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

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2016 Intel Corporation
   3  */
   4
   5 #ifndef _GNU_SOURCE
   6 #define _GNU_SOURCE
   7 #endif
   8
   9 #include <stdio.h>
  10 #include <stdlib.h>
  11 #include <stdint.h>
  12 #include <inttypes.h>
  13 #include <sys/types.h>
  14 #include <string.h>
  15 #include <sys/queue.h>
  16 #include <stdarg.h>
  17 #include <errno.h>
  18 #include <getopt.h>
  19 #include <sched.h>
  20
  21 #include <rte_common.h>
  22 #include <rte_vect.h>
  23 #include <rte_byteorder.h>
  24 #include <rte_log.h>
  25 #include <rte_memory.h>
  26 #include <rte_memcpy.h>
  27 #include <rte_eal.h>
  28 #include <rte_launch.h>
  29 #include <rte_cycles.h>
  30 #include <rte_prefetch.h>
  31 #include <rte_lcore.h>
  32 #include <rte_per_lcore.h>
  33 #include <rte_branch_prediction.h>
  34 #include <rte_interrupts.h>
  35 #include <rte_random.h>
  36 #include <rte_debug.h>
  37 #include <rte_ether.h>
  38 #include <rte_ethdev.h>
  39 #include <rte_ring.h>
  40 #include <rte_mempool.h>
  41 #include <rte_mbuf.h>
  42 #include <rte_ip.h>
  43 #include <rte_tcp.h>
  44 #include <rte_udp.h>
  45 #include <rte_string_fns.h>
  46 #include <rte_pause.h>
  47 #include <rte_timer.h>
  48
  49 #include <cmdline_parse.h>
  50 #include <cmdline_parse_etheraddr.h>
  51
  52 #include <lthread_api.h>
  53
  54 #define APP_LOOKUP_EXACT_MATCH          0
  55 #define APP_LOOKUP_LPM                  1
  56 #define DO_RFC_1812_CHECKS
  57
  58 /* Enable cpu-load stats 0-off, 1-on */
  59 #define APP_CPU_LOAD                 1
  60
  61 #ifndef APP_LOOKUP_METHOD
  62 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  63 #endif
  64
  65 #ifndef __GLIBC__ /* sched_getcpu() is glibc specific */
  66 #define sched_getcpu() rte_lcore_id()
  67 #endif
  68
  69 static int
  70 check_ptype(int portid)
  71 {
  72         int i, ret;
  73         int ipv4 = 0, ipv6 = 0;
  74
  75         ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK, NULL,
  76                         0);
  77         if (ret <= 0)
  78                 return 0;
  79
  80         uint32_t ptypes[ret];
  81
  82         ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK,
  83                         ptypes, ret);
  84         for (i = 0; i < ret; ++i) {
  85                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
  86                         ipv4 = 1;
  87                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
  88                         ipv6 = 1;
  89         }
  90
  91         if (ipv4 && ipv6)
  92                 return 1;
  93
  94         return 0;
  95 }
  96
  97 static inline void
  98 parse_ptype(struct rte_mbuf *m)
  99 {
 100         struct rte_ether_hdr *eth_hdr;
 101         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 102         uint16_t ether_type;
 103
 104         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 105         ether_type = eth_hdr->ether_type;
 106         if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
 107                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 108         else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6))
 109                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 110
 111         m->packet_type = packet_type;
 112 }
 113
 114 static uint16_t
 115 cb_parse_ptype(__rte_unused uint16_t port, __rte_unused uint16_t queue,
 116                 struct rte_mbuf *pkts[], uint16_t nb_pkts,
 117                 __rte_unused uint16_t max_pkts, __rte_unused void *user_param)
 118 {
 119         unsigned int i;
 120
 121         for (i = 0; i < nb_pkts; i++)
 122                 parse_ptype(pkts[i]);
 123
 124         return nb_pkts;
 125 }
 126
 127 /*
 128  *  When set to zero, simple forwaring path is eanbled.
 129  *  When set to one, optimized forwarding path is enabled.
 130  *  Note that LPM optimisation path uses SSE4.1 instructions.
 131  */
 132 #define ENABLE_MULTI_BUFFER_OPTIMIZE    1
 133
 134 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 135 #include <rte_hash.h>
 136 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 137 #include <rte_lpm.h>
 138 #include <rte_lpm6.h>
 139 #else
 140 #error "APP_LOOKUP_METHOD set to incorrect value"
 141 #endif
 142
 143 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
 144
 145 #define MAX_JUMBO_PKT_LEN  9600
 146
 147 #define IPV6_ADDR_LEN 16
 148
 149 #define MEMPOOL_CACHE_SIZE 256
 150
 151 /*
 152  * This expression is used to calculate the number of mbufs needed depending on
 153  * user input, taking into account memory for rx and tx hardware rings, cache
 154  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 155  * NB_MBUF never goes below a minimum value of 8192
 156  */
 157
 158 #define NB_MBUF RTE_MAX(\
 159                 (nb_ports*nb_rx_queue*nb_rxd +      \
 160                 nb_ports*nb_lcores*MAX_PKT_BURST +  \
 161                 nb_ports*n_tx_queue*nb_txd +        \
 162                 nb_lcores*MEMPOOL_CACHE_SIZE),      \
 163                 (unsigned)8192)
 164
 165 #define MAX_PKT_BURST     32
 166 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 167
 168 /*
 169  * Try to avoid TX buffering if we have at least MAX_TX_BURST packets to send.
 170  */
 171 #define MAX_TX_BURST  (MAX_PKT_BURST / 2)
 172 #define BURST_SIZE    MAX_TX_BURST
 173
 174 #define NB_SOCKETS 8
 175
 176 /* Configure how many packets ahead to prefetch, when reading packets */
 177 #define PREFETCH_OFFSET 3
 178
 179 /* Used to mark destination port as 'invalid'. */
 180 #define BAD_PORT        ((uint16_t)-1)
 181
 182 #define FWDSTEP 4
 183
 184 /*
 185  * Configurable number of RX/TX ring descriptors
 186  */
 187 #define RTE_TEST_RX_DESC_DEFAULT 1024
 188 #define RTE_TEST_TX_DESC_DEFAULT 1024
 189 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 190 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 191
 192 /* ethernet addresses of ports */
 193 static uint64_t dest_eth_addr[RTE_MAX_ETHPORTS];
 194 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 195
 196 static xmm_t val_eth[RTE_MAX_ETHPORTS];
 197
 198 /* replace first 12B of the ethernet header. */
 199 #define MASK_ETH 0x3f
 200
 201 /* mask of enabled ports */
 202 static uint32_t enabled_port_mask;
 203 static int promiscuous_on; /**< Set in promiscuous mode off by default. */
 204 static int numa_on = 1;    /**< NUMA is enabled by default. */
 205 static int parse_ptype_on;
 206
 207 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 208 static int ipv6;           /**< ipv6 is false by default. */
 209 #endif
 210
 211 #if (APP_CPU_LOAD == 1)
 212
 213 #define MAX_CPU RTE_MAX_LCORE
 214 #define CPU_LOAD_TIMEOUT_US (5 * 1000 * 1000)  /**< Timeout for collecting 5s */
 215
 216 #define CPU_PROCESS     0
 217 #define CPU_POLL        1
 218 #define MAX_CPU_COUNTER 2
 219
 220 struct cpu_load {
 221         uint16_t       n_cpu;
 222         uint64_t       counter;
 223         uint64_t       hits[MAX_CPU_COUNTER][MAX_CPU];
 224 } __rte_cache_aligned;
 225
 226 static struct cpu_load cpu_load;
 227 static int cpu_load_lcore_id = -1;
 228
 229 #define SET_CPU_BUSY(thread, counter) \
 230                 thread->conf.busy[counter] = 1
 231
 232 #define SET_CPU_IDLE(thread, counter) \
 233                 thread->conf.busy[counter] = 0
 234
 235 #define IS_CPU_BUSY(thread, counter) \
 236                 (thread->conf.busy[counter] > 0)
 237
 238 #else
 239
 240 #define SET_CPU_BUSY(thread, counter)
 241 #define SET_CPU_IDLE(thread, counter)
 242 #define IS_CPU_BUSY(thread, counter) 0
 243
 244 #endif
 245
 246 struct mbuf_table {
 247         uint16_t len;
 248         struct rte_mbuf *m_table[MAX_PKT_BURST];
 249 };
 250
 251 struct lcore_rx_queue {
 252         uint16_t port_id;
 253         uint8_t queue_id;
 254 } __rte_cache_aligned;
 255
 256 #define MAX_RX_QUEUE_PER_LCORE 16
 257 #define MAX_TX_QUEUE_PER_PORT  RTE_MAX_ETHPORTS
 258 #define MAX_RX_QUEUE_PER_PORT  128
 259
 260 #define MAX_LCORE_PARAMS       1024
 261 struct rx_thread_params {
 262         uint16_t port_id;
 263         uint8_t queue_id;
 264         uint8_t lcore_id;
 265         uint8_t thread_id;
 266 } __rte_cache_aligned;
 267
 268 static struct rx_thread_params rx_thread_params_array[MAX_LCORE_PARAMS];
 269 static struct rx_thread_params rx_thread_params_array_default[] = {
 270         {0, 0, 2, 0},
 271         {0, 1, 2, 1},
 272         {0, 2, 2, 2},
 273         {1, 0, 2, 3},
 274         {1, 1, 2, 4},
 275         {1, 2, 2, 5},
 276         {2, 0, 2, 6},
 277         {3, 0, 3, 7},
 278         {3, 1, 3, 8},
 279 };
 280
 281 static struct rx_thread_params *rx_thread_params =
 282                 rx_thread_params_array_default;
 283 static uint16_t nb_rx_thread_params = RTE_DIM(rx_thread_params_array_default);
 284
 285 struct tx_thread_params {
 286         uint8_t lcore_id;
 287         uint8_t thread_id;
 288 } __rte_cache_aligned;
 289
 290 static struct tx_thread_params tx_thread_params_array[MAX_LCORE_PARAMS];
 291 static struct tx_thread_params tx_thread_params_array_default[] = {
 292         {4, 0},
 293         {5, 1},
 294         {6, 2},
 295         {7, 3},
 296         {8, 4},
 297         {9, 5},
 298         {10, 6},
 299         {11, 7},
 300         {12, 8},
 301 };
 302
 303 static struct tx_thread_params *tx_thread_params =
 304                 tx_thread_params_array_default;
 305 static uint16_t nb_tx_thread_params = RTE_DIM(tx_thread_params_array_default);
 306
 307 static struct rte_eth_conf port_conf = {
 308         .rxmode = {
 309                 .mq_mode = ETH_MQ_RX_RSS,
 310                 .split_hdr_size = 0,
 311                 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
 312         },
 313         .rx_adv_conf = {
 314                 .rss_conf = {
 315                         .rss_key = NULL,
 316                         .rss_hf = ETH_RSS_TCP,
 317                 },
 318         },
 319         .txmode = {
 320                 .mq_mode = ETH_MQ_TX_NONE,
 321         },
 322 };
 323
 324 static uint32_t max_pkt_len;
 325
 326 static struct rte_mempool *pktmbuf_pool[NB_SOCKETS];
 327
 328 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 329
 330 #include <rte_hash_crc.h>
 331 #define DEFAULT_HASH_FUNC       rte_hash_crc
 332
 333 struct ipv4_5tuple {
 334         uint32_t ip_dst;
 335         uint32_t ip_src;
 336         uint16_t port_dst;
 337         uint16_t port_src;
 338         uint8_t  proto;
 339 } __rte_packed;
 340
 341 union ipv4_5tuple_host {
 342         struct {
 343                 uint8_t  pad0;
 344                 uint8_t  proto;
 345                 uint16_t pad1;
 346                 uint32_t ip_src;
 347                 uint32_t ip_dst;
 348                 uint16_t port_src;
 349                 uint16_t port_dst;
 350         };
 351         __m128i xmm;
 352 };
 353
 354 #define XMM_NUM_IN_IPV6_5TUPLE 3
 355
 356 struct ipv6_5tuple {
 357         uint8_t  ip_dst[IPV6_ADDR_LEN];
 358         uint8_t  ip_src[IPV6_ADDR_LEN];
 359         uint16_t port_dst;
 360         uint16_t port_src;
 361         uint8_t  proto;
 362 } __rte_packed;
 363
 364 union ipv6_5tuple_host {
 365         struct {
 366                 uint16_t pad0;
 367                 uint8_t  proto;
 368                 uint8_t  pad1;
 369                 uint8_t  ip_src[IPV6_ADDR_LEN];
 370                 uint8_t  ip_dst[IPV6_ADDR_LEN];
 371                 uint16_t port_src;
 372                 uint16_t port_dst;
 373                 uint64_t reserve;
 374         };
 375         __m128i xmm[XMM_NUM_IN_IPV6_5TUPLE];
 376 };
 377
 378 struct ipv4_l3fwd_route {
 379         struct ipv4_5tuple key;
 380         uint8_t if_out;
 381 };
 382
 383 struct ipv6_l3fwd_route {
 384         struct ipv6_5tuple key;
 385         uint8_t if_out;
 386 };
 387
 388 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 389         {{RTE_IPV4(101, 0, 0, 0), RTE_IPV4(100, 10, 0, 1),  101, 11, IPPROTO_TCP}, 0},
 390         {{RTE_IPV4(201, 0, 0, 0), RTE_IPV4(200, 20, 0, 1),  102, 12, IPPROTO_TCP}, 1},
 391         {{RTE_IPV4(111, 0, 0, 0), RTE_IPV4(100, 30, 0, 1),  101, 11, IPPROTO_TCP}, 2},
 392         {{RTE_IPV4(211, 0, 0, 0), RTE_IPV4(200, 40, 0, 1),  102, 12, IPPROTO_TCP}, 3},
 393 };
 394
 395 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 396         {{
 397         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 398         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 399                         0x05},
 400         101, 11, IPPROTO_TCP}, 0},
 401
 402         {{
 403         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 404         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 405                         0x05},
 406         102, 12, IPPROTO_TCP}, 1},
 407
 408         {{
 409         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 410         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 411                         0x05},
 412         101, 11, IPPROTO_TCP}, 2},
 413
 414         {{
 415         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 416         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38,
 417                         0x05},
 418         102, 12, IPPROTO_TCP}, 3},
 419 };
 420
 421 typedef struct rte_hash lookup_struct_t;
 422 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 423 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 424
 425 #ifdef RTE_ARCH_X86_64
 426 /* default to 4 million hash entries (approx) */
 427 #define L3FWD_HASH_ENTRIES (1024*1024*4)
 428 #else
 429 /* 32-bit has less address-space for hugepage memory, limit to 1M entries */
 430 #define L3FWD_HASH_ENTRIES (1024*1024*1)
 431 #endif
 432 #define HASH_ENTRY_NUMBER_DEFAULT 4
 433
 434 static uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT;
 435
 436 static inline uint32_t
 437 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
 438                 uint32_t init_val)
 439 {
 440         const union ipv4_5tuple_host *k;
 441         uint32_t t;
 442         const uint32_t *p;
 443
 444         k = data;
 445         t = k->proto;
 446         p = (const uint32_t *)&k->port_src;
 447
 448         init_val = rte_hash_crc_4byte(t, init_val);
 449         init_val = rte_hash_crc_4byte(k->ip_src, init_val);
 450         init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
 451         init_val = rte_hash_crc_4byte(*p, init_val);
 452         return init_val;
 453 }
 454
 455 static inline uint32_t
 456 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len,
 457                 uint32_t init_val)
 458 {
 459         const union ipv6_5tuple_host *k;
 460         uint32_t t;
 461         const uint32_t *p;
 462         const uint32_t *ip_src0, *ip_src1, *ip_src2, *ip_src3;
 463         const uint32_t *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
 464
 465         k = data;
 466         t = k->proto;
 467         p = (const uint32_t *)&k->port_src;
 468
 469         ip_src0 = (const uint32_t *) k->ip_src;
 470         ip_src1 = (const uint32_t *)(k->ip_src + 4);
 471         ip_src2 = (const uint32_t *)(k->ip_src + 8);
 472         ip_src3 = (const uint32_t *)(k->ip_src + 12);
 473         ip_dst0 = (const uint32_t *) k->ip_dst;
 474         ip_dst1 = (const uint32_t *)(k->ip_dst + 4);
 475         ip_dst2 = (const uint32_t *)(k->ip_dst + 8);
 476         ip_dst3 = (const uint32_t *)(k->ip_dst + 12);
 477         init_val = rte_hash_crc_4byte(t, init_val);
 478         init_val = rte_hash_crc_4byte(*ip_src0, init_val);
 479         init_val = rte_hash_crc_4byte(*ip_src1, init_val);
 480         init_val = rte_hash_crc_4byte(*ip_src2, init_val);
 481         init_val = rte_hash_crc_4byte(*ip_src3, init_val);
 482         init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
 483         init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
 484         init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
 485         init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
 486         init_val = rte_hash_crc_4byte(*p, init_val);
 487         return init_val;
 488 }
 489
 490 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
 491 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
 492
 493 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 494 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 495
 496 #endif
 497
 498 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 499 struct ipv4_l3fwd_route {
 500         uint32_t ip;
 501         uint8_t  depth;
 502         uint8_t  if_out;
 503 };
 504
 505 struct ipv6_l3fwd_route {
 506         uint8_t ip[16];
 507         uint8_t depth;
 508         uint8_t if_out;
 509 };
 510
 511 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 512         {RTE_IPV4(1, 1, 1, 0), 24, 0},
 513         {RTE_IPV4(2, 1, 1, 0), 24, 1},
 514         {RTE_IPV4(3, 1, 1, 0), 24, 2},
 515         {RTE_IPV4(4, 1, 1, 0), 24, 3},
 516         {RTE_IPV4(5, 1, 1, 0), 24, 4},
 517         {RTE_IPV4(6, 1, 1, 0), 24, 5},
 518         {RTE_IPV4(7, 1, 1, 0), 24, 6},
 519         {RTE_IPV4(8, 1, 1, 0), 24, 7},
 520 };
 521
 522 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 523         {{1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 0},
 524         {{2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 1},
 525         {{3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 2},
 526         {{4, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 3},
 527         {{5, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 4},
 528         {{6, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 5},
 529         {{7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 6},
 530         {{8, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 7},
 531 };
 532
 533 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array)
 534 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array)
 535
 536 #define IPV4_L3FWD_LPM_MAX_RULES         1024
 537 #define IPV6_L3FWD_LPM_MAX_RULES         1024
 538 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
 539
 540 typedef struct rte_lpm lookup_struct_t;
 541 typedef struct rte_lpm6 lookup6_struct_t;
 542 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 543 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 544 #endif
 545
 546 struct lcore_conf {
 547         lookup_struct_t *ipv4_lookup_struct;
 548 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 549         lookup6_struct_t *ipv6_lookup_struct;
 550 #else
 551         lookup_struct_t *ipv6_lookup_struct;
 552 #endif
 553         void *data;
 554 } __rte_cache_aligned;
 555
 556 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 557 RTE_DEFINE_PER_LCORE(struct lcore_conf *, lcore_conf);
 558
 559 #define MAX_RX_QUEUE_PER_THREAD 16
 560 #define MAX_TX_PORT_PER_THREAD  RTE_MAX_ETHPORTS
 561 #define MAX_TX_QUEUE_PER_PORT   RTE_MAX_ETHPORTS
 562 #define MAX_RX_QUEUE_PER_PORT   128
 563
 564 #define MAX_RX_THREAD 1024
 565 #define MAX_TX_THREAD 1024
 566 #define MAX_THREAD    (MAX_RX_THREAD + MAX_TX_THREAD)
 567
 568 /**
 569  * Producers and consumers threads configuration
 570  */
 571 static int lthreads_on = 1; /**< Use lthreads for processing*/
 572
 573 uint16_t rx_counter;  /**< Number of spawned rx threads */
 574 uint16_t tx_counter;  /**< Number of spawned tx threads */
 575
 576 struct thread_conf {
 577         uint16_t lcore_id;      /**< Initial lcore for rx thread */
 578         uint16_t cpu_id;        /**< Cpu id for cpu load stats counter */
 579         uint16_t thread_id;     /**< Thread ID */
 580
 581 #if (APP_CPU_LOAD > 0)
 582         int busy[MAX_CPU_COUNTER];
 583 #endif
 584 };
 585
 586 struct thread_rx_conf {
 587         struct thread_conf conf;
 588
 589         uint16_t n_rx_queue;
 590         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 591
 592         uint16_t n_ring;        /**< Number of output rings */
 593         struct rte_ring *ring[RTE_MAX_LCORE];
 594         struct lthread_cond *ready[RTE_MAX_LCORE];
 595
 596 #if (APP_CPU_LOAD > 0)
 597         int busy[MAX_CPU_COUNTER];
 598 #endif
 599 } __rte_cache_aligned;
 600
 601 uint16_t n_rx_thread;
 602 struct thread_rx_conf rx_thread[MAX_RX_THREAD];
 603
 604 struct thread_tx_conf {
 605         struct thread_conf conf;
 606
 607         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 608         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 609
 610         struct rte_ring *ring;
 611         struct lthread_cond **ready;
 612
 613 } __rte_cache_aligned;
 614
 615 uint16_t n_tx_thread;
 616 struct thread_tx_conf tx_thread[MAX_TX_THREAD];
 617
 618 /* Send burst of packets on an output interface */
 619 static inline int
 620 send_burst(struct thread_tx_conf *qconf, uint16_t n, uint16_t port)
 621 {
 622         struct rte_mbuf **m_table;
 623         int ret;
 624         uint16_t queueid;
 625
 626         queueid = qconf->tx_queue_id[port];
 627         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 628
 629         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 630         if (unlikely(ret < n)) {
 631                 do {
 632                         rte_pktmbuf_free(m_table[ret]);
 633                 } while (++ret < n);
 634         }
 635
 636         return 0;
 637 }
 638
 639 /* Enqueue a single packet, and send burst if queue is filled */
 640 static inline int
 641 send_single_packet(struct rte_mbuf *m, uint16_t port)
 642 {
 643         uint16_t len;
 644         struct thread_tx_conf *qconf;
 645
 646         if (lthreads_on)
 647                 qconf = (struct thread_tx_conf *)lthread_get_data();
 648         else
 649                 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
 650
 651         len = qconf->tx_mbufs[port].len;
 652         qconf->tx_mbufs[port].m_table[len] = m;
 653         len++;
 654
 655         /* enough pkts to be sent */
 656         if (unlikely(len == MAX_PKT_BURST)) {
 657                 send_burst(qconf, MAX_PKT_BURST, port);
 658                 len = 0;
 659         }
 660
 661         qconf->tx_mbufs[port].len = len;
 662         return 0;
 663 }
 664
 665 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
 666         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
 667 static __rte_always_inline void
 668 send_packetsx4(uint16_t port,
 669         struct rte_mbuf *m[], uint32_t num)
 670 {
 671         uint32_t len, j, n;
 672         struct thread_tx_conf *qconf;
 673
 674         if (lthreads_on)
 675                 qconf = (struct thread_tx_conf *)lthread_get_data();
 676         else
 677                 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data;
 678
 679         len = qconf->tx_mbufs[port].len;
 680
 681         /*
 682          * If TX buffer for that queue is empty, and we have enough packets,
 683          * then send them straightway.
 684          */
 685         if (num >= MAX_TX_BURST && len == 0) {
 686                 n = rte_eth_tx_burst(port, qconf->tx_queue_id[port], m, num);
 687                 if (unlikely(n < num)) {
 688                         do {
 689                                 rte_pktmbuf_free(m[n]);
 690                         } while (++n < num);
 691                 }
 692                 return;
 693         }
 694
 695         /*
 696          * Put packets into TX buffer for that queue.
 697          */
 698
 699         n = len + num;
 700         n = (n > MAX_PKT_BURST) ? MAX_PKT_BURST - len : num;
 701
 702         j = 0;
 703         switch (n % FWDSTEP) {
 704         while (j < n) {
 705         case 0:
 706                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 707                 j++;
 708                 /* fall-through */
 709         case 3:
 710                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 711                 j++;
 712                 /* fall-through */
 713         case 2:
 714                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 715                 j++;
 716                 /* fall-through */
 717         case 1:
 718                 qconf->tx_mbufs[port].m_table[len + j] = m[j];
 719                 j++;
 720         }
 721         }
 722
 723         len += n;
 724
 725         /* enough pkts to be sent */
 726         if (unlikely(len == MAX_PKT_BURST)) {
 727
 728                 send_burst(qconf, MAX_PKT_BURST, port);
 729
 730                 /* copy rest of the packets into the TX buffer. */
 731                 len = num - n;
 732                 j = 0;
 733                 switch (len % FWDSTEP) {
 734                 while (j < len) {
 735                 case 0:
 736                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 737                         j++;
 738                         /* fall-through */
 739                 case 3:
 740                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 741                         j++;
 742                         /* fall-through */
 743                 case 2:
 744                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 745                         j++;
 746                         /* fall-through */
 747                 case 1:
 748                         qconf->tx_mbufs[port].m_table[j] = m[n + j];
 749                         j++;
 750                 }
 751                 }
 752         }
 753
 754         qconf->tx_mbufs[port].len = len;
 755 }
 756 #endif /* APP_LOOKUP_LPM */
 757
 758 #ifdef DO_RFC_1812_CHECKS
 759 static inline int
 760 is_valid_ipv4_pkt(struct rte_ipv4_hdr *pkt, uint32_t link_len)
 761 {
 762         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 763         /*
 764          * 1. The packet length reported by the Link Layer must be large
 765          * enough to hold the minimum length legal IP datagram (20 bytes).
 766          */
 767         if (link_len < sizeof(struct rte_ipv4_hdr))
 768                 return -1;
 769
 770         /* 2. The IP checksum must be correct. */
 771         /* this is checked in H/W */
 772
 773         /*
 774          * 3. The IP version number must be 4. If the version number is not 4
 775          * then the packet may be another version of IP, such as IPng or
 776          * ST-II.
 777          */
 778         if (((pkt->version_ihl) >> 4) != 4)
 779                 return -3;
 780         /*
 781          * 4. The IP header length field must be large enough to hold the
 782          * minimum length legal IP datagram (20 bytes = 5 words).
 783          */
 784         if ((pkt->version_ihl & 0xf) < 5)
 785                 return -4;
 786
 787         /*
 788          * 5. The IP total length field must be large enough to hold the IP
 789          * datagram header, whose length is specified in the IP header length
 790          * field.
 791          */
 792         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct rte_ipv4_hdr))
 793                 return -5;
 794
 795         return 0;
 796 }
 797 #endif
 798
 799 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 800
 801 static __m128i mask0;
 802 static __m128i mask1;
 803 static __m128i mask2;
 804 static inline uint16_t
 805 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid,
 806                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 807 {
 808         int ret = 0;
 809         union ipv4_5tuple_host key;
 810
 811         ipv4_hdr = (uint8_t *)ipv4_hdr +
 812                 offsetof(struct rte_ipv4_hdr, time_to_live);
 813         __m128i data = _mm_loadu_si128((__m128i *)(ipv4_hdr));
 814         /* Get 5 tuple: dst port, src port, dst IP address, src IP address and
 815            protocol */
 816         key.xmm = _mm_and_si128(data, mask0);
 817         /* Find destination port */
 818         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 819         return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
 820 }
 821
 822 static inline uint16_t
 823 get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid,
 824                 lookup_struct_t *ipv6_l3fwd_lookup_struct)
 825 {
 826         int ret = 0;
 827         union ipv6_5tuple_host key;
 828
 829         ipv6_hdr = (uint8_t *)ipv6_hdr +
 830                 offsetof(struct rte_ipv6_hdr, payload_len);
 831         __m128i data0 = _mm_loadu_si128((__m128i *)(ipv6_hdr));
 832         __m128i data1 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
 833                         sizeof(__m128i)));
 834         __m128i data2 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) +
 835                         sizeof(__m128i) + sizeof(__m128i)));
 836         /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
 837         key.xmm[0] = _mm_and_si128(data0, mask1);
 838         /* Get part of 5 tuple: dst IP address lower 96 bits and src IP address
 839            higher 32 bits */
 840         key.xmm[1] = data1;
 841         /* Get part of 5 tuple: dst port and src port and dst IP address higher
 842            32 bits */
 843         key.xmm[2] = _mm_and_si128(data2, mask2);
 844
 845         /* Find destination port */
 846         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 847         return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
 848 }
 849 #endif
 850
 851 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 852
 853 static inline uint16_t
 854 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid,
 855                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 856 {
 857         uint32_t next_hop;
 858
 859         return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 860                 rte_be_to_cpu_32(((struct rte_ipv4_hdr *)ipv4_hdr)->dst_addr),
 861                 &next_hop) == 0) ? next_hop : portid);
 862 }
 863
 864 static inline uint16_t
 865 get_ipv6_dst_port(void *ipv6_hdr,  uint16_t portid,
 866                 lookup6_struct_t *ipv6_l3fwd_lookup_struct)
 867 {
 868         uint32_t next_hop;
 869
 870         return ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
 871                 ((struct rte_ipv6_hdr *)ipv6_hdr)->dst_addr, &next_hop) == 0) ?
 872                 next_hop : portid);
 873 }
 874 #endif
 875
 876 static inline void l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid)
 877                 __rte_unused;
 878
 879 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) && \
 880         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
 881
 882 #define MASK_ALL_PKTS   0xff
 883 #define EXCLUDE_1ST_PKT 0xfe
 884 #define EXCLUDE_2ND_PKT 0xfd
 885 #define EXCLUDE_3RD_PKT 0xfb
 886 #define EXCLUDE_4TH_PKT 0xf7
 887 #define EXCLUDE_5TH_PKT 0xef
 888 #define EXCLUDE_6TH_PKT 0xdf
 889 #define EXCLUDE_7TH_PKT 0xbf
 890 #define EXCLUDE_8TH_PKT 0x7f
 891
 892 static inline void
 893 simple_ipv4_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid)
 894 {
 895         struct rte_ether_hdr *eth_hdr[8];
 896         struct rte_ipv4_hdr *ipv4_hdr[8];
 897         uint16_t dst_port[8];
 898         int32_t ret[8];
 899         union ipv4_5tuple_host key[8];
 900         __m128i data[8];
 901
 902         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct rte_ether_hdr *);
 903         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct rte_ether_hdr *);
 904         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct rte_ether_hdr *);
 905         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct rte_ether_hdr *);
 906         eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct rte_ether_hdr *);
 907         eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct rte_ether_hdr *);
 908         eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct rte_ether_hdr *);
 909         eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct rte_ether_hdr *);
 910
 911         /* Handle IPv4 headers.*/
 912         ipv4_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct rte_ipv4_hdr *,
 913                         sizeof(struct rte_ether_hdr));
 914         ipv4_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct rte_ipv4_hdr *,
 915                         sizeof(struct rte_ether_hdr));
 916         ipv4_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct rte_ipv4_hdr *,
 917                         sizeof(struct rte_ether_hdr));
 918         ipv4_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct rte_ipv4_hdr *,
 919                         sizeof(struct rte_ether_hdr));
 920         ipv4_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct rte_ipv4_hdr *,
 921                         sizeof(struct rte_ether_hdr));
 922         ipv4_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct rte_ipv4_hdr *,
 923                         sizeof(struct rte_ether_hdr));
 924         ipv4_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct rte_ipv4_hdr *,
 925                         sizeof(struct rte_ether_hdr));
 926         ipv4_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct rte_ipv4_hdr *,
 927                         sizeof(struct rte_ether_hdr));
 928
 929 #ifdef DO_RFC_1812_CHECKS
 930         /* Check to make sure the packet is valid (RFC1812) */
 931         uint8_t valid_mask = MASK_ALL_PKTS;
 932
 933         if (is_valid_ipv4_pkt(ipv4_hdr[0], m[0]->pkt_len) < 0) {
 934                 rte_pktmbuf_free(m[0]);
 935                 valid_mask &= EXCLUDE_1ST_PKT;
 936         }
 937         if (is_valid_ipv4_pkt(ipv4_hdr[1], m[1]->pkt_len) < 0) {
 938                 rte_pktmbuf_free(m[1]);
 939                 valid_mask &= EXCLUDE_2ND_PKT;
 940         }
 941         if (is_valid_ipv4_pkt(ipv4_hdr[2], m[2]->pkt_len) < 0) {
 942                 rte_pktmbuf_free(m[2]);
 943                 valid_mask &= EXCLUDE_3RD_PKT;
 944         }
 945         if (is_valid_ipv4_pkt(ipv4_hdr[3], m[3]->pkt_len) < 0) {
 946                 rte_pktmbuf_free(m[3]);
 947                 valid_mask &= EXCLUDE_4TH_PKT;
 948         }
 949         if (is_valid_ipv4_pkt(ipv4_hdr[4], m[4]->pkt_len) < 0) {
 950                 rte_pktmbuf_free(m[4]);
 951                 valid_mask &= EXCLUDE_5TH_PKT;
 952         }
 953         if (is_valid_ipv4_pkt(ipv4_hdr[5], m[5]->pkt_len) < 0) {
 954                 rte_pktmbuf_free(m[5]);
 955                 valid_mask &= EXCLUDE_6TH_PKT;
 956         }
 957         if (is_valid_ipv4_pkt(ipv4_hdr[6], m[6]->pkt_len) < 0) {
 958                 rte_pktmbuf_free(m[6]);
 959                 valid_mask &= EXCLUDE_7TH_PKT;
 960         }
 961         if (is_valid_ipv4_pkt(ipv4_hdr[7], m[7]->pkt_len) < 0) {
 962                 rte_pktmbuf_free(m[7]);
 963                 valid_mask &= EXCLUDE_8TH_PKT;
 964         }
 965         if (unlikely(valid_mask != MASK_ALL_PKTS)) {
 966                 if (valid_mask == 0)
 967                         return;
 968
 969                 uint8_t i = 0;
 970
 971                 for (i = 0; i < 8; i++)
 972                         if ((0x1 << i) & valid_mask)
 973                                 l3fwd_simple_forward(m[i], portid);
 974         }
 975 #endif /* End of #ifdef DO_RFC_1812_CHECKS */
 976
 977         data[0] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[0], __m128i *,
 978                         sizeof(struct rte_ether_hdr) +
 979                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 980         data[1] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[1], __m128i *,
 981                         sizeof(struct rte_ether_hdr) +
 982                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 983         data[2] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[2], __m128i *,
 984                         sizeof(struct rte_ether_hdr) +
 985                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 986         data[3] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[3], __m128i *,
 987                         sizeof(struct rte_ether_hdr) +
 988                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 989         data[4] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[4], __m128i *,
 990                         sizeof(struct rte_ether_hdr) +
 991                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 992         data[5] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[5], __m128i *,
 993                         sizeof(struct rte_ether_hdr) +
 994                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 995         data[6] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[6], __m128i *,
 996                         sizeof(struct rte_ether_hdr) +
 997                         offsetof(struct rte_ipv4_hdr, time_to_live)));
 998         data[7] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[7], __m128i *,
 999                         sizeof(struct rte_ether_hdr) +
1000                         offsetof(struct rte_ipv4_hdr, time_to_live)));
1001
1002         key[0].xmm = _mm_and_si128(data[0], mask0);
1003         key[1].xmm = _mm_and_si128(data[1], mask0);
1004         key[2].xmm = _mm_and_si128(data[2], mask0);
1005         key[3].xmm = _mm_and_si128(data[3], mask0);
1006         key[4].xmm = _mm_and_si128(data[4], mask0);
1007         key[5].xmm = _mm_and_si128(data[5], mask0);
1008         key[6].xmm = _mm_and_si128(data[6], mask0);
1009         key[7].xmm = _mm_and_si128(data[7], mask0);
1010
1011         const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1012                         &key[4], &key[5], &key[6], &key[7]};
1013
1014         rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct,
1015                         &key_array[0], 8, ret);
1016         dst_port[0] = ((ret[0] < 0) ? portid : ipv4_l3fwd_out_if[ret[0]]);
1017         dst_port[1] = ((ret[1] < 0) ? portid : ipv4_l3fwd_out_if[ret[1]]);
1018         dst_port[2] = ((ret[2] < 0) ? portid : ipv4_l3fwd_out_if[ret[2]]);
1019         dst_port[3] = ((ret[3] < 0) ? portid : ipv4_l3fwd_out_if[ret[3]]);
1020         dst_port[4] = ((ret[4] < 0) ? portid : ipv4_l3fwd_out_if[ret[4]]);
1021         dst_port[5] = ((ret[5] < 0) ? portid : ipv4_l3fwd_out_if[ret[5]]);
1022         dst_port[6] = ((ret[6] < 0) ? portid : ipv4_l3fwd_out_if[ret[6]]);
1023         dst_port[7] = ((ret[7] < 0) ? portid : ipv4_l3fwd_out_if[ret[7]]);
1024
1025         if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1026                         (enabled_port_mask & 1 << dst_port[0]) == 0)
1027                 dst_port[0] = portid;
1028         if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1029                         (enabled_port_mask & 1 << dst_port[1]) == 0)
1030                 dst_port[1] = portid;
1031         if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1032                         (enabled_port_mask & 1 << dst_port[2]) == 0)
1033                 dst_port[2] = portid;
1034         if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1035                         (enabled_port_mask & 1 << dst_port[3]) == 0)
1036                 dst_port[3] = portid;
1037         if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1038                         (enabled_port_mask & 1 << dst_port[4]) == 0)
1039                 dst_port[4] = portid;
1040         if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1041                         (enabled_port_mask & 1 << dst_port[5]) == 0)
1042                 dst_port[5] = portid;
1043         if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1044                         (enabled_port_mask & 1 << dst_port[6]) == 0)
1045                 dst_port[6] = portid;
1046         if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1047                         (enabled_port_mask & 1 << dst_port[7]) == 0)
1048                 dst_port[7] = portid;
1049
1050 #ifdef DO_RFC_1812_CHECKS
1051         /* Update time to live and header checksum */
1052         --(ipv4_hdr[0]->time_to_live);
1053         --(ipv4_hdr[1]->time_to_live);
1054         --(ipv4_hdr[2]->time_to_live);
1055         --(ipv4_hdr[3]->time_to_live);
1056         ++(ipv4_hdr[0]->hdr_checksum);
1057         ++(ipv4_hdr[1]->hdr_checksum);
1058         ++(ipv4_hdr[2]->hdr_checksum);
1059         ++(ipv4_hdr[3]->hdr_checksum);
1060         --(ipv4_hdr[4]->time_to_live);
1061         --(ipv4_hdr[5]->time_to_live);
1062         --(ipv4_hdr[6]->time_to_live);
1063         --(ipv4_hdr[7]->time_to_live);
1064         ++(ipv4_hdr[4]->hdr_checksum);
1065         ++(ipv4_hdr[5]->hdr_checksum);
1066         ++(ipv4_hdr[6]->hdr_checksum);
1067         ++(ipv4_hdr[7]->hdr_checksum);
1068 #endif
1069
1070         /* dst addr */
1071         *(uint64_t *)&eth_hdr[0]->dst_addr = dest_eth_addr[dst_port[0]];
1072         *(uint64_t *)&eth_hdr[1]->dst_addr = dest_eth_addr[dst_port[1]];
1073         *(uint64_t *)&eth_hdr[2]->dst_addr = dest_eth_addr[dst_port[2]];
1074         *(uint64_t *)&eth_hdr[3]->dst_addr = dest_eth_addr[dst_port[3]];
1075         *(uint64_t *)&eth_hdr[4]->dst_addr = dest_eth_addr[dst_port[4]];
1076         *(uint64_t *)&eth_hdr[5]->dst_addr = dest_eth_addr[dst_port[5]];
1077         *(uint64_t *)&eth_hdr[6]->dst_addr = dest_eth_addr[dst_port[6]];
1078         *(uint64_t *)&eth_hdr[7]->dst_addr = dest_eth_addr[dst_port[7]];
1079
1080         /* src addr */
1081         rte_ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->src_addr);
1082         rte_ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->src_addr);
1083         rte_ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->src_addr);
1084         rte_ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->src_addr);
1085         rte_ether_addr_copy(&ports_eth_addr[dst_port[4]], &eth_hdr[4]->src_addr);
1086         rte_ether_addr_copy(&ports_eth_addr[dst_port[5]], &eth_hdr[5]->src_addr);
1087         rte_ether_addr_copy(&ports_eth_addr[dst_port[6]], &eth_hdr[6]->src_addr);
1088         rte_ether_addr_copy(&ports_eth_addr[dst_port[7]], &eth_hdr[7]->src_addr);
1089
1090         send_single_packet(m[0], (uint8_t)dst_port[0]);
1091         send_single_packet(m[1], (uint8_t)dst_port[1]);
1092         send_single_packet(m[2], (uint8_t)dst_port[2]);
1093         send_single_packet(m[3], (uint8_t)dst_port[3]);
1094         send_single_packet(m[4], (uint8_t)dst_port[4]);
1095         send_single_packet(m[5], (uint8_t)dst_port[5]);
1096         send_single_packet(m[6], (uint8_t)dst_port[6]);
1097         send_single_packet(m[7], (uint8_t)dst_port[7]);
1098
1099 }
1100
1101 static inline void get_ipv6_5tuple(struct rte_mbuf *m0, __m128i mask0,
1102                 __m128i mask1, union ipv6_5tuple_host *key)
1103 {
1104         __m128i tmpdata0 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1105                         __m128i *, sizeof(struct rte_ether_hdr) +
1106                         offsetof(struct rte_ipv6_hdr, payload_len)));
1107         __m128i tmpdata1 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1108                         __m128i *, sizeof(struct rte_ether_hdr) +
1109                         offsetof(struct rte_ipv6_hdr, payload_len) +
1110                         sizeof(__m128i)));
1111         __m128i tmpdata2 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0,
1112                         __m128i *, sizeof(struct rte_ether_hdr) +
1113                         offsetof(struct rte_ipv6_hdr, payload_len) +
1114                         sizeof(__m128i) + sizeof(__m128i)));
1115         key->xmm[0] = _mm_and_si128(tmpdata0, mask0);
1116         key->xmm[1] = tmpdata1;
1117         key->xmm[2] = _mm_and_si128(tmpdata2, mask1);
1118 }
1119
1120 static inline void
1121 simple_ipv6_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid)
1122 {
1123         int32_t ret[8];
1124         uint16_t dst_port[8];
1125         struct rte_ether_hdr *eth_hdr[8];
1126         union ipv6_5tuple_host key[8];
1127
1128         __rte_unused struct rte_ipv6_hdr *ipv6_hdr[8];
1129
1130         eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct rte_ether_hdr *);
1131         eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct rte_ether_hdr *);
1132         eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct rte_ether_hdr *);
1133         eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct rte_ether_hdr *);
1134         eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct rte_ether_hdr *);
1135         eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct rte_ether_hdr *);
1136         eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct rte_ether_hdr *);
1137         eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct rte_ether_hdr *);
1138
1139         /* Handle IPv6 headers.*/
1140         ipv6_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct rte_ipv6_hdr *,
1141                         sizeof(struct rte_ether_hdr));
1142         ipv6_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct rte_ipv6_hdr *,
1143                         sizeof(struct rte_ether_hdr));
1144         ipv6_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct rte_ipv6_hdr *,
1145                         sizeof(struct rte_ether_hdr));
1146         ipv6_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct rte_ipv6_hdr *,
1147                         sizeof(struct rte_ether_hdr));
1148         ipv6_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct rte_ipv6_hdr *,
1149                         sizeof(struct rte_ether_hdr));
1150         ipv6_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct rte_ipv6_hdr *,
1151                         sizeof(struct rte_ether_hdr));
1152         ipv6_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct rte_ipv6_hdr *,
1153                         sizeof(struct rte_ether_hdr));
1154         ipv6_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct rte_ipv6_hdr *,
1155                         sizeof(struct rte_ether_hdr));
1156
1157         get_ipv6_5tuple(m[0], mask1, mask2, &key[0]);
1158         get_ipv6_5tuple(m[1], mask1, mask2, &key[1]);
1159         get_ipv6_5tuple(m[2], mask1, mask2, &key[2]);
1160         get_ipv6_5tuple(m[3], mask1, mask2, &key[3]);
1161         get_ipv6_5tuple(m[4], mask1, mask2, &key[4]);
1162         get_ipv6_5tuple(m[5], mask1, mask2, &key[5]);
1163         get_ipv6_5tuple(m[6], mask1, mask2, &key[6]);
1164         get_ipv6_5tuple(m[7], mask1, mask2, &key[7]);
1165
1166         const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3],
1167                         &key[4], &key[5], &key[6], &key[7]};
1168
1169         rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1170                         &key_array[0], 4, ret);
1171         dst_port[0] = ((ret[0] < 0) ? portid : ipv6_l3fwd_out_if[ret[0]]);
1172         dst_port[1] = ((ret[1] < 0) ? portid : ipv6_l3fwd_out_if[ret[1]]);
1173         dst_port[2] = ((ret[2] < 0) ? portid : ipv6_l3fwd_out_if[ret[2]]);
1174         dst_port[3] = ((ret[3] < 0) ? portid : ipv6_l3fwd_out_if[ret[3]]);
1175         dst_port[4] = ((ret[4] < 0) ? portid : ipv6_l3fwd_out_if[ret[4]]);
1176         dst_port[5] = ((ret[5] < 0) ? portid : ipv6_l3fwd_out_if[ret[5]]);
1177         dst_port[6] = ((ret[6] < 0) ? portid : ipv6_l3fwd_out_if[ret[6]]);
1178         dst_port[7] = ((ret[7] < 0) ? portid : ipv6_l3fwd_out_if[ret[7]]);
1179
1180         if (dst_port[0] >= RTE_MAX_ETHPORTS ||
1181                         (enabled_port_mask & 1 << dst_port[0]) == 0)
1182                 dst_port[0] = portid;
1183         if (dst_port[1] >= RTE_MAX_ETHPORTS ||
1184                         (enabled_port_mask & 1 << dst_port[1]) == 0)
1185                 dst_port[1] = portid;
1186         if (dst_port[2] >= RTE_MAX_ETHPORTS ||
1187                         (enabled_port_mask & 1 << dst_port[2]) == 0)
1188                 dst_port[2] = portid;
1189         if (dst_port[3] >= RTE_MAX_ETHPORTS ||
1190                         (enabled_port_mask & 1 << dst_port[3]) == 0)
1191                 dst_port[3] = portid;
1192         if (dst_port[4] >= RTE_MAX_ETHPORTS ||
1193                         (enabled_port_mask & 1 << dst_port[4]) == 0)
1194                 dst_port[4] = portid;
1195         if (dst_port[5] >= RTE_MAX_ETHPORTS ||
1196                         (enabled_port_mask & 1 << dst_port[5]) == 0)
1197                 dst_port[5] = portid;
1198         if (dst_port[6] >= RTE_MAX_ETHPORTS ||
1199                         (enabled_port_mask & 1 << dst_port[6]) == 0)
1200                 dst_port[6] = portid;
1201         if (dst_port[7] >= RTE_MAX_ETHPORTS ||
1202                         (enabled_port_mask & 1 << dst_port[7]) == 0)
1203                 dst_port[7] = portid;
1204
1205         /* dst addr */
1206         *(uint64_t *)&eth_hdr[0]->dst_addr = dest_eth_addr[dst_port[0]];
1207         *(uint64_t *)&eth_hdr[1]->dst_addr = dest_eth_addr[dst_port[1]];
1208         *(uint64_t *)&eth_hdr[2]->dst_addr = dest_eth_addr[dst_port[2]];
1209         *(uint64_t *)&eth_hdr[3]->dst_addr = dest_eth_addr[dst_port[3]];
1210         *(uint64_t *)&eth_hdr[4]->dst_addr = dest_eth_addr[dst_port[4]];
1211         *(uint64_t *)&eth_hdr[5]->dst_addr = dest_eth_addr[dst_port[5]];
1212         *(uint64_t *)&eth_hdr[6]->dst_addr = dest_eth_addr[dst_port[6]];
1213         *(uint64_t *)&eth_hdr[7]->dst_addr = dest_eth_addr[dst_port[7]];
1214
1215         /* src addr */
1216         rte_ether_addr_copy(&ports_eth_addr[dst_port[0]], &eth_hdr[0]->src_addr);
1217         rte_ether_addr_copy(&ports_eth_addr[dst_port[1]], &eth_hdr[1]->src_addr);
1218         rte_ether_addr_copy(&ports_eth_addr[dst_port[2]], &eth_hdr[2]->src_addr);
1219         rte_ether_addr_copy(&ports_eth_addr[dst_port[3]], &eth_hdr[3]->src_addr);
1220         rte_ether_addr_copy(&ports_eth_addr[dst_port[4]], &eth_hdr[4]->src_addr);
1221         rte_ether_addr_copy(&ports_eth_addr[dst_port[5]], &eth_hdr[5]->src_addr);
1222         rte_ether_addr_copy(&ports_eth_addr[dst_port[6]], &eth_hdr[6]->src_addr);
1223         rte_ether_addr_copy(&ports_eth_addr[dst_port[7]], &eth_hdr[7]->src_addr);
1224
1225         send_single_packet(m[0], dst_port[0]);
1226         send_single_packet(m[1], dst_port[1]);
1227         send_single_packet(m[2], dst_port[2]);
1228         send_single_packet(m[3], dst_port[3]);
1229         send_single_packet(m[4], dst_port[4]);
1230         send_single_packet(m[5], dst_port[5]);
1231         send_single_packet(m[6], dst_port[6]);
1232         send_single_packet(m[7], dst_port[7]);
1233
1234 }
1235 #endif /* APP_LOOKUP_METHOD */
1236
1237 static __rte_always_inline void
1238 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid)
1239 {
1240         struct rte_ether_hdr *eth_hdr;
1241         struct rte_ipv4_hdr *ipv4_hdr;
1242         uint16_t dst_port;
1243
1244         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
1245
1246         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
1247                 /* Handle IPv4 headers.*/
1248                 ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
1249                                 sizeof(struct rte_ether_hdr));
1250
1251 #ifdef DO_RFC_1812_CHECKS
1252                 /* Check to make sure the packet is valid (RFC1812) */
1253                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
1254                         rte_pktmbuf_free(m);
1255                         return;
1256                 }
1257 #endif
1258
1259                  dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
1260                         RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct);
1261                 if (dst_port >= RTE_MAX_ETHPORTS ||
1262                                 (enabled_port_mask & 1 << dst_port) == 0)
1263                         dst_port = portid;
1264
1265 #ifdef DO_RFC_1812_CHECKS
1266                 /* Update time to live and header checksum */
1267                 --(ipv4_hdr->time_to_live);
1268                 ++(ipv4_hdr->hdr_checksum);
1269 #endif
1270                 /* dst addr */
1271                 *(uint64_t *)&eth_hdr->dst_addr = dest_eth_addr[dst_port];
1272
1273                 /* src addr */
1274                 rte_ether_addr_copy(&ports_eth_addr[dst_port],
1275                                 &eth_hdr->src_addr);
1276
1277                 send_single_packet(m, dst_port);
1278         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
1279                 /* Handle IPv6 headers.*/
1280                 struct rte_ipv6_hdr *ipv6_hdr;
1281
1282                 ipv6_hdr = rte_pktmbuf_mtod_offset(m, struct rte_ipv6_hdr *,
1283                                 sizeof(struct rte_ether_hdr));
1284
1285                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
1286                                 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct);
1287
1288                 if (dst_port >= RTE_MAX_ETHPORTS ||
1289                                 (enabled_port_mask & 1 << dst_port) == 0)
1290                         dst_port = portid;
1291
1292                 /* dst addr */
1293                 *(uint64_t *)&eth_hdr->dst_addr = dest_eth_addr[dst_port];
1294
1295                 /* src addr */
1296                 rte_ether_addr_copy(&ports_eth_addr[dst_port],
1297                                 &eth_hdr->src_addr);
1298
1299                 send_single_packet(m, dst_port);
1300         } else
1301                 /* Free the mbuf that contains non-IPV4/IPV6 packet */
1302                 rte_pktmbuf_free(m);
1303 }
1304
1305 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1306         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1307 #ifdef DO_RFC_1812_CHECKS
1308
1309 #define IPV4_MIN_VER_IHL        0x45
1310 #define IPV4_MAX_VER_IHL        0x4f
1311 #define IPV4_MAX_VER_IHL_DIFF   (IPV4_MAX_VER_IHL - IPV4_MIN_VER_IHL)
1312
1313 /* Minimum value of IPV4 total length (20B) in network byte order. */
1314 #define IPV4_MIN_LEN_BE (sizeof(struct rte_ipv4_hdr) << 8)
1315
1316 /*
1317  * From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2:
1318  * - The IP version number must be 4.
1319  * - The IP header length field must be large enough to hold the
1320  *    minimum length legal IP datagram (20 bytes = 5 words).
1321  * - The IP total length field must be large enough to hold the IP
1322  *   datagram header, whose length is specified in the IP header length
1323  *   field.
1324  * If we encounter invalid IPV4 packet, then set destination port for it
1325  * to BAD_PORT value.
1326  */
1327 static __rte_always_inline void
1328 rfc1812_process(struct rte_ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t ptype)
1329 {
1330         uint8_t ihl;
1331
1332         if (RTE_ETH_IS_IPV4_HDR(ptype)) {
1333                 ihl = ipv4_hdr->version_ihl - IPV4_MIN_VER_IHL;
1334
1335                 ipv4_hdr->time_to_live--;
1336                 ipv4_hdr->hdr_checksum++;
1337
1338                 if (ihl > IPV4_MAX_VER_IHL_DIFF ||
1339                                 ((uint8_t)ipv4_hdr->total_length == 0 &&
1340                                 ipv4_hdr->total_length < IPV4_MIN_LEN_BE)) {
1341                         dp[0] = BAD_PORT;
1342                 }
1343         }
1344 }
1345
1346 #else
1347 #define rfc1812_process(mb, dp, ptype)  do { } while (0)
1348 #endif /* DO_RFC_1812_CHECKS */
1349 #endif /* APP_LOOKUP_LPM && ENABLE_MULTI_BUFFER_OPTIMIZE */
1350
1351
1352 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1353         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1354
1355 static __rte_always_inline uint16_t
1356 get_dst_port(struct rte_mbuf *pkt, uint32_t dst_ipv4, uint16_t portid)
1357 {
1358         uint32_t next_hop;
1359         struct rte_ipv6_hdr *ipv6_hdr;
1360         struct rte_ether_hdr *eth_hdr;
1361
1362         if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
1363                 return (uint16_t) ((rte_lpm_lookup(
1364                                 RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dst_ipv4,
1365                                 &next_hop) == 0) ? next_hop : portid);
1366
1367         } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
1368
1369                 eth_hdr = rte_pktmbuf_mtod(pkt, struct rte_ether_hdr *);
1370                 ipv6_hdr = (struct rte_ipv6_hdr *)(eth_hdr + 1);
1371
1372                 return (uint16_t) ((rte_lpm6_lookup(
1373                                 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct,
1374                                 ipv6_hdr->dst_addr, &next_hop) == 0) ?
1375                                 next_hop : portid);
1376
1377         }
1378
1379         return portid;
1380 }
1381
1382 static inline void
1383 process_packet(struct rte_mbuf *pkt, uint16_t *dst_port, uint16_t portid)
1384 {
1385         struct rte_ether_hdr *eth_hdr;
1386         struct rte_ipv4_hdr *ipv4_hdr;
1387         uint32_t dst_ipv4;
1388         uint16_t dp;
1389         __m128i te, ve;
1390
1391         eth_hdr = rte_pktmbuf_mtod(pkt, struct rte_ether_hdr *);
1392         ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
1393
1394         dst_ipv4 = ipv4_hdr->dst_addr;
1395         dst_ipv4 = rte_be_to_cpu_32(dst_ipv4);
1396         dp = get_dst_port(pkt, dst_ipv4, portid);
1397
1398         te = _mm_load_si128((__m128i *)eth_hdr);
1399         ve = val_eth[dp];
1400
1401         dst_port[0] = dp;
1402         rfc1812_process(ipv4_hdr, dst_port, pkt->packet_type);
1403
1404         te =  _mm_blend_epi16(te, ve, MASK_ETH);
1405         _mm_store_si128((__m128i *)eth_hdr, te);
1406 }
1407
1408 /*
1409  * Read packet_type and destination IPV4 addresses from 4 mbufs.
1410  */
1411 static inline void
1412 processx4_step1(struct rte_mbuf *pkt[FWDSTEP],
1413                 __m128i *dip,
1414                 uint32_t *ipv4_flag)
1415 {
1416         struct rte_ipv4_hdr *ipv4_hdr;
1417         struct rte_ether_hdr *eth_hdr;
1418         uint32_t x0, x1, x2, x3;
1419
1420         eth_hdr = rte_pktmbuf_mtod(pkt[0], struct rte_ether_hdr *);
1421         ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
1422         x0 = ipv4_hdr->dst_addr;
1423         ipv4_flag[0] = pkt[0]->packet_type & RTE_PTYPE_L3_IPV4;
1424
1425         eth_hdr = rte_pktmbuf_mtod(pkt[1], struct rte_ether_hdr *);
1426         ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
1427         x1 = ipv4_hdr->dst_addr;
1428         ipv4_flag[0] &= pkt[1]->packet_type;
1429
1430         eth_hdr = rte_pktmbuf_mtod(pkt[2], struct rte_ether_hdr *);
1431         ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
1432         x2 = ipv4_hdr->dst_addr;
1433         ipv4_flag[0] &= pkt[2]->packet_type;
1434
1435         eth_hdr = rte_pktmbuf_mtod(pkt[3], struct rte_ether_hdr *);
1436         ipv4_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
1437         x3 = ipv4_hdr->dst_addr;
1438         ipv4_flag[0] &= pkt[3]->packet_type;
1439
1440         dip[0] = _mm_set_epi32(x3, x2, x1, x0);
1441 }
1442
1443 /*
1444  * Lookup into LPM for destination port.
1445  * If lookup fails, use incoming port (portid) as destination port.
1446  */
1447 static inline void
1448 processx4_step2(__m128i dip,
1449                 uint32_t ipv4_flag,
1450                 uint16_t portid,
1451                 struct rte_mbuf *pkt[FWDSTEP],
1452                 uint16_t dprt[FWDSTEP])
1453 {
1454         rte_xmm_t dst;
1455         const __m128i bswap_mask = _mm_set_epi8(12, 13, 14, 15, 8, 9, 10, 11,
1456                         4, 5, 6, 7, 0, 1, 2, 3);
1457
1458         /* Byte swap 4 IPV4 addresses. */
1459         dip = _mm_shuffle_epi8(dip, bswap_mask);
1460
1461         /* if all 4 packets are IPV4. */
1462         if (likely(ipv4_flag)) {
1463                 rte_lpm_lookupx4(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dip,
1464                                 dst.u32, portid);
1465
1466                 /* get rid of unused upper 16 bit for each dport. */
1467                 dst.x = _mm_packs_epi32(dst.x, dst.x);
1468                 *(uint64_t *)dprt = dst.u64[0];
1469         } else {
1470                 dst.x = dip;
1471                 dprt[0] = get_dst_port(pkt[0], dst.u32[0], portid);
1472                 dprt[1] = get_dst_port(pkt[1], dst.u32[1], portid);
1473                 dprt[2] = get_dst_port(pkt[2], dst.u32[2], portid);
1474                 dprt[3] = get_dst_port(pkt[3], dst.u32[3], portid);
1475         }
1476 }
1477
1478 /*
1479  * Update source and destination MAC addresses in the ethernet header.
1480  * Perform RFC1812 checks and updates for IPV4 packets.
1481  */
1482 static inline void
1483 processx4_step3(struct rte_mbuf *pkt[FWDSTEP], uint16_t dst_port[FWDSTEP])
1484 {
1485         __m128i te[FWDSTEP];
1486         __m128i ve[FWDSTEP];
1487         __m128i *p[FWDSTEP];
1488
1489         p[0] = rte_pktmbuf_mtod(pkt[0], __m128i *);
1490         p[1] = rte_pktmbuf_mtod(pkt[1], __m128i *);
1491         p[2] = rte_pktmbuf_mtod(pkt[2], __m128i *);
1492         p[3] = rte_pktmbuf_mtod(pkt[3], __m128i *);
1493
1494         ve[0] = val_eth[dst_port[0]];
1495         te[0] = _mm_load_si128(p[0]);
1496
1497         ve[1] = val_eth[dst_port[1]];
1498         te[1] = _mm_load_si128(p[1]);
1499
1500         ve[2] = val_eth[dst_port[2]];
1501         te[2] = _mm_load_si128(p[2]);
1502
1503         ve[3] = val_eth[dst_port[3]];
1504         te[3] = _mm_load_si128(p[3]);
1505
1506         /* Update first 12 bytes, keep rest bytes intact. */
1507         te[0] =  _mm_blend_epi16(te[0], ve[0], MASK_ETH);
1508         te[1] =  _mm_blend_epi16(te[1], ve[1], MASK_ETH);
1509         te[2] =  _mm_blend_epi16(te[2], ve[2], MASK_ETH);
1510         te[3] =  _mm_blend_epi16(te[3], ve[3], MASK_ETH);
1511
1512         _mm_store_si128(p[0], te[0]);
1513         _mm_store_si128(p[1], te[1]);
1514         _mm_store_si128(p[2], te[2]);
1515         _mm_store_si128(p[3], te[3]);
1516
1517         rfc1812_process((struct rte_ipv4_hdr *)
1518                         ((struct rte_ether_hdr *)p[0] + 1),
1519                         &dst_port[0], pkt[0]->packet_type);
1520         rfc1812_process((struct rte_ipv4_hdr *)
1521                         ((struct rte_ether_hdr *)p[1] + 1),
1522                         &dst_port[1], pkt[1]->packet_type);
1523         rfc1812_process((struct rte_ipv4_hdr *)
1524                         ((struct rte_ether_hdr *)p[2] + 1),
1525                         &dst_port[2], pkt[2]->packet_type);
1526         rfc1812_process((struct rte_ipv4_hdr *)
1527                         ((struct rte_ether_hdr *)p[3] + 1),
1528                         &dst_port[3], pkt[3]->packet_type);
1529 }
1530
1531 /*
1532  * We group consecutive packets with the same destionation port into one burst.
1533  * To avoid extra latency this is done together with some other packet
1534  * processing, but after we made a final decision about packet's destination.
1535  * To do this we maintain:
1536  * pnum - array of number of consecutive packets with the same dest port for
1537  * each packet in the input burst.
1538  * lp - pointer to the last updated element in the pnum.
1539  * dlp - dest port value lp corresponds to.
1540  */
1541
1542 #define GRPSZ   (1 << FWDSTEP)
1543 #define GRPMSK  (GRPSZ - 1)
1544
1545 #define GROUP_PORT_STEP(dlp, dcp, lp, pn, idx)  do { \
1546         if (likely((dlp) == (dcp)[(idx)])) {         \
1547                 (lp)[0]++;                           \
1548         } else {                                     \
1549                 (dlp) = (dcp)[idx];                  \
1550                 (lp) = (pn) + (idx);                 \
1551                 (lp)[0] = 1;                         \
1552         }                                            \
1553 } while (0)
1554
1555 /*
1556  * Group consecutive packets with the same destination port in bursts of 4.
1557  * Suppose we have array of destionation ports:
1558  * dst_port[] = {a, b, c, d,, e, ... }
1559  * dp1 should contain: <a, b, c, d>, dp2: <b, c, d, e>.
1560  * We doing 4 comparisons at once and the result is 4 bit mask.
1561  * This mask is used as an index into prebuild array of pnum values.
1562  */
1563 static inline uint16_t *
1564 port_groupx4(uint16_t pn[FWDSTEP + 1], uint16_t *lp, __m128i dp1, __m128i dp2)
1565 {
1566         static const struct {
1567                 uint64_t pnum; /* prebuild 4 values for pnum[]. */
1568                 int32_t  idx;  /* index for new last updated elemnet. */
1569                 uint16_t lpv;  /* add value to the last updated element. */
1570         } gptbl[GRPSZ] = {
1571         {
1572                 /* 0: a != b, b != c, c != d, d != e */
1573                 .pnum = UINT64_C(0x0001000100010001),
1574                 .idx = 4,
1575                 .lpv = 0,
1576         },
1577         {
1578                 /* 1: a == b, b != c, c != d, d != e */
1579                 .pnum = UINT64_C(0x0001000100010002),
1580                 .idx = 4,
1581                 .lpv = 1,
1582         },
1583         {
1584                 /* 2: a != b, b == c, c != d, d != e */
1585                 .pnum = UINT64_C(0x0001000100020001),
1586                 .idx = 4,
1587                 .lpv = 0,
1588         },
1589         {
1590                 /* 3: a == b, b == c, c != d, d != e */
1591                 .pnum = UINT64_C(0x0001000100020003),
1592                 .idx = 4,
1593                 .lpv = 2,
1594         },
1595         {
1596                 /* 4: a != b, b != c, c == d, d != e */
1597                 .pnum = UINT64_C(0x0001000200010001),
1598                 .idx = 4,
1599                 .lpv = 0,
1600         },
1601         {
1602                 /* 5: a == b, b != c, c == d, d != e */
1603                 .pnum = UINT64_C(0x0001000200010002),
1604                 .idx = 4,
1605                 .lpv = 1,
1606         },
1607         {
1608                 /* 6: a != b, b == c, c == d, d != e */
1609                 .pnum = UINT64_C(0x0001000200030001),
1610                 .idx = 4,
1611                 .lpv = 0,
1612         },
1613         {
1614                 /* 7: a == b, b == c, c == d, d != e */
1615                 .pnum = UINT64_C(0x0001000200030004),
1616                 .idx = 4,
1617                 .lpv = 3,
1618         },
1619         {
1620                 /* 8: a != b, b != c, c != d, d == e */
1621                 .pnum = UINT64_C(0x0002000100010001),
1622                 .idx = 3,
1623                 .lpv = 0,
1624         },
1625         {
1626                 /* 9: a == b, b != c, c != d, d == e */
1627                 .pnum = UINT64_C(0x0002000100010002),
1628                 .idx = 3,
1629                 .lpv = 1,
1630         },
1631         {
1632                 /* 0xa: a != b, b == c, c != d, d == e */
1633                 .pnum = UINT64_C(0x0002000100020001),
1634                 .idx = 3,
1635                 .lpv = 0,
1636         },
1637         {
1638                 /* 0xb: a == b, b == c, c != d, d == e */
1639                 .pnum = UINT64_C(0x0002000100020003),
1640                 .idx = 3,
1641                 .lpv = 2,
1642         },
1643         {
1644                 /* 0xc: a != b, b != c, c == d, d == e */
1645                 .pnum = UINT64_C(0x0002000300010001),
1646                 .idx = 2,
1647                 .lpv = 0,
1648         },
1649         {
1650                 /* 0xd: a == b, b != c, c == d, d == e */
1651                 .pnum = UINT64_C(0x0002000300010002),
1652                 .idx = 2,
1653                 .lpv = 1,
1654         },
1655         {
1656                 /* 0xe: a != b, b == c, c == d, d == e */
1657                 .pnum = UINT64_C(0x0002000300040001),
1658                 .idx = 1,
1659                 .lpv = 0,
1660         },
1661         {
1662                 /* 0xf: a == b, b == c, c == d, d == e */
1663                 .pnum = UINT64_C(0x0002000300040005),
1664                 .idx = 0,
1665                 .lpv = 4,
1666         },
1667         };
1668
1669         union {
1670                 uint16_t u16[FWDSTEP + 1];
1671                 uint64_t u64;
1672         } *pnum = (void *)pn;
1673
1674         int32_t v;
1675
1676         dp1 = _mm_cmpeq_epi16(dp1, dp2);
1677         dp1 = _mm_unpacklo_epi16(dp1, dp1);
1678         v = _mm_movemask_ps((__m128)dp1);
1679
1680         /* update last port counter. */
1681         lp[0] += gptbl[v].lpv;
1682
1683         /* if dest port value has changed. */
1684         if (v != GRPMSK) {
1685                 pnum->u64 = gptbl[v].pnum;
1686                 pnum->u16[FWDSTEP] = 1;
1687                 lp = pnum->u16 + gptbl[v].idx;
1688         }
1689
1690         return lp;
1691 }
1692
1693 #endif /* APP_LOOKUP_METHOD */
1694
1695 static void
1696 process_burst(struct rte_mbuf *pkts_burst[MAX_PKT_BURST], int nb_rx,
1697                 uint16_t portid)
1698 {
1699
1700         int j;
1701
1702 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \
1703         (ENABLE_MULTI_BUFFER_OPTIMIZE == 1))
1704         int32_t k;
1705         uint16_t dlp;
1706         uint16_t *lp;
1707         uint16_t dst_port[MAX_PKT_BURST];
1708         __m128i dip[MAX_PKT_BURST / FWDSTEP];
1709         uint32_t ipv4_flag[MAX_PKT_BURST / FWDSTEP];
1710         uint16_t pnum[MAX_PKT_BURST + 1];
1711 #endif
1712
1713
1714 #if (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)
1715 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1716         {
1717                 /*
1718                  * Send nb_rx - nb_rx%8 packets
1719                  * in groups of 8.
1720                  */
1721                 int32_t n = RTE_ALIGN_FLOOR(nb_rx, 8);
1722
1723                 for (j = 0; j < n; j += 8) {
1724                         uint32_t pkt_type =
1725                                 pkts_burst[j]->packet_type &
1726                                 pkts_burst[j+1]->packet_type &
1727                                 pkts_burst[j+2]->packet_type &
1728                                 pkts_burst[j+3]->packet_type &
1729                                 pkts_burst[j+4]->packet_type &
1730                                 pkts_burst[j+5]->packet_type &
1731                                 pkts_burst[j+6]->packet_type &
1732                                 pkts_burst[j+7]->packet_type;
1733                         if (pkt_type & RTE_PTYPE_L3_IPV4) {
1734                                 simple_ipv4_fwd_8pkts(&pkts_burst[j], portid);
1735                         } else if (pkt_type &
1736                                 RTE_PTYPE_L3_IPV6) {
1737                                 simple_ipv6_fwd_8pkts(&pkts_burst[j], portid);
1738                         } else {
1739                                 l3fwd_simple_forward(pkts_burst[j], portid);
1740                                 l3fwd_simple_forward(pkts_burst[j+1], portid);
1741                                 l3fwd_simple_forward(pkts_burst[j+2], portid);
1742                                 l3fwd_simple_forward(pkts_burst[j+3], portid);
1743                                 l3fwd_simple_forward(pkts_burst[j+4], portid);
1744                                 l3fwd_simple_forward(pkts_burst[j+5], portid);
1745                                 l3fwd_simple_forward(pkts_burst[j+6], portid);
1746                                 l3fwd_simple_forward(pkts_burst[j+7], portid);
1747                         }
1748                 }
1749                 for (; j < nb_rx ; j++)
1750                         l3fwd_simple_forward(pkts_burst[j], portid);
1751         }
1752 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1753
1754         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1755         for (j = 0; j != k; j += FWDSTEP)
1756                 processx4_step1(&pkts_burst[j], &dip[j / FWDSTEP],
1757                                 &ipv4_flag[j / FWDSTEP]);
1758
1759         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1760         for (j = 0; j != k; j += FWDSTEP)
1761                 processx4_step2(dip[j / FWDSTEP], ipv4_flag[j / FWDSTEP],
1762                                 portid, &pkts_burst[j], &dst_port[j]);
1763
1764         /*
1765          * Finish packet processing and group consecutive
1766          * packets with the same destination port.
1767          */
1768         k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
1769         if (k != 0) {
1770                 __m128i dp1, dp2;
1771
1772                 lp = pnum;
1773                 lp[0] = 1;
1774
1775                 processx4_step3(pkts_burst, dst_port);
1776
1777                 /* dp1: <d[0], d[1], d[2], d[3], ... > */
1778                 dp1 = _mm_loadu_si128((__m128i *)dst_port);
1779
1780                 for (j = FWDSTEP; j != k; j += FWDSTEP) {
1781                         processx4_step3(&pkts_burst[j], &dst_port[j]);
1782
1783                         /*
1784                          * dp2:
1785                          * <d[j-3], d[j-2], d[j-1], d[j], ... >
1786                          */
1787                         dp2 = _mm_loadu_si128(
1788                                         (__m128i *)&dst_port[j - FWDSTEP + 1]);
1789                         lp  = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1790
1791                         /*
1792                          * dp1:
1793                          * <d[j], d[j+1], d[j+2], d[j+3], ... >
1794                          */
1795                         dp1 = _mm_srli_si128(dp2, (FWDSTEP - 1) *
1796                                         sizeof(dst_port[0]));
1797                 }
1798
1799                 /*
1800                  * dp2: <d[j-3], d[j-2], d[j-1], d[j-1], ... >
1801                  */
1802                 dp2 = _mm_shufflelo_epi16(dp1, 0xf9);
1803                 lp  = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2);
1804
1805                 /*
1806                  * remove values added by the last repeated
1807                  * dst port.
1808                  */
1809                 lp[0]--;
1810                 dlp = dst_port[j - 1];
1811         } else {
1812                 /* set dlp and lp to the never used values. */
1813                 dlp = BAD_PORT - 1;
1814                 lp = pnum + MAX_PKT_BURST;
1815         }
1816
1817         /* Process up to last 3 packets one by one. */
1818         switch (nb_rx % FWDSTEP) {
1819         case 3:
1820                 process_packet(pkts_burst[j], dst_port + j, portid);
1821                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1822                 j++;
1823                 /* fall-through */
1824         case 2:
1825                 process_packet(pkts_burst[j], dst_port + j, portid);
1826                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1827                 j++;
1828                 /* fall-through */
1829         case 1:
1830                 process_packet(pkts_burst[j], dst_port + j, portid);
1831                 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j);
1832                 j++;
1833         }
1834
1835         /*
1836          * Send packets out, through destination port.
1837          * Consecuteve pacekts with the same destination port
1838          * are already grouped together.
1839          * If destination port for the packet equals BAD_PORT,
1840          * then free the packet without sending it out.
1841          */
1842         for (j = 0; j < nb_rx; j += k) {
1843
1844                 int32_t m;
1845                 uint16_t pn;
1846
1847                 pn = dst_port[j];
1848                 k = pnum[j];
1849
1850                 if (likely(pn != BAD_PORT))
1851                         send_packetsx4(pn, pkts_burst + j, k);
1852                 else
1853                         for (m = j; m != j + k; m++)
1854                                 rte_pktmbuf_free(pkts_burst[m]);
1855
1856         }
1857
1858 #endif /* APP_LOOKUP_METHOD */
1859 #else /* ENABLE_MULTI_BUFFER_OPTIMIZE == 0 */
1860
1861         /* Prefetch first packets */
1862         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++)
1863                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[j], void *));
1864
1865         /* Prefetch and forward already prefetched packets */
1866         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1867                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1868                                 j + PREFETCH_OFFSET], void *));
1869                 l3fwd_simple_forward(pkts_burst[j], portid);
1870         }
1871
1872         /* Forward remaining prefetched packets */
1873         for (; j < nb_rx; j++)
1874                 l3fwd_simple_forward(pkts_burst[j], portid);
1875
1876 #endif /* ENABLE_MULTI_BUFFER_OPTIMIZE */
1877
1878 }
1879
1880 #if (APP_CPU_LOAD > 0)
1881
1882 /*
1883  * CPU-load stats collector
1884  */
1885 static int __rte_noreturn
1886 cpu_load_collector(__rte_unused void *arg) {
1887         unsigned i, j, k;
1888         uint64_t hits;
1889         uint64_t prev_tsc, diff_tsc, cur_tsc;
1890         uint64_t total[MAX_CPU] = { 0 };
1891         unsigned min_cpu = MAX_CPU;
1892         unsigned max_cpu = 0;
1893         unsigned cpu_id;
1894         int busy_total = 0;
1895         int busy_flag = 0;
1896
1897         unsigned int n_thread_per_cpu[MAX_CPU] = { 0 };
1898         struct thread_conf *thread_per_cpu[MAX_CPU][MAX_THREAD];
1899
1900         struct thread_conf *thread_conf;
1901
1902         const uint64_t interval_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1903                 US_PER_S * CPU_LOAD_TIMEOUT_US;
1904
1905         prev_tsc = 0;
1906         /*
1907          * Wait for all threads
1908          */
1909
1910         printf("Waiting for %d rx threads and %d tx threads\n", n_rx_thread,
1911                         n_tx_thread);
1912
1913         rte_wait_until_equal_16(&rx_counter, n_rx_thread, __ATOMIC_RELAXED);
1914         rte_wait_until_equal_16(&tx_counter, n_tx_thread, __ATOMIC_RELAXED);
1915
1916         for (i = 0; i < n_rx_thread; i++) {
1917
1918                 thread_conf = &rx_thread[i].conf;
1919                 cpu_id = thread_conf->cpu_id;
1920                 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1921
1922                 if (cpu_id > max_cpu)
1923                         max_cpu = cpu_id;
1924                 if (cpu_id < min_cpu)
1925                         min_cpu = cpu_id;
1926         }
1927         for (i = 0; i < n_tx_thread; i++) {
1928
1929                 thread_conf = &tx_thread[i].conf;
1930                 cpu_id = thread_conf->cpu_id;
1931                 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1932
1933                 if (thread_conf->cpu_id > max_cpu)
1934                         max_cpu = thread_conf->cpu_id;
1935                 if (thread_conf->cpu_id < min_cpu)
1936                         min_cpu = thread_conf->cpu_id;
1937         }
1938
1939         while (1) {
1940
1941                 cpu_load.counter++;
1942                 for (i = min_cpu; i <= max_cpu; i++) {
1943                         for (j = 0; j < MAX_CPU_COUNTER; j++) {
1944                                 for (k = 0; k < n_thread_per_cpu[i]; k++)
1945                                         if (thread_per_cpu[i][k]->busy[j]) {
1946                                                 busy_flag = 1;
1947                                                 break;
1948                                         }
1949                                 if (busy_flag) {
1950                                         cpu_load.hits[j][i]++;
1951                                         busy_total = 1;
1952                                         busy_flag = 0;
1953                                 }
1954                         }
1955
1956                         if (busy_total) {
1957                                 total[i]++;
1958                                 busy_total = 0;
1959                         }
1960                 }
1961
1962                 cur_tsc = rte_rdtsc();
1963
1964                 diff_tsc = cur_tsc - prev_tsc;
1965                 if (unlikely(diff_tsc > interval_tsc)) {
1966
1967                         printf("\033c");
1968
1969                         printf("Cpu usage for %d rx threads and %d tx threads:\n\n",
1970                                         n_rx_thread, n_tx_thread);
1971
1972                         printf("cpu#     proc%%  poll%%  overhead%%\n\n");
1973
1974                         for (i = min_cpu; i <= max_cpu; i++) {
1975                                 hits = 0;
1976                                 printf("CPU %d:", i);
1977                                 for (j = 0; j < MAX_CPU_COUNTER; j++) {
1978                                         printf("%7" PRIu64 "",
1979                                                         cpu_load.hits[j][i] * 100 / cpu_load.counter);
1980                                         hits += cpu_load.hits[j][i];
1981                                         cpu_load.hits[j][i] = 0;
1982                                 }
1983                                 printf("%7" PRIu64 "\n",
1984                                                 100 - total[i] * 100 / cpu_load.counter);
1985                                 total[i] = 0;
1986                         }
1987                         cpu_load.counter = 0;
1988
1989                         prev_tsc = cur_tsc;
1990                 }
1991
1992         }
1993 }
1994 #endif /* APP_CPU_LOAD */
1995
1996 /*
1997  * Null processing lthread loop
1998  *
1999  * This loop is used to start empty scheduler on lcore.
2000  */
2001 static void *
2002 lthread_null(__rte_unused void *args)
2003 {
2004         int lcore_id = rte_lcore_id();
2005
2006         RTE_LOG(INFO, L3FWD, "Starting scheduler on lcore %d.\n", lcore_id);
2007         lthread_exit(NULL);
2008         return NULL;
2009 }
2010
2011 /* main processing loop */
2012 static void *
2013 lthread_tx_per_ring(void *dummy)
2014 {
2015         int nb_rx;
2016         uint16_t portid;
2017         struct rte_ring *ring;
2018         struct thread_tx_conf *tx_conf;
2019         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2020         struct lthread_cond *ready;
2021
2022         tx_conf = (struct thread_tx_conf *)dummy;
2023         ring = tx_conf->ring;
2024         ready = *tx_conf->ready;
2025
2026         lthread_set_data((void *)tx_conf);
2027
2028         /*
2029          * Move this lthread to lcore
2030          */
2031         lthread_set_affinity(tx_conf->conf.lcore_id);
2032
2033         RTE_LOG(INFO, L3FWD, "entering main tx loop on lcore %u\n", rte_lcore_id());
2034
2035         nb_rx = 0;
2036         __atomic_fetch_add(&tx_counter, 1, __ATOMIC_RELAXED);
2037         while (1) {
2038
2039                 /*
2040                  * Read packet from ring
2041                  */
2042                 SET_CPU_BUSY(tx_conf, CPU_POLL);
2043                 nb_rx = rte_ring_sc_dequeue_burst(ring, (void **)pkts_burst,
2044                                 MAX_PKT_BURST, NULL);
2045                 SET_CPU_IDLE(tx_conf, CPU_POLL);
2046
2047                 if (nb_rx > 0) {
2048                         SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2049                         portid = pkts_burst[0]->port;
2050                         process_burst(pkts_burst, nb_rx, portid);
2051                         SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2052                         lthread_yield();
2053                 } else
2054                         lthread_cond_wait(ready, 0);
2055
2056         }
2057         return NULL;
2058 }
2059
2060 /*
2061  * Main tx-lthreads spawner lthread.
2062  *
2063  * This lthread is used to spawn one new lthread per ring from producers.
2064  *
2065  */
2066 static void *
2067 lthread_tx(void *args)
2068 {
2069         struct lthread *lt;
2070
2071         unsigned lcore_id;
2072         uint16_t portid;
2073         struct thread_tx_conf *tx_conf;
2074
2075         tx_conf = (struct thread_tx_conf *)args;
2076         lthread_set_data((void *)tx_conf);
2077
2078         /*
2079          * Move this lthread to the selected lcore
2080          */
2081         lthread_set_affinity(tx_conf->conf.lcore_id);
2082
2083         /*
2084          * Spawn tx readers (one per input ring)
2085          */
2086         lthread_create(&lt, tx_conf->conf.lcore_id, lthread_tx_per_ring,
2087                         (void *)tx_conf);
2088
2089         lcore_id = rte_lcore_id();
2090
2091         RTE_LOG(INFO, L3FWD, "Entering Tx main loop on lcore %u\n", lcore_id);
2092
2093         tx_conf->conf.cpu_id = sched_getcpu();
2094         while (1) {
2095
2096                 lthread_sleep(BURST_TX_DRAIN_US * 1000);
2097
2098                 /*
2099                  * TX burst queue drain
2100                  */
2101                 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2102                         if (tx_conf->tx_mbufs[portid].len == 0)
2103                                 continue;
2104                         SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2105                         send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2106                         SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2107                         tx_conf->tx_mbufs[portid].len = 0;
2108                 }
2109
2110         }
2111         return NULL;
2112 }
2113
2114 static void *
2115 lthread_rx(void *dummy)
2116 {
2117         int ret;
2118         uint16_t nb_rx;
2119         int i;
2120         uint16_t portid;
2121         uint8_t queueid;
2122         int worker_id;
2123         int len[RTE_MAX_LCORE] = { 0 };
2124         int old_len, new_len;
2125         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2126         struct thread_rx_conf *rx_conf;
2127
2128         rx_conf = (struct thread_rx_conf *)dummy;
2129         lthread_set_data((void *)rx_conf);
2130
2131         /*
2132          * Move this lthread to lcore
2133          */
2134         lthread_set_affinity(rx_conf->conf.lcore_id);
2135
2136         if (rx_conf->n_rx_queue == 0) {
2137                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", rte_lcore_id());
2138                 return NULL;
2139         }
2140
2141         RTE_LOG(INFO, L3FWD, "Entering main Rx loop on lcore %u\n", rte_lcore_id());
2142
2143         for (i = 0; i < rx_conf->n_rx_queue; i++) {
2144
2145                 portid = rx_conf->rx_queue_list[i].port_id;
2146                 queueid = rx_conf->rx_queue_list[i].queue_id;
2147                 RTE_LOG(INFO, L3FWD,
2148                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2149                                 rte_lcore_id(), portid, queueid);
2150         }
2151
2152         /*
2153          * Init all condition variables (one per rx thread)
2154          */
2155         for (i = 0; i < rx_conf->n_rx_queue; i++)
2156                 lthread_cond_init(NULL, &rx_conf->ready[i], NULL);
2157
2158         worker_id = 0;
2159
2160         rx_conf->conf.cpu_id = sched_getcpu();
2161         __atomic_fetch_add(&rx_counter, 1, __ATOMIC_RELAXED);
2162         while (1) {
2163
2164                 /*
2165                  * Read packet from RX queues
2166                  */
2167                 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2168                         portid = rx_conf->rx_queue_list[i].port_id;
2169                         queueid = rx_conf->rx_queue_list[i].queue_id;
2170
2171                         SET_CPU_BUSY(rx_conf, CPU_POLL);
2172                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2173                                 MAX_PKT_BURST);
2174                         SET_CPU_IDLE(rx_conf, CPU_POLL);
2175
2176                         if (nb_rx != 0) {
2177                                 worker_id = (worker_id + 1) % rx_conf->n_ring;
2178                                 old_len = len[worker_id];
2179
2180                                 SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2181                                 ret = rte_ring_sp_enqueue_burst(
2182                                                 rx_conf->ring[worker_id],
2183                                                 (void **) pkts_burst,
2184                                                 nb_rx, NULL);
2185
2186                                 new_len = old_len + ret;
2187
2188                                 if (new_len >= BURST_SIZE) {
2189                                         lthread_cond_signal(rx_conf->ready[worker_id]);
2190                                         new_len = 0;
2191                                 }
2192
2193                                 len[worker_id] = new_len;
2194
2195                                 if (unlikely(ret < nb_rx)) {
2196                                         uint32_t k;
2197
2198                                         for (k = ret; k < nb_rx; k++) {
2199                                                 struct rte_mbuf *m = pkts_burst[k];
2200
2201                                                 rte_pktmbuf_free(m);
2202                                         }
2203                                 }
2204                                 SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2205                         }
2206
2207                         lthread_yield();
2208                 }
2209         }
2210         return NULL;
2211 }
2212
2213 /*
2214  * Start scheduler with initial lthread on lcore
2215  *
2216  * This lthread loop spawns all rx and tx lthreads on main lcore
2217  */
2218
2219 static void *
2220 lthread_spawner(__rte_unused void *arg)
2221 {
2222         struct lthread *lt[MAX_THREAD];
2223         int i;
2224         int n_thread = 0;
2225
2226         printf("Entering lthread_spawner\n");
2227
2228         /*
2229          * Create producers (rx threads) on default lcore
2230          */
2231         for (i = 0; i < n_rx_thread; i++) {
2232                 rx_thread[i].conf.thread_id = i;
2233                 lthread_create(&lt[n_thread], -1, lthread_rx,
2234                                 (void *)&rx_thread[i]);
2235                 n_thread++;
2236         }
2237
2238         /*
2239          * Wait for all producers. Until some producers can be started on the same
2240          * scheduler as this lthread, yielding is required to let them to run and
2241          * prevent deadlock here.
2242          */
2243         while (__atomic_load_n(&rx_counter, __ATOMIC_RELAXED) < n_rx_thread)
2244                 lthread_sleep(100000);
2245
2246         /*
2247          * Create consumers (tx threads) on default lcore_id
2248          */
2249         for (i = 0; i < n_tx_thread; i++) {
2250                 tx_thread[i].conf.thread_id = i;
2251                 lthread_create(&lt[n_thread], -1, lthread_tx,
2252                                 (void *)&tx_thread[i]);
2253                 n_thread++;
2254         }
2255
2256         /*
2257          * Wait for all threads finished
2258          */
2259         for (i = 0; i < n_thread; i++)
2260                 lthread_join(lt[i], NULL);
2261
2262         return NULL;
2263 }
2264
2265 /*
2266  * Start main scheduler with initial lthread spawning rx and tx lthreads
2267  * (main_lthread_main).
2268  */
2269 static int
2270 lthread_main_spawner(__rte_unused void *arg) {
2271         struct lthread *lt;
2272         int lcore_id = rte_lcore_id();
2273
2274         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2275         lthread_create(&lt, -1, lthread_spawner, NULL);
2276         lthread_run();
2277
2278         return 0;
2279 }
2280
2281 /*
2282  * Start scheduler on lcore.
2283  */
2284 static int
2285 sched_spawner(__rte_unused void *arg) {
2286         struct lthread *lt;
2287         int lcore_id = rte_lcore_id();
2288
2289 #if (APP_CPU_LOAD)
2290         if (lcore_id == cpu_load_lcore_id) {
2291                 cpu_load_collector(arg);
2292                 return 0;
2293         }
2294 #endif /* APP_CPU_LOAD */
2295
2296         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2297         lthread_create(&lt, -1, lthread_null, NULL);
2298         lthread_run();
2299
2300         return 0;
2301 }
2302
2303 /* main processing loop */
2304 static int __rte_noreturn
2305 pthread_tx(void *dummy)
2306 {
2307         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2308         uint64_t prev_tsc, diff_tsc, cur_tsc;
2309         int nb_rx;
2310         uint16_t portid;
2311         struct thread_tx_conf *tx_conf;
2312
2313         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
2314                 US_PER_S * BURST_TX_DRAIN_US;
2315
2316         prev_tsc = 0;
2317
2318         tx_conf = (struct thread_tx_conf *)dummy;
2319
2320         RTE_LOG(INFO, L3FWD, "Entering main Tx loop on lcore %u\n", rte_lcore_id());
2321
2322         tx_conf->conf.cpu_id = sched_getcpu();
2323         __atomic_fetch_add(&tx_counter, 1, __ATOMIC_RELAXED);
2324         while (1) {
2325
2326                 cur_tsc = rte_rdtsc();
2327
2328                 /*
2329                  * TX burst queue drain
2330                  */
2331                 diff_tsc = cur_tsc - prev_tsc;
2332                 if (unlikely(diff_tsc > drain_tsc)) {
2333
2334                         /*
2335                          * This could be optimized (use queueid instead of
2336                          * portid), but it is not called so often
2337                          */
2338                         SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2339                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2340                                 if (tx_conf->tx_mbufs[portid].len == 0)
2341                                         continue;
2342                                 send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2343                                 tx_conf->tx_mbufs[portid].len = 0;
2344                         }
2345                         SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2346
2347                         prev_tsc = cur_tsc;
2348                 }
2349
2350                 /*
2351                  * Read packet from ring
2352                  */
2353                 SET_CPU_BUSY(tx_conf, CPU_POLL);
2354                 nb_rx = rte_ring_sc_dequeue_burst(tx_conf->ring,
2355                                 (void **)pkts_burst, MAX_PKT_BURST, NULL);
2356                 SET_CPU_IDLE(tx_conf, CPU_POLL);
2357
2358                 if (unlikely(nb_rx == 0)) {
2359                         sched_yield();
2360                         continue;
2361                 }
2362
2363                 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2364                 portid = pkts_burst[0]->port;
2365                 process_burst(pkts_burst, nb_rx, portid);
2366                 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2367
2368         }
2369 }
2370
2371 static int
2372 pthread_rx(void *dummy)
2373 {
2374         int i;
2375         int worker_id;
2376         uint32_t n;
2377         uint32_t nb_rx;
2378         unsigned lcore_id;
2379         uint8_t queueid;
2380         uint16_t portid;
2381         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2382
2383         struct thread_rx_conf *rx_conf;
2384
2385         lcore_id = rte_lcore_id();
2386         rx_conf = (struct thread_rx_conf *)dummy;
2387
2388         if (rx_conf->n_rx_queue == 0) {
2389                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
2390                 return 0;
2391         }
2392
2393         RTE_LOG(INFO, L3FWD, "entering main rx loop on lcore %u\n", lcore_id);
2394
2395         for (i = 0; i < rx_conf->n_rx_queue; i++) {
2396
2397                 portid = rx_conf->rx_queue_list[i].port_id;
2398                 queueid = rx_conf->rx_queue_list[i].queue_id;
2399                 RTE_LOG(INFO, L3FWD,
2400                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2401                                 lcore_id, portid, queueid);
2402         }
2403
2404         worker_id = 0;
2405         rx_conf->conf.cpu_id = sched_getcpu();
2406         __atomic_fetch_add(&rx_counter, 1, __ATOMIC_RELAXED);
2407         while (1) {
2408
2409                 /*
2410                  * Read packet from RX queues
2411                  */
2412                 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2413                         portid = rx_conf->rx_queue_list[i].port_id;
2414                         queueid = rx_conf->rx_queue_list[i].queue_id;
2415
2416                         SET_CPU_BUSY(rx_conf, CPU_POLL);
2417                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2418                                 MAX_PKT_BURST);
2419                         SET_CPU_IDLE(rx_conf, CPU_POLL);
2420
2421                         if (nb_rx == 0) {
2422                                 sched_yield();
2423                                 continue;
2424                         }
2425
2426                         SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2427                         worker_id = (worker_id + 1) % rx_conf->n_ring;
2428                         n = rte_ring_sp_enqueue_burst(rx_conf->ring[worker_id],
2429                                         (void **)pkts_burst, nb_rx, NULL);
2430
2431                         if (unlikely(n != nb_rx)) {
2432                                 uint32_t k;
2433
2434                                 for (k = n; k < nb_rx; k++) {
2435                                         struct rte_mbuf *m = pkts_burst[k];
2436
2437                                         rte_pktmbuf_free(m);
2438                                 }
2439                         }
2440
2441                         SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2442
2443                 }
2444         }
2445 }
2446
2447 /*
2448  * P-Thread spawner.
2449  */
2450 static int
2451 pthread_run(__rte_unused void *arg) {
2452         int lcore_id = rte_lcore_id();
2453         int i;
2454
2455         for (i = 0; i < n_rx_thread; i++)
2456                 if (rx_thread[i].conf.lcore_id == lcore_id) {
2457                         printf("Start rx thread on %d...\n", lcore_id);
2458                         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2459                         RTE_PER_LCORE(lcore_conf)->data = (void *)&rx_thread[i];
2460                         pthread_rx((void *)&rx_thread[i]);
2461                         return 0;
2462                 }
2463
2464         for (i = 0; i < n_tx_thread; i++)
2465                 if (tx_thread[i].conf.lcore_id == lcore_id) {
2466                         printf("Start tx thread on %d...\n", lcore_id);
2467                         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2468                         RTE_PER_LCORE(lcore_conf)->data = (void *)&tx_thread[i];
2469                         pthread_tx((void *)&tx_thread[i]);
2470                         return 0;
2471                 }
2472
2473 #if (APP_CPU_LOAD)
2474         if (lcore_id == cpu_load_lcore_id)
2475                 cpu_load_collector(arg);
2476 #endif /* APP_CPU_LOAD */
2477
2478         return 0;
2479 }
2480
2481 static int
2482 check_lcore_params(void)
2483 {
2484         uint8_t queue, lcore;
2485         uint16_t i;
2486         int socketid;
2487
2488         for (i = 0; i < nb_rx_thread_params; ++i) {
2489                 queue = rx_thread_params[i].queue_id;
2490                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
2491                         printf("invalid queue number: %hhu\n", queue);
2492                         return -1;
2493                 }
2494                 lcore = rx_thread_params[i].lcore_id;
2495                 if (!rte_lcore_is_enabled(lcore)) {
2496                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
2497                         return -1;
2498                 }
2499                 socketid = rte_lcore_to_socket_id(lcore);
2500                 if ((socketid != 0) && (numa_on == 0))
2501                         printf("warning: lcore %hhu is on socket %d with numa off\n",
2502                                 lcore, socketid);
2503         }
2504         return 0;
2505 }
2506
2507 static int
2508 check_port_config(void)
2509 {
2510         unsigned portid;
2511         uint16_t i;
2512
2513         for (i = 0; i < nb_rx_thread_params; ++i) {
2514                 portid = rx_thread_params[i].port_id;
2515                 if ((enabled_port_mask & (1 << portid)) == 0) {
2516                         printf("port %u is not enabled in port mask\n", portid);
2517                         return -1;
2518                 }
2519                 if (!rte_eth_dev_is_valid_port(portid)) {
2520                         printf("port %u is not present on the board\n", portid);
2521                         return -1;
2522                 }
2523         }
2524         return 0;
2525 }
2526
2527 static uint8_t
2528 get_port_n_rx_queues(const uint16_t port)
2529 {
2530         int queue = -1;
2531         uint16_t i;
2532
2533         for (i = 0; i < nb_rx_thread_params; ++i)
2534                 if (rx_thread_params[i].port_id == port &&
2535                                 rx_thread_params[i].queue_id > queue)
2536                         queue = rx_thread_params[i].queue_id;
2537
2538         return (uint8_t)(++queue);
2539 }
2540
2541 static int
2542 init_rx_rings(void)
2543 {
2544         unsigned socket_io;
2545         struct thread_rx_conf *rx_conf;
2546         struct thread_tx_conf *tx_conf;
2547         unsigned rx_thread_id, tx_thread_id;
2548         char name[256];
2549         struct rte_ring *ring = NULL;
2550
2551         for (tx_thread_id = 0; tx_thread_id < n_tx_thread; tx_thread_id++) {
2552
2553                 tx_conf = &tx_thread[tx_thread_id];
2554
2555                 printf("Connecting tx-thread %d with rx-thread %d\n", tx_thread_id,
2556                                 tx_conf->conf.thread_id);
2557
2558                 rx_thread_id = tx_conf->conf.thread_id;
2559                 if (rx_thread_id > n_tx_thread) {
2560                         printf("connection from tx-thread %u to rx-thread %u fails "
2561                                         "(rx-thread not defined)\n", tx_thread_id, rx_thread_id);
2562                         return -1;
2563                 }
2564
2565                 rx_conf = &rx_thread[rx_thread_id];
2566                 socket_io = rte_lcore_to_socket_id(rx_conf->conf.lcore_id);
2567
2568                 snprintf(name, sizeof(name), "app_ring_s%u_rx%u_tx%u",
2569                                 socket_io, rx_thread_id, tx_thread_id);
2570
2571                 ring = rte_ring_create(name, 1024 * 4, socket_io,
2572                                 RING_F_SP_ENQ | RING_F_SC_DEQ);
2573
2574                 if (ring == NULL) {
2575                         rte_panic("Cannot create ring to connect rx-thread %u "
2576                                         "with tx-thread %u\n", rx_thread_id, tx_thread_id);
2577                 }
2578
2579                 rx_conf->ring[rx_conf->n_ring] = ring;
2580
2581                 tx_conf->ring = ring;
2582                 tx_conf->ready = &rx_conf->ready[rx_conf->n_ring];
2583
2584                 rx_conf->n_ring++;
2585         }
2586         return 0;
2587 }
2588
2589 static int
2590 init_rx_queues(void)
2591 {
2592         uint16_t i, nb_rx_queue;
2593         uint8_t thread;
2594
2595         n_rx_thread = 0;
2596
2597         for (i = 0; i < nb_rx_thread_params; ++i) {
2598                 thread = rx_thread_params[i].thread_id;
2599                 nb_rx_queue = rx_thread[thread].n_rx_queue;
2600
2601                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
2602                         printf("error: too many queues (%u) for thread: %u\n",
2603                                 (unsigned)nb_rx_queue + 1, (unsigned)thread);
2604                         return -1;
2605                 }
2606
2607                 rx_thread[thread].conf.thread_id = thread;
2608                 rx_thread[thread].conf.lcore_id = rx_thread_params[i].lcore_id;
2609                 rx_thread[thread].rx_queue_list[nb_rx_queue].port_id =
2610                         rx_thread_params[i].port_id;
2611                 rx_thread[thread].rx_queue_list[nb_rx_queue].queue_id =
2612                         rx_thread_params[i].queue_id;
2613                 rx_thread[thread].n_rx_queue++;
2614
2615                 if (thread >= n_rx_thread)
2616                         n_rx_thread = thread + 1;
2617
2618         }
2619         return 0;
2620 }
2621
2622 static int
2623 init_tx_threads(void)
2624 {
2625         int i;
2626
2627         n_tx_thread = 0;
2628         for (i = 0; i < nb_tx_thread_params; ++i) {
2629                 tx_thread[n_tx_thread].conf.thread_id = tx_thread_params[i].thread_id;
2630                 tx_thread[n_tx_thread].conf.lcore_id = tx_thread_params[i].lcore_id;
2631                 n_tx_thread++;
2632         }
2633         return 0;
2634 }
2635
2636 /* display usage */
2637 static void
2638 print_usage(const char *prgname)
2639 {
2640         printf("%s [EAL options] -- -p PORTMASK -P"
2641                 "  [--rx (port,queue,lcore,thread)[,(port,queue,lcore,thread]]"
2642                 "  [--tx (lcore,thread)[,(lcore,thread]]"
2643                 "  [--max-pkt-len PKTLEN]"
2644                 "  [--parse-ptype]\n\n"
2645                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
2646                 "  -P : enable promiscuous mode\n"
2647                 "  --rx (port,queue,lcore,thread): rx queues configuration\n"
2648                 "  --tx (lcore,thread): tx threads configuration\n"
2649                 "  --stat-lcore LCORE: use lcore for stat collector\n"
2650                 "  --eth-dest=X,MM:MM:MM:MM:MM:MM: optional, ethernet destination for port X\n"
2651                 "  --no-numa: optional, disable numa awareness\n"
2652                 "  --ipv6: optional, specify it if running ipv6 packets\n"
2653                 "  --max-pkt-len PKTLEN: maximum packet length in decimal (64-9600)\n"
2654                 "  --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n"
2655                 "  --no-lthreads: turn off lthread model\n"
2656                 "  --parse-ptype: set to use software to analyze packet type\n\n",
2657                 prgname);
2658 }
2659
2660 static int parse_max_pkt_len(const char *pktlen)
2661 {
2662         char *end = NULL;
2663         unsigned long len;
2664
2665         /* parse decimal string */
2666         len = strtoul(pktlen, &end, 10);
2667         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
2668                 return -1;
2669
2670         if (len == 0)
2671                 return -1;
2672
2673         return len;
2674 }
2675
2676 static int
2677 parse_portmask(const char *portmask)
2678 {
2679         char *end = NULL;
2680         unsigned long pm;
2681
2682         /* parse hexadecimal string */
2683         pm = strtoul(portmask, &end, 16);
2684         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
2685                 return 0;
2686
2687         return pm;
2688 }
2689
2690 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2691 static int
2692 parse_hash_entry_number(const char *hash_entry_num)
2693 {
2694         char *end = NULL;
2695         unsigned long hash_en;
2696
2697         /* parse hexadecimal string */
2698         hash_en = strtoul(hash_entry_num, &end, 16);
2699         if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
2700                 return -1;
2701
2702         if (hash_en == 0)
2703                 return -1;
2704
2705         return hash_en;
2706 }
2707 #endif
2708
2709 static int
2710 parse_rx_config(const char *q_arg)
2711 {
2712         char s[256];
2713         const char *p, *p0 = q_arg;
2714         char *end;
2715         enum fieldnames {
2716                 FLD_PORT = 0,
2717                 FLD_QUEUE,
2718                 FLD_LCORE,
2719                 FLD_THREAD,
2720                 _NUM_FLD
2721         };
2722         unsigned long int_fld[_NUM_FLD];
2723         char *str_fld[_NUM_FLD];
2724         int i;
2725         unsigned size;
2726
2727         nb_rx_thread_params = 0;
2728
2729         while ((p = strchr(p0, '(')) != NULL) {
2730                 ++p;
2731                 p0 = strchr(p, ')');
2732                 if (p0 == NULL)
2733                         return -1;
2734
2735                 size = p0 - p;
2736                 if (size >= sizeof(s))
2737                         return -1;
2738
2739                 snprintf(s, sizeof(s), "%.*s", size, p);
2740                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2741                         return -1;
2742                 for (i = 0; i < _NUM_FLD; i++) {
2743                         errno = 0;
2744                         int_fld[i] = strtoul(str_fld[i], &end, 0);
2745                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2746                                 return -1;
2747                 }
2748                 if (nb_rx_thread_params >= MAX_LCORE_PARAMS) {
2749                         printf("exceeded max number of rx params: %hu\n",
2750                                         nb_rx_thread_params);
2751                         return -1;
2752                 }
2753                 rx_thread_params_array[nb_rx_thread_params].port_id =
2754                                 int_fld[FLD_PORT];
2755                 rx_thread_params_array[nb_rx_thread_params].queue_id =
2756                                 (uint8_t)int_fld[FLD_QUEUE];
2757                 rx_thread_params_array[nb_rx_thread_params].lcore_id =
2758                                 (uint8_t)int_fld[FLD_LCORE];
2759                 rx_thread_params_array[nb_rx_thread_params].thread_id =
2760                                 (uint8_t)int_fld[FLD_THREAD];
2761                 ++nb_rx_thread_params;
2762         }
2763         rx_thread_params = rx_thread_params_array;
2764         return 0;
2765 }
2766
2767 static int
2768 parse_tx_config(const char *q_arg)
2769 {
2770         char s[256];
2771         const char *p, *p0 = q_arg;
2772         char *end;
2773         enum fieldnames {
2774                 FLD_LCORE = 0,
2775                 FLD_THREAD,
2776                 _NUM_FLD
2777         };
2778         unsigned long int_fld[_NUM_FLD];
2779         char *str_fld[_NUM_FLD];
2780         int i;
2781         unsigned size;
2782
2783         nb_tx_thread_params = 0;
2784
2785         while ((p = strchr(p0, '(')) != NULL) {
2786                 ++p;
2787                 p0 = strchr(p, ')');
2788                 if (p0 == NULL)
2789                         return -1;
2790
2791                 size = p0 - p;
2792                 if (size >= sizeof(s))
2793                         return -1;
2794
2795                 snprintf(s, sizeof(s), "%.*s", size, p);
2796                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2797                         return -1;
2798                 for (i = 0; i < _NUM_FLD; i++) {
2799                         errno = 0;
2800                         int_fld[i] = strtoul(str_fld[i], &end, 0);
2801                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2802                                 return -1;
2803                 }
2804                 if (nb_tx_thread_params >= MAX_LCORE_PARAMS) {
2805                         printf("exceeded max number of tx params: %hu\n",
2806                                 nb_tx_thread_params);
2807                         return -1;
2808                 }
2809                 tx_thread_params_array[nb_tx_thread_params].lcore_id =
2810                                 (uint8_t)int_fld[FLD_LCORE];
2811                 tx_thread_params_array[nb_tx_thread_params].thread_id =
2812                                 (uint8_t)int_fld[FLD_THREAD];
2813                 ++nb_tx_thread_params;
2814         }
2815         tx_thread_params = tx_thread_params_array;
2816
2817         return 0;
2818 }
2819
2820 #if (APP_CPU_LOAD > 0)
2821 static int
2822 parse_stat_lcore(const char *stat_lcore)
2823 {
2824         char *end = NULL;
2825         unsigned long lcore_id;
2826
2827         lcore_id = strtoul(stat_lcore, &end, 10);
2828         if ((stat_lcore[0] == '\0') || (end == NULL) || (*end != '\0'))
2829                 return -1;
2830
2831         return lcore_id;
2832 }
2833 #endif
2834
2835 static void
2836 parse_eth_dest(const char *optarg)
2837 {
2838         uint16_t portid;
2839         char *port_end;
2840         uint8_t c, *dest, peer_addr[6];
2841
2842         errno = 0;
2843         portid = strtoul(optarg, &port_end, 10);
2844         if (errno != 0 || port_end == optarg || *port_end++ != ',')
2845                 rte_exit(EXIT_FAILURE,
2846                 "Invalid eth-dest: %s", optarg);
2847         if (portid >= RTE_MAX_ETHPORTS)
2848                 rte_exit(EXIT_FAILURE,
2849                 "eth-dest: port %d >= RTE_MAX_ETHPORTS(%d)\n",
2850                 portid, RTE_MAX_ETHPORTS);
2851
2852         if (cmdline_parse_etheraddr(NULL, port_end,
2853                 &peer_addr, sizeof(peer_addr)) < 0)
2854                 rte_exit(EXIT_FAILURE,
2855                 "Invalid ethernet address: %s\n",
2856                 port_end);
2857         dest = (uint8_t *)&dest_eth_addr[portid];
2858         for (c = 0; c < 6; c++)
2859                 dest[c] = peer_addr[c];
2860         *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
2861 }
2862
2863 enum {
2864 #define OPT_RX_CONFIG       "rx"
2865         OPT_RX_CONFIG_NUM = 256,
2866 #define OPT_TX_CONFIG       "tx"
2867         OPT_TX_CONFIG_NUM,
2868 #define OPT_STAT_LCORE      "stat-lcore"
2869         OPT_STAT_LCORE_NUM,
2870 #define OPT_ETH_DEST        "eth-dest"
2871         OPT_ETH_DEST_NUM,
2872 #define OPT_NO_NUMA         "no-numa"
2873         OPT_NO_NUMA_NUM,
2874 #define OPT_IPV6            "ipv6"
2875         OPT_IPV6_NUM,
2876 #define OPT_MAX_PKT_LEN "max-pkt-len"
2877         OPT_MAX_PKT_LEN_NUM,
2878 #define OPT_HASH_ENTRY_NUM  "hash-entry-num"
2879         OPT_HASH_ENTRY_NUM_NUM,
2880 #define OPT_NO_LTHREADS     "no-lthreads"
2881         OPT_NO_LTHREADS_NUM,
2882 #define OPT_PARSE_PTYPE     "parse-ptype"
2883         OPT_PARSE_PTYPE_NUM,
2884 };
2885
2886 /* Parse the argument given in the command line of the application */
2887 static int
2888 parse_args(int argc, char **argv)
2889 {
2890         int opt, ret;
2891         char **argvopt;
2892         int option_index;
2893         char *prgname = argv[0];
2894         static struct option lgopts[] = {
2895                 {OPT_RX_CONFIG,      1, NULL, OPT_RX_CONFIG_NUM      },
2896                 {OPT_TX_CONFIG,      1, NULL, OPT_TX_CONFIG_NUM      },
2897                 {OPT_STAT_LCORE,     1, NULL, OPT_STAT_LCORE_NUM     },
2898                 {OPT_ETH_DEST,       1, NULL, OPT_ETH_DEST_NUM       },
2899                 {OPT_NO_NUMA,        0, NULL, OPT_NO_NUMA_NUM        },
2900                 {OPT_IPV6,           0, NULL, OPT_IPV6_NUM           },
2901                 {OPT_MAX_PKT_LEN,    1, NULL, OPT_MAX_PKT_LEN_NUM    },
2902                 {OPT_HASH_ENTRY_NUM, 1, NULL, OPT_HASH_ENTRY_NUM_NUM },
2903                 {OPT_NO_LTHREADS,    0, NULL, OPT_NO_LTHREADS_NUM    },
2904                 {OPT_PARSE_PTYPE,    0, NULL, OPT_PARSE_PTYPE_NUM    },
2905                 {NULL,               0, 0,    0                      }
2906         };
2907
2908         argvopt = argv;
2909
2910         while ((opt = getopt_long(argc, argvopt, "p:P",
2911                                 lgopts, &option_index)) != EOF) {
2912
2913                 switch (opt) {
2914                 /* portmask */
2915                 case 'p':
2916                         enabled_port_mask = parse_portmask(optarg);
2917                         if (enabled_port_mask == 0) {
2918                                 printf("invalid portmask\n");
2919                                 print_usage(prgname);
2920                                 return -1;
2921                         }
2922                         break;
2923
2924                 case 'P':
2925                         printf("Promiscuous mode selected\n");
2926                         promiscuous_on = 1;
2927                         break;
2928
2929                 /* long options */
2930                 case OPT_RX_CONFIG_NUM:
2931                         ret = parse_rx_config(optarg);
2932                         if (ret) {
2933                                 printf("invalid rx-config\n");
2934                                 print_usage(prgname);
2935                                 return -1;
2936                         }
2937                         break;
2938
2939                 case OPT_TX_CONFIG_NUM:
2940                         ret = parse_tx_config(optarg);
2941                         if (ret) {
2942                                 printf("invalid tx-config\n");
2943                                 print_usage(prgname);
2944                                 return -1;
2945                         }
2946                         break;
2947
2948 #if (APP_CPU_LOAD > 0)
2949                 case OPT_STAT_LCORE_NUM:
2950                         cpu_load_lcore_id = parse_stat_lcore(optarg);
2951                         break;
2952 #endif
2953
2954                 case OPT_ETH_DEST_NUM:
2955                         parse_eth_dest(optarg);
2956                         break;
2957
2958                 case OPT_NO_NUMA_NUM:
2959                         printf("numa is disabled\n");
2960                         numa_on = 0;
2961                         break;
2962
2963 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2964                 case OPT_IPV6_NUM:
2965                         printf("ipv6 is specified\n");
2966                         ipv6 = 1;
2967                         break;
2968 #endif
2969
2970                 case OPT_NO_LTHREADS_NUM:
2971                         printf("l-threads model is disabled\n");
2972                         lthreads_on = 0;
2973                         break;
2974
2975                 case OPT_PARSE_PTYPE_NUM:
2976                         printf("software packet type parsing enabled\n");
2977                         parse_ptype_on = 1;
2978                         break;
2979
2980                 case OPT_MAX_PKT_LEN_NUM:
2981                         max_pkt_len = parse_max_pkt_len(optarg);
2982                         break;
2983
2984 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2985                 case OPT_HASH_ENTRY_NUM_NUM:
2986                         ret = parse_hash_entry_number(optarg);
2987                         if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
2988                                 hash_entry_number = ret;
2989                         } else {
2990                                 printf("invalid hash entry number\n");
2991                                 print_usage(prgname);
2992                                 return -1;
2993                         }
2994                         break;
2995 #endif
2996
2997                 default:
2998                         print_usage(prgname);
2999                         return -1;
3000                 }
3001         }
3002
3003         if (optind >= 0)
3004                 argv[optind-1] = prgname;
3005
3006         ret = optind-1;
3007         optind = 1; /* reset getopt lib */
3008         return ret;
3009 }
3010
3011 static void
3012 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
3013 {
3014         char buf[RTE_ETHER_ADDR_FMT_SIZE];
3015
3016         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
3017         printf("%s%s", name, buf);
3018 }
3019
3020 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
3021
3022 static void convert_ipv4_5tuple(struct ipv4_5tuple *key1,
3023                 union ipv4_5tuple_host *key2)
3024 {
3025         key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
3026         key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
3027         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3028         key2->port_src = rte_cpu_to_be_16(key1->port_src);
3029         key2->proto = key1->proto;
3030         key2->pad0 = 0;
3031         key2->pad1 = 0;
3032 }
3033
3034 static void convert_ipv6_5tuple(struct ipv6_5tuple *key1,
3035                 union ipv6_5tuple_host *key2)
3036 {
3037         uint32_t i;
3038
3039         for (i = 0; i < 16; i++) {
3040                 key2->ip_dst[i] = key1->ip_dst[i];
3041                 key2->ip_src[i] = key1->ip_src[i];
3042         }
3043         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3044         key2->port_src = rte_cpu_to_be_16(key1->port_src);
3045         key2->proto = key1->proto;
3046         key2->pad0 = 0;
3047         key2->pad1 = 0;
3048         key2->reserve = 0;
3049 }
3050
3051 #define BYTE_VALUE_MAX 256
3052 #define ALL_32_BITS 0xffffffff
3053 #define BIT_8_TO_15 0x0000ff00
3054 static inline void
3055 populate_ipv4_few_flow_into_table(const struct rte_hash *h)
3056 {
3057         uint32_t i;
3058         int32_t ret;
3059         uint32_t array_len = RTE_DIM(ipv4_l3fwd_route_array);
3060
3061         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3062         for (i = 0; i < array_len; i++) {
3063                 struct ipv4_l3fwd_route  entry;
3064                 union ipv4_5tuple_host newkey;
3065
3066                 entry = ipv4_l3fwd_route_array[i];
3067                 convert_ipv4_5tuple(&entry.key, &newkey);
3068                 ret = rte_hash_add_key(h, (void *)&newkey);
3069                 if (ret < 0) {
3070                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3071                                 " to the l3fwd hash.\n", i);
3072                 }
3073                 ipv4_l3fwd_out_if[ret] = entry.if_out;
3074         }
3075         printf("Hash: Adding 0x%" PRIx32 " keys\n", array_len);
3076 }
3077
3078 #define BIT_16_TO_23 0x00ff0000
3079 static inline void
3080 populate_ipv6_few_flow_into_table(const struct rte_hash *h)
3081 {
3082         uint32_t i;
3083         int32_t ret;
3084         uint32_t array_len = RTE_DIM(ipv6_l3fwd_route_array);
3085
3086         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3087         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3088         for (i = 0; i < array_len; i++) {
3089                 struct ipv6_l3fwd_route entry;
3090                 union ipv6_5tuple_host newkey;
3091
3092                 entry = ipv6_l3fwd_route_array[i];
3093                 convert_ipv6_5tuple(&entry.key, &newkey);
3094                 ret = rte_hash_add_key(h, (void *)&newkey);
3095                 if (ret < 0) {
3096                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3097                                 " to the l3fwd hash.\n", i);
3098                 }
3099                 ipv6_l3fwd_out_if[ret] = entry.if_out;
3100         }
3101         printf("Hash: Adding 0x%" PRIx32 "keys\n", array_len);
3102 }
3103
3104 #define NUMBER_PORT_USED 4
3105 static inline void
3106 populate_ipv4_many_flow_into_table(const struct rte_hash *h,
3107                 unsigned int nr_flow)
3108 {
3109         unsigned i;
3110
3111         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3112
3113         for (i = 0; i < nr_flow; i++) {
3114                 struct ipv4_l3fwd_route entry;
3115                 union ipv4_5tuple_host newkey;
3116                 uint8_t a = (uint8_t)((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3117                 uint8_t b = (uint8_t)(((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3118                                 BYTE_VALUE_MAX);
3119                 uint8_t c = (uint8_t)((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3120                                 BYTE_VALUE_MAX));
3121                 /* Create the ipv4 exact match flow */
3122                 memset(&entry, 0, sizeof(entry));
3123                 switch (i & (NUMBER_PORT_USED - 1)) {
3124                 case 0:
3125                         entry = ipv4_l3fwd_route_array[0];
3126                         entry.key.ip_dst = RTE_IPV4(101, c, b, a);
3127                         break;
3128                 case 1:
3129                         entry = ipv4_l3fwd_route_array[1];
3130                         entry.key.ip_dst = RTE_IPV4(201, c, b, a);
3131                         break;
3132                 case 2:
3133                         entry = ipv4_l3fwd_route_array[2];
3134                         entry.key.ip_dst = RTE_IPV4(111, c, b, a);
3135                         break;
3136                 case 3:
3137                         entry = ipv4_l3fwd_route_array[3];
3138                         entry.key.ip_dst = RTE_IPV4(211, c, b, a);
3139                         break;
3140                 };
3141                 convert_ipv4_5tuple(&entry.key, &newkey);
3142                 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3143
3144                 if (ret < 0)
3145                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3146
3147                 ipv4_l3fwd_out_if[ret] = (uint8_t)entry.if_out;
3148
3149         }
3150         printf("Hash: Adding 0x%x keys\n", nr_flow);
3151 }
3152
3153 static inline void
3154 populate_ipv6_many_flow_into_table(const struct rte_hash *h,
3155                 unsigned int nr_flow)
3156 {
3157         unsigned i;
3158
3159         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3160         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3161         for (i = 0; i < nr_flow; i++) {
3162                 struct ipv6_l3fwd_route entry;
3163                 union ipv6_5tuple_host newkey;
3164
3165                 uint8_t a = (uint8_t) ((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3166                 uint8_t b = (uint8_t) (((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3167                                 BYTE_VALUE_MAX);
3168                 uint8_t c = (uint8_t) ((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3169                                 BYTE_VALUE_MAX));
3170
3171                 /* Create the ipv6 exact match flow */
3172                 memset(&entry, 0, sizeof(entry));
3173                 switch (i & (NUMBER_PORT_USED - 1)) {
3174                 case 0:
3175                         entry = ipv6_l3fwd_route_array[0];
3176                         break;
3177                 case 1:
3178                         entry = ipv6_l3fwd_route_array[1];
3179                         break;
3180                 case 2:
3181                         entry = ipv6_l3fwd_route_array[2];
3182                         break;
3183                 case 3:
3184                         entry = ipv6_l3fwd_route_array[3];
3185                         break;
3186                 };
3187                 entry.key.ip_dst[13] = c;
3188                 entry.key.ip_dst[14] = b;
3189                 entry.key.ip_dst[15] = a;
3190                 convert_ipv6_5tuple(&entry.key, &newkey);
3191                 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3192
3193                 if (ret < 0)
3194                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3195
3196                 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
3197
3198         }
3199         printf("Hash: Adding 0x%x keys\n", nr_flow);
3200 }
3201
3202 static void
3203 setup_hash(int socketid)
3204 {
3205         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
3206                 .name = NULL,
3207                 .entries = L3FWD_HASH_ENTRIES,
3208                 .key_len = sizeof(union ipv4_5tuple_host),
3209                 .hash_func = ipv4_hash_crc,
3210                 .hash_func_init_val = 0,
3211         };
3212
3213         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
3214                 .name = NULL,
3215                 .entries = L3FWD_HASH_ENTRIES,
3216                 .key_len = sizeof(union ipv6_5tuple_host),
3217                 .hash_func = ipv6_hash_crc,
3218                 .hash_func_init_val = 0,
3219         };
3220
3221         char s[64];
3222
3223         /* create ipv4 hash */
3224         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
3225         ipv4_l3fwd_hash_params.name = s;
3226         ipv4_l3fwd_hash_params.socket_id = socketid;
3227         ipv4_l3fwd_lookup_struct[socketid] =
3228                         rte_hash_create(&ipv4_l3fwd_hash_params);
3229         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3230                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3231                                 "socket %d\n", socketid);
3232
3233         /* create ipv6 hash */
3234         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
3235         ipv6_l3fwd_hash_params.name = s;
3236         ipv6_l3fwd_hash_params.socket_id = socketid;
3237         ipv6_l3fwd_lookup_struct[socketid] =
3238                         rte_hash_create(&ipv6_l3fwd_hash_params);
3239         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3240                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3241                                 "socket %d\n", socketid);
3242
3243         if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
3244                 /* For testing hash matching with a large number of flows we
3245                  * generate millions of IP 5-tuples with an incremented dst
3246                  * address to initialize the hash table. */
3247                 if (ipv6 == 0) {
3248                         /* populate the ipv4 hash */
3249                         populate_ipv4_many_flow_into_table(
3250                                 ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
3251                 } else {
3252                         /* populate the ipv6 hash */
3253                         populate_ipv6_many_flow_into_table(
3254                                 ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
3255                 }
3256         } else {
3257                 /* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize
3258                  * the hash table */
3259                 if (ipv6 == 0) {
3260                         /* populate the ipv4 hash */
3261                         populate_ipv4_few_flow_into_table(
3262                                         ipv4_l3fwd_lookup_struct[socketid]);
3263                 } else {
3264                         /* populate the ipv6 hash */
3265                         populate_ipv6_few_flow_into_table(
3266                                         ipv6_l3fwd_lookup_struct[socketid]);
3267                 }
3268         }
3269 }
3270 #endif
3271
3272 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3273 static void
3274 setup_lpm(int socketid)
3275 {
3276         struct rte_lpm6_config config;
3277         struct rte_lpm_config lpm_ipv4_config;
3278         unsigned i;
3279         int ret;
3280         char s[64];
3281
3282         /* create the LPM table */
3283         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
3284         lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
3285         lpm_ipv4_config.number_tbl8s = 256;
3286         lpm_ipv4_config.flags = 0;
3287         ipv4_l3fwd_lookup_struct[socketid] =
3288                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
3289         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3290                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3291                                 " on socket %d\n", socketid);
3292
3293         /* populate the LPM table */
3294         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
3295
3296                 /* skip unused ports */
3297                 if ((1 << ipv4_l3fwd_route_array[i].if_out &
3298                                 enabled_port_mask) == 0)
3299                         continue;
3300
3301                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
3302                         ipv4_l3fwd_route_array[i].ip,
3303                         ipv4_l3fwd_route_array[i].depth,
3304                         ipv4_l3fwd_route_array[i].if_out);
3305
3306                 if (ret < 0) {
3307                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3308                                 "l3fwd LPM table on socket %d\n",
3309                                 i, socketid);
3310                 }
3311
3312                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
3313                         (unsigned)ipv4_l3fwd_route_array[i].ip,
3314                         ipv4_l3fwd_route_array[i].depth,
3315                         ipv4_l3fwd_route_array[i].if_out);
3316         }
3317
3318         /* create the LPM6 table */
3319         snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
3320
3321         config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
3322         config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
3323         config.flags = 0;
3324         ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
3325                                 &config);
3326         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3327                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3328                                 " on socket %d\n", socketid);
3329
3330         /* populate the LPM table */
3331         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
3332
3333                 /* skip unused ports */
3334                 if ((1 << ipv6_l3fwd_route_array[i].if_out &
3335                                 enabled_port_mask) == 0)
3336                         continue;
3337
3338                 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
3339                         ipv6_l3fwd_route_array[i].ip,
3340                         ipv6_l3fwd_route_array[i].depth,
3341                         ipv6_l3fwd_route_array[i].if_out);
3342
3343                 if (ret < 0) {
3344                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3345                                 "l3fwd LPM table on socket %d\n",
3346                                 i, socketid);
3347                 }
3348
3349                 printf("LPM: Adding route %s / %d (%d)\n",
3350                         "IPV6",
3351                         ipv6_l3fwd_route_array[i].depth,
3352                         ipv6_l3fwd_route_array[i].if_out);
3353         }
3354 }
3355 #endif
3356
3357 static int
3358 init_mem(unsigned nb_mbuf)
3359 {
3360         struct lcore_conf *qconf;
3361         int socketid;
3362         unsigned lcore_id;
3363         char s[64];
3364
3365         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3366                 if (rte_lcore_is_enabled(lcore_id) == 0)
3367                         continue;
3368
3369                 if (numa_on)
3370                         socketid = rte_lcore_to_socket_id(lcore_id);
3371                 else
3372                         socketid = 0;
3373
3374                 if (socketid >= NB_SOCKETS) {
3375                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
3376                                 socketid, lcore_id, NB_SOCKETS);
3377                 }
3378                 if (pktmbuf_pool[socketid] == NULL) {
3379                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
3380                         pktmbuf_pool[socketid] =
3381                                 rte_pktmbuf_pool_create(s, nb_mbuf,
3382                                         MEMPOOL_CACHE_SIZE, 0,
3383                                         RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
3384                         if (pktmbuf_pool[socketid] == NULL)
3385                                 rte_exit(EXIT_FAILURE,
3386                                                 "Cannot init mbuf pool on socket %d\n", socketid);
3387                         else
3388                                 printf("Allocated mbuf pool on socket %d\n", socketid);
3389
3390 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3391                         setup_lpm(socketid);
3392 #else
3393                         setup_hash(socketid);
3394 #endif
3395                 }
3396                 qconf = &lcore_conf[lcore_id];
3397                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
3398                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
3399         }
3400         return 0;
3401 }
3402
3403 /* Check the link status of all ports in up to 9s, and print them finally */
3404 static void
3405 check_all_ports_link_status(uint32_t port_mask)
3406 {
3407 #define CHECK_INTERVAL 100 /* 100ms */
3408 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
3409         uint16_t portid;
3410         uint8_t count, all_ports_up, print_flag = 0;
3411         struct rte_eth_link link;
3412         int ret;
3413         char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
3414
3415         printf("\nChecking link status");
3416         fflush(stdout);
3417         for (count = 0; count <= MAX_CHECK_TIME; count++) {
3418                 all_ports_up = 1;
3419                 RTE_ETH_FOREACH_DEV(portid) {
3420                         if ((port_mask & (1 << portid)) == 0)
3421                                 continue;
3422                         memset(&link, 0, sizeof(link));
3423                         ret = rte_eth_link_get_nowait(portid, &link);
3424                         if (ret < 0) {
3425                                 all_ports_up = 0;
3426                                 if (print_flag == 1)
3427                                         printf("Port %u link get failed: %s\n",
3428                                                 portid, rte_strerror(-ret));
3429                                 continue;
3430                         }
3431                         /* print link status if flag set */
3432                         if (print_flag == 1) {
3433                                 rte_eth_link_to_str(link_status_text,
3434                                         sizeof(link_status_text), &link);
3435                                 printf("Port %d %s\n", portid,
3436                                         link_status_text);
3437                                 continue;
3438                         }
3439                         /* clear all_ports_up flag if any link down */
3440                         if (link.link_status == ETH_LINK_DOWN) {
3441                                 all_ports_up = 0;
3442                                 break;
3443                         }
3444                 }
3445                 /* after finally printing all link status, get out */
3446                 if (print_flag == 1)
3447                         break;
3448
3449                 if (all_ports_up == 0) {
3450                         printf(".");
3451                         fflush(stdout);
3452                         rte_delay_ms(CHECK_INTERVAL);
3453                 }
3454
3455                 /* set the print_flag if all ports up or timeout */
3456                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
3457                         print_flag = 1;
3458                         printf("done\n");
3459                 }
3460         }
3461 }
3462
3463 static uint32_t
3464 eth_dev_get_overhead_len(uint32_t max_rx_pktlen, uint16_t max_mtu)
3465 {
3466         uint32_t overhead_len;
3467
3468         if (max_mtu != UINT16_MAX && max_rx_pktlen > max_mtu)
3469                 overhead_len = max_rx_pktlen - max_mtu;
3470         else
3471                 overhead_len = RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN;
3472
3473         return overhead_len;
3474 }
3475
3476 static int
3477 config_port_max_pkt_len(struct rte_eth_conf *conf,
3478                 struct rte_eth_dev_info *dev_info)
3479 {
3480         uint32_t overhead_len;
3481
3482         if (max_pkt_len == 0)
3483                 return 0;
3484
3485         if (max_pkt_len < RTE_ETHER_MIN_LEN || max_pkt_len > MAX_JUMBO_PKT_LEN)
3486                 return -1;
3487
3488         overhead_len = eth_dev_get_overhead_len(dev_info->max_rx_pktlen,
3489                         dev_info->max_mtu);
3490         conf->rxmode.mtu = max_pkt_len - overhead_len;
3491
3492         if (conf->rxmode.mtu > RTE_ETHER_MTU)
3493                 conf->txmode.offloads |= DEV_TX_OFFLOAD_MULTI_SEGS;
3494
3495         return 0;
3496 }
3497
3498 int
3499 main(int argc, char **argv)
3500 {
3501         struct rte_eth_dev_info dev_info;
3502         struct rte_eth_txconf *txconf;
3503         int ret;
3504         int i;
3505         unsigned nb_ports;
3506         uint16_t queueid, portid;
3507         unsigned lcore_id;
3508         uint32_t n_tx_queue, nb_lcores;
3509         uint8_t nb_rx_queue, queue, socketid;
3510
3511         /* init EAL */
3512         ret = rte_eal_init(argc, argv);
3513         if (ret < 0)
3514                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
3515         argc -= ret;
3516         argv += ret;
3517
3518         ret = rte_timer_subsystem_init();
3519         if (ret < 0)
3520                 rte_exit(EXIT_FAILURE, "Failed to initialize timer subystem\n");
3521
3522         /* pre-init dst MACs for all ports to 02:00:00:00:00:xx */
3523         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
3524                 dest_eth_addr[portid] = RTE_ETHER_LOCAL_ADMIN_ADDR +
3525                                 ((uint64_t)portid << 40);
3526                 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
3527         }
3528
3529         /* parse application arguments (after the EAL ones) */
3530         ret = parse_args(argc, argv);
3531         if (ret < 0)
3532                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
3533
3534         if (check_lcore_params() < 0)
3535                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
3536
3537         printf("Initializing rx-queues...\n");
3538         ret = init_rx_queues();
3539         if (ret < 0)
3540                 rte_exit(EXIT_FAILURE, "init_rx_queues failed\n");
3541
3542         printf("Initializing tx-threads...\n");
3543         ret = init_tx_threads();
3544         if (ret < 0)
3545                 rte_exit(EXIT_FAILURE, "init_tx_threads failed\n");
3546
3547         printf("Initializing rings...\n");
3548         ret = init_rx_rings();
3549         if (ret < 0)
3550                 rte_exit(EXIT_FAILURE, "init_rx_rings failed\n");
3551
3552         nb_ports = rte_eth_dev_count_avail();
3553
3554         if (check_port_config() < 0)
3555                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
3556
3557         nb_lcores = rte_lcore_count();
3558
3559         /* initialize all ports */
3560         RTE_ETH_FOREACH_DEV(portid) {
3561                 struct rte_eth_conf local_port_conf = port_conf;
3562
3563                 /* skip ports that are not enabled */
3564                 if ((enabled_port_mask & (1 << portid)) == 0) {
3565                         printf("\nSkipping disabled port %d\n", portid);
3566                         continue;
3567                 }
3568
3569                 /* init port */
3570                 printf("Initializing port %d ... ", portid);
3571                 fflush(stdout);
3572
3573                 nb_rx_queue = get_port_n_rx_queues(portid);
3574                 n_tx_queue = nb_lcores;
3575                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
3576                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
3577                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
3578                         nb_rx_queue, (unsigned)n_tx_queue);
3579
3580                 ret = rte_eth_dev_info_get(portid, &dev_info);
3581                 if (ret != 0)
3582                         rte_exit(EXIT_FAILURE,
3583                                 "Error during getting device (port %u) info: %s\n",
3584                                 portid, strerror(-ret));
3585
3586                 ret = config_port_max_pkt_len(&local_port_conf, &dev_info);
3587                 if (ret != 0)
3588                         rte_exit(EXIT_FAILURE,
3589                                 "Invalid max packet length: %u (port %u)\n",
3590                                 max_pkt_len, portid);
3591
3592                 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
3593                         local_port_conf.txmode.offloads |=
3594                                 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
3595
3596                 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
3597                         dev_info.flow_type_rss_offloads;
3598                 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
3599                                 port_conf.rx_adv_conf.rss_conf.rss_hf) {
3600                         printf("Port %u modified RSS hash function based on hardware support,"
3601                                 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
3602                                 portid,
3603                                 port_conf.rx_adv_conf.rss_conf.rss_hf,
3604                                 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
3605                 }
3606
3607                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
3608                                         (uint16_t)n_tx_queue, &local_port_conf);
3609                 if (ret < 0)
3610                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
3611                                 ret, portid);
3612
3613                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
3614                                                        &nb_txd);
3615                 if (ret < 0)
3616                         rte_exit(EXIT_FAILURE,
3617                                  "rte_eth_dev_adjust_nb_rx_tx_desc: err=%d, port=%d\n",
3618                                  ret, portid);
3619
3620                 ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
3621                 if (ret < 0)
3622                         rte_exit(EXIT_FAILURE,
3623                                  "rte_eth_macaddr_get: err=%d, port=%d\n",
3624                                  ret, portid);
3625
3626                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
3627                 printf(", ");
3628                 print_ethaddr("Destination:",
3629                         (const struct rte_ether_addr *)&dest_eth_addr[portid]);
3630                 printf(", ");
3631
3632                 /*
3633                  * prepare src MACs for each port.
3634                  */
3635                 rte_ether_addr_copy(&ports_eth_addr[portid],
3636                         (struct rte_ether_addr *)(val_eth + portid) + 1);
3637
3638                 /* init memory */
3639                 ret = init_mem(NB_MBUF);
3640                 if (ret < 0)
3641                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
3642
3643                 /* init one TX queue per couple (lcore,port) */
3644                 queueid = 0;
3645                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3646                         if (rte_lcore_is_enabled(lcore_id) == 0)
3647                                 continue;
3648
3649                         if (numa_on)
3650                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3651                         else
3652                                 socketid = 0;
3653
3654                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
3655                         fflush(stdout);
3656
3657                         txconf = &dev_info.default_txconf;
3658                         txconf->offloads = local_port_conf.txmode.offloads;
3659                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
3660                                                      socketid, txconf);
3661                         if (ret < 0)
3662                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
3663                                         "port=%d\n", ret, portid);
3664
3665                         tx_thread[lcore_id].tx_queue_id[portid] = queueid;
3666                         queueid++;
3667                 }
3668                 printf("\n");
3669         }
3670
3671         for (i = 0; i < n_rx_thread; i++) {
3672                 lcore_id = rx_thread[i].conf.lcore_id;
3673
3674                 if (rte_lcore_is_enabled(lcore_id) == 0) {
3675                         rte_exit(EXIT_FAILURE,
3676                                         "Cannot start Rx thread on lcore %u: lcore disabled\n",
3677                                         lcore_id
3678                                 );
3679                 }
3680
3681                 printf("\nInitializing rx queues for Rx thread %d on lcore %u ... ",
3682                                 i, lcore_id);
3683                 fflush(stdout);
3684
3685                 /* init RX queues */
3686                 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3687                         struct rte_eth_rxconf rxq_conf;
3688
3689                         portid = rx_thread[i].rx_queue_list[queue].port_id;
3690                         queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3691
3692                         if (numa_on)
3693                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3694                         else
3695                                 socketid = 0;
3696
3697                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
3698                         fflush(stdout);
3699
3700                         ret = rte_eth_dev_info_get(portid, &dev_info);
3701                         if (ret != 0)
3702                                 rte_exit(EXIT_FAILURE,
3703                                         "Error during getting device (port %u) info: %s\n",
3704                                         portid, strerror(-ret));
3705
3706                         rxq_conf = dev_info.default_rxconf;
3707                         rxq_conf.offloads = port_conf.rxmode.offloads;
3708                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
3709                                         socketid,
3710                                         &rxq_conf,
3711                                         pktmbuf_pool[socketid]);
3712                         if (ret < 0)
3713                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, "
3714                                                 "port=%d\n", ret, portid);
3715                 }
3716         }
3717
3718         printf("\n");
3719
3720         /* start ports */
3721         RTE_ETH_FOREACH_DEV(portid) {
3722                 if ((enabled_port_mask & (1 << portid)) == 0)
3723                         continue;
3724
3725                 /* Start device */
3726                 ret = rte_eth_dev_start(portid);
3727                 if (ret < 0)
3728                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
3729                                 ret, portid);
3730
3731                 /*
3732                  * If enabled, put device in promiscuous mode.
3733                  * This allows IO forwarding mode to forward packets
3734                  * to itself through 2 cross-connected  ports of the
3735                  * target machine.
3736                  */
3737                 if (promiscuous_on) {
3738                         ret = rte_eth_promiscuous_enable(portid);
3739                         if (ret != 0)
3740                                 rte_exit(EXIT_FAILURE,
3741                                         "rte_eth_promiscuous_enable: err=%s, port=%u\n",
3742                                         rte_strerror(-ret), portid);
3743                 }
3744         }
3745
3746         for (i = 0; i < n_rx_thread; i++) {
3747                 lcore_id = rx_thread[i].conf.lcore_id;
3748                 if (rte_lcore_is_enabled(lcore_id) == 0)
3749                         continue;
3750
3751                 /* check if hw packet type is supported */
3752                 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3753                         portid = rx_thread[i].rx_queue_list[queue].port_id;
3754                         queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3755
3756                         if (parse_ptype_on) {
3757                                 if (!rte_eth_add_rx_callback(portid, queueid,
3758                                                 cb_parse_ptype, NULL))
3759                                         rte_exit(EXIT_FAILURE,
3760                                                 "Failed to add rx callback: "
3761                                                 "port=%d\n", portid);
3762                         } else if (!check_ptype(portid))
3763                                 rte_exit(EXIT_FAILURE,
3764                                         "Port %d cannot parse packet type.\n\n"
3765                                         "Please add --parse-ptype to use sw "
3766                                         "packet type analyzer.\n\n",
3767                                         portid);
3768                 }
3769         }
3770
3771         check_all_ports_link_status(enabled_port_mask);
3772
3773         if (lthreads_on) {
3774                 printf("Starting L-Threading Model\n");
3775
3776 #if (APP_CPU_LOAD > 0)
3777                 if (cpu_load_lcore_id > 0)
3778                         /* Use one lcore for cpu load collector */
3779                         nb_lcores--;
3780 #endif
3781
3782                 lthread_num_schedulers_set(nb_lcores);
3783                 rte_eal_mp_remote_launch(sched_spawner, NULL, SKIP_MAIN);
3784                 lthread_main_spawner(NULL);
3785
3786         } else {
3787                 printf("Starting P-Threading Model\n");
3788                 /* launch per-lcore init on every lcore */
3789                 rte_eal_mp_remote_launch(pthread_run, NULL, CALL_MAIN);
3790                 RTE_LCORE_FOREACH_WORKER(lcore_id) {
3791                         if (rte_eal_wait_lcore(lcore_id) < 0)
3792                                 return -1;
3793                 }
3794         }
3795
3796         /* clean up the EAL */
3797         rte_eal_cleanup();
3798
3799         return 0;
3800 }