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 #include "l3fwd_event.h"
  30
  31 #if defined(RTE_ARCH_X86) || defined(RTE_MACHINE_CPUFLAG_CRC32)
  32 #define EM_HASH_CRC 1
  33 #endif
  34
  35 #ifdef EM_HASH_CRC
  36 #include <rte_hash_crc.h>
  37 #define DEFAULT_HASH_FUNC       rte_hash_crc
  38 #else
  39 #include <rte_jhash.h>
  40 #define DEFAULT_HASH_FUNC       rte_jhash
  41 #endif
  42
  43 #define IPV6_ADDR_LEN 16
  44
  45 struct ipv4_5tuple {
  46         uint32_t ip_dst;
  47         uint32_t ip_src;
  48         uint16_t port_dst;
  49         uint16_t port_src;
  50         uint8_t  proto;
  51 } __attribute__((__packed__));
  52
  53 union ipv4_5tuple_host {
  54         struct {
  55                 uint8_t  pad0;
  56                 uint8_t  proto;
  57                 uint16_t pad1;
  58                 uint32_t ip_src;
  59                 uint32_t ip_dst;
  60                 uint16_t port_src;
  61                 uint16_t port_dst;
  62         };
  63         xmm_t xmm;
  64 };
  65
  66 #define XMM_NUM_IN_IPV6_5TUPLE 3
  67
  68 struct ipv6_5tuple {
  69         uint8_t  ip_dst[IPV6_ADDR_LEN];
  70         uint8_t  ip_src[IPV6_ADDR_LEN];
  71         uint16_t port_dst;
  72         uint16_t port_src;
  73         uint8_t  proto;
  74 } __attribute__((__packed__));
  75
  76 union ipv6_5tuple_host {
  77         struct {
  78                 uint16_t pad0;
  79                 uint8_t  proto;
  80                 uint8_t  pad1;
  81                 uint8_t  ip_src[IPV6_ADDR_LEN];
  82                 uint8_t  ip_dst[IPV6_ADDR_LEN];
  83                 uint16_t port_src;
  84                 uint16_t port_dst;
  85                 uint64_t reserve;
  86         };
  87         xmm_t xmm[XMM_NUM_IN_IPV6_5TUPLE];
  88 };
  89
  90
  91
  92 struct ipv4_l3fwd_em_route {
  93         struct ipv4_5tuple key;
  94         uint8_t if_out;
  95 };
  96
  97 struct ipv6_l3fwd_em_route {
  98         struct ipv6_5tuple key;
  99         uint8_t if_out;
 100 };
 101
 102 static struct ipv4_l3fwd_em_route ipv4_l3fwd_em_route_array[] = {
 103         {{RTE_IPV4(101, 0, 0, 0), RTE_IPV4(100, 10, 0, 1),  101, 11, IPPROTO_TCP}, 0},
 104         {{RTE_IPV4(201, 0, 0, 0), RTE_IPV4(200, 20, 0, 1),  102, 12, IPPROTO_TCP}, 1},
 105         {{RTE_IPV4(111, 0, 0, 0), RTE_IPV4(100, 30, 0, 1),  101, 11, IPPROTO_TCP}, 2},
 106         {{RTE_IPV4(211, 0, 0, 0), RTE_IPV4(200, 40, 0, 1),  102, 12, IPPROTO_TCP}, 3},
 107 };
 108
 109 static struct ipv6_l3fwd_em_route ipv6_l3fwd_em_route_array[] = {
 110         {{
 111         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 112         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 113         101, 11, IPPROTO_TCP}, 0},
 114
 115         {{
 116         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 117         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 118         102, 12, IPPROTO_TCP}, 1},
 119
 120         {{
 121         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 122         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 123         101, 11, IPPROTO_TCP}, 2},
 124
 125         {{
 126         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 127         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 128         102, 12, IPPROTO_TCP}, 3},
 129 };
 130
 131 struct rte_hash *ipv4_l3fwd_em_lookup_struct[NB_SOCKETS];
 132 struct rte_hash *ipv6_l3fwd_em_lookup_struct[NB_SOCKETS];
 133
 134 static inline uint32_t
 135 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
 136                 uint32_t init_val)
 137 {
 138         const union ipv4_5tuple_host *k;
 139         uint32_t t;
 140         const uint32_t *p;
 141
 142         k = data;
 143         t = k->proto;
 144         p = (const uint32_t *)&k->port_src;
 145
 146 #ifdef EM_HASH_CRC
 147         init_val = rte_hash_crc_4byte(t, init_val);
 148         init_val = rte_hash_crc_4byte(k->ip_src, init_val);
 149         init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
 150         init_val = rte_hash_crc_4byte(*p, init_val);
 151 #else
 152         init_val = rte_jhash_1word(t, init_val);
 153         init_val = rte_jhash_1word(k->ip_src, init_val);
 154         init_val = rte_jhash_1word(k->ip_dst, init_val);
 155         init_val = rte_jhash_1word(*p, init_val);
 156 #endif
 157
 158         return init_val;
 159 }
 160
 161 static inline uint32_t
 162 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len,
 163                 uint32_t init_val)
 164 {
 165         const union ipv6_5tuple_host *k;
 166         uint32_t t;
 167         const uint32_t *p;
 168 #ifdef EM_HASH_CRC
 169         const uint32_t  *ip_src0, *ip_src1, *ip_src2, *ip_src3;
 170         const uint32_t  *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
 171 #endif
 172
 173         k = data;
 174         t = k->proto;
 175         p = (const uint32_t *)&k->port_src;
 176
 177 #ifdef EM_HASH_CRC
 178         ip_src0 = (const uint32_t *) k->ip_src;
 179         ip_src1 = (const uint32_t *)(k->ip_src+4);
 180         ip_src2 = (const uint32_t *)(k->ip_src+8);
 181         ip_src3 = (const uint32_t *)(k->ip_src+12);
 182         ip_dst0 = (const uint32_t *) k->ip_dst;
 183         ip_dst1 = (const uint32_t *)(k->ip_dst+4);
 184         ip_dst2 = (const uint32_t *)(k->ip_dst+8);
 185         ip_dst3 = (const uint32_t *)(k->ip_dst+12);
 186         init_val = rte_hash_crc_4byte(t, init_val);
 187         init_val = rte_hash_crc_4byte(*ip_src0, init_val);
 188         init_val = rte_hash_crc_4byte(*ip_src1, init_val);
 189         init_val = rte_hash_crc_4byte(*ip_src2, init_val);
 190         init_val = rte_hash_crc_4byte(*ip_src3, init_val);
 191         init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
 192         init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
 193         init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
 194         init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
 195         init_val = rte_hash_crc_4byte(*p, init_val);
 196 #else
 197         init_val = rte_jhash_1word(t, init_val);
 198         init_val = rte_jhash(k->ip_src,
 199                         sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
 200         init_val = rte_jhash(k->ip_dst,
 201                         sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
 202         init_val = rte_jhash_1word(*p, init_val);
 203 #endif
 204         return init_val;
 205 }
 206
 207 #define IPV4_L3FWD_EM_NUM_ROUTES \
 208         (sizeof(ipv4_l3fwd_em_route_array) / sizeof(ipv4_l3fwd_em_route_array[0]))
 209
 210 #define IPV6_L3FWD_EM_NUM_ROUTES \
 211         (sizeof(ipv6_l3fwd_em_route_array) / sizeof(ipv6_l3fwd_em_route_array[0]))
 212
 213 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 214 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 215
 216 static rte_xmm_t mask0;
 217 static rte_xmm_t mask1;
 218 static rte_xmm_t mask2;
 219
 220 #if defined(RTE_MACHINE_CPUFLAG_SSE2)
 221 static inline xmm_t
 222 em_mask_key(void *key, xmm_t mask)
 223 {
 224         __m128i data = _mm_loadu_si128((__m128i *)(key));
 225
 226         return _mm_and_si128(data, mask);
 227 }
 228 #elif defined(RTE_MACHINE_CPUFLAG_NEON)
 229 static inline xmm_t
 230 em_mask_key(void *key, xmm_t mask)
 231 {
 232         int32x4_t data = vld1q_s32((int32_t *)key);
 233
 234         return vandq_s32(data, mask);
 235 }
 236 #elif defined(RTE_MACHINE_CPUFLAG_ALTIVEC)
 237 static inline xmm_t
 238 em_mask_key(void *key, xmm_t mask)
 239 {
 240         xmm_t data = vec_ld(0, (xmm_t *)(key));
 241
 242         return vec_and(data, mask);
 243 }
 244 #else
 245 #error No vector engine (SSE, NEON, ALTIVEC) available, check your toolchain
 246 #endif
 247
 248 static inline uint16_t
 249 em_get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid, void *lookup_struct)
 250 {
 251         int ret = 0;
 252         union ipv4_5tuple_host key;
 253         struct rte_hash *ipv4_l3fwd_lookup_struct =
 254                 (struct rte_hash *)lookup_struct;
 255
 256         ipv4_hdr = (uint8_t *)ipv4_hdr +
 257                 offsetof(struct rte_ipv4_hdr, time_to_live);
 258
 259         /*
 260          * Get 5 tuple: dst port, src port, dst IP address,
 261          * src IP address and protocol.
 262          */
 263         key.xmm = em_mask_key(ipv4_hdr, mask0.x);
 264
 265         /* Find destination port */
 266         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 267         return (ret < 0) ? portid : ipv4_l3fwd_out_if[ret];
 268 }
 269
 270 static inline uint16_t
 271 em_get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid, void *lookup_struct)
 272 {
 273         int ret = 0;
 274         union ipv6_5tuple_host key;
 275         struct rte_hash *ipv6_l3fwd_lookup_struct =
 276                 (struct rte_hash *)lookup_struct;
 277
 278         ipv6_hdr = (uint8_t *)ipv6_hdr +
 279                 offsetof(struct rte_ipv6_hdr, payload_len);
 280         void *data0 = ipv6_hdr;
 281         void *data1 = ((uint8_t *)ipv6_hdr) + sizeof(xmm_t);
 282         void *data2 = ((uint8_t *)ipv6_hdr) + sizeof(xmm_t) + sizeof(xmm_t);
 283
 284         /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
 285         key.xmm[0] = em_mask_key(data0, mask1.x);
 286
 287         /*
 288          * Get part of 5 tuple: dst IP address lower 96 bits
 289          * and src IP address higher 32 bits.
 290          */
 291 #if defined RTE_ARCH_X86
 292         key.xmm[1] = _mm_loadu_si128(data1);
 293 #else
 294         key.xmm[1] = *(xmm_t *)data1;
 295 #endif
 296
 297         /*
 298          * Get part of 5 tuple: dst port and src port
 299          * and dst IP address higher 32 bits.
 300          */
 301         key.xmm[2] = em_mask_key(data2, mask2.x);
 302
 303         /* Find destination port */
 304         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 305         return (ret < 0) ? portid : ipv6_l3fwd_out_if[ret];
 306 }
 307
 308 #if defined RTE_ARCH_X86 || defined RTE_MACHINE_CPUFLAG_NEON
 309 #if defined(NO_HASH_MULTI_LOOKUP)
 310 #include "l3fwd_em_sequential.h"
 311 #else
 312 #include "l3fwd_em_hlm.h"
 313 #endif
 314 #else
 315 #include "l3fwd_em.h"
 316 #endif
 317
 318 static void
 319 convert_ipv4_5tuple(struct ipv4_5tuple *key1,
 320                 union ipv4_5tuple_host *key2)
 321 {
 322         key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
 323         key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
 324         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
 325         key2->port_src = rte_cpu_to_be_16(key1->port_src);
 326         key2->proto = key1->proto;
 327         key2->pad0 = 0;
 328         key2->pad1 = 0;
 329 }
 330
 331 static void
 332 convert_ipv6_5tuple(struct ipv6_5tuple *key1,
 333                 union ipv6_5tuple_host *key2)
 334 {
 335         uint32_t i;
 336
 337         for (i = 0; i < 16; i++) {
 338                 key2->ip_dst[i] = key1->ip_dst[i];
 339                 key2->ip_src[i] = key1->ip_src[i];
 340         }
 341         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
 342         key2->port_src = rte_cpu_to_be_16(key1->port_src);
 343         key2->proto = key1->proto;
 344         key2->pad0 = 0;
 345         key2->pad1 = 0;
 346         key2->reserve = 0;
 347 }
 348
 349 #define BYTE_VALUE_MAX 256
 350 #define ALL_32_BITS 0xffffffff
 351 #define BIT_8_TO_15 0x0000ff00
 352
 353 static inline void
 354 populate_ipv4_few_flow_into_table(const struct rte_hash *h)
 355 {
 356         uint32_t i;
 357         int32_t ret;
 358
 359         mask0 = (rte_xmm_t){.u32 = {BIT_8_TO_15, ALL_32_BITS,
 360                                 ALL_32_BITS, ALL_32_BITS} };
 361
 362         for (i = 0; i < IPV4_L3FWD_EM_NUM_ROUTES; i++) {
 363                 struct ipv4_l3fwd_em_route  entry;
 364                 union ipv4_5tuple_host newkey;
 365
 366                 entry = ipv4_l3fwd_em_route_array[i];
 367                 convert_ipv4_5tuple(&entry.key, &newkey);
 368                 ret = rte_hash_add_key(h, (void *) &newkey);
 369                 if (ret < 0) {
 370                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
 371                                 " to the l3fwd hash.\n", i);
 372                 }
 373                 ipv4_l3fwd_out_if[ret] = entry.if_out;
 374         }
 375         printf("Hash: Adding 0x%" PRIx64 " keys\n",
 376                 (uint64_t)IPV4_L3FWD_EM_NUM_ROUTES);
 377 }
 378
 379 #define BIT_16_TO_23 0x00ff0000
 380 static inline void
 381 populate_ipv6_few_flow_into_table(const struct rte_hash *h)
 382 {
 383         uint32_t i;
 384         int32_t ret;
 385
 386         mask1 = (rte_xmm_t){.u32 = {BIT_16_TO_23, ALL_32_BITS,
 387                                 ALL_32_BITS, ALL_32_BITS} };
 388
 389         mask2 = (rte_xmm_t){.u32 = {ALL_32_BITS, ALL_32_BITS, 0, 0} };
 390
 391         for (i = 0; i < IPV6_L3FWD_EM_NUM_ROUTES; i++) {
 392                 struct ipv6_l3fwd_em_route entry;
 393                 union ipv6_5tuple_host newkey;
 394
 395                 entry = ipv6_l3fwd_em_route_array[i];
 396                 convert_ipv6_5tuple(&entry.key, &newkey);
 397                 ret = rte_hash_add_key(h, (void *) &newkey);
 398                 if (ret < 0) {
 399                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
 400                                 " to the l3fwd hash.\n", i);
 401                 }
 402                 ipv6_l3fwd_out_if[ret] = entry.if_out;
 403         }
 404         printf("Hash: Adding 0x%" PRIx64 "keys\n",
 405                 (uint64_t)IPV6_L3FWD_EM_NUM_ROUTES);
 406 }
 407
 408 #define NUMBER_PORT_USED 4
 409 static inline void
 410 populate_ipv4_many_flow_into_table(const struct rte_hash *h,
 411                 unsigned int nr_flow)
 412 {
 413         unsigned i;
 414
 415         mask0 = (rte_xmm_t){.u32 = {BIT_8_TO_15, ALL_32_BITS,
 416                                 ALL_32_BITS, ALL_32_BITS} };
 417
 418         for (i = 0; i < nr_flow; i++) {
 419                 struct ipv4_l3fwd_em_route entry;
 420                 union ipv4_5tuple_host newkey;
 421
 422                 uint8_t a = (uint8_t)
 423                         ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
 424                 uint8_t b = (uint8_t)
 425                         (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
 426                 uint8_t c = (uint8_t)
 427                         ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
 428
 429                 /* Create the ipv4 exact match flow */
 430                 memset(&entry, 0, sizeof(entry));
 431                 switch (i & (NUMBER_PORT_USED - 1)) {
 432                 case 0:
 433                         entry = ipv4_l3fwd_em_route_array[0];
 434                         entry.key.ip_dst = RTE_IPV4(101, c, b, a);
 435                         break;
 436                 case 1:
 437                         entry = ipv4_l3fwd_em_route_array[1];
 438                         entry.key.ip_dst = RTE_IPV4(201, c, b, a);
 439                         break;
 440                 case 2:
 441                         entry = ipv4_l3fwd_em_route_array[2];
 442                         entry.key.ip_dst = RTE_IPV4(111, c, b, a);
 443                         break;
 444                 case 3:
 445                         entry = ipv4_l3fwd_em_route_array[3];
 446                         entry.key.ip_dst = RTE_IPV4(211, c, b, a);
 447                         break;
 448                 };
 449                 convert_ipv4_5tuple(&entry.key, &newkey);
 450                 int32_t ret = rte_hash_add_key(h, (void *) &newkey);
 451
 452                 if (ret < 0)
 453                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
 454
 455                 ipv4_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
 456
 457         }
 458         printf("Hash: Adding 0x%x keys\n", nr_flow);
 459 }
 460
 461 static inline void
 462 populate_ipv6_many_flow_into_table(const struct rte_hash *h,
 463                 unsigned int nr_flow)
 464 {
 465         unsigned i;
 466
 467         mask1 = (rte_xmm_t){.u32 = {BIT_16_TO_23, ALL_32_BITS,
 468                                 ALL_32_BITS, ALL_32_BITS} };
 469         mask2 = (rte_xmm_t){.u32 = {ALL_32_BITS, ALL_32_BITS, 0, 0} };
 470
 471         for (i = 0; i < nr_flow; i++) {
 472                 struct ipv6_l3fwd_em_route entry;
 473                 union ipv6_5tuple_host newkey;
 474
 475                 uint8_t a = (uint8_t)
 476                         ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
 477                 uint8_t b = (uint8_t)
 478                         (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
 479                 uint8_t c = (uint8_t)
 480                         ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
 481
 482                 /* Create the ipv6 exact match flow */
 483                 memset(&entry, 0, sizeof(entry));
 484                 switch (i & (NUMBER_PORT_USED - 1)) {
 485                 case 0:
 486                         entry = ipv6_l3fwd_em_route_array[0];
 487                         break;
 488                 case 1:
 489                         entry = ipv6_l3fwd_em_route_array[1];
 490                         break;
 491                 case 2:
 492                         entry = ipv6_l3fwd_em_route_array[2];
 493                         break;
 494                 case 3:
 495                         entry = ipv6_l3fwd_em_route_array[3];
 496                         break;
 497                 };
 498                 entry.key.ip_dst[13] = c;
 499                 entry.key.ip_dst[14] = b;
 500                 entry.key.ip_dst[15] = a;
 501                 convert_ipv6_5tuple(&entry.key, &newkey);
 502                 int32_t ret = rte_hash_add_key(h, (void *) &newkey);
 503
 504                 if (ret < 0)
 505                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
 506
 507                 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
 508
 509         }
 510         printf("Hash: Adding 0x%x keys\n", nr_flow);
 511 }
 512
 513 /* Requirements:
 514  * 1. IP packets without extension;
 515  * 2. L4 payload should be either TCP or UDP.
 516  */
 517 int
 518 em_check_ptype(int portid)
 519 {
 520         int i, ret;
 521         int ptype_l3_ipv4_ext = 0;
 522         int ptype_l3_ipv6_ext = 0;
 523         int ptype_l4_tcp = 0;
 524         int ptype_l4_udp = 0;
 525         uint32_t ptype_mask = RTE_PTYPE_L3_MASK | RTE_PTYPE_L4_MASK;
 526
 527         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
 528         if (ret <= 0)
 529                 return 0;
 530
 531         uint32_t ptypes[ret];
 532
 533         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
 534         for (i = 0; i < ret; ++i) {
 535                 switch (ptypes[i]) {
 536                 case RTE_PTYPE_L3_IPV4_EXT:
 537                         ptype_l3_ipv4_ext = 1;
 538                         break;
 539                 case RTE_PTYPE_L3_IPV6_EXT:
 540                         ptype_l3_ipv6_ext = 1;
 541                         break;
 542                 case RTE_PTYPE_L4_TCP:
 543                         ptype_l4_tcp = 1;
 544                         break;
 545                 case RTE_PTYPE_L4_UDP:
 546                         ptype_l4_udp = 1;
 547                         break;
 548                 }
 549         }
 550
 551         if (ptype_l3_ipv4_ext == 0)
 552                 printf("port %d cannot parse RTE_PTYPE_L3_IPV4_EXT\n", portid);
 553         if (ptype_l3_ipv6_ext == 0)
 554                 printf("port %d cannot parse RTE_PTYPE_L3_IPV6_EXT\n", portid);
 555         if (!ptype_l3_ipv4_ext || !ptype_l3_ipv6_ext)
 556                 return 0;
 557
 558         if (ptype_l4_tcp == 0)
 559                 printf("port %d cannot parse RTE_PTYPE_L4_TCP\n", portid);
 560         if (ptype_l4_udp == 0)
 561                 printf("port %d cannot parse RTE_PTYPE_L4_UDP\n", portid);
 562         if (ptype_l4_tcp && ptype_l4_udp)
 563                 return 1;
 564
 565         return 0;
 566 }
 567
 568 static inline void
 569 em_parse_ptype(struct rte_mbuf *m)
 570 {
 571         struct rte_ether_hdr *eth_hdr;
 572         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 573         uint16_t ether_type;
 574         void *l3;
 575         int hdr_len;
 576         struct rte_ipv4_hdr *ipv4_hdr;
 577         struct rte_ipv6_hdr *ipv6_hdr;
 578
 579         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 580         ether_type = eth_hdr->ether_type;
 581         l3 = (uint8_t *)eth_hdr + sizeof(struct rte_ether_hdr);
 582         if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
 583                 ipv4_hdr = (struct rte_ipv4_hdr *)l3;
 584                 hdr_len = (ipv4_hdr->version_ihl & RTE_IPV4_HDR_IHL_MASK) *
 585                           RTE_IPV4_IHL_MULTIPLIER;
 586                 if (hdr_len == sizeof(struct rte_ipv4_hdr)) {
 587                         packet_type |= RTE_PTYPE_L3_IPV4;
 588                         if (ipv4_hdr->next_proto_id == IPPROTO_TCP)
 589                                 packet_type |= RTE_PTYPE_L4_TCP;
 590                         else if (ipv4_hdr->next_proto_id == IPPROTO_UDP)
 591                                 packet_type |= RTE_PTYPE_L4_UDP;
 592                 } else
 593                         packet_type |= RTE_PTYPE_L3_IPV4_EXT;
 594         } else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
 595                 ipv6_hdr = (struct rte_ipv6_hdr *)l3;
 596                 if (ipv6_hdr->proto == IPPROTO_TCP)
 597                         packet_type |= RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_TCP;
 598                 else if (ipv6_hdr->proto == IPPROTO_UDP)
 599                         packet_type |= RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_UDP;
 600                 else
 601                         packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 602         }
 603
 604         m->packet_type = packet_type;
 605 }
 606
 607 uint16_t
 608 em_cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
 609                   struct rte_mbuf *pkts[], uint16_t nb_pkts,
 610                   uint16_t max_pkts __rte_unused,
 611                   void *user_param __rte_unused)
 612 {
 613         unsigned i;
 614
 615         for (i = 0; i < nb_pkts; ++i)
 616                 em_parse_ptype(pkts[i]);
 617
 618         return nb_pkts;
 619 }
 620
 621 /* main processing loop */
 622 int
 623 em_main_loop(__attribute__((unused)) void *dummy)
 624 {
 625         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 626         unsigned lcore_id;
 627         uint64_t prev_tsc, diff_tsc, cur_tsc;
 628         int i, nb_rx;
 629         uint8_t queueid;
 630         uint16_t portid;
 631         struct lcore_conf *qconf;
 632         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
 633                 US_PER_S * BURST_TX_DRAIN_US;
 634
 635         prev_tsc = 0;
 636
 637         lcore_id = rte_lcore_id();
 638         qconf = &lcore_conf[lcore_id];
 639
 640         if (qconf->n_rx_queue == 0) {
 641                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
 642                 return 0;
 643         }
 644
 645         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
 646
 647         for (i = 0; i < qconf->n_rx_queue; i++) {
 648
 649                 portid = qconf->rx_queue_list[i].port_id;
 650                 queueid = qconf->rx_queue_list[i].queue_id;
 651                 RTE_LOG(INFO, L3FWD,
 652                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
 653                         lcore_id, portid, queueid);
 654         }
 655
 656         while (!force_quit) {
 657
 658                 cur_tsc = rte_rdtsc();
 659
 660                 /*
 661                  * TX burst queue drain
 662                  */
 663                 diff_tsc = cur_tsc - prev_tsc;
 664                 if (unlikely(diff_tsc > drain_tsc)) {
 665
 666                         for (i = 0; i < qconf->n_tx_port; ++i) {
 667                                 portid = qconf->tx_port_id[i];
 668                                 if (qconf->tx_mbufs[portid].len == 0)
 669                                         continue;
 670                                 send_burst(qconf,
 671                                         qconf->tx_mbufs[portid].len,
 672                                         portid);
 673                                 qconf->tx_mbufs[portid].len = 0;
 674                         }
 675
 676                         prev_tsc = cur_tsc;
 677                 }
 678
 679                 /*
 680                  * Read packet from RX queues
 681                  */
 682                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 683                         portid = qconf->rx_queue_list[i].port_id;
 684                         queueid = qconf->rx_queue_list[i].queue_id;
 685                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 686                                 MAX_PKT_BURST);
 687                         if (nb_rx == 0)
 688                                 continue;
 689
 690 #if defined RTE_ARCH_X86 || defined RTE_MACHINE_CPUFLAG_NEON
 691                         l3fwd_em_send_packets(nb_rx, pkts_burst,
 692                                                         portid, qconf);
 693 #else
 694                         l3fwd_em_no_opt_send_packets(nb_rx, pkts_burst,
 695                                                         portid, qconf);
 696 #endif
 697                 }
 698         }
 699
 700         return 0;
 701 }
 702
 703 static __rte_always_inline void
 704 em_event_loop_single(struct l3fwd_event_resources *evt_rsrc,
 705                 const uint8_t flags)
 706 {
 707         const int event_p_id = l3fwd_get_free_event_port(evt_rsrc);
 708         const uint8_t tx_q_id = evt_rsrc->evq.event_q_id[
 709                 evt_rsrc->evq.nb_queues - 1];
 710         const uint8_t event_d_id = evt_rsrc->event_d_id;
 711         struct lcore_conf *lconf;
 712         unsigned int lcore_id;
 713         struct rte_event ev;
 714
 715         if (event_p_id < 0)
 716                 return;
 717
 718         lcore_id = rte_lcore_id();
 719         lconf = &lcore_conf[lcore_id];
 720
 721         RTE_LOG(INFO, L3FWD, "entering %s on lcore %u\n", __func__, lcore_id);
 722         while (!force_quit) {
 723                 if (!rte_event_dequeue_burst(event_d_id, event_p_id, &ev, 1, 0))
 724                         continue;
 725
 726                 struct rte_mbuf *mbuf = ev.mbuf;
 727
 728 #if defined RTE_ARCH_X86 || defined RTE_MACHINE_CPUFLAG_NEON
 729                 mbuf->port = em_get_dst_port(lconf, mbuf, mbuf->port);
 730                 process_packet(mbuf, &mbuf->port);
 731 #else
 732                 l3fwd_em_simple_process(mbuf, lconf);
 733 #endif
 734                 if (mbuf->port == BAD_PORT) {
 735                         rte_pktmbuf_free(mbuf);
 736                         continue;
 737                 }
 738
 739                 if (flags & L3FWD_EVENT_TX_ENQ) {
 740                         ev.queue_id = tx_q_id;
 741                         ev.op = RTE_EVENT_OP_FORWARD;
 742                         while (rte_event_enqueue_burst(event_d_id, event_p_id,
 743                                                 &ev, 1) && !force_quit)
 744                                 ;
 745                 }
 746
 747                 if (flags & L3FWD_EVENT_TX_DIRECT) {
 748                         rte_event_eth_tx_adapter_txq_set(mbuf, 0);
 749                         while (!rte_event_eth_tx_adapter_enqueue(event_d_id,
 750                                                 event_p_id, &ev, 1, 0) &&
 751                                         !force_quit)
 752                                 ;
 753                 }
 754         }
 755 }
 756
 757 static __rte_always_inline void
 758 em_event_loop_burst(struct l3fwd_event_resources *evt_rsrc,
 759                 const uint8_t flags)
 760 {
 761         const int event_p_id = l3fwd_get_free_event_port(evt_rsrc);
 762         const uint8_t tx_q_id = evt_rsrc->evq.event_q_id[
 763                 evt_rsrc->evq.nb_queues - 1];
 764         const uint8_t event_d_id = evt_rsrc->event_d_id;
 765         const uint16_t deq_len = evt_rsrc->deq_depth;
 766         struct rte_event events[MAX_PKT_BURST];
 767         struct lcore_conf *lconf;
 768         unsigned int lcore_id;
 769         int i, nb_enq, nb_deq;
 770
 771         if (event_p_id < 0)
 772                 return;
 773
 774         lcore_id = rte_lcore_id();
 775
 776         lconf = &lcore_conf[lcore_id];
 777
 778         RTE_LOG(INFO, L3FWD, "entering %s on lcore %u\n", __func__, lcore_id);
 779
 780         while (!force_quit) {
 781                 /* Read events from RX queues */
 782                 nb_deq = rte_event_dequeue_burst(event_d_id, event_p_id,
 783                                 events, deq_len, 0);
 784                 if (nb_deq == 0) {
 785                         rte_pause();
 786                         continue;
 787                 }
 788
 789 #if defined RTE_ARCH_X86 || defined RTE_MACHINE_CPUFLAG_NEON
 790                 l3fwd_em_process_events(nb_deq, (struct rte_event **)&events,
 791                                         lconf);
 792 #else
 793                 l3fwd_em_no_opt_process_events(nb_deq,
 794                                                (struct rte_event **)&events,
 795                                                lconf);
 796 #endif
 797                 for (i = 0; i < nb_deq; i++) {
 798                         if (flags & L3FWD_EVENT_TX_ENQ) {
 799                                 events[i].queue_id = tx_q_id;
 800                                 events[i].op = RTE_EVENT_OP_FORWARD;
 801                         }
 802
 803                         if (flags & L3FWD_EVENT_TX_DIRECT)
 804                                 rte_event_eth_tx_adapter_txq_set(events[i].mbuf,
 805                                                                  0);
 806                 }
 807
 808                 if (flags & L3FWD_EVENT_TX_ENQ) {
 809                         nb_enq = rte_event_enqueue_burst(event_d_id, event_p_id,
 810                                         events, nb_deq);
 811                         while (nb_enq < nb_deq && !force_quit)
 812                                 nb_enq += rte_event_enqueue_burst(event_d_id,
 813                                                 event_p_id, events + nb_enq,
 814                                                 nb_deq - nb_enq);
 815                 }
 816
 817                 if (flags & L3FWD_EVENT_TX_DIRECT) {
 818                         nb_enq = rte_event_eth_tx_adapter_enqueue(event_d_id,
 819                                         event_p_id, events, nb_deq, 0);
 820                         while (nb_enq < nb_deq && !force_quit)
 821                                 nb_enq += rte_event_eth_tx_adapter_enqueue(
 822                                                 event_d_id, event_p_id,
 823                                                 events + nb_enq,
 824                                                 nb_deq - nb_enq, 0);
 825                 }
 826         }
 827 }
 828
 829 static __rte_always_inline void
 830 em_event_loop(struct l3fwd_event_resources *evt_rsrc,
 831                  const uint8_t flags)
 832 {
 833         if (flags & L3FWD_EVENT_SINGLE)
 834                 em_event_loop_single(evt_rsrc, flags);
 835         if (flags & L3FWD_EVENT_BURST)
 836                 em_event_loop_burst(evt_rsrc, flags);
 837 }
 838
 839 int __rte_noinline
 840 em_event_main_loop_tx_d(__attribute__((unused)) void *dummy)
 841 {
 842         struct l3fwd_event_resources *evt_rsrc =
 843                                         l3fwd_get_eventdev_rsrc();
 844
 845         em_event_loop(evt_rsrc, L3FWD_EVENT_TX_DIRECT | L3FWD_EVENT_SINGLE);
 846         return 0;
 847 }
 848
 849 int __rte_noinline
 850 em_event_main_loop_tx_d_burst(__attribute__((unused)) void *dummy)
 851 {
 852         struct l3fwd_event_resources *evt_rsrc =
 853                                         l3fwd_get_eventdev_rsrc();
 854
 855         em_event_loop(evt_rsrc, L3FWD_EVENT_TX_DIRECT | L3FWD_EVENT_BURST);
 856         return 0;
 857 }
 858
 859 int __rte_noinline
 860 em_event_main_loop_tx_q(__attribute__((unused)) void *dummy)
 861 {
 862         struct l3fwd_event_resources *evt_rsrc =
 863                                         l3fwd_get_eventdev_rsrc();
 864
 865         em_event_loop(evt_rsrc, L3FWD_EVENT_TX_ENQ | L3FWD_EVENT_SINGLE);
 866         return 0;
 867 }
 868
 869 int __rte_noinline
 870 em_event_main_loop_tx_q_burst(__attribute__((unused)) void *dummy)
 871 {
 872         struct l3fwd_event_resources *evt_rsrc =
 873                                         l3fwd_get_eventdev_rsrc();
 874
 875         em_event_loop(evt_rsrc, L3FWD_EVENT_TX_ENQ | L3FWD_EVENT_BURST);
 876         return 0;
 877 }
 878
 879 /*
 880  * Initialize exact match (hash) parameters.
 881  */
 882 void
 883 setup_hash(const int socketid)
 884 {
 885         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
 886                 .name = NULL,
 887                 .entries = L3FWD_HASH_ENTRIES,
 888                 .key_len = sizeof(union ipv4_5tuple_host),
 889                 .hash_func = ipv4_hash_crc,
 890                 .hash_func_init_val = 0,
 891         };
 892
 893         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
 894                 .name = NULL,
 895                 .entries = L3FWD_HASH_ENTRIES,
 896                 .key_len = sizeof(union ipv6_5tuple_host),
 897                 .hash_func = ipv6_hash_crc,
 898                 .hash_func_init_val = 0,
 899         };
 900
 901         char s[64];
 902
 903         /* create ipv4 hash */
 904         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
 905         ipv4_l3fwd_hash_params.name = s;
 906         ipv4_l3fwd_hash_params.socket_id = socketid;
 907         ipv4_l3fwd_em_lookup_struct[socketid] =
 908                 rte_hash_create(&ipv4_l3fwd_hash_params);
 909         if (ipv4_l3fwd_em_lookup_struct[socketid] == NULL)
 910                 rte_exit(EXIT_FAILURE,
 911                         "Unable to create the l3fwd hash on socket %d\n",
 912                         socketid);
 913
 914         /* create ipv6 hash */
 915         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
 916         ipv6_l3fwd_hash_params.name = s;
 917         ipv6_l3fwd_hash_params.socket_id = socketid;
 918         ipv6_l3fwd_em_lookup_struct[socketid] =
 919                 rte_hash_create(&ipv6_l3fwd_hash_params);
 920         if (ipv6_l3fwd_em_lookup_struct[socketid] == NULL)
 921                 rte_exit(EXIT_FAILURE,
 922                         "Unable to create the l3fwd hash on socket %d\n",
 923                         socketid);
 924
 925         if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
 926                 /* For testing hash matching with a large number of flows we
 927                  * generate millions of IP 5-tuples with an incremented dst
 928                  * address to initialize the hash table. */
 929                 if (ipv6 == 0) {
 930                         /* populate the ipv4 hash */
 931                         populate_ipv4_many_flow_into_table(
 932                                 ipv4_l3fwd_em_lookup_struct[socketid],
 933                                 hash_entry_number);
 934                 } else {
 935                         /* populate the ipv6 hash */
 936                         populate_ipv6_many_flow_into_table(
 937                                 ipv6_l3fwd_em_lookup_struct[socketid],
 938                                 hash_entry_number);
 939                 }
 940         } else {
 941                 /*
 942                  * Use data in ipv4/ipv6 l3fwd lookup table
 943                  * directly to initialize the hash table.
 944                  */
 945                 if (ipv6 == 0) {
 946                         /* populate the ipv4 hash */
 947                         populate_ipv4_few_flow_into_table(
 948                                 ipv4_l3fwd_em_lookup_struct[socketid]);
 949                 } else {
 950                         /* populate the ipv6 hash */
 951                         populate_ipv6_few_flow_into_table(
 952                                 ipv6_l3fwd_em_lookup_struct[socketid]);
 953                 }
 954         }
 955 }
 956
 957 /* Return ipv4/ipv6 em fwd lookup struct. */
 958 void *
 959 em_get_ipv4_l3fwd_lookup_struct(const int socketid)
 960 {
 961         return ipv4_l3fwd_em_lookup_struct[socketid];
 962 }
 963
 964 void *
 965 em_get_ipv6_l3fwd_lookup_struct(const int socketid)
 966 {
 967         return ipv6_l3fwd_em_lookup_struct[socketid];
 968 }