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 = RTE_ETH_MQ_RX_RSS,
 310                 .split_hdr_size = 0,
 311                 .offloads = RTE_ETH_RX_OFFLOAD_CHECKSUM,
 312         },
 313         .rx_adv_conf = {
 314                 .rss_conf = {
 315                         .rss_key = NULL,
 316                         .rss_hf = RTE_ETH_RSS_TCP,
 317                 },
 318         },
 319         .txmode = {
 320                 .mq_mode = RTE_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 prev_tsc, diff_tsc, cur_tsc;
1889         uint64_t total[MAX_CPU] = { 0 };
1890         unsigned min_cpu = MAX_CPU;
1891         unsigned max_cpu = 0;
1892         unsigned cpu_id;
1893         int busy_total = 0;
1894         int busy_flag = 0;
1895
1896         unsigned int n_thread_per_cpu[MAX_CPU] = { 0 };
1897         struct thread_conf *thread_per_cpu[MAX_CPU][MAX_THREAD];
1898
1899         struct thread_conf *thread_conf;
1900
1901         const uint64_t interval_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1902                 US_PER_S * CPU_LOAD_TIMEOUT_US;
1903
1904         prev_tsc = 0;
1905         /*
1906          * Wait for all threads
1907          */
1908
1909         printf("Waiting for %d rx threads and %d tx threads\n", n_rx_thread,
1910                         n_tx_thread);
1911
1912         rte_wait_until_equal_16(&rx_counter, n_rx_thread, __ATOMIC_RELAXED);
1913         rte_wait_until_equal_16(&tx_counter, n_tx_thread, __ATOMIC_RELAXED);
1914
1915         for (i = 0; i < n_rx_thread; i++) {
1916
1917                 thread_conf = &rx_thread[i].conf;
1918                 cpu_id = thread_conf->cpu_id;
1919                 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1920
1921                 if (cpu_id > max_cpu)
1922                         max_cpu = cpu_id;
1923                 if (cpu_id < min_cpu)
1924                         min_cpu = cpu_id;
1925         }
1926         for (i = 0; i < n_tx_thread; i++) {
1927
1928                 thread_conf = &tx_thread[i].conf;
1929                 cpu_id = thread_conf->cpu_id;
1930                 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf;
1931
1932                 if (thread_conf->cpu_id > max_cpu)
1933                         max_cpu = thread_conf->cpu_id;
1934                 if (thread_conf->cpu_id < min_cpu)
1935                         min_cpu = thread_conf->cpu_id;
1936         }
1937
1938         while (1) {
1939
1940                 cpu_load.counter++;
1941                 for (i = min_cpu; i <= max_cpu; i++) {
1942                         for (j = 0; j < MAX_CPU_COUNTER; j++) {
1943                                 for (k = 0; k < n_thread_per_cpu[i]; k++)
1944                                         if (thread_per_cpu[i][k]->busy[j]) {
1945                                                 busy_flag = 1;
1946                                                 break;
1947                                         }
1948                                 if (busy_flag) {
1949                                         cpu_load.hits[j][i]++;
1950                                         busy_total = 1;
1951                                         busy_flag = 0;
1952                                 }
1953                         }
1954
1955                         if (busy_total) {
1956                                 total[i]++;
1957                                 busy_total = 0;
1958                         }
1959                 }
1960
1961                 cur_tsc = rte_rdtsc();
1962
1963                 diff_tsc = cur_tsc - prev_tsc;
1964                 if (unlikely(diff_tsc > interval_tsc)) {
1965
1966                         printf("\033c");
1967
1968                         printf("Cpu usage for %d rx threads and %d tx threads:\n\n",
1969                                         n_rx_thread, n_tx_thread);
1970
1971                         printf("cpu#     proc%%  poll%%  overhead%%\n\n");
1972
1973                         for (i = min_cpu; i <= max_cpu; i++) {
1974                                 printf("CPU %d:", i);
1975                                 for (j = 0; j < MAX_CPU_COUNTER; j++) {
1976                                         printf("%7" PRIu64 "",
1977                                                         cpu_load.hits[j][i] * 100 / cpu_load.counter);
1978                                         cpu_load.hits[j][i] = 0;
1979                                 }
1980                                 printf("%7" PRIu64 "\n",
1981                                                 100 - total[i] * 100 / cpu_load.counter);
1982                                 total[i] = 0;
1983                         }
1984                         cpu_load.counter = 0;
1985
1986                         prev_tsc = cur_tsc;
1987                 }
1988
1989         }
1990 }
1991 #endif /* APP_CPU_LOAD */
1992
1993 /*
1994  * Null processing lthread loop
1995  *
1996  * This loop is used to start empty scheduler on lcore.
1997  */
1998 static void *
1999 lthread_null(__rte_unused void *args)
2000 {
2001         int lcore_id = rte_lcore_id();
2002
2003         RTE_LOG(INFO, L3FWD, "Starting scheduler on lcore %d.\n", lcore_id);
2004         lthread_exit(NULL);
2005         return NULL;
2006 }
2007
2008 /* main processing loop */
2009 static void *
2010 lthread_tx_per_ring(void *dummy)
2011 {
2012         int nb_rx;
2013         uint16_t portid;
2014         struct rte_ring *ring;
2015         struct thread_tx_conf *tx_conf;
2016         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2017         struct lthread_cond *ready;
2018
2019         tx_conf = (struct thread_tx_conf *)dummy;
2020         ring = tx_conf->ring;
2021         ready = *tx_conf->ready;
2022
2023         lthread_set_data((void *)tx_conf);
2024
2025         /*
2026          * Move this lthread to lcore
2027          */
2028         lthread_set_affinity(tx_conf->conf.lcore_id);
2029
2030         RTE_LOG(INFO, L3FWD, "entering main tx loop on lcore %u\n", rte_lcore_id());
2031
2032         nb_rx = 0;
2033         __atomic_fetch_add(&tx_counter, 1, __ATOMIC_RELAXED);
2034         while (1) {
2035
2036                 /*
2037                  * Read packet from ring
2038                  */
2039                 SET_CPU_BUSY(tx_conf, CPU_POLL);
2040                 nb_rx = rte_ring_sc_dequeue_burst(ring, (void **)pkts_burst,
2041                                 MAX_PKT_BURST, NULL);
2042                 SET_CPU_IDLE(tx_conf, CPU_POLL);
2043
2044                 if (nb_rx > 0) {
2045                         SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2046                         portid = pkts_burst[0]->port;
2047                         process_burst(pkts_burst, nb_rx, portid);
2048                         SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2049                         lthread_yield();
2050                 } else
2051                         lthread_cond_wait(ready, 0);
2052
2053         }
2054         return NULL;
2055 }
2056
2057 /*
2058  * Main tx-lthreads spawner lthread.
2059  *
2060  * This lthread is used to spawn one new lthread per ring from producers.
2061  *
2062  */
2063 static void *
2064 lthread_tx(void *args)
2065 {
2066         struct lthread *lt;
2067
2068         unsigned lcore_id;
2069         uint16_t portid;
2070         struct thread_tx_conf *tx_conf;
2071
2072         tx_conf = (struct thread_tx_conf *)args;
2073         lthread_set_data((void *)tx_conf);
2074
2075         /*
2076          * Move this lthread to the selected lcore
2077          */
2078         lthread_set_affinity(tx_conf->conf.lcore_id);
2079
2080         /*
2081          * Spawn tx readers (one per input ring)
2082          */
2083         lthread_create(&lt, tx_conf->conf.lcore_id, lthread_tx_per_ring,
2084                         (void *)tx_conf);
2085
2086         lcore_id = rte_lcore_id();
2087
2088         RTE_LOG(INFO, L3FWD, "Entering Tx main loop on lcore %u\n", lcore_id);
2089
2090         tx_conf->conf.cpu_id = sched_getcpu();
2091         while (1) {
2092
2093                 lthread_sleep(BURST_TX_DRAIN_US * 1000);
2094
2095                 /*
2096                  * TX burst queue drain
2097                  */
2098                 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2099                         if (tx_conf->tx_mbufs[portid].len == 0)
2100                                 continue;
2101                         SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2102                         send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2103                         SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2104                         tx_conf->tx_mbufs[portid].len = 0;
2105                 }
2106
2107         }
2108         return NULL;
2109 }
2110
2111 static void *
2112 lthread_rx(void *dummy)
2113 {
2114         int ret;
2115         uint16_t nb_rx;
2116         int i;
2117         uint16_t portid;
2118         uint8_t queueid;
2119         int worker_id;
2120         int len[RTE_MAX_LCORE] = { 0 };
2121         int old_len, new_len;
2122         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2123         struct thread_rx_conf *rx_conf;
2124
2125         rx_conf = (struct thread_rx_conf *)dummy;
2126         lthread_set_data((void *)rx_conf);
2127
2128         /*
2129          * Move this lthread to lcore
2130          */
2131         lthread_set_affinity(rx_conf->conf.lcore_id);
2132
2133         if (rx_conf->n_rx_queue == 0) {
2134                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", rte_lcore_id());
2135                 return NULL;
2136         }
2137
2138         RTE_LOG(INFO, L3FWD, "Entering main Rx loop on lcore %u\n", rte_lcore_id());
2139
2140         for (i = 0; i < rx_conf->n_rx_queue; i++) {
2141
2142                 portid = rx_conf->rx_queue_list[i].port_id;
2143                 queueid = rx_conf->rx_queue_list[i].queue_id;
2144                 RTE_LOG(INFO, L3FWD,
2145                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2146                                 rte_lcore_id(), portid, queueid);
2147         }
2148
2149         /*
2150          * Init all condition variables (one per rx thread)
2151          */
2152         for (i = 0; i < rx_conf->n_rx_queue; i++)
2153                 lthread_cond_init(NULL, &rx_conf->ready[i], NULL);
2154
2155         worker_id = 0;
2156
2157         rx_conf->conf.cpu_id = sched_getcpu();
2158         __atomic_fetch_add(&rx_counter, 1, __ATOMIC_RELAXED);
2159         while (1) {
2160
2161                 /*
2162                  * Read packet from RX queues
2163                  */
2164                 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2165                         portid = rx_conf->rx_queue_list[i].port_id;
2166                         queueid = rx_conf->rx_queue_list[i].queue_id;
2167
2168                         SET_CPU_BUSY(rx_conf, CPU_POLL);
2169                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2170                                 MAX_PKT_BURST);
2171                         SET_CPU_IDLE(rx_conf, CPU_POLL);
2172
2173                         if (nb_rx != 0) {
2174                                 worker_id = (worker_id + 1) % rx_conf->n_ring;
2175                                 old_len = len[worker_id];
2176
2177                                 SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2178                                 ret = rte_ring_sp_enqueue_burst(
2179                                                 rx_conf->ring[worker_id],
2180                                                 (void **) pkts_burst,
2181                                                 nb_rx, NULL);
2182
2183                                 new_len = old_len + ret;
2184
2185                                 if (new_len >= BURST_SIZE) {
2186                                         lthread_cond_signal(rx_conf->ready[worker_id]);
2187                                         new_len = 0;
2188                                 }
2189
2190                                 len[worker_id] = new_len;
2191
2192                                 if (unlikely(ret < nb_rx)) {
2193                                         uint32_t k;
2194
2195                                         for (k = ret; k < nb_rx; k++) {
2196                                                 struct rte_mbuf *m = pkts_burst[k];
2197
2198                                                 rte_pktmbuf_free(m);
2199                                         }
2200                                 }
2201                                 SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2202                         }
2203
2204                         lthread_yield();
2205                 }
2206         }
2207         return NULL;
2208 }
2209
2210 /*
2211  * Start scheduler with initial lthread on lcore
2212  *
2213  * This lthread loop spawns all rx and tx lthreads on main lcore
2214  */
2215
2216 static void *
2217 lthread_spawner(__rte_unused void *arg)
2218 {
2219         struct lthread *lt[MAX_THREAD];
2220         int i;
2221         int n_thread = 0;
2222
2223         printf("Entering lthread_spawner\n");
2224
2225         /*
2226          * Create producers (rx threads) on default lcore
2227          */
2228         for (i = 0; i < n_rx_thread; i++) {
2229                 rx_thread[i].conf.thread_id = i;
2230                 lthread_create(&lt[n_thread], -1, lthread_rx,
2231                                 (void *)&rx_thread[i]);
2232                 n_thread++;
2233         }
2234
2235         /*
2236          * Wait for all producers. Until some producers can be started on the same
2237          * scheduler as this lthread, yielding is required to let them to run and
2238          * prevent deadlock here.
2239          */
2240         while (__atomic_load_n(&rx_counter, __ATOMIC_RELAXED) < n_rx_thread)
2241                 lthread_sleep(100000);
2242
2243         /*
2244          * Create consumers (tx threads) on default lcore_id
2245          */
2246         for (i = 0; i < n_tx_thread; i++) {
2247                 tx_thread[i].conf.thread_id = i;
2248                 lthread_create(&lt[n_thread], -1, lthread_tx,
2249                                 (void *)&tx_thread[i]);
2250                 n_thread++;
2251         }
2252
2253         /*
2254          * Wait for all threads finished
2255          */
2256         for (i = 0; i < n_thread; i++)
2257                 lthread_join(lt[i], NULL);
2258
2259         return NULL;
2260 }
2261
2262 /*
2263  * Start main scheduler with initial lthread spawning rx and tx lthreads
2264  * (main_lthread_main).
2265  */
2266 static int
2267 lthread_main_spawner(__rte_unused void *arg) {
2268         struct lthread *lt;
2269         int lcore_id = rte_lcore_id();
2270
2271         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2272         lthread_create(&lt, -1, lthread_spawner, NULL);
2273         lthread_run();
2274
2275         return 0;
2276 }
2277
2278 /*
2279  * Start scheduler on lcore.
2280  */
2281 static int
2282 sched_spawner(__rte_unused void *arg) {
2283         struct lthread *lt;
2284         int lcore_id = rte_lcore_id();
2285
2286 #if (APP_CPU_LOAD)
2287         if (lcore_id == cpu_load_lcore_id) {
2288                 cpu_load_collector(arg);
2289                 return 0;
2290         }
2291 #endif /* APP_CPU_LOAD */
2292
2293         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2294         lthread_create(&lt, -1, lthread_null, NULL);
2295         lthread_run();
2296
2297         return 0;
2298 }
2299
2300 /* main processing loop */
2301 static int __rte_noreturn
2302 pthread_tx(void *dummy)
2303 {
2304         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2305         uint64_t prev_tsc, diff_tsc, cur_tsc;
2306         int nb_rx;
2307         uint16_t portid;
2308         struct thread_tx_conf *tx_conf;
2309
2310         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
2311                 US_PER_S * BURST_TX_DRAIN_US;
2312
2313         prev_tsc = 0;
2314
2315         tx_conf = (struct thread_tx_conf *)dummy;
2316
2317         RTE_LOG(INFO, L3FWD, "Entering main Tx loop on lcore %u\n", rte_lcore_id());
2318
2319         tx_conf->conf.cpu_id = sched_getcpu();
2320         __atomic_fetch_add(&tx_counter, 1, __ATOMIC_RELAXED);
2321         while (1) {
2322
2323                 cur_tsc = rte_rdtsc();
2324
2325                 /*
2326                  * TX burst queue drain
2327                  */
2328                 diff_tsc = cur_tsc - prev_tsc;
2329                 if (unlikely(diff_tsc > drain_tsc)) {
2330
2331                         /*
2332                          * This could be optimized (use queueid instead of
2333                          * portid), but it is not called so often
2334                          */
2335                         SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2336                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
2337                                 if (tx_conf->tx_mbufs[portid].len == 0)
2338                                         continue;
2339                                 send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid);
2340                                 tx_conf->tx_mbufs[portid].len = 0;
2341                         }
2342                         SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2343
2344                         prev_tsc = cur_tsc;
2345                 }
2346
2347                 /*
2348                  * Read packet from ring
2349                  */
2350                 SET_CPU_BUSY(tx_conf, CPU_POLL);
2351                 nb_rx = rte_ring_sc_dequeue_burst(tx_conf->ring,
2352                                 (void **)pkts_burst, MAX_PKT_BURST, NULL);
2353                 SET_CPU_IDLE(tx_conf, CPU_POLL);
2354
2355                 if (unlikely(nb_rx == 0)) {
2356                         sched_yield();
2357                         continue;
2358                 }
2359
2360                 SET_CPU_BUSY(tx_conf, CPU_PROCESS);
2361                 portid = pkts_burst[0]->port;
2362                 process_burst(pkts_burst, nb_rx, portid);
2363                 SET_CPU_IDLE(tx_conf, CPU_PROCESS);
2364
2365         }
2366 }
2367
2368 static int
2369 pthread_rx(void *dummy)
2370 {
2371         int i;
2372         int worker_id;
2373         uint32_t n;
2374         uint32_t nb_rx;
2375         unsigned lcore_id;
2376         uint8_t queueid;
2377         uint16_t portid;
2378         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
2379
2380         struct thread_rx_conf *rx_conf;
2381
2382         lcore_id = rte_lcore_id();
2383         rx_conf = (struct thread_rx_conf *)dummy;
2384
2385         if (rx_conf->n_rx_queue == 0) {
2386                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
2387                 return 0;
2388         }
2389
2390         RTE_LOG(INFO, L3FWD, "entering main rx loop on lcore %u\n", lcore_id);
2391
2392         for (i = 0; i < rx_conf->n_rx_queue; i++) {
2393
2394                 portid = rx_conf->rx_queue_list[i].port_id;
2395                 queueid = rx_conf->rx_queue_list[i].queue_id;
2396                 RTE_LOG(INFO, L3FWD,
2397                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
2398                                 lcore_id, portid, queueid);
2399         }
2400
2401         worker_id = 0;
2402         rx_conf->conf.cpu_id = sched_getcpu();
2403         __atomic_fetch_add(&rx_counter, 1, __ATOMIC_RELAXED);
2404         while (1) {
2405
2406                 /*
2407                  * Read packet from RX queues
2408                  */
2409                 for (i = 0; i < rx_conf->n_rx_queue; ++i) {
2410                         portid = rx_conf->rx_queue_list[i].port_id;
2411                         queueid = rx_conf->rx_queue_list[i].queue_id;
2412
2413                         SET_CPU_BUSY(rx_conf, CPU_POLL);
2414                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
2415                                 MAX_PKT_BURST);
2416                         SET_CPU_IDLE(rx_conf, CPU_POLL);
2417
2418                         if (nb_rx == 0) {
2419                                 sched_yield();
2420                                 continue;
2421                         }
2422
2423                         SET_CPU_BUSY(rx_conf, CPU_PROCESS);
2424                         worker_id = (worker_id + 1) % rx_conf->n_ring;
2425                         n = rte_ring_sp_enqueue_burst(rx_conf->ring[worker_id],
2426                                         (void **)pkts_burst, nb_rx, NULL);
2427
2428                         if (unlikely(n != nb_rx)) {
2429                                 uint32_t k;
2430
2431                                 for (k = n; k < nb_rx; k++) {
2432                                         struct rte_mbuf *m = pkts_burst[k];
2433
2434                                         rte_pktmbuf_free(m);
2435                                 }
2436                         }
2437
2438                         SET_CPU_IDLE(rx_conf, CPU_PROCESS);
2439
2440                 }
2441         }
2442 }
2443
2444 /*
2445  * P-Thread spawner.
2446  */
2447 static int
2448 pthread_run(__rte_unused void *arg) {
2449         int lcore_id = rte_lcore_id();
2450         int i;
2451
2452         for (i = 0; i < n_rx_thread; i++)
2453                 if (rx_thread[i].conf.lcore_id == lcore_id) {
2454                         printf("Start rx thread on %d...\n", lcore_id);
2455                         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2456                         RTE_PER_LCORE(lcore_conf)->data = (void *)&rx_thread[i];
2457                         pthread_rx((void *)&rx_thread[i]);
2458                         return 0;
2459                 }
2460
2461         for (i = 0; i < n_tx_thread; i++)
2462                 if (tx_thread[i].conf.lcore_id == lcore_id) {
2463                         printf("Start tx thread on %d...\n", lcore_id);
2464                         RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id];
2465                         RTE_PER_LCORE(lcore_conf)->data = (void *)&tx_thread[i];
2466                         pthread_tx((void *)&tx_thread[i]);
2467                         return 0;
2468                 }
2469
2470 #if (APP_CPU_LOAD)
2471         if (lcore_id == cpu_load_lcore_id)
2472                 cpu_load_collector(arg);
2473 #endif /* APP_CPU_LOAD */
2474
2475         return 0;
2476 }
2477
2478 static int
2479 check_lcore_params(void)
2480 {
2481         uint8_t queue, lcore;
2482         uint16_t i;
2483         int socketid;
2484
2485         for (i = 0; i < nb_rx_thread_params; ++i) {
2486                 queue = rx_thread_params[i].queue_id;
2487                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
2488                         printf("invalid queue number: %hhu\n", queue);
2489                         return -1;
2490                 }
2491                 lcore = rx_thread_params[i].lcore_id;
2492                 if (!rte_lcore_is_enabled(lcore)) {
2493                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
2494                         return -1;
2495                 }
2496                 socketid = rte_lcore_to_socket_id(lcore);
2497                 if ((socketid != 0) && (numa_on == 0))
2498                         printf("warning: lcore %hhu is on socket %d with numa off\n",
2499                                 lcore, socketid);
2500         }
2501         return 0;
2502 }
2503
2504 static int
2505 check_port_config(void)
2506 {
2507         unsigned portid;
2508         uint16_t i;
2509
2510         for (i = 0; i < nb_rx_thread_params; ++i) {
2511                 portid = rx_thread_params[i].port_id;
2512                 if ((enabled_port_mask & (1 << portid)) == 0) {
2513                         printf("port %u is not enabled in port mask\n", portid);
2514                         return -1;
2515                 }
2516                 if (!rte_eth_dev_is_valid_port(portid)) {
2517                         printf("port %u is not present on the board\n", portid);
2518                         return -1;
2519                 }
2520         }
2521         return 0;
2522 }
2523
2524 static uint8_t
2525 get_port_n_rx_queues(const uint16_t port)
2526 {
2527         int queue = -1;
2528         uint16_t i;
2529
2530         for (i = 0; i < nb_rx_thread_params; ++i)
2531                 if (rx_thread_params[i].port_id == port &&
2532                                 rx_thread_params[i].queue_id > queue)
2533                         queue = rx_thread_params[i].queue_id;
2534
2535         return (uint8_t)(++queue);
2536 }
2537
2538 static int
2539 init_rx_rings(void)
2540 {
2541         unsigned socket_io;
2542         struct thread_rx_conf *rx_conf;
2543         struct thread_tx_conf *tx_conf;
2544         unsigned rx_thread_id, tx_thread_id;
2545         char name[256];
2546         struct rte_ring *ring = NULL;
2547
2548         for (tx_thread_id = 0; tx_thread_id < n_tx_thread; tx_thread_id++) {
2549
2550                 tx_conf = &tx_thread[tx_thread_id];
2551
2552                 printf("Connecting tx-thread %d with rx-thread %d\n", tx_thread_id,
2553                                 tx_conf->conf.thread_id);
2554
2555                 rx_thread_id = tx_conf->conf.thread_id;
2556                 if (rx_thread_id > n_tx_thread) {
2557                         printf("connection from tx-thread %u to rx-thread %u fails "
2558                                         "(rx-thread not defined)\n", tx_thread_id, rx_thread_id);
2559                         return -1;
2560                 }
2561
2562                 rx_conf = &rx_thread[rx_thread_id];
2563                 socket_io = rte_lcore_to_socket_id(rx_conf->conf.lcore_id);
2564
2565                 snprintf(name, sizeof(name), "app_ring_s%u_rx%u_tx%u",
2566                                 socket_io, rx_thread_id, tx_thread_id);
2567
2568                 ring = rte_ring_create(name, 1024 * 4, socket_io,
2569                                 RING_F_SP_ENQ | RING_F_SC_DEQ);
2570
2571                 if (ring == NULL) {
2572                         rte_panic("Cannot create ring to connect rx-thread %u "
2573                                         "with tx-thread %u\n", rx_thread_id, tx_thread_id);
2574                 }
2575
2576                 rx_conf->ring[rx_conf->n_ring] = ring;
2577
2578                 tx_conf->ring = ring;
2579                 tx_conf->ready = &rx_conf->ready[rx_conf->n_ring];
2580
2581                 rx_conf->n_ring++;
2582         }
2583         return 0;
2584 }
2585
2586 static int
2587 init_rx_queues(void)
2588 {
2589         uint16_t i, nb_rx_queue;
2590         uint8_t thread;
2591
2592         n_rx_thread = 0;
2593
2594         for (i = 0; i < nb_rx_thread_params; ++i) {
2595                 thread = rx_thread_params[i].thread_id;
2596                 nb_rx_queue = rx_thread[thread].n_rx_queue;
2597
2598                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
2599                         printf("error: too many queues (%u) for thread: %u\n",
2600                                 (unsigned)nb_rx_queue + 1, (unsigned)thread);
2601                         return -1;
2602                 }
2603
2604                 rx_thread[thread].conf.thread_id = thread;
2605                 rx_thread[thread].conf.lcore_id = rx_thread_params[i].lcore_id;
2606                 rx_thread[thread].rx_queue_list[nb_rx_queue].port_id =
2607                         rx_thread_params[i].port_id;
2608                 rx_thread[thread].rx_queue_list[nb_rx_queue].queue_id =
2609                         rx_thread_params[i].queue_id;
2610                 rx_thread[thread].n_rx_queue++;
2611
2612                 if (thread >= n_rx_thread)
2613                         n_rx_thread = thread + 1;
2614
2615         }
2616         return 0;
2617 }
2618
2619 static int
2620 init_tx_threads(void)
2621 {
2622         int i;
2623
2624         n_tx_thread = 0;
2625         for (i = 0; i < nb_tx_thread_params; ++i) {
2626                 tx_thread[n_tx_thread].conf.thread_id = tx_thread_params[i].thread_id;
2627                 tx_thread[n_tx_thread].conf.lcore_id = tx_thread_params[i].lcore_id;
2628                 n_tx_thread++;
2629         }
2630         return 0;
2631 }
2632
2633 /* display usage */
2634 static void
2635 print_usage(const char *prgname)
2636 {
2637         printf("%s [EAL options] -- -p PORTMASK -P"
2638                 "  [--rx (port,queue,lcore,thread)[,(port,queue,lcore,thread]]"
2639                 "  [--tx (lcore,thread)[,(lcore,thread]]"
2640                 "  [--max-pkt-len PKTLEN]"
2641                 "  [--parse-ptype]\n\n"
2642                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
2643                 "  -P : enable promiscuous mode\n"
2644                 "  --rx (port,queue,lcore,thread): rx queues configuration\n"
2645                 "  --tx (lcore,thread): tx threads configuration\n"
2646                 "  --stat-lcore LCORE: use lcore for stat collector\n"
2647                 "  --eth-dest=X,MM:MM:MM:MM:MM:MM: optional, ethernet destination for port X\n"
2648                 "  --no-numa: optional, disable numa awareness\n"
2649                 "  --ipv6: optional, specify it if running ipv6 packets\n"
2650                 "  --max-pkt-len PKTLEN: maximum packet length in decimal (64-9600)\n"
2651                 "  --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n"
2652                 "  --no-lthreads: turn off lthread model\n"
2653                 "  --parse-ptype: set to use software to analyze packet type\n\n",
2654                 prgname);
2655 }
2656
2657 static int parse_max_pkt_len(const char *pktlen)
2658 {
2659         char *end = NULL;
2660         unsigned long len;
2661
2662         /* parse decimal string */
2663         len = strtoul(pktlen, &end, 10);
2664         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
2665                 return -1;
2666
2667         if (len == 0)
2668                 return -1;
2669
2670         return len;
2671 }
2672
2673 static int
2674 parse_portmask(const char *portmask)
2675 {
2676         char *end = NULL;
2677         unsigned long pm;
2678
2679         /* parse hexadecimal string */
2680         pm = strtoul(portmask, &end, 16);
2681         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
2682                 return 0;
2683
2684         return pm;
2685 }
2686
2687 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2688 static int
2689 parse_hash_entry_number(const char *hash_entry_num)
2690 {
2691         char *end = NULL;
2692         unsigned long hash_en;
2693
2694         /* parse hexadecimal string */
2695         hash_en = strtoul(hash_entry_num, &end, 16);
2696         if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0'))
2697                 return -1;
2698
2699         if (hash_en == 0)
2700                 return -1;
2701
2702         return hash_en;
2703 }
2704 #endif
2705
2706 static int
2707 parse_rx_config(const char *q_arg)
2708 {
2709         char s[256];
2710         const char *p, *p0 = q_arg;
2711         char *end;
2712         enum fieldnames {
2713                 FLD_PORT = 0,
2714                 FLD_QUEUE,
2715                 FLD_LCORE,
2716                 FLD_THREAD,
2717                 _NUM_FLD
2718         };
2719         unsigned long int_fld[_NUM_FLD];
2720         char *str_fld[_NUM_FLD];
2721         int i;
2722         unsigned size;
2723
2724         nb_rx_thread_params = 0;
2725
2726         while ((p = strchr(p0, '(')) != NULL) {
2727                 ++p;
2728                 p0 = strchr(p, ')');
2729                 if (p0 == NULL)
2730                         return -1;
2731
2732                 size = p0 - p;
2733                 if (size >= sizeof(s))
2734                         return -1;
2735
2736                 snprintf(s, sizeof(s), "%.*s", size, p);
2737                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2738                         return -1;
2739                 for (i = 0; i < _NUM_FLD; i++) {
2740                         errno = 0;
2741                         int_fld[i] = strtoul(str_fld[i], &end, 0);
2742                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2743                                 return -1;
2744                 }
2745                 if (nb_rx_thread_params >= MAX_LCORE_PARAMS) {
2746                         printf("exceeded max number of rx params: %hu\n",
2747                                         nb_rx_thread_params);
2748                         return -1;
2749                 }
2750                 rx_thread_params_array[nb_rx_thread_params].port_id =
2751                                 int_fld[FLD_PORT];
2752                 rx_thread_params_array[nb_rx_thread_params].queue_id =
2753                                 (uint8_t)int_fld[FLD_QUEUE];
2754                 rx_thread_params_array[nb_rx_thread_params].lcore_id =
2755                                 (uint8_t)int_fld[FLD_LCORE];
2756                 rx_thread_params_array[nb_rx_thread_params].thread_id =
2757                                 (uint8_t)int_fld[FLD_THREAD];
2758                 ++nb_rx_thread_params;
2759         }
2760         rx_thread_params = rx_thread_params_array;
2761         return 0;
2762 }
2763
2764 static int
2765 parse_tx_config(const char *q_arg)
2766 {
2767         char s[256];
2768         const char *p, *p0 = q_arg;
2769         char *end;
2770         enum fieldnames {
2771                 FLD_LCORE = 0,
2772                 FLD_THREAD,
2773                 _NUM_FLD
2774         };
2775         unsigned long int_fld[_NUM_FLD];
2776         char *str_fld[_NUM_FLD];
2777         int i;
2778         unsigned size;
2779
2780         nb_tx_thread_params = 0;
2781
2782         while ((p = strchr(p0, '(')) != NULL) {
2783                 ++p;
2784                 p0 = strchr(p, ')');
2785                 if (p0 == NULL)
2786                         return -1;
2787
2788                 size = p0 - p;
2789                 if (size >= sizeof(s))
2790                         return -1;
2791
2792                 snprintf(s, sizeof(s), "%.*s", size, p);
2793                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
2794                         return -1;
2795                 for (i = 0; i < _NUM_FLD; i++) {
2796                         errno = 0;
2797                         int_fld[i] = strtoul(str_fld[i], &end, 0);
2798                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
2799                                 return -1;
2800                 }
2801                 if (nb_tx_thread_params >= MAX_LCORE_PARAMS) {
2802                         printf("exceeded max number of tx params: %hu\n",
2803                                 nb_tx_thread_params);
2804                         return -1;
2805                 }
2806                 tx_thread_params_array[nb_tx_thread_params].lcore_id =
2807                                 (uint8_t)int_fld[FLD_LCORE];
2808                 tx_thread_params_array[nb_tx_thread_params].thread_id =
2809                                 (uint8_t)int_fld[FLD_THREAD];
2810                 ++nb_tx_thread_params;
2811         }
2812         tx_thread_params = tx_thread_params_array;
2813
2814         return 0;
2815 }
2816
2817 #if (APP_CPU_LOAD > 0)
2818 static int
2819 parse_stat_lcore(const char *stat_lcore)
2820 {
2821         char *end = NULL;
2822         unsigned long lcore_id;
2823
2824         lcore_id = strtoul(stat_lcore, &end, 10);
2825         if ((stat_lcore[0] == '\0') || (end == NULL) || (*end != '\0'))
2826                 return -1;
2827
2828         return lcore_id;
2829 }
2830 #endif
2831
2832 static void
2833 parse_eth_dest(const char *optarg)
2834 {
2835         uint16_t portid;
2836         char *port_end;
2837         uint8_t c, *dest, peer_addr[6];
2838
2839         errno = 0;
2840         portid = strtoul(optarg, &port_end, 10);
2841         if (errno != 0 || port_end == optarg || *port_end++ != ',')
2842                 rte_exit(EXIT_FAILURE,
2843                 "Invalid eth-dest: %s", optarg);
2844         if (portid >= RTE_MAX_ETHPORTS)
2845                 rte_exit(EXIT_FAILURE,
2846                 "eth-dest: port %d >= RTE_MAX_ETHPORTS(%d)\n",
2847                 portid, RTE_MAX_ETHPORTS);
2848
2849         if (cmdline_parse_etheraddr(NULL, port_end,
2850                 &peer_addr, sizeof(peer_addr)) < 0)
2851                 rte_exit(EXIT_FAILURE,
2852                 "Invalid ethernet address: %s\n",
2853                 port_end);
2854         dest = (uint8_t *)&dest_eth_addr[portid];
2855         for (c = 0; c < 6; c++)
2856                 dest[c] = peer_addr[c];
2857         *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
2858 }
2859
2860 enum {
2861 #define OPT_RX_CONFIG       "rx"
2862         OPT_RX_CONFIG_NUM = 256,
2863 #define OPT_TX_CONFIG       "tx"
2864         OPT_TX_CONFIG_NUM,
2865 #define OPT_STAT_LCORE      "stat-lcore"
2866         OPT_STAT_LCORE_NUM,
2867 #define OPT_ETH_DEST        "eth-dest"
2868         OPT_ETH_DEST_NUM,
2869 #define OPT_NO_NUMA         "no-numa"
2870         OPT_NO_NUMA_NUM,
2871 #define OPT_IPV6            "ipv6"
2872         OPT_IPV6_NUM,
2873 #define OPT_MAX_PKT_LEN "max-pkt-len"
2874         OPT_MAX_PKT_LEN_NUM,
2875 #define OPT_HASH_ENTRY_NUM  "hash-entry-num"
2876         OPT_HASH_ENTRY_NUM_NUM,
2877 #define OPT_NO_LTHREADS     "no-lthreads"
2878         OPT_NO_LTHREADS_NUM,
2879 #define OPT_PARSE_PTYPE     "parse-ptype"
2880         OPT_PARSE_PTYPE_NUM,
2881 };
2882
2883 /* Parse the argument given in the command line of the application */
2884 static int
2885 parse_args(int argc, char **argv)
2886 {
2887         int opt, ret;
2888         char **argvopt;
2889         int option_index;
2890         char *prgname = argv[0];
2891         static struct option lgopts[] = {
2892                 {OPT_RX_CONFIG,      1, NULL, OPT_RX_CONFIG_NUM      },
2893                 {OPT_TX_CONFIG,      1, NULL, OPT_TX_CONFIG_NUM      },
2894                 {OPT_STAT_LCORE,     1, NULL, OPT_STAT_LCORE_NUM     },
2895                 {OPT_ETH_DEST,       1, NULL, OPT_ETH_DEST_NUM       },
2896                 {OPT_NO_NUMA,        0, NULL, OPT_NO_NUMA_NUM        },
2897                 {OPT_IPV6,           0, NULL, OPT_IPV6_NUM           },
2898                 {OPT_MAX_PKT_LEN,    1, NULL, OPT_MAX_PKT_LEN_NUM    },
2899                 {OPT_HASH_ENTRY_NUM, 1, NULL, OPT_HASH_ENTRY_NUM_NUM },
2900                 {OPT_NO_LTHREADS,    0, NULL, OPT_NO_LTHREADS_NUM    },
2901                 {OPT_PARSE_PTYPE,    0, NULL, OPT_PARSE_PTYPE_NUM    },
2902                 {NULL,               0, 0,    0                      }
2903         };
2904
2905         argvopt = argv;
2906
2907         while ((opt = getopt_long(argc, argvopt, "p:P",
2908                                 lgopts, &option_index)) != EOF) {
2909
2910                 switch (opt) {
2911                 /* portmask */
2912                 case 'p':
2913                         enabled_port_mask = parse_portmask(optarg);
2914                         if (enabled_port_mask == 0) {
2915                                 printf("invalid portmask\n");
2916                                 print_usage(prgname);
2917                                 return -1;
2918                         }
2919                         break;
2920
2921                 case 'P':
2922                         printf("Promiscuous mode selected\n");
2923                         promiscuous_on = 1;
2924                         break;
2925
2926                 /* long options */
2927                 case OPT_RX_CONFIG_NUM:
2928                         ret = parse_rx_config(optarg);
2929                         if (ret) {
2930                                 printf("invalid rx-config\n");
2931                                 print_usage(prgname);
2932                                 return -1;
2933                         }
2934                         break;
2935
2936                 case OPT_TX_CONFIG_NUM:
2937                         ret = parse_tx_config(optarg);
2938                         if (ret) {
2939                                 printf("invalid tx-config\n");
2940                                 print_usage(prgname);
2941                                 return -1;
2942                         }
2943                         break;
2944
2945 #if (APP_CPU_LOAD > 0)
2946                 case OPT_STAT_LCORE_NUM:
2947                         cpu_load_lcore_id = parse_stat_lcore(optarg);
2948                         break;
2949 #endif
2950
2951                 case OPT_ETH_DEST_NUM:
2952                         parse_eth_dest(optarg);
2953                         break;
2954
2955                 case OPT_NO_NUMA_NUM:
2956                         printf("numa is disabled\n");
2957                         numa_on = 0;
2958                         break;
2959
2960 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2961                 case OPT_IPV6_NUM:
2962                         printf("ipv6 is specified\n");
2963                         ipv6 = 1;
2964                         break;
2965 #endif
2966
2967                 case OPT_NO_LTHREADS_NUM:
2968                         printf("l-threads model is disabled\n");
2969                         lthreads_on = 0;
2970                         break;
2971
2972                 case OPT_PARSE_PTYPE_NUM:
2973                         printf("software packet type parsing enabled\n");
2974                         parse_ptype_on = 1;
2975                         break;
2976
2977                 case OPT_MAX_PKT_LEN_NUM:
2978                         max_pkt_len = parse_max_pkt_len(optarg);
2979                         break;
2980
2981 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2982                 case OPT_HASH_ENTRY_NUM_NUM:
2983                         ret = parse_hash_entry_number(optarg);
2984                         if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) {
2985                                 hash_entry_number = ret;
2986                         } else {
2987                                 printf("invalid hash entry number\n");
2988                                 print_usage(prgname);
2989                                 return -1;
2990                         }
2991                         break;
2992 #endif
2993
2994                 default:
2995                         print_usage(prgname);
2996                         return -1;
2997                 }
2998         }
2999
3000         if (optind >= 0)
3001                 argv[optind-1] = prgname;
3002
3003         ret = optind-1;
3004         optind = 1; /* reset getopt lib */
3005         return ret;
3006 }
3007
3008 static void
3009 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
3010 {
3011         char buf[RTE_ETHER_ADDR_FMT_SIZE];
3012
3013         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
3014         printf("%s%s", name, buf);
3015 }
3016
3017 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
3018
3019 static void convert_ipv4_5tuple(struct ipv4_5tuple *key1,
3020                 union ipv4_5tuple_host *key2)
3021 {
3022         key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
3023         key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
3024         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3025         key2->port_src = rte_cpu_to_be_16(key1->port_src);
3026         key2->proto = key1->proto;
3027         key2->pad0 = 0;
3028         key2->pad1 = 0;
3029 }
3030
3031 static void convert_ipv6_5tuple(struct ipv6_5tuple *key1,
3032                 union ipv6_5tuple_host *key2)
3033 {
3034         uint32_t i;
3035
3036         for (i = 0; i < 16; i++) {
3037                 key2->ip_dst[i] = key1->ip_dst[i];
3038                 key2->ip_src[i] = key1->ip_src[i];
3039         }
3040         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
3041         key2->port_src = rte_cpu_to_be_16(key1->port_src);
3042         key2->proto = key1->proto;
3043         key2->pad0 = 0;
3044         key2->pad1 = 0;
3045         key2->reserve = 0;
3046 }
3047
3048 #define BYTE_VALUE_MAX 256
3049 #define ALL_32_BITS 0xffffffff
3050 #define BIT_8_TO_15 0x0000ff00
3051 static inline void
3052 populate_ipv4_few_flow_into_table(const struct rte_hash *h)
3053 {
3054         uint32_t i;
3055         int32_t ret;
3056         uint32_t array_len = RTE_DIM(ipv4_l3fwd_route_array);
3057
3058         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3059         for (i = 0; i < array_len; i++) {
3060                 struct ipv4_l3fwd_route  entry;
3061                 union ipv4_5tuple_host newkey;
3062
3063                 entry = ipv4_l3fwd_route_array[i];
3064                 convert_ipv4_5tuple(&entry.key, &newkey);
3065                 ret = rte_hash_add_key(h, (void *)&newkey);
3066                 if (ret < 0) {
3067                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3068                                 " to the l3fwd hash.\n", i);
3069                 }
3070                 ipv4_l3fwd_out_if[ret] = entry.if_out;
3071         }
3072         printf("Hash: Adding 0x%" PRIx32 " keys\n", array_len);
3073 }
3074
3075 #define BIT_16_TO_23 0x00ff0000
3076 static inline void
3077 populate_ipv6_few_flow_into_table(const struct rte_hash *h)
3078 {
3079         uint32_t i;
3080         int32_t ret;
3081         uint32_t array_len = RTE_DIM(ipv6_l3fwd_route_array);
3082
3083         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3084         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3085         for (i = 0; i < array_len; i++) {
3086                 struct ipv6_l3fwd_route entry;
3087                 union ipv6_5tuple_host newkey;
3088
3089                 entry = ipv6_l3fwd_route_array[i];
3090                 convert_ipv6_5tuple(&entry.key, &newkey);
3091                 ret = rte_hash_add_key(h, (void *)&newkey);
3092                 if (ret < 0) {
3093                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
3094                                 " to the l3fwd hash.\n", i);
3095                 }
3096                 ipv6_l3fwd_out_if[ret] = entry.if_out;
3097         }
3098         printf("Hash: Adding 0x%" PRIx32 "keys\n", array_len);
3099 }
3100
3101 #define NUMBER_PORT_USED 4
3102 static inline void
3103 populate_ipv4_many_flow_into_table(const struct rte_hash *h,
3104                 unsigned int nr_flow)
3105 {
3106         unsigned i;
3107
3108         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15);
3109
3110         for (i = 0; i < nr_flow; i++) {
3111                 struct ipv4_l3fwd_route entry;
3112                 union ipv4_5tuple_host newkey;
3113                 uint8_t a = (uint8_t)((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3114                 uint8_t b = (uint8_t)(((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3115                                 BYTE_VALUE_MAX);
3116                 uint8_t c = (uint8_t)((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3117                                 BYTE_VALUE_MAX));
3118                 /* Create the ipv4 exact match flow */
3119                 memset(&entry, 0, sizeof(entry));
3120                 switch (i & (NUMBER_PORT_USED - 1)) {
3121                 case 0:
3122                         entry = ipv4_l3fwd_route_array[0];
3123                         entry.key.ip_dst = RTE_IPV4(101, c, b, a);
3124                         break;
3125                 case 1:
3126                         entry = ipv4_l3fwd_route_array[1];
3127                         entry.key.ip_dst = RTE_IPV4(201, c, b, a);
3128                         break;
3129                 case 2:
3130                         entry = ipv4_l3fwd_route_array[2];
3131                         entry.key.ip_dst = RTE_IPV4(111, c, b, a);
3132                         break;
3133                 case 3:
3134                         entry = ipv4_l3fwd_route_array[3];
3135                         entry.key.ip_dst = RTE_IPV4(211, c, b, a);
3136                         break;
3137                 };
3138                 convert_ipv4_5tuple(&entry.key, &newkey);
3139                 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3140
3141                 if (ret < 0)
3142                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3143
3144                 ipv4_l3fwd_out_if[ret] = (uint8_t)entry.if_out;
3145
3146         }
3147         printf("Hash: Adding 0x%x keys\n", nr_flow);
3148 }
3149
3150 static inline void
3151 populate_ipv6_many_flow_into_table(const struct rte_hash *h,
3152                 unsigned int nr_flow)
3153 {
3154         unsigned i;
3155
3156         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23);
3157         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
3158         for (i = 0; i < nr_flow; i++) {
3159                 struct ipv6_l3fwd_route entry;
3160                 union ipv6_5tuple_host newkey;
3161
3162                 uint8_t a = (uint8_t) ((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX);
3163                 uint8_t b = (uint8_t) (((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) %
3164                                 BYTE_VALUE_MAX);
3165                 uint8_t c = (uint8_t) ((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX *
3166                                 BYTE_VALUE_MAX));
3167
3168                 /* Create the ipv6 exact match flow */
3169                 memset(&entry, 0, sizeof(entry));
3170                 switch (i & (NUMBER_PORT_USED - 1)) {
3171                 case 0:
3172                         entry = ipv6_l3fwd_route_array[0];
3173                         break;
3174                 case 1:
3175                         entry = ipv6_l3fwd_route_array[1];
3176                         break;
3177                 case 2:
3178                         entry = ipv6_l3fwd_route_array[2];
3179                         break;
3180                 case 3:
3181                         entry = ipv6_l3fwd_route_array[3];
3182                         break;
3183                 };
3184                 entry.key.ip_dst[13] = c;
3185                 entry.key.ip_dst[14] = b;
3186                 entry.key.ip_dst[15] = a;
3187                 convert_ipv6_5tuple(&entry.key, &newkey);
3188                 int32_t ret = rte_hash_add_key(h, (void *)&newkey);
3189
3190                 if (ret < 0)
3191                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
3192
3193                 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
3194
3195         }
3196         printf("Hash: Adding 0x%x keys\n", nr_flow);
3197 }
3198
3199 static void
3200 setup_hash(int socketid)
3201 {
3202         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
3203                 .name = NULL,
3204                 .entries = L3FWD_HASH_ENTRIES,
3205                 .key_len = sizeof(union ipv4_5tuple_host),
3206                 .hash_func = ipv4_hash_crc,
3207                 .hash_func_init_val = 0,
3208         };
3209
3210         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
3211                 .name = NULL,
3212                 .entries = L3FWD_HASH_ENTRIES,
3213                 .key_len = sizeof(union ipv6_5tuple_host),
3214                 .hash_func = ipv6_hash_crc,
3215                 .hash_func_init_val = 0,
3216         };
3217
3218         char s[64];
3219
3220         /* create ipv4 hash */
3221         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
3222         ipv4_l3fwd_hash_params.name = s;
3223         ipv4_l3fwd_hash_params.socket_id = socketid;
3224         ipv4_l3fwd_lookup_struct[socketid] =
3225                         rte_hash_create(&ipv4_l3fwd_hash_params);
3226         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3227                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3228                                 "socket %d\n", socketid);
3229
3230         /* create ipv6 hash */
3231         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
3232         ipv6_l3fwd_hash_params.name = s;
3233         ipv6_l3fwd_hash_params.socket_id = socketid;
3234         ipv6_l3fwd_lookup_struct[socketid] =
3235                         rte_hash_create(&ipv6_l3fwd_hash_params);
3236         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3237                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
3238                                 "socket %d\n", socketid);
3239
3240         if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
3241                 /* For testing hash matching with a large number of flows we
3242                  * generate millions of IP 5-tuples with an incremented dst
3243                  * address to initialize the hash table. */
3244                 if (ipv6 == 0) {
3245                         /* populate the ipv4 hash */
3246                         populate_ipv4_many_flow_into_table(
3247                                 ipv4_l3fwd_lookup_struct[socketid], hash_entry_number);
3248                 } else {
3249                         /* populate the ipv6 hash */
3250                         populate_ipv6_many_flow_into_table(
3251                                 ipv6_l3fwd_lookup_struct[socketid], hash_entry_number);
3252                 }
3253         } else {
3254                 /* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize
3255                  * the hash table */
3256                 if (ipv6 == 0) {
3257                         /* populate the ipv4 hash */
3258                         populate_ipv4_few_flow_into_table(
3259                                         ipv4_l3fwd_lookup_struct[socketid]);
3260                 } else {
3261                         /* populate the ipv6 hash */
3262                         populate_ipv6_few_flow_into_table(
3263                                         ipv6_l3fwd_lookup_struct[socketid]);
3264                 }
3265         }
3266 }
3267 #endif
3268
3269 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3270 static void
3271 setup_lpm(int socketid)
3272 {
3273         struct rte_lpm6_config config;
3274         struct rte_lpm_config lpm_ipv4_config;
3275         unsigned i;
3276         int ret;
3277         char s[64];
3278
3279         /* create the LPM table */
3280         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
3281         lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
3282         lpm_ipv4_config.number_tbl8s = 256;
3283         lpm_ipv4_config.flags = 0;
3284         ipv4_l3fwd_lookup_struct[socketid] =
3285                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
3286         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
3287                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3288                                 " on socket %d\n", socketid);
3289
3290         /* populate the LPM table */
3291         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
3292
3293                 /* skip unused ports */
3294                 if ((1 << ipv4_l3fwd_route_array[i].if_out &
3295                                 enabled_port_mask) == 0)
3296                         continue;
3297
3298                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
3299                         ipv4_l3fwd_route_array[i].ip,
3300                         ipv4_l3fwd_route_array[i].depth,
3301                         ipv4_l3fwd_route_array[i].if_out);
3302
3303                 if (ret < 0) {
3304                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3305                                 "l3fwd LPM table on socket %d\n",
3306                                 i, socketid);
3307                 }
3308
3309                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
3310                         (unsigned)ipv4_l3fwd_route_array[i].ip,
3311                         ipv4_l3fwd_route_array[i].depth,
3312                         ipv4_l3fwd_route_array[i].if_out);
3313         }
3314
3315         /* create the LPM6 table */
3316         snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
3317
3318         config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
3319         config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
3320         config.flags = 0;
3321         ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
3322                                 &config);
3323         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
3324                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
3325                                 " on socket %d\n", socketid);
3326
3327         /* populate the LPM table */
3328         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
3329
3330                 /* skip unused ports */
3331                 if ((1 << ipv6_l3fwd_route_array[i].if_out &
3332                                 enabled_port_mask) == 0)
3333                         continue;
3334
3335                 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
3336                         ipv6_l3fwd_route_array[i].ip,
3337                         ipv6_l3fwd_route_array[i].depth,
3338                         ipv6_l3fwd_route_array[i].if_out);
3339
3340                 if (ret < 0) {
3341                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
3342                                 "l3fwd LPM table on socket %d\n",
3343                                 i, socketid);
3344                 }
3345
3346                 printf("LPM: Adding route %s / %d (%d)\n",
3347                         "IPV6",
3348                         ipv6_l3fwd_route_array[i].depth,
3349                         ipv6_l3fwd_route_array[i].if_out);
3350         }
3351 }
3352 #endif
3353
3354 static int
3355 init_mem(unsigned nb_mbuf)
3356 {
3357         struct lcore_conf *qconf;
3358         int socketid;
3359         unsigned lcore_id;
3360         char s[64];
3361
3362         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3363                 if (rte_lcore_is_enabled(lcore_id) == 0)
3364                         continue;
3365
3366                 if (numa_on)
3367                         socketid = rte_lcore_to_socket_id(lcore_id);
3368                 else
3369                         socketid = 0;
3370
3371                 if (socketid >= NB_SOCKETS) {
3372                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
3373                                 socketid, lcore_id, NB_SOCKETS);
3374                 }
3375                 if (pktmbuf_pool[socketid] == NULL) {
3376                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
3377                         pktmbuf_pool[socketid] =
3378                                 rte_pktmbuf_pool_create(s, nb_mbuf,
3379                                         MEMPOOL_CACHE_SIZE, 0,
3380                                         RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
3381                         if (pktmbuf_pool[socketid] == NULL)
3382                                 rte_exit(EXIT_FAILURE,
3383                                                 "Cannot init mbuf pool on socket %d\n", socketid);
3384                         else
3385                                 printf("Allocated mbuf pool on socket %d\n", socketid);
3386
3387 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
3388                         setup_lpm(socketid);
3389 #else
3390                         setup_hash(socketid);
3391 #endif
3392                 }
3393                 qconf = &lcore_conf[lcore_id];
3394                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
3395                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
3396         }
3397         return 0;
3398 }
3399
3400 /* Check the link status of all ports in up to 9s, and print them finally */
3401 static void
3402 check_all_ports_link_status(uint32_t port_mask)
3403 {
3404 #define CHECK_INTERVAL 100 /* 100ms */
3405 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
3406         uint16_t portid;
3407         uint8_t count, all_ports_up, print_flag = 0;
3408         struct rte_eth_link link;
3409         int ret;
3410         char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
3411
3412         printf("\nChecking link status");
3413         fflush(stdout);
3414         for (count = 0; count <= MAX_CHECK_TIME; count++) {
3415                 all_ports_up = 1;
3416                 RTE_ETH_FOREACH_DEV(portid) {
3417                         if ((port_mask & (1 << portid)) == 0)
3418                                 continue;
3419                         memset(&link, 0, sizeof(link));
3420                         ret = rte_eth_link_get_nowait(portid, &link);
3421                         if (ret < 0) {
3422                                 all_ports_up = 0;
3423                                 if (print_flag == 1)
3424                                         printf("Port %u link get failed: %s\n",
3425                                                 portid, rte_strerror(-ret));
3426                                 continue;
3427                         }
3428                         /* print link status if flag set */
3429                         if (print_flag == 1) {
3430                                 rte_eth_link_to_str(link_status_text,
3431                                         sizeof(link_status_text), &link);
3432                                 printf("Port %d %s\n", portid,
3433                                         link_status_text);
3434                                 continue;
3435                         }
3436                         /* clear all_ports_up flag if any link down */
3437                         if (link.link_status == RTE_ETH_LINK_DOWN) {
3438                                 all_ports_up = 0;
3439                                 break;
3440                         }
3441                 }
3442                 /* after finally printing all link status, get out */
3443                 if (print_flag == 1)
3444                         break;
3445
3446                 if (all_ports_up == 0) {
3447                         printf(".");
3448                         fflush(stdout);
3449                         rte_delay_ms(CHECK_INTERVAL);
3450                 }
3451
3452                 /* set the print_flag if all ports up or timeout */
3453                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
3454                         print_flag = 1;
3455                         printf("done\n");
3456                 }
3457         }
3458 }
3459
3460 static uint32_t
3461 eth_dev_get_overhead_len(uint32_t max_rx_pktlen, uint16_t max_mtu)
3462 {
3463         uint32_t overhead_len;
3464
3465         if (max_mtu != UINT16_MAX && max_rx_pktlen > max_mtu)
3466                 overhead_len = max_rx_pktlen - max_mtu;
3467         else
3468                 overhead_len = RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN;
3469
3470         return overhead_len;
3471 }
3472
3473 static int
3474 config_port_max_pkt_len(struct rte_eth_conf *conf,
3475                 struct rte_eth_dev_info *dev_info)
3476 {
3477         uint32_t overhead_len;
3478
3479         if (max_pkt_len == 0)
3480                 return 0;
3481
3482         if (max_pkt_len < RTE_ETHER_MIN_LEN || max_pkt_len > MAX_JUMBO_PKT_LEN)
3483                 return -1;
3484
3485         overhead_len = eth_dev_get_overhead_len(dev_info->max_rx_pktlen,
3486                         dev_info->max_mtu);
3487         conf->rxmode.mtu = max_pkt_len - overhead_len;
3488
3489         if (conf->rxmode.mtu > RTE_ETHER_MTU)
3490                 conf->txmode.offloads |= RTE_ETH_TX_OFFLOAD_MULTI_SEGS;
3491
3492         return 0;
3493 }
3494
3495 int
3496 main(int argc, char **argv)
3497 {
3498         struct rte_eth_dev_info dev_info;
3499         struct rte_eth_txconf *txconf;
3500         int ret;
3501         int i;
3502         unsigned nb_ports;
3503         uint16_t queueid, portid;
3504         unsigned lcore_id;
3505         uint32_t n_tx_queue, nb_lcores;
3506         uint8_t nb_rx_queue, queue, socketid;
3507
3508         /* init EAL */
3509         ret = rte_eal_init(argc, argv);
3510         if (ret < 0)
3511                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
3512         argc -= ret;
3513         argv += ret;
3514
3515         ret = rte_timer_subsystem_init();
3516         if (ret < 0)
3517                 rte_exit(EXIT_FAILURE, "Failed to initialize timer subystem\n");
3518
3519         /* pre-init dst MACs for all ports to 02:00:00:00:00:xx */
3520         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
3521                 dest_eth_addr[portid] = RTE_ETHER_LOCAL_ADMIN_ADDR +
3522                                 ((uint64_t)portid << 40);
3523                 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid];
3524         }
3525
3526         /* parse application arguments (after the EAL ones) */
3527         ret = parse_args(argc, argv);
3528         if (ret < 0)
3529                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
3530
3531         if (check_lcore_params() < 0)
3532                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
3533
3534         printf("Initializing rx-queues...\n");
3535         ret = init_rx_queues();
3536         if (ret < 0)
3537                 rte_exit(EXIT_FAILURE, "init_rx_queues failed\n");
3538
3539         printf("Initializing tx-threads...\n");
3540         ret = init_tx_threads();
3541         if (ret < 0)
3542                 rte_exit(EXIT_FAILURE, "init_tx_threads failed\n");
3543
3544         printf("Initializing rings...\n");
3545         ret = init_rx_rings();
3546         if (ret < 0)
3547                 rte_exit(EXIT_FAILURE, "init_rx_rings failed\n");
3548
3549         nb_ports = rte_eth_dev_count_avail();
3550
3551         if (check_port_config() < 0)
3552                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
3553
3554         nb_lcores = rte_lcore_count();
3555
3556         /* initialize all ports */
3557         RTE_ETH_FOREACH_DEV(portid) {
3558                 struct rte_eth_conf local_port_conf = port_conf;
3559
3560                 /* skip ports that are not enabled */
3561                 if ((enabled_port_mask & (1 << portid)) == 0) {
3562                         printf("\nSkipping disabled port %d\n", portid);
3563                         continue;
3564                 }
3565
3566                 /* init port */
3567                 printf("Initializing port %d ... ", portid);
3568                 fflush(stdout);
3569
3570                 nb_rx_queue = get_port_n_rx_queues(portid);
3571                 n_tx_queue = nb_lcores;
3572                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
3573                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
3574                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
3575                         nb_rx_queue, (unsigned)n_tx_queue);
3576
3577                 ret = rte_eth_dev_info_get(portid, &dev_info);
3578                 if (ret != 0)
3579                         rte_exit(EXIT_FAILURE,
3580                                 "Error during getting device (port %u) info: %s\n",
3581                                 portid, strerror(-ret));
3582
3583                 ret = config_port_max_pkt_len(&local_port_conf, &dev_info);
3584                 if (ret != 0)
3585                         rte_exit(EXIT_FAILURE,
3586                                 "Invalid max packet length: %u (port %u)\n",
3587                                 max_pkt_len, portid);
3588
3589                 if (dev_info.tx_offload_capa & RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE)
3590                         local_port_conf.txmode.offloads |=
3591                                 RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE;
3592
3593                 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
3594                         dev_info.flow_type_rss_offloads;
3595                 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
3596                                 port_conf.rx_adv_conf.rss_conf.rss_hf) {
3597                         printf("Port %u modified RSS hash function based on hardware support,"
3598                                 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
3599                                 portid,
3600                                 port_conf.rx_adv_conf.rss_conf.rss_hf,
3601                                 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
3602                 }
3603
3604                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
3605                                         (uint16_t)n_tx_queue, &local_port_conf);
3606                 if (ret < 0)
3607                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
3608                                 ret, portid);
3609
3610                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
3611                                                        &nb_txd);
3612                 if (ret < 0)
3613                         rte_exit(EXIT_FAILURE,
3614                                  "rte_eth_dev_adjust_nb_rx_tx_desc: err=%d, port=%d\n",
3615                                  ret, portid);
3616
3617                 ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
3618                 if (ret < 0)
3619                         rte_exit(EXIT_FAILURE,
3620                                  "rte_eth_macaddr_get: err=%d, port=%d\n",
3621                                  ret, portid);
3622
3623                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
3624                 printf(", ");
3625                 print_ethaddr("Destination:",
3626                         (const struct rte_ether_addr *)&dest_eth_addr[portid]);
3627                 printf(", ");
3628
3629                 /*
3630                  * prepare src MACs for each port.
3631                  */
3632                 rte_ether_addr_copy(&ports_eth_addr[portid],
3633                         (struct rte_ether_addr *)(val_eth + portid) + 1);
3634
3635                 /* init memory */
3636                 ret = init_mem(NB_MBUF);
3637                 if (ret < 0)
3638                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
3639
3640                 /* init one TX queue per couple (lcore,port) */
3641                 queueid = 0;
3642                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3643                         if (rte_lcore_is_enabled(lcore_id) == 0)
3644                                 continue;
3645
3646                         if (numa_on)
3647                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3648                         else
3649                                 socketid = 0;
3650
3651                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
3652                         fflush(stdout);
3653
3654                         txconf = &dev_info.default_txconf;
3655                         txconf->offloads = local_port_conf.txmode.offloads;
3656                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
3657                                                      socketid, txconf);
3658                         if (ret < 0)
3659                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
3660                                         "port=%d\n", ret, portid);
3661
3662                         tx_thread[lcore_id].tx_queue_id[portid] = queueid;
3663                         queueid++;
3664                 }
3665                 printf("\n");
3666         }
3667
3668         for (i = 0; i < n_rx_thread; i++) {
3669                 lcore_id = rx_thread[i].conf.lcore_id;
3670
3671                 if (rte_lcore_is_enabled(lcore_id) == 0) {
3672                         rte_exit(EXIT_FAILURE,
3673                                         "Cannot start Rx thread on lcore %u: lcore disabled\n",
3674                                         lcore_id
3675                                 );
3676                 }
3677
3678                 printf("\nInitializing rx queues for Rx thread %d on lcore %u ... ",
3679                                 i, lcore_id);
3680                 fflush(stdout);
3681
3682                 /* init RX queues */
3683                 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3684                         struct rte_eth_rxconf rxq_conf;
3685
3686                         portid = rx_thread[i].rx_queue_list[queue].port_id;
3687                         queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3688
3689                         if (numa_on)
3690                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3691                         else
3692                                 socketid = 0;
3693
3694                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
3695                         fflush(stdout);
3696
3697                         ret = rte_eth_dev_info_get(portid, &dev_info);
3698                         if (ret != 0)
3699                                 rte_exit(EXIT_FAILURE,
3700                                         "Error during getting device (port %u) info: %s\n",
3701                                         portid, strerror(-ret));
3702
3703                         rxq_conf = dev_info.default_rxconf;
3704                         rxq_conf.offloads = port_conf.rxmode.offloads;
3705                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
3706                                         socketid,
3707                                         &rxq_conf,
3708                                         pktmbuf_pool[socketid]);
3709                         if (ret < 0)
3710                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, "
3711                                                 "port=%d\n", ret, portid);
3712                 }
3713         }
3714
3715         printf("\n");
3716
3717         /* start ports */
3718         RTE_ETH_FOREACH_DEV(portid) {
3719                 if ((enabled_port_mask & (1 << portid)) == 0)
3720                         continue;
3721
3722                 /* Start device */
3723                 ret = rte_eth_dev_start(portid);
3724                 if (ret < 0)
3725                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
3726                                 ret, portid);
3727
3728                 /*
3729                  * If enabled, put device in promiscuous mode.
3730                  * This allows IO forwarding mode to forward packets
3731                  * to itself through 2 cross-connected  ports of the
3732                  * target machine.
3733                  */
3734                 if (promiscuous_on) {
3735                         ret = rte_eth_promiscuous_enable(portid);
3736                         if (ret != 0)
3737                                 rte_exit(EXIT_FAILURE,
3738                                         "rte_eth_promiscuous_enable: err=%s, port=%u\n",
3739                                         rte_strerror(-ret), portid);
3740                 }
3741         }
3742
3743         for (i = 0; i < n_rx_thread; i++) {
3744                 lcore_id = rx_thread[i].conf.lcore_id;
3745                 if (rte_lcore_is_enabled(lcore_id) == 0)
3746                         continue;
3747
3748                 /* check if hw packet type is supported */
3749                 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) {
3750                         portid = rx_thread[i].rx_queue_list[queue].port_id;
3751                         queueid = rx_thread[i].rx_queue_list[queue].queue_id;
3752
3753                         if (parse_ptype_on) {
3754                                 if (!rte_eth_add_rx_callback(portid, queueid,
3755                                                 cb_parse_ptype, NULL))
3756                                         rte_exit(EXIT_FAILURE,
3757                                                 "Failed to add rx callback: "
3758                                                 "port=%d\n", portid);
3759                         } else if (!check_ptype(portid))
3760                                 rte_exit(EXIT_FAILURE,
3761                                         "Port %d cannot parse packet type.\n\n"
3762                                         "Please add --parse-ptype to use sw "
3763                                         "packet type analyzer.\n\n",
3764                                         portid);
3765                 }
3766         }
3767
3768         check_all_ports_link_status(enabled_port_mask);
3769
3770         if (lthreads_on) {
3771                 printf("Starting L-Threading Model\n");
3772
3773 #if (APP_CPU_LOAD > 0)
3774                 if (cpu_load_lcore_id > 0)
3775                         /* Use one lcore for cpu load collector */
3776                         nb_lcores--;
3777 #endif
3778
3779                 lthread_num_schedulers_set(nb_lcores);
3780                 rte_eal_mp_remote_launch(sched_spawner, NULL, SKIP_MAIN);
3781                 lthread_main_spawner(NULL);
3782
3783         } else {
3784                 printf("Starting P-Threading Model\n");
3785                 /* launch per-lcore init on every lcore */
3786                 rte_eal_mp_remote_launch(pthread_run, NULL, CALL_MAIN);
3787                 RTE_LCORE_FOREACH_WORKER(lcore_id) {
3788                         if (rte_eal_wait_lcore(lcore_id) < 0)
3789                                 return -1;
3790                 }
3791         }
3792
3793         /* clean up the EAL */
3794         rte_eal_cleanup();
3795
3796         return 0;
3797 }