1 /* 2 * Copyright (c) 2004 Topspin Communications. All rights reserved. 3 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved. 4 * 5 * This software is available to you under a choice of one of two 6 * licenses. You may choose to be licensed under the terms of the GNU 7 * General Public License (GPL) Version 2, available from the file 8 * COPYING in the main directory of this source tree, or the 9 * OpenIB.org BSD license below: 10 * 11 * Redistribution and use in source and binary forms, with or 12 * without modification, are permitted provided that the following 13 * conditions are met: 14 * 15 * - Redistributions of source code must retain the above 16 * copyright notice, this list of conditions and the following 17 * disclaimer. 18 * 19 * - Redistributions in binary form must reproduce the above 20 * copyright notice, this list of conditions and the following 21 * disclaimer in the documentation and/or other materials 22 * provided with the distribution. 23 * 24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 31 * SOFTWARE. 32 * 33 * $Id: device.c 1349 2004-12-16 21:09:43Z roland $ 34 */ 35 36 #include <linux/module.h> 37 #include <linux/string.h> 38 #include <linux/errno.h> 39 #include <linux/kernel.h> 40 #include <linux/slab.h> 41 #include <linux/init.h> 42 #include <linux/mutex.h> 43 #include <linux/workqueue.h> 44 45 #include "core_priv.h" 46 47 MODULE_AUTHOR("Roland Dreier"); 48 MODULE_DESCRIPTION("core kernel InfiniBand API"); 49 MODULE_LICENSE("Dual BSD/GPL"); 50 51 struct ib_client_data { 52 struct list_head list; 53 struct ib_client *client; 54 void * data; 55 }; 56 57 static LIST_HEAD(device_list); 58 static LIST_HEAD(client_list); 59 60 /* 61 * device_mutex protects access to both device_list and client_list. 62 * There's no real point to using multiple locks or something fancier 63 * like an rwsem: we always access both lists, and we're always 64 * modifying one list or the other list. In any case this is not a 65 * hot path so there's no point in trying to optimize. 66 */ 67 static DEFINE_MUTEX(device_mutex); 68 69 static int ib_device_check_mandatory(struct ib_device *device) 70 { 71 #define IB_MANDATORY_FUNC(x) { offsetof(struct ib_device, x), #x } 72 static const struct { 73 size_t offset; 74 char *name; 75 } mandatory_table[] = { 76 IB_MANDATORY_FUNC(query_device), 77 IB_MANDATORY_FUNC(query_port), 78 IB_MANDATORY_FUNC(query_pkey), 79 IB_MANDATORY_FUNC(query_gid), 80 IB_MANDATORY_FUNC(alloc_pd), 81 IB_MANDATORY_FUNC(dealloc_pd), 82 IB_MANDATORY_FUNC(create_ah), 83 IB_MANDATORY_FUNC(destroy_ah), 84 IB_MANDATORY_FUNC(create_qp), 85 IB_MANDATORY_FUNC(modify_qp), 86 IB_MANDATORY_FUNC(destroy_qp), 87 IB_MANDATORY_FUNC(post_send), 88 IB_MANDATORY_FUNC(post_recv), 89 IB_MANDATORY_FUNC(create_cq), 90 IB_MANDATORY_FUNC(destroy_cq), 91 IB_MANDATORY_FUNC(poll_cq), 92 IB_MANDATORY_FUNC(req_notify_cq), 93 IB_MANDATORY_FUNC(get_dma_mr), 94 IB_MANDATORY_FUNC(dereg_mr) 95 }; 96 int i; 97 98 for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) { 99 if (!*(void **) ((void *) device + mandatory_table[i].offset)) { 100 printk(KERN_WARNING "Device %s is missing mandatory function %s\n", 101 device->name, mandatory_table[i].name); 102 return -EINVAL; 103 } 104 } 105 106 return 0; 107 } 108 109 static struct ib_device *__ib_device_get_by_name(const char *name) 110 { 111 struct ib_device *device; 112 113 list_for_each_entry(device, &device_list, core_list) 114 if (!strncmp(name, device->name, IB_DEVICE_NAME_MAX)) 115 return device; 116 117 return NULL; 118 } 119 120 121 static int alloc_name(char *name) 122 { 123 long *inuse; 124 char buf[IB_DEVICE_NAME_MAX]; 125 struct ib_device *device; 126 int i; 127 128 inuse = (long *) get_zeroed_page(GFP_KERNEL); 129 if (!inuse) 130 return -ENOMEM; 131 132 list_for_each_entry(device, &device_list, core_list) { 133 if (!sscanf(device->name, name, &i)) 134 continue; 135 if (i < 0 || i >= PAGE_SIZE * 8) 136 continue; 137 snprintf(buf, sizeof buf, name, i); 138 if (!strncmp(buf, device->name, IB_DEVICE_NAME_MAX)) 139 set_bit(i, inuse); 140 } 141 142 i = find_first_zero_bit(inuse, PAGE_SIZE * 8); 143 free_page((unsigned long) inuse); 144 snprintf(buf, sizeof buf, name, i); 145 146 if (__ib_device_get_by_name(buf)) 147 return -ENFILE; 148 149 strlcpy(name, buf, IB_DEVICE_NAME_MAX); 150 return 0; 151 } 152 153 static int start_port(struct ib_device *device) 154 { 155 return (device->node_type == RDMA_NODE_IB_SWITCH) ? 0 : 1; 156 } 157 158 159 static int end_port(struct ib_device *device) 160 { 161 return (device->node_type == RDMA_NODE_IB_SWITCH) ? 162 0 : device->phys_port_cnt; 163 } 164 165 /** 166 * ib_alloc_device - allocate an IB device struct 167 * @size:size of structure to allocate 168 * 169 * Low-level drivers should use ib_alloc_device() to allocate &struct 170 * ib_device. @size is the size of the structure to be allocated, 171 * including any private data used by the low-level driver. 172 * ib_dealloc_device() must be used to free structures allocated with 173 * ib_alloc_device(). 174 */ 175 struct ib_device *ib_alloc_device(size_t size) 176 { 177 BUG_ON(size < sizeof (struct ib_device)); 178 179 return kzalloc(size, GFP_KERNEL); 180 } 181 EXPORT_SYMBOL(ib_alloc_device); 182 183 /** 184 * ib_dealloc_device - free an IB device struct 185 * @device:structure to free 186 * 187 * Free a structure allocated with ib_alloc_device(). 188 */ 189 void ib_dealloc_device(struct ib_device *device) 190 { 191 if (device->reg_state == IB_DEV_UNINITIALIZED) { 192 kfree(device); 193 return; 194 } 195 196 BUG_ON(device->reg_state != IB_DEV_UNREGISTERED); 197 198 ib_device_unregister_sysfs(device); 199 } 200 EXPORT_SYMBOL(ib_dealloc_device); 201 202 static int add_client_context(struct ib_device *device, struct ib_client *client) 203 { 204 struct ib_client_data *context; 205 unsigned long flags; 206 207 context = kmalloc(sizeof *context, GFP_KERNEL); 208 if (!context) { 209 printk(KERN_WARNING "Couldn't allocate client context for %s/%s\n", 210 device->name, client->name); 211 return -ENOMEM; 212 } 213 214 context->client = client; 215 context->data = NULL; 216 217 spin_lock_irqsave(&device->client_data_lock, flags); 218 list_add(&context->list, &device->client_data_list); 219 spin_unlock_irqrestore(&device->client_data_lock, flags); 220 221 return 0; 222 } 223 224 static int read_port_table_lengths(struct ib_device *device) 225 { 226 struct ib_port_attr *tprops = NULL; 227 int num_ports, ret = -ENOMEM; 228 u8 port_index; 229 230 tprops = kmalloc(sizeof *tprops, GFP_KERNEL); 231 if (!tprops) 232 goto out; 233 234 num_ports = end_port(device) - start_port(device) + 1; 235 236 device->pkey_tbl_len = kmalloc(sizeof *device->pkey_tbl_len * num_ports, 237 GFP_KERNEL); 238 device->gid_tbl_len = kmalloc(sizeof *device->gid_tbl_len * num_ports, 239 GFP_KERNEL); 240 if (!device->pkey_tbl_len || !device->gid_tbl_len) 241 goto err; 242 243 for (port_index = 0; port_index < num_ports; ++port_index) { 244 ret = ib_query_port(device, port_index + start_port(device), 245 tprops); 246 if (ret) 247 goto err; 248 device->pkey_tbl_len[port_index] = tprops->pkey_tbl_len; 249 device->gid_tbl_len[port_index] = tprops->gid_tbl_len; 250 } 251 252 ret = 0; 253 goto out; 254 255 err: 256 kfree(device->gid_tbl_len); 257 kfree(device->pkey_tbl_len); 258 out: 259 kfree(tprops); 260 return ret; 261 } 262 263 /** 264 * ib_register_device - Register an IB device with IB core 265 * @device:Device to register 266 * 267 * Low-level drivers use ib_register_device() to register their 268 * devices with the IB core. All registered clients will receive a 269 * callback for each device that is added. @device must be allocated 270 * with ib_alloc_device(). 271 */ 272 int ib_register_device(struct ib_device *device) 273 { 274 int ret; 275 276 mutex_lock(&device_mutex); 277 278 if (strchr(device->name, '%')) { 279 ret = alloc_name(device->name); 280 if (ret) 281 goto out; 282 } 283 284 if (ib_device_check_mandatory(device)) { 285 ret = -EINVAL; 286 goto out; 287 } 288 289 INIT_LIST_HEAD(&device->event_handler_list); 290 INIT_LIST_HEAD(&device->client_data_list); 291 spin_lock_init(&device->event_handler_lock); 292 spin_lock_init(&device->client_data_lock); 293 294 ret = read_port_table_lengths(device); 295 if (ret) { 296 printk(KERN_WARNING "Couldn't create table lengths cache for device %s\n", 297 device->name); 298 goto out; 299 } 300 301 ret = ib_device_register_sysfs(device); 302 if (ret) { 303 printk(KERN_WARNING "Couldn't register device %s with driver model\n", 304 device->name); 305 kfree(device->gid_tbl_len); 306 kfree(device->pkey_tbl_len); 307 goto out; 308 } 309 310 list_add_tail(&device->core_list, &device_list); 311 312 device->reg_state = IB_DEV_REGISTERED; 313 314 { 315 struct ib_client *client; 316 317 list_for_each_entry(client, &client_list, list) 318 if (client->add && !add_client_context(device, client)) 319 client->add(device); 320 } 321 322 out: 323 mutex_unlock(&device_mutex); 324 return ret; 325 } 326 EXPORT_SYMBOL(ib_register_device); 327 328 /** 329 * ib_unregister_device - Unregister an IB device 330 * @device:Device to unregister 331 * 332 * Unregister an IB device. All clients will receive a remove callback. 333 */ 334 void ib_unregister_device(struct ib_device *device) 335 { 336 struct ib_client *client; 337 struct ib_client_data *context, *tmp; 338 unsigned long flags; 339 340 mutex_lock(&device_mutex); 341 342 list_for_each_entry_reverse(client, &client_list, list) 343 if (client->remove) 344 client->remove(device); 345 346 list_del(&device->core_list); 347 348 kfree(device->gid_tbl_len); 349 kfree(device->pkey_tbl_len); 350 351 mutex_unlock(&device_mutex); 352 353 spin_lock_irqsave(&device->client_data_lock, flags); 354 list_for_each_entry_safe(context, tmp, &device->client_data_list, list) 355 kfree(context); 356 spin_unlock_irqrestore(&device->client_data_lock, flags); 357 358 device->reg_state = IB_DEV_UNREGISTERED; 359 } 360 EXPORT_SYMBOL(ib_unregister_device); 361 362 /** 363 * ib_register_client - Register an IB client 364 * @client:Client to register 365 * 366 * Upper level users of the IB drivers can use ib_register_client() to 367 * register callbacks for IB device addition and removal. When an IB 368 * device is added, each registered client's add method will be called 369 * (in the order the clients were registered), and when a device is 370 * removed, each client's remove method will be called (in the reverse 371 * order that clients were registered). In addition, when 372 * ib_register_client() is called, the client will receive an add 373 * callback for all devices already registered. 374 */ 375 int ib_register_client(struct ib_client *client) 376 { 377 struct ib_device *device; 378 379 mutex_lock(&device_mutex); 380 381 list_add_tail(&client->list, &client_list); 382 list_for_each_entry(device, &device_list, core_list) 383 if (client->add && !add_client_context(device, client)) 384 client->add(device); 385 386 mutex_unlock(&device_mutex); 387 388 return 0; 389 } 390 EXPORT_SYMBOL(ib_register_client); 391 392 /** 393 * ib_unregister_client - Unregister an IB client 394 * @client:Client to unregister 395 * 396 * Upper level users use ib_unregister_client() to remove their client 397 * registration. When ib_unregister_client() is called, the client 398 * will receive a remove callback for each IB device still registered. 399 */ 400 void ib_unregister_client(struct ib_client *client) 401 { 402 struct ib_client_data *context, *tmp; 403 struct ib_device *device; 404 unsigned long flags; 405 406 mutex_lock(&device_mutex); 407 408 list_for_each_entry(device, &device_list, core_list) { 409 if (client->remove) 410 client->remove(device); 411 412 spin_lock_irqsave(&device->client_data_lock, flags); 413 list_for_each_entry_safe(context, tmp, &device->client_data_list, list) 414 if (context->client == client) { 415 list_del(&context->list); 416 kfree(context); 417 } 418 spin_unlock_irqrestore(&device->client_data_lock, flags); 419 } 420 list_del(&client->list); 421 422 mutex_unlock(&device_mutex); 423 } 424 EXPORT_SYMBOL(ib_unregister_client); 425 426 /** 427 * ib_get_client_data - Get IB client context 428 * @device:Device to get context for 429 * @client:Client to get context for 430 * 431 * ib_get_client_data() returns client context set with 432 * ib_set_client_data(). 433 */ 434 void *ib_get_client_data(struct ib_device *device, struct ib_client *client) 435 { 436 struct ib_client_data *context; 437 void *ret = NULL; 438 unsigned long flags; 439 440 spin_lock_irqsave(&device->client_data_lock, flags); 441 list_for_each_entry(context, &device->client_data_list, list) 442 if (context->client == client) { 443 ret = context->data; 444 break; 445 } 446 spin_unlock_irqrestore(&device->client_data_lock, flags); 447 448 return ret; 449 } 450 EXPORT_SYMBOL(ib_get_client_data); 451 452 /** 453 * ib_set_client_data - Set IB client context 454 * @device:Device to set context for 455 * @client:Client to set context for 456 * @data:Context to set 457 * 458 * ib_set_client_data() sets client context that can be retrieved with 459 * ib_get_client_data(). 460 */ 461 void ib_set_client_data(struct ib_device *device, struct ib_client *client, 462 void *data) 463 { 464 struct ib_client_data *context; 465 unsigned long flags; 466 467 spin_lock_irqsave(&device->client_data_lock, flags); 468 list_for_each_entry(context, &device->client_data_list, list) 469 if (context->client == client) { 470 context->data = data; 471 goto out; 472 } 473 474 printk(KERN_WARNING "No client context found for %s/%s\n", 475 device->name, client->name); 476 477 out: 478 spin_unlock_irqrestore(&device->client_data_lock, flags); 479 } 480 EXPORT_SYMBOL(ib_set_client_data); 481 482 /** 483 * ib_register_event_handler - Register an IB event handler 484 * @event_handler:Handler to register 485 * 486 * ib_register_event_handler() registers an event handler that will be 487 * called back when asynchronous IB events occur (as defined in 488 * chapter 11 of the InfiniBand Architecture Specification). This 489 * callback may occur in interrupt context. 490 */ 491 int ib_register_event_handler (struct ib_event_handler *event_handler) 492 { 493 unsigned long flags; 494 495 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags); 496 list_add_tail(&event_handler->list, 497 &event_handler->device->event_handler_list); 498 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags); 499 500 return 0; 501 } 502 EXPORT_SYMBOL(ib_register_event_handler); 503 504 /** 505 * ib_unregister_event_handler - Unregister an event handler 506 * @event_handler:Handler to unregister 507 * 508 * Unregister an event handler registered with 509 * ib_register_event_handler(). 510 */ 511 int ib_unregister_event_handler(struct ib_event_handler *event_handler) 512 { 513 unsigned long flags; 514 515 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags); 516 list_del(&event_handler->list); 517 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags); 518 519 return 0; 520 } 521 EXPORT_SYMBOL(ib_unregister_event_handler); 522 523 /** 524 * ib_dispatch_event - Dispatch an asynchronous event 525 * @event:Event to dispatch 526 * 527 * Low-level drivers must call ib_dispatch_event() to dispatch the 528 * event to all registered event handlers when an asynchronous event 529 * occurs. 530 */ 531 void ib_dispatch_event(struct ib_event *event) 532 { 533 unsigned long flags; 534 struct ib_event_handler *handler; 535 536 spin_lock_irqsave(&event->device->event_handler_lock, flags); 537 538 list_for_each_entry(handler, &event->device->event_handler_list, list) 539 handler->handler(handler, event); 540 541 spin_unlock_irqrestore(&event->device->event_handler_lock, flags); 542 } 543 EXPORT_SYMBOL(ib_dispatch_event); 544 545 /** 546 * ib_query_device - Query IB device attributes 547 * @device:Device to query 548 * @device_attr:Device attributes 549 * 550 * ib_query_device() returns the attributes of a device through the 551 * @device_attr pointer. 552 */ 553 int ib_query_device(struct ib_device *device, 554 struct ib_device_attr *device_attr) 555 { 556 return device->query_device(device, device_attr); 557 } 558 EXPORT_SYMBOL(ib_query_device); 559 560 /** 561 * ib_query_port - Query IB port attributes 562 * @device:Device to query 563 * @port_num:Port number to query 564 * @port_attr:Port attributes 565 * 566 * ib_query_port() returns the attributes of a port through the 567 * @port_attr pointer. 568 */ 569 int ib_query_port(struct ib_device *device, 570 u8 port_num, 571 struct ib_port_attr *port_attr) 572 { 573 if (port_num < start_port(device) || port_num > end_port(device)) 574 return -EINVAL; 575 576 return device->query_port(device, port_num, port_attr); 577 } 578 EXPORT_SYMBOL(ib_query_port); 579 580 /** 581 * ib_query_gid - Get GID table entry 582 * @device:Device to query 583 * @port_num:Port number to query 584 * @index:GID table index to query 585 * @gid:Returned GID 586 * 587 * ib_query_gid() fetches the specified GID table entry. 588 */ 589 int ib_query_gid(struct ib_device *device, 590 u8 port_num, int index, union ib_gid *gid) 591 { 592 return device->query_gid(device, port_num, index, gid); 593 } 594 EXPORT_SYMBOL(ib_query_gid); 595 596 /** 597 * ib_query_pkey - Get P_Key table entry 598 * @device:Device to query 599 * @port_num:Port number to query 600 * @index:P_Key table index to query 601 * @pkey:Returned P_Key 602 * 603 * ib_query_pkey() fetches the specified P_Key table entry. 604 */ 605 int ib_query_pkey(struct ib_device *device, 606 u8 port_num, u16 index, u16 *pkey) 607 { 608 return device->query_pkey(device, port_num, index, pkey); 609 } 610 EXPORT_SYMBOL(ib_query_pkey); 611 612 /** 613 * ib_modify_device - Change IB device attributes 614 * @device:Device to modify 615 * @device_modify_mask:Mask of attributes to change 616 * @device_modify:New attribute values 617 * 618 * ib_modify_device() changes a device's attributes as specified by 619 * the @device_modify_mask and @device_modify structure. 620 */ 621 int ib_modify_device(struct ib_device *device, 622 int device_modify_mask, 623 struct ib_device_modify *device_modify) 624 { 625 return device->modify_device(device, device_modify_mask, 626 device_modify); 627 } 628 EXPORT_SYMBOL(ib_modify_device); 629 630 /** 631 * ib_modify_port - Modifies the attributes for the specified port. 632 * @device: The device to modify. 633 * @port_num: The number of the port to modify. 634 * @port_modify_mask: Mask used to specify which attributes of the port 635 * to change. 636 * @port_modify: New attribute values for the port. 637 * 638 * ib_modify_port() changes a port's attributes as specified by the 639 * @port_modify_mask and @port_modify structure. 640 */ 641 int ib_modify_port(struct ib_device *device, 642 u8 port_num, int port_modify_mask, 643 struct ib_port_modify *port_modify) 644 { 645 if (port_num < start_port(device) || port_num > end_port(device)) 646 return -EINVAL; 647 648 return device->modify_port(device, port_num, port_modify_mask, 649 port_modify); 650 } 651 EXPORT_SYMBOL(ib_modify_port); 652 653 /** 654 * ib_find_gid - Returns the port number and GID table index where 655 * a specified GID value occurs. 656 * @device: The device to query. 657 * @gid: The GID value to search for. 658 * @port_num: The port number of the device where the GID value was found. 659 * @index: The index into the GID table where the GID was found. This 660 * parameter may be NULL. 661 */ 662 int ib_find_gid(struct ib_device *device, union ib_gid *gid, 663 u8 *port_num, u16 *index) 664 { 665 union ib_gid tmp_gid; 666 int ret, port, i; 667 668 for (port = start_port(device); port <= end_port(device); ++port) { 669 for (i = 0; i < device->gid_tbl_len[port - start_port(device)]; ++i) { 670 ret = ib_query_gid(device, port, i, &tmp_gid); 671 if (ret) 672 return ret; 673 if (!memcmp(&tmp_gid, gid, sizeof *gid)) { 674 *port_num = port; 675 if (index) 676 *index = i; 677 return 0; 678 } 679 } 680 } 681 682 return -ENOENT; 683 } 684 EXPORT_SYMBOL(ib_find_gid); 685 686 /** 687 * ib_find_pkey - Returns the PKey table index where a specified 688 * PKey value occurs. 689 * @device: The device to query. 690 * @port_num: The port number of the device to search for the PKey. 691 * @pkey: The PKey value to search for. 692 * @index: The index into the PKey table where the PKey was found. 693 */ 694 int ib_find_pkey(struct ib_device *device, 695 u8 port_num, u16 pkey, u16 *index) 696 { 697 int ret, i; 698 u16 tmp_pkey; 699 700 for (i = 0; i < device->pkey_tbl_len[port_num - start_port(device)]; ++i) { 701 ret = ib_query_pkey(device, port_num, i, &tmp_pkey); 702 if (ret) 703 return ret; 704 705 if (pkey == tmp_pkey) { 706 *index = i; 707 return 0; 708 } 709 } 710 711 return -ENOENT; 712 } 713 EXPORT_SYMBOL(ib_find_pkey); 714 715 static int __init ib_core_init(void) 716 { 717 int ret; 718 719 ret = ib_sysfs_setup(); 720 if (ret) 721 printk(KERN_WARNING "Couldn't create InfiniBand device class\n"); 722 723 ret = ib_cache_setup(); 724 if (ret) { 725 printk(KERN_WARNING "Couldn't set up InfiniBand P_Key/GID cache\n"); 726 ib_sysfs_cleanup(); 727 } 728 729 return ret; 730 } 731 732 static void __exit ib_core_cleanup(void) 733 { 734 ib_cache_cleanup(); 735 ib_sysfs_cleanup(); 736 /* Make sure that any pending umem accounting work is done. */ 737 flush_scheduled_work(); 738 } 739 740 module_init(ib_core_init); 741 module_exit(ib_core_cleanup); 742