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 * $Id: sa_query.c 2811 2005-07-06 18:11:43Z halr $ 35 */ 36 37 #include <linux/module.h> 38 #include <linux/init.h> 39 #include <linux/err.h> 40 #include <linux/random.h> 41 #include <linux/spinlock.h> 42 #include <linux/slab.h> 43 #include <linux/dma-mapping.h> 44 #include <linux/kref.h> 45 #include <linux/idr.h> 46 #include <linux/workqueue.h> 47 48 #include <rdma/ib_pack.h> 49 #include <rdma/ib_cache.h> 50 #include "sa.h" 51 52 MODULE_AUTHOR("Roland Dreier"); 53 MODULE_DESCRIPTION("InfiniBand subnet administration query support"); 54 MODULE_LICENSE("Dual BSD/GPL"); 55 56 struct ib_sa_sm_ah { 57 struct ib_ah *ah; 58 struct kref ref; 59 u16 pkey_index; 60 u8 src_path_mask; 61 }; 62 63 struct ib_sa_port { 64 struct ib_mad_agent *agent; 65 struct ib_sa_sm_ah *sm_ah; 66 struct work_struct update_task; 67 spinlock_t ah_lock; 68 u8 port_num; 69 }; 70 71 struct ib_sa_device { 72 int start_port, end_port; 73 struct ib_event_handler event_handler; 74 struct ib_sa_port port[0]; 75 }; 76 77 struct ib_sa_query { 78 void (*callback)(struct ib_sa_query *, int, struct ib_sa_mad *); 79 void (*release)(struct ib_sa_query *); 80 struct ib_sa_client *client; 81 struct ib_sa_port *port; 82 struct ib_mad_send_buf *mad_buf; 83 struct ib_sa_sm_ah *sm_ah; 84 int id; 85 }; 86 87 struct ib_sa_service_query { 88 void (*callback)(int, struct ib_sa_service_rec *, void *); 89 void *context; 90 struct ib_sa_query sa_query; 91 }; 92 93 struct ib_sa_path_query { 94 void (*callback)(int, struct ib_sa_path_rec *, void *); 95 void *context; 96 struct ib_sa_query sa_query; 97 }; 98 99 struct ib_sa_mcmember_query { 100 void (*callback)(int, struct ib_sa_mcmember_rec *, void *); 101 void *context; 102 struct ib_sa_query sa_query; 103 }; 104 105 static void ib_sa_add_one(struct ib_device *device); 106 static void ib_sa_remove_one(struct ib_device *device); 107 108 static struct ib_client sa_client = { 109 .name = "sa", 110 .add = ib_sa_add_one, 111 .remove = ib_sa_remove_one 112 }; 113 114 static spinlock_t idr_lock; 115 static DEFINE_IDR(query_idr); 116 117 static spinlock_t tid_lock; 118 static u32 tid; 119 120 #define PATH_REC_FIELD(field) \ 121 .struct_offset_bytes = offsetof(struct ib_sa_path_rec, field), \ 122 .struct_size_bytes = sizeof ((struct ib_sa_path_rec *) 0)->field, \ 123 .field_name = "sa_path_rec:" #field 124 125 static const struct ib_field path_rec_table[] = { 126 { RESERVED, 127 .offset_words = 0, 128 .offset_bits = 0, 129 .size_bits = 32 }, 130 { RESERVED, 131 .offset_words = 1, 132 .offset_bits = 0, 133 .size_bits = 32 }, 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 { RESERVED, 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 static void free_sm_ah(struct kref *kref) 357 { 358 struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref); 359 360 ib_destroy_ah(sm_ah->ah); 361 kfree(sm_ah); 362 } 363 364 static void update_sm_ah(struct work_struct *work) 365 { 366 struct ib_sa_port *port = 367 container_of(work, struct ib_sa_port, update_task); 368 struct ib_sa_sm_ah *new_ah, *old_ah; 369 struct ib_port_attr port_attr; 370 struct ib_ah_attr ah_attr; 371 372 if (ib_query_port(port->agent->device, port->port_num, &port_attr)) { 373 printk(KERN_WARNING "Couldn't query port\n"); 374 return; 375 } 376 377 new_ah = kmalloc(sizeof *new_ah, GFP_KERNEL); 378 if (!new_ah) { 379 printk(KERN_WARNING "Couldn't allocate new SM AH\n"); 380 return; 381 } 382 383 kref_init(&new_ah->ref); 384 new_ah->src_path_mask = (1 << port_attr.lmc) - 1; 385 386 new_ah->pkey_index = 0; 387 if (ib_find_pkey(port->agent->device, port->port_num, 388 IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index) && 389 ib_find_pkey(port->agent->device, port->port_num, 390 IB_DEFAULT_PKEY_PARTIAL, &new_ah->pkey_index)) 391 printk(KERN_ERR "Couldn't find index for default PKey\n"); 392 393 memset(&ah_attr, 0, sizeof ah_attr); 394 ah_attr.dlid = port_attr.sm_lid; 395 ah_attr.sl = port_attr.sm_sl; 396 ah_attr.port_num = port->port_num; 397 398 new_ah->ah = ib_create_ah(port->agent->qp->pd, &ah_attr); 399 if (IS_ERR(new_ah->ah)) { 400 printk(KERN_WARNING "Couldn't create new SM AH\n"); 401 kfree(new_ah); 402 return; 403 } 404 405 spin_lock_irq(&port->ah_lock); 406 old_ah = port->sm_ah; 407 port->sm_ah = new_ah; 408 spin_unlock_irq(&port->ah_lock); 409 410 if (old_ah) 411 kref_put(&old_ah->ref, free_sm_ah); 412 } 413 414 static void ib_sa_event(struct ib_event_handler *handler, struct ib_event *event) 415 { 416 if (event->event == IB_EVENT_PORT_ERR || 417 event->event == IB_EVENT_PORT_ACTIVE || 418 event->event == IB_EVENT_LID_CHANGE || 419 event->event == IB_EVENT_PKEY_CHANGE || 420 event->event == IB_EVENT_SM_CHANGE || 421 event->event == IB_EVENT_CLIENT_REREGISTER) { 422 struct ib_sa_device *sa_dev; 423 sa_dev = container_of(handler, typeof(*sa_dev), event_handler); 424 425 schedule_work(&sa_dev->port[event->element.port_num - 426 sa_dev->start_port].update_task); 427 } 428 } 429 430 void ib_sa_register_client(struct ib_sa_client *client) 431 { 432 atomic_set(&client->users, 1); 433 init_completion(&client->comp); 434 } 435 EXPORT_SYMBOL(ib_sa_register_client); 436 437 void ib_sa_unregister_client(struct ib_sa_client *client) 438 { 439 ib_sa_client_put(client); 440 wait_for_completion(&client->comp); 441 } 442 EXPORT_SYMBOL(ib_sa_unregister_client); 443 444 /** 445 * ib_sa_cancel_query - try to cancel an SA query 446 * @id:ID of query to cancel 447 * @query:query pointer to cancel 448 * 449 * Try to cancel an SA query. If the id and query don't match up or 450 * the query has already completed, nothing is done. Otherwise the 451 * query is canceled and will complete with a status of -EINTR. 452 */ 453 void ib_sa_cancel_query(int id, struct ib_sa_query *query) 454 { 455 unsigned long flags; 456 struct ib_mad_agent *agent; 457 struct ib_mad_send_buf *mad_buf; 458 459 spin_lock_irqsave(&idr_lock, flags); 460 if (idr_find(&query_idr, id) != query) { 461 spin_unlock_irqrestore(&idr_lock, flags); 462 return; 463 } 464 agent = query->port->agent; 465 mad_buf = query->mad_buf; 466 spin_unlock_irqrestore(&idr_lock, flags); 467 468 ib_cancel_mad(agent, mad_buf); 469 } 470 EXPORT_SYMBOL(ib_sa_cancel_query); 471 472 static u8 get_src_path_mask(struct ib_device *device, u8 port_num) 473 { 474 struct ib_sa_device *sa_dev; 475 struct ib_sa_port *port; 476 unsigned long flags; 477 u8 src_path_mask; 478 479 sa_dev = ib_get_client_data(device, &sa_client); 480 if (!sa_dev) 481 return 0x7f; 482 483 port = &sa_dev->port[port_num - sa_dev->start_port]; 484 spin_lock_irqsave(&port->ah_lock, flags); 485 src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f; 486 spin_unlock_irqrestore(&port->ah_lock, flags); 487 488 return src_path_mask; 489 } 490 491 int ib_init_ah_from_path(struct ib_device *device, u8 port_num, 492 struct ib_sa_path_rec *rec, struct ib_ah_attr *ah_attr) 493 { 494 int ret; 495 u16 gid_index; 496 497 memset(ah_attr, 0, sizeof *ah_attr); 498 ah_attr->dlid = be16_to_cpu(rec->dlid); 499 ah_attr->sl = rec->sl; 500 ah_attr->src_path_bits = be16_to_cpu(rec->slid) & 501 get_src_path_mask(device, port_num); 502 ah_attr->port_num = port_num; 503 ah_attr->static_rate = rec->rate; 504 505 if (rec->hop_limit > 1) { 506 ah_attr->ah_flags = IB_AH_GRH; 507 ah_attr->grh.dgid = rec->dgid; 508 509 ret = ib_find_cached_gid(device, &rec->sgid, &port_num, 510 &gid_index); 511 if (ret) 512 return ret; 513 514 ah_attr->grh.sgid_index = gid_index; 515 ah_attr->grh.flow_label = be32_to_cpu(rec->flow_label); 516 ah_attr->grh.hop_limit = rec->hop_limit; 517 ah_attr->grh.traffic_class = rec->traffic_class; 518 } 519 return 0; 520 } 521 EXPORT_SYMBOL(ib_init_ah_from_path); 522 523 static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask) 524 { 525 unsigned long flags; 526 527 spin_lock_irqsave(&query->port->ah_lock, flags); 528 kref_get(&query->port->sm_ah->ref); 529 query->sm_ah = query->port->sm_ah; 530 spin_unlock_irqrestore(&query->port->ah_lock, flags); 531 532 query->mad_buf = ib_create_send_mad(query->port->agent, 1, 533 query->sm_ah->pkey_index, 534 0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA, 535 gfp_mask); 536 if (!query->mad_buf) { 537 kref_put(&query->sm_ah->ref, free_sm_ah); 538 return -ENOMEM; 539 } 540 541 query->mad_buf->ah = query->sm_ah->ah; 542 543 return 0; 544 } 545 546 static void free_mad(struct ib_sa_query *query) 547 { 548 ib_free_send_mad(query->mad_buf); 549 kref_put(&query->sm_ah->ref, free_sm_ah); 550 } 551 552 static void init_mad(struct ib_sa_mad *mad, struct ib_mad_agent *agent) 553 { 554 unsigned long flags; 555 556 memset(mad, 0, sizeof *mad); 557 558 mad->mad_hdr.base_version = IB_MGMT_BASE_VERSION; 559 mad->mad_hdr.mgmt_class = IB_MGMT_CLASS_SUBN_ADM; 560 mad->mad_hdr.class_version = IB_SA_CLASS_VERSION; 561 562 spin_lock_irqsave(&tid_lock, flags); 563 mad->mad_hdr.tid = 564 cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++); 565 spin_unlock_irqrestore(&tid_lock, flags); 566 } 567 568 static int send_mad(struct ib_sa_query *query, int timeout_ms, gfp_t gfp_mask) 569 { 570 unsigned long flags; 571 int ret, id; 572 573 retry: 574 if (!idr_pre_get(&query_idr, gfp_mask)) 575 return -ENOMEM; 576 spin_lock_irqsave(&idr_lock, flags); 577 ret = idr_get_new(&query_idr, query, &id); 578 spin_unlock_irqrestore(&idr_lock, flags); 579 if (ret == -EAGAIN) 580 goto retry; 581 if (ret) 582 return ret; 583 584 query->mad_buf->timeout_ms = timeout_ms; 585 query->mad_buf->context[0] = query; 586 query->id = id; 587 588 ret = ib_post_send_mad(query->mad_buf, NULL); 589 if (ret) { 590 spin_lock_irqsave(&idr_lock, flags); 591 idr_remove(&query_idr, id); 592 spin_unlock_irqrestore(&idr_lock, flags); 593 } 594 595 /* 596 * It's not safe to dereference query any more, because the 597 * send may already have completed and freed the query in 598 * another context. 599 */ 600 return ret ? ret : id; 601 } 602 603 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query, 604 int status, 605 struct ib_sa_mad *mad) 606 { 607 struct ib_sa_path_query *query = 608 container_of(sa_query, struct ib_sa_path_query, sa_query); 609 610 if (mad) { 611 struct ib_sa_path_rec rec; 612 613 ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), 614 mad->data, &rec); 615 query->callback(status, &rec, query->context); 616 } else 617 query->callback(status, NULL, query->context); 618 } 619 620 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query) 621 { 622 kfree(container_of(sa_query, struct ib_sa_path_query, sa_query)); 623 } 624 625 /** 626 * ib_sa_path_rec_get - Start a Path get query 627 * @client:SA client 628 * @device:device to send query on 629 * @port_num: port number to send query on 630 * @rec:Path Record to send in query 631 * @comp_mask:component mask to send in query 632 * @timeout_ms:time to wait for response 633 * @gfp_mask:GFP mask to use for internal allocations 634 * @callback:function called when query completes, times out or is 635 * canceled 636 * @context:opaque user context passed to callback 637 * @sa_query:query context, used to cancel query 638 * 639 * Send a Path Record Get query to the SA to look up a path. The 640 * callback function will be called when the query completes (or 641 * fails); status is 0 for a successful response, -EINTR if the query 642 * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error 643 * occurred sending the query. The resp parameter of the callback is 644 * only valid if status is 0. 645 * 646 * If the return value of ib_sa_path_rec_get() is negative, it is an 647 * error code. Otherwise it is a query ID that can be used to cancel 648 * the query. 649 */ 650 int ib_sa_path_rec_get(struct ib_sa_client *client, 651 struct ib_device *device, u8 port_num, 652 struct ib_sa_path_rec *rec, 653 ib_sa_comp_mask comp_mask, 654 int timeout_ms, gfp_t gfp_mask, 655 void (*callback)(int status, 656 struct ib_sa_path_rec *resp, 657 void *context), 658 void *context, 659 struct ib_sa_query **sa_query) 660 { 661 struct ib_sa_path_query *query; 662 struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client); 663 struct ib_sa_port *port; 664 struct ib_mad_agent *agent; 665 struct ib_sa_mad *mad; 666 int ret; 667 668 if (!sa_dev) 669 return -ENODEV; 670 671 port = &sa_dev->port[port_num - sa_dev->start_port]; 672 agent = port->agent; 673 674 query = kmalloc(sizeof *query, gfp_mask); 675 if (!query) 676 return -ENOMEM; 677 678 query->sa_query.port = port; 679 ret = alloc_mad(&query->sa_query, gfp_mask); 680 if (ret) 681 goto err1; 682 683 ib_sa_client_get(client); 684 query->sa_query.client = client; 685 query->callback = callback; 686 query->context = context; 687 688 mad = query->sa_query.mad_buf->mad; 689 init_mad(mad, agent); 690 691 query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL; 692 query->sa_query.release = ib_sa_path_rec_release; 693 mad->mad_hdr.method = IB_MGMT_METHOD_GET; 694 mad->mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_PATH_REC); 695 mad->sa_hdr.comp_mask = comp_mask; 696 697 ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, mad->data); 698 699 *sa_query = &query->sa_query; 700 701 ret = send_mad(&query->sa_query, timeout_ms, gfp_mask); 702 if (ret < 0) 703 goto err2; 704 705 return ret; 706 707 err2: 708 *sa_query = NULL; 709 ib_sa_client_put(query->sa_query.client); 710 free_mad(&query->sa_query); 711 712 err1: 713 kfree(query); 714 return ret; 715 } 716 EXPORT_SYMBOL(ib_sa_path_rec_get); 717 718 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query, 719 int status, 720 struct ib_sa_mad *mad) 721 { 722 struct ib_sa_service_query *query = 723 container_of(sa_query, struct ib_sa_service_query, sa_query); 724 725 if (mad) { 726 struct ib_sa_service_rec rec; 727 728 ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table), 729 mad->data, &rec); 730 query->callback(status, &rec, query->context); 731 } else 732 query->callback(status, NULL, query->context); 733 } 734 735 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query) 736 { 737 kfree(container_of(sa_query, struct ib_sa_service_query, sa_query)); 738 } 739 740 /** 741 * ib_sa_service_rec_query - Start Service Record operation 742 * @client:SA client 743 * @device:device to send request on 744 * @port_num: port number to send request on 745 * @method:SA method - should be get, set, or delete 746 * @rec:Service Record to send in request 747 * @comp_mask:component mask to send in request 748 * @timeout_ms:time to wait for response 749 * @gfp_mask:GFP mask to use for internal allocations 750 * @callback:function called when request completes, times out or is 751 * canceled 752 * @context:opaque user context passed to callback 753 * @sa_query:request context, used to cancel request 754 * 755 * Send a Service Record set/get/delete to the SA to register, 756 * unregister or query a service record. 757 * The callback function will be called when the request completes (or 758 * fails); status is 0 for a successful response, -EINTR if the query 759 * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error 760 * occurred sending the query. The resp parameter of the callback is 761 * only valid if status is 0. 762 * 763 * If the return value of ib_sa_service_rec_query() is negative, it is an 764 * error code. Otherwise it is a request ID that can be used to cancel 765 * the query. 766 */ 767 int ib_sa_service_rec_query(struct ib_sa_client *client, 768 struct ib_device *device, u8 port_num, u8 method, 769 struct ib_sa_service_rec *rec, 770 ib_sa_comp_mask comp_mask, 771 int timeout_ms, gfp_t gfp_mask, 772 void (*callback)(int status, 773 struct ib_sa_service_rec *resp, 774 void *context), 775 void *context, 776 struct ib_sa_query **sa_query) 777 { 778 struct ib_sa_service_query *query; 779 struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client); 780 struct ib_sa_port *port; 781 struct ib_mad_agent *agent; 782 struct ib_sa_mad *mad; 783 int ret; 784 785 if (!sa_dev) 786 return -ENODEV; 787 788 port = &sa_dev->port[port_num - sa_dev->start_port]; 789 agent = port->agent; 790 791 if (method != IB_MGMT_METHOD_GET && 792 method != IB_MGMT_METHOD_SET && 793 method != IB_SA_METHOD_DELETE) 794 return -EINVAL; 795 796 query = kmalloc(sizeof *query, gfp_mask); 797 if (!query) 798 return -ENOMEM; 799 800 query->sa_query.port = port; 801 ret = alloc_mad(&query->sa_query, gfp_mask); 802 if (ret) 803 goto err1; 804 805 ib_sa_client_get(client); 806 query->sa_query.client = client; 807 query->callback = callback; 808 query->context = context; 809 810 mad = query->sa_query.mad_buf->mad; 811 init_mad(mad, agent); 812 813 query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL; 814 query->sa_query.release = ib_sa_service_rec_release; 815 mad->mad_hdr.method = method; 816 mad->mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_SERVICE_REC); 817 mad->sa_hdr.comp_mask = comp_mask; 818 819 ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table), 820 rec, mad->data); 821 822 *sa_query = &query->sa_query; 823 824 ret = send_mad(&query->sa_query, timeout_ms, gfp_mask); 825 if (ret < 0) 826 goto err2; 827 828 return ret; 829 830 err2: 831 *sa_query = NULL; 832 ib_sa_client_put(query->sa_query.client); 833 free_mad(&query->sa_query); 834 835 err1: 836 kfree(query); 837 return ret; 838 } 839 EXPORT_SYMBOL(ib_sa_service_rec_query); 840 841 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query, 842 int status, 843 struct ib_sa_mad *mad) 844 { 845 struct ib_sa_mcmember_query *query = 846 container_of(sa_query, struct ib_sa_mcmember_query, sa_query); 847 848 if (mad) { 849 struct ib_sa_mcmember_rec rec; 850 851 ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table), 852 mad->data, &rec); 853 query->callback(status, &rec, query->context); 854 } else 855 query->callback(status, NULL, query->context); 856 } 857 858 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query) 859 { 860 kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query)); 861 } 862 863 int ib_sa_mcmember_rec_query(struct ib_sa_client *client, 864 struct ib_device *device, u8 port_num, 865 u8 method, 866 struct ib_sa_mcmember_rec *rec, 867 ib_sa_comp_mask comp_mask, 868 int timeout_ms, gfp_t gfp_mask, 869 void (*callback)(int status, 870 struct ib_sa_mcmember_rec *resp, 871 void *context), 872 void *context, 873 struct ib_sa_query **sa_query) 874 { 875 struct ib_sa_mcmember_query *query; 876 struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client); 877 struct ib_sa_port *port; 878 struct ib_mad_agent *agent; 879 struct ib_sa_mad *mad; 880 int ret; 881 882 if (!sa_dev) 883 return -ENODEV; 884 885 port = &sa_dev->port[port_num - sa_dev->start_port]; 886 agent = port->agent; 887 888 query = kmalloc(sizeof *query, gfp_mask); 889 if (!query) 890 return -ENOMEM; 891 892 query->sa_query.port = port; 893 ret = alloc_mad(&query->sa_query, gfp_mask); 894 if (ret) 895 goto err1; 896 897 ib_sa_client_get(client); 898 query->sa_query.client = client; 899 query->callback = callback; 900 query->context = context; 901 902 mad = query->sa_query.mad_buf->mad; 903 init_mad(mad, agent); 904 905 query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL; 906 query->sa_query.release = ib_sa_mcmember_rec_release; 907 mad->mad_hdr.method = method; 908 mad->mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC); 909 mad->sa_hdr.comp_mask = comp_mask; 910 911 ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table), 912 rec, mad->data); 913 914 *sa_query = &query->sa_query; 915 916 ret = send_mad(&query->sa_query, timeout_ms, gfp_mask); 917 if (ret < 0) 918 goto err2; 919 920 return ret; 921 922 err2: 923 *sa_query = NULL; 924 ib_sa_client_put(query->sa_query.client); 925 free_mad(&query->sa_query); 926 927 err1: 928 kfree(query); 929 return ret; 930 } 931 932 static void send_handler(struct ib_mad_agent *agent, 933 struct ib_mad_send_wc *mad_send_wc) 934 { 935 struct ib_sa_query *query = mad_send_wc->send_buf->context[0]; 936 unsigned long flags; 937 938 if (query->callback) 939 switch (mad_send_wc->status) { 940 case IB_WC_SUCCESS: 941 /* No callback -- already got recv */ 942 break; 943 case IB_WC_RESP_TIMEOUT_ERR: 944 query->callback(query, -ETIMEDOUT, NULL); 945 break; 946 case IB_WC_WR_FLUSH_ERR: 947 query->callback(query, -EINTR, NULL); 948 break; 949 default: 950 query->callback(query, -EIO, NULL); 951 break; 952 } 953 954 spin_lock_irqsave(&idr_lock, flags); 955 idr_remove(&query_idr, query->id); 956 spin_unlock_irqrestore(&idr_lock, flags); 957 958 free_mad(query); 959 ib_sa_client_put(query->client); 960 query->release(query); 961 } 962 963 static void recv_handler(struct ib_mad_agent *mad_agent, 964 struct ib_mad_recv_wc *mad_recv_wc) 965 { 966 struct ib_sa_query *query; 967 struct ib_mad_send_buf *mad_buf; 968 969 mad_buf = (void *) (unsigned long) mad_recv_wc->wc->wr_id; 970 query = mad_buf->context[0]; 971 972 if (query->callback) { 973 if (mad_recv_wc->wc->status == IB_WC_SUCCESS) 974 query->callback(query, 975 mad_recv_wc->recv_buf.mad->mad_hdr.status ? 976 -EINVAL : 0, 977 (struct ib_sa_mad *) mad_recv_wc->recv_buf.mad); 978 else 979 query->callback(query, -EIO, NULL); 980 } 981 982 ib_free_recv_mad(mad_recv_wc); 983 } 984 985 static void ib_sa_add_one(struct ib_device *device) 986 { 987 struct ib_sa_device *sa_dev; 988 int s, e, i; 989 990 if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB) 991 return; 992 993 if (device->node_type == RDMA_NODE_IB_SWITCH) 994 s = e = 0; 995 else { 996 s = 1; 997 e = device->phys_port_cnt; 998 } 999 1000 sa_dev = kmalloc(sizeof *sa_dev + 1001 (e - s + 1) * sizeof (struct ib_sa_port), 1002 GFP_KERNEL); 1003 if (!sa_dev) 1004 return; 1005 1006 sa_dev->start_port = s; 1007 sa_dev->end_port = e; 1008 1009 for (i = 0; i <= e - s; ++i) { 1010 sa_dev->port[i].sm_ah = NULL; 1011 sa_dev->port[i].port_num = i + s; 1012 spin_lock_init(&sa_dev->port[i].ah_lock); 1013 1014 sa_dev->port[i].agent = 1015 ib_register_mad_agent(device, i + s, IB_QPT_GSI, 1016 NULL, 0, send_handler, 1017 recv_handler, sa_dev); 1018 if (IS_ERR(sa_dev->port[i].agent)) 1019 goto err; 1020 1021 INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah); 1022 } 1023 1024 ib_set_client_data(device, &sa_client, sa_dev); 1025 1026 /* 1027 * We register our event handler after everything is set up, 1028 * and then update our cached info after the event handler is 1029 * registered to avoid any problems if a port changes state 1030 * during our initialization. 1031 */ 1032 1033 INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event); 1034 if (ib_register_event_handler(&sa_dev->event_handler)) 1035 goto err; 1036 1037 for (i = 0; i <= e - s; ++i) 1038 update_sm_ah(&sa_dev->port[i].update_task); 1039 1040 return; 1041 1042 err: 1043 while (--i >= 0) 1044 ib_unregister_mad_agent(sa_dev->port[i].agent); 1045 1046 kfree(sa_dev); 1047 1048 return; 1049 } 1050 1051 static void ib_sa_remove_one(struct ib_device *device) 1052 { 1053 struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client); 1054 int i; 1055 1056 if (!sa_dev) 1057 return; 1058 1059 ib_unregister_event_handler(&sa_dev->event_handler); 1060 1061 flush_scheduled_work(); 1062 1063 for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) { 1064 ib_unregister_mad_agent(sa_dev->port[i].agent); 1065 kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah); 1066 } 1067 1068 kfree(sa_dev); 1069 } 1070 1071 static int __init ib_sa_init(void) 1072 { 1073 int ret; 1074 1075 spin_lock_init(&idr_lock); 1076 spin_lock_init(&tid_lock); 1077 1078 get_random_bytes(&tid, sizeof tid); 1079 1080 ret = ib_register_client(&sa_client); 1081 if (ret) { 1082 printk(KERN_ERR "Couldn't register ib_sa client\n"); 1083 goto err1; 1084 } 1085 1086 ret = mcast_init(); 1087 if (ret) { 1088 printk(KERN_ERR "Couldn't initialize multicast handling\n"); 1089 goto err2; 1090 } 1091 1092 return 0; 1093 err2: 1094 ib_unregister_client(&sa_client); 1095 err1: 1096 return ret; 1097 } 1098 1099 static void __exit ib_sa_cleanup(void) 1100 { 1101 mcast_cleanup(); 1102 ib_unregister_client(&sa_client); 1103 idr_destroy(&query_idr); 1104 } 1105 1106 module_init(ib_sa_init); 1107 module_exit(ib_sa_cleanup); 1108