1 /* 2 * Copyright (C) 2017 THL A29 Limited, a Tencent company. 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions are met: 7 * 8 * 1. Redistributions of source code must retain the above copyright notice, this 9 * list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright notice, 11 * this list of conditions and the following disclaimer in the documentation 12 * and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED 16 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 17 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR 18 * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 19 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 20 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND 21 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 22 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 23 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 24 * 25 */ 26 #include <assert.h> 27 #include <unistd.h> 28 #include <sys/mman.h> 29 #include <errno.h> 30 31 #include <rte_common.h> 32 #include <rte_byteorder.h> 33 #include <rte_log.h> 34 #include <rte_memory.h> 35 #include <rte_memcpy.h> 36 #include <rte_memzone.h> 37 #include <rte_config.h> 38 #include <rte_eal.h> 39 #include <rte_pci.h> 40 #include <rte_mbuf.h> 41 #include <rte_memory.h> 42 #include <rte_lcore.h> 43 #include <rte_launch.h> 44 #include <rte_ethdev.h> 45 #include <rte_debug.h> 46 #include <rte_common.h> 47 #include <rte_ether.h> 48 #include <rte_malloc.h> 49 #include <rte_cycles.h> 50 #include <rte_timer.h> 51 #include <rte_thash.h> 52 #include <rte_ip.h> 53 #include <rte_tcp.h> 54 #include <rte_udp.h> 55 #include <rte_eth_bond.h> 56 57 #include "ff_dpdk_if.h" 58 #include "ff_dpdk_pcap.h" 59 #include "ff_dpdk_kni.h" 60 #include "ff_config.h" 61 #include "ff_veth.h" 62 #include "ff_host_interface.h" 63 #include "ff_msg.h" 64 #include "ff_api.h" 65 #include "ff_memory.h" 66 67 #ifdef FF_KNI 68 #define KNI_MBUF_MAX 2048 69 #define KNI_QUEUE_SIZE 2048 70 71 int enable_kni; 72 static int kni_accept; 73 #endif 74 75 static int numa_on; 76 77 static unsigned idle_sleep; 78 static unsigned pkt_tx_delay; 79 80 static struct rte_timer freebsd_clock; 81 82 // Mellanox Linux's driver key 83 static uint8_t default_rsskey_40bytes[40] = { 84 0xd1, 0x81, 0xc6, 0x2c, 0xf7, 0xf4, 0xdb, 0x5b, 85 0x19, 0x83, 0xa2, 0xfc, 0x94, 0x3e, 0x1a, 0xdb, 86 0xd9, 0x38, 0x9e, 0x6b, 0xd1, 0x03, 0x9c, 0x2c, 87 0xa7, 0x44, 0x99, 0xad, 0x59, 0x3d, 0x56, 0xd9, 88 0xf3, 0x25, 0x3c, 0x06, 0x2a, 0xdc, 0x1f, 0xfc 89 }; 90 91 static int use_rsskey_52bytes = 0; 92 static uint8_t default_rsskey_52bytes[52] = { 93 0x44, 0x39, 0x79, 0x6b, 0xb5, 0x4c, 0x50, 0x23, 94 0xb6, 0x75, 0xea, 0x5b, 0x12, 0x4f, 0x9f, 0x30, 95 0xb8, 0xa2, 0xc0, 0x3d, 0xdf, 0xdc, 0x4d, 0x02, 96 0xa0, 0x8c, 0x9b, 0x33, 0x4a, 0xf6, 0x4a, 0x4c, 97 0x05, 0xc6, 0xfa, 0x34, 0x39, 0x58, 0xd8, 0x55, 98 0x7d, 0x99, 0x58, 0x3a, 0xe1, 0x38, 0xc9, 0x2e, 99 0x81, 0x15, 0x03, 0x66 100 }; 101 102 struct lcore_conf lcore_conf; 103 104 struct rte_mempool *pktmbuf_pool[NB_SOCKETS]; 105 106 static struct rte_ring **dispatch_ring[RTE_MAX_ETHPORTS]; 107 static dispatch_func_t packet_dispatcher; 108 109 static uint16_t rss_reta_size[RTE_MAX_ETHPORTS]; 110 111 #define BOND_DRIVER_NAME "net_bonding" 112 113 static inline int send_single_packet(struct rte_mbuf *m, uint8_t port); 114 115 struct ff_msg_ring { 116 char ring_name[FF_MSG_NUM][RTE_RING_NAMESIZE]; 117 /* ring[0] for lcore recv msg, other send */ 118 /* ring[1] for lcore send msg, other read */ 119 struct rte_ring *ring[FF_MSG_NUM]; 120 } __rte_cache_aligned; 121 122 static struct ff_msg_ring msg_ring[RTE_MAX_LCORE]; 123 static struct rte_mempool *message_pool; 124 static struct ff_dpdk_if_context *veth_ctx[RTE_MAX_ETHPORTS]; 125 126 static struct ff_top_args ff_top_status; 127 static struct ff_traffic_args ff_traffic; 128 extern void ff_hardclock(void); 129 130 static void 131 ff_hardclock_job(__rte_unused struct rte_timer *timer, 132 __rte_unused void *arg) { 133 ff_hardclock(); 134 ff_update_current_ts(); 135 } 136 137 struct ff_dpdk_if_context * 138 ff_dpdk_register_if(void *sc, void *ifp, struct ff_port_cfg *cfg) 139 { 140 struct ff_dpdk_if_context *ctx; 141 142 ctx = calloc(1, sizeof(struct ff_dpdk_if_context)); 143 if (ctx == NULL) 144 return NULL; 145 146 ctx->sc = sc; 147 ctx->ifp = ifp; 148 ctx->port_id = cfg->port_id; 149 ctx->hw_features = cfg->hw_features; 150 151 return ctx; 152 } 153 154 void 155 ff_dpdk_deregister_if(struct ff_dpdk_if_context *ctx) 156 { 157 free(ctx); 158 } 159 160 static void 161 check_all_ports_link_status(void) 162 { 163 #define CHECK_INTERVAL 100 /* 100ms */ 164 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */ 165 166 uint16_t portid; 167 uint8_t count, all_ports_up, print_flag = 0; 168 struct rte_eth_link link; 169 170 printf("\nChecking link status"); 171 fflush(stdout); 172 173 int i, nb_ports; 174 nb_ports = ff_global_cfg.dpdk.nb_ports; 175 for (count = 0; count <= MAX_CHECK_TIME; count++) { 176 all_ports_up = 1; 177 for (i = 0; i < nb_ports; i++) { 178 uint16_t portid = ff_global_cfg.dpdk.portid_list[i]; 179 memset(&link, 0, sizeof(link)); 180 rte_eth_link_get_nowait(portid, &link); 181 182 /* print link status if flag set */ 183 if (print_flag == 1) { 184 if (link.link_status) { 185 printf("Port %d Link Up - speed %u " 186 "Mbps - %s\n", (int)portid, 187 (unsigned)link.link_speed, 188 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ? 189 ("full-duplex") : ("half-duplex\n")); 190 } else { 191 printf("Port %d Link Down\n", (int)portid); 192 } 193 continue; 194 } 195 /* clear all_ports_up flag if any link down */ 196 if (link.link_status == 0) { 197 all_ports_up = 0; 198 break; 199 } 200 } 201 202 /* after finally printing all link status, get out */ 203 if (print_flag == 1) 204 break; 205 206 if (all_ports_up == 0) { 207 printf("."); 208 fflush(stdout); 209 rte_delay_ms(CHECK_INTERVAL); 210 } 211 212 /* set the print_flag if all ports up or timeout */ 213 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) { 214 print_flag = 1; 215 printf("done\n"); 216 } 217 } 218 } 219 220 static int 221 init_lcore_conf(void) 222 { 223 uint8_t nb_dev_ports = rte_eth_dev_count_avail(); 224 if (nb_dev_ports == 0) { 225 rte_exit(EXIT_FAILURE, "No probed ethernet devices\n"); 226 } 227 228 if (ff_global_cfg.dpdk.max_portid >= nb_dev_ports) { 229 rte_exit(EXIT_FAILURE, "this machine doesn't have port %d.\n", 230 ff_global_cfg.dpdk.max_portid); 231 } 232 233 lcore_conf.port_cfgs = ff_global_cfg.dpdk.port_cfgs; 234 lcore_conf.proc_id = ff_global_cfg.dpdk.proc_id; 235 236 uint16_t proc_id; 237 for (proc_id = 0; proc_id < ff_global_cfg.dpdk.nb_procs; proc_id++) { 238 uint16_t lcore_id = ff_global_cfg.dpdk.proc_lcore[proc_id]; 239 if (!lcore_config[lcore_id].detected) { 240 rte_exit(EXIT_FAILURE, "lcore %u unavailable\n", lcore_id); 241 } 242 } 243 244 uint16_t socket_id = 0; 245 if (numa_on) { 246 socket_id = rte_lcore_to_socket_id(rte_lcore_id()); 247 } 248 249 lcore_conf.socket_id = socket_id; 250 251 uint16_t lcore_id = ff_global_cfg.dpdk.proc_lcore[lcore_conf.proc_id]; 252 int j; 253 for (j = 0; j < ff_global_cfg.dpdk.nb_ports; ++j) { 254 uint16_t port_id = ff_global_cfg.dpdk.portid_list[j]; 255 struct ff_port_cfg *pconf = &ff_global_cfg.dpdk.port_cfgs[port_id]; 256 257 int queueid = -1; 258 int i; 259 for (i = 0; i < pconf->nb_lcores; i++) { 260 if (pconf->lcore_list[i] == lcore_id) { 261 queueid = i; 262 } 263 } 264 if (queueid < 0) { 265 continue; 266 } 267 printf("lcore: %u, port: %u, queue: %u\n", lcore_id, port_id, queueid); 268 uint16_t nb_rx_queue = lcore_conf.nb_rx_queue; 269 lcore_conf.rx_queue_list[nb_rx_queue].port_id = port_id; 270 lcore_conf.rx_queue_list[nb_rx_queue].queue_id = queueid; 271 lcore_conf.nb_rx_queue++; 272 273 lcore_conf.tx_queue_id[port_id] = queueid; 274 lcore_conf.tx_port_id[lcore_conf.nb_tx_port] = port_id; 275 lcore_conf.nb_tx_port++; 276 277 lcore_conf.pcap[port_id] = pconf->pcap; 278 lcore_conf.nb_queue_list[port_id] = pconf->nb_lcores; 279 } 280 281 if (lcore_conf.nb_rx_queue == 0) { 282 rte_exit(EXIT_FAILURE, "lcore %u has nothing to do\n", lcore_id); 283 } 284 285 return 0; 286 } 287 288 static int 289 init_mem_pool(void) 290 { 291 uint8_t nb_ports = ff_global_cfg.dpdk.nb_ports; 292 uint32_t nb_lcores = ff_global_cfg.dpdk.nb_procs; 293 uint32_t nb_tx_queue = nb_lcores; 294 uint32_t nb_rx_queue = lcore_conf.nb_rx_queue * nb_lcores; 295 uint16_t max_portid = ff_global_cfg.dpdk.max_portid; 296 297 unsigned nb_mbuf = RTE_ALIGN_CEIL ( 298 (nb_rx_queue * (max_portid + 1) * 2 * RX_QUEUE_SIZE + 299 nb_ports * (max_portid + 1) * 2 * nb_lcores * MAX_PKT_BURST + 300 nb_ports * (max_portid + 1) * 2 * nb_tx_queue * TX_QUEUE_SIZE + 301 nb_lcores * MEMPOOL_CACHE_SIZE + 302 #ifdef FF_KNI 303 nb_ports * KNI_MBUF_MAX + 304 nb_ports * KNI_QUEUE_SIZE + 305 #endif 306 nb_lcores * nb_ports * DISPATCH_RING_SIZE), 307 (unsigned)8192); 308 309 unsigned socketid = 0; 310 uint16_t i, lcore_id; 311 char s[64]; 312 313 for (i = 0; i < ff_global_cfg.dpdk.nb_procs; i++) { 314 lcore_id = ff_global_cfg.dpdk.proc_lcore[i]; 315 if (numa_on) { 316 socketid = rte_lcore_to_socket_id(lcore_id); 317 } 318 319 if (socketid >= NB_SOCKETS) { 320 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n", 321 socketid, i, NB_SOCKETS); 322 } 323 324 if (pktmbuf_pool[socketid] != NULL) { 325 continue; 326 } 327 328 if (rte_eal_process_type() == RTE_PROC_PRIMARY) { 329 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid); 330 pktmbuf_pool[socketid] = 331 rte_pktmbuf_pool_create(s, nb_mbuf, 332 MEMPOOL_CACHE_SIZE, 0, 333 RTE_MBUF_DEFAULT_BUF_SIZE, socketid); 334 } else { 335 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid); 336 pktmbuf_pool[socketid] = rte_mempool_lookup(s); 337 } 338 339 if (pktmbuf_pool[socketid] == NULL) { 340 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool on socket %d\n", socketid); 341 } else { 342 printf("create mbuf pool on socket %d\n", socketid); 343 } 344 345 #ifdef FF_USE_PAGE_ARRAY 346 nb_mbuf = RTE_ALIGN_CEIL ( 347 nb_ports*nb_lcores*MAX_PKT_BURST + 348 nb_ports*nb_tx_queue*TX_QUEUE_SIZE + 349 nb_lcores*MEMPOOL_CACHE_SIZE, 350 (unsigned)4096); 351 ff_init_ref_pool(nb_mbuf, socketid); 352 #endif 353 } 354 355 return 0; 356 } 357 358 static struct rte_ring * 359 create_ring(const char *name, unsigned count, int socket_id, unsigned flags) 360 { 361 struct rte_ring *ring; 362 363 if (name == NULL) { 364 rte_exit(EXIT_FAILURE, "create ring failed, no name!\n"); 365 } 366 367 if (rte_eal_process_type() == RTE_PROC_PRIMARY) { 368 ring = rte_ring_create(name, count, socket_id, flags); 369 } else { 370 ring = rte_ring_lookup(name); 371 } 372 373 if (ring == NULL) { 374 rte_exit(EXIT_FAILURE, "create ring:%s failed!\n", name); 375 } 376 377 return ring; 378 } 379 380 static int 381 init_dispatch_ring(void) 382 { 383 int j; 384 char name_buf[RTE_RING_NAMESIZE]; 385 int queueid; 386 387 unsigned socketid = lcore_conf.socket_id; 388 389 /* Create ring according to ports actually being used. */ 390 int nb_ports = ff_global_cfg.dpdk.nb_ports; 391 for (j = 0; j < nb_ports; j++) { 392 uint16_t portid = ff_global_cfg.dpdk.portid_list[j]; 393 struct ff_port_cfg *pconf = &ff_global_cfg.dpdk.port_cfgs[portid]; 394 int nb_queues = pconf->nb_lcores; 395 if (dispatch_ring[portid] == NULL) { 396 snprintf(name_buf, RTE_RING_NAMESIZE, "ring_ptr_p%d", portid); 397 398 dispatch_ring[portid] = rte_zmalloc(name_buf, 399 sizeof(struct rte_ring *) * nb_queues, 400 RTE_CACHE_LINE_SIZE); 401 if (dispatch_ring[portid] == NULL) { 402 rte_exit(EXIT_FAILURE, "rte_zmalloc(%s (struct rte_ring*)) " 403 "failed\n", name_buf); 404 } 405 } 406 407 for(queueid = 0; queueid < nb_queues; ++queueid) { 408 snprintf(name_buf, RTE_RING_NAMESIZE, "dispatch_ring_p%d_q%d", 409 portid, queueid); 410 dispatch_ring[portid][queueid] = create_ring(name_buf, 411 DISPATCH_RING_SIZE, socketid, RING_F_SC_DEQ); 412 413 if (dispatch_ring[portid][queueid] == NULL) 414 rte_panic("create ring:%s failed!\n", name_buf); 415 416 printf("create ring:%s success, %u ring entries are now free!\n", 417 name_buf, rte_ring_free_count(dispatch_ring[portid][queueid])); 418 } 419 } 420 421 return 0; 422 } 423 424 static void 425 ff_msg_init(struct rte_mempool *mp, 426 __attribute__((unused)) void *opaque_arg, 427 void *obj, __attribute__((unused)) unsigned i) 428 { 429 struct ff_msg *msg = (struct ff_msg *)obj; 430 msg->msg_type = FF_UNKNOWN; 431 msg->buf_addr = (char *)msg + sizeof(struct ff_msg); 432 msg->buf_len = mp->elt_size - sizeof(struct ff_msg); 433 } 434 435 static int 436 init_msg_ring(void) 437 { 438 uint16_t i, j; 439 uint16_t nb_procs = ff_global_cfg.dpdk.nb_procs; 440 unsigned socketid = lcore_conf.socket_id; 441 442 /* Create message buffer pool */ 443 if (rte_eal_process_type() == RTE_PROC_PRIMARY) { 444 message_pool = rte_mempool_create(FF_MSG_POOL, 445 MSG_RING_SIZE * 2 * nb_procs, 446 MAX_MSG_BUF_SIZE, MSG_RING_SIZE / 2, 0, 447 NULL, NULL, ff_msg_init, NULL, 448 socketid, 0); 449 } else { 450 message_pool = rte_mempool_lookup(FF_MSG_POOL); 451 } 452 453 if (message_pool == NULL) { 454 rte_panic("Create msg mempool failed\n"); 455 } 456 457 for(i = 0; i < nb_procs; ++i) { 458 snprintf(msg_ring[i].ring_name[0], RTE_RING_NAMESIZE, 459 "%s%u", FF_MSG_RING_IN, i); 460 msg_ring[i].ring[0] = create_ring(msg_ring[i].ring_name[0], 461 MSG_RING_SIZE, socketid, RING_F_SP_ENQ | RING_F_SC_DEQ); 462 if (msg_ring[i].ring[0] == NULL) 463 rte_panic("create ring::%s failed!\n", msg_ring[i].ring_name[0]); 464 465 for (j = FF_SYSCTL; j < FF_MSG_NUM; j++) { 466 snprintf(msg_ring[i].ring_name[j], RTE_RING_NAMESIZE, 467 "%s%u_%u", FF_MSG_RING_OUT, i, j); 468 msg_ring[i].ring[j] = create_ring(msg_ring[i].ring_name[j], 469 MSG_RING_SIZE, socketid, RING_F_SP_ENQ | RING_F_SC_DEQ); 470 if (msg_ring[i].ring[j] == NULL) 471 rte_panic("create ring::%s failed!\n", msg_ring[i].ring_name[j]); 472 } 473 } 474 475 return 0; 476 } 477 478 #ifdef FF_KNI 479 static int 480 init_kni(void) 481 { 482 int nb_ports = rte_eth_dev_count_avail(); 483 kni_accept = 0; 484 if(strcasecmp(ff_global_cfg.kni.method, "accept") == 0) 485 kni_accept = 1; 486 487 ff_kni_init(nb_ports, ff_global_cfg.kni.tcp_port, 488 ff_global_cfg.kni.udp_port); 489 490 unsigned socket_id = lcore_conf.socket_id; 491 struct rte_mempool *mbuf_pool = pktmbuf_pool[socket_id]; 492 493 nb_ports = ff_global_cfg.dpdk.nb_ports; 494 int i, ret; 495 for (i = 0; i < nb_ports; i++) { 496 uint16_t port_id = ff_global_cfg.dpdk.portid_list[i]; 497 ff_kni_alloc(port_id, socket_id, mbuf_pool, KNI_QUEUE_SIZE); 498 } 499 500 return 0; 501 } 502 #endif 503 504 static void 505 set_rss_table(uint16_t port_id, uint16_t reta_size, uint16_t nb_queues) 506 { 507 if (reta_size == 0) { 508 return; 509 } 510 511 int reta_conf_size = RTE_MAX(1, reta_size / RTE_RETA_GROUP_SIZE); 512 struct rte_eth_rss_reta_entry64 reta_conf[reta_conf_size]; 513 514 /* config HW indirection table */ 515 unsigned i, j, hash=0; 516 for (i = 0; i < reta_conf_size; i++) { 517 reta_conf[i].mask = ~0ULL; 518 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++) { 519 reta_conf[i].reta[j] = hash++ % nb_queues; 520 } 521 } 522 523 if (rte_eth_dev_rss_reta_update(port_id, reta_conf, reta_size)) { 524 rte_exit(EXIT_FAILURE, "port[%d], failed to update rss table\n", 525 port_id); 526 } 527 } 528 529 static int 530 init_port_start(void) 531 { 532 int nb_ports = ff_global_cfg.dpdk.nb_ports; 533 unsigned socketid = 0; 534 struct rte_mempool *mbuf_pool; 535 uint16_t i, j; 536 537 for (i = 0; i < nb_ports; i++) { 538 uint16_t port_id, u_port_id = ff_global_cfg.dpdk.portid_list[i]; 539 struct ff_port_cfg *pconf = &ff_global_cfg.dpdk.port_cfgs[u_port_id]; 540 uint16_t nb_queues = pconf->nb_lcores; 541 542 for (j=0; j<=pconf->nb_slaves; j++) { 543 if (j < pconf->nb_slaves) { 544 port_id = pconf->slave_portid_list[j]; 545 printf("To init %s's %d'st slave port[%d]\n", 546 ff_global_cfg.dpdk.bond_cfgs->name, 547 j, port_id); 548 } else { 549 port_id = u_port_id; 550 } 551 552 struct rte_eth_dev_info dev_info; 553 struct rte_eth_conf port_conf = {0}; 554 struct rte_eth_rxconf rxq_conf; 555 struct rte_eth_txconf txq_conf; 556 557 rte_eth_dev_info_get(port_id, &dev_info); 558 559 if (nb_queues > dev_info.max_rx_queues) { 560 rte_exit(EXIT_FAILURE, "num_procs[%d] bigger than max_rx_queues[%d]\n", 561 nb_queues, 562 dev_info.max_rx_queues); 563 } 564 565 if (nb_queues > dev_info.max_tx_queues) { 566 rte_exit(EXIT_FAILURE, "num_procs[%d] bigger than max_tx_queues[%d]\n", 567 nb_queues, 568 dev_info.max_tx_queues); 569 } 570 571 /* Set RSS mode */ 572 uint64_t default_rss_hf = ETH_RSS_PROTO_MASK; 573 port_conf.rxmode.mq_mode = ETH_MQ_RX_RSS; 574 port_conf.rx_adv_conf.rss_conf.rss_hf = default_rss_hf; 575 if (dev_info.hash_key_size == 52) { 576 port_conf.rx_adv_conf.rss_conf.rss_key = default_rsskey_52bytes; 577 port_conf.rx_adv_conf.rss_conf.rss_key_len = 52; 578 use_rsskey_52bytes = 1; 579 } else { 580 port_conf.rx_adv_conf.rss_conf.rss_key = default_rsskey_40bytes; 581 port_conf.rx_adv_conf.rss_conf.rss_key_len = 40; 582 } 583 port_conf.rx_adv_conf.rss_conf.rss_hf &= dev_info.flow_type_rss_offloads; 584 if (port_conf.rx_adv_conf.rss_conf.rss_hf != 585 ETH_RSS_PROTO_MASK) { 586 printf("Port %u modified RSS hash function based on hardware support," 587 "requested:%#"PRIx64" configured:%#"PRIx64"\n", 588 port_id, default_rss_hf, 589 port_conf.rx_adv_conf.rss_conf.rss_hf); 590 } 591 592 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE) { 593 port_conf.txmode.offloads |= 594 DEV_TX_OFFLOAD_MBUF_FAST_FREE; 595 } 596 597 /* Set Rx VLAN stripping */ 598 if (ff_global_cfg.dpdk.vlan_strip) { 599 if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_VLAN_STRIP) { 600 port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_VLAN_STRIP; 601 } 602 } 603 604 /* Enable HW CRC stripping */ 605 port_conf.rxmode.offloads &= ~DEV_RX_OFFLOAD_KEEP_CRC; 606 607 /* FIXME: Enable TCP LRO ?*/ 608 #if 0 609 if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_TCP_LRO) { 610 printf("LRO is supported\n"); 611 port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_TCP_LRO; 612 pconf->hw_features.rx_lro = 1; 613 } 614 #endif 615 616 /* Set Rx checksum checking */ 617 if ((dev_info.rx_offload_capa & DEV_RX_OFFLOAD_IPV4_CKSUM) && 618 (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_UDP_CKSUM) && 619 (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_TCP_CKSUM)) { 620 printf("RX checksum offload supported\n"); 621 port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_CHECKSUM; 622 pconf->hw_features.rx_csum = 1; 623 } 624 625 if ((dev_info.tx_offload_capa & DEV_TX_OFFLOAD_IPV4_CKSUM)) { 626 printf("TX ip checksum offload supported\n"); 627 port_conf.txmode.offloads |= DEV_TX_OFFLOAD_IPV4_CKSUM; 628 pconf->hw_features.tx_csum_ip = 1; 629 } 630 631 if ((dev_info.tx_offload_capa & DEV_TX_OFFLOAD_UDP_CKSUM) && 632 (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_TCP_CKSUM)) { 633 printf("TX TCP&UDP checksum offload supported\n"); 634 port_conf.txmode.offloads |= DEV_TX_OFFLOAD_UDP_CKSUM | DEV_TX_OFFLOAD_TCP_CKSUM; 635 pconf->hw_features.tx_csum_l4 = 1; 636 } 637 638 if (ff_global_cfg.dpdk.tso) { 639 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_TCP_TSO) { 640 printf("TSO is supported\n"); 641 port_conf.txmode.offloads |= DEV_TX_OFFLOAD_TCP_TSO; 642 pconf->hw_features.tx_tso = 1; 643 } 644 } else { 645 printf("TSO is disabled\n"); 646 } 647 648 if (dev_info.reta_size) { 649 /* reta size must be power of 2 */ 650 assert((dev_info.reta_size & (dev_info.reta_size - 1)) == 0); 651 652 rss_reta_size[port_id] = dev_info.reta_size; 653 printf("port[%d]: rss table size: %d\n", port_id, 654 dev_info.reta_size); 655 } 656 657 if (rte_eal_process_type() != RTE_PROC_PRIMARY) { 658 continue; 659 } 660 661 int ret = rte_eth_dev_configure(port_id, nb_queues, nb_queues, &port_conf); 662 if (ret != 0) { 663 return ret; 664 } 665 666 static uint16_t nb_rxd = RX_QUEUE_SIZE; 667 static uint16_t nb_txd = TX_QUEUE_SIZE; 668 ret = rte_eth_dev_adjust_nb_rx_tx_desc(port_id, &nb_rxd, &nb_txd); 669 if (ret < 0) 670 printf("Could not adjust number of descriptors " 671 "for port%u (%d)\n", (unsigned)port_id, ret); 672 673 uint16_t q; 674 for (q = 0; q < nb_queues; q++) { 675 if (numa_on) { 676 uint16_t lcore_id = lcore_conf.port_cfgs[port_id].lcore_list[q]; 677 socketid = rte_lcore_to_socket_id(lcore_id); 678 } 679 mbuf_pool = pktmbuf_pool[socketid]; 680 681 txq_conf = dev_info.default_txconf; 682 txq_conf.offloads = port_conf.txmode.offloads; 683 ret = rte_eth_tx_queue_setup(port_id, q, nb_txd, 684 socketid, &txq_conf); 685 if (ret < 0) { 686 return ret; 687 } 688 689 rxq_conf = dev_info.default_rxconf; 690 rxq_conf.offloads = port_conf.rxmode.offloads; 691 ret = rte_eth_rx_queue_setup(port_id, q, nb_rxd, 692 socketid, &rxq_conf, mbuf_pool); 693 if (ret < 0) { 694 return ret; 695 } 696 } 697 698 struct ether_addr addr; 699 rte_eth_macaddr_get(port_id, &addr); 700 printf("Port %u MAC: %02" PRIx8 " %02" PRIx8 " %02" PRIx8 701 " %02" PRIx8 " %02" PRIx8 " %02" PRIx8 "\n", 702 (unsigned)port_id, 703 addr.addr_bytes[0], addr.addr_bytes[1], 704 addr.addr_bytes[2], addr.addr_bytes[3], 705 addr.addr_bytes[4], addr.addr_bytes[5]); 706 707 rte_memcpy(pconf->mac, 708 addr.addr_bytes, ETHER_ADDR_LEN); 709 710 if (strncmp(dev_info.driver_name, BOND_DRIVER_NAME, 711 strlen(dev_info.driver_name)) == 0) { 712 int mode, count, x; 713 uint16_t slaves[RTE_MAX_ETHPORTS], len = RTE_MAX_ETHPORTS; 714 715 mode = rte_eth_bond_mode_get(port_id); 716 printf("Port %u, bond mode:%d\n", mode); 717 718 count = rte_eth_bond_slaves_get(port_id, slaves, len); 719 printf("Port %u, %s's slave ports count:%d\n", 720 ff_global_cfg.dpdk.bond_cfgs->name, count); 721 for (x=0; x<count; x++) { 722 printf("Port %u, %s's slave port[%u]\n", 723 ff_global_cfg.dpdk.bond_cfgs->name, slaves[x]); 724 } 725 } 726 727 ret = rte_eth_dev_start(port_id); 728 if (ret < 0) { 729 return ret; 730 } 731 732 if (nb_queues > 1) { 733 /* set HW rss hash function to Toeplitz. */ 734 if (!rte_eth_dev_filter_supported(port_id, RTE_ETH_FILTER_HASH)) { 735 struct rte_eth_hash_filter_info info = {0}; 736 info.info_type = RTE_ETH_HASH_FILTER_GLOBAL_CONFIG; 737 info.info.global_conf.hash_func = RTE_ETH_HASH_FUNCTION_TOEPLITZ; 738 739 if (rte_eth_dev_filter_ctrl(port_id, RTE_ETH_FILTER_HASH, 740 RTE_ETH_FILTER_SET, &info) < 0) { 741 rte_exit(EXIT_FAILURE, "port[%d] set hash func failed\n", 742 port_id); 743 } 744 } 745 746 set_rss_table(port_id, dev_info.reta_size, nb_queues); 747 } 748 749 /* Enable RX in promiscuous mode for the Ethernet device. */ 750 if (ff_global_cfg.dpdk.promiscuous) { 751 rte_eth_promiscuous_enable(port_id); 752 ret = rte_eth_promiscuous_get(port_id); 753 if (ret == 1) { 754 printf("set port %u to promiscuous mode ok\n", port_id); 755 } else { 756 printf("set port %u to promiscuous mode error\n", port_id); 757 } 758 } 759 760 /* Enable pcap dump */ 761 if (pconf->pcap) { 762 ff_enable_pcap(pconf->pcap); 763 } 764 } 765 } 766 767 if (rte_eal_process_type() == RTE_PROC_PRIMARY) { 768 check_all_ports_link_status(); 769 } 770 771 return 0; 772 } 773 774 static int 775 init_clock(void) 776 { 777 rte_timer_subsystem_init(); 778 uint64_t hz = rte_get_timer_hz(); 779 uint64_t intrs = MS_PER_S/ff_global_cfg.freebsd.hz; 780 uint64_t tsc = (hz + MS_PER_S - 1) / MS_PER_S*intrs; 781 782 rte_timer_init(&freebsd_clock); 783 rte_timer_reset(&freebsd_clock, tsc, PERIODICAL, 784 rte_lcore_id(), &ff_hardclock_job, NULL); 785 786 ff_update_current_ts(); 787 788 return 0; 789 } 790 791 int 792 ff_dpdk_init(int argc, char **argv) 793 { 794 if (ff_global_cfg.dpdk.nb_procs < 1 || 795 ff_global_cfg.dpdk.nb_procs > RTE_MAX_LCORE || 796 ff_global_cfg.dpdk.proc_id >= ff_global_cfg.dpdk.nb_procs || 797 ff_global_cfg.dpdk.proc_id < 0) { 798 printf("param num_procs[%d] or proc_id[%d] error!\n", 799 ff_global_cfg.dpdk.nb_procs, 800 ff_global_cfg.dpdk.proc_id); 801 exit(1); 802 } 803 804 int ret = rte_eal_init(argc, argv); 805 if (ret < 0) { 806 rte_exit(EXIT_FAILURE, "Error with EAL initialization\n"); 807 } 808 809 numa_on = ff_global_cfg.dpdk.numa_on; 810 811 idle_sleep = ff_global_cfg.dpdk.idle_sleep; 812 pkt_tx_delay = ff_global_cfg.dpdk.pkt_tx_delay > BURST_TX_DRAIN_US ? \ 813 BURST_TX_DRAIN_US : ff_global_cfg.dpdk.pkt_tx_delay; 814 815 init_lcore_conf(); 816 817 init_mem_pool(); 818 819 init_dispatch_ring(); 820 821 init_msg_ring(); 822 823 #ifdef FF_KNI 824 enable_kni = ff_global_cfg.kni.enable; 825 if (enable_kni) { 826 init_kni(); 827 } 828 #endif 829 830 #ifdef FF_USE_PAGE_ARRAY 831 ff_mmap_init(); 832 #endif 833 834 ret = init_port_start(); 835 if (ret < 0) { 836 rte_exit(EXIT_FAILURE, "init_port_start failed\n"); 837 } 838 839 init_clock(); 840 841 return 0; 842 } 843 844 static void 845 ff_veth_input(const struct ff_dpdk_if_context *ctx, struct rte_mbuf *pkt) 846 { 847 uint8_t rx_csum = ctx->hw_features.rx_csum; 848 if (rx_csum) { 849 if (pkt->ol_flags & (PKT_RX_IP_CKSUM_BAD | PKT_RX_L4_CKSUM_BAD)) { 850 rte_pktmbuf_free(pkt); 851 return; 852 } 853 } 854 855 void *data = rte_pktmbuf_mtod(pkt, void*); 856 uint16_t len = rte_pktmbuf_data_len(pkt); 857 858 void *hdr = ff_mbuf_gethdr(pkt, pkt->pkt_len, data, len, rx_csum); 859 if (hdr == NULL) { 860 rte_pktmbuf_free(pkt); 861 return; 862 } 863 864 if (pkt->ol_flags & PKT_RX_VLAN_STRIPPED) { 865 ff_mbuf_set_vlan_info(hdr, pkt->vlan_tci); 866 } 867 868 struct rte_mbuf *pn = pkt->next; 869 void *prev = hdr; 870 while(pn != NULL) { 871 data = rte_pktmbuf_mtod(pn, void*); 872 len = rte_pktmbuf_data_len(pn); 873 874 void *mb = ff_mbuf_get(prev, data, len); 875 if (mb == NULL) { 876 ff_mbuf_free(hdr); 877 rte_pktmbuf_free(pkt); 878 return; 879 } 880 pn = pn->next; 881 prev = mb; 882 } 883 884 ff_veth_process_packet(ctx->ifp, hdr); 885 } 886 887 static enum FilterReturn 888 protocol_filter(const void *data, uint16_t len) 889 { 890 if(len < ETHER_HDR_LEN) 891 return FILTER_UNKNOWN; 892 893 const struct ether_hdr *hdr; 894 const struct vlan_hdr *vlanhdr; 895 hdr = (const struct ether_hdr *)data; 896 uint16_t ether_type = rte_be_to_cpu_16(hdr->ether_type); 897 data += ETHER_HDR_LEN; 898 len -= ETHER_HDR_LEN; 899 900 if (ether_type == ETHER_TYPE_VLAN) { 901 vlanhdr = (struct vlan_hdr *)data; 902 ether_type = rte_be_to_cpu_16(vlanhdr->eth_proto); 903 data += sizeof(struct vlan_hdr); 904 len -= sizeof(struct vlan_hdr); 905 } 906 907 if(ether_type == ETHER_TYPE_ARP) 908 return FILTER_ARP; 909 910 #ifdef INET6 911 if (ether_type == ETHER_TYPE_IPv6) { 912 return ff_kni_proto_filter(data, 913 len, ether_type); 914 } 915 #endif 916 917 #ifndef FF_KNI 918 return FILTER_UNKNOWN; 919 #else 920 if (!enable_kni) { 921 return FILTER_UNKNOWN; 922 } 923 924 if(ether_type != ETHER_TYPE_IPv4) 925 return FILTER_UNKNOWN; 926 927 return ff_kni_proto_filter(data, 928 len, ether_type); 929 #endif 930 } 931 932 static inline void 933 pktmbuf_deep_attach(struct rte_mbuf *mi, const struct rte_mbuf *m) 934 { 935 struct rte_mbuf *md; 936 void *src, *dst; 937 938 dst = rte_pktmbuf_mtod(mi, void *); 939 src = rte_pktmbuf_mtod(m, void *); 940 941 mi->data_len = m->data_len; 942 rte_memcpy(dst, src, m->data_len); 943 944 mi->port = m->port; 945 mi->vlan_tci = m->vlan_tci; 946 mi->vlan_tci_outer = m->vlan_tci_outer; 947 mi->tx_offload = m->tx_offload; 948 mi->hash = m->hash; 949 mi->ol_flags = m->ol_flags; 950 mi->packet_type = m->packet_type; 951 } 952 953 /* copied from rte_pktmbuf_clone */ 954 static inline struct rte_mbuf * 955 pktmbuf_deep_clone(const struct rte_mbuf *md, 956 struct rte_mempool *mp) 957 { 958 struct rte_mbuf *mc, *mi, **prev; 959 uint32_t pktlen; 960 uint8_t nseg; 961 962 if (unlikely ((mc = rte_pktmbuf_alloc(mp)) == NULL)) 963 return NULL; 964 965 mi = mc; 966 prev = &mi->next; 967 pktlen = md->pkt_len; 968 nseg = 0; 969 970 do { 971 nseg++; 972 pktmbuf_deep_attach(mi, md); 973 *prev = mi; 974 prev = &mi->next; 975 } while ((md = md->next) != NULL && 976 (mi = rte_pktmbuf_alloc(mp)) != NULL); 977 978 *prev = NULL; 979 mc->nb_segs = nseg; 980 mc->pkt_len = pktlen; 981 982 /* Allocation of new indirect segment failed */ 983 if (unlikely (mi == NULL)) { 984 rte_pktmbuf_free(mc); 985 return NULL; 986 } 987 988 __rte_mbuf_sanity_check(mc, 1); 989 return mc; 990 } 991 992 static inline void 993 process_packets(uint16_t port_id, uint16_t queue_id, struct rte_mbuf **bufs, 994 uint16_t count, const struct ff_dpdk_if_context *ctx, int pkts_from_ring) 995 { 996 struct lcore_conf *qconf = &lcore_conf; 997 uint16_t nb_queues = qconf->nb_queue_list[port_id]; 998 999 uint16_t i; 1000 for (i = 0; i < count; i++) { 1001 struct rte_mbuf *rtem = bufs[i]; 1002 1003 if (unlikely(qconf->pcap[port_id] != NULL)) { 1004 if (!pkts_from_ring) { 1005 ff_dump_packets(qconf->pcap[port_id], rtem); 1006 } 1007 } 1008 1009 void *data = rte_pktmbuf_mtod(rtem, void*); 1010 uint16_t len = rte_pktmbuf_data_len(rtem); 1011 1012 if (!pkts_from_ring) { 1013 ff_traffic.rx_packets++; 1014 ff_traffic.rx_bytes += len; 1015 } 1016 1017 if (!pkts_from_ring && packet_dispatcher) { 1018 int ret = (*packet_dispatcher)(data, &len, queue_id, nb_queues); 1019 if (ret == FF_DISPATCH_RESPONSE) { 1020 rte_pktmbuf_pkt_len(rtem) = rte_pktmbuf_data_len(rtem) = len; 1021 1022 /* 1023 * We have not support vlan out strip 1024 */ 1025 if (rtem->vlan_tci) { 1026 data = rte_pktmbuf_prepend(rtem, sizeof(struct vlan_hdr)); 1027 if (data != NULL) { 1028 memmove(data, data + sizeof(struct vlan_hdr), ETHER_HDR_LEN); 1029 struct ether_hdr *etherhdr = (struct ether_hdr *)data; 1030 struct vlan_hdr *vlanhdr = (struct vlan_hdr *)(data + ETHER_HDR_LEN); 1031 vlanhdr->vlan_tci = rte_cpu_to_be_16(rtem->vlan_tci); 1032 vlanhdr->eth_proto = etherhdr->ether_type; 1033 etherhdr->ether_type = rte_cpu_to_be_16(ETHER_TYPE_VLAN); 1034 } 1035 } 1036 send_single_packet(rtem, port_id); 1037 continue; 1038 } 1039 1040 if (ret == FF_DISPATCH_ERROR || ret >= nb_queues) { 1041 rte_pktmbuf_free(rtem); 1042 continue; 1043 } 1044 1045 if (ret != queue_id) { 1046 ret = rte_ring_enqueue(dispatch_ring[port_id][ret], rtem); 1047 if (ret < 0) 1048 rte_pktmbuf_free(rtem); 1049 1050 continue; 1051 } 1052 } 1053 1054 enum FilterReturn filter = protocol_filter(data, len); 1055 #ifdef INET6 1056 if (filter == FILTER_ARP || filter == FILTER_NDP) { 1057 #else 1058 if (filter == FILTER_ARP) { 1059 #endif 1060 struct rte_mempool *mbuf_pool; 1061 struct rte_mbuf *mbuf_clone; 1062 if (!pkts_from_ring) { 1063 uint16_t j; 1064 for(j = 0; j < nb_queues; ++j) { 1065 if(j == queue_id) 1066 continue; 1067 1068 unsigned socket_id = 0; 1069 if (numa_on) { 1070 uint16_t lcore_id = qconf->port_cfgs[port_id].lcore_list[j]; 1071 socket_id = rte_lcore_to_socket_id(lcore_id); 1072 } 1073 mbuf_pool = pktmbuf_pool[socket_id]; 1074 mbuf_clone = pktmbuf_deep_clone(rtem, mbuf_pool); 1075 if(mbuf_clone) { 1076 int ret = rte_ring_enqueue(dispatch_ring[port_id][j], 1077 mbuf_clone); 1078 if (ret < 0) 1079 rte_pktmbuf_free(mbuf_clone); 1080 } 1081 } 1082 } 1083 1084 #ifdef FF_KNI 1085 if (enable_kni && rte_eal_process_type() == RTE_PROC_PRIMARY) { 1086 mbuf_pool = pktmbuf_pool[qconf->socket_id]; 1087 mbuf_clone = pktmbuf_deep_clone(rtem, mbuf_pool); 1088 if(mbuf_clone) { 1089 ff_kni_enqueue(port_id, mbuf_clone); 1090 } 1091 } 1092 #endif 1093 ff_veth_input(ctx, rtem); 1094 #ifdef FF_KNI 1095 } else if (enable_kni && 1096 ((filter == FILTER_KNI && kni_accept) || 1097 (filter == FILTER_UNKNOWN && !kni_accept)) ) { 1098 ff_kni_enqueue(port_id, rtem); 1099 #endif 1100 } else { 1101 ff_veth_input(ctx, rtem); 1102 } 1103 } 1104 } 1105 1106 static inline int 1107 process_dispatch_ring(uint16_t port_id, uint16_t queue_id, 1108 struct rte_mbuf **pkts_burst, const struct ff_dpdk_if_context *ctx) 1109 { 1110 /* read packet from ring buf and to process */ 1111 uint16_t nb_rb; 1112 nb_rb = rte_ring_dequeue_burst(dispatch_ring[port_id][queue_id], 1113 (void **)pkts_burst, MAX_PKT_BURST, NULL); 1114 1115 if(nb_rb > 0) { 1116 process_packets(port_id, queue_id, pkts_burst, nb_rb, ctx, 1); 1117 } 1118 1119 return 0; 1120 } 1121 1122 static inline void 1123 handle_sysctl_msg(struct ff_msg *msg) 1124 { 1125 int ret = ff_sysctl(msg->sysctl.name, msg->sysctl.namelen, 1126 msg->sysctl.old, msg->sysctl.oldlenp, msg->sysctl.new, 1127 msg->sysctl.newlen); 1128 1129 if (ret < 0) { 1130 msg->result = errno; 1131 } else { 1132 msg->result = 0; 1133 } 1134 } 1135 1136 static inline void 1137 handle_ioctl_msg(struct ff_msg *msg) 1138 { 1139 int fd, ret; 1140 #ifdef INET6 1141 if (msg->msg_type == FF_IOCTL6) { 1142 fd = ff_socket(AF_INET6, SOCK_DGRAM, 0); 1143 } else 1144 #endif 1145 fd = ff_socket(AF_INET, SOCK_DGRAM, 0); 1146 1147 if (fd < 0) { 1148 ret = -1; 1149 goto done; 1150 } 1151 1152 ret = ff_ioctl_freebsd(fd, msg->ioctl.cmd, msg->ioctl.data); 1153 1154 ff_close(fd); 1155 1156 done: 1157 if (ret < 0) { 1158 msg->result = errno; 1159 } else { 1160 msg->result = 0; 1161 } 1162 } 1163 1164 static inline void 1165 handle_route_msg(struct ff_msg *msg) 1166 { 1167 int ret = ff_rtioctl(msg->route.fib, msg->route.data, 1168 &msg->route.len, msg->route.maxlen); 1169 if (ret < 0) { 1170 msg->result = errno; 1171 } else { 1172 msg->result = 0; 1173 } 1174 } 1175 1176 static inline void 1177 handle_top_msg(struct ff_msg *msg) 1178 { 1179 msg->top = ff_top_status; 1180 msg->result = 0; 1181 } 1182 1183 #ifdef FF_NETGRAPH 1184 static inline void 1185 handle_ngctl_msg(struct ff_msg *msg) 1186 { 1187 int ret = ff_ngctl(msg->ngctl.cmd, msg->ngctl.data); 1188 if (ret < 0) { 1189 msg->result = errno; 1190 } else { 1191 msg->result = 0; 1192 msg->ngctl.ret = ret; 1193 } 1194 } 1195 #endif 1196 1197 #ifdef FF_IPFW 1198 static inline void 1199 handle_ipfw_msg(struct ff_msg *msg) 1200 { 1201 int fd, ret; 1202 fd = ff_socket(AF_INET, SOCK_RAW, IPPROTO_RAW); 1203 if (fd < 0) { 1204 ret = -1; 1205 goto done; 1206 } 1207 1208 switch (msg->ipfw.cmd) { 1209 case FF_IPFW_GET: 1210 ret = ff_getsockopt_freebsd(fd, msg->ipfw.level, 1211 msg->ipfw.optname, msg->ipfw.optval, 1212 msg->ipfw.optlen); 1213 break; 1214 case FF_IPFW_SET: 1215 ret = ff_setsockopt_freebsd(fd, msg->ipfw.level, 1216 msg->ipfw.optname, msg->ipfw.optval, 1217 *(msg->ipfw.optlen)); 1218 break; 1219 default: 1220 ret = -1; 1221 errno = ENOTSUP; 1222 break; 1223 } 1224 1225 ff_close(fd); 1226 1227 done: 1228 if (ret < 0) { 1229 msg->result = errno; 1230 } else { 1231 msg->result = 0; 1232 } 1233 } 1234 #endif 1235 1236 static inline void 1237 handle_traffic_msg(struct ff_msg *msg) 1238 { 1239 msg->traffic = ff_traffic; 1240 msg->result = 0; 1241 } 1242 1243 static inline void 1244 handle_default_msg(struct ff_msg *msg) 1245 { 1246 msg->result = ENOTSUP; 1247 } 1248 1249 static inline void 1250 handle_msg(struct ff_msg *msg, uint16_t proc_id) 1251 { 1252 switch (msg->msg_type) { 1253 case FF_SYSCTL: 1254 handle_sysctl_msg(msg); 1255 break; 1256 case FF_IOCTL: 1257 #ifdef INET6 1258 case FF_IOCTL6: 1259 #endif 1260 handle_ioctl_msg(msg); 1261 break; 1262 case FF_ROUTE: 1263 handle_route_msg(msg); 1264 break; 1265 case FF_TOP: 1266 handle_top_msg(msg); 1267 break; 1268 #ifdef FF_NETGRAPH 1269 case FF_NGCTL: 1270 handle_ngctl_msg(msg); 1271 break; 1272 #endif 1273 #ifdef FF_IPFW 1274 case FF_IPFW_CTL: 1275 handle_ipfw_msg(msg); 1276 break; 1277 #endif 1278 case FF_TRAFFIC: 1279 handle_traffic_msg(msg); 1280 break; 1281 default: 1282 handle_default_msg(msg); 1283 break; 1284 } 1285 rte_ring_enqueue(msg_ring[proc_id].ring[msg->msg_type], msg); 1286 } 1287 1288 static inline int 1289 process_msg_ring(uint16_t proc_id) 1290 { 1291 void *msg; 1292 int ret = rte_ring_dequeue(msg_ring[proc_id].ring[0], &msg); 1293 1294 if (unlikely(ret == 0)) { 1295 handle_msg((struct ff_msg *)msg, proc_id); 1296 } 1297 1298 return 0; 1299 } 1300 1301 /* Send burst of packets on an output interface */ 1302 static inline int 1303 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port) 1304 { 1305 struct rte_mbuf **m_table; 1306 int ret; 1307 uint16_t queueid; 1308 1309 queueid = qconf->tx_queue_id[port]; 1310 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table; 1311 1312 if (unlikely(qconf->pcap[port] != NULL)) { 1313 uint16_t i; 1314 for (i = 0; i < n; i++) { 1315 ff_dump_packets(qconf->pcap[port], m_table[i]); 1316 } 1317 } 1318 1319 ret = rte_eth_tx_burst(port, queueid, m_table, n); 1320 ff_traffic.tx_packets += ret; 1321 uint16_t i; 1322 for (i = 0; i < ret; i++) { 1323 ff_traffic.tx_bytes += rte_pktmbuf_pkt_len(m_table[i]); 1324 #ifdef FF_USE_PAGE_ARRAY 1325 if (qconf->tx_mbufs[port].bsd_m_table[i]) 1326 ff_enq_tx_bsdmbuf(port, qconf->tx_mbufs[port].bsd_m_table[i], m_table[i]->nb_segs); 1327 #endif 1328 } 1329 if (unlikely(ret < n)) { 1330 do { 1331 rte_pktmbuf_free(m_table[ret]); 1332 #ifdef FF_USE_PAGE_ARRAY 1333 if ( qconf->tx_mbufs[port].bsd_m_table[ret] ) 1334 ff_mbuf_free(qconf->tx_mbufs[port].bsd_m_table[ret]); 1335 #endif 1336 } while (++ret < n); 1337 } 1338 return 0; 1339 } 1340 1341 /* Enqueue a single packet, and send burst if queue is filled */ 1342 static inline int 1343 send_single_packet(struct rte_mbuf *m, uint8_t port) 1344 { 1345 uint16_t len; 1346 struct lcore_conf *qconf; 1347 1348 qconf = &lcore_conf; 1349 len = qconf->tx_mbufs[port].len; 1350 qconf->tx_mbufs[port].m_table[len] = m; 1351 len++; 1352 1353 /* enough pkts to be sent */ 1354 if (unlikely(len == MAX_PKT_BURST)) { 1355 send_burst(qconf, MAX_PKT_BURST, port); 1356 len = 0; 1357 } 1358 1359 qconf->tx_mbufs[port].len = len; 1360 return 0; 1361 } 1362 1363 int 1364 ff_dpdk_if_send(struct ff_dpdk_if_context *ctx, void *m, 1365 int total) 1366 { 1367 #ifdef FF_USE_PAGE_ARRAY 1368 struct lcore_conf *qconf = &lcore_conf; 1369 int len = 0; 1370 1371 len = ff_if_send_onepkt(ctx, m,total); 1372 if (unlikely(len == MAX_PKT_BURST)) { 1373 send_burst(qconf, MAX_PKT_BURST, ctx->port_id); 1374 len = 0; 1375 } 1376 qconf->tx_mbufs[ctx->port_id].len = len; 1377 return 0; 1378 #endif 1379 struct rte_mempool *mbuf_pool = pktmbuf_pool[lcore_conf.socket_id]; 1380 struct rte_mbuf *head = rte_pktmbuf_alloc(mbuf_pool); 1381 if (head == NULL) { 1382 ff_mbuf_free(m); 1383 return -1; 1384 } 1385 1386 head->pkt_len = total; 1387 head->nb_segs = 0; 1388 1389 int off = 0; 1390 struct rte_mbuf *cur = head, *prev = NULL; 1391 while(total > 0) { 1392 if (cur == NULL) { 1393 cur = rte_pktmbuf_alloc(mbuf_pool); 1394 if (cur == NULL) { 1395 rte_pktmbuf_free(head); 1396 ff_mbuf_free(m); 1397 return -1; 1398 } 1399 } 1400 1401 if (prev != NULL) { 1402 prev->next = cur; 1403 } 1404 head->nb_segs++; 1405 1406 prev = cur; 1407 void *data = rte_pktmbuf_mtod(cur, void*); 1408 int len = total > RTE_MBUF_DEFAULT_DATAROOM ? RTE_MBUF_DEFAULT_DATAROOM : total; 1409 int ret = ff_mbuf_copydata(m, data, off, len); 1410 if (ret < 0) { 1411 rte_pktmbuf_free(head); 1412 ff_mbuf_free(m); 1413 return -1; 1414 } 1415 1416 1417 cur->data_len = len; 1418 off += len; 1419 total -= len; 1420 cur = NULL; 1421 } 1422 1423 struct ff_tx_offload offload = {0}; 1424 ff_mbuf_tx_offload(m, &offload); 1425 1426 void *data = rte_pktmbuf_mtod(head, void*); 1427 1428 if (offload.ip_csum) { 1429 /* ipv6 not supported yet */ 1430 struct ipv4_hdr *iph; 1431 int iph_len; 1432 iph = (struct ipv4_hdr *)(data + ETHER_HDR_LEN); 1433 iph_len = (iph->version_ihl & 0x0f) << 2; 1434 1435 head->ol_flags |= PKT_TX_IP_CKSUM | PKT_TX_IPV4; 1436 head->l2_len = ETHER_HDR_LEN; 1437 head->l3_len = iph_len; 1438 } 1439 1440 if (ctx->hw_features.tx_csum_l4) { 1441 struct ipv4_hdr *iph; 1442 int iph_len; 1443 iph = (struct ipv4_hdr *)(data + ETHER_HDR_LEN); 1444 iph_len = (iph->version_ihl & 0x0f) << 2; 1445 1446 if (offload.tcp_csum) { 1447 head->ol_flags |= PKT_TX_TCP_CKSUM; 1448 head->l2_len = ETHER_HDR_LEN; 1449 head->l3_len = iph_len; 1450 } 1451 1452 /* 1453 * TCP segmentation offload. 1454 * 1455 * - set the PKT_TX_TCP_SEG flag in mbuf->ol_flags (this flag 1456 * implies PKT_TX_TCP_CKSUM) 1457 * - set the flag PKT_TX_IPV4 or PKT_TX_IPV6 1458 * - if it's IPv4, set the PKT_TX_IP_CKSUM flag and 1459 * write the IP checksum to 0 in the packet 1460 * - fill the mbuf offload information: l2_len, 1461 * l3_len, l4_len, tso_segsz 1462 * - calculate the pseudo header checksum without taking ip_len 1463 * in account, and set it in the TCP header. Refer to 1464 * rte_ipv4_phdr_cksum() and rte_ipv6_phdr_cksum() that can be 1465 * used as helpers. 1466 */ 1467 if (offload.tso_seg_size) { 1468 struct tcp_hdr *tcph; 1469 int tcph_len; 1470 tcph = (struct tcp_hdr *)((char *)iph + iph_len); 1471 tcph_len = (tcph->data_off & 0xf0) >> 2; 1472 tcph->cksum = rte_ipv4_phdr_cksum(iph, PKT_TX_TCP_SEG); 1473 1474 head->ol_flags |= PKT_TX_TCP_SEG; 1475 head->l4_len = tcph_len; 1476 head->tso_segsz = offload.tso_seg_size; 1477 } 1478 1479 if (offload.udp_csum) { 1480 head->ol_flags |= PKT_TX_UDP_CKSUM; 1481 head->l2_len = ETHER_HDR_LEN; 1482 head->l3_len = iph_len; 1483 } 1484 } 1485 1486 ff_mbuf_free(m); 1487 1488 return send_single_packet(head, ctx->port_id); 1489 } 1490 1491 static int 1492 main_loop(void *arg) 1493 { 1494 struct loop_routine *lr = (struct loop_routine *)arg; 1495 1496 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 1497 uint64_t prev_tsc, diff_tsc, cur_tsc, usch_tsc, div_tsc, usr_tsc, sys_tsc, end_tsc, idle_sleep_tsc; 1498 int i, j, nb_rx, idle; 1499 uint16_t port_id, queue_id; 1500 struct lcore_conf *qconf; 1501 uint64_t drain_tsc = 0; 1502 struct ff_dpdk_if_context *ctx; 1503 1504 if (pkt_tx_delay) { 1505 drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * pkt_tx_delay; 1506 } 1507 1508 prev_tsc = 0; 1509 usch_tsc = 0; 1510 1511 qconf = &lcore_conf; 1512 1513 while (1) { 1514 cur_tsc = rte_rdtsc(); 1515 if (unlikely(freebsd_clock.expire < cur_tsc)) { 1516 rte_timer_manage(); 1517 } 1518 1519 idle = 1; 1520 sys_tsc = 0; 1521 usr_tsc = 0; 1522 1523 /* 1524 * TX burst queue drain 1525 */ 1526 diff_tsc = cur_tsc - prev_tsc; 1527 if (unlikely(diff_tsc >= drain_tsc)) { 1528 for (i = 0; i < qconf->nb_tx_port; i++) { 1529 port_id = qconf->tx_port_id[i]; 1530 if (qconf->tx_mbufs[port_id].len == 0) 1531 continue; 1532 1533 idle = 0; 1534 1535 send_burst(qconf, 1536 qconf->tx_mbufs[port_id].len, 1537 port_id); 1538 qconf->tx_mbufs[port_id].len = 0; 1539 } 1540 1541 prev_tsc = cur_tsc; 1542 } 1543 1544 /* 1545 * Read packet from RX queues 1546 */ 1547 for (i = 0; i < qconf->nb_rx_queue; ++i) { 1548 port_id = qconf->rx_queue_list[i].port_id; 1549 queue_id = qconf->rx_queue_list[i].queue_id; 1550 ctx = veth_ctx[port_id]; 1551 1552 #ifdef FF_KNI 1553 if (enable_kni && rte_eal_process_type() == RTE_PROC_PRIMARY) { 1554 ff_kni_process(port_id, queue_id, pkts_burst, MAX_PKT_BURST); 1555 } 1556 #endif 1557 1558 process_dispatch_ring(port_id, queue_id, pkts_burst, ctx); 1559 1560 nb_rx = rte_eth_rx_burst(port_id, queue_id, pkts_burst, 1561 MAX_PKT_BURST); 1562 if (nb_rx == 0) 1563 continue; 1564 1565 idle = 0; 1566 1567 /* Prefetch first packets */ 1568 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) { 1569 rte_prefetch0(rte_pktmbuf_mtod( 1570 pkts_burst[j], void *)); 1571 } 1572 1573 /* Prefetch and handle already prefetched packets */ 1574 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) { 1575 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[ 1576 j + PREFETCH_OFFSET], void *)); 1577 process_packets(port_id, queue_id, &pkts_burst[j], 1, ctx, 0); 1578 } 1579 1580 /* Handle remaining prefetched packets */ 1581 for (; j < nb_rx; j++) { 1582 process_packets(port_id, queue_id, &pkts_burst[j], 1, ctx, 0); 1583 } 1584 } 1585 1586 process_msg_ring(qconf->proc_id); 1587 1588 div_tsc = rte_rdtsc(); 1589 1590 if (likely(lr->loop != NULL && (!idle || cur_tsc - usch_tsc >= drain_tsc))) { 1591 usch_tsc = cur_tsc; 1592 lr->loop(lr->arg); 1593 } 1594 1595 idle_sleep_tsc = rte_rdtsc(); 1596 if (likely(idle && idle_sleep)) { 1597 usleep(idle_sleep); 1598 end_tsc = rte_rdtsc(); 1599 } else { 1600 end_tsc = idle_sleep_tsc; 1601 } 1602 1603 if (usch_tsc == cur_tsc) { 1604 usr_tsc = idle_sleep_tsc - div_tsc; 1605 } 1606 1607 if (!idle) { 1608 sys_tsc = div_tsc - cur_tsc; 1609 ff_top_status.sys_tsc += sys_tsc; 1610 } 1611 1612 ff_top_status.usr_tsc += usr_tsc; 1613 ff_top_status.work_tsc += end_tsc - cur_tsc; 1614 ff_top_status.idle_tsc += end_tsc - cur_tsc - usr_tsc - sys_tsc; 1615 1616 ff_top_status.loops++; 1617 } 1618 1619 return 0; 1620 } 1621 1622 int 1623 ff_dpdk_if_up(void) { 1624 int i; 1625 struct lcore_conf *qconf = &lcore_conf; 1626 for (i = 0; i < qconf->nb_tx_port; i++) { 1627 uint16_t port_id = qconf->tx_port_id[i]; 1628 1629 struct ff_port_cfg *pconf = &qconf->port_cfgs[port_id]; 1630 veth_ctx[port_id] = ff_veth_attach(pconf); 1631 if (veth_ctx[port_id] == NULL) { 1632 rte_exit(EXIT_FAILURE, "ff_veth_attach failed"); 1633 } 1634 } 1635 1636 return 0; 1637 } 1638 1639 void 1640 ff_dpdk_run(loop_func_t loop, void *arg) { 1641 struct loop_routine *lr = rte_malloc(NULL, 1642 sizeof(struct loop_routine), 0); 1643 lr->loop = loop; 1644 lr->arg = arg; 1645 rte_eal_mp_remote_launch(main_loop, lr, CALL_MASTER); 1646 rte_eal_mp_wait_lcore(); 1647 rte_free(lr); 1648 } 1649 1650 void 1651 ff_dpdk_pktmbuf_free(void *m) 1652 { 1653 rte_pktmbuf_free((struct rte_mbuf *)m); 1654 } 1655 1656 static uint32_t 1657 toeplitz_hash(unsigned keylen, const uint8_t *key, 1658 unsigned datalen, const uint8_t *data) 1659 { 1660 uint32_t hash = 0, v; 1661 u_int i, b; 1662 1663 /* XXXRW: Perhaps an assertion about key length vs. data length? */ 1664 1665 v = (key[0]<<24) + (key[1]<<16) + (key[2] <<8) + key[3]; 1666 for (i = 0; i < datalen; i++) { 1667 for (b = 0; b < 8; b++) { 1668 if (data[i] & (1<<(7-b))) 1669 hash ^= v; 1670 v <<= 1; 1671 if ((i + 4) < keylen && 1672 (key[i+4] & (1<<(7-b)))) 1673 v |= 1; 1674 } 1675 } 1676 return (hash); 1677 } 1678 1679 int 1680 ff_rss_check(void *softc, uint32_t saddr, uint32_t daddr, 1681 uint16_t sport, uint16_t dport) 1682 { 1683 struct lcore_conf *qconf = &lcore_conf; 1684 struct ff_dpdk_if_context *ctx = ff_veth_softc_to_hostc(softc); 1685 uint16_t nb_queues = qconf->nb_queue_list[ctx->port_id]; 1686 1687 if (nb_queues <= 1) { 1688 return 1; 1689 } 1690 1691 uint16_t reta_size = rss_reta_size[ctx->port_id]; 1692 uint16_t queueid = qconf->tx_queue_id[ctx->port_id]; 1693 1694 uint8_t data[sizeof(saddr) + sizeof(daddr) + sizeof(sport) + 1695 sizeof(dport)]; 1696 1697 unsigned datalen = 0; 1698 1699 bcopy(&saddr, &data[datalen], sizeof(saddr)); 1700 datalen += sizeof(saddr); 1701 1702 bcopy(&daddr, &data[datalen], sizeof(daddr)); 1703 datalen += sizeof(daddr); 1704 1705 bcopy(&sport, &data[datalen], sizeof(sport)); 1706 datalen += sizeof(sport); 1707 1708 bcopy(&dport, &data[datalen], sizeof(dport)); 1709 datalen += sizeof(dport); 1710 1711 uint32_t hash = 0; 1712 if ( !use_rsskey_52bytes ) 1713 hash = toeplitz_hash(sizeof(default_rsskey_40bytes), 1714 default_rsskey_40bytes, datalen, data); 1715 else 1716 hash = toeplitz_hash(sizeof(default_rsskey_52bytes), 1717 default_rsskey_52bytes, datalen, data); 1718 return ((hash & (reta_size - 1)) % nb_queues) == queueid; 1719 } 1720 1721 void 1722 ff_regist_packet_dispatcher(dispatch_func_t func) 1723 { 1724 packet_dispatcher = func; 1725 } 1726 1727 uint64_t 1728 ff_get_tsc_ns() 1729 { 1730 uint64_t cur_tsc = rte_rdtsc(); 1731 uint64_t hz = rte_get_tsc_hz(); 1732 return ((double)cur_tsc/(double)hz) * NS_PER_S; 1733 } 1734 1735