examples/l3fwd/l3fwd_em.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2016 Intel Corporation
   3  */
   4
   5 #include <stdio.h>
   6 #include <stdlib.h>
   7 #include <stdint.h>
   8 #include <inttypes.h>
   9 #include <sys/types.h>
  10 #include <string.h>
  11 #include <sys/queue.h>
  12 #include <stdarg.h>
  13 #include <errno.h>
  14 #include <getopt.h>
  15 #include <stdbool.h>
  16 #include <netinet/in.h>
  17
  18 #include <rte_debug.h>
  19 #include <rte_ether.h>
  20 #include <rte_ethdev.h>
  21 #include <rte_cycles.h>
  22 #include <rte_mbuf.h>
  23 #include <rte_ip.h>
  24 #include <rte_tcp.h>
  25 #include <rte_udp.h>
  26 #include <rte_hash.h>
  27
  28 #include "l3fwd.h"
  29
  30 #if defined(RTE_ARCH_X86) || defined(RTE_MACHINE_CPUFLAG_CRC32)
  31 #define EM_HASH_CRC 1
  32 #endif
  33
  34 #ifdef EM_HASH_CRC
  35 #include <rte_hash_crc.h>
  36 #define DEFAULT_HASH_FUNC       rte_hash_crc
  37 #else
  38 #include <rte_jhash.h>
  39 #define DEFAULT_HASH_FUNC       rte_jhash
  40 #endif
  41
  42 #define IPV6_ADDR_LEN 16
  43
  44 struct ipv4_5tuple {
  45         uint32_t ip_dst;
  46         uint32_t ip_src;
  47         uint16_t port_dst;
  48         uint16_t port_src;
  49         uint8_t  proto;
  50 } __attribute__((__packed__));
  51
  52 union ipv4_5tuple_host {
  53         struct {
  54                 uint8_t  pad0;
  55                 uint8_t  proto;
  56                 uint16_t pad1;
  57                 uint32_t ip_src;
  58                 uint32_t ip_dst;
  59                 uint16_t port_src;
  60                 uint16_t port_dst;
  61         };
  62         xmm_t xmm;
  63 };
  64
  65 #define XMM_NUM_IN_IPV6_5TUPLE 3
  66
  67 struct ipv6_5tuple {
  68         uint8_t  ip_dst[IPV6_ADDR_LEN];
  69         uint8_t  ip_src[IPV6_ADDR_LEN];
  70         uint16_t port_dst;
  71         uint16_t port_src;
  72         uint8_t  proto;
  73 } __attribute__((__packed__));
  74
  75 union ipv6_5tuple_host {
  76         struct {
  77                 uint16_t pad0;
  78                 uint8_t  proto;
  79                 uint8_t  pad1;
  80                 uint8_t  ip_src[IPV6_ADDR_LEN];
  81                 uint8_t  ip_dst[IPV6_ADDR_LEN];
  82                 uint16_t port_src;
  83                 uint16_t port_dst;
  84                 uint64_t reserve;
  85         };
  86         xmm_t xmm[XMM_NUM_IN_IPV6_5TUPLE];
  87 };
  88
  89
  90
  91 struct ipv4_l3fwd_em_route {
  92         struct ipv4_5tuple key;
  93         uint8_t if_out;
  94 };
  95
  96 struct ipv6_l3fwd_em_route {
  97         struct ipv6_5tuple key;
  98         uint8_t if_out;
  99 };
 100
 101 static struct ipv4_l3fwd_em_route ipv4_l3fwd_em_route_array[] = {
 102         {{RTE_IPV4(101, 0, 0, 0), RTE_IPV4(100, 10, 0, 1),  101, 11, IPPROTO_TCP}, 0},
 103         {{RTE_IPV4(201, 0, 0, 0), RTE_IPV4(200, 20, 0, 1),  102, 12, IPPROTO_TCP}, 1},
 104         {{RTE_IPV4(111, 0, 0, 0), RTE_IPV4(100, 30, 0, 1),  101, 11, IPPROTO_TCP}, 2},
 105         {{RTE_IPV4(211, 0, 0, 0), RTE_IPV4(200, 40, 0, 1),  102, 12, IPPROTO_TCP}, 3},
 106 };
 107
 108 static struct ipv6_l3fwd_em_route ipv6_l3fwd_em_route_array[] = {
 109         {{
 110         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 111         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 112         101, 11, IPPROTO_TCP}, 0},
 113
 114         {{
 115         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 116         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 117         102, 12, IPPROTO_TCP}, 1},
 118
 119         {{
 120         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 121         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 122         101, 11, IPPROTO_TCP}, 2},
 123
 124         {{
 125         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 126         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 127         102, 12, IPPROTO_TCP}, 3},
 128 };
 129
 130 struct rte_hash *ipv4_l3fwd_em_lookup_struct[NB_SOCKETS];
 131 struct rte_hash *ipv6_l3fwd_em_lookup_struct[NB_SOCKETS];
 132
 133 static inline uint32_t
 134 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
 135                 uint32_t init_val)
 136 {
 137         const union ipv4_5tuple_host *k;
 138         uint32_t t;
 139         const uint32_t *p;
 140
 141         k = data;
 142         t = k->proto;
 143         p = (const uint32_t *)&k->port_src;
 144
 145 #ifdef EM_HASH_CRC
 146         init_val = rte_hash_crc_4byte(t, init_val);
 147         init_val = rte_hash_crc_4byte(k->ip_src, init_val);
 148         init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
 149         init_val = rte_hash_crc_4byte(*p, init_val);
 150 #else
 151         init_val = rte_jhash_1word(t, init_val);
 152         init_val = rte_jhash_1word(k->ip_src, init_val);
 153         init_val = rte_jhash_1word(k->ip_dst, init_val);
 154         init_val = rte_jhash_1word(*p, init_val);
 155 #endif
 156
 157         return init_val;
 158 }
 159
 160 static inline uint32_t
 161 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len,
 162                 uint32_t init_val)
 163 {
 164         const union ipv6_5tuple_host *k;
 165         uint32_t t;
 166         const uint32_t *p;
 167 #ifdef EM_HASH_CRC
 168         const uint32_t  *ip_src0, *ip_src1, *ip_src2, *ip_src3;
 169         const uint32_t  *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
 170 #endif
 171
 172         k = data;
 173         t = k->proto;
 174         p = (const uint32_t *)&k->port_src;
 175
 176 #ifdef EM_HASH_CRC
 177         ip_src0 = (const uint32_t *) k->ip_src;
 178         ip_src1 = (const uint32_t *)(k->ip_src+4);
 179         ip_src2 = (const uint32_t *)(k->ip_src+8);
 180         ip_src3 = (const uint32_t *)(k->ip_src+12);
 181         ip_dst0 = (const uint32_t *) k->ip_dst;
 182         ip_dst1 = (const uint32_t *)(k->ip_dst+4);
 183         ip_dst2 = (const uint32_t *)(k->ip_dst+8);
 184         ip_dst3 = (const uint32_t *)(k->ip_dst+12);
 185         init_val = rte_hash_crc_4byte(t, init_val);
 186         init_val = rte_hash_crc_4byte(*ip_src0, init_val);
 187         init_val = rte_hash_crc_4byte(*ip_src1, init_val);
 188         init_val = rte_hash_crc_4byte(*ip_src2, init_val);
 189         init_val = rte_hash_crc_4byte(*ip_src3, init_val);
 190         init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
 191         init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
 192         init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
 193         init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
 194         init_val = rte_hash_crc_4byte(*p, init_val);
 195 #else
 196         init_val = rte_jhash_1word(t, init_val);
 197         init_val = rte_jhash(k->ip_src,
 198                         sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
 199         init_val = rte_jhash(k->ip_dst,
 200                         sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
 201         init_val = rte_jhash_1word(*p, init_val);
 202 #endif
 203         return init_val;
 204 }
 205
 206 #define IPV4_L3FWD_EM_NUM_ROUTES \
 207         (sizeof(ipv4_l3fwd_em_route_array) / sizeof(ipv4_l3fwd_em_route_array[0]))
 208
 209 #define IPV6_L3FWD_EM_NUM_ROUTES \
 210         (sizeof(ipv6_l3fwd_em_route_array) / sizeof(ipv6_l3fwd_em_route_array[0]))
 211
 212 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 213 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 214
 215 static rte_xmm_t mask0;
 216 static rte_xmm_t mask1;
 217 static rte_xmm_t mask2;
 218
 219 #if defined(RTE_MACHINE_CPUFLAG_SSE2)
 220 static inline xmm_t
 221 em_mask_key(void *key, xmm_t mask)
 222 {
 223         __m128i data = _mm_loadu_si128((__m128i *)(key));
 224
 225         return _mm_and_si128(data, mask);
 226 }
 227 #elif defined(RTE_MACHINE_CPUFLAG_NEON)
 228 static inline xmm_t
 229 em_mask_key(void *key, xmm_t mask)
 230 {
 231         int32x4_t data = vld1q_s32((int32_t *)key);
 232
 233         return vandq_s32(data, mask);
 234 }
 235 #elif defined(RTE_MACHINE_CPUFLAG_ALTIVEC)
 236 static inline xmm_t
 237 em_mask_key(void *key, xmm_t mask)
 238 {
 239         xmm_t data = vec_ld(0, (xmm_t *)(key));
 240
 241         return vec_and(data, mask);
 242 }
 243 #else
 244 #error No vector engine (SSE, NEON, ALTIVEC) available, check your toolchain
 245 #endif
 246
 247 static inline uint16_t
 248 em_get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid, void *lookup_struct)
 249 {
 250         int ret = 0;
 251         union ipv4_5tuple_host key;
 252         struct rte_hash *ipv4_l3fwd_lookup_struct =
 253                 (struct rte_hash *)lookup_struct;
 254
 255         ipv4_hdr = (uint8_t *)ipv4_hdr +
 256                 offsetof(struct rte_ipv4_hdr, time_to_live);
 257
 258         /*
 259          * Get 5 tuple: dst port, src port, dst IP address,
 260          * src IP address and protocol.
 261          */
 262         key.xmm = em_mask_key(ipv4_hdr, mask0.x);
 263
 264         /* Find destination port */
 265         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 266         return (ret < 0) ? portid : ipv4_l3fwd_out_if[ret];
 267 }
 268
 269 static inline uint16_t
 270 em_get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid, void *lookup_struct)
 271 {
 272         int ret = 0;
 273         union ipv6_5tuple_host key;
 274         struct rte_hash *ipv6_l3fwd_lookup_struct =
 275                 (struct rte_hash *)lookup_struct;
 276
 277         ipv6_hdr = (uint8_t *)ipv6_hdr +
 278                 offsetof(struct rte_ipv6_hdr, payload_len);
 279         void *data0 = ipv6_hdr;
 280         void *data1 = ((uint8_t *)ipv6_hdr) + sizeof(xmm_t);
 281         void *data2 = ((uint8_t *)ipv6_hdr) + sizeof(xmm_t) + sizeof(xmm_t);
 282
 283         /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
 284         key.xmm[0] = em_mask_key(data0, mask1.x);
 285
 286         /*
 287          * Get part of 5 tuple: dst IP address lower 96 bits
 288          * and src IP address higher 32 bits.
 289          */
 290 #if defined RTE_ARCH_X86
 291         key.xmm[1] = _mm_loadu_si128(data1);
 292 #else
 293         key.xmm[1] = *(xmm_t *)data1;
 294 #endif
 295
 296         /*
 297          * Get part of 5 tuple: dst port and src port
 298          * and dst IP address higher 32 bits.
 299          */
 300         key.xmm[2] = em_mask_key(data2, mask2.x);
 301
 302         /* Find destination port */
 303         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 304         return (ret < 0) ? portid : ipv6_l3fwd_out_if[ret];
 305 }
 306
 307 #if defined RTE_ARCH_X86 || defined RTE_MACHINE_CPUFLAG_NEON
 308 #if defined(NO_HASH_MULTI_LOOKUP)
 309 #include "l3fwd_em_sequential.h"
 310 #else
 311 #include "l3fwd_em_hlm.h"
 312 #endif
 313 #else
 314 #include "l3fwd_em.h"
 315 #endif
 316
 317 static void
 318 convert_ipv4_5tuple(struct ipv4_5tuple *key1,
 319                 union ipv4_5tuple_host *key2)
 320 {
 321         key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
 322         key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
 323         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
 324         key2->port_src = rte_cpu_to_be_16(key1->port_src);
 325         key2->proto = key1->proto;
 326         key2->pad0 = 0;
 327         key2->pad1 = 0;
 328 }
 329
 330 static void
 331 convert_ipv6_5tuple(struct ipv6_5tuple *key1,
 332                 union ipv6_5tuple_host *key2)
 333 {
 334         uint32_t i;
 335
 336         for (i = 0; i < 16; i++) {
 337                 key2->ip_dst[i] = key1->ip_dst[i];
 338                 key2->ip_src[i] = key1->ip_src[i];
 339         }
 340         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
 341         key2->port_src = rte_cpu_to_be_16(key1->port_src);
 342         key2->proto = key1->proto;
 343         key2->pad0 = 0;
 344         key2->pad1 = 0;
 345         key2->reserve = 0;
 346 }
 347
 348 #define BYTE_VALUE_MAX 256
 349 #define ALL_32_BITS 0xffffffff
 350 #define BIT_8_TO_15 0x0000ff00
 351
 352 static inline void
 353 populate_ipv4_few_flow_into_table(const struct rte_hash *h)
 354 {
 355         uint32_t i;
 356         int32_t ret;
 357
 358         mask0 = (rte_xmm_t){.u32 = {BIT_8_TO_15, ALL_32_BITS,
 359                                 ALL_32_BITS, ALL_32_BITS} };
 360
 361         for (i = 0; i < IPV4_L3FWD_EM_NUM_ROUTES; i++) {
 362                 struct ipv4_l3fwd_em_route  entry;
 363                 union ipv4_5tuple_host newkey;
 364
 365                 entry = ipv4_l3fwd_em_route_array[i];
 366                 convert_ipv4_5tuple(&entry.key, &newkey);
 367                 ret = rte_hash_add_key(h, (void *) &newkey);
 368                 if (ret < 0) {
 369                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
 370                                 " to the l3fwd hash.\n", i);
 371                 }
 372                 ipv4_l3fwd_out_if[ret] = entry.if_out;
 373         }
 374         printf("Hash: Adding 0x%" PRIx64 " keys\n",
 375                 (uint64_t)IPV4_L3FWD_EM_NUM_ROUTES);
 376 }
 377
 378 #define BIT_16_TO_23 0x00ff0000
 379 static inline void
 380 populate_ipv6_few_flow_into_table(const struct rte_hash *h)
 381 {
 382         uint32_t i;
 383         int32_t ret;
 384
 385         mask1 = (rte_xmm_t){.u32 = {BIT_16_TO_23, ALL_32_BITS,
 386                                 ALL_32_BITS, ALL_32_BITS} };
 387
 388         mask2 = (rte_xmm_t){.u32 = {ALL_32_BITS, ALL_32_BITS, 0, 0} };
 389
 390         for (i = 0; i < IPV6_L3FWD_EM_NUM_ROUTES; i++) {
 391                 struct ipv6_l3fwd_em_route entry;
 392                 union ipv6_5tuple_host newkey;
 393
 394                 entry = ipv6_l3fwd_em_route_array[i];
 395                 convert_ipv6_5tuple(&entry.key, &newkey);
 396                 ret = rte_hash_add_key(h, (void *) &newkey);
 397                 if (ret < 0) {
 398                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
 399                                 " to the l3fwd hash.\n", i);
 400                 }
 401                 ipv6_l3fwd_out_if[ret] = entry.if_out;
 402         }
 403         printf("Hash: Adding 0x%" PRIx64 "keys\n",
 404                 (uint64_t)IPV6_L3FWD_EM_NUM_ROUTES);
 405 }
 406
 407 #define NUMBER_PORT_USED 4
 408 static inline void
 409 populate_ipv4_many_flow_into_table(const struct rte_hash *h,
 410                 unsigned int nr_flow)
 411 {
 412         unsigned i;
 413
 414         mask0 = (rte_xmm_t){.u32 = {BIT_8_TO_15, ALL_32_BITS,
 415                                 ALL_32_BITS, ALL_32_BITS} };
 416
 417         for (i = 0; i < nr_flow; i++) {
 418                 struct ipv4_l3fwd_em_route entry;
 419                 union ipv4_5tuple_host newkey;
 420
 421                 uint8_t a = (uint8_t)
 422                         ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
 423                 uint8_t b = (uint8_t)
 424                         (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
 425                 uint8_t c = (uint8_t)
 426                         ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
 427
 428                 /* Create the ipv4 exact match flow */
 429                 memset(&entry, 0, sizeof(entry));
 430                 switch (i & (NUMBER_PORT_USED - 1)) {
 431                 case 0:
 432                         entry = ipv4_l3fwd_em_route_array[0];
 433                         entry.key.ip_dst = RTE_IPV4(101, c, b, a);
 434                         break;
 435                 case 1:
 436                         entry = ipv4_l3fwd_em_route_array[1];
 437                         entry.key.ip_dst = RTE_IPV4(201, c, b, a);
 438                         break;
 439                 case 2:
 440                         entry = ipv4_l3fwd_em_route_array[2];
 441                         entry.key.ip_dst = RTE_IPV4(111, c, b, a);
 442                         break;
 443                 case 3:
 444                         entry = ipv4_l3fwd_em_route_array[3];
 445                         entry.key.ip_dst = RTE_IPV4(211, c, b, a);
 446                         break;
 447                 };
 448                 convert_ipv4_5tuple(&entry.key, &newkey);
 449                 int32_t ret = rte_hash_add_key(h, (void *) &newkey);
 450
 451                 if (ret < 0)
 452                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
 453
 454                 ipv4_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
 455
 456         }
 457         printf("Hash: Adding 0x%x keys\n", nr_flow);
 458 }
 459
 460 static inline void
 461 populate_ipv6_many_flow_into_table(const struct rte_hash *h,
 462                 unsigned int nr_flow)
 463 {
 464         unsigned i;
 465
 466         mask1 = (rte_xmm_t){.u32 = {BIT_16_TO_23, ALL_32_BITS,
 467                                 ALL_32_BITS, ALL_32_BITS} };
 468         mask2 = (rte_xmm_t){.u32 = {ALL_32_BITS, ALL_32_BITS, 0, 0} };
 469
 470         for (i = 0; i < nr_flow; i++) {
 471                 struct ipv6_l3fwd_em_route entry;
 472                 union ipv6_5tuple_host newkey;
 473
 474                 uint8_t a = (uint8_t)
 475                         ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
 476                 uint8_t b = (uint8_t)
 477                         (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
 478                 uint8_t c = (uint8_t)
 479                         ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
 480
 481                 /* Create the ipv6 exact match flow */
 482                 memset(&entry, 0, sizeof(entry));
 483                 switch (i & (NUMBER_PORT_USED - 1)) {
 484                 case 0:
 485                         entry = ipv6_l3fwd_em_route_array[0];
 486                         break;
 487                 case 1:
 488                         entry = ipv6_l3fwd_em_route_array[1];
 489                         break;
 490                 case 2:
 491                         entry = ipv6_l3fwd_em_route_array[2];
 492                         break;
 493                 case 3:
 494                         entry = ipv6_l3fwd_em_route_array[3];
 495                         break;
 496                 };
 497                 entry.key.ip_dst[13] = c;
 498                 entry.key.ip_dst[14] = b;
 499                 entry.key.ip_dst[15] = a;
 500                 convert_ipv6_5tuple(&entry.key, &newkey);
 501                 int32_t ret = rte_hash_add_key(h, (void *) &newkey);
 502
 503                 if (ret < 0)
 504                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
 505
 506                 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
 507
 508         }
 509         printf("Hash: Adding 0x%x keys\n", nr_flow);
 510 }
 511
 512 /* Requirements:
 513  * 1. IP packets without extension;
 514  * 2. L4 payload should be either TCP or UDP.
 515  */
 516 int
 517 em_check_ptype(int portid)
 518 {
 519         int i, ret;
 520         int ptype_l3_ipv4_ext = 0;
 521         int ptype_l3_ipv6_ext = 0;
 522         int ptype_l4_tcp = 0;
 523         int ptype_l4_udp = 0;
 524         uint32_t ptype_mask = RTE_PTYPE_L3_MASK | RTE_PTYPE_L4_MASK;
 525
 526         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
 527         if (ret <= 0)
 528                 return 0;
 529
 530         uint32_t ptypes[ret];
 531
 532         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
 533         for (i = 0; i < ret; ++i) {
 534                 switch (ptypes[i]) {
 535                 case RTE_PTYPE_L3_IPV4_EXT:
 536                         ptype_l3_ipv4_ext = 1;
 537                         break;
 538                 case RTE_PTYPE_L3_IPV6_EXT:
 539                         ptype_l3_ipv6_ext = 1;
 540                         break;
 541                 case RTE_PTYPE_L4_TCP:
 542                         ptype_l4_tcp = 1;
 543                         break;
 544                 case RTE_PTYPE_L4_UDP:
 545                         ptype_l4_udp = 1;
 546                         break;
 547                 }
 548         }
 549
 550         if (ptype_l3_ipv4_ext == 0)
 551                 printf("port %d cannot parse RTE_PTYPE_L3_IPV4_EXT\n", portid);
 552         if (ptype_l3_ipv6_ext == 0)
 553                 printf("port %d cannot parse RTE_PTYPE_L3_IPV6_EXT\n", portid);
 554         if (!ptype_l3_ipv4_ext || !ptype_l3_ipv6_ext)
 555                 return 0;
 556
 557         if (ptype_l4_tcp == 0)
 558                 printf("port %d cannot parse RTE_PTYPE_L4_TCP\n", portid);
 559         if (ptype_l4_udp == 0)
 560                 printf("port %d cannot parse RTE_PTYPE_L4_UDP\n", portid);
 561         if (ptype_l4_tcp && ptype_l4_udp)
 562                 return 1;
 563
 564         return 0;
 565 }
 566
 567 static inline void
 568 em_parse_ptype(struct rte_mbuf *m)
 569 {
 570         struct rte_ether_hdr *eth_hdr;
 571         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 572         uint16_t ether_type;
 573         void *l3;
 574         int hdr_len;
 575         struct rte_ipv4_hdr *ipv4_hdr;
 576         struct rte_ipv6_hdr *ipv6_hdr;
 577
 578         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 579         ether_type = eth_hdr->ether_type;
 580         l3 = (uint8_t *)eth_hdr + sizeof(struct rte_ether_hdr);
 581         if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
 582                 ipv4_hdr = (struct rte_ipv4_hdr *)l3;
 583                 hdr_len = (ipv4_hdr->version_ihl & RTE_IPV4_HDR_IHL_MASK) *
 584                           RTE_IPV4_IHL_MULTIPLIER;
 585                 if (hdr_len == sizeof(struct rte_ipv4_hdr)) {
 586                         packet_type |= RTE_PTYPE_L3_IPV4;
 587                         if (ipv4_hdr->next_proto_id == IPPROTO_TCP)
 588                                 packet_type |= RTE_PTYPE_L4_TCP;
 589                         else if (ipv4_hdr->next_proto_id == IPPROTO_UDP)
 590                                 packet_type |= RTE_PTYPE_L4_UDP;
 591                 } else
 592                         packet_type |= RTE_PTYPE_L3_IPV4_EXT;
 593         } else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
 594                 ipv6_hdr = (struct rte_ipv6_hdr *)l3;
 595                 if (ipv6_hdr->proto == IPPROTO_TCP)
 596                         packet_type |= RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_TCP;
 597                 else if (ipv6_hdr->proto == IPPROTO_UDP)
 598                         packet_type |= RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_UDP;
 599                 else
 600                         packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 601         }
 602
 603         m->packet_type = packet_type;
 604 }
 605
 606 uint16_t
 607 em_cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
 608                   struct rte_mbuf *pkts[], uint16_t nb_pkts,
 609                   uint16_t max_pkts __rte_unused,
 610                   void *user_param __rte_unused)
 611 {
 612         unsigned i;
 613
 614         for (i = 0; i < nb_pkts; ++i)
 615                 em_parse_ptype(pkts[i]);
 616
 617         return nb_pkts;
 618 }
 619
 620 /* main processing loop */
 621 int
 622 em_main_loop(__attribute__((unused)) void *dummy)
 623 {
 624         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 625         unsigned lcore_id;
 626         uint64_t prev_tsc, diff_tsc, cur_tsc;
 627         int i, nb_rx;
 628         uint8_t queueid;
 629         uint16_t portid;
 630         struct lcore_conf *qconf;
 631         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
 632                 US_PER_S * BURST_TX_DRAIN_US;
 633
 634         prev_tsc = 0;
 635
 636         lcore_id = rte_lcore_id();
 637         qconf = &lcore_conf[lcore_id];
 638
 639         if (qconf->n_rx_queue == 0) {
 640                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
 641                 return 0;
 642         }
 643
 644         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
 645
 646         for (i = 0; i < qconf->n_rx_queue; i++) {
 647
 648                 portid = qconf->rx_queue_list[i].port_id;
 649                 queueid = qconf->rx_queue_list[i].queue_id;
 650                 RTE_LOG(INFO, L3FWD,
 651                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
 652                         lcore_id, portid, queueid);
 653         }
 654
 655         while (!force_quit) {
 656
 657                 cur_tsc = rte_rdtsc();
 658
 659                 /*
 660                  * TX burst queue drain
 661                  */
 662                 diff_tsc = cur_tsc - prev_tsc;
 663                 if (unlikely(diff_tsc > drain_tsc)) {
 664
 665                         for (i = 0; i < qconf->n_tx_port; ++i) {
 666                                 portid = qconf->tx_port_id[i];
 667                                 if (qconf->tx_mbufs[portid].len == 0)
 668                                         continue;
 669                                 send_burst(qconf,
 670                                         qconf->tx_mbufs[portid].len,
 671                                         portid);
 672                                 qconf->tx_mbufs[portid].len = 0;
 673                         }
 674
 675                         prev_tsc = cur_tsc;
 676                 }
 677
 678                 /*
 679                  * Read packet from RX queues
 680                  */
 681                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 682                         portid = qconf->rx_queue_list[i].port_id;
 683                         queueid = qconf->rx_queue_list[i].queue_id;
 684                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 685                                 MAX_PKT_BURST);
 686                         if (nb_rx == 0)
 687                                 continue;
 688
 689 #if defined RTE_ARCH_X86 || defined RTE_MACHINE_CPUFLAG_NEON
 690                         l3fwd_em_send_packets(nb_rx, pkts_burst,
 691                                                         portid, qconf);
 692 #else
 693                         l3fwd_em_no_opt_send_packets(nb_rx, pkts_burst,
 694                                                         portid, qconf);
 695 #endif
 696                 }
 697         }
 698
 699         return 0;
 700 }
 701
 702 /*
 703  * Initialize exact match (hash) parameters.
 704  */
 705 void
 706 setup_hash(const int socketid)
 707 {
 708         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
 709                 .name = NULL,
 710                 .entries = L3FWD_HASH_ENTRIES,
 711                 .key_len = sizeof(union ipv4_5tuple_host),
 712                 .hash_func = ipv4_hash_crc,
 713                 .hash_func_init_val = 0,
 714         };
 715
 716         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
 717                 .name = NULL,
 718                 .entries = L3FWD_HASH_ENTRIES,
 719                 .key_len = sizeof(union ipv6_5tuple_host),
 720                 .hash_func = ipv6_hash_crc,
 721                 .hash_func_init_val = 0,
 722         };
 723
 724         char s[64];
 725
 726         /* create ipv4 hash */
 727         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
 728         ipv4_l3fwd_hash_params.name = s;
 729         ipv4_l3fwd_hash_params.socket_id = socketid;
 730         ipv4_l3fwd_em_lookup_struct[socketid] =
 731                 rte_hash_create(&ipv4_l3fwd_hash_params);
 732         if (ipv4_l3fwd_em_lookup_struct[socketid] == NULL)
 733                 rte_exit(EXIT_FAILURE,
 734                         "Unable to create the l3fwd hash on socket %d\n",
 735                         socketid);
 736
 737         /* create ipv6 hash */
 738         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
 739         ipv6_l3fwd_hash_params.name = s;
 740         ipv6_l3fwd_hash_params.socket_id = socketid;
 741         ipv6_l3fwd_em_lookup_struct[socketid] =
 742                 rte_hash_create(&ipv6_l3fwd_hash_params);
 743         if (ipv6_l3fwd_em_lookup_struct[socketid] == NULL)
 744                 rte_exit(EXIT_FAILURE,
 745                         "Unable to create the l3fwd hash on socket %d\n",
 746                         socketid);
 747
 748         if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
 749                 /* For testing hash matching with a large number of flows we
 750                  * generate millions of IP 5-tuples with an incremented dst
 751                  * address to initialize the hash table. */
 752                 if (ipv6 == 0) {
 753                         /* populate the ipv4 hash */
 754                         populate_ipv4_many_flow_into_table(
 755                                 ipv4_l3fwd_em_lookup_struct[socketid],
 756                                 hash_entry_number);
 757                 } else {
 758                         /* populate the ipv6 hash */
 759                         populate_ipv6_many_flow_into_table(
 760                                 ipv6_l3fwd_em_lookup_struct[socketid],
 761                                 hash_entry_number);
 762                 }
 763         } else {
 764                 /*
 765                  * Use data in ipv4/ipv6 l3fwd lookup table
 766                  * directly to initialize the hash table.
 767                  */
 768                 if (ipv6 == 0) {
 769                         /* populate the ipv4 hash */
 770                         populate_ipv4_few_flow_into_table(
 771                                 ipv4_l3fwd_em_lookup_struct[socketid]);
 772                 } else {
 773                         /* populate the ipv6 hash */
 774                         populate_ipv6_few_flow_into_table(
 775                                 ipv6_l3fwd_em_lookup_struct[socketid]);
 776                 }
 777         }
 778 }
 779
 780 /* Return ipv4/ipv6 em fwd lookup struct. */
 781 void *
 782 em_get_ipv4_l3fwd_lookup_struct(const int socketid)
 783 {
 784         return ipv4_l3fwd_em_lookup_struct[socketid];
 785 }
 786
 787 void *
 788 em_get_ipv6_l3fwd_lookup_struct(const int socketid)
 789 {
 790         return ipv6_l3fwd_em_lookup_struct[socketid];
 791 }