examples/ipsec-secgw/ipsec-secgw.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 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
  45 #include <rte_common.h>
  46 #include <rte_byteorder.h>
  47 #include <rte_log.h>
  48 #include <rte_eal.h>
  49 #include <rte_launch.h>
  50 #include <rte_atomic.h>
  51 #include <rte_cycles.h>
  52 #include <rte_prefetch.h>
  53 #include <rte_lcore.h>
  54 #include <rte_per_lcore.h>
  55 #include <rte_branch_prediction.h>
  56 #include <rte_interrupts.h>
  57 #include <rte_pci.h>
  58 #include <rte_random.h>
  59 #include <rte_debug.h>
  60 #include <rte_ether.h>
  61 #include <rte_ethdev.h>
  62 #include <rte_mempool.h>
  63 #include <rte_mbuf.h>
  64 #include <rte_acl.h>
  65 #include <rte_lpm.h>
  66 #include <rte_hash.h>
  67 #include <rte_jhash.h>
  68 #include <rte_cryptodev.h>
  69
  70 #include "ipsec.h"
  71
  72 #define RTE_LOGTYPE_IPSEC RTE_LOGTYPE_USER1
  73
  74 #define MAX_JUMBO_PKT_LEN  9600
  75
  76 #define MEMPOOL_CACHE_SIZE 256
  77
  78 #define NB_MBUF (32000)
  79
  80 #define CDEV_MAP_ENTRIES 1024
  81 #define CDEV_MP_NB_OBJS 2048
  82 #define CDEV_MP_CACHE_SZ 64
  83 #define MAX_QUEUE_PAIRS 1
  84
  85 #define OPTION_CONFIG           "config"
  86 #define OPTION_SINGLE_SA        "single-sa"
  87 #define OPTION_EP0              "ep0"
  88 #define OPTION_EP1              "ep1"
  89
  90 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
  91
  92 #define NB_SOCKETS 4
  93
  94 /* Configure how many packets ahead to prefetch, when reading packets */
  95 #define PREFETCH_OFFSET 3
  96
  97 #define MAX_RX_QUEUE_PER_LCORE 16
  98
  99 #define MAX_LCORE_PARAMS 1024
 100
 101 #define UNPROTECTED_PORT(port) (unprotected_port_mask & (1 << portid))
 102
 103 /*
 104  * Configurable number of RX/TX ring descriptors
 105  */
 106 #define IPSEC_SECGW_RX_DESC_DEFAULT 128
 107 #define IPSEC_SECGW_TX_DESC_DEFAULT 512
 108 static uint16_t nb_rxd = IPSEC_SECGW_RX_DESC_DEFAULT;
 109 static uint16_t nb_txd = IPSEC_SECGW_TX_DESC_DEFAULT;
 110
 111 #if RTE_BYTE_ORDER != RTE_LITTLE_ENDIAN
 112 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
 113         (((uint64_t)((a) & 0xff) << 56) | \
 114         ((uint64_t)((b) & 0xff) << 48) | \
 115         ((uint64_t)((c) & 0xff) << 40) | \
 116         ((uint64_t)((d) & 0xff) << 32) | \
 117         ((uint64_t)((e) & 0xff) << 24) | \
 118         ((uint64_t)((f) & 0xff) << 16) | \
 119         ((uint64_t)((g) & 0xff) << 8)  | \
 120         ((uint64_t)(h) & 0xff))
 121 #else
 122 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
 123         (((uint64_t)((h) & 0xff) << 56) | \
 124         ((uint64_t)((g) & 0xff) << 48) | \
 125         ((uint64_t)((f) & 0xff) << 40) | \
 126         ((uint64_t)((e) & 0xff) << 32) | \
 127         ((uint64_t)((d) & 0xff) << 24) | \
 128         ((uint64_t)((c) & 0xff) << 16) | \
 129         ((uint64_t)((b) & 0xff) << 8) | \
 130         ((uint64_t)(a) & 0xff))
 131 #endif
 132 #define ETHADDR(a, b, c, d, e, f) (__BYTES_TO_UINT64(a, b, c, d, e, f, 0, 0))
 133
 134 #define ETHADDR_TO_UINT64(addr) __BYTES_TO_UINT64( \
 135                 addr.addr_bytes[0], addr.addr_bytes[1], \
 136                 addr.addr_bytes[2], addr.addr_bytes[3], \
 137                 addr.addr_bytes[4], addr.addr_bytes[5], \
 138                 0, 0)
 139
 140 /* port/source ethernet addr and destination ethernet addr */
 141 struct ethaddr_info {
 142         uint64_t src, dst;
 143 };
 144
 145 struct ethaddr_info ethaddr_tbl[RTE_MAX_ETHPORTS] = {
 146         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x7e, 0x94, 0x9a) },
 147         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x22, 0xa1, 0xd9) },
 148         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x08, 0x69, 0x26) },
 149         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x49, 0x9e, 0xdd) }
 150 };
 151
 152 /* mask of enabled ports */
 153 static uint32_t enabled_port_mask;
 154 static uint32_t unprotected_port_mask;
 155 static int32_t promiscuous_on = 1;
 156 static int32_t numa_on = 1; /**< NUMA is enabled by default. */
 157 static int32_t ep = -1; /**< Endpoint configuration (0 or 1) */
 158 static uint32_t nb_lcores;
 159 static uint32_t single_sa;
 160 static uint32_t single_sa_idx;
 161
 162 struct lcore_rx_queue {
 163         uint8_t port_id;
 164         uint8_t queue_id;
 165 } __rte_cache_aligned;
 166
 167 struct lcore_params {
 168         uint8_t port_id;
 169         uint8_t queue_id;
 170         uint8_t lcore_id;
 171 } __rte_cache_aligned;
 172
 173 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 174
 175 static struct lcore_params *lcore_params;
 176 static uint16_t nb_lcore_params;
 177
 178 static struct rte_hash *cdev_map_in;
 179 static struct rte_hash *cdev_map_out;
 180
 181 struct buffer {
 182         uint16_t len;
 183         struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
 184 };
 185
 186 struct lcore_conf {
 187         uint16_t nb_rx_queue;
 188         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 189         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 190         struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
 191         struct ipsec_ctx inbound;
 192         struct ipsec_ctx outbound;
 193         struct rt_ctx *rt_ctx;
 194 } __rte_cache_aligned;
 195
 196 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 197
 198 static struct rte_eth_conf port_conf = {
 199         .rxmode = {
 200                 .mq_mode        = ETH_MQ_RX_RSS,
 201                 .max_rx_pkt_len = ETHER_MAX_LEN,
 202                 .split_hdr_size = 0,
 203                 .header_split   = 0, /**< Header Split disabled */
 204                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 205                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 206                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 207                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 208         },
 209         .rx_adv_conf = {
 210                 .rss_conf = {
 211                         .rss_key = NULL,
 212                         .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
 213                                 ETH_RSS_TCP | ETH_RSS_SCTP,
 214                 },
 215         },
 216         .txmode = {
 217                 .mq_mode = ETH_MQ_TX_NONE,
 218         },
 219 };
 220
 221 static struct socket_ctx socket_ctx[NB_SOCKETS];
 222
 223 struct traffic_type {
 224         const uint8_t *data[MAX_PKT_BURST * 2];
 225         struct rte_mbuf *pkts[MAX_PKT_BURST * 2];
 226         uint32_t res[MAX_PKT_BURST * 2];
 227         uint32_t num;
 228 };
 229
 230 struct ipsec_traffic {
 231         struct traffic_type ipsec4;
 232         struct traffic_type ipv4;
 233 };
 234
 235 static inline void
 236 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
 237 {
 238         uint8_t *nlp;
 239
 240         if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
 241                 rte_pktmbuf_adj(pkt, ETHER_HDR_LEN);
 242                 nlp = rte_pktmbuf_mtod_offset(pkt, uint8_t *,
 243                                 offsetof(struct ip, ip_p));
 244                 if (*nlp == IPPROTO_ESP)
 245                         t->ipsec4.pkts[(t->ipsec4.num)++] = pkt;
 246                 else {
 247                         t->ipv4.data[t->ipv4.num] = nlp;
 248                         t->ipv4.pkts[(t->ipv4.num)++] = pkt;
 249                 }
 250         } else {
 251                 /* Unknown/Unsupported type, drop the packet */
 252                 rte_pktmbuf_free(pkt);
 253         }
 254 }
 255
 256 static inline void
 257 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
 258                 uint16_t nb_pkts)
 259 {
 260         int32_t i;
 261
 262         t->ipsec4.num = 0;
 263         t->ipv4.num = 0;
 264
 265         for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
 266                 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
 267                                         void *));
 268                 prepare_one_packet(pkts[i], t);
 269         }
 270         /* Process left packets */
 271         for (; i < nb_pkts; i++)
 272                 prepare_one_packet(pkts[i], t);
 273 }
 274
 275 static inline void
 276 prepare_tx_pkt(struct rte_mbuf *pkt, uint8_t port)
 277 {
 278         pkt->ol_flags |= PKT_TX_IP_CKSUM | PKT_TX_IPV4;
 279         pkt->l3_len = sizeof(struct ip);
 280         pkt->l2_len = ETHER_HDR_LEN;
 281
 282         struct ether_hdr *ethhdr = (struct ether_hdr *)rte_pktmbuf_prepend(pkt,
 283                         ETHER_HDR_LEN);
 284
 285         ethhdr->ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv4);
 286         memcpy(&ethhdr->s_addr, &ethaddr_tbl[port].src,
 287                         sizeof(struct ether_addr));
 288         memcpy(&ethhdr->d_addr, &ethaddr_tbl[port].dst,
 289                         sizeof(struct ether_addr));
 290 }
 291
 292 static inline void
 293 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint8_t port)
 294 {
 295         int32_t i;
 296         const int32_t prefetch_offset = 2;
 297
 298         for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
 299                 rte_prefetch0(pkts[i + prefetch_offset]->cacheline1);
 300                 prepare_tx_pkt(pkts[i], port);
 301         }
 302         /* Process left packets */
 303         for (; i < nb_pkts; i++)
 304                 prepare_tx_pkt(pkts[i], port);
 305 }
 306
 307 /* Send burst of packets on an output interface */
 308 static inline int32_t
 309 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 310 {
 311         struct rte_mbuf **m_table;
 312         int32_t ret;
 313         uint16_t queueid;
 314
 315         queueid = qconf->tx_queue_id[port];
 316         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 317
 318         prepare_tx_burst(m_table, n, port);
 319
 320         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 321         if (unlikely(ret < n)) {
 322                 do {
 323                         rte_pktmbuf_free(m_table[ret]);
 324                 } while (++ret < n);
 325         }
 326
 327         return 0;
 328 }
 329
 330 /* Enqueue a single packet, and send burst if queue is filled */
 331 static inline int32_t
 332 send_single_packet(struct rte_mbuf *m, uint8_t port)
 333 {
 334         uint32_t lcore_id;
 335         uint16_t len;
 336         struct lcore_conf *qconf;
 337
 338         lcore_id = rte_lcore_id();
 339
 340         qconf = &lcore_conf[lcore_id];
 341         len = qconf->tx_mbufs[port].len;
 342         qconf->tx_mbufs[port].m_table[len] = m;
 343         len++;
 344
 345         /* enough pkts to be sent */
 346         if (unlikely(len == MAX_PKT_BURST)) {
 347                 send_burst(qconf, MAX_PKT_BURST, port);
 348                 len = 0;
 349         }
 350
 351         qconf->tx_mbufs[port].len = len;
 352         return 0;
 353 }
 354
 355 static inline void
 356 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
 357                 struct ipsec_traffic *traffic)
 358 {
 359         struct rte_mbuf *m;
 360         uint16_t idx, nb_pkts_in, i, j;
 361         uint32_t sa_idx, res;
 362
 363         nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec4.pkts,
 364                         traffic->ipsec4.num, MAX_PKT_BURST);
 365
 366         /* SP/ACL Inbound check ipsec and ipv4 */
 367         for (i = 0; i < nb_pkts_in; i++) {
 368                 idx = traffic->ipv4.num++;
 369                 m = traffic->ipsec4.pkts[i];
 370                 traffic->ipv4.pkts[idx] = m;
 371                 traffic->ipv4.data[idx] = rte_pktmbuf_mtod_offset(m,
 372                                 uint8_t *, offsetof(struct ip, ip_p));
 373         }
 374
 375         rte_acl_classify((struct rte_acl_ctx *)ipsec_ctx->sp_ctx,
 376                         traffic->ipv4.data, traffic->ipv4.res,
 377                         traffic->ipv4.num, DEFAULT_MAX_CATEGORIES);
 378
 379         j = 0;
 380         for (i = 0; i < traffic->ipv4.num - nb_pkts_in; i++) {
 381                 m = traffic->ipv4.pkts[i];
 382                 res = traffic->ipv4.res[i];
 383                 if (res & ~BYPASS) {
 384                         rte_pktmbuf_free(m);
 385                         continue;
 386                 }
 387                 traffic->ipv4.pkts[j++] = m;
 388         }
 389         /* Check return SA SPI matches pkt SPI */
 390         for ( ; i < traffic->ipv4.num; i++) {
 391                 m = traffic->ipv4.pkts[i];
 392                 sa_idx = traffic->ipv4.res[i] & PROTECT_MASK;
 393                 if (sa_idx == 0 || !inbound_sa_check(ipsec_ctx->sa_ctx,
 394                                         m, sa_idx)) {
 395                         rte_pktmbuf_free(m);
 396                         continue;
 397                 }
 398                 traffic->ipv4.pkts[j++] = m;
 399         }
 400         traffic->ipv4.num = j;
 401 }
 402
 403 static inline void
 404 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
 405                 struct ipsec_traffic *traffic)
 406 {
 407         struct rte_mbuf *m;
 408         uint16_t idx, nb_pkts_out, i, j;
 409         uint32_t sa_idx, res;
 410
 411         rte_acl_classify((struct rte_acl_ctx *)ipsec_ctx->sp_ctx,
 412                         traffic->ipv4.data, traffic->ipv4.res,
 413                         traffic->ipv4.num, DEFAULT_MAX_CATEGORIES);
 414
 415         /* Drop any IPsec traffic from protected ports */
 416         for (i = 0; i < traffic->ipsec4.num; i++)
 417                 rte_pktmbuf_free(traffic->ipsec4.pkts[i]);
 418
 419         traffic->ipsec4.num = 0;
 420
 421         j = 0;
 422         for (i = 0; i < traffic->ipv4.num; i++) {
 423                 m = traffic->ipv4.pkts[i];
 424                 res = traffic->ipv4.res[i];
 425                 sa_idx = res & PROTECT_MASK;
 426                 if ((res == 0) || (res & DISCARD))
 427                         rte_pktmbuf_free(m);
 428                 else if (sa_idx != 0) {
 429                         traffic->ipsec4.res[traffic->ipsec4.num] = sa_idx;
 430                         traffic->ipsec4.pkts[traffic->ipsec4.num++] = m;
 431                 } else /* BYPASS */
 432                         traffic->ipv4.pkts[j++] = m;
 433         }
 434         traffic->ipv4.num = j;
 435
 436         nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec4.pkts,
 437                         traffic->ipsec4.res, traffic->ipsec4.num,
 438                         MAX_PKT_BURST);
 439
 440         for (i = 0; i < nb_pkts_out; i++) {
 441                 idx = traffic->ipv4.num++;
 442                 m = traffic->ipsec4.pkts[i];
 443                 traffic->ipv4.pkts[idx] = m;
 444         }
 445 }
 446
 447 static inline void
 448 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
 449                 struct ipsec_traffic *traffic)
 450 {
 451         uint16_t nb_pkts_in, i;
 452
 453         /* Drop any IPv4 traffic from unprotected ports */
 454         for (i = 0; i < traffic->ipv4.num; i++)
 455                 rte_pktmbuf_free(traffic->ipv4.pkts[i]);
 456
 457         traffic->ipv4.num = 0;
 458
 459         nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec4.pkts,
 460                         traffic->ipsec4.num, MAX_PKT_BURST);
 461
 462         for (i = 0; i < nb_pkts_in; i++)
 463                 traffic->ipv4.pkts[i] = traffic->ipsec4.pkts[i];
 464
 465         traffic->ipv4.num = nb_pkts_in;
 466 }
 467
 468 static inline void
 469 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
 470                 struct ipsec_traffic *traffic)
 471 {
 472         uint16_t nb_pkts_out, i;
 473
 474         /* Drop any IPsec traffic from protected ports */
 475         for (i = 0; i < traffic->ipsec4.num; i++)
 476                 rte_pktmbuf_free(traffic->ipsec4.pkts[i]);
 477
 478         traffic->ipsec4.num = 0;
 479
 480         for (i = 0; i < traffic->ipv4.num; i++)
 481                 traffic->ipv4.res[i] = single_sa_idx;
 482
 483         nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipv4.pkts,
 484                         traffic->ipv4.res, traffic->ipv4.num,
 485                         MAX_PKT_BURST);
 486
 487         traffic->ipv4.num = nb_pkts_out;
 488 }
 489
 490 static inline void
 491 route_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 492 {
 493         uint32_t hop[MAX_PKT_BURST * 2];
 494         uint32_t dst_ip[MAX_PKT_BURST * 2];
 495         uint16_t i, offset;
 496
 497         if (nb_pkts == 0)
 498                 return;
 499
 500         for (i = 0; i < nb_pkts; i++) {
 501                 offset = offsetof(struct ip, ip_dst);
 502                 dst_ip[i] = *rte_pktmbuf_mtod_offset(pkts[i],
 503                                 uint32_t *, offset);
 504                 dst_ip[i] = rte_be_to_cpu_32(dst_ip[i]);
 505         }
 506
 507         rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, nb_pkts);
 508
 509         for (i = 0; i < nb_pkts; i++) {
 510                 if ((hop[i] & RTE_LPM_LOOKUP_SUCCESS) == 0) {
 511                         rte_pktmbuf_free(pkts[i]);
 512                         continue;
 513                 }
 514                 send_single_packet(pkts[i], hop[i] & 0xff);
 515         }
 516 }
 517
 518 static inline void
 519 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
 520                 uint8_t nb_pkts, uint8_t portid)
 521 {
 522         struct ipsec_traffic traffic;
 523
 524         prepare_traffic(pkts, &traffic, nb_pkts);
 525
 526         if (single_sa) {
 527                 if (UNPROTECTED_PORT(portid))
 528                         process_pkts_inbound_nosp(&qconf->inbound, &traffic);
 529                 else
 530                         process_pkts_outbound_nosp(&qconf->outbound, &traffic);
 531         } else {
 532                 if (UNPROTECTED_PORT(portid))
 533                         process_pkts_inbound(&qconf->inbound, &traffic);
 534                 else
 535                         process_pkts_outbound(&qconf->outbound, &traffic);
 536         }
 537
 538         route_pkts(qconf->rt_ctx, traffic.ipv4.pkts, traffic.ipv4.num);
 539 }
 540
 541 static inline void
 542 drain_buffers(struct lcore_conf *qconf)
 543 {
 544         struct buffer *buf;
 545         uint32_t portid;
 546
 547         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 548                 buf = &qconf->tx_mbufs[portid];
 549                 if (buf->len == 0)
 550                         continue;
 551                 send_burst(qconf, buf->len, portid);
 552                 buf->len = 0;
 553         }
 554 }
 555
 556 /* main processing loop */
 557 static int32_t
 558 main_loop(__attribute__((unused)) void *dummy)
 559 {
 560         struct rte_mbuf *pkts[MAX_PKT_BURST];
 561         uint32_t lcore_id;
 562         uint64_t prev_tsc, diff_tsc, cur_tsc;
 563         int32_t i, nb_rx;
 564         uint8_t portid, queueid;
 565         struct lcore_conf *qconf;
 566         int32_t socket_id;
 567         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
 568                         / US_PER_S * BURST_TX_DRAIN_US;
 569         struct lcore_rx_queue *rxql;
 570
 571         prev_tsc = 0;
 572         lcore_id = rte_lcore_id();
 573         qconf = &lcore_conf[lcore_id];
 574         rxql = qconf->rx_queue_list;
 575         socket_id = rte_lcore_to_socket_id(lcore_id);
 576
 577         qconf->rt_ctx = socket_ctx[socket_id].rt_ipv4;
 578         qconf->inbound.sp_ctx = socket_ctx[socket_id].sp_ipv4_in;
 579         qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_ipv4_in;
 580         qconf->inbound.cdev_map = cdev_map_in;
 581         qconf->outbound.sp_ctx = socket_ctx[socket_id].sp_ipv4_out;
 582         qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_ipv4_out;
 583         qconf->outbound.cdev_map = cdev_map_out;
 584
 585         if (qconf->nb_rx_queue == 0) {
 586                 RTE_LOG(INFO, IPSEC, "lcore %u has nothing to do\n", lcore_id);
 587                 return 0;
 588         }
 589
 590         RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
 591
 592         for (i = 0; i < qconf->nb_rx_queue; i++) {
 593                 portid = rxql[i].port_id;
 594                 queueid = rxql[i].queue_id;
 595                 RTE_LOG(INFO, IPSEC,
 596                         " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n",
 597                         lcore_id, portid, queueid);
 598         }
 599
 600         while (1) {
 601                 cur_tsc = rte_rdtsc();
 602
 603                 /* TX queue buffer drain */
 604                 diff_tsc = cur_tsc - prev_tsc;
 605
 606                 if (unlikely(diff_tsc > drain_tsc)) {
 607                         drain_buffers(qconf);
 608                         prev_tsc = cur_tsc;
 609                 }
 610
 611                 /* Read packet from RX queues */
 612                 for (i = 0; i < qconf->nb_rx_queue; ++i) {
 613                         portid = rxql[i].port_id;
 614                         queueid = rxql[i].queue_id;
 615                         nb_rx = rte_eth_rx_burst(portid, queueid,
 616                                         pkts, MAX_PKT_BURST);
 617
 618                         if (nb_rx > 0)
 619                                 process_pkts(qconf, pkts, nb_rx, portid);
 620                 }
 621         }
 622 }
 623
 624 static int32_t
 625 check_params(void)
 626 {
 627         uint8_t lcore, portid, nb_ports;
 628         uint16_t i;
 629         int32_t socket_id;
 630
 631         if (lcore_params == NULL) {
 632                 printf("Error: No port/queue/core mappings\n");
 633                 return -1;
 634         }
 635
 636         nb_ports = rte_eth_dev_count();
 637         if (nb_ports > RTE_MAX_ETHPORTS)
 638                 nb_ports = RTE_MAX_ETHPORTS;
 639
 640         for (i = 0; i < nb_lcore_params; ++i) {
 641                 lcore = lcore_params[i].lcore_id;
 642                 if (!rte_lcore_is_enabled(lcore)) {
 643                         printf("error: lcore %hhu is not enabled in "
 644                                 "lcore mask\n", lcore);
 645                         return -1;
 646                 }
 647                 socket_id = rte_lcore_to_socket_id(lcore);
 648                 if (socket_id != 0 && numa_on == 0) {
 649                         printf("warning: lcore %hhu is on socket %d "
 650                                 "with numa off\n",
 651                                 lcore, socket_id);
 652                 }
 653                 portid = lcore_params[i].port_id;
 654                 if ((enabled_port_mask & (1 << portid)) == 0) {
 655                         printf("port %u is not enabled in port mask\n", portid);
 656                         return -1;
 657                 }
 658                 if (portid >= nb_ports) {
 659                         printf("port %u is not present on the board\n", portid);
 660                         return -1;
 661                 }
 662         }
 663         return 0;
 664 }
 665
 666 static uint8_t
 667 get_port_nb_rx_queues(const uint8_t port)
 668 {
 669         int32_t queue = -1;
 670         uint16_t i;
 671
 672         for (i = 0; i < nb_lcore_params; ++i) {
 673                 if (lcore_params[i].port_id == port &&
 674                                 lcore_params[i].queue_id > queue)
 675                         queue = lcore_params[i].queue_id;
 676         }
 677         return (uint8_t)(++queue);
 678 }
 679
 680 static int32_t
 681 init_lcore_rx_queues(void)
 682 {
 683         uint16_t i, nb_rx_queue;
 684         uint8_t lcore;
 685
 686         for (i = 0; i < nb_lcore_params; ++i) {
 687                 lcore = lcore_params[i].lcore_id;
 688                 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
 689                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
 690                         printf("error: too many queues (%u) for lcore: %u\n",
 691                                         nb_rx_queue + 1, lcore);
 692                         return -1;
 693                 }
 694                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
 695                         lcore_params[i].port_id;
 696                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
 697                         lcore_params[i].queue_id;
 698                 lcore_conf[lcore].nb_rx_queue++;
 699         }
 700         return 0;
 701 }
 702
 703 /* display usage */
 704 static void
 705 print_usage(const char *prgname)
 706 {
 707         printf("%s [EAL options] -- -p PORTMASK -P -u PORTMASK"
 708                 "  --"OPTION_CONFIG" (port,queue,lcore)[,(port,queue,lcore]"
 709                 " --single-sa SAIDX --ep0|--ep1\n"
 710                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 711                 "  -P : enable promiscuous mode\n"
 712                 "  -u PORTMASK: hexadecimal bitmask of unprotected ports\n"
 713                 "  --"OPTION_CONFIG": (port,queue,lcore): "
 714                 "rx queues configuration\n"
 715                 "  --single-sa SAIDX: use single SA index for outbound, "
 716                 "bypassing the SP\n"
 717                 "  --ep0: Configure as Endpoint 0\n"
 718                 "  --ep1: Configure as Endpoint 1\n", prgname);
 719 }
 720
 721 static int32_t
 722 parse_portmask(const char *portmask)
 723 {
 724         char *end = NULL;
 725         unsigned long pm;
 726
 727         /* parse hexadecimal string */
 728         pm = strtoul(portmask, &end, 16);
 729         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 730                 return -1;
 731
 732         if ((pm == 0) && errno)
 733                 return -1;
 734
 735         return pm;
 736 }
 737
 738 static int32_t
 739 parse_decimal(const char *str)
 740 {
 741         char *end = NULL;
 742         unsigned long num;
 743
 744         num = strtoul(str, &end, 10);
 745         if ((str[0] == '\0') || (end == NULL) || (*end != '\0'))
 746                 return -1;
 747
 748         return num;
 749 }
 750
 751 static int32_t
 752 parse_config(const char *q_arg)
 753 {
 754         char s[256];
 755         const char *p, *p0 = q_arg;
 756         char *end;
 757         enum fieldnames {
 758                 FLD_PORT = 0,
 759                 FLD_QUEUE,
 760                 FLD_LCORE,
 761                 _NUM_FLD
 762         };
 763         int long int_fld[_NUM_FLD];
 764         char *str_fld[_NUM_FLD];
 765         int32_t i;
 766         uint32_t size;
 767
 768         nb_lcore_params = 0;
 769
 770         while ((p = strchr(p0, '(')) != NULL) {
 771                 ++p;
 772                 p0 = strchr(p, ')');
 773                 if (p0 == NULL)
 774                         return -1;
 775
 776                 size = p0 - p;
 777                 if (size >= sizeof(s))
 778                         return -1;
 779
 780                 snprintf(s, sizeof(s), "%.*s", size, p);
 781                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
 782                                 _NUM_FLD)
 783                         return -1;
 784                 for (i = 0; i < _NUM_FLD; i++) {
 785                         errno = 0;
 786                         int_fld[i] = strtoul(str_fld[i], &end, 0);
 787                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
 788                                 return -1;
 789                 }
 790                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
 791                         printf("exceeded max number of lcore params: %hu\n",
 792                                 nb_lcore_params);
 793                         return -1;
 794                 }
 795                 lcore_params_array[nb_lcore_params].port_id =
 796                         (uint8_t)int_fld[FLD_PORT];
 797                 lcore_params_array[nb_lcore_params].queue_id =
 798                         (uint8_t)int_fld[FLD_QUEUE];
 799                 lcore_params_array[nb_lcore_params].lcore_id =
 800                         (uint8_t)int_fld[FLD_LCORE];
 801                 ++nb_lcore_params;
 802         }
 803         lcore_params = lcore_params_array;
 804         return 0;
 805 }
 806
 807 #define __STRNCMP(name, opt) (!strncmp(name, opt, sizeof(opt)))
 808 static int32_t
 809 parse_args_long_options(struct option *lgopts, int32_t option_index)
 810 {
 811         int32_t ret = -1;
 812         const char *optname = lgopts[option_index].name;
 813
 814         if (__STRNCMP(optname, OPTION_CONFIG)) {
 815                 ret = parse_config(optarg);
 816                 if (ret)
 817                         printf("invalid config\n");
 818         }
 819
 820         if (__STRNCMP(optname, OPTION_SINGLE_SA)) {
 821                 ret = parse_decimal(optarg);
 822                 if (ret != -1) {
 823                         single_sa = 1;
 824                         single_sa_idx = ret;
 825                         printf("Configured with single SA index %u\n",
 826                                         single_sa_idx);
 827                         ret = 0;
 828                 }
 829         }
 830
 831         if (__STRNCMP(optname, OPTION_EP0)) {
 832                 printf("endpoint 0\n");
 833                 ep = 0;
 834                 ret = 0;
 835         }
 836
 837         if (__STRNCMP(optname, OPTION_EP1)) {
 838                 printf("endpoint 1\n");
 839                 ep = 1;
 840                 ret = 0;
 841         }
 842
 843         return ret;
 844 }
 845 #undef __STRNCMP
 846
 847 static int32_t
 848 parse_args(int32_t argc, char **argv)
 849 {
 850         int32_t opt, ret;
 851         char **argvopt;
 852         int32_t option_index;
 853         char *prgname = argv[0];
 854         static struct option lgopts[] = {
 855                 {OPTION_CONFIG, 1, 0, 0},
 856                 {OPTION_SINGLE_SA, 1, 0, 0},
 857                 {OPTION_EP0, 0, 0, 0},
 858                 {OPTION_EP1, 0, 0, 0},
 859                 {NULL, 0, 0, 0}
 860         };
 861
 862         argvopt = argv;
 863
 864         while ((opt = getopt_long(argc, argvopt, "p:Pu:",
 865                                 lgopts, &option_index)) != EOF) {
 866
 867                 switch (opt) {
 868                 case 'p':
 869                         enabled_port_mask = parse_portmask(optarg);
 870                         if (enabled_port_mask == 0) {
 871                                 printf("invalid portmask\n");
 872                                 print_usage(prgname);
 873                                 return -1;
 874                         }
 875                         break;
 876                 case 'P':
 877                         printf("Promiscuous mode selected\n");
 878                         promiscuous_on = 1;
 879                         break;
 880                 case 'u':
 881                         unprotected_port_mask = parse_portmask(optarg);
 882                         if (unprotected_port_mask == 0) {
 883                                 printf("invalid unprotected portmask\n");
 884                                 print_usage(prgname);
 885                                 return -1;
 886                         }
 887                         break;
 888                 case 0:
 889                         if (parse_args_long_options(lgopts, option_index)) {
 890                                 print_usage(prgname);
 891                                 return -1;
 892                         }
 893                         break;
 894                 default:
 895                         print_usage(prgname);
 896                         return -1;
 897                 }
 898         }
 899
 900         if (optind >= 0)
 901                 argv[optind-1] = prgname;
 902
 903         ret = optind-1;
 904         optind = 0; /* reset getopt lib */
 905         return ret;
 906 }
 907
 908 static void
 909 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
 910 {
 911         char buf[ETHER_ADDR_FMT_SIZE];
 912         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
 913         printf("%s%s", name, buf);
 914 }
 915
 916 /* Check the link status of all ports in up to 9s, and print them finally */
 917 static void
 918 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
 919 {
 920 #define CHECK_INTERVAL 100 /* 100ms */
 921 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
 922         uint8_t portid, count, all_ports_up, print_flag = 0;
 923         struct rte_eth_link link;
 924
 925         printf("\nChecking link status");
 926         fflush(stdout);
 927         for (count = 0; count <= MAX_CHECK_TIME; count++) {
 928                 all_ports_up = 1;
 929                 for (portid = 0; portid < port_num; portid++) {
 930                         if ((port_mask & (1 << portid)) == 0)
 931                                 continue;
 932                         memset(&link, 0, sizeof(link));
 933                         rte_eth_link_get_nowait(portid, &link);
 934                         /* print link status if flag set */
 935                         if (print_flag == 1) {
 936                                 if (link.link_status)
 937                                         printf("Port %d Link Up - speed %u "
 938                                                 "Mbps - %s\n", (uint8_t)portid,
 939                                                 (uint32_t)link.link_speed,
 940                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
 941                                         ("full-duplex") : ("half-duplex\n"));
 942                                 else
 943                                         printf("Port %d Link Down\n",
 944                                                 (uint8_t)portid);
 945                                 continue;
 946                         }
 947                         /* clear all_ports_up flag if any link down */
 948                         if (link.link_status == 0) {
 949                                 all_ports_up = 0;
 950                                 break;
 951                         }
 952                 }
 953                 /* after finally printing all link status, get out */
 954                 if (print_flag == 1)
 955                         break;
 956
 957                 if (all_ports_up == 0) {
 958                         printf(".");
 959                         fflush(stdout);
 960                         rte_delay_ms(CHECK_INTERVAL);
 961                 }
 962
 963                 /* set the print_flag if all ports up or timeout */
 964                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
 965                         print_flag = 1;
 966                         printf("done\n");
 967                 }
 968         }
 969 }
 970
 971 static int32_t
 972 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
 973                 uint16_t qp, struct lcore_params *params,
 974                 struct ipsec_ctx *ipsec_ctx,
 975                 const struct rte_cryptodev_capabilities *cipher,
 976                 const struct rte_cryptodev_capabilities *auth)
 977 {
 978         int32_t ret = 0;
 979         unsigned long i;
 980         struct cdev_key key = { 0 };
 981
 982         key.lcore_id = params->lcore_id;
 983         if (cipher)
 984                 key.cipher_algo = cipher->sym.cipher.algo;
 985         if (auth)
 986                 key.auth_algo = auth->sym.auth.algo;
 987
 988         ret = rte_hash_lookup(map, &key);
 989         if (ret != -ENOENT)
 990                 return 0;
 991
 992         for (i = 0; i < ipsec_ctx->nb_qps; i++)
 993                 if (ipsec_ctx->tbl[i].id == cdev_id)
 994                         break;
 995
 996         if (i == ipsec_ctx->nb_qps) {
 997                 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
 998                         printf("Maximum number of crypto devices assigned to "
 999                                 "a core, increase MAX_QP_PER_LCORE value\n");
1000                         return 0;
1001                 }
1002                 ipsec_ctx->tbl[i].id = cdev_id;
1003                 ipsec_ctx->tbl[i].qp = qp;
1004                 ipsec_ctx->nb_qps++;
1005                 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1006                                 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1007                                 cdev_id, qp, i);
1008         }
1009
1010         ret = rte_hash_add_key_data(map, &key, (void *)i);
1011         if (ret < 0) {
1012                 printf("Faled to insert cdev mapping for (lcore %u, "
1013                                 "cdev %u, qp %u), errno %d\n",
1014                                 key.lcore_id, ipsec_ctx->tbl[i].id,
1015                                 ipsec_ctx->tbl[i].qp, ret);
1016                 return 0;
1017         }
1018
1019         return 1;
1020 }
1021
1022 static int32_t
1023 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1024                 uint16_t qp, struct lcore_params *params)
1025 {
1026         int32_t ret = 0;
1027         const struct rte_cryptodev_capabilities *i, *j;
1028         struct rte_hash *map;
1029         struct lcore_conf *qconf;
1030         struct ipsec_ctx *ipsec_ctx;
1031         const char *str;
1032
1033         qconf = &lcore_conf[params->lcore_id];
1034
1035         if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
1036                 map = cdev_map_out;
1037                 ipsec_ctx = &qconf->outbound;
1038                 str = "Outbound";
1039         } else {
1040                 map = cdev_map_in;
1041                 ipsec_ctx = &qconf->inbound;
1042                 str = "Inbound";
1043         }
1044
1045         /* Required cryptodevs with operation chainning */
1046         if (!(dev_info->feature_flags &
1047                                 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
1048                 return ret;
1049
1050         for (i = dev_info->capabilities;
1051                         i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
1052                 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1053                         continue;
1054
1055                 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
1056                         continue;
1057
1058                 for (j = dev_info->capabilities;
1059                                 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
1060                         if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1061                                 continue;
1062
1063                         if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
1064                                 continue;
1065
1066                         ret |= add_mapping(map, str, cdev_id, qp, params,
1067                                         ipsec_ctx, i, j);
1068                 }
1069         }
1070
1071         return ret;
1072 }
1073
1074 static int32_t
1075 cryptodevs_init(void)
1076 {
1077         struct rte_cryptodev_config dev_conf;
1078         struct rte_cryptodev_qp_conf qp_conf;
1079         uint16_t idx, max_nb_qps, qp, i;
1080         int16_t cdev_id;
1081         struct rte_hash_parameters params = { 0 };
1082
1083         params.entries = CDEV_MAP_ENTRIES;
1084         params.key_len = sizeof(struct cdev_key);
1085         params.hash_func = rte_jhash;
1086         params.hash_func_init_val = 0;
1087         params.socket_id = rte_socket_id();
1088
1089         params.name = "cdev_map_in";
1090         cdev_map_in = rte_hash_create(&params);
1091         if (cdev_map_in == NULL)
1092                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1093                                 rte_errno);
1094
1095         params.name = "cdev_map_out";
1096         cdev_map_out = rte_hash_create(&params);
1097         if (cdev_map_out == NULL)
1098                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1099                                 rte_errno);
1100
1101         printf("lcore/cryptodev/qp mappings:\n");
1102
1103         idx = 0;
1104         /* Start from last cdev id to give HW priority */
1105         for (cdev_id = rte_cryptodev_count() - 1; cdev_id >= 0; cdev_id--) {
1106                 struct rte_cryptodev_info cdev_info;
1107
1108                 rte_cryptodev_info_get(cdev_id, &cdev_info);
1109
1110                 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
1111                         max_nb_qps = cdev_info.max_nb_queue_pairs;
1112                 else
1113                         max_nb_qps = nb_lcore_params;
1114
1115                 qp = 0;
1116                 i = 0;
1117                 while (qp < max_nb_qps && i < nb_lcore_params) {
1118                         if (add_cdev_mapping(&cdev_info, cdev_id, qp,
1119                                                 &lcore_params[idx]))
1120                                 qp++;
1121                         idx++;
1122                         idx = idx % nb_lcore_params;
1123                         i++;
1124                 }
1125
1126                 if (qp == 0)
1127                         continue;
1128
1129                 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
1130                 dev_conf.nb_queue_pairs = qp;
1131                 dev_conf.session_mp.nb_objs = CDEV_MP_NB_OBJS;
1132                 dev_conf.session_mp.cache_size = CDEV_MP_CACHE_SZ;
1133
1134                 if (rte_cryptodev_configure(cdev_id, &dev_conf))
1135                         rte_panic("Failed to initialize crypodev %u\n",
1136                                         cdev_id);
1137
1138                 qp_conf.nb_descriptors = CDEV_MP_NB_OBJS;
1139                 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
1140                         if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
1141                                                 &qp_conf, dev_conf.socket_id))
1142                                 rte_panic("Failed to setup queue %u for "
1143                                                 "cdev_id %u\n", 0, cdev_id);
1144         }
1145
1146         printf("\n");
1147
1148         return 0;
1149 }
1150
1151 static void
1152 port_init(uint8_t portid)
1153 {
1154         struct rte_eth_dev_info dev_info;
1155         struct rte_eth_txconf *txconf;
1156         uint16_t nb_tx_queue, nb_rx_queue;
1157         uint16_t tx_queueid, rx_queueid, queue, lcore_id;
1158         int32_t ret, socket_id;
1159         struct lcore_conf *qconf;
1160         struct ether_addr ethaddr;
1161
1162         rte_eth_dev_info_get(portid, &dev_info);
1163
1164         printf("Configuring device port %u:\n", portid);
1165
1166         rte_eth_macaddr_get(portid, &ethaddr);
1167         ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(ethaddr);
1168         print_ethaddr("Address: ", &ethaddr);
1169         printf("\n");
1170
1171         nb_rx_queue = get_port_nb_rx_queues(portid);
1172         nb_tx_queue = nb_lcores;
1173
1174         if (nb_rx_queue > dev_info.max_rx_queues)
1175                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1176                                 "(max rx queue is %u)\n",
1177                                 nb_rx_queue, dev_info.max_rx_queues);
1178
1179         if (nb_tx_queue > dev_info.max_tx_queues)
1180                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1181                                 "(max tx queue is %u)\n",
1182                                 nb_tx_queue, dev_info.max_tx_queues);
1183
1184         printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
1185                         nb_rx_queue, nb_tx_queue);
1186
1187         ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
1188                         &port_conf);
1189         if (ret < 0)
1190                 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1191                                 "err=%d, port=%d\n", ret, portid);
1192
1193         /* init one TX queue per lcore */
1194         tx_queueid = 0;
1195         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1196                 if (rte_lcore_is_enabled(lcore_id) == 0)
1197                         continue;
1198
1199                 if (numa_on)
1200                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1201                 else
1202                         socket_id = 0;
1203
1204                 /* init TX queue */
1205                 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
1206
1207                 txconf = &dev_info.default_txconf;
1208                 txconf->txq_flags = 0;
1209
1210                 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
1211                                 socket_id, txconf);
1212                 if (ret < 0)
1213                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
1214                                         "err=%d, port=%d\n", ret, portid);
1215
1216                 qconf = &lcore_conf[lcore_id];
1217                 qconf->tx_queue_id[portid] = tx_queueid;
1218                 tx_queueid++;
1219
1220                 /* init RX queues */
1221                 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
1222                         if (portid != qconf->rx_queue_list[queue].port_id)
1223                                 continue;
1224
1225                         rx_queueid = qconf->rx_queue_list[queue].queue_id;
1226
1227                         printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
1228                                         socket_id);
1229
1230                         ret = rte_eth_rx_queue_setup(portid, rx_queueid,
1231                                         nb_rxd, socket_id, NULL,
1232                                         socket_ctx[socket_id].mbuf_pool);
1233                         if (ret < 0)
1234                                 rte_exit(EXIT_FAILURE,
1235                                         "rte_eth_rx_queue_setup: err=%d, "
1236                                         "port=%d\n", ret, portid);
1237                 }
1238         }
1239         printf("\n");
1240 }
1241
1242 static void
1243 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
1244 {
1245         char s[64];
1246
1247         snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
1248         ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
1249                         MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
1250                         RTE_MBUF_DEFAULT_BUF_SIZE,
1251                         socket_id);
1252         if (ctx->mbuf_pool == NULL)
1253                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
1254                                 socket_id);
1255         else
1256                 printf("Allocated mbuf pool on socket %d\n", socket_id);
1257 }
1258
1259 int32_t
1260 main(int32_t argc, char **argv)
1261 {
1262         int32_t ret;
1263         uint32_t lcore_id, nb_ports;
1264         uint8_t portid, socket_id;
1265
1266         /* init EAL */
1267         ret = rte_eal_init(argc, argv);
1268         if (ret < 0)
1269                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1270         argc -= ret;
1271         argv += ret;
1272
1273         /* parse application arguments (after the EAL ones) */
1274         ret = parse_args(argc, argv);
1275         if (ret < 0)
1276                 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
1277
1278         if (ep < 0)
1279                 rte_exit(EXIT_FAILURE, "need to choose either EP0 or EP1\n");
1280
1281         if ((unprotected_port_mask & enabled_port_mask) !=
1282                         unprotected_port_mask)
1283                 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
1284                                 unprotected_port_mask);
1285
1286         nb_ports = rte_eth_dev_count();
1287         if (nb_ports > RTE_MAX_ETHPORTS)
1288                 nb_ports = RTE_MAX_ETHPORTS;
1289
1290         if (check_params() < 0)
1291                 rte_exit(EXIT_FAILURE, "check_params failed\n");
1292
1293         ret = init_lcore_rx_queues();
1294         if (ret < 0)
1295                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1296
1297         nb_lcores = rte_lcore_count();
1298
1299         /* Replicate each contex per socket */
1300         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1301                 if (rte_lcore_is_enabled(lcore_id) == 0)
1302                         continue;
1303
1304                 if (numa_on)
1305                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1306                 else
1307                         socket_id = 0;
1308
1309                 if (socket_ctx[socket_id].mbuf_pool)
1310                         continue;
1311
1312                 sa_init(&socket_ctx[socket_id], socket_id, ep);
1313
1314                 sp_init(&socket_ctx[socket_id], socket_id, ep);
1315
1316                 rt_init(&socket_ctx[socket_id], socket_id, ep);
1317
1318                 pool_init(&socket_ctx[socket_id], socket_id, NB_MBUF);
1319         }
1320
1321         for (portid = 0; portid < nb_ports; portid++) {
1322                 if ((enabled_port_mask & (1 << portid)) == 0)
1323                         continue;
1324
1325                 port_init(portid);
1326         }
1327
1328         cryptodevs_init();
1329
1330         /* start ports */
1331         for (portid = 0; portid < nb_ports; portid++) {
1332                 if ((enabled_port_mask & (1 << portid)) == 0)
1333                         continue;
1334
1335                 /* Start device */
1336                 ret = rte_eth_dev_start(portid);
1337                 if (ret < 0)
1338                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
1339                                         "err=%d, port=%d\n", ret, portid);
1340                 /*
1341                  * If enabled, put device in promiscuous mode.
1342                  * This allows IO forwarding mode to forward packets
1343                  * to itself through 2 cross-connected  ports of the
1344                  * target machine.
1345                  */
1346                 if (promiscuous_on)
1347                         rte_eth_promiscuous_enable(portid);
1348         }
1349
1350         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1351
1352         /* launch per-lcore init on every lcore */
1353         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1354         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1355                 if (rte_eal_wait_lcore(lcore_id) < 0)
1356                         return -1;
1357         }
1358
1359         return 0;
1360 }