1 /* 2 * Copyright (c) 2004 Topspin Communications. All rights reserved. 3 * Copyright (c) 2005 Voltaire, Inc. All rights reserved. 4 * Copyright (c) 2006 Intel Corporation. All rights reserved. 5 * 6 * This software is available to you under a choice of one of two 7 * licenses. You may choose to be licensed under the terms of the GNU 8 * General Public License (GPL) Version 2, available from the file 9 * COPYING in the main directory of this source tree, or the 10 * OpenIB.org BSD license below: 11 * 12 * Redistribution and use in source and binary forms, with or 13 * without modification, are permitted provided that the following 14 * conditions are met: 15 * 16 * - Redistributions of source code must retain the above 17 * copyright notice, this list of conditions and the following 18 * disclaimer. 19 * 20 * - Redistributions in binary form must reproduce the above 21 * copyright notice, this list of conditions and the following 22 * disclaimer in the documentation and/or other materials 23 * provided with the distribution. 24 * 25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 32 * SOFTWARE. 33 */ 34 35 #include <linux/module.h> 36 #include <linux/init.h> 37 #include <linux/err.h> 38 #include <linux/random.h> 39 #include <linux/spinlock.h> 40 #include <linux/slab.h> 41 #include <linux/dma-mapping.h> 42 #include <linux/kref.h> 43 #include <linux/idr.h> 44 #include <linux/workqueue.h> 45 46 #include <rdma/ib_pack.h> 47 #include <rdma/ib_cache.h> 48 #include "sa.h" 49 50 MODULE_AUTHOR("Roland Dreier"); 51 MODULE_DESCRIPTION("InfiniBand subnet administration query support"); 52 MODULE_LICENSE("Dual BSD/GPL"); 53 54 struct ib_sa_sm_ah { 55 struct ib_ah *ah; 56 struct kref ref; 57 u16 pkey_index; 58 u8 src_path_mask; 59 }; 60 61 struct ib_sa_port { 62 struct ib_mad_agent *agent; 63 struct ib_sa_sm_ah *sm_ah; 64 struct work_struct update_task; 65 spinlock_t ah_lock; 66 u8 port_num; 67 }; 68 69 struct ib_sa_device { 70 int start_port, end_port; 71 struct ib_event_handler event_handler; 72 struct ib_sa_port port[0]; 73 }; 74 75 struct ib_sa_query { 76 void (*callback)(struct ib_sa_query *, int, struct ib_sa_mad *); 77 void (*release)(struct ib_sa_query *); 78 struct ib_sa_client *client; 79 struct ib_sa_port *port; 80 struct ib_mad_send_buf *mad_buf; 81 struct ib_sa_sm_ah *sm_ah; 82 int id; 83 }; 84 85 struct ib_sa_service_query { 86 void (*callback)(int, struct ib_sa_service_rec *, void *); 87 void *context; 88 struct ib_sa_query sa_query; 89 }; 90 91 struct ib_sa_path_query { 92 void (*callback)(int, struct ib_sa_path_rec *, void *); 93 void *context; 94 struct ib_sa_query sa_query; 95 }; 96 97 struct ib_sa_guidinfo_query { 98 void (*callback)(int, struct ib_sa_guidinfo_rec *, void *); 99 void *context; 100 struct ib_sa_query sa_query; 101 }; 102 103 struct ib_sa_mcmember_query { 104 void (*callback)(int, struct ib_sa_mcmember_rec *, void *); 105 void *context; 106 struct ib_sa_query sa_query; 107 }; 108 109 static void ib_sa_add_one(struct ib_device *device); 110 static void ib_sa_remove_one(struct ib_device *device); 111 112 static struct ib_client sa_client = { 113 .name = "sa", 114 .add = ib_sa_add_one, 115 .remove = ib_sa_remove_one 116 }; 117 118 static DEFINE_SPINLOCK(idr_lock); 119 static DEFINE_IDR(query_idr); 120 121 static DEFINE_SPINLOCK(tid_lock); 122 static u32 tid; 123 124 #define PATH_REC_FIELD(field) \ 125 .struct_offset_bytes = offsetof(struct ib_sa_path_rec, field), \ 126 .struct_size_bytes = sizeof ((struct ib_sa_path_rec *) 0)->field, \ 127 .field_name = "sa_path_rec:" #field 128 129 static const struct ib_field path_rec_table[] = { 130 { PATH_REC_FIELD(service_id), 131 .offset_words = 0, 132 .offset_bits = 0, 133 .size_bits = 64 }, 134 { PATH_REC_FIELD(dgid), 135 .offset_words = 2, 136 .offset_bits = 0, 137 .size_bits = 128 }, 138 { PATH_REC_FIELD(sgid), 139 .offset_words = 6, 140 .offset_bits = 0, 141 .size_bits = 128 }, 142 { PATH_REC_FIELD(dlid), 143 .offset_words = 10, 144 .offset_bits = 0, 145 .size_bits = 16 }, 146 { PATH_REC_FIELD(slid), 147 .offset_words = 10, 148 .offset_bits = 16, 149 .size_bits = 16 }, 150 { PATH_REC_FIELD(raw_traffic), 151 .offset_words = 11, 152 .offset_bits = 0, 153 .size_bits = 1 }, 154 { RESERVED, 155 .offset_words = 11, 156 .offset_bits = 1, 157 .size_bits = 3 }, 158 { PATH_REC_FIELD(flow_label), 159 .offset_words = 11, 160 .offset_bits = 4, 161 .size_bits = 20 }, 162 { PATH_REC_FIELD(hop_limit), 163 .offset_words = 11, 164 .offset_bits = 24, 165 .size_bits = 8 }, 166 { PATH_REC_FIELD(traffic_class), 167 .offset_words = 12, 168 .offset_bits = 0, 169 .size_bits = 8 }, 170 { PATH_REC_FIELD(reversible), 171 .offset_words = 12, 172 .offset_bits = 8, 173 .size_bits = 1 }, 174 { PATH_REC_FIELD(numb_path), 175 .offset_words = 12, 176 .offset_bits = 9, 177 .size_bits = 7 }, 178 { PATH_REC_FIELD(pkey), 179 .offset_words = 12, 180 .offset_bits = 16, 181 .size_bits = 16 }, 182 { PATH_REC_FIELD(qos_class), 183 .offset_words = 13, 184 .offset_bits = 0, 185 .size_bits = 12 }, 186 { PATH_REC_FIELD(sl), 187 .offset_words = 13, 188 .offset_bits = 12, 189 .size_bits = 4 }, 190 { PATH_REC_FIELD(mtu_selector), 191 .offset_words = 13, 192 .offset_bits = 16, 193 .size_bits = 2 }, 194 { PATH_REC_FIELD(mtu), 195 .offset_words = 13, 196 .offset_bits = 18, 197 .size_bits = 6 }, 198 { PATH_REC_FIELD(rate_selector), 199 .offset_words = 13, 200 .offset_bits = 24, 201 .size_bits = 2 }, 202 { PATH_REC_FIELD(rate), 203 .offset_words = 13, 204 .offset_bits = 26, 205 .size_bits = 6 }, 206 { PATH_REC_FIELD(packet_life_time_selector), 207 .offset_words = 14, 208 .offset_bits = 0, 209 .size_bits = 2 }, 210 { PATH_REC_FIELD(packet_life_time), 211 .offset_words = 14, 212 .offset_bits = 2, 213 .size_bits = 6 }, 214 { PATH_REC_FIELD(preference), 215 .offset_words = 14, 216 .offset_bits = 8, 217 .size_bits = 8 }, 218 { RESERVED, 219 .offset_words = 14, 220 .offset_bits = 16, 221 .size_bits = 48 }, 222 }; 223 224 #define MCMEMBER_REC_FIELD(field) \ 225 .struct_offset_bytes = offsetof(struct ib_sa_mcmember_rec, field), \ 226 .struct_size_bytes = sizeof ((struct ib_sa_mcmember_rec *) 0)->field, \ 227 .field_name = "sa_mcmember_rec:" #field 228 229 static const struct ib_field mcmember_rec_table[] = { 230 { MCMEMBER_REC_FIELD(mgid), 231 .offset_words = 0, 232 .offset_bits = 0, 233 .size_bits = 128 }, 234 { MCMEMBER_REC_FIELD(port_gid), 235 .offset_words = 4, 236 .offset_bits = 0, 237 .size_bits = 128 }, 238 { MCMEMBER_REC_FIELD(qkey), 239 .offset_words = 8, 240 .offset_bits = 0, 241 .size_bits = 32 }, 242 { MCMEMBER_REC_FIELD(mlid), 243 .offset_words = 9, 244 .offset_bits = 0, 245 .size_bits = 16 }, 246 { MCMEMBER_REC_FIELD(mtu_selector), 247 .offset_words = 9, 248 .offset_bits = 16, 249 .size_bits = 2 }, 250 { MCMEMBER_REC_FIELD(mtu), 251 .offset_words = 9, 252 .offset_bits = 18, 253 .size_bits = 6 }, 254 { MCMEMBER_REC_FIELD(traffic_class), 255 .offset_words = 9, 256 .offset_bits = 24, 257 .size_bits = 8 }, 258 { MCMEMBER_REC_FIELD(pkey), 259 .offset_words = 10, 260 .offset_bits = 0, 261 .size_bits = 16 }, 262 { MCMEMBER_REC_FIELD(rate_selector), 263 .offset_words = 10, 264 .offset_bits = 16, 265 .size_bits = 2 }, 266 { MCMEMBER_REC_FIELD(rate), 267 .offset_words = 10, 268 .offset_bits = 18, 269 .size_bits = 6 }, 270 { MCMEMBER_REC_FIELD(packet_life_time_selector), 271 .offset_words = 10, 272 .offset_bits = 24, 273 .size_bits = 2 }, 274 { MCMEMBER_REC_FIELD(packet_life_time), 275 .offset_words = 10, 276 .offset_bits = 26, 277 .size_bits = 6 }, 278 { MCMEMBER_REC_FIELD(sl), 279 .offset_words = 11, 280 .offset_bits = 0, 281 .size_bits = 4 }, 282 { MCMEMBER_REC_FIELD(flow_label), 283 .offset_words = 11, 284 .offset_bits = 4, 285 .size_bits = 20 }, 286 { MCMEMBER_REC_FIELD(hop_limit), 287 .offset_words = 11, 288 .offset_bits = 24, 289 .size_bits = 8 }, 290 { MCMEMBER_REC_FIELD(scope), 291 .offset_words = 12, 292 .offset_bits = 0, 293 .size_bits = 4 }, 294 { MCMEMBER_REC_FIELD(join_state), 295 .offset_words = 12, 296 .offset_bits = 4, 297 .size_bits = 4 }, 298 { MCMEMBER_REC_FIELD(proxy_join), 299 .offset_words = 12, 300 .offset_bits = 8, 301 .size_bits = 1 }, 302 { RESERVED, 303 .offset_words = 12, 304 .offset_bits = 9, 305 .size_bits = 23 }, 306 }; 307 308 #define SERVICE_REC_FIELD(field) \ 309 .struct_offset_bytes = offsetof(struct ib_sa_service_rec, field), \ 310 .struct_size_bytes = sizeof ((struct ib_sa_service_rec *) 0)->field, \ 311 .field_name = "sa_service_rec:" #field 312 313 static const struct ib_field service_rec_table[] = { 314 { SERVICE_REC_FIELD(id), 315 .offset_words = 0, 316 .offset_bits = 0, 317 .size_bits = 64 }, 318 { SERVICE_REC_FIELD(gid), 319 .offset_words = 2, 320 .offset_bits = 0, 321 .size_bits = 128 }, 322 { SERVICE_REC_FIELD(pkey), 323 .offset_words = 6, 324 .offset_bits = 0, 325 .size_bits = 16 }, 326 { SERVICE_REC_FIELD(lease), 327 .offset_words = 7, 328 .offset_bits = 0, 329 .size_bits = 32 }, 330 { SERVICE_REC_FIELD(key), 331 .offset_words = 8, 332 .offset_bits = 0, 333 .size_bits = 128 }, 334 { SERVICE_REC_FIELD(name), 335 .offset_words = 12, 336 .offset_bits = 0, 337 .size_bits = 64*8 }, 338 { SERVICE_REC_FIELD(data8), 339 .offset_words = 28, 340 .offset_bits = 0, 341 .size_bits = 16*8 }, 342 { SERVICE_REC_FIELD(data16), 343 .offset_words = 32, 344 .offset_bits = 0, 345 .size_bits = 8*16 }, 346 { SERVICE_REC_FIELD(data32), 347 .offset_words = 36, 348 .offset_bits = 0, 349 .size_bits = 4*32 }, 350 { SERVICE_REC_FIELD(data64), 351 .offset_words = 40, 352 .offset_bits = 0, 353 .size_bits = 2*64 }, 354 }; 355 356 #define GUIDINFO_REC_FIELD(field) \ 357 .struct_offset_bytes = offsetof(struct ib_sa_guidinfo_rec, field), \ 358 .struct_size_bytes = sizeof((struct ib_sa_guidinfo_rec *) 0)->field, \ 359 .field_name = "sa_guidinfo_rec:" #field 360 361 static const struct ib_field guidinfo_rec_table[] = { 362 { GUIDINFO_REC_FIELD(lid), 363 .offset_words = 0, 364 .offset_bits = 0, 365 .size_bits = 16 }, 366 { GUIDINFO_REC_FIELD(block_num), 367 .offset_words = 0, 368 .offset_bits = 16, 369 .size_bits = 8 }, 370 { GUIDINFO_REC_FIELD(res1), 371 .offset_words = 0, 372 .offset_bits = 24, 373 .size_bits = 8 }, 374 { GUIDINFO_REC_FIELD(res2), 375 .offset_words = 1, 376 .offset_bits = 0, 377 .size_bits = 32 }, 378 { GUIDINFO_REC_FIELD(guid_info_list), 379 .offset_words = 2, 380 .offset_bits = 0, 381 .size_bits = 512 }, 382 }; 383 384 static void free_sm_ah(struct kref *kref) 385 { 386 struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref); 387 388 ib_destroy_ah(sm_ah->ah); 389 kfree(sm_ah); 390 } 391 392 static void update_sm_ah(struct work_struct *work) 393 { 394 struct ib_sa_port *port = 395 container_of(work, struct ib_sa_port, update_task); 396 struct ib_sa_sm_ah *new_ah; 397 struct ib_port_attr port_attr; 398 struct ib_ah_attr ah_attr; 399 400 if (ib_query_port(port->agent->device, port->port_num, &port_attr)) { 401 printk(KERN_WARNING "Couldn't query port\n"); 402 return; 403 } 404 405 new_ah = kmalloc(sizeof *new_ah, GFP_KERNEL); 406 if (!new_ah) { 407 printk(KERN_WARNING "Couldn't allocate new SM AH\n"); 408 return; 409 } 410 411 kref_init(&new_ah->ref); 412 new_ah->src_path_mask = (1 << port_attr.lmc) - 1; 413 414 new_ah->pkey_index = 0; 415 if (ib_find_pkey(port->agent->device, port->port_num, 416 IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index)) 417 printk(KERN_ERR "Couldn't find index for default PKey\n"); 418 419 memset(&ah_attr, 0, sizeof ah_attr); 420 ah_attr.dlid = port_attr.sm_lid; 421 ah_attr.sl = port_attr.sm_sl; 422 ah_attr.port_num = port->port_num; 423 424 new_ah->ah = ib_create_ah(port->agent->qp->pd, &ah_attr); 425 if (IS_ERR(new_ah->ah)) { 426 printk(KERN_WARNING "Couldn't create new SM AH\n"); 427 kfree(new_ah); 428 return; 429 } 430 431 spin_lock_irq(&port->ah_lock); 432 if (port->sm_ah) 433 kref_put(&port->sm_ah->ref, free_sm_ah); 434 port->sm_ah = new_ah; 435 spin_unlock_irq(&port->ah_lock); 436 437 } 438 439 static void ib_sa_event(struct ib_event_handler *handler, struct ib_event *event) 440 { 441 if (event->event == IB_EVENT_PORT_ERR || 442 event->event == IB_EVENT_PORT_ACTIVE || 443 event->event == IB_EVENT_LID_CHANGE || 444 event->event == IB_EVENT_PKEY_CHANGE || 445 event->event == IB_EVENT_SM_CHANGE || 446 event->event == IB_EVENT_CLIENT_REREGISTER) { 447 unsigned long flags; 448 struct ib_sa_device *sa_dev = 449 container_of(handler, typeof(*sa_dev), event_handler); 450 struct ib_sa_port *port = 451 &sa_dev->port[event->element.port_num - sa_dev->start_port]; 452 453 if (rdma_port_get_link_layer(handler->device, port->port_num) != IB_LINK_LAYER_INFINIBAND) 454 return; 455 456 spin_lock_irqsave(&port->ah_lock, flags); 457 if (port->sm_ah) 458 kref_put(&port->sm_ah->ref, free_sm_ah); 459 port->sm_ah = NULL; 460 spin_unlock_irqrestore(&port->ah_lock, flags); 461 462 queue_work(ib_wq, &sa_dev->port[event->element.port_num - 463 sa_dev->start_port].update_task); 464 } 465 } 466 467 void ib_sa_register_client(struct ib_sa_client *client) 468 { 469 atomic_set(&client->users, 1); 470 init_completion(&client->comp); 471 } 472 EXPORT_SYMBOL(ib_sa_register_client); 473 474 void ib_sa_unregister_client(struct ib_sa_client *client) 475 { 476 ib_sa_client_put(client); 477 wait_for_completion(&client->comp); 478 } 479 EXPORT_SYMBOL(ib_sa_unregister_client); 480 481 /** 482 * ib_sa_cancel_query - try to cancel an SA query 483 * @id:ID of query to cancel 484 * @query:query pointer to cancel 485 * 486 * Try to cancel an SA query. If the id and query don't match up or 487 * the query has already completed, nothing is done. Otherwise the 488 * query is canceled and will complete with a status of -EINTR. 489 */ 490 void ib_sa_cancel_query(int id, struct ib_sa_query *query) 491 { 492 unsigned long flags; 493 struct ib_mad_agent *agent; 494 struct ib_mad_send_buf *mad_buf; 495 496 spin_lock_irqsave(&idr_lock, flags); 497 if (idr_find(&query_idr, id) != query) { 498 spin_unlock_irqrestore(&idr_lock, flags); 499 return; 500 } 501 agent = query->port->agent; 502 mad_buf = query->mad_buf; 503 spin_unlock_irqrestore(&idr_lock, flags); 504 505 ib_cancel_mad(agent, mad_buf); 506 } 507 EXPORT_SYMBOL(ib_sa_cancel_query); 508 509 static u8 get_src_path_mask(struct ib_device *device, u8 port_num) 510 { 511 struct ib_sa_device *sa_dev; 512 struct ib_sa_port *port; 513 unsigned long flags; 514 u8 src_path_mask; 515 516 sa_dev = ib_get_client_data(device, &sa_client); 517 if (!sa_dev) 518 return 0x7f; 519 520 port = &sa_dev->port[port_num - sa_dev->start_port]; 521 spin_lock_irqsave(&port->ah_lock, flags); 522 src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f; 523 spin_unlock_irqrestore(&port->ah_lock, flags); 524 525 return src_path_mask; 526 } 527 528 int ib_init_ah_from_path(struct ib_device *device, u8 port_num, 529 struct ib_sa_path_rec *rec, struct ib_ah_attr *ah_attr) 530 { 531 int ret; 532 u16 gid_index; 533 int force_grh; 534 535 memset(ah_attr, 0, sizeof *ah_attr); 536 ah_attr->dlid = be16_to_cpu(rec->dlid); 537 ah_attr->sl = rec->sl; 538 ah_attr->src_path_bits = be16_to_cpu(rec->slid) & 539 get_src_path_mask(device, port_num); 540 ah_attr->port_num = port_num; 541 ah_attr->static_rate = rec->rate; 542 543 force_grh = rdma_port_get_link_layer(device, port_num) == IB_LINK_LAYER_ETHERNET; 544 545 if (rec->hop_limit > 1 || force_grh) { 546 ah_attr->ah_flags = IB_AH_GRH; 547 ah_attr->grh.dgid = rec->dgid; 548 549 ret = ib_find_cached_gid(device, &rec->sgid, &port_num, 550 &gid_index); 551 if (ret) 552 return ret; 553 554 ah_attr->grh.sgid_index = gid_index; 555 ah_attr->grh.flow_label = be32_to_cpu(rec->flow_label); 556 ah_attr->grh.hop_limit = rec->hop_limit; 557 ah_attr->grh.traffic_class = rec->traffic_class; 558 } 559 return 0; 560 } 561 EXPORT_SYMBOL(ib_init_ah_from_path); 562 563 static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask) 564 { 565 unsigned long flags; 566 567 spin_lock_irqsave(&query->port->ah_lock, flags); 568 if (!query->port->sm_ah) { 569 spin_unlock_irqrestore(&query->port->ah_lock, flags); 570 return -EAGAIN; 571 } 572 kref_get(&query->port->sm_ah->ref); 573 query->sm_ah = query->port->sm_ah; 574 spin_unlock_irqrestore(&query->port->ah_lock, flags); 575 576 query->mad_buf = ib_create_send_mad(query->port->agent, 1, 577 query->sm_ah->pkey_index, 578 0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA, 579 gfp_mask); 580 if (IS_ERR(query->mad_buf)) { 581 kref_put(&query->sm_ah->ref, free_sm_ah); 582 return -ENOMEM; 583 } 584 585 query->mad_buf->ah = query->sm_ah->ah; 586 587 return 0; 588 } 589 590 static void free_mad(struct ib_sa_query *query) 591 { 592 ib_free_send_mad(query->mad_buf); 593 kref_put(&query->sm_ah->ref, free_sm_ah); 594 } 595 596 static void init_mad(struct ib_sa_mad *mad, struct ib_mad_agent *agent) 597 { 598 unsigned long flags; 599 600 memset(mad, 0, sizeof *mad); 601 602 mad->mad_hdr.base_version = IB_MGMT_BASE_VERSION; 603 mad->mad_hdr.mgmt_class = IB_MGMT_CLASS_SUBN_ADM; 604 mad->mad_hdr.class_version = IB_SA_CLASS_VERSION; 605 606 spin_lock_irqsave(&tid_lock, flags); 607 mad->mad_hdr.tid = 608 cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++); 609 spin_unlock_irqrestore(&tid_lock, flags); 610 } 611 612 static int send_mad(struct ib_sa_query *query, int timeout_ms, gfp_t gfp_mask) 613 { 614 bool preload = gfp_mask & __GFP_WAIT; 615 unsigned long flags; 616 int ret, id; 617 618 if (preload) 619 idr_preload(gfp_mask); 620 spin_lock_irqsave(&idr_lock, flags); 621 622 id = idr_alloc(&query_idr, query, 0, 0, GFP_NOWAIT); 623 624 spin_unlock_irqrestore(&idr_lock, flags); 625 if (preload) 626 idr_preload_end(); 627 if (id < 0) 628 return id; 629 630 query->mad_buf->timeout_ms = timeout_ms; 631 query->mad_buf->context[0] = query; 632 query->id = id; 633 634 ret = ib_post_send_mad(query->mad_buf, NULL); 635 if (ret) { 636 spin_lock_irqsave(&idr_lock, flags); 637 idr_remove(&query_idr, id); 638 spin_unlock_irqrestore(&idr_lock, flags); 639 } 640 641 /* 642 * It's not safe to dereference query any more, because the 643 * send may already have completed and freed the query in 644 * another context. 645 */ 646 return ret ? ret : id; 647 } 648 649 void ib_sa_unpack_path(void *attribute, struct ib_sa_path_rec *rec) 650 { 651 ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), attribute, rec); 652 } 653 EXPORT_SYMBOL(ib_sa_unpack_path); 654 655 void ib_sa_pack_path(struct ib_sa_path_rec *rec, void *attribute) 656 { 657 ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, attribute); 658 } 659 EXPORT_SYMBOL(ib_sa_pack_path); 660 661 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query, 662 int status, 663 struct ib_sa_mad *mad) 664 { 665 struct ib_sa_path_query *query = 666 container_of(sa_query, struct ib_sa_path_query, sa_query); 667 668 if (mad) { 669 struct ib_sa_path_rec rec; 670 671 ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), 672 mad->data, &rec); 673 query->callback(status, &rec, query->context); 674 } else 675 query->callback(status, NULL, query->context); 676 } 677 678 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query) 679 { 680 kfree(container_of(sa_query, struct ib_sa_path_query, sa_query)); 681 } 682 683 /** 684 * ib_sa_path_rec_get - Start a Path get query 685 * @client:SA client 686 * @device:device to send query on 687 * @port_num: port number to send query on 688 * @rec:Path Record to send in query 689 * @comp_mask:component mask to send in query 690 * @timeout_ms:time to wait for response 691 * @gfp_mask:GFP mask to use for internal allocations 692 * @callback:function called when query completes, times out or is 693 * canceled 694 * @context:opaque user context passed to callback 695 * @sa_query:query context, used to cancel query 696 * 697 * Send a Path Record Get query to the SA to look up a path. The 698 * callback function will be called when the query completes (or 699 * fails); status is 0 for a successful response, -EINTR if the query 700 * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error 701 * occurred sending the query. The resp parameter of the callback is 702 * only valid if status is 0. 703 * 704 * If the return value of ib_sa_path_rec_get() is negative, it is an 705 * error code. Otherwise it is a query ID that can be used to cancel 706 * the query. 707 */ 708 int ib_sa_path_rec_get(struct ib_sa_client *client, 709 struct ib_device *device, u8 port_num, 710 struct ib_sa_path_rec *rec, 711 ib_sa_comp_mask comp_mask, 712 int timeout_ms, gfp_t gfp_mask, 713 void (*callback)(int status, 714 struct ib_sa_path_rec *resp, 715 void *context), 716 void *context, 717 struct ib_sa_query **sa_query) 718 { 719 struct ib_sa_path_query *query; 720 struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client); 721 struct ib_sa_port *port; 722 struct ib_mad_agent *agent; 723 struct ib_sa_mad *mad; 724 int ret; 725 726 if (!sa_dev) 727 return -ENODEV; 728 729 port = &sa_dev->port[port_num - sa_dev->start_port]; 730 agent = port->agent; 731 732 query = kmalloc(sizeof *query, gfp_mask); 733 if (!query) 734 return -ENOMEM; 735 736 query->sa_query.port = port; 737 ret = alloc_mad(&query->sa_query, gfp_mask); 738 if (ret) 739 goto err1; 740 741 ib_sa_client_get(client); 742 query->sa_query.client = client; 743 query->callback = callback; 744 query->context = context; 745 746 mad = query->sa_query.mad_buf->mad; 747 init_mad(mad, agent); 748 749 query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL; 750 query->sa_query.release = ib_sa_path_rec_release; 751 mad->mad_hdr.method = IB_MGMT_METHOD_GET; 752 mad->mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_PATH_REC); 753 mad->sa_hdr.comp_mask = comp_mask; 754 755 ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, mad->data); 756 757 *sa_query = &query->sa_query; 758 759 ret = send_mad(&query->sa_query, timeout_ms, gfp_mask); 760 if (ret < 0) 761 goto err2; 762 763 return ret; 764 765 err2: 766 *sa_query = NULL; 767 ib_sa_client_put(query->sa_query.client); 768 free_mad(&query->sa_query); 769 770 err1: 771 kfree(query); 772 return ret; 773 } 774 EXPORT_SYMBOL(ib_sa_path_rec_get); 775 776 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query, 777 int status, 778 struct ib_sa_mad *mad) 779 { 780 struct ib_sa_service_query *query = 781 container_of(sa_query, struct ib_sa_service_query, sa_query); 782 783 if (mad) { 784 struct ib_sa_service_rec rec; 785 786 ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table), 787 mad->data, &rec); 788 query->callback(status, &rec, query->context); 789 } else 790 query->callback(status, NULL, query->context); 791 } 792 793 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query) 794 { 795 kfree(container_of(sa_query, struct ib_sa_service_query, sa_query)); 796 } 797 798 /** 799 * ib_sa_service_rec_query - Start Service Record operation 800 * @client:SA client 801 * @device:device to send request on 802 * @port_num: port number to send request on 803 * @method:SA method - should be get, set, or delete 804 * @rec:Service Record to send in request 805 * @comp_mask:component mask to send in request 806 * @timeout_ms:time to wait for response 807 * @gfp_mask:GFP mask to use for internal allocations 808 * @callback:function called when request completes, times out or is 809 * canceled 810 * @context:opaque user context passed to callback 811 * @sa_query:request context, used to cancel request 812 * 813 * Send a Service Record set/get/delete to the SA to register, 814 * unregister or query a service record. 815 * The callback function will be called when the request completes (or 816 * fails); status is 0 for a successful response, -EINTR if the query 817 * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error 818 * occurred sending the query. The resp parameter of the callback is 819 * only valid if status is 0. 820 * 821 * If the return value of ib_sa_service_rec_query() is negative, it is an 822 * error code. Otherwise it is a request ID that can be used to cancel 823 * the query. 824 */ 825 int ib_sa_service_rec_query(struct ib_sa_client *client, 826 struct ib_device *device, u8 port_num, u8 method, 827 struct ib_sa_service_rec *rec, 828 ib_sa_comp_mask comp_mask, 829 int timeout_ms, gfp_t gfp_mask, 830 void (*callback)(int status, 831 struct ib_sa_service_rec *resp, 832 void *context), 833 void *context, 834 struct ib_sa_query **sa_query) 835 { 836 struct ib_sa_service_query *query; 837 struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client); 838 struct ib_sa_port *port; 839 struct ib_mad_agent *agent; 840 struct ib_sa_mad *mad; 841 int ret; 842 843 if (!sa_dev) 844 return -ENODEV; 845 846 port = &sa_dev->port[port_num - sa_dev->start_port]; 847 agent = port->agent; 848 849 if (method != IB_MGMT_METHOD_GET && 850 method != IB_MGMT_METHOD_SET && 851 method != IB_SA_METHOD_DELETE) 852 return -EINVAL; 853 854 query = kmalloc(sizeof *query, gfp_mask); 855 if (!query) 856 return -ENOMEM; 857 858 query->sa_query.port = port; 859 ret = alloc_mad(&query->sa_query, gfp_mask); 860 if (ret) 861 goto err1; 862 863 ib_sa_client_get(client); 864 query->sa_query.client = client; 865 query->callback = callback; 866 query->context = context; 867 868 mad = query->sa_query.mad_buf->mad; 869 init_mad(mad, agent); 870 871 query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL; 872 query->sa_query.release = ib_sa_service_rec_release; 873 mad->mad_hdr.method = method; 874 mad->mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_SERVICE_REC); 875 mad->sa_hdr.comp_mask = comp_mask; 876 877 ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table), 878 rec, mad->data); 879 880 *sa_query = &query->sa_query; 881 882 ret = send_mad(&query->sa_query, timeout_ms, gfp_mask); 883 if (ret < 0) 884 goto err2; 885 886 return ret; 887 888 err2: 889 *sa_query = NULL; 890 ib_sa_client_put(query->sa_query.client); 891 free_mad(&query->sa_query); 892 893 err1: 894 kfree(query); 895 return ret; 896 } 897 EXPORT_SYMBOL(ib_sa_service_rec_query); 898 899 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query, 900 int status, 901 struct ib_sa_mad *mad) 902 { 903 struct ib_sa_mcmember_query *query = 904 container_of(sa_query, struct ib_sa_mcmember_query, sa_query); 905 906 if (mad) { 907 struct ib_sa_mcmember_rec rec; 908 909 ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table), 910 mad->data, &rec); 911 query->callback(status, &rec, query->context); 912 } else 913 query->callback(status, NULL, query->context); 914 } 915 916 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query) 917 { 918 kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query)); 919 } 920 921 int ib_sa_mcmember_rec_query(struct ib_sa_client *client, 922 struct ib_device *device, u8 port_num, 923 u8 method, 924 struct ib_sa_mcmember_rec *rec, 925 ib_sa_comp_mask comp_mask, 926 int timeout_ms, gfp_t gfp_mask, 927 void (*callback)(int status, 928 struct ib_sa_mcmember_rec *resp, 929 void *context), 930 void *context, 931 struct ib_sa_query **sa_query) 932 { 933 struct ib_sa_mcmember_query *query; 934 struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client); 935 struct ib_sa_port *port; 936 struct ib_mad_agent *agent; 937 struct ib_sa_mad *mad; 938 int ret; 939 940 if (!sa_dev) 941 return -ENODEV; 942 943 port = &sa_dev->port[port_num - sa_dev->start_port]; 944 agent = port->agent; 945 946 query = kmalloc(sizeof *query, gfp_mask); 947 if (!query) 948 return -ENOMEM; 949 950 query->sa_query.port = port; 951 ret = alloc_mad(&query->sa_query, gfp_mask); 952 if (ret) 953 goto err1; 954 955 ib_sa_client_get(client); 956 query->sa_query.client = client; 957 query->callback = callback; 958 query->context = context; 959 960 mad = query->sa_query.mad_buf->mad; 961 init_mad(mad, agent); 962 963 query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL; 964 query->sa_query.release = ib_sa_mcmember_rec_release; 965 mad->mad_hdr.method = method; 966 mad->mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC); 967 mad->sa_hdr.comp_mask = comp_mask; 968 969 ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table), 970 rec, mad->data); 971 972 *sa_query = &query->sa_query; 973 974 ret = send_mad(&query->sa_query, timeout_ms, gfp_mask); 975 if (ret < 0) 976 goto err2; 977 978 return ret; 979 980 err2: 981 *sa_query = NULL; 982 ib_sa_client_put(query->sa_query.client); 983 free_mad(&query->sa_query); 984 985 err1: 986 kfree(query); 987 return ret; 988 } 989 990 /* Support GuidInfoRecord */ 991 static void ib_sa_guidinfo_rec_callback(struct ib_sa_query *sa_query, 992 int status, 993 struct ib_sa_mad *mad) 994 { 995 struct ib_sa_guidinfo_query *query = 996 container_of(sa_query, struct ib_sa_guidinfo_query, sa_query); 997 998 if (mad) { 999 struct ib_sa_guidinfo_rec rec; 1000 1001 ib_unpack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table), 1002 mad->data, &rec); 1003 query->callback(status, &rec, query->context); 1004 } else 1005 query->callback(status, NULL, query->context); 1006 } 1007 1008 static void ib_sa_guidinfo_rec_release(struct ib_sa_query *sa_query) 1009 { 1010 kfree(container_of(sa_query, struct ib_sa_guidinfo_query, sa_query)); 1011 } 1012 1013 int ib_sa_guid_info_rec_query(struct ib_sa_client *client, 1014 struct ib_device *device, u8 port_num, 1015 struct ib_sa_guidinfo_rec *rec, 1016 ib_sa_comp_mask comp_mask, u8 method, 1017 int timeout_ms, gfp_t gfp_mask, 1018 void (*callback)(int status, 1019 struct ib_sa_guidinfo_rec *resp, 1020 void *context), 1021 void *context, 1022 struct ib_sa_query **sa_query) 1023 { 1024 struct ib_sa_guidinfo_query *query; 1025 struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client); 1026 struct ib_sa_port *port; 1027 struct ib_mad_agent *agent; 1028 struct ib_sa_mad *mad; 1029 int ret; 1030 1031 if (!sa_dev) 1032 return -ENODEV; 1033 1034 if (method != IB_MGMT_METHOD_GET && 1035 method != IB_MGMT_METHOD_SET && 1036 method != IB_SA_METHOD_DELETE) { 1037 return -EINVAL; 1038 } 1039 1040 port = &sa_dev->port[port_num - sa_dev->start_port]; 1041 agent = port->agent; 1042 1043 query = kmalloc(sizeof *query, gfp_mask); 1044 if (!query) 1045 return -ENOMEM; 1046 1047 query->sa_query.port = port; 1048 ret = alloc_mad(&query->sa_query, gfp_mask); 1049 if (ret) 1050 goto err1; 1051 1052 ib_sa_client_get(client); 1053 query->sa_query.client = client; 1054 query->callback = callback; 1055 query->context = context; 1056 1057 mad = query->sa_query.mad_buf->mad; 1058 init_mad(mad, agent); 1059 1060 query->sa_query.callback = callback ? ib_sa_guidinfo_rec_callback : NULL; 1061 query->sa_query.release = ib_sa_guidinfo_rec_release; 1062 1063 mad->mad_hdr.method = method; 1064 mad->mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_GUID_INFO_REC); 1065 mad->sa_hdr.comp_mask = comp_mask; 1066 1067 ib_pack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table), rec, 1068 mad->data); 1069 1070 *sa_query = &query->sa_query; 1071 1072 ret = send_mad(&query->sa_query, timeout_ms, gfp_mask); 1073 if (ret < 0) 1074 goto err2; 1075 1076 return ret; 1077 1078 err2: 1079 *sa_query = NULL; 1080 ib_sa_client_put(query->sa_query.client); 1081 free_mad(&query->sa_query); 1082 1083 err1: 1084 kfree(query); 1085 return ret; 1086 } 1087 EXPORT_SYMBOL(ib_sa_guid_info_rec_query); 1088 1089 static void send_handler(struct ib_mad_agent *agent, 1090 struct ib_mad_send_wc *mad_send_wc) 1091 { 1092 struct ib_sa_query *query = mad_send_wc->send_buf->context[0]; 1093 unsigned long flags; 1094 1095 if (query->callback) 1096 switch (mad_send_wc->status) { 1097 case IB_WC_SUCCESS: 1098 /* No callback -- already got recv */ 1099 break; 1100 case IB_WC_RESP_TIMEOUT_ERR: 1101 query->callback(query, -ETIMEDOUT, NULL); 1102 break; 1103 case IB_WC_WR_FLUSH_ERR: 1104 query->callback(query, -EINTR, NULL); 1105 break; 1106 default: 1107 query->callback(query, -EIO, NULL); 1108 break; 1109 } 1110 1111 spin_lock_irqsave(&idr_lock, flags); 1112 idr_remove(&query_idr, query->id); 1113 spin_unlock_irqrestore(&idr_lock, flags); 1114 1115 free_mad(query); 1116 ib_sa_client_put(query->client); 1117 query->release(query); 1118 } 1119 1120 static void recv_handler(struct ib_mad_agent *mad_agent, 1121 struct ib_mad_recv_wc *mad_recv_wc) 1122 { 1123 struct ib_sa_query *query; 1124 struct ib_mad_send_buf *mad_buf; 1125 1126 mad_buf = (void *) (unsigned long) mad_recv_wc->wc->wr_id; 1127 query = mad_buf->context[0]; 1128 1129 if (query->callback) { 1130 if (mad_recv_wc->wc->status == IB_WC_SUCCESS) 1131 query->callback(query, 1132 mad_recv_wc->recv_buf.mad->mad_hdr.status ? 1133 -EINVAL : 0, 1134 (struct ib_sa_mad *) mad_recv_wc->recv_buf.mad); 1135 else 1136 query->callback(query, -EIO, NULL); 1137 } 1138 1139 ib_free_recv_mad(mad_recv_wc); 1140 } 1141 1142 static void ib_sa_add_one(struct ib_device *device) 1143 { 1144 struct ib_sa_device *sa_dev; 1145 int s, e, i; 1146 1147 if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB) 1148 return; 1149 1150 if (device->node_type == RDMA_NODE_IB_SWITCH) 1151 s = e = 0; 1152 else { 1153 s = 1; 1154 e = device->phys_port_cnt; 1155 } 1156 1157 sa_dev = kzalloc(sizeof *sa_dev + 1158 (e - s + 1) * sizeof (struct ib_sa_port), 1159 GFP_KERNEL); 1160 if (!sa_dev) 1161 return; 1162 1163 sa_dev->start_port = s; 1164 sa_dev->end_port = e; 1165 1166 for (i = 0; i <= e - s; ++i) { 1167 spin_lock_init(&sa_dev->port[i].ah_lock); 1168 if (rdma_port_get_link_layer(device, i + 1) != IB_LINK_LAYER_INFINIBAND) 1169 continue; 1170 1171 sa_dev->port[i].sm_ah = NULL; 1172 sa_dev->port[i].port_num = i + s; 1173 1174 sa_dev->port[i].agent = 1175 ib_register_mad_agent(device, i + s, IB_QPT_GSI, 1176 NULL, 0, send_handler, 1177 recv_handler, sa_dev); 1178 if (IS_ERR(sa_dev->port[i].agent)) 1179 goto err; 1180 1181 INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah); 1182 } 1183 1184 ib_set_client_data(device, &sa_client, sa_dev); 1185 1186 /* 1187 * We register our event handler after everything is set up, 1188 * and then update our cached info after the event handler is 1189 * registered to avoid any problems if a port changes state 1190 * during our initialization. 1191 */ 1192 1193 INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event); 1194 if (ib_register_event_handler(&sa_dev->event_handler)) 1195 goto err; 1196 1197 for (i = 0; i <= e - s; ++i) 1198 if (rdma_port_get_link_layer(device, i + 1) == IB_LINK_LAYER_INFINIBAND) 1199 update_sm_ah(&sa_dev->port[i].update_task); 1200 1201 return; 1202 1203 err: 1204 while (--i >= 0) 1205 if (rdma_port_get_link_layer(device, i + 1) == IB_LINK_LAYER_INFINIBAND) 1206 ib_unregister_mad_agent(sa_dev->port[i].agent); 1207 1208 kfree(sa_dev); 1209 1210 return; 1211 } 1212 1213 static void ib_sa_remove_one(struct ib_device *device) 1214 { 1215 struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client); 1216 int i; 1217 1218 if (!sa_dev) 1219 return; 1220 1221 ib_unregister_event_handler(&sa_dev->event_handler); 1222 1223 flush_workqueue(ib_wq); 1224 1225 for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) { 1226 if (rdma_port_get_link_layer(device, i + 1) == IB_LINK_LAYER_INFINIBAND) { 1227 ib_unregister_mad_agent(sa_dev->port[i].agent); 1228 if (sa_dev->port[i].sm_ah) 1229 kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah); 1230 } 1231 1232 } 1233 1234 kfree(sa_dev); 1235 } 1236 1237 static int __init ib_sa_init(void) 1238 { 1239 int ret; 1240 1241 get_random_bytes(&tid, sizeof tid); 1242 1243 ret = ib_register_client(&sa_client); 1244 if (ret) { 1245 printk(KERN_ERR "Couldn't register ib_sa client\n"); 1246 goto err1; 1247 } 1248 1249 ret = mcast_init(); 1250 if (ret) { 1251 printk(KERN_ERR "Couldn't initialize multicast handling\n"); 1252 goto err2; 1253 } 1254 1255 return 0; 1256 err2: 1257 ib_unregister_client(&sa_client); 1258 err1: 1259 return ret; 1260 } 1261 1262 static void __exit ib_sa_cleanup(void) 1263 { 1264 mcast_cleanup(); 1265 ib_unregister_client(&sa_client); 1266 idr_destroy(&query_idr); 1267 } 1268 1269 module_init(ib_sa_init); 1270 module_exit(ib_sa_cleanup); 1271