1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright 2022 Microsoft Corporation 3 */ 4 5 #include <unistd.h> 6 #include <dirent.h> 7 #include <fcntl.h> 8 #include <sys/mman.h> 9 #include <sys/ioctl.h> 10 #include <net/if.h> 11 12 #include <ethdev_driver.h> 13 #include <ethdev_pci.h> 14 #include <rte_kvargs.h> 15 #include <rte_eal_paging.h> 16 #include <rte_pci.h> 17 18 #include <infiniband/verbs.h> 19 #include <infiniband/manadv.h> 20 21 #include <assert.h> 22 23 #include "mana.h" 24 25 /* Shared memory between primary/secondary processes, per driver */ 26 /* Data to track primary/secondary usage */ 27 struct mana_shared_data *mana_shared_data; 28 static struct mana_shared_data mana_local_data; 29 30 /* The memory region for the above data */ 31 static const struct rte_memzone *mana_shared_mz; 32 static const char *MZ_MANA_SHARED_DATA = "mana_shared_data"; 33 34 /* Spinlock for mana_shared_data */ 35 static rte_spinlock_t mana_shared_data_lock = RTE_SPINLOCK_INITIALIZER; 36 37 /* Allocate a buffer on the stack and fill it with a printf format string. */ 38 #define MANA_MKSTR(name, ...) \ 39 int mkstr_size_##name = snprintf(NULL, 0, "" __VA_ARGS__); \ 40 char name[mkstr_size_##name + 1]; \ 41 \ 42 memset(name, 0, mkstr_size_##name + 1); \ 43 snprintf(name, sizeof(name), "" __VA_ARGS__) 44 45 int mana_logtype_driver; 46 int mana_logtype_init; 47 48 /* 49 * Callback from rdma-core to allocate a buffer for a queue. 50 */ 51 void * 52 mana_alloc_verbs_buf(size_t size, void *data) 53 { 54 void *ret; 55 size_t alignment = rte_mem_page_size(); 56 int socket = (int)(uintptr_t)data; 57 58 DRV_LOG(DEBUG, "size=%zu socket=%d", size, socket); 59 60 if (alignment == (size_t)-1) { 61 DRV_LOG(ERR, "Failed to get mem page size"); 62 rte_errno = ENOMEM; 63 return NULL; 64 } 65 66 ret = rte_zmalloc_socket("mana_verb_buf", size, alignment, socket); 67 if (!ret && size) 68 rte_errno = ENOMEM; 69 return ret; 70 } 71 72 void 73 mana_free_verbs_buf(void *ptr, void *data __rte_unused) 74 { 75 rte_free(ptr); 76 } 77 78 static int 79 mana_dev_configure(struct rte_eth_dev *dev) 80 { 81 struct mana_priv *priv = dev->data->dev_private; 82 struct rte_eth_conf *dev_conf = &dev->data->dev_conf; 83 84 if (dev_conf->rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG) 85 dev_conf->rxmode.offloads |= RTE_ETH_RX_OFFLOAD_RSS_HASH; 86 87 if (dev->data->nb_rx_queues != dev->data->nb_tx_queues) { 88 DRV_LOG(ERR, "Only support equal number of rx/tx queues"); 89 return -EINVAL; 90 } 91 92 if (!rte_is_power_of_2(dev->data->nb_rx_queues)) { 93 DRV_LOG(ERR, "number of TX/RX queues must be power of 2"); 94 return -EINVAL; 95 } 96 97 priv->num_queues = dev->data->nb_rx_queues; 98 99 manadv_set_context_attr(priv->ib_ctx, MANADV_CTX_ATTR_BUF_ALLOCATORS, 100 (void *)((uintptr_t)&(struct manadv_ctx_allocators){ 101 .alloc = &mana_alloc_verbs_buf, 102 .free = &mana_free_verbs_buf, 103 .data = 0, 104 })); 105 106 return 0; 107 } 108 109 static void 110 rx_intr_vec_disable(struct mana_priv *priv) 111 { 112 struct rte_intr_handle *intr_handle = priv->intr_handle; 113 114 rte_intr_free_epoll_fd(intr_handle); 115 rte_intr_vec_list_free(intr_handle); 116 rte_intr_nb_efd_set(intr_handle, 0); 117 } 118 119 static int 120 rx_intr_vec_enable(struct mana_priv *priv) 121 { 122 unsigned int i; 123 unsigned int rxqs_n = priv->dev_data->nb_rx_queues; 124 unsigned int n = RTE_MIN(rxqs_n, (uint32_t)RTE_MAX_RXTX_INTR_VEC_ID); 125 struct rte_intr_handle *intr_handle = priv->intr_handle; 126 int ret; 127 128 rx_intr_vec_disable(priv); 129 130 if (rte_intr_vec_list_alloc(intr_handle, NULL, n)) { 131 DRV_LOG(ERR, "Failed to allocate memory for interrupt vector"); 132 return -ENOMEM; 133 } 134 135 for (i = 0; i < n; i++) { 136 struct mana_rxq *rxq = priv->dev_data->rx_queues[i]; 137 138 ret = rte_intr_vec_list_index_set(intr_handle, i, 139 RTE_INTR_VEC_RXTX_OFFSET + i); 140 if (ret) { 141 DRV_LOG(ERR, "Failed to set intr vec %u", i); 142 return ret; 143 } 144 145 ret = rte_intr_efds_index_set(intr_handle, i, rxq->channel->fd); 146 if (ret) { 147 DRV_LOG(ERR, "Failed to set FD at intr %u", i); 148 return ret; 149 } 150 } 151 152 return rte_intr_nb_efd_set(intr_handle, n); 153 } 154 155 static void 156 rxq_intr_disable(struct mana_priv *priv) 157 { 158 int err = rte_errno; 159 160 rx_intr_vec_disable(priv); 161 rte_errno = err; 162 } 163 164 static int 165 rxq_intr_enable(struct mana_priv *priv) 166 { 167 const struct rte_eth_intr_conf *const intr_conf = 168 &priv->dev_data->dev_conf.intr_conf; 169 170 if (!intr_conf->rxq) 171 return 0; 172 173 return rx_intr_vec_enable(priv); 174 } 175 176 static int 177 mana_dev_start(struct rte_eth_dev *dev) 178 { 179 int ret; 180 struct mana_priv *priv = dev->data->dev_private; 181 182 rte_spinlock_init(&priv->mr_btree_lock); 183 ret = mana_mr_btree_init(&priv->mr_btree, MANA_MR_BTREE_CACHE_N, 184 dev->device->numa_node); 185 if (ret) { 186 DRV_LOG(ERR, "Failed to init device MR btree %d", ret); 187 return ret; 188 } 189 190 ret = mana_start_tx_queues(dev); 191 if (ret) { 192 DRV_LOG(ERR, "failed to start tx queues %d", ret); 193 goto failed_tx; 194 } 195 196 ret = mana_start_rx_queues(dev); 197 if (ret) { 198 DRV_LOG(ERR, "failed to start rx queues %d", ret); 199 goto failed_rx; 200 } 201 202 rte_wmb(); 203 204 dev->tx_pkt_burst = mana_tx_burst; 205 dev->rx_pkt_burst = mana_rx_burst; 206 207 DRV_LOG(INFO, "TX/RX queues have started"); 208 209 /* Enable datapath for secondary processes */ 210 mana_mp_req_on_rxtx(dev, MANA_MP_REQ_START_RXTX); 211 212 ret = rxq_intr_enable(priv); 213 if (ret) { 214 DRV_LOG(ERR, "Failed to enable RX interrupts"); 215 goto failed_intr; 216 } 217 218 return 0; 219 220 failed_intr: 221 mana_stop_rx_queues(dev); 222 223 failed_rx: 224 mana_stop_tx_queues(dev); 225 226 failed_tx: 227 mana_mr_btree_free(&priv->mr_btree); 228 229 return ret; 230 } 231 232 static int 233 mana_dev_stop(struct rte_eth_dev *dev) 234 { 235 int ret; 236 struct mana_priv *priv = dev->data->dev_private; 237 238 rxq_intr_disable(priv); 239 240 dev->tx_pkt_burst = mana_tx_burst_removed; 241 dev->rx_pkt_burst = mana_rx_burst_removed; 242 243 /* Stop datapath on secondary processes */ 244 mana_mp_req_on_rxtx(dev, MANA_MP_REQ_STOP_RXTX); 245 246 rte_wmb(); 247 248 ret = mana_stop_tx_queues(dev); 249 if (ret) { 250 DRV_LOG(ERR, "failed to stop tx queues"); 251 return ret; 252 } 253 254 ret = mana_stop_rx_queues(dev); 255 if (ret) { 256 DRV_LOG(ERR, "failed to stop tx queues"); 257 return ret; 258 } 259 260 return 0; 261 } 262 263 static int mana_intr_uninstall(struct mana_priv *priv); 264 265 static int 266 mana_dev_close(struct rte_eth_dev *dev) 267 { 268 struct mana_priv *priv = dev->data->dev_private; 269 int ret; 270 271 mana_remove_all_mr(priv); 272 273 ret = mana_intr_uninstall(priv); 274 if (ret) 275 return ret; 276 277 ret = ibv_close_device(priv->ib_ctx); 278 if (ret) { 279 ret = errno; 280 return ret; 281 } 282 283 return 0; 284 } 285 286 static int 287 mana_dev_info_get(struct rte_eth_dev *dev, 288 struct rte_eth_dev_info *dev_info) 289 { 290 struct mana_priv *priv = dev->data->dev_private; 291 292 dev_info->min_mtu = RTE_ETHER_MIN_MTU; 293 dev_info->max_mtu = MANA_MAX_MTU; 294 295 /* RX params */ 296 dev_info->min_rx_bufsize = MIN_RX_BUF_SIZE; 297 dev_info->max_rx_pktlen = MANA_MAX_MTU + RTE_ETHER_HDR_LEN; 298 299 dev_info->max_rx_queues = RTE_MIN(priv->max_rx_queues, UINT16_MAX); 300 dev_info->max_tx_queues = RTE_MIN(priv->max_tx_queues, UINT16_MAX); 301 302 dev_info->max_mac_addrs = MANA_MAX_MAC_ADDR; 303 dev_info->max_hash_mac_addrs = 0; 304 305 dev_info->max_vfs = 1; 306 307 /* Offload params */ 308 dev_info->rx_offload_capa = MANA_DEV_RX_OFFLOAD_SUPPORT; 309 310 dev_info->tx_offload_capa = MANA_DEV_TX_OFFLOAD_SUPPORT; 311 312 /* RSS */ 313 dev_info->reta_size = INDIRECTION_TABLE_NUM_ELEMENTS; 314 dev_info->hash_key_size = TOEPLITZ_HASH_KEY_SIZE_IN_BYTES; 315 dev_info->flow_type_rss_offloads = MANA_ETH_RSS_SUPPORT; 316 317 /* Thresholds */ 318 dev_info->default_rxconf = (struct rte_eth_rxconf){ 319 .rx_thresh = { 320 .pthresh = 8, 321 .hthresh = 8, 322 .wthresh = 0, 323 }, 324 .rx_free_thresh = 32, 325 /* If no descriptors available, pkts are dropped by default */ 326 .rx_drop_en = 1, 327 }; 328 329 dev_info->default_txconf = (struct rte_eth_txconf){ 330 .tx_thresh = { 331 .pthresh = 32, 332 .hthresh = 0, 333 .wthresh = 0, 334 }, 335 .tx_rs_thresh = 32, 336 .tx_free_thresh = 32, 337 }; 338 339 /* Buffer limits */ 340 dev_info->rx_desc_lim.nb_min = MIN_BUFFERS_PER_QUEUE; 341 dev_info->rx_desc_lim.nb_max = RTE_MIN(priv->max_rx_desc, UINT16_MAX); 342 dev_info->rx_desc_lim.nb_align = MIN_BUFFERS_PER_QUEUE; 343 dev_info->rx_desc_lim.nb_seg_max = 344 RTE_MIN(priv->max_recv_sge, UINT16_MAX); 345 dev_info->rx_desc_lim.nb_mtu_seg_max = 346 RTE_MIN(priv->max_recv_sge, UINT16_MAX); 347 348 dev_info->tx_desc_lim.nb_min = MIN_BUFFERS_PER_QUEUE; 349 dev_info->tx_desc_lim.nb_max = RTE_MIN(priv->max_tx_desc, UINT16_MAX); 350 dev_info->tx_desc_lim.nb_align = MIN_BUFFERS_PER_QUEUE; 351 dev_info->tx_desc_lim.nb_seg_max = 352 RTE_MIN(priv->max_send_sge, UINT16_MAX); 353 dev_info->tx_desc_lim.nb_mtu_seg_max = 354 RTE_MIN(priv->max_send_sge, UINT16_MAX); 355 356 /* Speed */ 357 dev_info->speed_capa = RTE_ETH_LINK_SPEED_100G; 358 359 /* RX params */ 360 dev_info->default_rxportconf.burst_size = 1; 361 dev_info->default_rxportconf.ring_size = MAX_RECEIVE_BUFFERS_PER_QUEUE; 362 dev_info->default_rxportconf.nb_queues = 1; 363 364 /* TX params */ 365 dev_info->default_txportconf.burst_size = 1; 366 dev_info->default_txportconf.ring_size = MAX_SEND_BUFFERS_PER_QUEUE; 367 dev_info->default_txportconf.nb_queues = 1; 368 369 return 0; 370 } 371 372 static void 373 mana_dev_tx_queue_info(struct rte_eth_dev *dev, uint16_t queue_id, 374 struct rte_eth_txq_info *qinfo) 375 { 376 struct mana_txq *txq = dev->data->tx_queues[queue_id]; 377 378 qinfo->conf.offloads = dev->data->dev_conf.txmode.offloads; 379 qinfo->nb_desc = txq->num_desc; 380 } 381 382 static void 383 mana_dev_rx_queue_info(struct rte_eth_dev *dev, uint16_t queue_id, 384 struct rte_eth_rxq_info *qinfo) 385 { 386 struct mana_rxq *rxq = dev->data->rx_queues[queue_id]; 387 388 qinfo->mp = rxq->mp; 389 qinfo->nb_desc = rxq->num_desc; 390 qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads; 391 } 392 393 static const uint32_t * 394 mana_supported_ptypes(struct rte_eth_dev *dev __rte_unused) 395 { 396 static const uint32_t ptypes[] = { 397 RTE_PTYPE_L2_ETHER, 398 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN, 399 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN, 400 RTE_PTYPE_L4_FRAG, 401 RTE_PTYPE_L4_TCP, 402 RTE_PTYPE_L4_UDP, 403 RTE_PTYPE_UNKNOWN 404 }; 405 406 return ptypes; 407 } 408 409 static int 410 mana_rss_hash_update(struct rte_eth_dev *dev, 411 struct rte_eth_rss_conf *rss_conf) 412 { 413 struct mana_priv *priv = dev->data->dev_private; 414 415 /* Currently can only update RSS hash when device is stopped */ 416 if (dev->data->dev_started) { 417 DRV_LOG(ERR, "Can't update RSS after device has started"); 418 return -ENODEV; 419 } 420 421 if (rss_conf->rss_hf & ~MANA_ETH_RSS_SUPPORT) { 422 DRV_LOG(ERR, "Port %u invalid RSS HF 0x%" PRIx64, 423 dev->data->port_id, rss_conf->rss_hf); 424 return -EINVAL; 425 } 426 427 if (rss_conf->rss_key && rss_conf->rss_key_len) { 428 if (rss_conf->rss_key_len != TOEPLITZ_HASH_KEY_SIZE_IN_BYTES) { 429 DRV_LOG(ERR, "Port %u key len must be %u long", 430 dev->data->port_id, 431 TOEPLITZ_HASH_KEY_SIZE_IN_BYTES); 432 return -EINVAL; 433 } 434 435 priv->rss_conf.rss_key_len = rss_conf->rss_key_len; 436 priv->rss_conf.rss_key = 437 rte_zmalloc("mana_rss", rss_conf->rss_key_len, 438 RTE_CACHE_LINE_SIZE); 439 if (!priv->rss_conf.rss_key) 440 return -ENOMEM; 441 memcpy(priv->rss_conf.rss_key, rss_conf->rss_key, 442 rss_conf->rss_key_len); 443 } 444 priv->rss_conf.rss_hf = rss_conf->rss_hf; 445 446 return 0; 447 } 448 449 static int 450 mana_rss_hash_conf_get(struct rte_eth_dev *dev, 451 struct rte_eth_rss_conf *rss_conf) 452 { 453 struct mana_priv *priv = dev->data->dev_private; 454 455 if (!rss_conf) 456 return -EINVAL; 457 458 if (rss_conf->rss_key && 459 rss_conf->rss_key_len >= priv->rss_conf.rss_key_len) { 460 memcpy(rss_conf->rss_key, priv->rss_conf.rss_key, 461 priv->rss_conf.rss_key_len); 462 } 463 464 rss_conf->rss_key_len = priv->rss_conf.rss_key_len; 465 rss_conf->rss_hf = priv->rss_conf.rss_hf; 466 467 return 0; 468 } 469 470 static int 471 mana_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx, 472 uint16_t nb_desc, unsigned int socket_id, 473 const struct rte_eth_txconf *tx_conf __rte_unused) 474 475 { 476 struct mana_priv *priv = dev->data->dev_private; 477 struct mana_txq *txq; 478 int ret; 479 480 txq = rte_zmalloc_socket("mana_txq", sizeof(*txq), 0, socket_id); 481 if (!txq) { 482 DRV_LOG(ERR, "failed to allocate txq"); 483 return -ENOMEM; 484 } 485 486 txq->socket = socket_id; 487 488 txq->desc_ring = rte_malloc_socket("mana_tx_desc_ring", 489 sizeof(struct mana_txq_desc) * 490 nb_desc, 491 RTE_CACHE_LINE_SIZE, socket_id); 492 if (!txq->desc_ring) { 493 DRV_LOG(ERR, "failed to allocate txq desc_ring"); 494 ret = -ENOMEM; 495 goto fail; 496 } 497 498 txq->gdma_comp_buf = rte_malloc_socket("mana_txq_comp", 499 sizeof(*txq->gdma_comp_buf) * nb_desc, 500 RTE_CACHE_LINE_SIZE, socket_id); 501 if (!txq->gdma_comp_buf) { 502 DRV_LOG(ERR, "failed to allocate txq comp"); 503 ret = -ENOMEM; 504 goto fail; 505 } 506 507 ret = mana_mr_btree_init(&txq->mr_btree, 508 MANA_MR_BTREE_PER_QUEUE_N, socket_id); 509 if (ret) { 510 DRV_LOG(ERR, "Failed to init TXQ MR btree"); 511 goto fail; 512 } 513 514 DRV_LOG(DEBUG, "idx %u nb_desc %u socket %u txq->desc_ring %p", 515 queue_idx, nb_desc, socket_id, txq->desc_ring); 516 517 txq->desc_ring_head = 0; 518 txq->desc_ring_tail = 0; 519 txq->priv = priv; 520 txq->num_desc = nb_desc; 521 dev->data->tx_queues[queue_idx] = txq; 522 523 return 0; 524 525 fail: 526 rte_free(txq->gdma_comp_buf); 527 rte_free(txq->desc_ring); 528 rte_free(txq); 529 return ret; 530 } 531 532 static void 533 mana_dev_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid) 534 { 535 struct mana_txq *txq = dev->data->tx_queues[qid]; 536 537 mana_mr_btree_free(&txq->mr_btree); 538 539 rte_free(txq->gdma_comp_buf); 540 rte_free(txq->desc_ring); 541 rte_free(txq); 542 } 543 544 static int 545 mana_dev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx, 546 uint16_t nb_desc, unsigned int socket_id, 547 const struct rte_eth_rxconf *rx_conf __rte_unused, 548 struct rte_mempool *mp) 549 { 550 struct mana_priv *priv = dev->data->dev_private; 551 struct mana_rxq *rxq; 552 int ret; 553 554 rxq = rte_zmalloc_socket("mana_rxq", sizeof(*rxq), 0, socket_id); 555 if (!rxq) { 556 DRV_LOG(ERR, "failed to allocate rxq"); 557 return -ENOMEM; 558 } 559 560 DRV_LOG(DEBUG, "idx %u nb_desc %u socket %u", 561 queue_idx, nb_desc, socket_id); 562 563 rxq->socket = socket_id; 564 565 rxq->desc_ring = rte_zmalloc_socket("mana_rx_mbuf_ring", 566 sizeof(struct mana_rxq_desc) * 567 nb_desc, 568 RTE_CACHE_LINE_SIZE, socket_id); 569 570 if (!rxq->desc_ring) { 571 DRV_LOG(ERR, "failed to allocate rxq desc_ring"); 572 ret = -ENOMEM; 573 goto fail; 574 } 575 576 rxq->desc_ring_head = 0; 577 rxq->desc_ring_tail = 0; 578 579 rxq->gdma_comp_buf = rte_malloc_socket("mana_rxq_comp", 580 sizeof(*rxq->gdma_comp_buf) * nb_desc, 581 RTE_CACHE_LINE_SIZE, socket_id); 582 if (!rxq->gdma_comp_buf) { 583 DRV_LOG(ERR, "failed to allocate rxq comp"); 584 ret = -ENOMEM; 585 goto fail; 586 } 587 588 ret = mana_mr_btree_init(&rxq->mr_btree, 589 MANA_MR_BTREE_PER_QUEUE_N, socket_id); 590 if (ret) { 591 DRV_LOG(ERR, "Failed to init RXQ MR btree"); 592 goto fail; 593 } 594 595 rxq->priv = priv; 596 rxq->num_desc = nb_desc; 597 rxq->mp = mp; 598 dev->data->rx_queues[queue_idx] = rxq; 599 600 return 0; 601 602 fail: 603 rte_free(rxq->gdma_comp_buf); 604 rte_free(rxq->desc_ring); 605 rte_free(rxq); 606 return ret; 607 } 608 609 static void 610 mana_dev_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid) 611 { 612 struct mana_rxq *rxq = dev->data->rx_queues[qid]; 613 614 mana_mr_btree_free(&rxq->mr_btree); 615 616 rte_free(rxq->gdma_comp_buf); 617 rte_free(rxq->desc_ring); 618 rte_free(rxq); 619 } 620 621 static int 622 mana_dev_link_update(struct rte_eth_dev *dev, 623 int wait_to_complete __rte_unused) 624 { 625 struct rte_eth_link link; 626 627 /* MANA has no concept of carrier state, always reporting UP */ 628 link = (struct rte_eth_link) { 629 .link_duplex = RTE_ETH_LINK_FULL_DUPLEX, 630 .link_autoneg = RTE_ETH_LINK_SPEED_FIXED, 631 .link_speed = RTE_ETH_SPEED_NUM_100G, 632 .link_status = RTE_ETH_LINK_UP, 633 }; 634 635 return rte_eth_linkstatus_set(dev, &link); 636 } 637 638 static int 639 mana_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats) 640 { 641 unsigned int i; 642 643 for (i = 0; i < dev->data->nb_tx_queues; i++) { 644 struct mana_txq *txq = dev->data->tx_queues[i]; 645 646 if (!txq) 647 continue; 648 649 stats->opackets += txq->stats.packets; 650 stats->obytes += txq->stats.bytes; 651 stats->oerrors += txq->stats.errors; 652 653 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) { 654 stats->q_opackets[i] = txq->stats.packets; 655 stats->q_obytes[i] = txq->stats.bytes; 656 } 657 } 658 659 stats->rx_nombuf = 0; 660 for (i = 0; i < dev->data->nb_rx_queues; i++) { 661 struct mana_rxq *rxq = dev->data->rx_queues[i]; 662 663 if (!rxq) 664 continue; 665 666 stats->ipackets += rxq->stats.packets; 667 stats->ibytes += rxq->stats.bytes; 668 stats->ierrors += rxq->stats.errors; 669 670 /* There is no good way to get stats->imissed, not setting it */ 671 672 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) { 673 stats->q_ipackets[i] = rxq->stats.packets; 674 stats->q_ibytes[i] = rxq->stats.bytes; 675 } 676 677 stats->rx_nombuf += rxq->stats.nombuf; 678 } 679 680 return 0; 681 } 682 683 static int 684 mana_dev_stats_reset(struct rte_eth_dev *dev __rte_unused) 685 { 686 unsigned int i; 687 688 PMD_INIT_FUNC_TRACE(); 689 690 for (i = 0; i < dev->data->nb_tx_queues; i++) { 691 struct mana_txq *txq = dev->data->tx_queues[i]; 692 693 if (!txq) 694 continue; 695 696 memset(&txq->stats, 0, sizeof(txq->stats)); 697 } 698 699 for (i = 0; i < dev->data->nb_rx_queues; i++) { 700 struct mana_rxq *rxq = dev->data->rx_queues[i]; 701 702 if (!rxq) 703 continue; 704 705 memset(&rxq->stats, 0, sizeof(rxq->stats)); 706 } 707 708 return 0; 709 } 710 711 static int 712 mana_get_ifname(const struct mana_priv *priv, char (*ifname)[IF_NAMESIZE]) 713 { 714 int ret; 715 DIR *dir; 716 struct dirent *dent; 717 718 MANA_MKSTR(dirpath, "%s/device/net", priv->ib_ctx->device->ibdev_path); 719 720 dir = opendir(dirpath); 721 if (dir == NULL) 722 return -ENODEV; 723 724 while ((dent = readdir(dir)) != NULL) { 725 char *name = dent->d_name; 726 FILE *file; 727 struct rte_ether_addr addr; 728 char *mac = NULL; 729 730 if ((name[0] == '.') && 731 ((name[1] == '\0') || 732 ((name[1] == '.') && (name[2] == '\0')))) 733 continue; 734 735 MANA_MKSTR(path, "%s/%s/address", dirpath, name); 736 737 file = fopen(path, "r"); 738 if (!file) { 739 ret = -ENODEV; 740 break; 741 } 742 743 ret = fscanf(file, "%ms", &mac); 744 fclose(file); 745 746 if (ret <= 0) { 747 ret = -EINVAL; 748 break; 749 } 750 751 ret = rte_ether_unformat_addr(mac, &addr); 752 free(mac); 753 if (ret) 754 break; 755 756 if (rte_is_same_ether_addr(&addr, priv->dev_data->mac_addrs)) { 757 strlcpy(*ifname, name, sizeof(*ifname)); 758 ret = 0; 759 break; 760 } 761 } 762 763 closedir(dir); 764 return ret; 765 } 766 767 static int 768 mana_ifreq(const struct mana_priv *priv, int req, struct ifreq *ifr) 769 { 770 int sock, ret; 771 772 sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP); 773 if (sock == -1) 774 return -errno; 775 776 ret = mana_get_ifname(priv, &ifr->ifr_name); 777 if (ret) { 778 close(sock); 779 return ret; 780 } 781 782 if (ioctl(sock, req, ifr) == -1) 783 ret = -errno; 784 785 close(sock); 786 787 return ret; 788 } 789 790 static int 791 mana_mtu_set(struct rte_eth_dev *dev, uint16_t mtu) 792 { 793 struct mana_priv *priv = dev->data->dev_private; 794 struct ifreq request = { .ifr_mtu = mtu, }; 795 796 return mana_ifreq(priv, SIOCSIFMTU, &request); 797 } 798 799 static const struct eth_dev_ops mana_dev_ops = { 800 .dev_configure = mana_dev_configure, 801 .dev_start = mana_dev_start, 802 .dev_stop = mana_dev_stop, 803 .dev_close = mana_dev_close, 804 .dev_infos_get = mana_dev_info_get, 805 .txq_info_get = mana_dev_tx_queue_info, 806 .rxq_info_get = mana_dev_rx_queue_info, 807 .dev_supported_ptypes_get = mana_supported_ptypes, 808 .rss_hash_update = mana_rss_hash_update, 809 .rss_hash_conf_get = mana_rss_hash_conf_get, 810 .tx_queue_setup = mana_dev_tx_queue_setup, 811 .tx_queue_release = mana_dev_tx_queue_release, 812 .rx_queue_setup = mana_dev_rx_queue_setup, 813 .rx_queue_release = mana_dev_rx_queue_release, 814 .rx_queue_intr_enable = mana_rx_intr_enable, 815 .rx_queue_intr_disable = mana_rx_intr_disable, 816 .link_update = mana_dev_link_update, 817 .stats_get = mana_dev_stats_get, 818 .stats_reset = mana_dev_stats_reset, 819 .mtu_set = mana_mtu_set, 820 }; 821 822 static const struct eth_dev_ops mana_dev_secondary_ops = { 823 .stats_get = mana_dev_stats_get, 824 .stats_reset = mana_dev_stats_reset, 825 .dev_infos_get = mana_dev_info_get, 826 }; 827 828 uint16_t 829 mana_rx_burst_removed(void *dpdk_rxq __rte_unused, 830 struct rte_mbuf **pkts __rte_unused, 831 uint16_t pkts_n __rte_unused) 832 { 833 rte_mb(); 834 return 0; 835 } 836 837 uint16_t 838 mana_tx_burst_removed(void *dpdk_rxq __rte_unused, 839 struct rte_mbuf **pkts __rte_unused, 840 uint16_t pkts_n __rte_unused) 841 { 842 rte_mb(); 843 return 0; 844 } 845 846 #define ETH_MANA_MAC_ARG "mac" 847 static const char * const mana_init_args[] = { 848 ETH_MANA_MAC_ARG, 849 NULL, 850 }; 851 852 /* Support of parsing up to 8 mac address from EAL command line */ 853 #define MAX_NUM_ADDRESS 8 854 struct mana_conf { 855 struct rte_ether_addr mac_array[MAX_NUM_ADDRESS]; 856 unsigned int index; 857 }; 858 859 static int 860 mana_arg_parse_callback(const char *key, const char *val, void *private) 861 { 862 struct mana_conf *conf = (struct mana_conf *)private; 863 int ret; 864 865 DRV_LOG(INFO, "key=%s value=%s index=%d", key, val, conf->index); 866 867 if (conf->index >= MAX_NUM_ADDRESS) { 868 DRV_LOG(ERR, "Exceeding max MAC address"); 869 return 1; 870 } 871 872 ret = rte_ether_unformat_addr(val, &conf->mac_array[conf->index]); 873 if (ret) { 874 DRV_LOG(ERR, "Invalid MAC address %s", val); 875 return ret; 876 } 877 878 conf->index++; 879 880 return 0; 881 } 882 883 static int 884 mana_parse_args(struct rte_devargs *devargs, struct mana_conf *conf) 885 { 886 struct rte_kvargs *kvlist; 887 unsigned int arg_count; 888 int ret = 0; 889 890 kvlist = rte_kvargs_parse(devargs->drv_str, mana_init_args); 891 if (!kvlist) { 892 DRV_LOG(ERR, "failed to parse kvargs args=%s", devargs->drv_str); 893 return -EINVAL; 894 } 895 896 arg_count = rte_kvargs_count(kvlist, mana_init_args[0]); 897 if (arg_count > MAX_NUM_ADDRESS) { 898 ret = -EINVAL; 899 goto free_kvlist; 900 } 901 ret = rte_kvargs_process(kvlist, mana_init_args[0], 902 mana_arg_parse_callback, conf); 903 if (ret) { 904 DRV_LOG(ERR, "error parsing args"); 905 goto free_kvlist; 906 } 907 908 free_kvlist: 909 rte_kvargs_free(kvlist); 910 return ret; 911 } 912 913 static int 914 get_port_mac(struct ibv_device *device, unsigned int port, 915 struct rte_ether_addr *addr) 916 { 917 FILE *file; 918 int ret = 0; 919 DIR *dir; 920 struct dirent *dent; 921 unsigned int dev_port; 922 923 MANA_MKSTR(path, "%s/device/net", device->ibdev_path); 924 925 dir = opendir(path); 926 if (!dir) 927 return -ENOENT; 928 929 while ((dent = readdir(dir))) { 930 char *name = dent->d_name; 931 char *mac = NULL; 932 933 MANA_MKSTR(port_path, "%s/%s/dev_port", path, name); 934 935 /* Ignore . and .. */ 936 if ((name[0] == '.') && 937 ((name[1] == '\0') || 938 ((name[1] == '.') && (name[2] == '\0')))) 939 continue; 940 941 file = fopen(port_path, "r"); 942 if (!file) 943 continue; 944 945 ret = fscanf(file, "%u", &dev_port); 946 fclose(file); 947 948 if (ret != 1) 949 continue; 950 951 /* Ethernet ports start at 0, IB port start at 1 */ 952 if (dev_port == port - 1) { 953 MANA_MKSTR(address_path, "%s/%s/address", path, name); 954 955 file = fopen(address_path, "r"); 956 if (!file) 957 continue; 958 959 ret = fscanf(file, "%ms", &mac); 960 fclose(file); 961 962 if (ret < 0) 963 break; 964 965 ret = rte_ether_unformat_addr(mac, addr); 966 if (ret) 967 DRV_LOG(ERR, "unrecognized mac addr %s", mac); 968 969 free(mac); 970 break; 971 } 972 } 973 974 closedir(dir); 975 return ret; 976 } 977 978 static int 979 mana_ibv_device_to_pci_addr(const struct ibv_device *device, 980 struct rte_pci_addr *pci_addr) 981 { 982 FILE *file; 983 char *line = NULL; 984 size_t len = 0; 985 986 MANA_MKSTR(path, "%s/device/uevent", device->ibdev_path); 987 988 file = fopen(path, "r"); 989 if (!file) 990 return -errno; 991 992 while (getline(&line, &len, file) != -1) { 993 /* Extract information. */ 994 if (sscanf(line, 995 "PCI_SLOT_NAME=" 996 "%" SCNx32 ":%" SCNx8 ":%" SCNx8 ".%" SCNx8 "\n", 997 &pci_addr->domain, 998 &pci_addr->bus, 999 &pci_addr->devid, 1000 &pci_addr->function) == 4) { 1001 break; 1002 } 1003 } 1004 1005 free(line); 1006 fclose(file); 1007 return 0; 1008 } 1009 1010 /* 1011 * Interrupt handler from IB layer to notify this device is being removed. 1012 */ 1013 static void 1014 mana_intr_handler(void *arg) 1015 { 1016 struct mana_priv *priv = arg; 1017 struct ibv_context *ctx = priv->ib_ctx; 1018 struct ibv_async_event event; 1019 1020 /* Read and ack all messages from IB device */ 1021 while (true) { 1022 if (ibv_get_async_event(ctx, &event)) 1023 break; 1024 1025 if (event.event_type == IBV_EVENT_DEVICE_FATAL) { 1026 struct rte_eth_dev *dev; 1027 1028 dev = &rte_eth_devices[priv->port_id]; 1029 if (dev->data->dev_conf.intr_conf.rmv) 1030 rte_eth_dev_callback_process(dev, 1031 RTE_ETH_EVENT_INTR_RMV, NULL); 1032 } 1033 1034 ibv_ack_async_event(&event); 1035 } 1036 } 1037 1038 static int 1039 mana_intr_uninstall(struct mana_priv *priv) 1040 { 1041 int ret; 1042 1043 ret = rte_intr_callback_unregister(priv->intr_handle, 1044 mana_intr_handler, priv); 1045 if (ret <= 0) { 1046 DRV_LOG(ERR, "Failed to unregister intr callback ret %d", ret); 1047 return ret; 1048 } 1049 1050 rte_intr_instance_free(priv->intr_handle); 1051 1052 return 0; 1053 } 1054 1055 int 1056 mana_fd_set_non_blocking(int fd) 1057 { 1058 int ret = fcntl(fd, F_GETFL); 1059 1060 if (ret != -1 && !fcntl(fd, F_SETFL, ret | O_NONBLOCK)) 1061 return 0; 1062 1063 rte_errno = errno; 1064 return -rte_errno; 1065 } 1066 1067 static int 1068 mana_intr_install(struct rte_eth_dev *eth_dev, struct mana_priv *priv) 1069 { 1070 int ret; 1071 struct ibv_context *ctx = priv->ib_ctx; 1072 1073 priv->intr_handle = rte_intr_instance_alloc(RTE_INTR_INSTANCE_F_SHARED); 1074 if (!priv->intr_handle) { 1075 DRV_LOG(ERR, "Failed to allocate intr_handle"); 1076 rte_errno = ENOMEM; 1077 return -ENOMEM; 1078 } 1079 1080 ret = rte_intr_fd_set(priv->intr_handle, -1); 1081 if (ret) 1082 goto free_intr; 1083 1084 ret = mana_fd_set_non_blocking(ctx->async_fd); 1085 if (ret) { 1086 DRV_LOG(ERR, "Failed to change async_fd to NONBLOCK"); 1087 goto free_intr; 1088 } 1089 1090 ret = rte_intr_fd_set(priv->intr_handle, ctx->async_fd); 1091 if (ret) 1092 goto free_intr; 1093 1094 ret = rte_intr_type_set(priv->intr_handle, RTE_INTR_HANDLE_EXT); 1095 if (ret) 1096 goto free_intr; 1097 1098 ret = rte_intr_callback_register(priv->intr_handle, 1099 mana_intr_handler, priv); 1100 if (ret) { 1101 DRV_LOG(ERR, "Failed to register intr callback"); 1102 rte_intr_fd_set(priv->intr_handle, -1); 1103 goto free_intr; 1104 } 1105 1106 eth_dev->intr_handle = priv->intr_handle; 1107 return 0; 1108 1109 free_intr: 1110 rte_intr_instance_free(priv->intr_handle); 1111 priv->intr_handle = NULL; 1112 1113 return ret; 1114 } 1115 1116 static int 1117 mana_proc_priv_init(struct rte_eth_dev *dev) 1118 { 1119 struct mana_process_priv *priv; 1120 1121 priv = rte_zmalloc_socket("mana_proc_priv", 1122 sizeof(struct mana_process_priv), 1123 RTE_CACHE_LINE_SIZE, 1124 dev->device->numa_node); 1125 if (!priv) 1126 return -ENOMEM; 1127 1128 dev->process_private = priv; 1129 return 0; 1130 } 1131 1132 /* 1133 * Map the doorbell page for the secondary process through IB device handle. 1134 */ 1135 static int 1136 mana_map_doorbell_secondary(struct rte_eth_dev *eth_dev, int fd) 1137 { 1138 struct mana_process_priv *priv = eth_dev->process_private; 1139 1140 void *addr; 1141 1142 addr = mmap(NULL, rte_mem_page_size(), PROT_WRITE, MAP_SHARED, fd, 0); 1143 if (addr == MAP_FAILED) { 1144 DRV_LOG(ERR, "Failed to map secondary doorbell port %u", 1145 eth_dev->data->port_id); 1146 return -ENOMEM; 1147 } 1148 1149 DRV_LOG(INFO, "Secondary doorbell mapped to %p", addr); 1150 1151 priv->db_page = addr; 1152 1153 return 0; 1154 } 1155 1156 /* Initialize shared data for the driver (all devices) */ 1157 static int 1158 mana_init_shared_data(void) 1159 { 1160 int ret = 0; 1161 const struct rte_memzone *secondary_mz; 1162 1163 rte_spinlock_lock(&mana_shared_data_lock); 1164 1165 /* Skip if shared data is already initialized */ 1166 if (mana_shared_data) 1167 goto exit; 1168 1169 if (rte_eal_process_type() == RTE_PROC_PRIMARY) { 1170 mana_shared_mz = rte_memzone_reserve(MZ_MANA_SHARED_DATA, 1171 sizeof(*mana_shared_data), 1172 SOCKET_ID_ANY, 0); 1173 if (!mana_shared_mz) { 1174 DRV_LOG(ERR, "Cannot allocate mana shared data"); 1175 ret = -rte_errno; 1176 goto exit; 1177 } 1178 1179 mana_shared_data = mana_shared_mz->addr; 1180 memset(mana_shared_data, 0, sizeof(*mana_shared_data)); 1181 rte_spinlock_init(&mana_shared_data->lock); 1182 } else { 1183 secondary_mz = rte_memzone_lookup(MZ_MANA_SHARED_DATA); 1184 if (!secondary_mz) { 1185 DRV_LOG(ERR, "Cannot attach mana shared data"); 1186 ret = -rte_errno; 1187 goto exit; 1188 } 1189 1190 mana_shared_data = secondary_mz->addr; 1191 memset(&mana_local_data, 0, sizeof(mana_local_data)); 1192 } 1193 1194 exit: 1195 rte_spinlock_unlock(&mana_shared_data_lock); 1196 1197 return ret; 1198 } 1199 1200 /* 1201 * Init the data structures for use in primary and secondary processes. 1202 */ 1203 static int 1204 mana_init_once(void) 1205 { 1206 int ret; 1207 1208 ret = mana_init_shared_data(); 1209 if (ret) 1210 return ret; 1211 1212 rte_spinlock_lock(&mana_shared_data->lock); 1213 1214 switch (rte_eal_process_type()) { 1215 case RTE_PROC_PRIMARY: 1216 if (mana_shared_data->init_done) 1217 break; 1218 1219 ret = mana_mp_init_primary(); 1220 if (ret) 1221 break; 1222 DRV_LOG(ERR, "MP INIT PRIMARY"); 1223 1224 mana_shared_data->init_done = 1; 1225 break; 1226 1227 case RTE_PROC_SECONDARY: 1228 1229 if (mana_local_data.init_done) 1230 break; 1231 1232 ret = mana_mp_init_secondary(); 1233 if (ret) 1234 break; 1235 1236 DRV_LOG(ERR, "MP INIT SECONDARY"); 1237 1238 mana_local_data.init_done = 1; 1239 break; 1240 1241 default: 1242 /* Impossible, internal error */ 1243 ret = -EPROTO; 1244 break; 1245 } 1246 1247 rte_spinlock_unlock(&mana_shared_data->lock); 1248 1249 return ret; 1250 } 1251 1252 /* 1253 * Probe an IB port 1254 * Return value: 1255 * positive value: successfully probed port 1256 * 0: port not matching specified MAC address 1257 * negative value: error code 1258 */ 1259 static int 1260 mana_probe_port(struct ibv_device *ibdev, struct ibv_device_attr_ex *dev_attr, 1261 uint8_t port, struct rte_pci_device *pci_dev, struct rte_ether_addr *addr) 1262 { 1263 struct mana_priv *priv = NULL; 1264 struct rte_eth_dev *eth_dev = NULL; 1265 struct ibv_parent_domain_init_attr attr = {0}; 1266 char address[64]; 1267 char name[RTE_ETH_NAME_MAX_LEN]; 1268 int ret; 1269 struct ibv_context *ctx = NULL; 1270 1271 rte_ether_format_addr(address, sizeof(address), addr); 1272 DRV_LOG(INFO, "device located port %u address %s", port, address); 1273 1274 priv = rte_zmalloc_socket(NULL, sizeof(*priv), RTE_CACHE_LINE_SIZE, 1275 SOCKET_ID_ANY); 1276 if (!priv) 1277 return -ENOMEM; 1278 1279 snprintf(name, sizeof(name), "%s_port%d", pci_dev->device.name, port); 1280 1281 if (rte_eal_process_type() == RTE_PROC_SECONDARY) { 1282 int fd; 1283 1284 eth_dev = rte_eth_dev_attach_secondary(name); 1285 if (!eth_dev) { 1286 DRV_LOG(ERR, "Can't attach to dev %s", name); 1287 ret = -ENOMEM; 1288 goto failed; 1289 } 1290 1291 eth_dev->device = &pci_dev->device; 1292 eth_dev->dev_ops = &mana_dev_secondary_ops; 1293 ret = mana_proc_priv_init(eth_dev); 1294 if (ret) 1295 goto failed; 1296 priv->process_priv = eth_dev->process_private; 1297 1298 /* Get the IB FD from the primary process */ 1299 fd = mana_mp_req_verbs_cmd_fd(eth_dev); 1300 if (fd < 0) { 1301 DRV_LOG(ERR, "Failed to get FD %d", fd); 1302 ret = -ENODEV; 1303 goto failed; 1304 } 1305 1306 ret = mana_map_doorbell_secondary(eth_dev, fd); 1307 if (ret) { 1308 DRV_LOG(ERR, "Failed secondary map %d", fd); 1309 goto failed; 1310 } 1311 1312 /* fd is no not used after mapping doorbell */ 1313 close(fd); 1314 1315 eth_dev->tx_pkt_burst = mana_tx_burst_removed; 1316 eth_dev->rx_pkt_burst = mana_rx_burst_removed; 1317 1318 rte_spinlock_lock(&mana_shared_data->lock); 1319 mana_shared_data->secondary_cnt++; 1320 mana_local_data.secondary_cnt++; 1321 rte_spinlock_unlock(&mana_shared_data->lock); 1322 1323 rte_eth_copy_pci_info(eth_dev, pci_dev); 1324 rte_eth_dev_probing_finish(eth_dev); 1325 1326 return 0; 1327 } 1328 1329 ctx = ibv_open_device(ibdev); 1330 if (!ctx) { 1331 DRV_LOG(ERR, "Failed to open IB device %s", ibdev->name); 1332 ret = -ENODEV; 1333 goto failed; 1334 } 1335 1336 eth_dev = rte_eth_dev_allocate(name); 1337 if (!eth_dev) { 1338 ret = -ENOMEM; 1339 goto failed; 1340 } 1341 1342 eth_dev->data->mac_addrs = 1343 rte_calloc("mana_mac", 1, 1344 sizeof(struct rte_ether_addr), 0); 1345 if (!eth_dev->data->mac_addrs) { 1346 ret = -ENOMEM; 1347 goto failed; 1348 } 1349 1350 rte_ether_addr_copy(addr, eth_dev->data->mac_addrs); 1351 1352 priv->ib_pd = ibv_alloc_pd(ctx); 1353 if (!priv->ib_pd) { 1354 DRV_LOG(ERR, "ibv_alloc_pd failed port %d", port); 1355 ret = -ENOMEM; 1356 goto failed; 1357 } 1358 1359 /* Create a parent domain with the port number */ 1360 attr.pd = priv->ib_pd; 1361 attr.comp_mask = IBV_PARENT_DOMAIN_INIT_ATTR_PD_CONTEXT; 1362 attr.pd_context = (void *)(uintptr_t)port; 1363 priv->ib_parent_pd = ibv_alloc_parent_domain(ctx, &attr); 1364 if (!priv->ib_parent_pd) { 1365 DRV_LOG(ERR, "ibv_alloc_parent_domain failed port %d", port); 1366 ret = -ENOMEM; 1367 goto failed; 1368 } 1369 1370 priv->ib_ctx = ctx; 1371 priv->port_id = eth_dev->data->port_id; 1372 priv->dev_port = port; 1373 eth_dev->data->dev_private = priv; 1374 priv->dev_data = eth_dev->data; 1375 1376 priv->max_rx_queues = dev_attr->orig_attr.max_qp; 1377 priv->max_tx_queues = dev_attr->orig_attr.max_qp; 1378 1379 priv->max_rx_desc = 1380 RTE_MIN(dev_attr->orig_attr.max_qp_wr, 1381 dev_attr->orig_attr.max_cqe); 1382 priv->max_tx_desc = 1383 RTE_MIN(dev_attr->orig_attr.max_qp_wr, 1384 dev_attr->orig_attr.max_cqe); 1385 1386 priv->max_send_sge = dev_attr->orig_attr.max_sge; 1387 priv->max_recv_sge = dev_attr->orig_attr.max_sge; 1388 1389 priv->max_mr = dev_attr->orig_attr.max_mr; 1390 priv->max_mr_size = dev_attr->orig_attr.max_mr_size; 1391 1392 DRV_LOG(INFO, "dev %s max queues %d desc %d sge %d mr %" PRIu64, 1393 name, priv->max_rx_queues, priv->max_rx_desc, 1394 priv->max_send_sge, priv->max_mr_size); 1395 1396 rte_eth_copy_pci_info(eth_dev, pci_dev); 1397 1398 /* Create async interrupt handler */ 1399 ret = mana_intr_install(eth_dev, priv); 1400 if (ret) { 1401 DRV_LOG(ERR, "Failed to install intr handler"); 1402 goto failed; 1403 } 1404 1405 rte_spinlock_lock(&mana_shared_data->lock); 1406 mana_shared_data->primary_cnt++; 1407 rte_spinlock_unlock(&mana_shared_data->lock); 1408 1409 eth_dev->device = &pci_dev->device; 1410 1411 DRV_LOG(INFO, "device %s at port %u", name, eth_dev->data->port_id); 1412 1413 eth_dev->rx_pkt_burst = mana_rx_burst_removed; 1414 eth_dev->tx_pkt_burst = mana_tx_burst_removed; 1415 eth_dev->dev_ops = &mana_dev_ops; 1416 1417 rte_eth_dev_probing_finish(eth_dev); 1418 1419 return 0; 1420 1421 failed: 1422 /* Free the resource for the port failed */ 1423 if (priv) { 1424 if (priv->ib_parent_pd) 1425 ibv_dealloc_pd(priv->ib_parent_pd); 1426 1427 if (priv->ib_pd) 1428 ibv_dealloc_pd(priv->ib_pd); 1429 } 1430 1431 if (eth_dev) 1432 rte_eth_dev_release_port(eth_dev); 1433 1434 rte_free(priv); 1435 1436 if (ctx) 1437 ibv_close_device(ctx); 1438 1439 return ret; 1440 } 1441 1442 /* 1443 * Goes through the IB device list to look for the IB port matching the 1444 * mac_addr. If found, create a rte_eth_dev for it. 1445 * Return value: number of successfully probed devices 1446 */ 1447 static int 1448 mana_pci_probe_mac(struct rte_pci_device *pci_dev, 1449 struct rte_ether_addr *mac_addr) 1450 { 1451 struct ibv_device **ibv_list; 1452 int ibv_idx; 1453 struct ibv_context *ctx; 1454 int num_devices; 1455 int ret; 1456 uint8_t port; 1457 int count = 0; 1458 1459 ibv_list = ibv_get_device_list(&num_devices); 1460 for (ibv_idx = 0; ibv_idx < num_devices; ibv_idx++) { 1461 struct ibv_device *ibdev = ibv_list[ibv_idx]; 1462 struct rte_pci_addr pci_addr; 1463 struct ibv_device_attr_ex dev_attr; 1464 1465 DRV_LOG(INFO, "Probe device name %s dev_name %s ibdev_path %s", 1466 ibdev->name, ibdev->dev_name, ibdev->ibdev_path); 1467 1468 if (mana_ibv_device_to_pci_addr(ibdev, &pci_addr)) 1469 continue; 1470 1471 /* Ignore if this IB device is not this PCI device */ 1472 if (rte_pci_addr_cmp(&pci_dev->addr, &pci_addr) != 0) 1473 continue; 1474 1475 ctx = ibv_open_device(ibdev); 1476 if (!ctx) { 1477 DRV_LOG(ERR, "Failed to open IB device %s", 1478 ibdev->name); 1479 continue; 1480 } 1481 ret = ibv_query_device_ex(ctx, NULL, &dev_attr); 1482 ibv_close_device(ctx); 1483 1484 if (ret) { 1485 DRV_LOG(ERR, "Failed to query IB device %s", 1486 ibdev->name); 1487 continue; 1488 } 1489 1490 for (port = 1; port <= dev_attr.orig_attr.phys_port_cnt; 1491 port++) { 1492 struct rte_ether_addr addr; 1493 ret = get_port_mac(ibdev, port, &addr); 1494 if (ret) 1495 continue; 1496 1497 if (mac_addr && !rte_is_same_ether_addr(&addr, mac_addr)) 1498 continue; 1499 1500 ret = mana_probe_port(ibdev, &dev_attr, port, pci_dev, &addr); 1501 if (ret) { 1502 DRV_LOG(ERR, "Probe on IB port %u failed %d", port, ret); 1503 } else { 1504 count++; 1505 DRV_LOG(INFO, "Successfully probed on IB port %u", port); 1506 } 1507 } 1508 } 1509 1510 ibv_free_device_list(ibv_list); 1511 return count; 1512 } 1513 1514 /* 1515 * Main callback function from PCI bus to probe a device. 1516 */ 1517 static int 1518 mana_pci_probe(struct rte_pci_driver *pci_drv __rte_unused, 1519 struct rte_pci_device *pci_dev) 1520 { 1521 struct rte_devargs *args = pci_dev->device.devargs; 1522 struct mana_conf conf = {0}; 1523 unsigned int i; 1524 int ret; 1525 int count = 0; 1526 1527 if (args && args->drv_str) { 1528 ret = mana_parse_args(args, &conf); 1529 if (ret) { 1530 DRV_LOG(ERR, "Failed to parse parameters args = %s", 1531 args->drv_str); 1532 return ret; 1533 } 1534 } 1535 1536 ret = mana_init_once(); 1537 if (ret) { 1538 DRV_LOG(ERR, "Failed to init PMD global data %d", ret); 1539 return ret; 1540 } 1541 1542 /* If there are no driver parameters, probe on all ports */ 1543 if (conf.index) { 1544 for (i = 0; i < conf.index; i++) 1545 count += mana_pci_probe_mac(pci_dev, 1546 &conf.mac_array[i]); 1547 } else { 1548 count = mana_pci_probe_mac(pci_dev, NULL); 1549 } 1550 1551 if (!count) { 1552 rte_memzone_free(mana_shared_mz); 1553 mana_shared_mz = NULL; 1554 ret = -ENODEV; 1555 } 1556 1557 return ret; 1558 } 1559 1560 static int 1561 mana_dev_uninit(struct rte_eth_dev *dev) 1562 { 1563 return mana_dev_close(dev); 1564 } 1565 1566 /* 1567 * Callback from PCI to remove this device. 1568 */ 1569 static int 1570 mana_pci_remove(struct rte_pci_device *pci_dev) 1571 { 1572 if (rte_eal_process_type() == RTE_PROC_PRIMARY) { 1573 rte_spinlock_lock(&mana_shared_data_lock); 1574 1575 rte_spinlock_lock(&mana_shared_data->lock); 1576 1577 RTE_VERIFY(mana_shared_data->primary_cnt > 0); 1578 mana_shared_data->primary_cnt--; 1579 if (!mana_shared_data->primary_cnt) { 1580 DRV_LOG(DEBUG, "mp uninit primary"); 1581 mana_mp_uninit_primary(); 1582 } 1583 1584 rte_spinlock_unlock(&mana_shared_data->lock); 1585 1586 /* Also free the shared memory if this is the last */ 1587 if (!mana_shared_data->primary_cnt) { 1588 DRV_LOG(DEBUG, "free shared memezone data"); 1589 rte_memzone_free(mana_shared_mz); 1590 mana_shared_mz = NULL; 1591 } 1592 1593 rte_spinlock_unlock(&mana_shared_data_lock); 1594 } else { 1595 rte_spinlock_lock(&mana_shared_data_lock); 1596 1597 rte_spinlock_lock(&mana_shared_data->lock); 1598 RTE_VERIFY(mana_shared_data->secondary_cnt > 0); 1599 mana_shared_data->secondary_cnt--; 1600 rte_spinlock_unlock(&mana_shared_data->lock); 1601 1602 RTE_VERIFY(mana_local_data.secondary_cnt > 0); 1603 mana_local_data.secondary_cnt--; 1604 if (!mana_local_data.secondary_cnt) { 1605 DRV_LOG(DEBUG, "mp uninit secondary"); 1606 mana_mp_uninit_secondary(); 1607 } 1608 1609 rte_spinlock_unlock(&mana_shared_data_lock); 1610 } 1611 1612 return rte_eth_dev_pci_generic_remove(pci_dev, mana_dev_uninit); 1613 } 1614 1615 static const struct rte_pci_id mana_pci_id_map[] = { 1616 { 1617 RTE_PCI_DEVICE(PCI_VENDOR_ID_MICROSOFT, 1618 PCI_DEVICE_ID_MICROSOFT_MANA) 1619 }, 1620 { 1621 .vendor_id = 0 1622 }, 1623 }; 1624 1625 static struct rte_pci_driver mana_pci_driver = { 1626 .id_table = mana_pci_id_map, 1627 .probe = mana_pci_probe, 1628 .remove = mana_pci_remove, 1629 .drv_flags = RTE_PCI_DRV_INTR_RMV, 1630 }; 1631 1632 RTE_PMD_REGISTER_PCI(net_mana, mana_pci_driver); 1633 RTE_PMD_REGISTER_PCI_TABLE(net_mana, mana_pci_id_map); 1634 RTE_PMD_REGISTER_KMOD_DEP(net_mana, "* ib_uverbs & mana_ib"); 1635 RTE_LOG_REGISTER_SUFFIX(mana_logtype_init, init, NOTICE); 1636 RTE_LOG_REGISTER_SUFFIX(mana_logtype_driver, driver, NOTICE); 1637 RTE_PMD_REGISTER_PARAM_STRING(net_mana, ETH_MANA_MAC_ARG "=<mac_addr>"); 1638