examples/l3fwd/l3fwd_em.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <stdio.h>
  35 #include <stdlib.h>
  36 #include <stdint.h>
  37 #include <inttypes.h>
  38 #include <sys/types.h>
  39 #include <string.h>
  40 #include <sys/queue.h>
  41 #include <stdarg.h>
  42 #include <errno.h>
  43 #include <getopt.h>
  44 #include <stdbool.h>
  45
  46 #include <rte_debug.h>
  47 #include <rte_ether.h>
  48 #include <rte_ethdev.h>
  49 #include <rte_ring.h>
  50 #include <rte_mempool.h>
  51 #include <rte_cycles.h>
  52 #include <rte_mbuf.h>
  53 #include <rte_ip.h>
  54 #include <rte_tcp.h>
  55 #include <rte_udp.h>
  56 #include <rte_hash.h>
  57
  58 #include "l3fwd.h"
  59
  60 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
  61 #include <rte_hash_crc.h>
  62 #define DEFAULT_HASH_FUNC       rte_hash_crc
  63 #else
  64 #include <rte_jhash.h>
  65 #define DEFAULT_HASH_FUNC       rte_jhash
  66 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
  67
  68 #define IPV6_ADDR_LEN 16
  69
  70 struct ipv4_5tuple {
  71         uint32_t ip_dst;
  72         uint32_t ip_src;
  73         uint16_t port_dst;
  74         uint16_t port_src;
  75         uint8_t  proto;
  76 } __attribute__((__packed__));
  77
  78 union ipv4_5tuple_host {
  79         struct {
  80                 uint8_t  pad0;
  81                 uint8_t  proto;
  82                 uint16_t pad1;
  83                 uint32_t ip_src;
  84                 uint32_t ip_dst;
  85                 uint16_t port_src;
  86                 uint16_t port_dst;
  87         };
  88         __m128i xmm;
  89 };
  90
  91 #define XMM_NUM_IN_IPV6_5TUPLE 3
  92
  93 struct ipv6_5tuple {
  94         uint8_t  ip_dst[IPV6_ADDR_LEN];
  95         uint8_t  ip_src[IPV6_ADDR_LEN];
  96         uint16_t port_dst;
  97         uint16_t port_src;
  98         uint8_t  proto;
  99 } __attribute__((__packed__));
 100
 101 union ipv6_5tuple_host {
 102         struct {
 103                 uint16_t pad0;
 104                 uint8_t  proto;
 105                 uint8_t  pad1;
 106                 uint8_t  ip_src[IPV6_ADDR_LEN];
 107                 uint8_t  ip_dst[IPV6_ADDR_LEN];
 108                 uint16_t port_src;
 109                 uint16_t port_dst;
 110                 uint64_t reserve;
 111         };
 112         __m128i xmm[XMM_NUM_IN_IPV6_5TUPLE];
 113 };
 114
 115 struct ipv4_l3fwd_em_route {
 116         struct ipv4_5tuple key;
 117         uint8_t if_out;
 118 };
 119
 120 struct ipv6_l3fwd_em_route {
 121         struct ipv6_5tuple key;
 122         uint8_t if_out;
 123 };
 124
 125 static struct ipv4_l3fwd_em_route ipv4_l3fwd_em_route_array[] = {
 126         {{IPv4(101, 0, 0, 0), IPv4(100, 10, 0, 1),  101, 11, IPPROTO_TCP}, 0},
 127         {{IPv4(201, 0, 0, 0), IPv4(200, 20, 0, 1),  102, 12, IPPROTO_TCP}, 1},
 128         {{IPv4(111, 0, 0, 0), IPv4(100, 30, 0, 1),  101, 11, IPPROTO_TCP}, 2},
 129         {{IPv4(211, 0, 0, 0), IPv4(200, 40, 0, 1),  102, 12, IPPROTO_TCP}, 3},
 130 };
 131
 132 static struct ipv6_l3fwd_em_route ipv6_l3fwd_em_route_array[] = {
 133         {{
 134         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 135         {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 136         101, 11, IPPROTO_TCP}, 0},
 137
 138         {{
 139         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 140         {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 141         102, 12, IPPROTO_TCP}, 1},
 142
 143         {{
 144         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 145         {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 146         101, 11, IPPROTO_TCP}, 2},
 147
 148         {{
 149         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0},
 150         {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 151         102, 12, IPPROTO_TCP}, 3},
 152 };
 153
 154 struct rte_hash *ipv4_l3fwd_em_lookup_struct[NB_SOCKETS];
 155 struct rte_hash *ipv6_l3fwd_em_lookup_struct[NB_SOCKETS];
 156
 157 static inline uint32_t
 158 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len,
 159                 uint32_t init_val)
 160 {
 161         const union ipv4_5tuple_host *k;
 162         uint32_t t;
 163         const uint32_t *p;
 164
 165         k = data;
 166         t = k->proto;
 167         p = (const uint32_t *)&k->port_src;
 168
 169 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 170         init_val = rte_hash_crc_4byte(t, init_val);
 171         init_val = rte_hash_crc_4byte(k->ip_src, init_val);
 172         init_val = rte_hash_crc_4byte(k->ip_dst, init_val);
 173         init_val = rte_hash_crc_4byte(*p, init_val);
 174 #else /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 175         init_val = rte_jhash_1word(t, init_val);
 176         init_val = rte_jhash_1word(k->ip_src, init_val);
 177         init_val = rte_jhash_1word(k->ip_dst, init_val);
 178         init_val = rte_jhash_1word(*p, init_val);
 179 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 180
 181         return init_val;
 182 }
 183
 184 static inline uint32_t
 185 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len,
 186                 uint32_t init_val)
 187 {
 188         const union ipv6_5tuple_host *k;
 189         uint32_t t;
 190         const uint32_t *p;
 191 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 192         const uint32_t  *ip_src0, *ip_src1, *ip_src2, *ip_src3;
 193         const uint32_t  *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3;
 194 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 195
 196         k = data;
 197         t = k->proto;
 198         p = (const uint32_t *)&k->port_src;
 199
 200 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 201         ip_src0 = (const uint32_t *) k->ip_src;
 202         ip_src1 = (const uint32_t *)(k->ip_src+4);
 203         ip_src2 = (const uint32_t *)(k->ip_src+8);
 204         ip_src3 = (const uint32_t *)(k->ip_src+12);
 205         ip_dst0 = (const uint32_t *) k->ip_dst;
 206         ip_dst1 = (const uint32_t *)(k->ip_dst+4);
 207         ip_dst2 = (const uint32_t *)(k->ip_dst+8);
 208         ip_dst3 = (const uint32_t *)(k->ip_dst+12);
 209         init_val = rte_hash_crc_4byte(t, init_val);
 210         init_val = rte_hash_crc_4byte(*ip_src0, init_val);
 211         init_val = rte_hash_crc_4byte(*ip_src1, init_val);
 212         init_val = rte_hash_crc_4byte(*ip_src2, init_val);
 213         init_val = rte_hash_crc_4byte(*ip_src3, init_val);
 214         init_val = rte_hash_crc_4byte(*ip_dst0, init_val);
 215         init_val = rte_hash_crc_4byte(*ip_dst1, init_val);
 216         init_val = rte_hash_crc_4byte(*ip_dst2, init_val);
 217         init_val = rte_hash_crc_4byte(*ip_dst3, init_val);
 218         init_val = rte_hash_crc_4byte(*p, init_val);
 219 #else /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 220         init_val = rte_jhash_1word(t, init_val);
 221         init_val = rte_jhash(k->ip_src,
 222                         sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
 223         init_val = rte_jhash(k->ip_dst,
 224                         sizeof(uint8_t) * IPV6_ADDR_LEN, init_val);
 225         init_val = rte_jhash_1word(*p, init_val);
 226 #endif /* RTE_MACHINE_CPUFLAG_SSE4_2 */
 227         return init_val;
 228 }
 229
 230 #define IPV4_L3FWD_EM_NUM_ROUTES \
 231         (sizeof(ipv4_l3fwd_em_route_array) / sizeof(ipv4_l3fwd_em_route_array[0]))
 232
 233 #define IPV6_L3FWD_EM_NUM_ROUTES \
 234         (sizeof(ipv6_l3fwd_em_route_array) / sizeof(ipv6_l3fwd_em_route_array[0]))
 235
 236 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 237 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 238
 239 static __m128i mask0;
 240 static __m128i mask1;
 241 static __m128i mask2;
 242
 243 static inline uint8_t
 244 em_get_ipv4_dst_port(void *ipv4_hdr, uint8_t portid, void *lookup_struct)
 245 {
 246         int ret = 0;
 247         union ipv4_5tuple_host key;
 248         struct rte_hash *ipv4_l3fwd_lookup_struct =
 249                 (struct rte_hash *)lookup_struct;
 250
 251         ipv4_hdr = (uint8_t *)ipv4_hdr + offsetof(struct ipv4_hdr, time_to_live);
 252         __m128i data = _mm_loadu_si128((__m128i *)(ipv4_hdr));
 253
 254         /*
 255          * Get 5 tuple: dst port, src port, dst IP address,
 256          * src IP address and protocol.
 257          */
 258         key.xmm = _mm_and_si128(data, mask0);
 259
 260         /* Find destination port */
 261         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 262         return (uint8_t)((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
 263 }
 264
 265 static inline uint8_t
 266 em_get_ipv6_dst_port(void *ipv6_hdr,  uint8_t portid, void *lookup_struct)
 267 {
 268         int ret = 0;
 269         union ipv6_5tuple_host key;
 270         struct rte_hash *ipv6_l3fwd_lookup_struct =
 271                 (struct rte_hash *)lookup_struct;
 272
 273         ipv6_hdr = (uint8_t *)ipv6_hdr + offsetof(struct ipv6_hdr, payload_len);
 274         __m128i data0 =
 275                 _mm_loadu_si128((__m128i *)(ipv6_hdr));
 276         __m128i data1 =
 277                 _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr)+
 278                                         sizeof(__m128i)));
 279         __m128i data2 =
 280                 _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr)+
 281                                         sizeof(__m128i)+sizeof(__m128i)));
 282
 283         /* Get part of 5 tuple: src IP address lower 96 bits and protocol */
 284         key.xmm[0] = _mm_and_si128(data0, mask1);
 285
 286         /*
 287          * Get part of 5 tuple: dst IP address lower 96 bits
 288          * and src IP address higher 32 bits.
 289          */
 290         key.xmm[1] = data1;
 291
 292         /*
 293          * Get part of 5 tuple: dst port and src port
 294          * and dst IP address higher 32 bits.
 295          */
 296         key.xmm[2] = _mm_and_si128(data2, mask2);
 297
 298         /* Find destination port */
 299         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 300         return (uint8_t)((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
 301 }
 302
 303 static inline __attribute__((always_inline)) void
 304 l3fwd_em_simple_forward(struct rte_mbuf *m, uint8_t portid,
 305                 struct lcore_conf *qconf)
 306 {
 307         struct ether_hdr *eth_hdr;
 308         struct ipv4_hdr *ipv4_hdr;
 309         uint8_t dst_port;
 310
 311         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 312
 313         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 314                 /* Handle IPv4 headers.*/
 315                 ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
 316                                                    sizeof(struct ether_hdr));
 317
 318 #ifdef DO_RFC_1812_CHECKS
 319                 /* Check to make sure the packet is valid (RFC1812) */
 320                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 321                         rte_pktmbuf_free(m);
 322                         return;
 323                 }
 324 #endif
 325                  dst_port = em_get_ipv4_dst_port(ipv4_hdr, portid,
 326                                                 qconf->ipv4_lookup_struct);
 327
 328                 if (dst_port >= RTE_MAX_ETHPORTS ||
 329                         (enabled_port_mask & 1 << dst_port) == 0)
 330                         dst_port = portid;
 331
 332 #ifdef DO_RFC_1812_CHECKS
 333                 /* Update time to live and header checksum */
 334                 --(ipv4_hdr->time_to_live);
 335                 ++(ipv4_hdr->hdr_checksum);
 336 #endif
 337                 /* dst addr */
 338                 *(uint64_t *)&eth_hdr->d_addr = dest_eth_addr[dst_port];
 339
 340                 /* src addr */
 341                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 342
 343                 send_single_packet(qconf, m, dst_port);
 344         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 345                 /* Handle IPv6 headers.*/
 346                 struct ipv6_hdr *ipv6_hdr;
 347
 348                 ipv6_hdr = rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
 349                                                    sizeof(struct ether_hdr));
 350
 351                 dst_port = em_get_ipv6_dst_port(ipv6_hdr, portid,
 352                                         qconf->ipv6_lookup_struct);
 353
 354                 if (dst_port >= RTE_MAX_ETHPORTS ||
 355                         (enabled_port_mask & 1 << dst_port) == 0)
 356                         dst_port = portid;
 357
 358                 /* dst addr */
 359                 *(uint64_t *)&eth_hdr->d_addr = dest_eth_addr[dst_port];
 360
 361                 /* src addr */
 362                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 363
 364                 send_single_packet(qconf, m, dst_port);
 365         } else {
 366                 /* Free the mbuf that contains non-IPV4/IPV6 packet */
 367                 rte_pktmbuf_free(m);
 368         }
 369 }
 370
 371 /*
 372  * Include header file if SSE4_1 is enabled for
 373  * buffer optimization i.e. ENABLE_MULTI_BUFFER_OPTIMIZE=1.
 374  */
 375 #if defined(__SSE4_1__)
 376 #include "l3fwd_em_sse.h"
 377 #else
 378 #include "l3fwd_em.h"
 379 #endif
 380
 381 static void
 382 convert_ipv4_5tuple(struct ipv4_5tuple *key1,
 383                 union ipv4_5tuple_host *key2)
 384 {
 385         key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst);
 386         key2->ip_src = rte_cpu_to_be_32(key1->ip_src);
 387         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
 388         key2->port_src = rte_cpu_to_be_16(key1->port_src);
 389         key2->proto = key1->proto;
 390         key2->pad0 = 0;
 391         key2->pad1 = 0;
 392 }
 393
 394 static void
 395 convert_ipv6_5tuple(struct ipv6_5tuple *key1,
 396                 union ipv6_5tuple_host *key2)
 397 {
 398         uint32_t i;
 399
 400         for (i = 0; i < 16; i++) {
 401                 key2->ip_dst[i] = key1->ip_dst[i];
 402                 key2->ip_src[i] = key1->ip_src[i];
 403         }
 404         key2->port_dst = rte_cpu_to_be_16(key1->port_dst);
 405         key2->port_src = rte_cpu_to_be_16(key1->port_src);
 406         key2->proto = key1->proto;
 407         key2->pad0 = 0;
 408         key2->pad1 = 0;
 409         key2->reserve = 0;
 410 }
 411
 412 #define BYTE_VALUE_MAX 256
 413 #define ALL_32_BITS 0xffffffff
 414 #define BIT_8_TO_15 0x0000ff00
 415 static inline void
 416 populate_ipv4_few_flow_into_table(const struct rte_hash *h)
 417 {
 418         uint32_t i;
 419         int32_t ret;
 420
 421         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS,
 422                                 ALL_32_BITS, BIT_8_TO_15);
 423
 424         for (i = 0; i < IPV4_L3FWD_EM_NUM_ROUTES; i++) {
 425                 struct ipv4_l3fwd_em_route  entry;
 426                 union ipv4_5tuple_host newkey;
 427
 428                 entry = ipv4_l3fwd_em_route_array[i];
 429                 convert_ipv4_5tuple(&entry.key, &newkey);
 430                 ret = rte_hash_add_key(h, (void *) &newkey);
 431                 if (ret < 0) {
 432                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
 433                                 " to the l3fwd hash.\n", i);
 434                 }
 435                 ipv4_l3fwd_out_if[ret] = entry.if_out;
 436         }
 437         printf("Hash: Adding 0x%" PRIx64 " keys\n",
 438                 (uint64_t)IPV4_L3FWD_EM_NUM_ROUTES);
 439 }
 440
 441 #define BIT_16_TO_23 0x00ff0000
 442 static inline void
 443 populate_ipv6_few_flow_into_table(const struct rte_hash *h)
 444 {
 445         uint32_t i;
 446         int32_t ret;
 447
 448         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS,
 449                                 ALL_32_BITS, BIT_16_TO_23);
 450
 451         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
 452
 453         for (i = 0; i < IPV6_L3FWD_EM_NUM_ROUTES; i++) {
 454                 struct ipv6_l3fwd_em_route entry;
 455                 union ipv6_5tuple_host newkey;
 456
 457                 entry = ipv6_l3fwd_em_route_array[i];
 458                 convert_ipv6_5tuple(&entry.key, &newkey);
 459                 ret = rte_hash_add_key(h, (void *) &newkey);
 460                 if (ret < 0) {
 461                         rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32
 462                                 " to the l3fwd hash.\n", i);
 463                 }
 464                 ipv6_l3fwd_out_if[ret] = entry.if_out;
 465         }
 466         printf("Hash: Adding 0x%" PRIx64 "keys\n",
 467                 (uint64_t)IPV6_L3FWD_EM_NUM_ROUTES);
 468 }
 469
 470 #define NUMBER_PORT_USED 4
 471 static inline void
 472 populate_ipv4_many_flow_into_table(const struct rte_hash *h,
 473                 unsigned int nr_flow)
 474 {
 475         unsigned i;
 476
 477         mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS,
 478                                 ALL_32_BITS, BIT_8_TO_15);
 479
 480         for (i = 0; i < nr_flow; i++) {
 481                 struct ipv4_l3fwd_em_route entry;
 482                 union ipv4_5tuple_host newkey;
 483
 484                 uint8_t a = (uint8_t)
 485                         ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
 486                 uint8_t b = (uint8_t)
 487                         (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
 488                 uint8_t c = (uint8_t)
 489                         ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
 490
 491                 /* Create the ipv4 exact match flow */
 492                 memset(&entry, 0, sizeof(entry));
 493                 switch (i & (NUMBER_PORT_USED - 1)) {
 494                 case 0:
 495                         entry = ipv4_l3fwd_em_route_array[0];
 496                         entry.key.ip_dst = IPv4(101, c, b, a);
 497                         break;
 498                 case 1:
 499                         entry = ipv4_l3fwd_em_route_array[1];
 500                         entry.key.ip_dst = IPv4(201, c, b, a);
 501                         break;
 502                 case 2:
 503                         entry = ipv4_l3fwd_em_route_array[2];
 504                         entry.key.ip_dst = IPv4(111, c, b, a);
 505                         break;
 506                 case 3:
 507                         entry = ipv4_l3fwd_em_route_array[3];
 508                         entry.key.ip_dst = IPv4(211, c, b, a);
 509                         break;
 510                 };
 511                 convert_ipv4_5tuple(&entry.key, &newkey);
 512                 int32_t ret = rte_hash_add_key(h, (void *) &newkey);
 513
 514                 if (ret < 0)
 515                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
 516
 517                 ipv4_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
 518
 519         }
 520         printf("Hash: Adding 0x%x keys\n", nr_flow);
 521 }
 522
 523 static inline void
 524 populate_ipv6_many_flow_into_table(const struct rte_hash *h,
 525                 unsigned int nr_flow)
 526 {
 527         unsigned i;
 528
 529         mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS,
 530                                 ALL_32_BITS, BIT_16_TO_23);
 531         mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS);
 532
 533         for (i = 0; i < nr_flow; i++) {
 534                 struct ipv6_l3fwd_em_route entry;
 535                 union ipv6_5tuple_host newkey;
 536
 537                 uint8_t a = (uint8_t)
 538                         ((i/NUMBER_PORT_USED)%BYTE_VALUE_MAX);
 539                 uint8_t b = (uint8_t)
 540                         (((i/NUMBER_PORT_USED)/BYTE_VALUE_MAX)%BYTE_VALUE_MAX);
 541                 uint8_t c = (uint8_t)
 542                         ((i/NUMBER_PORT_USED)/(BYTE_VALUE_MAX*BYTE_VALUE_MAX));
 543
 544                 /* Create the ipv6 exact match flow */
 545                 memset(&entry, 0, sizeof(entry));
 546                 switch (i & (NUMBER_PORT_USED - 1)) {
 547                 case 0:
 548                         entry = ipv6_l3fwd_em_route_array[0];
 549                         break;
 550                 case 1:
 551                         entry = ipv6_l3fwd_em_route_array[1];
 552                         break;
 553                 case 2:
 554                         entry = ipv6_l3fwd_em_route_array[2];
 555                         break;
 556                 case 3:
 557                         entry = ipv6_l3fwd_em_route_array[3];
 558                         break;
 559                 };
 560                 entry.key.ip_dst[13] = c;
 561                 entry.key.ip_dst[14] = b;
 562                 entry.key.ip_dst[15] = a;
 563                 convert_ipv6_5tuple(&entry.key, &newkey);
 564                 int32_t ret = rte_hash_add_key(h, (void *) &newkey);
 565
 566                 if (ret < 0)
 567                         rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i);
 568
 569                 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out;
 570
 571         }
 572         printf("Hash: Adding 0x%x keys\n", nr_flow);
 573 }
 574
 575 int
 576 em_main_loop(__attribute__((unused)) void *dummy)
 577 {
 578         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 579         unsigned lcore_id;
 580         uint64_t prev_tsc, diff_tsc, cur_tsc;
 581         int i, nb_rx;
 582         uint8_t portid, queueid;
 583         struct lcore_conf *qconf;
 584         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
 585                 US_PER_S * BURST_TX_DRAIN_US;
 586
 587         prev_tsc = 0;
 588
 589         lcore_id = rte_lcore_id();
 590         qconf = &lcore_conf[lcore_id];
 591
 592         if (qconf->n_rx_queue == 0) {
 593                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
 594                 return 0;
 595         }
 596
 597         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
 598
 599         for (i = 0; i < qconf->n_rx_queue; i++) {
 600
 601                 portid = qconf->rx_queue_list[i].port_id;
 602                 queueid = qconf->rx_queue_list[i].queue_id;
 603                 RTE_LOG(INFO, L3FWD,
 604                         " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n",
 605                         lcore_id, portid, queueid);
 606         }
 607
 608         while (!force_quit) {
 609
 610                 cur_tsc = rte_rdtsc();
 611
 612                 /*
 613                  * TX burst queue drain
 614                  */
 615                 diff_tsc = cur_tsc - prev_tsc;
 616                 if (unlikely(diff_tsc > drain_tsc)) {
 617
 618                         /*
 619                          * This could be optimized (use queueid instead of
 620                          * portid), but it is not called so often
 621                          */
 622                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 623                                 if (qconf->tx_mbufs[portid].len == 0)
 624                                         continue;
 625                                 send_burst(qconf,
 626                                         qconf->tx_mbufs[portid].len,
 627                                         portid);
 628                                 qconf->tx_mbufs[portid].len = 0;
 629                         }
 630
 631                         prev_tsc = cur_tsc;
 632                 }
 633
 634                 /*
 635                  * Read packet from RX queues
 636                  */
 637                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 638                         portid = qconf->rx_queue_list[i].port_id;
 639                         queueid = qconf->rx_queue_list[i].queue_id;
 640                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 641                                 MAX_PKT_BURST);
 642                         if (nb_rx == 0)
 643                                 continue;
 644
 645                         /*
 646                          * For SSE4_1 use ENABLE_MULTI_BUFFER_OPTIMIZE=1
 647                          * code.
 648                          */
 649 #if defined(__SSE4_1__)
 650                         l3fwd_em_send_packets(nb_rx, pkts_burst,
 651                                                         portid, qconf);
 652 #else
 653                         l3fwd_em_no_opt_send_packets(nb_rx, pkts_burst,
 654                                                         portid, qconf);
 655 #endif /* __SSE_4_1__ */
 656                 }
 657         }
 658
 659         return 0;
 660 }
 661
 662 /*
 663  * Initialize exact match (hash) parameters.
 664  */
 665 void
 666 setup_hash(const int socketid)
 667 {
 668         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
 669                 .name = NULL,
 670                 .entries = L3FWD_HASH_ENTRIES,
 671                 .key_len = sizeof(union ipv4_5tuple_host),
 672                 .hash_func = ipv4_hash_crc,
 673                 .hash_func_init_val = 0,
 674         };
 675
 676         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
 677                 .name = NULL,
 678                 .entries = L3FWD_HASH_ENTRIES,
 679                 .key_len = sizeof(union ipv6_5tuple_host),
 680                 .hash_func = ipv6_hash_crc,
 681                 .hash_func_init_val = 0,
 682         };
 683
 684         char s[64];
 685
 686         /* create ipv4 hash */
 687         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
 688         ipv4_l3fwd_hash_params.name = s;
 689         ipv4_l3fwd_hash_params.socket_id = socketid;
 690         ipv4_l3fwd_em_lookup_struct[socketid] =
 691                 rte_hash_create(&ipv4_l3fwd_hash_params);
 692         if (ipv4_l3fwd_em_lookup_struct[socketid] == NULL)
 693                 rte_exit(EXIT_FAILURE,
 694                         "Unable to create the l3fwd hash on socket %d\n",
 695                         socketid);
 696
 697         /* create ipv6 hash */
 698         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
 699         ipv6_l3fwd_hash_params.name = s;
 700         ipv6_l3fwd_hash_params.socket_id = socketid;
 701         ipv6_l3fwd_em_lookup_struct[socketid] =
 702                 rte_hash_create(&ipv6_l3fwd_hash_params);
 703         if (ipv6_l3fwd_em_lookup_struct[socketid] == NULL)
 704                 rte_exit(EXIT_FAILURE,
 705                         "Unable to create the l3fwd hash on socket %d\n",
 706                         socketid);
 707
 708         if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) {
 709                 /* For testing hash matching with a large number of flows we
 710                  * generate millions of IP 5-tuples with an incremented dst
 711                  * address to initialize the hash table. */
 712                 if (ipv6 == 0) {
 713                         /* populate the ipv4 hash */
 714                         populate_ipv4_many_flow_into_table(
 715                                 ipv4_l3fwd_em_lookup_struct[socketid],
 716                                 hash_entry_number);
 717                 } else {
 718                         /* populate the ipv6 hash */
 719                         populate_ipv6_many_flow_into_table(
 720                                 ipv6_l3fwd_em_lookup_struct[socketid],
 721                                 hash_entry_number);
 722                 }
 723         } else {
 724                 /*
 725                  * Use data in ipv4/ipv6 l3fwd lookup table
 726                  * directly to initialize the hash table.
 727                  */
 728                 if (ipv6 == 0) {
 729                         /* populate the ipv4 hash */
 730                         populate_ipv4_few_flow_into_table(
 731                                 ipv4_l3fwd_em_lookup_struct[socketid]);
 732                 } else {
 733                         /* populate the ipv6 hash */
 734                         populate_ipv6_few_flow_into_table(
 735                                 ipv6_l3fwd_em_lookup_struct[socketid]);
 736                 }
 737         }
 738 }
 739
 740 /* Return ipv4/ipv6 em fwd lookup struct. */
 741 void *
 742 em_get_ipv4_l3fwd_lookup_struct(const int socketid)
 743 {
 744         return ipv4_l3fwd_em_lookup_struct[socketid];
 745 }
 746
 747 void *
 748 em_get_ipv6_l3fwd_lookup_struct(const int socketid)
 749 {
 750         return ipv6_l3fwd_em_lookup_struct[socketid];
 751 }