},
 524         {{8, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 7},
 525 };
 526
 527 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
 528 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
 529
 530 #define IPV4_L3FWD_LPM_MAX_RULES         1024
 531 #define IPV6_L3FWD_LPM_MAX_RULES         1024
 532 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
 533
 534 typedef struct rte_lpm lookup_struct_t;
 535 typedef struct rte_lpm6 lookup6_struct_t;
 536 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 537 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 538 #endif
 539
 540 struct lcore_conf {
 541         lookup_struct_t *ipv4_lookup_struct;
 542 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 543         lookup6_struct_t *ipv6_lookup_struct;
 544 #else
 545         lookup_struct_t *ipv6_lookup_struct;
 546 #endif
 547         void *data;
 548 } __rte_cache_aligned;
 549
 550 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 551 RTE_DEFINE_PER_LCORE(struct lcore_conf *, lcore_conf);
 552
 553 #define MAX_RX_QUEUE_PER_THREAD 16
 554 #define MAX_TX_PORT_PER_THREAD  RTE_MAX_ETHPORTS
 555 #define MAX_TX_QUEUE_PER_PORT   RTE_MAX_ETHPORTS
 556 #define MAX_RX_QUEUE_PER_PORT   128
 557
 558 #define MAX_RX_THREAD 1024
 559 #define MAX_TX_THREAD 1024
 560 #define MAX_THREAD    (MAX_RX_THREAD + MAX_TX_THREAD)
 561
 562 /**
 563  * Producers and consumers threads configuration
 564  */
 565 static int lthreads_on = 1; /**< Use lthreads for processing*/
 566
 567 rte_atomic16_t rx_counter;  /**< Number of spawned rx threads */
 568 rte_atomic16_t tx_counter;  /**< Number of spawned tx threads */
 569
 570 struct thread_conf {
 571         uint16_t lcore_id;      /**< Initial lcore for rx thread */
 572         uint16_t cpu_id;        /**< Cpu id for cpu load stats counter */
 573         uint16_t thread_id;     /**< Thread ID */
 574
 575 #if (APP_CPU_LOAD > 0)
 576         int busy[MAX_CPU_COUNTER];
 577 #endif
 578 };
 579
 580 struct thread_rx_conf {
 581         struct thread_conf conf;
 582
 583         uint16_t n_rx_queue;
 584         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 585
 586         uint16_t n_ring;        /**< Number of output rings */
 587         struct rte_ring *ring[RTE_MAX_LCORE];
 588         struct lthread_cond *ready[RTE_MAX_LCORE];
 589
 590 #if (APP_CPU_LOAD > 0)
 591         int busy[MAX_CPU_COUNTER];
 592 #endif
 593 } __rte_cache_aligned;
 594
 595 uint16_t n_rx_thread;
 596 struct thread_rx_conf rx_thread[MAX_RX_THREAD];
 597
 598 struct thread_tx_conf {
 599         struct thread_conf conf;
 600
 601         uint16_t tx_queue_id[RTE_MAX_LCORE];
 602         struct mbuf_table tx_mbufs[RTE_MAX_LCORE];
 603
 604         struct rte_ring *ring;
 605         struct lthread_cond **ready;
 606
 607 } __rte_cache_aligned;
 608
 609 uint16_t n_tx_thread;
 610 struct thread_tx_conf tx_thread[MAX_TX_THREAD];
 611
 612 /* Send burst of packets on an output interface */
 613 static inline int
 614 send_burst(struct thread_tx_conf *qconf, uint16_t n, uint16_t port)
 615 {
 616         struct rte_mbuf **m_table;
 617         int ret;
 618         uint16_t queueid;
 619
 620         queueid = qconf->tx_queue_id[port];
 621         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 622
 623         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 624         if (unlikely(ret < n)) {
 625                 do {
 626                         rte_pktmbuf_free(m_table[ret]);
 627                 } while (++ret < n);
 628         }
 629
 630         return 0;
 631 }
 632
 633 /* Enqueue a single packet, and send burst if queue is filled */
 634 static inline int
 635 send_single_packet(struct rte_mbuf *m, uint16_t port)
 636 {
 637         uint16_t len;
 638         struct thread_tx_conf *qconf;
 639
 640         if (lthreads_on)
 641                 qconf = (struct thread_tx_conf *)lthread_get_data();
 642         else
 643                 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
 644
 645         len = qconf->tx_mbufs[port].len;
 646         qconf->tx_mbufs[port].m_table[len] = m;
 647         len++;
 648
 649         /* enough pkts to be sent */
 650         if (unlikely(len == MAX_PKT_BURST)) {
 651                 send_burst(qconf, MAX_PKT_BURST, port);
 652                 len = 0;
 653         }
 654
 655         qconf->tx_mbufs[port].len = len;
 656         return 0;
 657 }
 658
 659 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
 660         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
 661 static __rte_always_inline void
 662 send_packetsx4(uint16_t port,
 663         struct rte_mbuf *m[], uint32_t num)
 664 {
 665         uint32_t len, j, n;
 666         struct thread_tx_conf *qconf;
 667
 668         if (lthreads_on)
 669                 qconf = (struct thread_tx_conf *)lthread_get_data();
 670         else
 671                 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
 672
 673         len = qconf->tx_mbufs[port].len;
 674
 675         /*
 676          * If TX buffer for that queue is empty, and we have enough packets,
 677          * then send them straightway.
 678          */
 679         if (num >= MAX_TX_BURST && len == 0) {
 680                 n = rte_eth_tx_burst(port, qconf->tx_queue_id[port], m, num);
 681                 if (unlikely(n < num)) {
 682                         do {
 683                                 rte_pktmbuf_free(m[n]);
 684                         } while (++n < num);
 685                 }
 686                 return;
 687         }
 688
 689         /*
 690          * Put packets into TX buffer for that queue.
 691          */
 692
 693         n = len + num;
 694         n = (n > MAX_PKT_BURST) ? MAX_PKT_BURST - len : num;
 695
 696         j = 0;
 697         switch (n % FWDSTEP) {
 698         while (j < n) {
 699         case 0:
 700                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 701                 j++;
 702                 /* fall-through */
 703         case 3:
 704                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 705                 j++;
 706                 /* fall-through */
 707         case 2:
 708                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 709                 j++;
 710                 /* fall-through */
 711         case 1:
 712                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 713                 j++;
 714         }
 715         }
 716
 717         len += n;
 718
 719         /* enough pkts to be sent */
 720         if (unlikely(len == MAX_PKT_BURST)) {
 721
 722                 send_burst(qconf, MAX_PKT_BURST, port);
 723
 724                 /* copy rest of the packets into the TX buffer. */
 725                 len = num - n;
 726                 j = 0;
 727                 switch (len % FWDSTEP) {
 728                 while (j < len) {
 729                 case 0:
 730                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 731                         j++;
 732                         /* fall-through */
 733                 case 3:
 734                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 735                         j++;
 736                         /* fall-through */
 737                 case 2:
 738                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 739                         j++;
 740                         /* fall-through */
 741                 case 1:
 742                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 743                         j++;
 744                 }
 745                 }
 746         }
 747
 748         qconf->tx_mbufs[port].len = len;
 749 }
 750 #endif /* APP_LOOKUP_LPM */
 751
 752 #ifdef DO_RFC_1812_CHECKS
 753 static inline int
 754 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 755 {
 756         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 757         /*
 758          * 1. The packet length reported by the Link Layer must be large
 759          * enough to hold the minimum length legal IP datagram (20 bytes).
 760          */
 761         if (link_len < sizeof(struct ipv4_hdr))
 762                 return -1;
 763
 764         /* 2. The IP checksum must be correct. */
 765         /* this is checked in H/W */
 766
 767         /*
 768          * 3. The IP version number must be 4. If the version number is not 4
 769          * then the packet may be another version of IP, such as IPng or
 770          * ST-II.
 771          */
 772         if (((pkt->version_ihl) >> 4) != 4)
 773                 return -3;
 774         /*
 775          * 4. The IP header length field must be large enough to hold the
 776          * minimum length legal IP datagram (20 bytes = 5 words).
 777          */
 778         if ((pkt->version_ihl & 0xf) < 5)
 779                 return -4;
 780
 781         /*
 782          * 5. The IP total length field must be large enough to hold the IP
 783          * datagram header, whose length is specified in the IP header length
 784          * field.
 785          */
 786         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 787                 return -5;
 788
 789         return 0;
 790 }
 791 #endif
 792
 793 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 794
 795 static __m128i mask0;
 796 static __m128i mask1;
 797 static __m128i mask2;
 798 static inline uint16_t
 799 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid,
 800                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 801 {
 802         int ret = 0;
 803         union ipv4_5tuple_host key;
 804
 805         ipv4_hdr = (uint8_t *)ipv4_hdr + offsetof(struct ipv4_hdr, time_to_live);
 806         __m128i data = _mm_loadu_si128((__m128i *)(ipv4_hdr));
 807         /* Get 5 tuple: dst port, src port, dst IP address, src IP address and
 808            protocol */
 809         key.xmm = _mm_and_si128(data, mask0);
 810         /* Find destination port */
 811         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 812         return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
 813 }
 814
 815 static inline uint16_t
 816 get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid,
 817                 lookup_struct_t *ipv6_l3fwd_lookup_struct)
 818 {
 819         int ret = 0;
 820         union ipv6_5tuple_host key;
 821
 822         ipv6_hdr = (uint8_t *)ipv6_hdr + offsetof(struct ipv6_hdr, payload_len);
 823         __m128i data0 = _mm_loadu_si128((__m128i *)(ipv6_hdr));
 824         __m128i data1 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
 825                         sizeof(__m128i)));
 826         __m128i data2 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
 827                         sizeof(__m128i) + sizeof(__m128i)));
 828         /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
 829         key.xmm[0] = _mm_and_si128(data0, mask1);
 830         /* Get part of 5 tuple: dst IP address lower 96 bits and src IP address
 831            higher 32 bits */
 832         key.xmm[1] = data1;
 833         /* Get part of 5 tuple: dst port and src port and dst IP address higher
 834            32 bits */
 835         key.xmm[2] = _mm_and_si128(data2, mask2);
 836
 837         /* Find destination port */
 838         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 839         return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
 840 }
 841 #endif
 842
 843 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 844
 845 static inline uint16_t
 846 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid,
 847                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 848 {
 849         uint32_t next_hop;
 850
 851         return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 852                 rte_be_to_cpu_32(((struct ipv4_hdr *)ipv4_hdr)->dst_addr),
 853                 &next_hop) == 0) ? next_hop : portid);
 854 }
 855
 856 static inline uint16_t
 857 get_ipv6_dst_port(void *ipv6_hdr,  uint16_t portid,
 858                 lookup6_struct_t *ipv6_l3fwd_lookup_struct)
 859 {
 860         uint32_t next_hop;
 861
 862         return ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
 863                         ((struct ipv6_hdr *)ipv6_hdr)->dst_addr, &next_hop) == 0) ?
 864                         next_hop : portid);
 865 }
 866 #endif
 867
 868 static inline void l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid)
 869                 __attribute__((unused));
 870
 871 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) && \
 872         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
 873
 874 #define MASK_ALL_PKTS   0xff
 875 #define EXCLUDE_1ST_PKT 0xfe
 876 #define EXCLUDE_2ND_PKT 0xfd
 877 #define EXCLUDE_3RD_PKT 0xfb
 878 #define EXCLUDE_4TH_PKT 0xf7
 879 #define EXCLUDE_5TH_PKT 0xef
 880 #define EXCLUDE_6TH_PKT 0xdf
 881 #define EXCLUDE_7TH_PKT 0xbf
 882 #define EXCLUDE_8TH_PKT 0x7f
 883
 884 static inline void
 885 simple_ipv4_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid)
 886 {
 887         struct rte_ether_hdr *eth_hdr[8];
 888         struct ipv4_hdr *ipv4_hdr[8];
 889         uint16_t dst_port[8];
 890         int32_t ret[8];
 891         union ipv4_5tuple_host key[8];
 892         __m128i data[8];
 893
 894         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct rte_ether_hdr *);
 895         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct rte_ether_hdr *);
 896         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct rte_ether_hdr *);
 897         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct rte_ether_hdr *);
 898         eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct rte_ether_hdr *);
 899         eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct rte_ether_hdr *);
 900         eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct rte_ether_hdr *);
 901         eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct rte_ether_hdr *);
 902
 903         /* Handle IPv4 headers.*/
 904         ipv4_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv4_hdr *,
 905                         sizeof(struct rte_ether_hdr));
 906         ipv4_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv4_hdr *,
 907                         sizeof(struct rte_ether_hdr));
 908         ipv4_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv4_hdr *,
 909                         sizeof(struct rte_ether_hdr));
 910         ipv4_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv4_hdr *,
 911                         sizeof(struct rte_ether_hdr));
 912         ipv4_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv4_hdr *,
 913                         sizeof(struct rte_ether_hdr));
 914         ipv4_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv4_hdr *,
 915                         sizeof(struct rte_ether_hdr));
 916         ipv4_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv4_hdr *,
 917                         sizeof(struct rte_ether_hdr));
 918         ipv4_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv4_hdr *,
 919                         sizeof(struct rte_ether_hdr));
 920
 921 #ifdef DO_RFC_1812_CHECKS
 922         /* Check to make sure the packet is valid (RFC1812) */
 923         uint8_t valid_mask = MASK_ALL_PKTS;
 924
 925         if (is_valid_ipv4_pkt(ipv4_hdr[0], m[0]->pkt_len) < 0) {
 926                 rte_pktmbuf_free(m[0]);
 927                 valid_mask &= EXCLUDE_1ST_PKT;
 928         }
 929         if (is_valid_ipv4_pkt(ipv4_hdr[1], m[1]->pkt_len) < 0) {
 930                 rte_pktmbuf_free(m[1]);
 931                 valid_mask &= EXCLUDE_2ND_PKT;
 932         }
 933         if (is_valid_ipv4_pkt(ipv4_hdr[2], m[2]->pkt_len) < 0) {
 934                 rte_pktmbuf_free(m[2]);
 935                 valid_mask &= EXCLUDE_3RD_PKT;
 936         }
 937         if (is_valid_ipv4_pkt(ipv4_hdr[3], m[3]->pkt_len) < 0) {
 938                 rte_pktmbuf_free(m[3]);
 939                 valid_mask &= EXCLUDE_4TH_PKT;
 940         }
 941         if (is_valid_ipv4_pkt(ipv4_hdr[4], m[4]->pkt_len) < 0) {
 942                 rte_pktmbuf_free(m[4]);
 943                 valid_mask &= EXCLUDE_5TH_PKT;
 944         }
 945         if (is_valid_ipv4_pkt(ipv4_hdr[5], m[5]->pkt_len) < 0) {
 946                 rte_pktmbuf_free(m[5]);
 947                 valid_mask &= EXCLUDE_6TH_PKT;
 948         }
 949         if (is_valid_ipv4_pkt(ipv4_hdr[6], m[6]->pkt_len) < 0) {
 950                 rte_pktmbuf_free(m[6]);
 951                 valid_mask &= EXCLUDE_7TH_PKT;
 952         }
 953         if (is_valid_ipv4_pkt(ipv4_hdr[7], m[7]->pkt_len) < 0) {
 954                 rte_pktmbuf_free(m[7]);
 955                 valid_mask &= EXCLUDE_8TH_PKT;
 956         }
 957         if (unlikely(valid_mask != MASK_ALL_PKTS)) {
 958                 if (valid_mask == 0)
 959                         return;
 960
 961                 uint8_t i = 0;
 962
 963                 for (i = 0; i < 8; i++)
 964                         if ((0x1 << i) & valid_mask)
 965                                 l3fwd_simple_forward(m[i], portid);
 966         }
 967 #endif /* End of #ifdef DO_RFC_1812_CHECKS */
 968
 969         data[0] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[0], __m128i *,
 970                         sizeof(struct rte_ether_hdr) +
 971                         offsetof(struct ipv4_hdr, time_to_live)));
 972         data[1] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[1], __m128i *,
 973                         sizeof(struct rte_ether_hdr) +
 974                         offsetof(struct ipv4_hdr, time_to_live)));
 975         data[2] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[2], __m128i *,
 976                         sizeof(struct rte_ether_hdr) +
 977                         offsetof(struct ipv4_hdr, time_to_live)));
 978         data[3] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[3], __m128i *,
 979                         sizeof(struct rte_ether_hdr) +
 980                         offsetof(struct ipv4_hdr, time_to_live)));
 981         data[4] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[4], __m128i *,
 982                         sizeof(struct rte_ether_hdr) +
 983                         offsetof(struct ipv4_hdr, time_to_live)));
 984         data[5] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[5], __m128i *,
 985                         sizeof(struct rte_ether_hdr) +
 986                         offsetof(struct ipv4_hdr, time_to_live)));
 987         data[6] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[6], __m128i *,
 988                         sizeof(struct rte_ether_hdr) +
 989                         offsetof(struct ipv4_hdr, time_to_live)));
 990         data[7] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[7], __m128i *,
 991                         sizeof(struct rte_ether_hdr) +
 992                         offsetof(struct ipv4_hdr, time_to_live)));
 993
 994         key[0].xmm = _mm_and_si128(data[0], mask0);
 995         key[1].xmm = _mm_and_si128(data[1], mask0);
 996         key[2].xmm = _mm_and_si128(data[2], mask0);
 997         key[3].xmm = _mm_and_si128(data[3], mask0);
 998         key[4].xmm = _mm_and_si128(data[4], mask0);
 999         key[5].xmm = _mm_and_si128(data[5], mask0);
1000         key[6].xmm = _mm_and_si128(data[6], mask0);
1001         key[7].xmm = _mm_and_si128(data[7], mask0);
1002
1003         const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1004                         &key[4], &key[5], &key[6], &key[7]};
1005
1006         rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct,
1007                         &key_array[0], 8, ret);
1008         dst_port[0] = ((ret[0] < 0) ? portid : ipv4_l3fwd_out_if[ret[0]]);
1009         dst_port[1] = ((ret[1] < 0) ? portid : ipv4_l3fwd_out_if[ret[1]]);
1010         dst_port[2] = ((ret[2] < 0) ? portid : ipv4_l3fwd_out_if[ret[2]]);
1011         dst_port[3] = ((ret[3] < 0) ? portid : ipv4_l3fwd_out_if[ret[3]]);
1012         dst_port[4] = ((ret[4] < 0) ? portid : ipv4_l3fwd_out_if[ret[4]]);
1013         dst_port[5] = ((ret[5] < 0) ? portid : ipv4_l3fwd_out_if[ret[5]]);
1014         dst_port[6] = ((ret[6] < 0) ? portid : ipv4_l3fwd_out_if[ret[6]]);
1015         dst_port[7] = ((ret[7] < 0) ? portid : ipv4_l3fwd_out_if[ret[7]]);
1016
1017         if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1018                         (enabled_port_mask & 1 << dst_port[0]) == 0)
1019                 dst_port[0] = portid;
1020         if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1021                         (enabled_port_mask & 1 << dst_port[1]) == 0)
1022                 dst_port[1] = portid;
1023         if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1024                         (enabled_port_mask & 1 << dst_port[2]) == 0)
1025                 dst_port[2] = portid;
1026         if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1027                         (enabled_port_mask & 1 << dst_port[3]) == 0)
1028                 dst_port[3] = portid;
1029         if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1030                         (enabled_port_mask & 1 << dst_port[4]) == 0)
1031                 dst_port[4] = portid;
1032         if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1033                         (enabled_port_mask & 1 << dst_port[5]) == 0)
1034                 dst_port[5] = portid;
1035         if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1036                         (enabled_port_mask & 1 << dst_port[6]) == 0)
1037                 dst_port[6] = portid;
1038         if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1039                         (enabled_port_mask & 1 << dst_port[7]) == 0)
1040                 dst_port[7] = portid;
1041
1042 #ifdef DO_RFC_1812_CHECKS
1043         /* Update time to live and header checksum */
1044         --(ipv4_hdr[0]->time_to_live);
1045         --(ipv4_hdr[1]->time_to_live);
1046         --(ipv4_hdr[2]->time_to_live);
1047         --(ipv4_hdr[3]->time_to_live);
1048         ++(ipv4_hdr[0]->hdr_checksum);
1049         ++(ipv4_hdr[1]->hdr_checksum);
1050         ++(ipv4_hdr[2]->hdr_checksum);
1051         ++(ipv4_hdr[3]->hdr_checksum);
1052         --(ipv4_hdr[4]->time_to_live);
1053         --(ipv4_hdr[5]->time_to_live);
1054         --(ipv4_hdr[6]->time_to_live);
1055         --(ipv4_hdr[7]->time_to_live);
1056         ++(ipv4_hdr[4]->hdr_checksum);
1057         ++(ipv4_hdr[5]->hdr_checksum);
1058         ++(ipv4_hdr[6]->hdr_checksum);
1059         ++(ipv4_hdr[7]->hdr_checksum);
1060 #endif
1061
1062         /* dst addr */
1063         *(uint64_t *)&eth_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1064         *(uint64_t *)&eth_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1065         *(uint64_t *)&eth_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1066         *(uint64_t *)&eth_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1067         *(uint64_t *)&eth_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1068         *(uint64_t *)&eth_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1069         *(uint64_t *)&eth_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1070         *(uint64_t *)&eth_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1071
1072         /* src addr */
1073         rte_ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
1074         rte_ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
1075         rte_ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
1076         rte_ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
1077         rte_ether_addr_copy(&ports_eth_addr[dst_port[4]], &eth_hdr[4]->s_addr);
1078         rte_ether_addr_copy(&ports_eth_addr[dst_port[5]], &eth_hdr[5]->s_addr);
1079         rte_ether_addr_copy(&ports_eth_addr[dst_port[6]], &eth_hdr[6]->s_addr);
1080         rte_ether_addr_copy(&ports_eth_addr[dst_port[7]], &eth_hdr[7]->s_addr);
1081
1082         send_single_packet(m[0], (uint8_t)dst_port[0]);
1083         send_single_packet(m[1], (uint8_t)dst_port[1]);
1084         send_single_packet(m[2], (uint8_t)dst_port[2]);
1085         send_single_packet(m[3], (uint8_t)dst_port[3]);
1086         send_single_packet(m[4], (uint8_t)dst_port[4]);
1087         send_single_packet(m[5], (uint8_t)dst_port[5]);
1088         send_single_packet(m[6], (uint8_t)dst_port[6]);
1089         send_single_packet(m[7], (uint8_t)dst_port[7]);
1090
1091 }
1092
1093 static inline void get_ipv6_5tuple(struct rte_mbuf *m0, __m128i mask0,
1094                 __m128i mask1, union ipv6_5tuple_host *key)
1095 {
1096         __m128i tmpdata0 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1097                         __m128i *, sizeof(struct rte_ether_hdr) +
1098                         offsetof(struct ipv6_hdr, payload_len)));
1099         __m128i tmpdata1 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1100                         __m128i *, sizeof(struct rte_ether_hdr) +
1101                         offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i)));
1102         __m128i tmpdata2 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1103                         __m128i *, sizeof(struct rte_ether_hdr) +
1104                         offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i) +
1105                         sizeof(__m128i)));
1106         key->xmm[0] = _mm_and_si128(tmpdata0, mask0);
1107         key->xmm[1] = tmpdata1;
1108         key->xmm[2] = _mm_and_si128(tmpdata2, mask1);
1109 }
1110
1111 static inline void
1112 simple_ipv6_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid)
1113 {
1114         int32_t ret[8];
1115         uint16_t dst_port[8];
1116         struct rte_ether_hdr *eth_hdr[8];
1117         union ipv6_5tuple_host key[8];
1118
1119         __attribute__((unused)) struct ipv6_hdr *ipv6_hdr[8];
1120
1121         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct rte_ether_hdr *);
1122         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct rte_ether_hdr *);
1123         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct rte_ether_hdr *);
1124         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct rte_ether_hdr *);
1125         eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct rte_ether_hdr *);
1126         eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct rte_ether_hdr *);
1127         eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct rte_ether_hdr *);
1128         eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct rte_ether_hdr *);
1129
1130         /* Handle IPv6 headers.*/
1131         ipv6_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv6_hdr *,
1132                         sizeof(struct rte_ether_hdr));
1133         ipv6_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv6_hdr *,
1134                         sizeof(struct rte_ether_hdr));
1135         ipv6_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv6_hdr *,
1136                         sizeof(struct rte_ether_hdr));
1137         ipv6_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv6_hdr *,
1138                         sizeof(struct rte_ether_hdr));
1139         ipv6_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv6_hdr *,
1140                         sizeof(struct rte_ether_hdr));
1141         ipv6_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv6_hdr *,
1142                         sizeof(struct rte_ether_hdr));
1143         ipv6_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv6_hdr *,
1144                         sizeof(struct rte_ether_hdr));
1145         ipv6_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv6_hdr *,
1146                         sizeof(struct rte_ether_hdr));
1147
1148         get_ipv6_5tuple(m[0], mask1, mask2, &key[0]);
1149         get_ipv6_5tuple(m[1], mask1, mask2, &key[1]);
1150         get_ipv6_5tuple(m[2], mask1, mask2, &key[2]);
1151         get_ipv6_5tuple(m[3], mask1, mask2, &key[3]);
1152         get_ipv6_5tuple(m[4], mask1, mask2, &key[4]);
1153         get_ipv6_5tuple(m[5], mask1, mask2, &key[5]);
1154         get_ipv6_5tuple(m[6], mask1, mask2, &key[6]);
1155         get_ipv6_5tuple(m[7], mask1, mask2, &key[7]);
1156
1157         const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1158                         &key[4], &key[5], &key[6], &key[7]};
1159
1160         rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1161                         &key_array[0], 4, ret);
1162         dst_port[0] = ((ret[0] < 0) ? portid : ipv6_l3fwd_out_if[ret[0]]);
1163         dst_port[1] = ((ret[1] < 0) ? portid : ipv6_l3fwd_out_if[ret[1]]);
1164         dst_port[2] = ((ret[2] < 0) ? portid : ipv6_l3fwd_out_if[ret[2]]);
1165         dst_port[3] = ((ret[3] < 0) ? portid : ipv6_l3fwd_out_if[ret[3]]);
1166         dst_port[4] = ((ret[4] < 0) ? portid : ipv6_l3fwd_out_if[ret[4]]);
1167         dst_port[5] = ((ret[5] < 0) ? portid : ipv6_l3fwd_out_if[ret[5]]);
1168         dst_port[6] = ((ret[6] < 0) ? portid : ipv6_l3fwd_out_if[ret[6]]);
1169         dst_port[7] = ((ret[7] < 0) ? portid : ipv6_l3fwd_out_if[ret[7]]);
1170
1171         if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1172                         (enabled_port_mask & 1 << dst_port[0]) == 0)
1173                 dst_port[0] = portid;
1174         if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1175                         (enabled_port_mask & 1 << dst_port[1]) == 0)
1176                 dst_port[1] = portid;
1177         if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1178                         (enabled_port_mask & 1 << dst_port[2]) == 0)
1179                 dst_port[2] = portid;
1180         if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1181                         (enabled_port_mask & 1 << dst_port[3]) == 0)
1182                 dst_port[3] = portid;
1183         if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1184                         (enabled_port_mask & 1 << dst_port[4]) == 0)
1185                 dst_port[4] = portid;
1186         if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1187                         (enabled_port_mask & 1 << dst_port[5]) == 0)
1188                 dst_port[5] = portid;
1189         if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1190                         (enabled_port_mask & 1 << dst_port[6]) == 0)
1191                 dst_port[6] = portid;
1192         if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1193                         (enabled_port_mask & 1 << dst_port[7]) == 0)
1194                 dst_port[7] = portid;
1195
1196         /* dst addr */
1197         *(uint64_t *)&eth_hdr[0]->d_addr = dest_eth_addr[dst_port[0]];
1198         *(uint64_t *)&eth_hdr[1]->d_addr = dest_eth_addr[dst_port[1]];
1199         *(uint64_t *)&eth_hdr[2]->d_addr = dest_eth_addr[dst_port[2]];
1200         *(uint64_t *)&eth_hdr[3]->d_addr = dest_eth_addr[dst_port[3]];
1201         *(uint64_t *)&eth_hdr[4]->d_addr = dest_eth_addr[dst_port[4]];
1202         *(uint64_t *)&eth_hdr[5]->d_addr = dest_eth_addr[dst_port[5]];
1203         *(uint64_t *)&eth_hdr[6]->d_addr = dest_eth_addr[dst_port[6]];
1204         *(uint64_t *)&eth_hdr[7]->d_addr = dest_eth_addr[dst_port[7]];
1205
1206         /* src addr */
1207         rte_ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->s_addr);
1208         rte_ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->s_addr);
1209         rte_ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->s_addr);
1210         rte_ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->s_addr);
1211         rte_ether_addr_copy(&ports_eth_addr[dst_port[4]], &eth_hdr[4]->s_addr);
1212         rte_ether_addr_copy(&ports_eth_addr[dst_port[5]], &eth_hdr[5]->s_addr);
1213         rte_ether_addr_copy(&ports_eth_addr[dst_port[6]], &eth_hdr[6]->s_addr);
1214         rte_ether_addr_copy(&ports_eth_addr[dst_port[7]], &eth_hdr[7]->s_addr);
1215
1216         send_single_packet(m[0], dst_port[0]);
1217         send_single_packet(m[1], dst_port[1]);
1218         send_single_packet(m[2], dst_port[2]);
1219         send_single_packet(m[3], dst_port[3]);
1220         send_single_packet(m[4], dst_port[4]);
1221         send_single_packet(m[5], dst_port[5]);
1222         send_single_packet(m[6], dst_port[6]);
1223         send_single_packet(m[7], dst_port[7]);
1224
1225 }
1226 #endif /* APP_LOOKUP_METHOD */
1227
1228 static __rte_always_inline void
1229 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid)
1230 {
1231         struct rte_ether_hdr *eth_hdr;
1232         struct ipv4_hdr *ipv4_hdr;
1233         uint16_t dst_port;
1234
1235         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
1236
1237         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
1238                 /* Handle IPv4 headers.*/
1239                 ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
1240                                 sizeof(struct rte_ether_hdr));
1241
1242 #ifdef DO_RFC_1812_CHECKS
1243                 /* Check to make sure the packet is valid (RFC1812) */
1244                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
1245                         rte_pktmbuf_free(m);
1246                         return;
1247                 }
1248 #endif
1249
1250                  dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
1251                         RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct);
1252                 if (dst_port >= RTE_MAX_ETHPORTS ||
1253                                 (enabled_port_mask & 1 << dst_port) == 0)
1254                         dst_port = portid;
1255
1256 #ifdef DO_RFC_1812_CHECKS
1257                 /* Update time to live and header checksum */
1258                 --(ipv4_hdr->time_to_live);
1259                 ++(ipv4_hdr->hdr_checksum);
1260 #endif
1261                 /* dst addr */
1262                 *(uint64_t *)&eth_hdr->d_addr = dest_eth_addr[dst_port];
1263
1264                 /* src addr */
1265                 rte_ether_addr_copy(&ports_eth_addr[dst_port],
1266                                 &eth_hdr->s_addr);
1267
1268                 send_single_packet(m, dst_port);
1269         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
1270                 /* Handle IPv6 headers.*/
1271                 struct ipv6_hdr *ipv6_hdr;
1272
1273                 ipv6_hdr = rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
1274                                 sizeof(struct rte_ether_hdr));
1275
1276                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
1277                                 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct);
1278
1279                 if (dst_port >= RTE_MAX_ETHPORTS ||
1280                                 (enabled_port_mask & 1 << dst_port) == 0)
1281                         dst_port = portid;
1282
1283                 /* dst addr */
1284                 *(uint64_t *)&eth_hdr->d_addr = dest_eth_addr[dst_port];
1285
1286                 /* src addr */
1287                 rte_ether_addr_copy(&ports_eth_addr[dst_port],
1288                                 &eth_hdr->s_addr);
1289
1290                 send_single_packet(m, dst_port);
1291         } else
1292                 /* Free the mbuf that contains non-IPV4/IPV6 packet */
1293                 rte_pktmbuf_free(m);
1294 }
1295
1296 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1297         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1298 #ifdef DO_RFC_1812_CHECKS
1299
1300 #define IPV4_MIN_VER_IHL        0x45
1301 #define IPV4_MAX_VER_IHL        0x4f
1302 #define IPV4_MAX_VER_IHL_DIFF   (IPV4_MAX_VER_IHL - IPV4_MIN_VER_IHL)
1303
1304 /* Minimum value of IPV4 total length (20B) in network byte order. */
1305 #define IPV4_MIN_LEN_BE (sizeof(struct ipv4_hdr) << 8)
1306
1307 /*
1308  * From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2:
1309  * - The IP version number must be 4.
1310  * - The IP header length field must be large enough to hold the
1311  *    minimum length legal IP datagram (20 bytes = 5 words).
1312  * - The IP total length field must be large enough to hold the IP
1313  *   datagram header, whose length is specified in the IP header length
1314  *   field.
1315  * If we encounter invalid IPV4 packet, then set destination port for it
1316  * to BAD_PORT value.
1317  */
1318 static __rte_always_inline void
1319 rfc1812_process(struct ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t ptype)
1320 {
1321         uint8_t ihl;
1322
1323         if (RTE_ETH_IS_IPV4_HDR(ptype)) {
1324                 ihl = ipv4_hdr->version_ihl - IPV4_MIN_VER_IHL;
1325
1326                 ipv4_hdr->time_to_live--;
1327                 ipv4_hdr->hdr_checksum++;
1328
1329                 if (ihl > IPV4_MAX_VER_IHL_DIFF ||
1330                                 ((uint8_t)ipv4_hdr->total_length == 0 &&
1331                                 ipv4_hdr->total_length < IPV4_MIN_LEN_BE)) {
1332                         dp[0] = BAD_PORT;
1333                 }
1334         }
1335 }
1336
1337 #else
1338 #define rfc1812_process(mb, dp, ptype)  do { } while (0)
1339 #endif /* DO_RFC_1812_CHECKS */
1340 #endif /* APP_LOOKUP_LPM && ENABLE_MULTI_BUFFER_OPTIMIZE */
1341
1342
1343 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1344         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1345
1346 static __rte_always_inline uint16_t
1347 get_dst_port(struct rte_mbuf *pkt, uint32_t dst_ipv4, uint16_t portid)
1348 {
1349         uint32_t next_hop;
1350         struct ipv6_hdr *ipv6_hdr;
1351         struct rte_ether_hdr *eth_hdr;
1352
1353         if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
1354                 return (uint16_t) ((rte_lpm_lookup(
1355                                 RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dst_ipv4,
1356                                 &next_hop) == 0) ? next_hop : portid);
1357
1358         } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
1359
1360                 eth_hdr = rte_pktmbuf_mtod(pkt, struct rte_ether_hdr *);
1361                 ipv6_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
1362
1363                 return (uint16_t) ((rte_lpm6_lookup(
1364                                 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1365                                 ipv6_hdr->dst_addr, &next_hop) == 0) ?
1366                                 next_hop : portid);
1367
1368         }
1369
1370         return portid;
1371 }
1372
1373 static inline void
1374 process_packet(struct rte_mbuf *pkt, uint16_t *dst_port, uint16_t portid)
1375 {
1376         struct rte_ether_hdr *eth_hdr;
1377         struct ipv4_hdr *ipv4_hdr;
1378         uint32_t dst_ipv4;
1379         uint16_t dp;
1380         __m128i te, ve;
1381
1382         eth_hdr = rte_pktmbuf_mtod(pkt, struct rte_ether_hdr *);
1383         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1384
1385         dst_ipv4 = ipv4_hdr->dst_addr;
1386         dst_ipv4 = rte_be_to_cpu_32(dst_ipv4);
1387         dp = get_dst_port(pkt, dst_ipv4, portid);
1388
1389         te = _mm_load_si128((__m128i *)eth_hdr);
1390         ve = val_eth[dp];
1391
1392         dst_port[0] = dp;
1393         rfc1812_process(ipv4_hdr, dst_port, pkt->packet_type);
1394
1395         te =  _mm_blend_epi16(te, ve, MASK_ETH);
1396         _mm_store_si128((__m128i *)eth_hdr, te);
1397 }
1398
1399 /*
1400  * Read packet_type and destination IPV4 addresses from 4 mbufs.
1401  */
1402 static inline void
1403 processx4_step1(struct rte_mbuf *pkt[FWDSTEP],
1404                 __m128i *dip,
1405                 uint32_t *ipv4_flag)
1406 {
1407         struct ipv4_hdr *ipv4_hdr;
1408         struct rte_ether_hdr *eth_hdr;
1409         uint32_t x0, x1, x2, x3;
1410
1411         eth_hdr = rte_pktmbuf_mtod(pkt[0], struct rte_ether_hdr *);
1412         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1413         x0 = ipv4_hdr->dst_addr;
1414         ipv4_flag[0] = pkt[0]->packet_type & RTE_PTYPE_L3_IPV4;
1415
1416         eth_hdr = rte_pktmbuf_mtod(pkt[1], struct rte_ether_hdr *);
1417         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1418         x1 = ipv4_hdr->dst_addr;
1419         ipv4_flag[0] &= pkt[1]->packet_type;
1420
1421         eth_hdr = rte_pktmbuf_mtod(pkt[2], struct rte_ether_hdr *);
1422         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1423         x2 = ipv4_hdr->dst_addr;
1424         ipv4_flag[0] &= pkt[2]->packet_type;
1425
1426         eth_hdr = rte_pktmbuf_mtod(pkt[3], struct rte_ether_hdr *);
1427         ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
1428         x3 = ipv4_hdr->dst_addr;
1429         ipv4_flag[0] &= pkt[3]->packet_type;
1430
1431         dip[0] = _mm_set_epi32(x3, x2, x1, x0);
1432 }
1433
1434 /*
1435  * Lookup into LPM for destination port.
1436  * If lookup fails, use incoming port (portid) as destination port.
1437  */
1438 static inline void
1439 processx4_step2(__m128i dip,
1440                 uint32_t ipv4_flag,
1441                 uint16_t portid,
1442                 struct rte_mbuf *pkt[FWDSTEP],
1443                 uint16_t dprt[FWDSTEP])
1444 {
1445         rte_xmm_t dst;
1446         const __m128i bswap_mask = _mm_set_epi8(12, 13, 14, 15, 8, 9, 10, 11,
1447                         4, 5, 6, 7, 0, 1, 2, 3);
1448
1449         /* Byte swap 4 IPV4 addresses. */
1450         dip = _mm_shuffle_epi8(dip, bswap_mask);
1451
1452         /* if all 4 packets are IPV4. */
1453         if (likely(ipv4_flag)) {
1454                 rte_lpm_lookupx4(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dip,
1455                                 dst.u32, portid);
1456
1457                 /* get rid of unused upper 16 bit for each dport. */
1458                 dst.x = _mm_packs_epi32(dst.x, dst.x);
1459                 *(uint64_t *)dprt = dst.u64[0];
1460         } else {
1461                 dst.x = dip;
1462                 dprt[0] = get_dst_port(pkt[0], dst.u32[0], portid);
1463                 dprt[1] = get_dst_port(pkt[1], dst.u32[1], portid);
1464                 dprt[2] = get_dst_port(pkt[2], dst.u32[2], portid);
1465                 dprt[3] = get_dst_port(pkt[3], dst.u32[3], portid);
1466         }
1467 }
1468
1469 /*
1470  * Update source and destination MAC addresses in the ethernet header.
1471  * Perform RFC1812 checks and updates for IPV4 packets.
1472  */
1473 static inline void
1474 processx4_step3(struct rte_mbuf *pkt[FWDSTEP], uint16_t dst_port[FWDSTEP])
1475 {
1476         __m128i te[FWDSTEP];
1477         __m128i ve[FWDSTEP];
1478         __m128i *p[FWDSTEP];
1479
1480         p[0] = rte_pktmbuf_mtod(pkt[0], __m128i *);
1481         p[1] = rte_pktmbuf_mtod(pkt[1], __m128i *);
1482         p[2] = rte_pktmbuf_mtod(pkt[2], __m128i *);
1483         p[3] = rte_pktmbuf_mtod(pkt[3], __m128i *);
1484
1485         ve[0] = val_eth[dst_port[0]];
1486         te[0] = _mm_load_si128(p[0]);
1487
1488         ve[1] = val_eth[dst_port[1]];
1489         te[1] = _mm_load_si128(p[1]);
1490
1491         ve[2] = val_eth[dst_port[2]];
1492         te[2] = _mm_load_si128(p[2]);
1493
1494         ve[3] = val_eth[dst_port[3]];
1495         te[3] = _mm_load_si128(p[3]);
1496
1497         /* Update first 12 bytes, keep rest bytes intact. */
1498         te[0] =  _mm_blend_epi16(te[0], ve[0], MASK_ETH);
1499         te[1] =  _mm_blend_epi16(te[1], ve[1], MASK_ETH);
1500         te[2] =  _mm_blend_epi16(te[2], ve[2], MASK_ETH);
1501         te[3] =  _mm_blend_epi16(te[3], ve[3], MASK_ETH);
1502
1503         _mm_store_si128(p[0], te[0]);
1504         _mm_store_si128(p[1], te[1]);
1505         _mm_store_si128(p[2], te[2]);
1506         _mm_store_si128(p[3], te[3]);
1507
1508         rfc1812_process((struct ipv4_hdr *)((struct rte_ether_hdr *)p[0] + 1),
1509                         &dst_port[0], pkt[0]->packet_type);
1510         rfc1812_process((struct ipv4_hdr *)((struct rte_ether_hdr *)p[1] + 1),
1511                         &dst_port[1], pkt[1]->packet_type);
1512         rfc1812_process((struct ipv4_hdr *)((struct rte_ether_hdr *)p[2] + 1),
1513                         &dst_port[2], pkt[2]->packet_type);
1514         rfc1812_process((struct ipv4_hdr *)((struct rte_ether_hdr *)p[3] + 1),
1515                         &dst_port[3], pkt[3]->packet_type);
1516 }
1517
1518 /*
1519  * We group consecutive packets with the same destionation port into one burst.
1520  * To avoid extra latency this is done together with some other packet
1521  * processing, but after we made a final decision about packet's destination.
1522  * To do this we maintain:
1523  * pnum - array of number of consecutive packets with the same dest port for
1524  * each packet in the input burst.
1525  * lp - pointer to the last updated element in the pnum.
1526  * dlp - dest port value lp corresponds to.
1527  */
1528
1529 #define GRPSZ   (1 << FWDSTEP)
1530 #define GRPMSK  (GRPSZ - 1)
1531
1532 #define GROUP_PORT_STEP(dlp, dcp, lp, pn, idx)  do { \
1533         if (likely((dlp) == (dcp)[(idx)])) {         \
1534                 (lp)[0]++;                           \
1535         } else {                                     \
1536                 (dlp) = (dcp)[idx];                  \
1537                 (lp) = (pn) + (idx);                 \
1538                 (lp)[0] = 1;                         \
1539         }                                            \
1540 } while (0)
1541
1542 /*
1543  * Group consecutive packets with the same destination port in bursts of 4.
1544  * Suppose we have array of destionation ports:
1545  * dst_port[] = {a, b, c, d,, e, ... }
1546  * dp1 should contain: <a, b, c, d>, dp2: <b, c, d, e>.
1547  * We doing 4 comparisons at once and the result is 4 bit mask.
1548  * This mask is used as an index into prebuild array of pnum values.
1549  */
1550 static inline uint16_t *
1551 port_groupx4(uint16_t pn[FWDSTEP + 1], uint16_t *lp, __m128i dp1, __m128i dp2)
1552 {
1553         static const struct {
1554                 uint64_t pnum; /* prebuild 4 values for pnum[]. */
1555                 int32_t  idx;  /* index for new last updated elemnet. */
1556                 uint16_t lpv;  /* add value to the last updated element. */
1557         } gptbl[GRPSZ] = {
1558         {
1559                 /* 0: a != b, b != c, c != d, d != e */
1560                 .pnum = UINT64_C(0x0001000100010001),
1561                 .idx = 4,
1562                 .lpv = 0,
1563         },
1564         {
1565                 /* 1: a == b, b != c, c != d, d != e */
1566                 .pnum = UINT64_C(0x0001000100010002),
1567                 .idx = 4,
1568                 .lpv = 1,
1569         },
1570         {
1571                 /* 2: a != b, b == c, c != d, d != e */
1572                 .pnum = UINT64_C(0x0001000100020001),
1573                 .idx = 4,
1574                 .lpv = 0,
1575         },
1576         {
1577                 /* 3: a == b, b == c, c != d, d != e */
1578                 .pnum = UINT64_C(0x0001000100020003),
1579                 .idx = 4,
1580                 .lpv = 2,
1581         },
1582         {
1583                 /* 4: a != b, b != c, c == d, d != e */
1584                 .pnum = UINT64_C(0x0001000200010001),
1585                 .idx = 4,
1586                 .lpv = 0,
1587         },
1588         {
1589                 /* 5: a == b, b != c, c == d, d != e */
1590                 .pnum = UINT64_C(0x0001000200010002),
1591                 .idx = 4,
1592                 .lpv = 1,
1593         },
1594         {
1595                 /* 6: a != b, b == c, c == d, d != e */
1596                 .pnum = UINT64_C(0x0001000200030001),
1597                 .idx = 4,
1598                 .lpv = 0,
1599         },
1600         {
1601                 /* 7: a == b, b == c, c == d, d != e */
1602                 .pnum = UINT64_C(0x0001000200030004),
1603                 .idx = 4,
1604                 .lpv = 3,
1605         },
1606         {
1607                 /* 8: a != b, b != c, c != d, d == e */
1608                 .pnum = UINT64_C(0x0002000100010001),
1609                 .idx = 3,
1610                 .lpv = 0,
1611         },
1612         {
1613                 /* 9: a == b, b != c, c != d, d == e */
1614                 .pnum = UINT64_C(0x0002000100010002),
1615                 .idx = 3,
1616                 .lpv = 1,
1617         },
1618         {
1619                 /* 0xa: a != b, b == c, c != d, d == e */
1620                 .pnum = UINT64_C(0x0002000100020001),
1621                 .idx = 3,
1622                 .lpv = 0,
1623         },
1624         {
1625                 /* 0xb: a == b, b == c, c != d, d == e */
1626                 .pnum = UINT64_C(0x0002000100020003),
1627                 .idx = 3,
1628                 .lpv = 2,
1629         },
1630         {
1631                 /* 0xc: a != b, b != c, c == d, d == e */
1632                 .pnum = UINT64_C(0x0002000300010001),
1633                 .idx = 2,
1634                 .lpv = 0,
1635         },
1636         {
1637                 /* 0xd: a == b, b != c, c == d, d == e */
1638                 .pnum = UINT64_C(0x0002000300010002),
1639                 .idx = 2,
1640                 .lpv = 1,
1641         },
1642         {
1643                 /* 0xe: a != b, b == c, c == d, d == e */
1644                 .pnum = UINT64_C(0x0002000300040001),
1645                 .idx = 1,
1646                 .lpv = 0,
1647         },
1648         {
1649                 /* 0xf: a == b, b == c, c == d, d == e */
1650                 .pnum = UINT64_C(0x0002000300040005),
1651                 .idx = 0,
1652                 .lpv = 4,
1653         },
1654         };
1655
1656         union {
1657                 uint16_t u16[FWDSTEP + 1];
1658                 uint64_t u64;
1659         } *pnum = (void *)pn;
1660
1661         int32_t v;
1662
1663         dp1 = _mm_cmpeq_epi16(dp1, dp2);
1664         dp1 = _mm_unpacklo_epi16(dp1, dp1);
1665         v = _mm_movemask_ps((__m128)dp1);
1666
1667         /* update last port counter. */
1668         lp[0] += gptbl[v].lpv;
1669
1670         /* if dest port value has changed. */
1671         if (v != GRPMSK) {
1672                 pnum->u64 = gptbl[v].pnum;
1673                 pnum->u16[FWDSTEP] = 1;
1674                 lp = pnum->u16 + gptbl[v].idx;
1675         }
1676
1677         return lp;
1678 }
1679
1680 #endif /* APP_LOOKUP_METHOD */
1681
1682 static void
1683 process_burst(struct rte_mbuf *pkts_burst[MAX_PKT_BURST], int nb_rx,
1684                 uint16_t portid)
1685 {
1686
1687         int j;
1688
1689 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1690         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1691         int32_t k;
1692         uint16_t dlp;
1693         uint16_t *lp;
1694         uint16_t dst_port[MAX_PKT_BURST];
1695         __m128i dip[MAX_PKT_BURST / FWDSTEP];
1696         uint32_t ipv4_flag[MAX_PKT_BURST / FWDSTEP];
1697         uint16_t pnum[MAX_PKT_BURST + 1];
1698 #endif
1699
1700
1701 #if (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
1702 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1703         {
1704                 /*
1705                  * Send nb_rx - nb_rx%8 packets
1706                  * in groups of 8.
1707                  */
1708                 int32_t n = RTE_ALIGN_FLOOR(nb_rx, 8);
1709
1710                 for (j = 0; j < n; j += 8) {
1711                         uint32_t pkt_type =
1712                                 pkts_burst[j]->packet_type &
1713                                 pkts_burst[j+1]->packet_type &
1714                                 pkts_burst[j+2]->packet_type &
1715                                 pkts_burst[j+3]->packet_type &
1716                                 pkts_burst[j+4]->packet_type &
1717                                 pkts_burst[j+5]->packet_type &
1718                                 pkts_burst[j+6]->packet_type &
1719                                 pkts_burst[j+7]->packet_type;
1720                         if (pkt_type & RTE_PTYPE_L3_IPV4) {
1721                                 simple_ipv4_fwd_8pkts(&pkts_burst[j], portid);
1722                         } else if (pkt_type &
1723                                 RTE_PTYPE_L3_IPV6) {
1724                                 simple_ipv6_fwd_8pkts(&pkts_burst[j], portid);
1725                         } else {
1726                                 l3fwd_simple_forward(pkts_burst[j], portid);
1727                                 l3fwd_simple_forward(pkts_burst[j+1], portid);
1728                                 l3fwd_simple_forward(pkts_burst[j+2], portid);
1729                                 l3fwd_simple_forward(pkts_burst[j+3], portid);
1730                                 l3fwd_simple_forward(pkts_burst[j+4], portid);
1731                                 l3fwd_simple_forward(pkts_burst[j+5], portid);
1732                                 l3fwd_simple_forward(pkts_burst[j+6], portid);
1733                                 l3fwd_simple_forward(pkts_burst[j+7], portid);
1734                         }
1735                 }
1736                 for (; j < nb_rx ; j++)
1737                         l3fwd_simple_forward(pkts_burst[j], portid);
1738         }
1739 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1740
1741         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1742         for (j = 0; j != k; j += FWDSTEP)
1743                 processx4_step1(&pkts_burst[j], &dip[j / FWDSTEP],
1744                                 &ipv4_flag[j / FWDSTEP]);
1745
1746         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1747         for (j = 0; j != k; j += FWDSTEP)
1748                 processx4_step2(dip[j / FWDSTEP], ipv4_flag[j / FWDSTEP],
1749                                 portid, &pkts_burst[j], &dst_port[j]);
1750
1751         /*
1752          * Finish packet processing and group consecutive
1753          * packets with the same destination port.
1754          */
1755         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1756         if (k != 0) {
1757                 __m128i dp1, dp2;
1758
1759                 lp = pnum;
1760                 lp[0] = 1;
1761
1762                 processx4_step3(pkts_burst, dst_port);
1763
1764                 /* dp1: <d[0], d[1], d[2], d[3], ... > */
1765                 dp1 = _mm_loadu_si128((__m128i *)dst_port);
1766
1767                 for (j = FWDSTEP; j != k; j += FWDSTEP) {
1768                         processx4_step3(&pkts_burst[j], &dst_port[j]);
1769
1770                         /*
1771                          * dp2:
1772                          * <d[j-3], d[j-2], d[j-1], d[j], ... >
1773                          */
1774                         dp2 = _mm_loadu_si128(
1775                                         (__m128i *)&dst_port[j - FWDSTEP + 1]);
1776                         lp  = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1777
1778                         /*
1779                          * dp1:
1780                          * <d[j], d[j+1], d[j+2], d[j+3], ... >
1781                          */
1782                         dp1 = _mm_srli_si128(dp2, (FWDSTEP - 1) *
1783                                         sizeof(dst_port[0]));
1784                 }
1785
1786                 /*
1787                  * dp2: <d[j-3], d[j-2], d[j-1], d[j-1], ... >
1788                  */
1789                 dp2 = _mm_shufflelo_epi16(dp1, 0xf9);
1790                 lp  = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1791
1792                 /*
1793                  * remove values added by the last repeated
1794                  * dst port.
1795                  */
1796                 lp[0]--;
1797                 dlp = dst_port[j - 1];
1798         } else {
1799                 /* set dlp and lp to the never used values. */
1800                 dlp = BAD_PORT - 1;
1801                 lp = pnum + MAX_PKT_BURST;
1802         }
1803
1804         /* Process up to last 3 packets one by one. */
1805         switch (nb_rx % FWDSTEP) {
1806         case 3:
1807                 process_packet(pkts_burst[j], dst_port + j, portid);
1808                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1809                 j++;
1810                 /* fall-through */
1811         case 2:
1812                 process_packet(pkts_burst[j], dst_port + j, portid);
1813                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1814                 j++;
1815                 /* fall-through */
1816         case 1:
1817                 process_packet(pkts_burst[j], dst_port + j, portid);
1818                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1819                 j++;
1820         }
1821
1822         /*
1823          * Send packets out, through destination port.
1824          * Consecuteve pacekts with the same destination port
1825          * are already grouped together.
1826          * If destination port for the packet equals BAD_PORT,
1827          * then free the packet without sending it out.
1828          */
1829         for (j = 0; j < nb_rx; j += k) {
1830
1831                 int32_t m;
1832                 uint16_t pn;
1833
1834                 pn = dst_port[j];
1835                 k = pnum[j];
1836
1837                 if (likely(pn != BAD_PORT))
1838                         send_packetsx4(pn, pkts_burst + j, k);
1839                 else
1840                         for (m = j; m != j + k; m++)
1841                                 rte_pktmbuf_free(pkts_burst[m]);
1842
1843         }
1844
1845 #endif /* APP_LOOKUP_METHOD */
1846 #else /* ENABLE_MULTI_BUFFER_OPTIMIZE == 0 */
1847
1848         /* Prefetch first packets */
1849         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++)
1850                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[j], void *));
1851
1852         /* Prefetch and forward already prefetched packets */
1853         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1854                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1855                                 j + PREFETCH_OFFSET], void *));
1856                 l3fwd_simple_forward(pkts_burst[j], portid);
1857         }
1858
1859         /* Forward remaining prefetched packets */
1860         for (; j < nb_rx; j++)
1861                 l3fwd_simple_forward(pkts_burst[j], portid);
1862
1863 #endif /* ENABLE_MULTI_BUFFER_OPTIMIZE */
1864
1865 }
1866
1867 #if (APP_CPU_LOAD > 0)
1868
1869 /*
1870  * CPU-load stats collector
1871  */
1872 static int
1873 cpu_load_collector(__rte_unused void *arg) {
1874         unsigned i, j, k;
1875         uint64_t hits;
1876         uint64_t prev_tsc, diff_tsc, cur_tsc;
1877         uint64_t total[MAX_CPU] = { 0 };
1878         unsigned min_cpu = MAX_CPU;
1879         unsigned max_cpu = 0;
1880         unsigned cpu_id;
1881         int busy_total = 0;
1882         int busy_flag = 0;
1883
1884         unsigned int n_thread_per_cpu[MAX_CPU] = { 0 };
1885         struct thread_conf *thread_per_cpu[MAX_CPU][MAX_THREAD];
1886
1887         struct thread_conf *thread_conf;
1888
1889         const uint64_t interval_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1890                 US_PER_S * CPU_LOAD_TIMEOUT_US;
1891
1892         prev_tsc = 0;
1893         /*
1894          * Wait for all threads
1895          */
1896
1897         printf("Waiting for %d rx threads and %d tx threads\n", n_rx_thread,
1898                         n_tx_thread);
1899
1900         while (rte_atomic16_read(&rx_counter) < n_rx_thread)
1901                 rte_pause();
1902
1903         while (rte_atomic16_read(&tx_counter) < n_tx_thread)
1904                 rte_pause();
1905
1906         for (i = 0; i < n_rx_thread; i++) {
1907
1908                 thread_conf = &rx_thread[i].conf;
1909                 cpu_id = thread_conf->cpu_id;
1910                 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1911
1912                 if (cpu_id > max_cpu)
1913                         max_cpu = cpu_id;
1914                 if (cpu_id < min_cpu)
1915                         min_cpu = cpu_id;
1916         }
1917         for (i = 0; i < n_tx_thread; i++) {
1918
1919                 thread_conf = &tx_thread[i].conf;
1920                 cpu_id = thread_conf->cpu_id;
1921                 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1922
1923                 if (thread_conf->cpu_id > max_cpu)
1924                         max_cpu = thread_conf->cpu_id;
1925                 if (thread_conf->cpu_id < min_cpu)
1926                         min_cpu = thread_conf->cpu_id;
1927         }
1928
1929         while (1) {
1930
1931                 cpu_load.counter++;
1932                 for (i = min_cpu; i <= max_cpu; i++) {
1933                         for (j = 0; j < MAX_CPU_COUNTER; j++) {
1934                                 for (k = 0; k < n_thread_per_cpu[i]; k++)
1935                                         if (thread_per_cpu[i][k]->busy[j]) {
1936                                                 busy_flag = 1;
1937                                                 break;
1938                                         }
1939                                 if (busy_flag) {
1940                                         cpu_load.hits[j][i]++;
1941                                         busy_total = 1;
1942                                         busy_flag = 0;
1943                                 }
1944                         }
1945
1946                         if (busy_total) {
1947                                 total[i]++;
1948                                 busy_total = 0;
1949                         }
1950                 }
1951
1952                 cur_tsc = rte_rdtsc();
1953
1954                 diff_tsc = cur_tsc - prev_tsc;
1955                 if (unlikely(diff_tsc > interval_tsc)) {
1956
1957                         printf("\033c");
1958
1959                         printf("Cpu usage for %d rx threads and %d tx threads:\n\n",
1960                                         n_rx_thread, n_tx_thread);
1961
1962                         printf("cpu#     proc%%  poll%%  overhead%%\n\n");
1963
1964                         for (i = min_cpu; i <= max_cpu; i++) {
1965                                 hits = 0;
1966                                 printf("CPU %d:", i);
1967                                 for (j = 0; j < MAX_CPU_COUNTER; j++) {
1968                                         printf("%7" PRIu64 "",
1969                                                         cpu_load.hits[j][i] * 100 / cpu_load.counter);
1970                                         hits += cpu_load.hits[j][i];
1971                                         cpu_load.hits[j][i] = 0;
1972                                 }
1973                                 printf("%7" PRIu64 "\n",
1974                                                 100 - total[i] * 100 / cpu_load.counter);
1975                                 total[i] = 0;
1976                         }
1977                         cpu_load.counter = 0;
1978
1979                         prev_tsc = cur_tsc;
1980                 }
1981
1982         }
1983 }
1984 #endif /* APP_CPU_LOAD */
1985
1986 /*
1987  * Null processing lthread loop
1988  *
1989  * This loop is used to start empty scheduler on lcore.
1990  */
1991 static void *
1992 lthread_null(__rte_unused void *args)
1993 {
1994         int lcore_id = rte_lcore_id();
1995
1996         RTE_LOG(INFO, L3FWD, "Starting scheduler on lcore %d.\n", lcore_id);
1997         lthread_exit(NULL);
1998         return NULL;
1999 }
2000
2001 /* main processing loop */
2002 static void *
2003 lthread_tx_per_ring(void *dummy)
2004 {
2005         int nb_rx;
2006         uint16_t portid;
2007         struct rte_ring *ring;
2008         struct thread_tx_conf *tx_conf;
2009         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2010         struct lthread_cond *ready;
2011
2012         tx_conf = (struct thread_tx_conf *)dummy;
2013         ring = tx_conf->ring;
2014         ready = *tx_conf->ready;
2015
2016         lthread_set_data((void *)tx_conf);
2017
2018         /*
2019          * Move this lthread to lcore
2020          */
2021         lthread_set_affinity(tx_conf->conf.lcore_id);
2022
2023         RTE_LOG(INFO, L3FWD, "entering main tx loop on lcore %u\n", rte_lcore_id());
2024
2025         nb_rx = 0;
2026         rte_atomic16_inc(&tx_counter);
2027         while (1) {
2028
2029                 /*
2030                  * Read packet from ring
2031                  */
2032                 SET_CPU_BUSY(tx_conf, CPU_POLL);
2033                 nb_rx = rte_ring_sc_dequeue_burst(ring, (void **)pkts_burst,
2034                                 MAX_PKT_BURST, NULL);
2035                 SET_CPU_IDLE(tx_conf, CPU_POLL);
2036
2037                 if (nb_rx > 0) {
2038                         SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2039                         portid = pkts_burst[0]->port;
2040                         process_burst(pkts_burst, nb_rx, portid);
2041                         SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2042                         lthread_yield();
2043                 } else
2044                         lthread_cond_wait(ready, 0);
2045
2046         }
2047         return NULL;
2048 }
2049
2050 /*
2051  * Main tx-lthreads spawner lthread.
2052  *
2053  * This lthread is used to spawn one new lthread per ring from producers.
2054  *
2055  */
2056 static void *
2057 lthread_tx(void *args)
2058 {
2059         struct lthread *lt;
2060
2061         unsigned lcore_id;
2062         uint16_t portid;
2063         struct thread_tx_conf *tx_conf;
2064
2065         tx_conf = (struct thread_tx_conf *)args;
2066         lthread_set_data((void *)tx_conf);
2067
2068         /*
2069          * Move this lthread to the selected lcore
2070          */
2071         lthread_set_affinity(tx_conf->conf.lcore_id);
2072
2073         /*
2074          * Spawn tx readers (one per input ring)
2075          */
2076         lthread_create(&lt, tx_conf->conf.lcore_id, lthread_tx_per_ring,
2077                         (void *)tx_conf);
2078
2079         lcore_id = rte_lcore_id();
2080
2081         RTE_LOG(INFO, L3FWD, "Entering Tx main loop on lcore %u\n", lcore_id);
2082
2083         tx_conf->conf.cpu_id = sched_getcpu();
2084         while (1) {
2085
2086                 lthread_sleep(BURST_TX_DRAIN_US * 1000);
2087
2088                 /*
2089                  * TX burst queue drain
2090                  */
2091                 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2092                         if (tx_conf->tx_mbufs[portid].len == 0)
2093                                 continue;
2094                         SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2095                         send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2096                         SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2097                         tx_conf->tx_mbufs[portid].len = 0;
2098                 }
2099
2100         }
2101         return NULL;
2102 }
2103
2104 static void *
2105 lthread_rx(void *dummy)
2106 {
2107         int ret;
2108         uint16_t nb_rx;
2109         int i;
2110         uint16_t portid;
2111         uint8_t queueid;
2112         int worker_id;
2113         int len[RTE_MAX_LCORE] = { 0 };
2114         int old_len, new_len;
2115         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2116         struct thread_rx_conf *rx_conf;
2117
2118         rx_conf = (struct thread_rx_conf *)dummy;
2119         lthread_set_data((void *)rx_conf);
2120
2121         /*
2122          * Move this lthread to lcore
2123          */
2124         lthread_set_affinity(rx_conf->conf.lcore_id);
2125
2126         if (rx_conf->n_rx_queue == 0) {
2127                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", rte_lcore_id());
2128                 return NULL;
2129         }
2130
2131         RTE_LOG(INFO, L3FWD, "Entering main Rx loop on lcore %u\n", rte_lcore_id());
2132
2133         for (i = 0; i < rx_conf->n_rx_queue; i++) {
2134
2135                 portid = rx_conf->rx_queue_list[i].port_id;
2136                 queueid = rx_conf->rx_queue_list[i].queue_id;
2137                 RTE_LOG(INFO, L3FWD,
2138                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2139                                 rte_lcore_id(), portid, queueid);
2140         }
2141
2142         /*
2143          * Init all condition variables (one per rx thread)
2144          */
2145         for (i = 0; i < rx_conf->n_rx_queue; i++)
2146                 lthread_cond_init(NULL, &rx_conf->ready[i], NULL);
2147
2148         worker_id = 0;
2149
2150         rx_conf->conf.cpu_id = sched_getcpu();
2151         rte_atomic16_inc(&rx_counter);
2152         while (1) {
2153
2154                 /*
2155                  * Read packet from RX queues
2156                  */
2157                 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2158                         portid = rx_conf->rx_queue_list[i].port_id;
2159                         queueid = rx_conf->rx_queue_list[i].queue_id;
2160
2161                         SET_CPU_BUSY(rx_conf, CPU_POLL);
2162                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2163                                 MAX_PKT_BURST);
2164                         SET_CPU_IDLE(rx_conf, CPU_POLL);
2165
2166                         if (nb_rx != 0) {
2167                                 worker_id = (worker_id + 1) % rx_conf->n_ring;
2168                                 old_len = len[worker_id];
2169
2170                                 SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2171                                 ret = rte_ring_sp_enqueue_burst(
2172                                                 rx_conf->ring[worker_id],
2173                                                 (void **) pkts_burst,
2174                                                 nb_rx, NULL);
2175
2176                                 new_len = old_len + ret;
2177
2178                                 if (new_len >= BURST_SIZE) {
2179                                         lthread_cond_signal(rx_conf->ready[worker_id]);
2180                                         new_len = 0;
2181                                 }
2182
2183                                 len[worker_id] = new_len;
2184
2185                                 if (unlikely(ret < nb_rx)) {
2186                                         uint32_t k;
2187
2188                                         for (k = ret; k < nb_rx; k++) {
2189                                                 struct rte_mbuf *m = pkts_burst[k];
2190
2191                                                 rte_pktmbuf_free(m);
2192                                         }
2193                                 }
2194                                 SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2195                         }
2196
2197                         lthread_yield();
2198                 }
2199         }
2200         return NULL;
2201 }
2202
2203 /*
2204  * Start scheduler with initial lthread on lcore
2205  *
2206  * This lthread loop spawns all rx and tx lthreads on master lcore
2207  */
2208
2209 static void *
2210 lthread_spawner(__rte_unused void *arg)
2211 {
2212         struct lthread *lt[MAX_THREAD];
2213         int i;
2214         int n_thread = 0;
2215
2216         printf("Entering lthread_spawner\n");
2217
2218         /*
2219          * Create producers (rx threads) on default lcore
2220          */
2221         for (i = 0; i < n_rx_thread; i++) {
2222                 rx_thread[i].conf.thread_id = i;
2223                 lthread_create(&lt[n_thread], -1, lthread_rx,
2224                                 (void *)&rx_thread[i]);
2225                 n_thread++;
2226         }
2227
2228         /*
2229          * Wait for all producers. Until some producers can be started on the same
2230          * scheduler as this lthread, yielding is required to let them to run and
2231          * prevent deadlock here.
2232          */
2233         while (rte_atomic16_read(&rx_counter) < n_rx_thread)
2234                 lthread_sleep(100000);
2235
2236         /*
2237          * Create consumers (tx threads) on default lcore_id
2238          */
2239         for (i = 0; i < n_tx_thread; i++) {
2240                 tx_thread[i].conf.thread_id = i;
2241                 lthread_create(&lt[n_thread], -1, lthread_tx,
2242                                 (void *)&tx_thread[i]);
2243                 n_thread++;
2244         }
2245
2246         /*
2247          * Wait for all threads finished
2248          */
2249         for (i = 0; i < n_thread; i++)
2250                 lthread_join(lt[i], NULL);
2251
2252         return NULL;
2253 }
2254
2255 /*
2256  * Start master scheduler with initial lthread spawning rx and tx lthreads
2257  * (main_lthread_master).
2258  */
2259 static int
2260 lthread_master_spawner(__rte_unused void *arg) {
2261         struct lthread *lt;
2262         int lcore_id = rte_lcore_id();
2263
2264         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2265         lthread_create(&lt, -1, lthread_spawner, NULL);
2266         lthread_run();
2267
2268         return 0;
2269 }
2270
2271 /*
2272  * Start scheduler on lcore.
2273  */
2274 static int
2275 sched_spawner(__rte_unused void *arg) {
2276         struct lthread *lt;
2277         int lcore_id = rte_lcore_id();
2278
2279 #if (APP_CPU_LOAD)
2280         if (lcore_id == cpu_load_lcore_id) {
2281                 cpu_load_collector(arg);
2282                 return 0;
2283         }
2284 #endif /* APP_CPU_LOAD */
2285
2286         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2287         lthread_create(&lt, -1, lthread_null, NULL);
2288         lthread_run();
2289
2290         return 0;
2291 }
2292
2293 /* main processing loop */
2294 static int
2295 pthread_tx(void *dummy)
2296 {
2297         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2298         uint64_t prev_tsc, diff_tsc, cur_tsc;
2299         int nb_rx;
2300         uint16_t portid;
2301         struct thread_tx_conf *tx_conf;
2302
2303         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
2304                 US_PER_S * BURST_TX_DRAIN_US;
2305
2306         prev_tsc = 0;
2307
2308         tx_conf = (struct thread_tx_conf *)dummy;
2309
2310         RTE_LOG(INFO, L3FWD, "Entering main Tx loop on lcore %u\n", rte_lcore_id());
2311
2312         tx_conf->conf.cpu_id = sched_getcpu();
2313         rte_atomic16_inc(&tx_counter);
2314         while (1) {
2315
2316                 cur_tsc = rte_rdtsc();
2317
2318                 /*
2319                  * TX burst queue drain
2320                  */
2321                 diff_tsc = cur_tsc - prev_tsc;
2322                 if (unlikely(diff_tsc > drain_tsc)) {
2323
2324                         /*
2325                          * This could be optimized (use queueid instead of
2326                          * portid), but it is not called so often
2327                          */
2328                         SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2329                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2330                                 if (tx_conf->tx_mbufs[portid].len == 0)
2331                                         continue;
2332                                 send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2333                                 tx_conf->tx_mbufs[portid].len = 0;
2334                         }
2335                         SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2336
2337                         prev_tsc = cur_tsc;
2338                 }
2339
2340                 /*
2341                  * Read packet from ring
2342                  */
2343                 SET_CPU_BUSY(tx_conf, CPU_POLL);
2344                 nb_rx = rte_ring_sc_dequeue_burst(tx_conf->ring,
2345                                 (void **)pkts_burst, MAX_PKT_BURST, NULL);
2346                 SET_CPU_IDLE(tx_conf, CPU_POLL);
2347
2348                 if (unlikely(nb_rx == 0)) {
2349                         sched_yield();
2350                         continue;
2351                 }
2352
2353                 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2354                 portid = pkts_burst[0]->port;
2355                 process_burst(pkts_burst, nb_rx, portid);
2356                 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2357
2358         }
2359 }
2360
2361 static int
2362 pthread_rx(void *dummy)
2363 {
2364         int i;
2365         int worker_id;
2366         uint32_t n;
2367         uint32_t nb_rx;
2368         unsigned lcore_id;
2369         uint8_t queueid;
2370         uint16_t portid;
2371         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2372
2373         struct thread_rx_conf *rx_conf;
2374
2375         lcore_id = rte_lcore_id();
2376         rx_conf = (struct thread_rx_conf *)dummy;
2377
2378         if (rx_conf->n_rx_queue == 0) {
2379                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
2380                 return 0;
2381         }
2382
2383         RTE_LOG(INFO, L3FWD, "entering main rx loop on lcore %u\n", lcore_id);
2384
2385         for (i = 0; i < rx_conf->n_rx_queue; i++) {
2386
2387                 portid = rx_conf->rx_queue_list[i].port_id;
2388                 queueid = rx_conf->rx_queue_list[i].queue_id;
2389                 RTE_LOG(INFO, L3FWD,
2390                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2391                                 lcore_id, portid, queueid);
2392         }
2393
2394         worker_id = 0;
2395         rx_conf->conf.cpu_id = sched_getcpu();
2396         rte_atomic16_inc(&rx_counter);
2397         while (1) {
2398
2399                 /*
2400                  * Read packet from RX queues
2401                  */
2402                 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2403                         portid = rx_conf->rx_queue_list[i].port_id;
2404                         queueid = rx_conf->rx_queue_list[i].queue_id;
2405
2406                         SET_CPU_BUSY(rx_conf, CPU_POLL);
2407                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2408                                 MAX_PKT_BURST);
2409                         SET_CPU_IDLE(rx_conf, CPU_POLL);
2410
2411                         if (nb_rx == 0) {
2412                                 sched_yield();
2413                                 continue;
2414                         }
2415
2416                         SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2417                         worker_id = (worker_id + 1) % rx_conf->n_ring;
2418                         n = rte_ring_sp_enqueue_burst(rx_conf->ring[worker_id],
2419                                         (void **)pkts_burst, nb_rx, NULL);
2420
2421                         if (unlikely(n != nb_rx)) {
2422                                 uint32_t k;
2423
2424                                 for (k = n; k < nb_rx; k++) {
2425                                         struct rte_mbuf *m = pkts_burst[k];
2426
2427                                         rte_pktmbuf_free(m);
2428                                 }
2429                         }
2430
2431                         SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2432
2433                 }
2434         }
2435 }
2436
2437 /*
2438  * P-Thread spawner.
2439  */
2440 static int
2441 pthread_run(__rte_unused void *arg) {
2442         int lcore_id = rte_lcore_id();
2443         int i;
2444
2445         for (i = 0; i < n_rx_thread; i++)
2446                 if (rx_thread[i].conf.lcore_id == lcore_id) {
2447                         printf("Start rx thread on %d...\n", lcore_id);
2448                         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2449                         RTE_PER_LCORE(lcore_conf)->data = (void *)&rx_thread[i];
2450                         pthread_rx((void *)&rx_thread[i]);
2451                         return 0;
2452                 }
2453
2454         for (i = 0; i < n_tx_thread; i++)
2455                 if (tx_thread[i].conf.lcore_id == lcore_id) {
2456                         printf("Start tx thread on %d...\n", lcore_id);
2457                         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2458                         RTE_PER_LCORE(lcore_conf)->data = (void *)&tx_thread[i];
2459                         pthread_tx((void *)&tx_thread[i]);
2460                         return 0;
2461                 }
2462
2463 #if (APP_CPU_LOAD)
2464         if (lcore_id == cpu_load_lcore_id)
2465                 cpu_load_collector(arg);
2466 #endif /* APP_CPU_LOAD */
2467
2468         return 0;
2469 }
2470
2471 static int
2472 check_lcore_params(void)
2473 {
2474         uint8_t queue, lcore;
2475         uint16_t i;
2476         int socketid;
2477
2478         for (i = 0; i < nb_rx_thread_params; ++i) {
2479                 queue = rx_thread_params[i].queue_id;
2480                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
2481                         printf("invalid queue number: %hhu\n", queue);
2482                         return -1;
2483                 }
2484                 lcore = rx_thread_params[i].lcore_id;
2485                 if (!rte_lcore_is_enabled(lcore)) {
2486                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
2487                         return -1;
2488                 }
2489                 socketid = rte_lcore_to_socket_id(lcore);
2490                 if ((socketid != 0) && (numa_on == 0))
2491                         printf("warning: lcore %hhu is on socket %d with numa off\n",
2492                                 lcore, socketid);
2493         }
2494         return 0;
2495 }
2496
2497 static int
2498 check_port_config(void)
2499 {
2500         unsigned portid;
2501         uint16_t i;
2502
2503         for (i = 0; i < nb_rx_thread_params; ++i) {
2504                 portid = rx_thread_params[i].port_id;
2505                 if ((enabled_port_mask & (1 << portid)) == 0) {
2506                         printf("port %u is not enabled in port mask\n", portid);
2507                         return -1;
2508                 }
2509                 if (!rte_eth_dev_is_valid_port(portid)) {
2510                         printf("port %u is not present on the board\n", portid);
2511                         return -1;
2512                 }
2513         }
2514         return 0;
2515 }
2516
2517 static uint8_t
2518 get_port_n_rx_queues(const uint16_t port)
2519 {
2520         int queue = -1;
2521         uint16_t i;
2522
2523         for (i = 0; i < nb_rx_thread_params; ++i)
2524                 if (rx_thread_params[i].port_id == port &&
2525                                 rx_thread_params[i].queue_id > queue)
2526                         queue = rx_thread_params[i].queue_id;
2527
2528         return (uint8_t)(++queue);
2529 }
2530
2531 static int
2532 init_rx_rings(void)
2533 {
2534         unsigned socket_io;
2535         struct thread_rx_conf *rx_conf;
2536         struct thread_tx_conf *tx_conf;
2537         unsigned rx_thread_id, tx_thread_id;
2538         char name[256];
2539         struct rte_ring *ring = NULL;
2540
2541         for (tx_thread_id = 0; tx_thread_id < n_tx_thread; tx_thread_id++) {
2542
2543                 tx_conf = &tx_thread[tx_thread_id];
2544
2545                 printf("Connecting tx-thread %d with rx-thread %d\n", tx_thread_id,
2546                                 tx_conf->conf.thread_id);
2547
2548                 rx_thread_id = tx_conf->conf.thread_id;
2549                 if (rx_thread_id > n_tx_thread) {
2550                         printf("connection from tx-thread %u to rx-thread %u fails "
2551                                         "(rx-thread not defined)\n", tx_thread_id, rx_thread_id);
2552                         return -1;
2553                 }
2554
2555                 rx_conf = &rx_thread[rx_thread_id];
2556                 socket_io = rte_lcore_to_socket_id(rx_conf->conf.lcore_id);
2557
2558                 snprintf(name, sizeof(name), "app_ring_s%u_rx%u_tx%u",
2559                                 socket_io, rx_thread_id, tx_thread_id);
2560
2561                 ring = rte_ring_create(name, 1024 * 4, socket_io,
2562                                 RING_F_SP_ENQ | RING_F_SC_DEQ);
2563
2564                 if (ring == NULL) {
2565                         rte_panic("Cannot create ring to connect rx-thread %u "
2566                                         "with tx-thread %u\n", rx_thread_id, tx_thread_id);
2567                 }
2568
2569                 rx_conf->ring[rx_conf->n_ring] = ring;
2570
2571                 tx_conf->ring = ring;
2572                 tx_conf->ready = &rx_conf->ready[rx_conf->n_ring];
2573
2574                 rx_conf->n_ring++;
2575         }
2576         return 0;
2577 }
2578
2579 static int
2580 init_rx_queues(void)
2581 {
2582         uint16_t i, nb_rx_queue;
2583         uint8_t thread;
2584
2585         n_rx_thread = 0;
2586
2587         for (i = 0; i < nb_rx_thread_params; ++i) {
2588                 thread = rx_thread_params[i].thread_id;
2589                 nb_rx_queue = rx_thread[thread].n_rx_queue;
2590
2591                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
2592                         printf("error: too many queues (%u) for thread: %u\n",
2593                                 (unsigned)nb_rx_queue + 1, (unsigned)thread);
2594                         return -1;
2595                 }
2596
2597                 rx_thread[thread].conf.thread_id = thread;
2598                 rx_thread[thread].conf.lcore_id = rx_thread_params[i].lcore_id;
2599                 rx_thread[thread].rx_queue_list[nb_rx_queue].port_id =
2600                         rx_thread_params[i].port_id;
2601                 rx_thread[thread].rx_queue_list[nb_rx_queue].queue_id =
2602                         rx_thread_params[i].queue_id;
2603                 rx_thread[thread].n_rx_queue++;
2604
2605                 if (thread >= n_rx_thread)
2606                         n_rx_thread = thread + 1;
2607
2608         }
2609         return 0;
2610 }
2611
2612 static int
2613 init_tx_threads(void)
2614 {
2615         int i;
2616
2617         n_tx_thread = 0;
2618         for (i = 0; i < nb_tx_thread_params; ++i) {
2619                 tx_thread[n_tx_thread].conf.thread_id = tx_thread_params[i].thread_id;
2620                 tx_thread[n_tx_thread].conf.lcore_id = tx_thread_params[i].lcore_id;
2621                 n_tx_thread++;
2622         }
2623         return 0;
2624 }
2625
2626 /* display usage */
2627 static void
2628 print_usage(const char *prgname)
2629 {
2630         printf("%s [EAL options] -- -p PORTMASK -P"
2631                 "  [--rx (port,queue,lcore,thread)[,(port,queue,lcore,thread]]"
2632                 "  [--tx (lcore,thread)[,(lcore,thread]]"
2633                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
2634                 "  [--parse-ptype]\n\n"
2635                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
2636                 "  -P : enable promiscuous mode\n"
2637                 "  --rx (port,queue,lcore,thread): rx queues configuration\n"
2638                 "  --tx (lcore,thread): tx threads configuration\n"
2639                 "  --stat-lcore LCORE: use lcore for stat collector\n"
2640                 "  --eth-dest=X,MM:MM:MM:MM:MM:MM: optional, ethernet destination for port X\n"
2641                 "  --no-numa: optional, disable numa awareness\n"
2642                 "  --ipv6: optional, specify it if running ipv6 packets\n"
2643                 "  --enable-jumbo: enable jumbo frame"
2644                 " which max packet len is PKTLEN in decimal (64-9600)\n"
2645                 "  --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n"
2646                 "  --no-lthreads: turn off lthread model\n"
2647                 "  --parse-ptype: set to use software to analyze packet type\n\n",
2648                 prgname);
2649 }
2650
2651 static int parse_max_pkt_len(const char *pktlen)
2652 {
2653         char *end = NULL;
2654         unsigned long len;
2655
2656         /* parse decimal string */
2657         len = strtoul(pktlen, &end, 10);
2658         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
2659                 return -1;
2660
2661         if (len == 0)
2662                 return -1;
2663
2664         return len;
2665 }
2666
2667 static int
2668 parse_portmask(const char *portmask)
2669 {
2670         char *end = NULL;
2671         unsigned long pm;
2672
2673         /* parse hexadecimal string */
2674         pm = strtoul(portmask, &end, 16);
2675         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
2676                 return -1;
2677
2678         if (pm == 0)
2679                 return -1;
2680
2681         return pm;
2682 }
2683
2684 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2685 static int
2686 parse_hash_entry_number(const char *hash_entry_num)
2687 {
2688         char *end = NULL;
2689         unsigned long hash_en;
2690
2691         /* parse hexadecimal string */
2692         hash_en = strtoul(hash_entry_num, &end, 16);
2693         if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
2694                 return -1;
2695
2696         if (hash_en == 0)
2697                 return -1;
2698
2699         return hash_en;
2700 }
2701 #endif
2702
2703 static int
2704 parse_rx_config(const char *q_arg)
2705 {
2706         char s[256];
2707         const char *p, *p0 = q_arg;
2708         char *end;
2709         enum fieldnames {
2710                 FLD_PORT = 0,
2711                 FLD_QUEUE,
2712                 FLD_LCORE,
2713                 FLD_THREAD,
2714                 _NUM_FLD
2715         };
2716         unsigned long int_fld[_NUM_FLD];
2717         char *str_fld[_NUM_FLD];
2718         int i;
2719         unsigned size;
2720
2721         nb_rx_thread_params = 0;
2722
2723         while ((p = strchr(p0, '(')) != NULL) {
2724                 ++p;
2725                 p0 = strchr(p, ')');
2726                 if (p0 == NULL)
2727                         return -1;
2728
2729                 size = p0 - p;
2730                 if (size >= sizeof(s))
2731                         return -1;
2732
2733                 snprintf(s, sizeof(s), "%.*s", size, p);
2734                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2735                         return -1;
2736                 for (i = 0; i < _NUM_FLD; i++) {
2737                         errno = 0;
2738                         int_fld[i] = strtoul(str_fld[i], &end, 0);
2739                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2740                                 return -1;
2741                 }
2742                 if (nb_rx_thread_params >= MAX_LCORE_PARAMS) {
2743                         printf("exceeded max number of rx params: %hu\n",
2744                                         nb_rx_thread_params);
2745                         return -1;
2746                 }
2747                 rx_thread_params_array[nb_rx_thread_params].port_id =
2748                                 int_fld[FLD_PORT];
2749                 rx_thread_params_array[nb_rx_thread_params].queue_id =
2750                                 (uint8_t)int_fld[FLD_QUEUE];
2751                 rx_thread_params_array[nb_rx_thread_params].lcore_id =
2752                                 (uint8_t)int_fld[FLD_LCORE];
2753                 rx_thread_params_array[nb_rx_thread_params].thread_id =
2754                                 (uint8_t)int_fld[FLD_THREAD];
2755                 ++nb_rx_thread_params;
2756         }
2757         rx_thread_params = rx_thread_params_array;
2758         return 0;
2759 }
2760
2761 static int
2762 parse_tx_config(const char *q_arg)
2763 {
2764         char s[256];
2765         const char *p, *p0 = q_arg;
2766         char *end;
2767         enum fieldnames {
2768                 FLD_LCORE = 0,
2769                 FLD_THREAD,
2770                 _NUM_FLD
2771         };
2772         unsigned long int_fld[_NUM_FLD];
2773         char *str_fld[_NUM_FLD];
2774         int i;
2775         unsigned size;
2776
2777         nb_tx_thread_params = 0;
2778
2779         while ((p = strchr(p0, '(')) != NULL) {
2780                 ++p;
2781                 p0 = strchr(p, ')');
2782                 if (p0 == NULL)
2783                         return -1;
2784
2785                 size = p0 - p;
2786                 if (size >= sizeof(s))
2787                         return -1;
2788
2789                 snprintf(s, sizeof(s), "%.*s", size, p);
2790                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2791                         return -1;
2792                 for (i = 0; i < _NUM_FLD; i++) {
2793                         errno = 0;
2794                         int_fld[i] = strtoul(str_fld[i], &end, 0);
2795                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2796                                 return -1;
2797                 }
2798                 if (nb_tx_thread_params >= MAX_LCORE_PARAMS) {
2799                         printf("exceeded max number of tx params: %hu\n",
2800                                 nb_tx_thread_params);
2801                         return -1;
2802                 }
2803                 tx_thread_params_array[nb_tx_thread_params].lcore_id =
2804                                 (uint8_t)int_fld[FLD_LCORE];
2805                 tx_thread_params_array[nb_tx_thread_params].thread_id =
2806                                 (uint8_t)int_fld[FLD_THREAD];
2807                 ++nb_tx_thread_params;
2808         }
2809         tx_thread_params = tx_thread_params_array;
2810
2811         return 0;
2812 }
2813
2814 #if (APP_CPU_LOAD > 0)
2815 static int
2816 parse_stat_lcore(const char *stat_lcore)
2817 {
2818         char *end = NULL;
2819         unsigned long lcore_id;
2820
2821         lcore_id = strtoul(stat_lcore, &end, 10);
2822         if ((stat_lcore[0] == '\0') || (end == NULL) || (*end != '\0'))
2823                 return -1;
2824
2825         return lcore_id;
2826 }
2827 #endif
2828
2829 static void
2830 parse_eth_dest(const char *optarg)
2831 {
2832         uint16_t portid;
2833         char *port_end;
2834         uint8_t c, *dest, peer_addr[6];
2835
2836         errno = 0;
2837         portid = strtoul(optarg, &port_end, 10);
2838         if (errno != 0 || port_end == optarg || *port_end++ != ',')
2839                 rte_exit(EXIT_FAILURE,
2840                 "Invalid eth-dest: %s", optarg);
2841         if (portid >= RTE_MAX_ETHPORTS)
2842                 rte_exit(EXIT_FAILURE,
2843                 "eth-dest: port %d >= RTE_MAX_ETHPORTS(%d)\n",
2844                 portid, RTE_MAX_ETHPORTS);
2845
2846         if (cmdline_parse_etheraddr(NULL, port_end,
2847                 &peer_addr, sizeof(peer_addr)) < 0)
2848                 rte_exit(EXIT_FAILURE,
2849                 "Invalid ethernet address: %s\n",
2850                 port_end);
2851         dest = (uint8_t *)&dest_eth_addr[portid];
2852         for (c = 0; c < 6; c++)
2853                 dest[c] = peer_addr[c];
2854         *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
2855 }
2856
2857 #define CMD_LINE_OPT_RX_CONFIG "rx"
2858 #define CMD_LINE_OPT_TX_CONFIG "tx"
2859 #define CMD_LINE_OPT_STAT_LCORE "stat-lcore"
2860 #define CMD_LINE_OPT_ETH_DEST "eth-dest"
2861 #define CMD_LINE_OPT_NO_NUMA "no-numa"
2862 #define CMD_LINE_OPT_IPV6 "ipv6"
2863 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo"
2864 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num"
2865 #define CMD_LINE_OPT_NO_LTHREADS "no-lthreads"
2866 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
2867
2868 /* Parse the argument given in the command line of the application */
2869 static int
2870 parse_args(int argc, char **argv)
2871 {
2872         int opt, ret;
2873         char **argvopt;
2874         int option_index;
2875         char *prgname = argv[0];
2876         static struct option lgopts[] = {
2877                 {CMD_LINE_OPT_RX_CONFIG, 1, 0, 0},
2878                 {CMD_LINE_OPT_TX_CONFIG, 1, 0, 0},
2879                 {CMD_LINE_OPT_STAT_LCORE, 1, 0, 0},
2880                 {CMD_LINE_OPT_ETH_DEST, 1, 0, 0},
2881                 {CMD_LINE_OPT_NO_NUMA, 0, 0, 0},
2882                 {CMD_LINE_OPT_IPV6, 0, 0, 0},
2883                 {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, 0},
2884                 {CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, 0},
2885                 {CMD_LINE_OPT_NO_LTHREADS, 0, 0, 0},
2886                 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
2887                 {NULL, 0, 0, 0}
2888         };
2889
2890         argvopt = argv;
2891
2892         while ((opt = getopt_long(argc, argvopt, "p:P",
2893                                 lgopts, &option_index)) != EOF) {
2894
2895                 switch (opt) {
2896                 /* portmask */
2897                 case 'p':
2898                         enabled_port_mask = parse_portmask(optarg);
2899                         if (enabled_port_mask == 0) {
2900                                 printf("invalid portmask\n");
2901                                 print_usage(prgname);
2902                                 return -1;
2903                         }
2904                         break;
2905                 case 'P':
2906                         printf("Promiscuous mode selected\n");
2907                         promiscuous_on = 1;
2908                         break;
2909
2910                 /* long options */
2911                 case 0:
2912                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_RX_CONFIG,
2913                                 sizeof(CMD_LINE_OPT_RX_CONFIG))) {
2914                                 ret = parse_rx_config(optarg);
2915                                 if (ret) {
2916                                         printf("invalid rx-config\n");
2917                                         print_usage(prgname);
2918                                         return -1;
2919                                 }
2920                         }
2921
2922                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_TX_CONFIG,
2923                                 sizeof(CMD_LINE_OPT_TX_CONFIG))) {
2924                                 ret = parse_tx_config(optarg);
2925                                 if (ret) {
2926                                         printf("invalid tx-config\n");
2927                                         print_usage(prgname);
2928                                         return -1;
2929                                 }
2930                         }
2931
2932 #if (APP_CPU_LOAD > 0)
2933                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_STAT_LCORE,
2934                                         sizeof(CMD_LINE_OPT_STAT_LCORE))) {
2935                                 cpu_load_lcore_id = parse_stat_lcore(optarg);
2936                         }
2937 #endif
2938
2939                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ETH_DEST,
2940                                 sizeof(CMD_LINE_OPT_ETH_DEST)))
2941                                         parse_eth_dest(optarg);
2942
2943                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_NUMA,
2944                                 sizeof(CMD_LINE_OPT_NO_NUMA))) {
2945                                 printf("numa is disabled\n");
2946                                 numa_on = 0;
2947                         }
2948
2949 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2950                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_IPV6,
2951                                 sizeof(CMD_LINE_OPT_IPV6))) {
2952                                 printf("ipv6 is specified\n");
2953                                 ipv6 = 1;
2954                         }
2955 #endif
2956
2957                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_LTHREADS,
2958                                         sizeof(CMD_LINE_OPT_NO_LTHREADS))) {
2959                                 printf("l-threads model is disabled\n");
2960                                 lthreads_on = 0;
2961                         }
2962
2963                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_PARSE_PTYPE,
2964                                         sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
2965                                 printf("software packet type parsing enabled\n");
2966                                 parse_ptype_on = 1;
2967                         }
2968
2969                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ENABLE_JUMBO,
2970                                 sizeof(CMD_LINE_OPT_ENABLE_JUMBO))) {
2971                                 struct option lenopts = {"max-pkt-len", required_argument, 0,
2972                                                 0};
2973
2974                                 printf("jumbo frame is enabled - disabling simple TX path\n");
2975                                 port_conf.rxmode.offloads |=
2976                                                 DEV_RX_OFFLOAD_JUMBO_FRAME;
2977                                 port_conf.txmode.offloads |=
2978                                                 DEV_TX_OFFLOAD_MULTI_SEGS;
2979
2980                                 /* if no max-pkt-len set, use the default value
2981                                  * RTE_ETHER_MAX_LEN
2982                                  */
2983                                 if (0 == getopt_long(argc, argvopt, "", &lenopts,
2984                                                 &option_index)) {
2985
2986                                         ret = parse_max_pkt_len(optarg);
2987                                         if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)) {
2988                                                 printf("invalid packet length\n");
2989                                                 print_usage(prgname);
2990                                                 return -1;
2991                                         }
2992                                         port_conf.rxmode.max_rx_pkt_len = ret;
2993                                 }
2994                                 printf("set jumbo frame max packet length to %u\n",
2995                                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
2996                         }
2997 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2998                         if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_HASH_ENTRY_NUM,
2999                                 sizeof(CMD_LINE_OPT_HASH_ENTRY_NUM))) {
3000                                 ret = parse_hash_entry_number(optarg);
3001                                 if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
3002                                         hash_entry_number = ret;
3003                                 } else {
3004                                         printf("invalid hash entry number\n");
3005                                         print_usage(prgname);
3006                                         return -1;
3007                                 }
3008                         }
3009 #endif
3010                         break;
3011
3012                 default:
3013                         print_usage(prgname);
3014                         return -1;
3015                 }
3016         }
3017
3018         if (optind >= 0)
3019                 argv[optind-1] = prgname;
3020
3021         ret = optind-1;
3022         optind = 1; /* reset getopt lib */
3023         return ret;
3024 }
3025
3026 static void
3027 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
3028 {
3029         char buf[RTE_ETHER_ADDR_FMT_SIZE];
3030
3031         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
3032         printf("%s%s", name, buf);
3033 }
3034
3035 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
3036
3037 static void convert_ipv4_5tuple(struct ipv4_5tuple *key1,
3038                 union ipv4_5tuple_host *key2)
3039 {
3040         key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
3041         key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
3042         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3043         key2->port_src = rte_cpu_to_be_16(key1->port_src);
3044         key2->proto = key1->proto;
3045         key2->pad0 = 0;
3046         key2->pad1 = 0;
3047 }
3048
3049 static void convert_ipv6_5tuple(struct ipv6_5tuple *key1,
3050                 union ipv6_5tuple_host *key2)
3051 {
3052         uint32_t i;
3053
3054         for (i = 0; i < 16; i++) {
3055                 key2->ip_dst[i] = key1->ip_dst[i];
3056                 key2->ip_src[i] = key1->ip_src[i];
3057         }
3058         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3059         key2->port_src = rte_cpu_to_be_16(key1->port_src);
3060         key2->proto = key1->proto;
3061         key2->pad0 = 0;
3062         key2->pad1 = 0;
3063         key2->reserve = 0;
3064 }
3065
3066 #define BYTE_VALUE_MAX 256
3067 #define ALL_32_BITS 0xffffffff
3068 #define BIT_8_TO_15 0x0000ff00
3069 static inline void
3070 populate_ipv4_few_flow_into_table(const struct rte_hash *h)
3071 {
3072         uint32_t i;
3073         int32_t ret;
3074         uint32_t array_len = RTE_DIM(ipv4_l3fwd_route_array);
3075
3076         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3077         for (i = 0; i < array_len; i++) {
3078                 struct ipv4_l3fwd_route  entry;
3079                 union ipv4_5tuple_host newkey;
3080
3081                 entry = ipv4_l3fwd_route_array[i];
3082                 convert_ipv4_5tuple(&entry.key, &newkey);
3083                 ret = rte_hash_add_key(h, (void *)&newkey);
3084                 if (ret < 0) {
3085                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3086                                 " to the l3fwd hash.\n", i);
3087                 }
3088                 ipv4_l3fwd_out_if[ret] = entry.if_out;
3089         }
3090         printf("Hash: Adding 0x%" PRIx32 " keys\n", array_len);
3091 }
3092
3093 #define BIT_16_TO_23 0x00ff0000
3094 static inline void
3095 populate_ipv6_few_flow_into_table(const struct rte_hash *h)
3096 {
3097         uint32_t i;
3098         int32_t ret;
3099         uint32_t array_len = RTE_DIM(ipv6_l3fwd_route_array);
3100
3101         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3102         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3103         for (i = 0; i < array_len; i++) {
3104                 struct ipv6_l3fwd_route entry;
3105                 union ipv6_5tuple_host newkey;
3106
3107                 entry = ipv6_l3fwd_route_array[i];
3108                 convert_ipv6_5tuple(&entry.key, &newkey);
3109                 ret = rte_hash_add_key(h, (void *)&newkey);
3110                 if (ret < 0) {
3111                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3112                                 " to the l3fwd hash.\n", i);
3113                 }
3114                 ipv6_l3fwd_out_if[ret] = entry.if_out;
3115         }
3116         printf("Hash: Adding 0x%" PRIx32 "keys\n", array_len);
3117 }
3118
3119 #define NUMBER_PORT_USED 4
3120 static inline void
3121 populate_ipv4_many_flow_into_table(const struct rte_hash *h,
3122                 unsigned int nr_flow)
3123 {
3124         unsigned i;
3125
3126         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3127
3128         for (i = 0; i < nr_flow; i++) {
3129                 struct ipv4_l3fwd_route entry;
3130                 union ipv4_5tuple_host newkey;
3131                 uint8_t a = (uint8_t)((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3132                 uint8_t b = (uint8_t)(((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3133                                 BYTE_VALUE_MAX);
3134                 uint8_t c = (uint8_t)((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3135                                 BYTE_VALUE_MAX));
3136                 /* Create the ipv4 exact match flow */
3137                 memset(&entry, 0, sizeof(entry));
3138                 switch (i & (NUMBER_PORT_USED - 1)) {
3139                 case 0:
3140                         entry = ipv4_l3fwd_route_array[0];
3141                         entry.key.ip_dst = IPv4(101, c, b, a);
3142                         break;
3143                 case 1:
3144                         entry = ipv4_l3fwd_route_array[1];
3145                         entry.key.ip_dst = IPv4(201, c, b, a);
3146                         break;
3147                 case 2:
3148                         entry = ipv4_l3fwd_route_array[2];
3149                         entry.key.ip_dst = IPv4(111, c, b, a);
3150                         break;
3151                 case 3:
3152                         entry = ipv4_l3fwd_route_array[3];
3153                         entry.key.ip_dst = IPv4(211, c, b, a);
3154                         break;
3155                 };
3156                 convert_ipv4_5tuple(&entry.key, &newkey);
3157                 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3158
3159                 if (ret < 0)
3160                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3161
3162                 ipv4_l3fwd_out_if[ret] = (uint8_t)entry.if_out;
3163
3164         }
3165         printf("Hash: Adding 0x%x keys\n", nr_flow);
3166 }
3167
3168 static inline void
3169 populate_ipv6_many_flow_into_table(const struct rte_hash *h,
3170                 unsigned int nr_flow)
3171 {
3172         unsigned i;
3173
3174         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3175         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3176         for (i = 0; i < nr_flow; i++) {
3177                 struct ipv6_l3fwd_route entry;
3178                 union ipv6_5tuple_host newkey;
3179
3180                 uint8_t a = (uint8_t) ((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3181                 uint8_t b = (uint8_t) (((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3182                                 BYTE_VALUE_MAX);
3183                 uint8_t c = (uint8_t) ((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3184                                 BYTE_VALUE_MAX));
3185
3186                 /* Create the ipv6 exact match flow */
3187                 memset(&entry, 0, sizeof(entry));
3188                 switch (i & (NUMBER_PORT_USED - 1)) {
3189                 case 0:
3190                         entry = ipv6_l3fwd_route_array[0];
3191                         break;
3192                 case 1:
3193                         entry = ipv6_l3fwd_route_array[1];
3194                         break;
3195                 case 2:
3196                         entry = ipv6_l3fwd_route_array[2];
3197                         break;
3198                 case 3:
3199                         entry = ipv6_l3fwd_route_array[3];
3200                         break;
3201                 };
3202                 entry.key.ip_dst[13] = c;
3203                 entry.key.ip_dst[14] = b;
3204                 entry.key.ip_dst[15] = a;
3205                 convert_ipv6_5tuple(&entry.key, &newkey);
3206                 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3207
3208                 if (ret < 0)
3209                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3210
3211                 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
3212
3213         }
3214         printf("Hash: Adding 0x%x keys\n", nr_flow);
3215 }
3216
3217 static void
3218 setup_hash(int socketid)
3219 {
3220         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
3221                 .name = NULL,
3222                 .entries = L3FWD_HASH_ENTRIES,
3223                 .key_len = sizeof(union ipv4_5tuple_host),
3224                 .hash_func = ipv4_hash_crc,
3225                 .hash_func_init_val = 0,
3226         };
3227
3228         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
3229                 .name = NULL,
3230                 .entries = L3FWD_HASH_ENTRIES,
3231                 .key_len = sizeof(union ipv6_5tuple_host),
3232                 .hash_func = ipv6_hash_crc,
3233                 .hash_func_init_val = 0,
3234         };
3235
3236         char s[64];
3237
3238         /* create ipv4 hash */
3239         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
3240         ipv4_l3fwd_hash_params.name = s;
3241         ipv4_l3fwd_hash_params.socket_id = socketid;
3242         ipv4_l3fwd_lookup_struct[socketid] =
3243                         rte_hash_create(&ipv4_l3fwd_hash_params);
3244         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3245                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3246                                 "socket %d\n", socketid);
3247
3248         /* create ipv6 hash */
3249         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
3250         ipv6_l3fwd_hash_params.name = s;
3251         ipv6_l3fwd_hash_params.socket_id = socketid;
3252         ipv6_l3fwd_lookup_struct[socketid] =
3253                         rte_hash_create(&ipv6_l3fwd_hash_params);
3254         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3255                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3256                                 "socket %d\n", socketid);
3257
3258         if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
3259                 /* For testing hash matching with a large number of flows we
3260                  * generate millions of IP 5-tuples with an incremented dst
3261                  * address to initialize the hash table. */
3262                 if (ipv6 == 0) {
3263                         /* populate the ipv4 hash */
3264                         populate_ipv4_many_flow_into_table(
3265                                 ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
3266                 } else {
3267                         /* populate the ipv6 hash */
3268                         populate_ipv6_many_flow_into_table(
3269                                 ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
3270                 }
3271         } else {
3272                 /* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize
3273                  * the hash table */
3274                 if (ipv6 == 0) {
3275                         /* populate the ipv4 hash */
3276                         populate_ipv4_few_flow_into_table(
3277                                         ipv4_l3fwd_lookup_struct[socketid]);
3278                 } else {
3279                         /* populate the ipv6 hash */
3280                         populate_ipv6_few_flow_into_table(
3281                                         ipv6_l3fwd_lookup_struct[socketid]);
3282                 }
3283         }
3284 }
3285 #endif
3286
3287 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3288 static void
3289 setup_lpm(int socketid)
3290 {
3291         struct rte_lpm6_config config;
3292         struct rte_lpm_config lpm_ipv4_config;
3293         unsigned i;
3294         int ret;
3295         char s[64];
3296
3297         /* create the LPM table */
3298         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
3299         lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
3300         lpm_ipv4_config.number_tbl8s = 256;
3301         lpm_ipv4_config.flags = 0;
3302         ipv4_l3fwd_lookup_struct[socketid] =
3303                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
3304         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3305                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3306                                 " on socket %d\n", socketid);
3307
3308         /* populate the LPM table */
3309         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
3310
3311                 /* skip unused ports */
3312                 if ((1 << ipv4_l3fwd_route_array[i].if_out &
3313                                 enabled_port_mask) == 0)
3314                         continue;
3315
3316                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
3317                         ipv4_l3fwd_route_array[i].ip,
3318                         ipv4_l3fwd_route_array[i].depth,
3319                         ipv4_l3fwd_route_array[i].if_out);
3320
3321                 if (ret < 0) {
3322                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3323                                 "l3fwd LPM table on socket %d\n",
3324                                 i, socketid);
3325                 }
3326
3327                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
3328                         (unsigned)ipv4_l3fwd_route_array[i].ip,
3329                         ipv4_l3fwd_route_array[i].depth,
3330                         ipv4_l3fwd_route_array[i].if_out);
3331         }
3332
3333         /* create the LPM6 table */
3334         snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
3335
3336         config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
3337         config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
3338         config.flags = 0;
3339         ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
3340                                 &config);
3341         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3342                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3343                                 " on socket %d\n", socketid);
3344
3345         /* populate the LPM table */
3346         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
3347
3348                 /* skip unused ports */
3349                 if ((1 << ipv6_l3fwd_route_array[i].if_out &
3350                                 enabled_port_mask) == 0)
3351                         continue;
3352
3353                 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
3354                         ipv6_l3fwd_route_array[i].ip,
3355                         ipv6_l3fwd_route_array[i].depth,
3356                         ipv6_l3fwd_route_array[i].if_out);
3357
3358                 if (ret < 0) {
3359                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3360                                 "l3fwd LPM table on socket %d\n",
3361                                 i, socketid);
3362                 }
3363
3364                 printf("LPM: Adding route %s / %d (%d)\n",
3365                         "IPV6",
3366                         ipv6_l3fwd_route_array[i].depth,
3367                         ipv6_l3fwd_route_array[i].if_out);
3368         }
3369 }
3370 #endif
3371
3372 static int
3373 init_mem(unsigned nb_mbuf)
3374 {
3375         struct lcore_conf *qconf;
3376         int socketid;
3377         unsigned lcore_id;
3378         char s[64];
3379
3380         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3381                 if (rte_lcore_is_enabled(lcore_id) == 0)
3382                         continue;
3383
3384                 if (numa_on)
3385                         socketid = rte_lcore_to_socket_id(lcore_id);
3386                 else
3387                         socketid = 0;
3388
3389                 if (socketid >= NB_SOCKETS) {
3390                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
3391                                 socketid, lcore_id, NB_SOCKETS);
3392                 }
3393                 if (pktmbuf_pool[socketid] == NULL) {
3394                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
3395                         pktmbuf_pool[socketid] =
3396                                 rte_pktmbuf_pool_create(s, nb_mbuf,
3397                                         MEMPOOL_CACHE_SIZE, 0,
3398                                         RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
3399                         if (pktmbuf_pool[socketid] == NULL)
3400                                 rte_exit(EXIT_FAILURE,
3401                                                 "Cannot init mbuf pool on socket %d\n", socketid);
3402                         else
3403                                 printf("Allocated mbuf pool on socket %d\n", socketid);
3404
3405 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3406                         setup_lpm(socketid);
3407 #else
3408                         setup_hash(socketid);
3409 #endif
3410                 }
3411                 qconf = &lcore_conf[lcore_id];
3412                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
3413                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
3414         }
3415         return 0;
3416 }
3417
3418 /* Check the link status of all ports in up to 9s, and print them finally */
3419 static void
3420 check_all_ports_link_status(uint32_t port_mask)
3421 {
3422 #define CHECK_INTERVAL 100 /* 100ms */
3423 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
3424         uint16_t portid;
3425         uint8_t count, all_ports_up, print_flag = 0;
3426         struct rte_eth_link link;
3427
3428         printf("\nChecking link status");
3429         fflush(stdout);
3430         for (count = 0; count <= MAX_CHECK_TIME; count++) {
3431                 all_ports_up = 1;
3432                 RTE_ETH_FOREACH_DEV(portid) {
3433                         if ((port_mask & (1 << portid)) == 0)
3434                                 continue;
3435                         memset(&link, 0, sizeof(link));
3436                         rte_eth_link_get_nowait(portid, &link);
3437                         /* print link status if flag set */
3438                         if (print_flag == 1) {
3439                                 if (link.link_status)
3440                                         printf(
3441                                         "Port%d Link Up. Speed %u Mbps - %s\n",
3442                                                 portid, link.link_speed,
3443                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
3444                                         ("full-duplex") : ("half-duplex\n"));
3445                                 else
3446                                         printf("Port %d Link Down\n", portid);
3447                                 continue;
3448                         }
3449                         /* clear all_ports_up flag if any link down */
3450                         if (link.link_status == ETH_LINK_DOWN) {
3451                                 all_ports_up = 0;
3452                                 break;
3453                         }
3454                 }
3455                 /* after finally printing all link status, get out */
3456                 if (print_flag == 1)
3457                         break;
3458
3459                 if (all_ports_up == 0) {
3460                         printf(".");
3461                         fflush(stdout);
3462                         rte_delay_ms(CHECK_INTERVAL);
3463                 }
3464
3465                 /* set the print_flag if all ports up or timeout */
3466                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
3467                         print_flag = 1;
3468                         printf("done\n");
3469                 }
3470         }
3471 }
3472
3473 int
3474 main(int argc, char **argv)
3475 {
3476         struct rte_eth_dev_info dev_info;
3477         struct rte_eth_txconf *txconf;
3478         int ret;
3479         int i;
3480         unsigned nb_ports;
3481         uint16_t queueid, portid;
3482         unsigned lcore_id;
3483         uint32_t n_tx_queue, nb_lcores;
3484         uint8_t nb_rx_queue, queue, socketid;
3485
3486         /* init EAL */
3487         ret = rte_eal_init(argc, argv);
3488         if (ret < 0)
3489                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
3490         argc -= ret;
3491         argv += ret;
3492
3493         /* pre-init dst MACs for all ports to 02:00:00:00:00:xx */
3494         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
3495                 dest_eth_addr[portid] = RTE_ETHER_LOCAL_ADMIN_ADDR +
3496                                 ((uint64_t)portid << 40);
3497                 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
3498         }
3499
3500         /* parse application arguments (after the EAL ones) */
3501         ret = parse_args(argc, argv);
3502         if (ret < 0)
3503                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
3504
3505         if (check_lcore_params() < 0)
3506                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
3507
3508         printf("Initializing rx-queues...\n");
3509         ret = init_rx_queues();
3510         if (ret < 0)
3511                 rte_exit(EXIT_FAILURE, "init_rx_queues failed\n");
3512
3513         printf("Initializing tx-threads...\n");
3514         ret = init_tx_threads();
3515         if (ret < 0)
3516                 rte_exit(EXIT_FAILURE, "init_tx_threads failed\n");
3517
3518         printf("Initializing rings...\n");
3519         ret = init_rx_rings();
3520         if (ret < 0)
3521                 rte_exit(EXIT_FAILURE, "init_rx_rings failed\n");
3522
3523         nb_ports = rte_eth_dev_count_avail();
3524
3525         if (check_port_config() < 0)
3526                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
3527
3528         nb_lcores = rte_lcore_count();
3529
3530         /* initialize all ports */
3531         RTE_ETH_FOREACH_DEV(portid) {
3532                 struct rte_eth_conf local_port_conf = port_conf;
3533
3534                 /* skip ports that are not enabled */
3535                 if ((enabled_port_mask & (1 << portid)) == 0) {
3536                         printf("\nSkipping disabled port %d\n", portid);
3537                         continue;
3538                 }
3539
3540                 /* init port */
3541                 printf("Initializing port %d ... ", portid);
3542                 fflush(stdout);
3543
3544                 nb_rx_queue = get_port_n_rx_queues(portid);
3545                 n_tx_queue = nb_lcores;
3546                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
3547                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
3548                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
3549                         nb_rx_queue, (unsigned)n_tx_queue);
3550                 rte_eth_dev_info_get(portid, &dev_info);
3551                 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
3552                         local_port_conf.txmode.offloads |=
3553                                 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
3554
3555                 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
3556                         dev_info.flow_type_rss_offloads;
3557                 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
3558                                 port_conf.rx_adv_conf.rss_conf.rss_hf) {
3559                         printf("Port %u modified RSS hash function based on hardware support,"
3560                                 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
3561                                 portid,
3562                                 port_conf.rx_adv_conf.rss_conf.rss_hf,
3563                                 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
3564                 }
3565
3566                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
3567                                         (uint16_t)n_tx_queue, &local_port_conf);
3568                 if (ret < 0)
3569                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
3570                                 ret, portid);
3571
3572                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
3573                                                        &nb_txd);
3574                 if (ret < 0)
3575                         rte_exit(EXIT_FAILURE,
3576                                  "rte_eth_dev_adjust_nb_rx_tx_desc: err=%d, port=%d\n",
3577                                  ret, portid);
3578
3579                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
3580                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
3581                 printf(", ");
3582                 print_ethaddr("Destination:",
3583                         (const struct rte_ether_addr *)&dest_eth_addr[portid]);
3584                 printf(", ");
3585
3586                 /*
3587                  * prepare src MACs for each port.
3588                  */
3589                 rte_ether_addr_copy(&ports_eth_addr[portid],
3590                         (struct rte_ether_addr *)(val_eth + portid) + 1);
3591
3592                 /* init memory */
3593                 ret = init_mem(NB_MBUF);
3594                 if (ret < 0)
3595                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
3596
3597                 /* init one TX queue per couple (lcore,port) */
3598                 queueid = 0;
3599                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3600                         if (rte_lcore_is_enabled(lcore_id) == 0)
3601                                 continue;
3602
3603                         if (numa_on)
3604                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3605                         else
3606                                 socketid = 0;
3607
3608                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
3609                         fflush(stdout);
3610
3611                         txconf = &dev_info.default_txconf;
3612                         txconf->offloads = local_port_conf.txmode.offloads;
3613                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
3614                                                      socketid, txconf);
3615                         if (ret < 0)
3616                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
3617                                         "port=%d\n", ret, portid);
3618
3619                         tx_thread[lcore_id].tx_queue_id[portid] = queueid;
3620                         queueid++;
3621                 }
3622                 printf("\n");
3623         }
3624
3625         for (i = 0; i < n_rx_thread; i++) {
3626                 lcore_id = rx_thread[i].conf.lcore_id;
3627
3628                 if (rte_lcore_is_enabled(lcore_id) == 0) {
3629                         rte_exit(EXIT_FAILURE,
3630                                         "Cannot start Rx thread on lcore %u: lcore disabled\n",
3631                                         lcore_id
3632                                 );
3633                 }
3634
3635                 printf("\nInitializing rx queues for Rx thread %d on lcore %u ... ",
3636                                 i, lcore_id);
3637                 fflush(stdout);
3638
3639                 /* init RX queues */
3640                 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3641                         struct rte_eth_dev *dev;
3642                         struct rte_eth_conf *conf;
3643                         struct rte_eth_rxconf rxq_conf;
3644
3645                         portid = rx_thread[i].rx_queue_list[queue].port_id;
3646                         queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3647                         dev = &rte_eth_devices[portid];
3648                         conf = &dev->data->dev_conf;
3649
3650                         if (numa_on)
3651                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3652                         else
3653                                 socketid = 0;
3654
3655                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
3656                         fflush(stdout);
3657
3658                         rte_eth_dev_info_get(portid, &dev_info);
3659                         rxq_conf = dev_info.default_rxconf;
3660                         rxq_conf.offloads = conf->rxmode.offloads;
3661                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
3662                                         socketid,
3663                                         &rxq_conf,
3664                                         pktmbuf_pool[socketid]);
3665                         if (ret < 0)
3666                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, "
3667                                                 "port=%d\n", ret, portid);
3668                 }
3669         }
3670
3671         printf("\n");
3672
3673         /* start ports */
3674         RTE_ETH_FOREACH_DEV(portid) {
3675                 if ((enabled_port_mask & (1 << portid)) == 0)
3676                         continue;
3677
3678                 /* Start device */
3679                 ret = rte_eth_dev_start(portid);
3680                 if (ret < 0)
3681                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
3682                                 ret, portid);
3683
3684                 /*
3685                  * If enabled, put device in promiscuous mode.
3686                  * This allows IO forwarding mode to forward packets
3687                  * to itself through 2 cross-connected  ports of the
3688                  * target machine.
3689                  */
3690                 if (promiscuous_on)
3691                         rte_eth_promiscuous_enable(portid);
3692         }
3693
3694         for (i = 0; i < n_rx_thread; i++) {
3695                 lcore_id = rx_thread[i].conf.lcore_id;
3696                 if (rte_lcore_is_enabled(lcore_id) == 0)
3697                         continue;
3698
3699                 /* check if hw packet type is supported */
3700                 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3701                         portid = rx_thread[i].rx_queue_list[queue].port_id;
3702                         queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3703
3704                         if (parse_ptype_on) {
3705                                 if (!rte_eth_add_rx_callback(portid, queueid,
3706                                                 cb_parse_ptype, NULL))
3707                                         rte_exit(EXIT_FAILURE,
3708                                                 "Failed to add rx callback: "
3709                                                 "port=%d\n", portid);
3710                         } else if (!check_ptype(portid))
3711                                 rte_exit(EXIT_FAILURE,
3712                                         "Port %d cannot parse packet type.\n\n"
3713                                         "Please add --parse-ptype to use sw "
3714                                         "packet type analyzer.\n\n",
3715                                         portid);
3716                 }
3717         }
3718
3719         check_all_ports_link_status(enabled_port_mask);
3720
3721         if (lthreads_on) {
3722                 printf("Starting L-Threading Model\n");
3723
3724 #if (APP_CPU_LOAD > 0)
3725                 if (cpu_load_lcore_id > 0)
3726                         /* Use one lcore for cpu load collector */
3727                         nb_lcores--;
3728 #endif
3729
3730                 lthread_num_schedulers_set(nb_lcores);
3731                 rte_eal_mp_remote_launch(sched_spawner, NULL, SKIP_MASTER);
3732                 lthread_master_spawner(NULL);
3733
3734         } else {
3735                 printf("Starting P-Threading Model\n");
3736                 /* launch per-lcore init on every lcore */
3737                 rte_eal_mp_remote_launch(pthread_run, NULL, CALL_MASTER);
3738                 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
3739                         if (rte_eal_wait_lcore(lcore_id) < 0)
3740                                 return -1;
3741                 }
3742         }
3743
3744         return 0;
3745 }