1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright(c) 2010-2015 Intel Corporation 3 */ 4 5 #include "rte_eth_ring.h" 6 #include <rte_mbuf.h> 7 #include <rte_ethdev_driver.h> 8 #include <rte_malloc.h> 9 #include <rte_memcpy.h> 10 #include <rte_string_fns.h> 11 #include <rte_bus_vdev.h> 12 #include <rte_kvargs.h> 13 #include <rte_errno.h> 14 15 #define ETH_RING_NUMA_NODE_ACTION_ARG "nodeaction" 16 #define ETH_RING_ACTION_CREATE "CREATE" 17 #define ETH_RING_ACTION_ATTACH "ATTACH" 18 #define ETH_RING_INTERNAL_ARG "internal" 19 20 static const char *valid_arguments[] = { 21 ETH_RING_NUMA_NODE_ACTION_ARG, 22 ETH_RING_INTERNAL_ARG, 23 NULL 24 }; 25 26 struct ring_internal_args { 27 struct rte_ring * const *rx_queues; 28 const unsigned int nb_rx_queues; 29 struct rte_ring * const *tx_queues; 30 const unsigned int nb_tx_queues; 31 const unsigned int numa_node; 32 void *addr; /* self addr for sanity check */ 33 }; 34 35 enum dev_action { 36 DEV_CREATE, 37 DEV_ATTACH 38 }; 39 40 struct ring_queue { 41 struct rte_ring *rng; 42 rte_atomic64_t rx_pkts; 43 rte_atomic64_t tx_pkts; 44 rte_atomic64_t err_pkts; 45 }; 46 47 struct pmd_internals { 48 unsigned int max_rx_queues; 49 unsigned int max_tx_queues; 50 51 struct ring_queue rx_ring_queues[RTE_PMD_RING_MAX_RX_RINGS]; 52 struct ring_queue tx_ring_queues[RTE_PMD_RING_MAX_TX_RINGS]; 53 54 struct ether_addr address; 55 enum dev_action action; 56 }; 57 58 static struct rte_eth_link pmd_link = { 59 .link_speed = ETH_SPEED_NUM_10G, 60 .link_duplex = ETH_LINK_FULL_DUPLEX, 61 .link_status = ETH_LINK_DOWN, 62 .link_autoneg = ETH_LINK_FIXED, 63 }; 64 65 static int eth_ring_logtype; 66 67 #define PMD_LOG(level, fmt, args...) \ 68 rte_log(RTE_LOG_ ## level, eth_ring_logtype, \ 69 "%s(): " fmt "\n", __func__, ##args) 70 71 static uint16_t 72 eth_ring_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs) 73 { 74 void **ptrs = (void *)&bufs[0]; 75 struct ring_queue *r = q; 76 const uint16_t nb_rx = (uint16_t)rte_ring_dequeue_burst(r->rng, 77 ptrs, nb_bufs, NULL); 78 if (r->rng->flags & RING_F_SC_DEQ) 79 r->rx_pkts.cnt += nb_rx; 80 else 81 rte_atomic64_add(&(r->rx_pkts), nb_rx); 82 return nb_rx; 83 } 84 85 static uint16_t 86 eth_ring_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs) 87 { 88 void **ptrs = (void *)&bufs[0]; 89 struct ring_queue *r = q; 90 const uint16_t nb_tx = (uint16_t)rte_ring_enqueue_burst(r->rng, 91 ptrs, nb_bufs, NULL); 92 if (r->rng->flags & RING_F_SP_ENQ) { 93 r->tx_pkts.cnt += nb_tx; 94 r->err_pkts.cnt += nb_bufs - nb_tx; 95 } else { 96 rte_atomic64_add(&(r->tx_pkts), nb_tx); 97 rte_atomic64_add(&(r->err_pkts), nb_bufs - nb_tx); 98 } 99 return nb_tx; 100 } 101 102 static int 103 eth_dev_configure(struct rte_eth_dev *dev __rte_unused) { return 0; } 104 105 static int 106 eth_dev_start(struct rte_eth_dev *dev) 107 { 108 dev->data->dev_link.link_status = ETH_LINK_UP; 109 return 0; 110 } 111 112 static void 113 eth_dev_stop(struct rte_eth_dev *dev) 114 { 115 dev->data->dev_link.link_status = ETH_LINK_DOWN; 116 } 117 118 static int 119 eth_dev_set_link_down(struct rte_eth_dev *dev) 120 { 121 dev->data->dev_link.link_status = ETH_LINK_DOWN; 122 return 0; 123 } 124 125 static int 126 eth_dev_set_link_up(struct rte_eth_dev *dev) 127 { 128 dev->data->dev_link.link_status = ETH_LINK_UP; 129 return 0; 130 } 131 132 static int 133 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id, 134 uint16_t nb_rx_desc __rte_unused, 135 unsigned int socket_id __rte_unused, 136 const struct rte_eth_rxconf *rx_conf __rte_unused, 137 struct rte_mempool *mb_pool __rte_unused) 138 { 139 struct pmd_internals *internals = dev->data->dev_private; 140 141 dev->data->rx_queues[rx_queue_id] = &internals->rx_ring_queues[rx_queue_id]; 142 return 0; 143 } 144 145 static int 146 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id, 147 uint16_t nb_tx_desc __rte_unused, 148 unsigned int socket_id __rte_unused, 149 const struct rte_eth_txconf *tx_conf __rte_unused) 150 { 151 struct pmd_internals *internals = dev->data->dev_private; 152 153 dev->data->tx_queues[tx_queue_id] = &internals->tx_ring_queues[tx_queue_id]; 154 return 0; 155 } 156 157 158 static void 159 eth_dev_info(struct rte_eth_dev *dev, 160 struct rte_eth_dev_info *dev_info) 161 { 162 struct pmd_internals *internals = dev->data->dev_private; 163 164 dev_info->max_mac_addrs = 1; 165 dev_info->max_rx_pktlen = (uint32_t)-1; 166 dev_info->max_rx_queues = (uint16_t)internals->max_rx_queues; 167 dev_info->max_tx_queues = (uint16_t)internals->max_tx_queues; 168 dev_info->min_rx_bufsize = 0; 169 } 170 171 static int 172 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats) 173 { 174 unsigned int i; 175 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0; 176 const struct pmd_internals *internal = dev->data->dev_private; 177 178 for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS && 179 i < dev->data->nb_rx_queues; i++) { 180 stats->q_ipackets[i] = internal->rx_ring_queues[i].rx_pkts.cnt; 181 rx_total += stats->q_ipackets[i]; 182 } 183 184 for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS && 185 i < dev->data->nb_tx_queues; i++) { 186 stats->q_opackets[i] = internal->tx_ring_queues[i].tx_pkts.cnt; 187 stats->q_errors[i] = internal->tx_ring_queues[i].err_pkts.cnt; 188 tx_total += stats->q_opackets[i]; 189 tx_err_total += stats->q_errors[i]; 190 } 191 192 stats->ipackets = rx_total; 193 stats->opackets = tx_total; 194 stats->oerrors = tx_err_total; 195 196 return 0; 197 } 198 199 static void 200 eth_stats_reset(struct rte_eth_dev *dev) 201 { 202 unsigned int i; 203 struct pmd_internals *internal = dev->data->dev_private; 204 205 for (i = 0; i < dev->data->nb_rx_queues; i++) 206 internal->rx_ring_queues[i].rx_pkts.cnt = 0; 207 for (i = 0; i < dev->data->nb_tx_queues; i++) { 208 internal->tx_ring_queues[i].tx_pkts.cnt = 0; 209 internal->tx_ring_queues[i].err_pkts.cnt = 0; 210 } 211 } 212 213 static void 214 eth_mac_addr_remove(struct rte_eth_dev *dev __rte_unused, 215 uint32_t index __rte_unused) 216 { 217 } 218 219 static int 220 eth_mac_addr_add(struct rte_eth_dev *dev __rte_unused, 221 struct ether_addr *mac_addr __rte_unused, 222 uint32_t index __rte_unused, 223 uint32_t vmdq __rte_unused) 224 { 225 return 0; 226 } 227 228 static void 229 eth_queue_release(void *q __rte_unused) { ; } 230 static int 231 eth_link_update(struct rte_eth_dev *dev __rte_unused, 232 int wait_to_complete __rte_unused) { return 0; } 233 234 static const struct eth_dev_ops ops = { 235 .dev_start = eth_dev_start, 236 .dev_stop = eth_dev_stop, 237 .dev_set_link_up = eth_dev_set_link_up, 238 .dev_set_link_down = eth_dev_set_link_down, 239 .dev_configure = eth_dev_configure, 240 .dev_infos_get = eth_dev_info, 241 .rx_queue_setup = eth_rx_queue_setup, 242 .tx_queue_setup = eth_tx_queue_setup, 243 .rx_queue_release = eth_queue_release, 244 .tx_queue_release = eth_queue_release, 245 .link_update = eth_link_update, 246 .stats_get = eth_stats_get, 247 .stats_reset = eth_stats_reset, 248 .mac_addr_remove = eth_mac_addr_remove, 249 .mac_addr_add = eth_mac_addr_add, 250 }; 251 252 static struct rte_vdev_driver pmd_ring_drv; 253 254 static int 255 do_eth_dev_ring_create(const char *name, 256 struct rte_ring * const rx_queues[], 257 const unsigned int nb_rx_queues, 258 struct rte_ring *const tx_queues[], 259 const unsigned int nb_tx_queues, 260 const unsigned int numa_node, enum dev_action action, 261 struct rte_eth_dev **eth_dev_p) 262 { 263 struct rte_eth_dev_data *data = NULL; 264 struct pmd_internals *internals = NULL; 265 struct rte_eth_dev *eth_dev = NULL; 266 void **rx_queues_local = NULL; 267 void **tx_queues_local = NULL; 268 unsigned int i; 269 270 PMD_LOG(INFO, "Creating rings-backed ethdev on numa socket %u", 271 numa_node); 272 273 rx_queues_local = rte_calloc_socket(name, nb_rx_queues, 274 sizeof(void *), 0, numa_node); 275 if (rx_queues_local == NULL) { 276 rte_errno = ENOMEM; 277 goto error; 278 } 279 280 tx_queues_local = rte_calloc_socket(name, nb_tx_queues, 281 sizeof(void *), 0, numa_node); 282 if (tx_queues_local == NULL) { 283 rte_errno = ENOMEM; 284 goto error; 285 } 286 287 internals = rte_zmalloc_socket(name, sizeof(*internals), 0, numa_node); 288 if (internals == NULL) { 289 rte_errno = ENOMEM; 290 goto error; 291 } 292 293 /* reserve an ethdev entry */ 294 eth_dev = rte_eth_dev_allocate(name); 295 if (eth_dev == NULL) { 296 rte_errno = ENOSPC; 297 goto error; 298 } 299 300 /* now put it all together 301 * - store queue data in internals, 302 * - store numa_node info in eth_dev_data 303 * - point eth_dev_data to internals 304 * - and point eth_dev structure to new eth_dev_data structure 305 */ 306 307 data = eth_dev->data; 308 data->rx_queues = rx_queues_local; 309 data->tx_queues = tx_queues_local; 310 311 internals->action = action; 312 internals->max_rx_queues = nb_rx_queues; 313 internals->max_tx_queues = nb_tx_queues; 314 for (i = 0; i < nb_rx_queues; i++) { 315 internals->rx_ring_queues[i].rng = rx_queues[i]; 316 data->rx_queues[i] = &internals->rx_ring_queues[i]; 317 } 318 for (i = 0; i < nb_tx_queues; i++) { 319 internals->tx_ring_queues[i].rng = tx_queues[i]; 320 data->tx_queues[i] = &internals->tx_ring_queues[i]; 321 } 322 323 data->dev_private = internals; 324 data->nb_rx_queues = (uint16_t)nb_rx_queues; 325 data->nb_tx_queues = (uint16_t)nb_tx_queues; 326 data->dev_link = pmd_link; 327 data->mac_addrs = &internals->address; 328 329 eth_dev->dev_ops = &ops; 330 data->kdrv = RTE_KDRV_NONE; 331 data->numa_node = numa_node; 332 333 /* finally assign rx and tx ops */ 334 eth_dev->rx_pkt_burst = eth_ring_rx; 335 eth_dev->tx_pkt_burst = eth_ring_tx; 336 337 rte_eth_dev_probing_finish(eth_dev); 338 *eth_dev_p = eth_dev; 339 340 return data->port_id; 341 342 error: 343 rte_free(rx_queues_local); 344 rte_free(tx_queues_local); 345 rte_free(internals); 346 347 return -1; 348 } 349 350 int 351 rte_eth_from_rings(const char *name, struct rte_ring *const rx_queues[], 352 const unsigned int nb_rx_queues, 353 struct rte_ring *const tx_queues[], 354 const unsigned int nb_tx_queues, 355 const unsigned int numa_node) 356 { 357 struct ring_internal_args args = { 358 .rx_queues = rx_queues, 359 .nb_rx_queues = nb_rx_queues, 360 .tx_queues = tx_queues, 361 .nb_tx_queues = nb_tx_queues, 362 .numa_node = numa_node, 363 .addr = &args, 364 }; 365 char args_str[32]; 366 char ring_name[RTE_RING_NAMESIZE]; 367 uint16_t port_id = RTE_MAX_ETHPORTS; 368 int ret; 369 370 /* do some parameter checking */ 371 if (rx_queues == NULL && nb_rx_queues > 0) { 372 rte_errno = EINVAL; 373 return -1; 374 } 375 if (tx_queues == NULL && nb_tx_queues > 0) { 376 rte_errno = EINVAL; 377 return -1; 378 } 379 if (nb_rx_queues > RTE_PMD_RING_MAX_RX_RINGS) { 380 rte_errno = EINVAL; 381 return -1; 382 } 383 384 snprintf(args_str, sizeof(args_str), "%s=%p", 385 ETH_RING_INTERNAL_ARG, &args); 386 387 ret = snprintf(ring_name, sizeof(ring_name), "net_ring_%s", name); 388 if (ret >= (int)sizeof(ring_name)) { 389 rte_errno = ENAMETOOLONG; 390 return -1; 391 } 392 393 ret = rte_vdev_init(ring_name, args_str); 394 if (ret) { 395 rte_errno = EINVAL; 396 return -1; 397 } 398 399 ret = rte_eth_dev_get_port_by_name(ring_name, &port_id); 400 if (ret) { 401 rte_errno = ENODEV; 402 return -1; 403 } 404 405 return port_id; 406 } 407 408 int 409 rte_eth_from_ring(struct rte_ring *r) 410 { 411 return rte_eth_from_rings(r->name, &r, 1, &r, 1, 412 r->memzone ? r->memzone->socket_id : SOCKET_ID_ANY); 413 } 414 415 static int 416 eth_dev_ring_create(const char *name, const unsigned int numa_node, 417 enum dev_action action, struct rte_eth_dev **eth_dev) 418 { 419 /* rx and tx are so-called from point of view of first port. 420 * They are inverted from the point of view of second port 421 */ 422 struct rte_ring *rxtx[RTE_PMD_RING_MAX_RX_RINGS]; 423 unsigned int i; 424 char rng_name[RTE_RING_NAMESIZE]; 425 unsigned int num_rings = RTE_MIN(RTE_PMD_RING_MAX_RX_RINGS, 426 RTE_PMD_RING_MAX_TX_RINGS); 427 428 for (i = 0; i < num_rings; i++) { 429 int cc; 430 431 cc = snprintf(rng_name, sizeof(rng_name), 432 "ETH_RXTX%u_%s", i, name); 433 if (cc >= (int)sizeof(rng_name)) { 434 rte_errno = ENAMETOOLONG; 435 return -1; 436 } 437 438 rxtx[i] = (action == DEV_CREATE) ? 439 rte_ring_create(rng_name, 1024, numa_node, 440 RING_F_SP_ENQ|RING_F_SC_DEQ) : 441 rte_ring_lookup(rng_name); 442 if (rxtx[i] == NULL) 443 return -1; 444 } 445 446 if (do_eth_dev_ring_create(name, rxtx, num_rings, rxtx, num_rings, 447 numa_node, action, eth_dev) < 0) 448 return -1; 449 450 return 0; 451 } 452 453 struct node_action_pair { 454 char name[PATH_MAX]; 455 unsigned int node; 456 enum dev_action action; 457 }; 458 459 struct node_action_list { 460 unsigned int total; 461 unsigned int count; 462 struct node_action_pair *list; 463 }; 464 465 static int parse_kvlist(const char *key __rte_unused, 466 const char *value, void *data) 467 { 468 struct node_action_list *info = data; 469 int ret; 470 char *name; 471 char *action; 472 char *node; 473 char *end; 474 475 name = strdup(value); 476 477 ret = -EINVAL; 478 479 if (!name) { 480 PMD_LOG(WARNING, "command line parameter is empty for ring pmd!"); 481 goto out; 482 } 483 484 node = strchr(name, ':'); 485 if (!node) { 486 PMD_LOG(WARNING, "could not parse node value from %s", 487 name); 488 goto out; 489 } 490 491 *node = '\0'; 492 node++; 493 494 action = strchr(node, ':'); 495 if (!action) { 496 PMD_LOG(WARNING, "could not parse action value from %s", 497 node); 498 goto out; 499 } 500 501 *action = '\0'; 502 action++; 503 504 /* 505 * Need to do some sanity checking here 506 */ 507 508 if (strcmp(action, ETH_RING_ACTION_ATTACH) == 0) 509 info->list[info->count].action = DEV_ATTACH; 510 else if (strcmp(action, ETH_RING_ACTION_CREATE) == 0) 511 info->list[info->count].action = DEV_CREATE; 512 else 513 goto out; 514 515 errno = 0; 516 info->list[info->count].node = strtol(node, &end, 10); 517 518 if ((errno != 0) || (*end != '\0')) { 519 PMD_LOG(WARNING, 520 "node value %s is unparseable as a number", node); 521 goto out; 522 } 523 524 snprintf(info->list[info->count].name, sizeof(info->list[info->count].name), "%s", name); 525 526 info->count++; 527 528 ret = 0; 529 out: 530 free(name); 531 return ret; 532 } 533 534 static int 535 parse_internal_args(const char *key __rte_unused, const char *value, 536 void *data) 537 { 538 struct ring_internal_args **internal_args = data; 539 void *args; 540 541 sscanf(value, "%p", &args); 542 543 *internal_args = args; 544 545 if ((*internal_args)->addr != args) 546 return -1; 547 548 return 0; 549 } 550 551 static int 552 rte_pmd_ring_probe(struct rte_vdev_device *dev) 553 { 554 const char *name, *params; 555 struct rte_kvargs *kvlist = NULL; 556 int ret = 0; 557 struct node_action_list *info = NULL; 558 struct rte_eth_dev *eth_dev = NULL; 559 struct ring_internal_args *internal_args; 560 561 name = rte_vdev_device_name(dev); 562 params = rte_vdev_device_args(dev); 563 564 PMD_LOG(INFO, "Initializing pmd_ring for %s", name); 565 566 if (params == NULL || params[0] == '\0') { 567 ret = eth_dev_ring_create(name, rte_socket_id(), DEV_CREATE, 568 ð_dev); 569 if (ret == -1) { 570 PMD_LOG(INFO, 571 "Attach to pmd_ring for %s", name); 572 ret = eth_dev_ring_create(name, rte_socket_id(), 573 DEV_ATTACH, ð_dev); 574 } 575 } else { 576 kvlist = rte_kvargs_parse(params, valid_arguments); 577 578 if (!kvlist) { 579 PMD_LOG(INFO, 580 "Ignoring unsupported parameters when creatingrings-backed ethernet device"); 581 ret = eth_dev_ring_create(name, rte_socket_id(), 582 DEV_CREATE, ð_dev); 583 if (ret == -1) { 584 PMD_LOG(INFO, 585 "Attach to pmd_ring for %s", 586 name); 587 ret = eth_dev_ring_create(name, rte_socket_id(), 588 DEV_ATTACH, ð_dev); 589 } 590 591 if (eth_dev) 592 eth_dev->device = &dev->device; 593 594 return ret; 595 } 596 597 if (rte_kvargs_count(kvlist, ETH_RING_INTERNAL_ARG) == 1) { 598 ret = rte_kvargs_process(kvlist, ETH_RING_INTERNAL_ARG, 599 parse_internal_args, 600 &internal_args); 601 if (ret < 0) 602 goto out_free; 603 604 ret = do_eth_dev_ring_create(name, 605 internal_args->rx_queues, 606 internal_args->nb_rx_queues, 607 internal_args->tx_queues, 608 internal_args->nb_tx_queues, 609 internal_args->numa_node, 610 DEV_ATTACH, 611 ð_dev); 612 if (ret >= 0) 613 ret = 0; 614 } else { 615 ret = rte_kvargs_count(kvlist, ETH_RING_NUMA_NODE_ACTION_ARG); 616 info = rte_zmalloc("struct node_action_list", 617 sizeof(struct node_action_list) + 618 (sizeof(struct node_action_pair) * ret), 619 0); 620 if (!info) 621 goto out_free; 622 623 info->total = ret; 624 info->list = (struct node_action_pair *)(info + 1); 625 626 ret = rte_kvargs_process(kvlist, ETH_RING_NUMA_NODE_ACTION_ARG, 627 parse_kvlist, info); 628 629 if (ret < 0) 630 goto out_free; 631 632 for (info->count = 0; info->count < info->total; info->count++) { 633 ret = eth_dev_ring_create(info->list[info->count].name, 634 info->list[info->count].node, 635 info->list[info->count].action, 636 ð_dev); 637 if ((ret == -1) && 638 (info->list[info->count].action == DEV_CREATE)) { 639 PMD_LOG(INFO, 640 "Attach to pmd_ring for %s", 641 name); 642 ret = eth_dev_ring_create(name, 643 info->list[info->count].node, 644 DEV_ATTACH, 645 ð_dev); 646 } 647 } 648 } 649 } 650 651 if (eth_dev) 652 eth_dev->device = &dev->device; 653 654 out_free: 655 rte_kvargs_free(kvlist); 656 rte_free(info); 657 return ret; 658 } 659 660 static int 661 rte_pmd_ring_remove(struct rte_vdev_device *dev) 662 { 663 const char *name = rte_vdev_device_name(dev); 664 struct rte_eth_dev *eth_dev = NULL; 665 struct pmd_internals *internals = NULL; 666 struct ring_queue *r = NULL; 667 uint16_t i; 668 669 PMD_LOG(INFO, "Un-Initializing pmd_ring for %s", name); 670 671 if (name == NULL) 672 return -EINVAL; 673 674 /* find an ethdev entry */ 675 eth_dev = rte_eth_dev_allocated(name); 676 if (eth_dev == NULL) 677 return -ENODEV; 678 679 eth_dev_stop(eth_dev); 680 681 internals = eth_dev->data->dev_private; 682 if (internals->action == DEV_CREATE) { 683 /* 684 * it is only necessary to delete the rings in rx_queues because 685 * they are the same used in tx_queues 686 */ 687 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) { 688 r = eth_dev->data->rx_queues[i]; 689 rte_ring_free(r->rng); 690 } 691 } 692 693 /* mac_addrs must not be freed alone because part of dev_private */ 694 eth_dev->data->mac_addrs = NULL; 695 rte_eth_dev_release_port(eth_dev); 696 return 0; 697 } 698 699 static struct rte_vdev_driver pmd_ring_drv = { 700 .probe = rte_pmd_ring_probe, 701 .remove = rte_pmd_ring_remove, 702 }; 703 704 RTE_PMD_REGISTER_VDEV(net_ring, pmd_ring_drv); 705 RTE_PMD_REGISTER_ALIAS(net_ring, eth_ring); 706 RTE_PMD_REGISTER_PARAM_STRING(net_ring, 707 ETH_RING_NUMA_NODE_ACTION_ARG "=name:node:action(ATTACH|CREATE)"); 708 709 RTE_INIT(eth_ring_init_log) 710 { 711 eth_ring_logtype = rte_log_register("pmd.net.ring"); 712 if (eth_ring_logtype >= 0) 713 rte_log_set_level(eth_ring_logtype, RTE_LOG_NOTICE); 714 } 715