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