Loading...
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
34#include <linux/module.h>
35#include <linux/string.h>
36#include <linux/errno.h>
37#include <linux/kernel.h>
38#include <linux/slab.h>
39#include <linux/init.h>
40#include <linux/mutex.h>
41#include <rdma/rdma_netlink.h>
42
43#include "core_priv.h"
44
45MODULE_AUTHOR("Roland Dreier");
46MODULE_DESCRIPTION("core kernel InfiniBand API");
47MODULE_LICENSE("Dual BSD/GPL");
48
49struct ib_client_data {
50 struct list_head list;
51 struct ib_client *client;
52 void * data;
53};
54
55struct workqueue_struct *ib_wq;
56EXPORT_SYMBOL_GPL(ib_wq);
57
58static LIST_HEAD(device_list);
59static LIST_HEAD(client_list);
60
61/*
62 * device_mutex protects access to both device_list and client_list.
63 * There's no real point to using multiple locks or something fancier
64 * like an rwsem: we always access both lists, and we're always
65 * modifying one list or the other list. In any case this is not a
66 * hot path so there's no point in trying to optimize.
67 */
68static DEFINE_MUTEX(device_mutex);
69
70static int ib_device_check_mandatory(struct ib_device *device)
71{
72#define IB_MANDATORY_FUNC(x) { offsetof(struct ib_device, x), #x }
73 static const struct {
74 size_t offset;
75 char *name;
76 } mandatory_table[] = {
77 IB_MANDATORY_FUNC(query_device),
78 IB_MANDATORY_FUNC(query_port),
79 IB_MANDATORY_FUNC(query_pkey),
80 IB_MANDATORY_FUNC(query_gid),
81 IB_MANDATORY_FUNC(alloc_pd),
82 IB_MANDATORY_FUNC(dealloc_pd),
83 IB_MANDATORY_FUNC(create_ah),
84 IB_MANDATORY_FUNC(destroy_ah),
85 IB_MANDATORY_FUNC(create_qp),
86 IB_MANDATORY_FUNC(modify_qp),
87 IB_MANDATORY_FUNC(destroy_qp),
88 IB_MANDATORY_FUNC(post_send),
89 IB_MANDATORY_FUNC(post_recv),
90 IB_MANDATORY_FUNC(create_cq),
91 IB_MANDATORY_FUNC(destroy_cq),
92 IB_MANDATORY_FUNC(poll_cq),
93 IB_MANDATORY_FUNC(req_notify_cq),
94 IB_MANDATORY_FUNC(get_dma_mr),
95 IB_MANDATORY_FUNC(dereg_mr)
96 };
97 int i;
98
99 for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) {
100 if (!*(void **) ((void *) device + mandatory_table[i].offset)) {
101 printk(KERN_WARNING "Device %s is missing mandatory function %s\n",
102 device->name, mandatory_table[i].name);
103 return -EINVAL;
104 }
105 }
106
107 return 0;
108}
109
110static struct ib_device *__ib_device_get_by_name(const char *name)
111{
112 struct ib_device *device;
113
114 list_for_each_entry(device, &device_list, core_list)
115 if (!strncmp(name, device->name, IB_DEVICE_NAME_MAX))
116 return device;
117
118 return NULL;
119}
120
121
122static int alloc_name(char *name)
123{
124 unsigned long *inuse;
125 char buf[IB_DEVICE_NAME_MAX];
126 struct ib_device *device;
127 int i;
128
129 inuse = (unsigned long *) get_zeroed_page(GFP_KERNEL);
130 if (!inuse)
131 return -ENOMEM;
132
133 list_for_each_entry(device, &device_list, core_list) {
134 if (!sscanf(device->name, name, &i))
135 continue;
136 if (i < 0 || i >= PAGE_SIZE * 8)
137 continue;
138 snprintf(buf, sizeof buf, name, i);
139 if (!strncmp(buf, device->name, IB_DEVICE_NAME_MAX))
140 set_bit(i, inuse);
141 }
142
143 i = find_first_zero_bit(inuse, PAGE_SIZE * 8);
144 free_page((unsigned long) inuse);
145 snprintf(buf, sizeof buf, name, i);
146
147 if (__ib_device_get_by_name(buf))
148 return -ENFILE;
149
150 strlcpy(name, buf, IB_DEVICE_NAME_MAX);
151 return 0;
152}
153
154static int start_port(struct ib_device *device)
155{
156 return (device->node_type == RDMA_NODE_IB_SWITCH) ? 0 : 1;
157}
158
159
160static int end_port(struct ib_device *device)
161{
162 return (device->node_type == RDMA_NODE_IB_SWITCH) ?
163 0 : device->phys_port_cnt;
164}
165
166/**
167 * ib_alloc_device - allocate an IB device struct
168 * @size:size of structure to allocate
169 *
170 * Low-level drivers should use ib_alloc_device() to allocate &struct
171 * ib_device. @size is the size of the structure to be allocated,
172 * including any private data used by the low-level driver.
173 * ib_dealloc_device() must be used to free structures allocated with
174 * ib_alloc_device().
175 */
176struct ib_device *ib_alloc_device(size_t size)
177{
178 BUG_ON(size < sizeof (struct ib_device));
179
180 return kzalloc(size, GFP_KERNEL);
181}
182EXPORT_SYMBOL(ib_alloc_device);
183
184/**
185 * ib_dealloc_device - free an IB device struct
186 * @device:structure to free
187 *
188 * Free a structure allocated with ib_alloc_device().
189 */
190void ib_dealloc_device(struct ib_device *device)
191{
192 if (device->reg_state == IB_DEV_UNINITIALIZED) {
193 kfree(device);
194 return;
195 }
196
197 BUG_ON(device->reg_state != IB_DEV_UNREGISTERED);
198
199 kobject_put(&device->dev.kobj);
200}
201EXPORT_SYMBOL(ib_dealloc_device);
202
203static int add_client_context(struct ib_device *device, struct ib_client *client)
204{
205 struct ib_client_data *context;
206 unsigned long flags;
207
208 context = kmalloc(sizeof *context, GFP_KERNEL);
209 if (!context) {
210 printk(KERN_WARNING "Couldn't allocate client context for %s/%s\n",
211 device->name, client->name);
212 return -ENOMEM;
213 }
214
215 context->client = client;
216 context->data = NULL;
217
218 spin_lock_irqsave(&device->client_data_lock, flags);
219 list_add(&context->list, &device->client_data_list);
220 spin_unlock_irqrestore(&device->client_data_lock, flags);
221
222 return 0;
223}
224
225static int read_port_table_lengths(struct ib_device *device)
226{
227 struct ib_port_attr *tprops = NULL;
228 int num_ports, ret = -ENOMEM;
229 u8 port_index;
230
231 tprops = kmalloc(sizeof *tprops, GFP_KERNEL);
232 if (!tprops)
233 goto out;
234
235 num_ports = end_port(device) - start_port(device) + 1;
236
237 device->pkey_tbl_len = kmalloc(sizeof *device->pkey_tbl_len * num_ports,
238 GFP_KERNEL);
239 device->gid_tbl_len = kmalloc(sizeof *device->gid_tbl_len * num_ports,
240 GFP_KERNEL);
241 if (!device->pkey_tbl_len || !device->gid_tbl_len)
242 goto err;
243
244 for (port_index = 0; port_index < num_ports; ++port_index) {
245 ret = ib_query_port(device, port_index + start_port(device),
246 tprops);
247 if (ret)
248 goto err;
249 device->pkey_tbl_len[port_index] = tprops->pkey_tbl_len;
250 device->gid_tbl_len[port_index] = tprops->gid_tbl_len;
251 }
252
253 ret = 0;
254 goto out;
255
256err:
257 kfree(device->gid_tbl_len);
258 kfree(device->pkey_tbl_len);
259out:
260 kfree(tprops);
261 return ret;
262}
263
264/**
265 * ib_register_device - Register an IB device with IB core
266 * @device:Device to register
267 *
268 * Low-level drivers use ib_register_device() to register their
269 * devices with the IB core. All registered clients will receive a
270 * callback for each device that is added. @device must be allocated
271 * with ib_alloc_device().
272 */
273int ib_register_device(struct ib_device *device,
274 int (*port_callback)(struct ib_device *,
275 u8, struct kobject *))
276{
277 int ret;
278
279 mutex_lock(&device_mutex);
280
281 if (strchr(device->name, '%')) {
282 ret = alloc_name(device->name);
283 if (ret)
284 goto out;
285 }
286
287 if (ib_device_check_mandatory(device)) {
288 ret = -EINVAL;
289 goto out;
290 }
291
292 INIT_LIST_HEAD(&device->event_handler_list);
293 INIT_LIST_HEAD(&device->client_data_list);
294 spin_lock_init(&device->event_handler_lock);
295 spin_lock_init(&device->client_data_lock);
296
297 ret = read_port_table_lengths(device);
298 if (ret) {
299 printk(KERN_WARNING "Couldn't create table lengths cache for device %s\n",
300 device->name);
301 goto out;
302 }
303
304 ret = ib_device_register_sysfs(device, port_callback);
305 if (ret) {
306 printk(KERN_WARNING "Couldn't register device %s with driver model\n",
307 device->name);
308 kfree(device->gid_tbl_len);
309 kfree(device->pkey_tbl_len);
310 goto out;
311 }
312
313 list_add_tail(&device->core_list, &device_list);
314
315 device->reg_state = IB_DEV_REGISTERED;
316
317 {
318 struct ib_client *client;
319
320 list_for_each_entry(client, &client_list, list)
321 if (client->add && !add_client_context(device, client))
322 client->add(device);
323 }
324
325 out:
326 mutex_unlock(&device_mutex);
327 return ret;
328}
329EXPORT_SYMBOL(ib_register_device);
330
331/**
332 * ib_unregister_device - Unregister an IB device
333 * @device:Device to unregister
334 *
335 * Unregister an IB device. All clients will receive a remove callback.
336 */
337void ib_unregister_device(struct ib_device *device)
338{
339 struct ib_client *client;
340 struct ib_client_data *context, *tmp;
341 unsigned long flags;
342
343 mutex_lock(&device_mutex);
344
345 list_for_each_entry_reverse(client, &client_list, list)
346 if (client->remove)
347 client->remove(device);
348
349 list_del(&device->core_list);
350
351 kfree(device->gid_tbl_len);
352 kfree(device->pkey_tbl_len);
353
354 mutex_unlock(&device_mutex);
355
356 ib_device_unregister_sysfs(device);
357
358 spin_lock_irqsave(&device->client_data_lock, flags);
359 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
360 kfree(context);
361 spin_unlock_irqrestore(&device->client_data_lock, flags);
362
363 device->reg_state = IB_DEV_UNREGISTERED;
364}
365EXPORT_SYMBOL(ib_unregister_device);
366
367/**
368 * ib_register_client - Register an IB client
369 * @client:Client to register
370 *
371 * Upper level users of the IB drivers can use ib_register_client() to
372 * register callbacks for IB device addition and removal. When an IB
373 * device is added, each registered client's add method will be called
374 * (in the order the clients were registered), and when a device is
375 * removed, each client's remove method will be called (in the reverse
376 * order that clients were registered). In addition, when
377 * ib_register_client() is called, the client will receive an add
378 * callback for all devices already registered.
379 */
380int ib_register_client(struct ib_client *client)
381{
382 struct ib_device *device;
383
384 mutex_lock(&device_mutex);
385
386 list_add_tail(&client->list, &client_list);
387 list_for_each_entry(device, &device_list, core_list)
388 if (client->add && !add_client_context(device, client))
389 client->add(device);
390
391 mutex_unlock(&device_mutex);
392
393 return 0;
394}
395EXPORT_SYMBOL(ib_register_client);
396
397/**
398 * ib_unregister_client - Unregister an IB client
399 * @client:Client to unregister
400 *
401 * Upper level users use ib_unregister_client() to remove their client
402 * registration. When ib_unregister_client() is called, the client
403 * will receive a remove callback for each IB device still registered.
404 */
405void ib_unregister_client(struct ib_client *client)
406{
407 struct ib_client_data *context, *tmp;
408 struct ib_device *device;
409 unsigned long flags;
410
411 mutex_lock(&device_mutex);
412
413 list_for_each_entry(device, &device_list, core_list) {
414 if (client->remove)
415 client->remove(device);
416
417 spin_lock_irqsave(&device->client_data_lock, flags);
418 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
419 if (context->client == client) {
420 list_del(&context->list);
421 kfree(context);
422 }
423 spin_unlock_irqrestore(&device->client_data_lock, flags);
424 }
425 list_del(&client->list);
426
427 mutex_unlock(&device_mutex);
428}
429EXPORT_SYMBOL(ib_unregister_client);
430
431/**
432 * ib_get_client_data - Get IB client context
433 * @device:Device to get context for
434 * @client:Client to get context for
435 *
436 * ib_get_client_data() returns client context set with
437 * ib_set_client_data().
438 */
439void *ib_get_client_data(struct ib_device *device, struct ib_client *client)
440{
441 struct ib_client_data *context;
442 void *ret = NULL;
443 unsigned long flags;
444
445 spin_lock_irqsave(&device->client_data_lock, flags);
446 list_for_each_entry(context, &device->client_data_list, list)
447 if (context->client == client) {
448 ret = context->data;
449 break;
450 }
451 spin_unlock_irqrestore(&device->client_data_lock, flags);
452
453 return ret;
454}
455EXPORT_SYMBOL(ib_get_client_data);
456
457/**
458 * ib_set_client_data - Set IB client context
459 * @device:Device to set context for
460 * @client:Client to set context for
461 * @data:Context to set
462 *
463 * ib_set_client_data() sets client context that can be retrieved with
464 * ib_get_client_data().
465 */
466void ib_set_client_data(struct ib_device *device, struct ib_client *client,
467 void *data)
468{
469 struct ib_client_data *context;
470 unsigned long flags;
471
472 spin_lock_irqsave(&device->client_data_lock, flags);
473 list_for_each_entry(context, &device->client_data_list, list)
474 if (context->client == client) {
475 context->data = data;
476 goto out;
477 }
478
479 printk(KERN_WARNING "No client context found for %s/%s\n",
480 device->name, client->name);
481
482out:
483 spin_unlock_irqrestore(&device->client_data_lock, flags);
484}
485EXPORT_SYMBOL(ib_set_client_data);
486
487/**
488 * ib_register_event_handler - Register an IB event handler
489 * @event_handler:Handler to register
490 *
491 * ib_register_event_handler() registers an event handler that will be
492 * called back when asynchronous IB events occur (as defined in
493 * chapter 11 of the InfiniBand Architecture Specification). This
494 * callback may occur in interrupt context.
495 */
496int ib_register_event_handler (struct ib_event_handler *event_handler)
497{
498 unsigned long flags;
499
500 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
501 list_add_tail(&event_handler->list,
502 &event_handler->device->event_handler_list);
503 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
504
505 return 0;
506}
507EXPORT_SYMBOL(ib_register_event_handler);
508
509/**
510 * ib_unregister_event_handler - Unregister an event handler
511 * @event_handler:Handler to unregister
512 *
513 * Unregister an event handler registered with
514 * ib_register_event_handler().
515 */
516int ib_unregister_event_handler(struct ib_event_handler *event_handler)
517{
518 unsigned long flags;
519
520 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
521 list_del(&event_handler->list);
522 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
523
524 return 0;
525}
526EXPORT_SYMBOL(ib_unregister_event_handler);
527
528/**
529 * ib_dispatch_event - Dispatch an asynchronous event
530 * @event:Event to dispatch
531 *
532 * Low-level drivers must call ib_dispatch_event() to dispatch the
533 * event to all registered event handlers when an asynchronous event
534 * occurs.
535 */
536void ib_dispatch_event(struct ib_event *event)
537{
538 unsigned long flags;
539 struct ib_event_handler *handler;
540
541 spin_lock_irqsave(&event->device->event_handler_lock, flags);
542
543 list_for_each_entry(handler, &event->device->event_handler_list, list)
544 handler->handler(handler, event);
545
546 spin_unlock_irqrestore(&event->device->event_handler_lock, flags);
547}
548EXPORT_SYMBOL(ib_dispatch_event);
549
550/**
551 * ib_query_device - Query IB device attributes
552 * @device:Device to query
553 * @device_attr:Device attributes
554 *
555 * ib_query_device() returns the attributes of a device through the
556 * @device_attr pointer.
557 */
558int ib_query_device(struct ib_device *device,
559 struct ib_device_attr *device_attr)
560{
561 return device->query_device(device, device_attr);
562}
563EXPORT_SYMBOL(ib_query_device);
564
565/**
566 * ib_query_port - Query IB port attributes
567 * @device:Device to query
568 * @port_num:Port number to query
569 * @port_attr:Port attributes
570 *
571 * ib_query_port() returns the attributes of a port through the
572 * @port_attr pointer.
573 */
574int ib_query_port(struct ib_device *device,
575 u8 port_num,
576 struct ib_port_attr *port_attr)
577{
578 if (port_num < start_port(device) || port_num > end_port(device))
579 return -EINVAL;
580
581 return device->query_port(device, port_num, port_attr);
582}
583EXPORT_SYMBOL(ib_query_port);
584
585/**
586 * ib_query_gid - Get GID table entry
587 * @device:Device to query
588 * @port_num:Port number to query
589 * @index:GID table index to query
590 * @gid:Returned GID
591 *
592 * ib_query_gid() fetches the specified GID table entry.
593 */
594int ib_query_gid(struct ib_device *device,
595 u8 port_num, int index, union ib_gid *gid)
596{
597 return device->query_gid(device, port_num, index, gid);
598}
599EXPORT_SYMBOL(ib_query_gid);
600
601/**
602 * ib_query_pkey - Get P_Key table entry
603 * @device:Device to query
604 * @port_num:Port number to query
605 * @index:P_Key table index to query
606 * @pkey:Returned P_Key
607 *
608 * ib_query_pkey() fetches the specified P_Key table entry.
609 */
610int ib_query_pkey(struct ib_device *device,
611 u8 port_num, u16 index, u16 *pkey)
612{
613 return device->query_pkey(device, port_num, index, pkey);
614}
615EXPORT_SYMBOL(ib_query_pkey);
616
617/**
618 * ib_modify_device - Change IB device attributes
619 * @device:Device to modify
620 * @device_modify_mask:Mask of attributes to change
621 * @device_modify:New attribute values
622 *
623 * ib_modify_device() changes a device's attributes as specified by
624 * the @device_modify_mask and @device_modify structure.
625 */
626int ib_modify_device(struct ib_device *device,
627 int device_modify_mask,
628 struct ib_device_modify *device_modify)
629{
630 if (!device->modify_device)
631 return -ENOSYS;
632
633 return device->modify_device(device, device_modify_mask,
634 device_modify);
635}
636EXPORT_SYMBOL(ib_modify_device);
637
638/**
639 * ib_modify_port - Modifies the attributes for the specified port.
640 * @device: The device to modify.
641 * @port_num: The number of the port to modify.
642 * @port_modify_mask: Mask used to specify which attributes of the port
643 * to change.
644 * @port_modify: New attribute values for the port.
645 *
646 * ib_modify_port() changes a port's attributes as specified by the
647 * @port_modify_mask and @port_modify structure.
648 */
649int ib_modify_port(struct ib_device *device,
650 u8 port_num, int port_modify_mask,
651 struct ib_port_modify *port_modify)
652{
653 if (!device->modify_port)
654 return -ENOSYS;
655
656 if (port_num < start_port(device) || port_num > end_port(device))
657 return -EINVAL;
658
659 return device->modify_port(device, port_num, port_modify_mask,
660 port_modify);
661}
662EXPORT_SYMBOL(ib_modify_port);
663
664/**
665 * ib_find_gid - Returns the port number and GID table index where
666 * a specified GID value occurs.
667 * @device: The device to query.
668 * @gid: The GID value to search for.
669 * @port_num: The port number of the device where the GID value was found.
670 * @index: The index into the GID table where the GID was found. This
671 * parameter may be NULL.
672 */
673int ib_find_gid(struct ib_device *device, union ib_gid *gid,
674 u8 *port_num, u16 *index)
675{
676 union ib_gid tmp_gid;
677 int ret, port, i;
678
679 for (port = start_port(device); port <= end_port(device); ++port) {
680 for (i = 0; i < device->gid_tbl_len[port - start_port(device)]; ++i) {
681 ret = ib_query_gid(device, port, i, &tmp_gid);
682 if (ret)
683 return ret;
684 if (!memcmp(&tmp_gid, gid, sizeof *gid)) {
685 *port_num = port;
686 if (index)
687 *index = i;
688 return 0;
689 }
690 }
691 }
692
693 return -ENOENT;
694}
695EXPORT_SYMBOL(ib_find_gid);
696
697/**
698 * ib_find_pkey - Returns the PKey table index where a specified
699 * PKey value occurs.
700 * @device: The device to query.
701 * @port_num: The port number of the device to search for the PKey.
702 * @pkey: The PKey value to search for.
703 * @index: The index into the PKey table where the PKey was found.
704 */
705int ib_find_pkey(struct ib_device *device,
706 u8 port_num, u16 pkey, u16 *index)
707{
708 int ret, i;
709 u16 tmp_pkey;
710
711 for (i = 0; i < device->pkey_tbl_len[port_num - start_port(device)]; ++i) {
712 ret = ib_query_pkey(device, port_num, i, &tmp_pkey);
713 if (ret)
714 return ret;
715
716 if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) {
717 *index = i;
718 return 0;
719 }
720 }
721
722 return -ENOENT;
723}
724EXPORT_SYMBOL(ib_find_pkey);
725
726static int __init ib_core_init(void)
727{
728 int ret;
729
730 ib_wq = alloc_workqueue("infiniband", 0, 0);
731 if (!ib_wq)
732 return -ENOMEM;
733
734 ret = ib_sysfs_setup();
735 if (ret) {
736 printk(KERN_WARNING "Couldn't create InfiniBand device class\n");
737 goto err;
738 }
739
740 ret = ibnl_init();
741 if (ret) {
742 printk(KERN_WARNING "Couldn't init IB netlink interface\n");
743 goto err_sysfs;
744 }
745
746 ret = ib_cache_setup();
747 if (ret) {
748 printk(KERN_WARNING "Couldn't set up InfiniBand P_Key/GID cache\n");
749 goto err_nl;
750 }
751
752 return 0;
753
754err_nl:
755 ibnl_cleanup();
756
757err_sysfs:
758 ib_sysfs_cleanup();
759
760err:
761 destroy_workqueue(ib_wq);
762 return ret;
763}
764
765static void __exit ib_core_cleanup(void)
766{
767 ib_cache_cleanup();
768 ibnl_cleanup();
769 ib_sysfs_cleanup();
770 /* Make sure that any pending umem accounting work is done. */
771 destroy_workqueue(ib_wq);
772}
773
774module_init(ib_core_init);
775module_exit(ib_core_cleanup);
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
34#include <linux/module.h>
35#include <linux/string.h>
36#include <linux/errno.h>
37#include <linux/kernel.h>
38#include <linux/slab.h>
39#include <linux/init.h>
40#include <linux/mutex.h>
41#include <linux/netdevice.h>
42#include <linux/security.h>
43#include <linux/notifier.h>
44#include <rdma/rdma_netlink.h>
45#include <rdma/ib_addr.h>
46#include <rdma/ib_cache.h>
47
48#include "core_priv.h"
49
50MODULE_AUTHOR("Roland Dreier");
51MODULE_DESCRIPTION("core kernel InfiniBand API");
52MODULE_LICENSE("Dual BSD/GPL");
53
54struct ib_client_data {
55 struct list_head list;
56 struct ib_client *client;
57 void * data;
58 /* The device or client is going down. Do not call client or device
59 * callbacks other than remove(). */
60 bool going_down;
61};
62
63struct workqueue_struct *ib_comp_wq;
64struct workqueue_struct *ib_wq;
65EXPORT_SYMBOL_GPL(ib_wq);
66
67/* The device_list and client_list contain devices and clients after their
68 * registration has completed, and the devices and clients are removed
69 * during unregistration. */
70static LIST_HEAD(device_list);
71static LIST_HEAD(client_list);
72
73/*
74 * device_mutex and lists_rwsem protect access to both device_list and
75 * client_list. device_mutex protects writer access by device and client
76 * registration / de-registration. lists_rwsem protects reader access to
77 * these lists. Iterators of these lists must lock it for read, while updates
78 * to the lists must be done with a write lock. A special case is when the
79 * device_mutex is locked. In this case locking the lists for read access is
80 * not necessary as the device_mutex implies it.
81 *
82 * lists_rwsem also protects access to the client data list.
83 */
84static DEFINE_MUTEX(device_mutex);
85static DECLARE_RWSEM(lists_rwsem);
86
87static int ib_security_change(struct notifier_block *nb, unsigned long event,
88 void *lsm_data);
89static void ib_policy_change_task(struct work_struct *work);
90static DECLARE_WORK(ib_policy_change_work, ib_policy_change_task);
91
92static struct notifier_block ibdev_lsm_nb = {
93 .notifier_call = ib_security_change,
94};
95
96static int ib_device_check_mandatory(struct ib_device *device)
97{
98#define IB_MANDATORY_FUNC(x) { offsetof(struct ib_device, x), #x }
99 static const struct {
100 size_t offset;
101 char *name;
102 } mandatory_table[] = {
103 IB_MANDATORY_FUNC(query_device),
104 IB_MANDATORY_FUNC(query_port),
105 IB_MANDATORY_FUNC(query_pkey),
106 IB_MANDATORY_FUNC(alloc_pd),
107 IB_MANDATORY_FUNC(dealloc_pd),
108 IB_MANDATORY_FUNC(create_ah),
109 IB_MANDATORY_FUNC(destroy_ah),
110 IB_MANDATORY_FUNC(create_qp),
111 IB_MANDATORY_FUNC(modify_qp),
112 IB_MANDATORY_FUNC(destroy_qp),
113 IB_MANDATORY_FUNC(post_send),
114 IB_MANDATORY_FUNC(post_recv),
115 IB_MANDATORY_FUNC(create_cq),
116 IB_MANDATORY_FUNC(destroy_cq),
117 IB_MANDATORY_FUNC(poll_cq),
118 IB_MANDATORY_FUNC(req_notify_cq),
119 IB_MANDATORY_FUNC(get_dma_mr),
120 IB_MANDATORY_FUNC(dereg_mr),
121 IB_MANDATORY_FUNC(get_port_immutable)
122 };
123 int i;
124
125 for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) {
126 if (!*(void **) ((void *) device + mandatory_table[i].offset)) {
127 pr_warn("Device %s is missing mandatory function %s\n",
128 device->name, mandatory_table[i].name);
129 return -EINVAL;
130 }
131 }
132
133 return 0;
134}
135
136static struct ib_device *__ib_device_get_by_index(u32 index)
137{
138 struct ib_device *device;
139
140 list_for_each_entry(device, &device_list, core_list)
141 if (device->index == index)
142 return device;
143
144 return NULL;
145}
146
147/*
148 * Caller is responsible to return refrerence count by calling put_device()
149 */
150struct ib_device *ib_device_get_by_index(u32 index)
151{
152 struct ib_device *device;
153
154 down_read(&lists_rwsem);
155 device = __ib_device_get_by_index(index);
156 if (device)
157 get_device(&device->dev);
158
159 up_read(&lists_rwsem);
160 return device;
161}
162
163static struct ib_device *__ib_device_get_by_name(const char *name)
164{
165 struct ib_device *device;
166
167 list_for_each_entry(device, &device_list, core_list)
168 if (!strncmp(name, device->name, IB_DEVICE_NAME_MAX))
169 return device;
170
171 return NULL;
172}
173
174static int alloc_name(char *name)
175{
176 unsigned long *inuse;
177 char buf[IB_DEVICE_NAME_MAX];
178 struct ib_device *device;
179 int i;
180
181 inuse = (unsigned long *) get_zeroed_page(GFP_KERNEL);
182 if (!inuse)
183 return -ENOMEM;
184
185 list_for_each_entry(device, &device_list, core_list) {
186 if (!sscanf(device->name, name, &i))
187 continue;
188 if (i < 0 || i >= PAGE_SIZE * 8)
189 continue;
190 snprintf(buf, sizeof buf, name, i);
191 if (!strncmp(buf, device->name, IB_DEVICE_NAME_MAX))
192 set_bit(i, inuse);
193 }
194
195 i = find_first_zero_bit(inuse, PAGE_SIZE * 8);
196 free_page((unsigned long) inuse);
197 snprintf(buf, sizeof buf, name, i);
198
199 if (__ib_device_get_by_name(buf))
200 return -ENFILE;
201
202 strlcpy(name, buf, IB_DEVICE_NAME_MAX);
203 return 0;
204}
205
206static void ib_device_release(struct device *device)
207{
208 struct ib_device *dev = container_of(device, struct ib_device, dev);
209
210 WARN_ON(dev->reg_state == IB_DEV_REGISTERED);
211 if (dev->reg_state == IB_DEV_UNREGISTERED) {
212 /*
213 * In IB_DEV_UNINITIALIZED state, cache or port table
214 * is not even created. Free cache and port table only when
215 * device reaches UNREGISTERED state.
216 */
217 ib_cache_release_one(dev);
218 kfree(dev->port_immutable);
219 }
220 kfree(dev);
221}
222
223static int ib_device_uevent(struct device *device,
224 struct kobj_uevent_env *env)
225{
226 struct ib_device *dev = container_of(device, struct ib_device, dev);
227
228 if (add_uevent_var(env, "NAME=%s", dev->name))
229 return -ENOMEM;
230
231 /*
232 * It would be nice to pass the node GUID with the event...
233 */
234
235 return 0;
236}
237
238static struct class ib_class = {
239 .name = "infiniband",
240 .dev_release = ib_device_release,
241 .dev_uevent = ib_device_uevent,
242};
243
244/**
245 * ib_alloc_device - allocate an IB device struct
246 * @size:size of structure to allocate
247 *
248 * Low-level drivers should use ib_alloc_device() to allocate &struct
249 * ib_device. @size is the size of the structure to be allocated,
250 * including any private data used by the low-level driver.
251 * ib_dealloc_device() must be used to free structures allocated with
252 * ib_alloc_device().
253 */
254struct ib_device *ib_alloc_device(size_t size)
255{
256 struct ib_device *device;
257
258 if (WARN_ON(size < sizeof(struct ib_device)))
259 return NULL;
260
261 device = kzalloc(size, GFP_KERNEL);
262 if (!device)
263 return NULL;
264
265 rdma_restrack_init(&device->res);
266
267 device->dev.class = &ib_class;
268 device_initialize(&device->dev);
269
270 dev_set_drvdata(&device->dev, device);
271
272 INIT_LIST_HEAD(&device->event_handler_list);
273 spin_lock_init(&device->event_handler_lock);
274 spin_lock_init(&device->client_data_lock);
275 INIT_LIST_HEAD(&device->client_data_list);
276 INIT_LIST_HEAD(&device->port_list);
277
278 return device;
279}
280EXPORT_SYMBOL(ib_alloc_device);
281
282/**
283 * ib_dealloc_device - free an IB device struct
284 * @device:structure to free
285 *
286 * Free a structure allocated with ib_alloc_device().
287 */
288void ib_dealloc_device(struct ib_device *device)
289{
290 WARN_ON(device->reg_state != IB_DEV_UNREGISTERED &&
291 device->reg_state != IB_DEV_UNINITIALIZED);
292 rdma_restrack_clean(&device->res);
293 put_device(&device->dev);
294}
295EXPORT_SYMBOL(ib_dealloc_device);
296
297static int add_client_context(struct ib_device *device, struct ib_client *client)
298{
299 struct ib_client_data *context;
300 unsigned long flags;
301
302 context = kmalloc(sizeof *context, GFP_KERNEL);
303 if (!context)
304 return -ENOMEM;
305
306 context->client = client;
307 context->data = NULL;
308 context->going_down = false;
309
310 down_write(&lists_rwsem);
311 spin_lock_irqsave(&device->client_data_lock, flags);
312 list_add(&context->list, &device->client_data_list);
313 spin_unlock_irqrestore(&device->client_data_lock, flags);
314 up_write(&lists_rwsem);
315
316 return 0;
317}
318
319static int verify_immutable(const struct ib_device *dev, u8 port)
320{
321 return WARN_ON(!rdma_cap_ib_mad(dev, port) &&
322 rdma_max_mad_size(dev, port) != 0);
323}
324
325static int read_port_immutable(struct ib_device *device)
326{
327 int ret;
328 u8 start_port = rdma_start_port(device);
329 u8 end_port = rdma_end_port(device);
330 u8 port;
331
332 /**
333 * device->port_immutable is indexed directly by the port number to make
334 * access to this data as efficient as possible.
335 *
336 * Therefore port_immutable is declared as a 1 based array with
337 * potential empty slots at the beginning.
338 */
339 device->port_immutable = kzalloc(sizeof(*device->port_immutable)
340 * (end_port + 1),
341 GFP_KERNEL);
342 if (!device->port_immutable)
343 return -ENOMEM;
344
345 for (port = start_port; port <= end_port; ++port) {
346 ret = device->get_port_immutable(device, port,
347 &device->port_immutable[port]);
348 if (ret)
349 return ret;
350
351 if (verify_immutable(device, port))
352 return -EINVAL;
353 }
354 return 0;
355}
356
357void ib_get_device_fw_str(struct ib_device *dev, char *str)
358{
359 if (dev->get_dev_fw_str)
360 dev->get_dev_fw_str(dev, str);
361 else
362 str[0] = '\0';
363}
364EXPORT_SYMBOL(ib_get_device_fw_str);
365
366static int setup_port_pkey_list(struct ib_device *device)
367{
368 int i;
369
370 /**
371 * device->port_pkey_list is indexed directly by the port number,
372 * Therefore it is declared as a 1 based array with potential empty
373 * slots at the beginning.
374 */
375 device->port_pkey_list = kcalloc(rdma_end_port(device) + 1,
376 sizeof(*device->port_pkey_list),
377 GFP_KERNEL);
378
379 if (!device->port_pkey_list)
380 return -ENOMEM;
381
382 for (i = 0; i < (rdma_end_port(device) + 1); i++) {
383 spin_lock_init(&device->port_pkey_list[i].list_lock);
384 INIT_LIST_HEAD(&device->port_pkey_list[i].pkey_list);
385 }
386
387 return 0;
388}
389
390static void ib_policy_change_task(struct work_struct *work)
391{
392 struct ib_device *dev;
393
394 down_read(&lists_rwsem);
395 list_for_each_entry(dev, &device_list, core_list) {
396 int i;
397
398 for (i = rdma_start_port(dev); i <= rdma_end_port(dev); i++) {
399 u64 sp;
400 int ret = ib_get_cached_subnet_prefix(dev,
401 i,
402 &sp);
403
404 WARN_ONCE(ret,
405 "ib_get_cached_subnet_prefix err: %d, this should never happen here\n",
406 ret);
407 if (!ret)
408 ib_security_cache_change(dev, i, sp);
409 }
410 }
411 up_read(&lists_rwsem);
412}
413
414static int ib_security_change(struct notifier_block *nb, unsigned long event,
415 void *lsm_data)
416{
417 if (event != LSM_POLICY_CHANGE)
418 return NOTIFY_DONE;
419
420 schedule_work(&ib_policy_change_work);
421
422 return NOTIFY_OK;
423}
424
425/**
426 * __dev_new_index - allocate an device index
427 *
428 * Returns a suitable unique value for a new device interface
429 * number. It assumes that there are less than 2^32-1 ib devices
430 * will be present in the system.
431 */
432static u32 __dev_new_index(void)
433{
434 /*
435 * The device index to allow stable naming.
436 * Similar to struct net -> ifindex.
437 */
438 static u32 index;
439
440 for (;;) {
441 if (!(++index))
442 index = 1;
443
444 if (!__ib_device_get_by_index(index))
445 return index;
446 }
447}
448
449/**
450 * ib_register_device - Register an IB device with IB core
451 * @device:Device to register
452 *
453 * Low-level drivers use ib_register_device() to register their
454 * devices with the IB core. All registered clients will receive a
455 * callback for each device that is added. @device must be allocated
456 * with ib_alloc_device().
457 */
458int ib_register_device(struct ib_device *device,
459 int (*port_callback)(struct ib_device *,
460 u8, struct kobject *))
461{
462 int ret;
463 struct ib_client *client;
464 struct ib_udata uhw = {.outlen = 0, .inlen = 0};
465 struct device *parent = device->dev.parent;
466
467 WARN_ON_ONCE(device->dma_device);
468 if (device->dev.dma_ops) {
469 /*
470 * The caller provided custom DMA operations. Copy the
471 * DMA-related fields that are used by e.g. dma_alloc_coherent()
472 * into device->dev.
473 */
474 device->dma_device = &device->dev;
475 if (!device->dev.dma_mask) {
476 if (parent)
477 device->dev.dma_mask = parent->dma_mask;
478 else
479 WARN_ON_ONCE(true);
480 }
481 if (!device->dev.coherent_dma_mask) {
482 if (parent)
483 device->dev.coherent_dma_mask =
484 parent->coherent_dma_mask;
485 else
486 WARN_ON_ONCE(true);
487 }
488 } else {
489 /*
490 * The caller did not provide custom DMA operations. Use the
491 * DMA mapping operations of the parent device.
492 */
493 WARN_ON_ONCE(!parent);
494 device->dma_device = parent;
495 }
496
497 mutex_lock(&device_mutex);
498
499 if (strchr(device->name, '%')) {
500 ret = alloc_name(device->name);
501 if (ret)
502 goto out;
503 }
504
505 if (ib_device_check_mandatory(device)) {
506 ret = -EINVAL;
507 goto out;
508 }
509
510 ret = read_port_immutable(device);
511 if (ret) {
512 pr_warn("Couldn't create per port immutable data %s\n",
513 device->name);
514 goto out;
515 }
516
517 ret = setup_port_pkey_list(device);
518 if (ret) {
519 pr_warn("Couldn't create per port_pkey_list\n");
520 goto out;
521 }
522
523 ret = ib_cache_setup_one(device);
524 if (ret) {
525 pr_warn("Couldn't set up InfiniBand P_Key/GID cache\n");
526 goto port_cleanup;
527 }
528
529 ret = ib_device_register_rdmacg(device);
530 if (ret) {
531 pr_warn("Couldn't register device with rdma cgroup\n");
532 goto cache_cleanup;
533 }
534
535 memset(&device->attrs, 0, sizeof(device->attrs));
536 ret = device->query_device(device, &device->attrs, &uhw);
537 if (ret) {
538 pr_warn("Couldn't query the device attributes\n");
539 goto cg_cleanup;
540 }
541
542 ret = ib_device_register_sysfs(device, port_callback);
543 if (ret) {
544 pr_warn("Couldn't register device %s with driver model\n",
545 device->name);
546 goto cg_cleanup;
547 }
548
549 device->reg_state = IB_DEV_REGISTERED;
550
551 list_for_each_entry(client, &client_list, list)
552 if (!add_client_context(device, client) && client->add)
553 client->add(device);
554
555 device->index = __dev_new_index();
556 down_write(&lists_rwsem);
557 list_add_tail(&device->core_list, &device_list);
558 up_write(&lists_rwsem);
559 mutex_unlock(&device_mutex);
560 return 0;
561
562cg_cleanup:
563 ib_device_unregister_rdmacg(device);
564cache_cleanup:
565 ib_cache_cleanup_one(device);
566 ib_cache_release_one(device);
567port_cleanup:
568 kfree(device->port_immutable);
569out:
570 mutex_unlock(&device_mutex);
571 return ret;
572}
573EXPORT_SYMBOL(ib_register_device);
574
575/**
576 * ib_unregister_device - Unregister an IB device
577 * @device:Device to unregister
578 *
579 * Unregister an IB device. All clients will receive a remove callback.
580 */
581void ib_unregister_device(struct ib_device *device)
582{
583 struct ib_client_data *context, *tmp;
584 unsigned long flags;
585
586 mutex_lock(&device_mutex);
587
588 down_write(&lists_rwsem);
589 list_del(&device->core_list);
590 spin_lock_irqsave(&device->client_data_lock, flags);
591 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
592 context->going_down = true;
593 spin_unlock_irqrestore(&device->client_data_lock, flags);
594 downgrade_write(&lists_rwsem);
595
596 list_for_each_entry_safe(context, tmp, &device->client_data_list,
597 list) {
598 if (context->client->remove)
599 context->client->remove(device, context->data);
600 }
601 up_read(&lists_rwsem);
602
603 ib_device_unregister_rdmacg(device);
604 ib_device_unregister_sysfs(device);
605
606 mutex_unlock(&device_mutex);
607
608 ib_cache_cleanup_one(device);
609
610 ib_security_destroy_port_pkey_list(device);
611 kfree(device->port_pkey_list);
612
613 down_write(&lists_rwsem);
614 spin_lock_irqsave(&device->client_data_lock, flags);
615 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
616 kfree(context);
617 spin_unlock_irqrestore(&device->client_data_lock, flags);
618 up_write(&lists_rwsem);
619
620 device->reg_state = IB_DEV_UNREGISTERED;
621}
622EXPORT_SYMBOL(ib_unregister_device);
623
624/**
625 * ib_register_client - Register an IB client
626 * @client:Client to register
627 *
628 * Upper level users of the IB drivers can use ib_register_client() to
629 * register callbacks for IB device addition and removal. When an IB
630 * device is added, each registered client's add method will be called
631 * (in the order the clients were registered), and when a device is
632 * removed, each client's remove method will be called (in the reverse
633 * order that clients were registered). In addition, when
634 * ib_register_client() is called, the client will receive an add
635 * callback for all devices already registered.
636 */
637int ib_register_client(struct ib_client *client)
638{
639 struct ib_device *device;
640
641 mutex_lock(&device_mutex);
642
643 list_for_each_entry(device, &device_list, core_list)
644 if (!add_client_context(device, client) && client->add)
645 client->add(device);
646
647 down_write(&lists_rwsem);
648 list_add_tail(&client->list, &client_list);
649 up_write(&lists_rwsem);
650
651 mutex_unlock(&device_mutex);
652
653 return 0;
654}
655EXPORT_SYMBOL(ib_register_client);
656
657/**
658 * ib_unregister_client - Unregister an IB client
659 * @client:Client to unregister
660 *
661 * Upper level users use ib_unregister_client() to remove their client
662 * registration. When ib_unregister_client() is called, the client
663 * will receive a remove callback for each IB device still registered.
664 */
665void ib_unregister_client(struct ib_client *client)
666{
667 struct ib_client_data *context, *tmp;
668 struct ib_device *device;
669 unsigned long flags;
670
671 mutex_lock(&device_mutex);
672
673 down_write(&lists_rwsem);
674 list_del(&client->list);
675 up_write(&lists_rwsem);
676
677 list_for_each_entry(device, &device_list, core_list) {
678 struct ib_client_data *found_context = NULL;
679
680 down_write(&lists_rwsem);
681 spin_lock_irqsave(&device->client_data_lock, flags);
682 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
683 if (context->client == client) {
684 context->going_down = true;
685 found_context = context;
686 break;
687 }
688 spin_unlock_irqrestore(&device->client_data_lock, flags);
689 up_write(&lists_rwsem);
690
691 if (client->remove)
692 client->remove(device, found_context ?
693 found_context->data : NULL);
694
695 if (!found_context) {
696 pr_warn("No client context found for %s/%s\n",
697 device->name, client->name);
698 continue;
699 }
700
701 down_write(&lists_rwsem);
702 spin_lock_irqsave(&device->client_data_lock, flags);
703 list_del(&found_context->list);
704 kfree(found_context);
705 spin_unlock_irqrestore(&device->client_data_lock, flags);
706 up_write(&lists_rwsem);
707 }
708
709 mutex_unlock(&device_mutex);
710}
711EXPORT_SYMBOL(ib_unregister_client);
712
713/**
714 * ib_get_client_data - Get IB client context
715 * @device:Device to get context for
716 * @client:Client to get context for
717 *
718 * ib_get_client_data() returns client context set with
719 * ib_set_client_data().
720 */
721void *ib_get_client_data(struct ib_device *device, struct ib_client *client)
722{
723 struct ib_client_data *context;
724 void *ret = NULL;
725 unsigned long flags;
726
727 spin_lock_irqsave(&device->client_data_lock, flags);
728 list_for_each_entry(context, &device->client_data_list, list)
729 if (context->client == client) {
730 ret = context->data;
731 break;
732 }
733 spin_unlock_irqrestore(&device->client_data_lock, flags);
734
735 return ret;
736}
737EXPORT_SYMBOL(ib_get_client_data);
738
739/**
740 * ib_set_client_data - Set IB client context
741 * @device:Device to set context for
742 * @client:Client to set context for
743 * @data:Context to set
744 *
745 * ib_set_client_data() sets client context that can be retrieved with
746 * ib_get_client_data().
747 */
748void ib_set_client_data(struct ib_device *device, struct ib_client *client,
749 void *data)
750{
751 struct ib_client_data *context;
752 unsigned long flags;
753
754 spin_lock_irqsave(&device->client_data_lock, flags);
755 list_for_each_entry(context, &device->client_data_list, list)
756 if (context->client == client) {
757 context->data = data;
758 goto out;
759 }
760
761 pr_warn("No client context found for %s/%s\n",
762 device->name, client->name);
763
764out:
765 spin_unlock_irqrestore(&device->client_data_lock, flags);
766}
767EXPORT_SYMBOL(ib_set_client_data);
768
769/**
770 * ib_register_event_handler - Register an IB event handler
771 * @event_handler:Handler to register
772 *
773 * ib_register_event_handler() registers an event handler that will be
774 * called back when asynchronous IB events occur (as defined in
775 * chapter 11 of the InfiniBand Architecture Specification). This
776 * callback may occur in interrupt context.
777 */
778void ib_register_event_handler(struct ib_event_handler *event_handler)
779{
780 unsigned long flags;
781
782 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
783 list_add_tail(&event_handler->list,
784 &event_handler->device->event_handler_list);
785 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
786}
787EXPORT_SYMBOL(ib_register_event_handler);
788
789/**
790 * ib_unregister_event_handler - Unregister an event handler
791 * @event_handler:Handler to unregister
792 *
793 * Unregister an event handler registered with
794 * ib_register_event_handler().
795 */
796void ib_unregister_event_handler(struct ib_event_handler *event_handler)
797{
798 unsigned long flags;
799
800 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
801 list_del(&event_handler->list);
802 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
803}
804EXPORT_SYMBOL(ib_unregister_event_handler);
805
806/**
807 * ib_dispatch_event - Dispatch an asynchronous event
808 * @event:Event to dispatch
809 *
810 * Low-level drivers must call ib_dispatch_event() to dispatch the
811 * event to all registered event handlers when an asynchronous event
812 * occurs.
813 */
814void ib_dispatch_event(struct ib_event *event)
815{
816 unsigned long flags;
817 struct ib_event_handler *handler;
818
819 spin_lock_irqsave(&event->device->event_handler_lock, flags);
820
821 list_for_each_entry(handler, &event->device->event_handler_list, list)
822 handler->handler(handler, event);
823
824 spin_unlock_irqrestore(&event->device->event_handler_lock, flags);
825}
826EXPORT_SYMBOL(ib_dispatch_event);
827
828/**
829 * ib_query_port - Query IB port attributes
830 * @device:Device to query
831 * @port_num:Port number to query
832 * @port_attr:Port attributes
833 *
834 * ib_query_port() returns the attributes of a port through the
835 * @port_attr pointer.
836 */
837int ib_query_port(struct ib_device *device,
838 u8 port_num,
839 struct ib_port_attr *port_attr)
840{
841 union ib_gid gid;
842 int err;
843
844 if (!rdma_is_port_valid(device, port_num))
845 return -EINVAL;
846
847 memset(port_attr, 0, sizeof(*port_attr));
848 err = device->query_port(device, port_num, port_attr);
849 if (err || port_attr->subnet_prefix)
850 return err;
851
852 if (rdma_port_get_link_layer(device, port_num) != IB_LINK_LAYER_INFINIBAND)
853 return 0;
854
855 err = device->query_gid(device, port_num, 0, &gid);
856 if (err)
857 return err;
858
859 port_attr->subnet_prefix = be64_to_cpu(gid.global.subnet_prefix);
860 return 0;
861}
862EXPORT_SYMBOL(ib_query_port);
863
864/**
865 * ib_query_gid - Get GID table entry
866 * @device:Device to query
867 * @port_num:Port number to query
868 * @index:GID table index to query
869 * @gid:Returned GID
870 * @attr: Returned GID attributes related to this GID index (only in RoCE).
871 * NULL means ignore.
872 *
873 * ib_query_gid() fetches the specified GID table entry from the cache.
874 */
875int ib_query_gid(struct ib_device *device,
876 u8 port_num, int index, union ib_gid *gid,
877 struct ib_gid_attr *attr)
878{
879 return ib_get_cached_gid(device, port_num, index, gid, attr);
880}
881EXPORT_SYMBOL(ib_query_gid);
882
883/**
884 * ib_enum_roce_netdev - enumerate all RoCE ports
885 * @ib_dev : IB device we want to query
886 * @filter: Should we call the callback?
887 * @filter_cookie: Cookie passed to filter
888 * @cb: Callback to call for each found RoCE ports
889 * @cookie: Cookie passed back to the callback
890 *
891 * Enumerates all of the physical RoCE ports of ib_dev
892 * which are related to netdevice and calls callback() on each
893 * device for which filter() function returns non zero.
894 */
895void ib_enum_roce_netdev(struct ib_device *ib_dev,
896 roce_netdev_filter filter,
897 void *filter_cookie,
898 roce_netdev_callback cb,
899 void *cookie)
900{
901 u8 port;
902
903 for (port = rdma_start_port(ib_dev); port <= rdma_end_port(ib_dev);
904 port++)
905 if (rdma_protocol_roce(ib_dev, port)) {
906 struct net_device *idev = NULL;
907
908 if (ib_dev->get_netdev)
909 idev = ib_dev->get_netdev(ib_dev, port);
910
911 if (idev &&
912 idev->reg_state >= NETREG_UNREGISTERED) {
913 dev_put(idev);
914 idev = NULL;
915 }
916
917 if (filter(ib_dev, port, idev, filter_cookie))
918 cb(ib_dev, port, idev, cookie);
919
920 if (idev)
921 dev_put(idev);
922 }
923}
924
925/**
926 * ib_enum_all_roce_netdevs - enumerate all RoCE devices
927 * @filter: Should we call the callback?
928 * @filter_cookie: Cookie passed to filter
929 * @cb: Callback to call for each found RoCE ports
930 * @cookie: Cookie passed back to the callback
931 *
932 * Enumerates all RoCE devices' physical ports which are related
933 * to netdevices and calls callback() on each device for which
934 * filter() function returns non zero.
935 */
936void ib_enum_all_roce_netdevs(roce_netdev_filter filter,
937 void *filter_cookie,
938 roce_netdev_callback cb,
939 void *cookie)
940{
941 struct ib_device *dev;
942
943 down_read(&lists_rwsem);
944 list_for_each_entry(dev, &device_list, core_list)
945 ib_enum_roce_netdev(dev, filter, filter_cookie, cb, cookie);
946 up_read(&lists_rwsem);
947}
948
949/**
950 * ib_enum_all_devs - enumerate all ib_devices
951 * @cb: Callback to call for each found ib_device
952 *
953 * Enumerates all ib_devices and calls callback() on each device.
954 */
955int ib_enum_all_devs(nldev_callback nldev_cb, struct sk_buff *skb,
956 struct netlink_callback *cb)
957{
958 struct ib_device *dev;
959 unsigned int idx = 0;
960 int ret = 0;
961
962 down_read(&lists_rwsem);
963 list_for_each_entry(dev, &device_list, core_list) {
964 ret = nldev_cb(dev, skb, cb, idx);
965 if (ret)
966 break;
967 idx++;
968 }
969
970 up_read(&lists_rwsem);
971 return ret;
972}
973
974/**
975 * ib_query_pkey - Get P_Key table entry
976 * @device:Device to query
977 * @port_num:Port number to query
978 * @index:P_Key table index to query
979 * @pkey:Returned P_Key
980 *
981 * ib_query_pkey() fetches the specified P_Key table entry.
982 */
983int ib_query_pkey(struct ib_device *device,
984 u8 port_num, u16 index, u16 *pkey)
985{
986 return device->query_pkey(device, port_num, index, pkey);
987}
988EXPORT_SYMBOL(ib_query_pkey);
989
990/**
991 * ib_modify_device - Change IB device attributes
992 * @device:Device to modify
993 * @device_modify_mask:Mask of attributes to change
994 * @device_modify:New attribute values
995 *
996 * ib_modify_device() changes a device's attributes as specified by
997 * the @device_modify_mask and @device_modify structure.
998 */
999int ib_modify_device(struct ib_device *device,
1000 int device_modify_mask,
1001 struct ib_device_modify *device_modify)
1002{
1003 if (!device->modify_device)
1004 return -ENOSYS;
1005
1006 return device->modify_device(device, device_modify_mask,
1007 device_modify);
1008}
1009EXPORT_SYMBOL(ib_modify_device);
1010
1011/**
1012 * ib_modify_port - Modifies the attributes for the specified port.
1013 * @device: The device to modify.
1014 * @port_num: The number of the port to modify.
1015 * @port_modify_mask: Mask used to specify which attributes of the port
1016 * to change.
1017 * @port_modify: New attribute values for the port.
1018 *
1019 * ib_modify_port() changes a port's attributes as specified by the
1020 * @port_modify_mask and @port_modify structure.
1021 */
1022int ib_modify_port(struct ib_device *device,
1023 u8 port_num, int port_modify_mask,
1024 struct ib_port_modify *port_modify)
1025{
1026 int rc;
1027
1028 if (!rdma_is_port_valid(device, port_num))
1029 return -EINVAL;
1030
1031 if (device->modify_port)
1032 rc = device->modify_port(device, port_num, port_modify_mask,
1033 port_modify);
1034 else
1035 rc = rdma_protocol_roce(device, port_num) ? 0 : -ENOSYS;
1036 return rc;
1037}
1038EXPORT_SYMBOL(ib_modify_port);
1039
1040/**
1041 * ib_find_gid - Returns the port number and GID table index where
1042 * a specified GID value occurs. Its searches only for IB link layer.
1043 * @device: The device to query.
1044 * @gid: The GID value to search for.
1045 * @port_num: The port number of the device where the GID value was found.
1046 * @index: The index into the GID table where the GID was found. This
1047 * parameter may be NULL.
1048 */
1049int ib_find_gid(struct ib_device *device, union ib_gid *gid,
1050 u8 *port_num, u16 *index)
1051{
1052 union ib_gid tmp_gid;
1053 int ret, port, i;
1054
1055 for (port = rdma_start_port(device); port <= rdma_end_port(device); ++port) {
1056 if (!rdma_protocol_ib(device, port))
1057 continue;
1058
1059 for (i = 0; i < device->port_immutable[port].gid_tbl_len; ++i) {
1060 ret = ib_query_gid(device, port, i, &tmp_gid, NULL);
1061 if (ret)
1062 return ret;
1063 if (!memcmp(&tmp_gid, gid, sizeof *gid)) {
1064 *port_num = port;
1065 if (index)
1066 *index = i;
1067 return 0;
1068 }
1069 }
1070 }
1071
1072 return -ENOENT;
1073}
1074EXPORT_SYMBOL(ib_find_gid);
1075
1076/**
1077 * ib_find_pkey - Returns the PKey table index where a specified
1078 * PKey value occurs.
1079 * @device: The device to query.
1080 * @port_num: The port number of the device to search for the PKey.
1081 * @pkey: The PKey value to search for.
1082 * @index: The index into the PKey table where the PKey was found.
1083 */
1084int ib_find_pkey(struct ib_device *device,
1085 u8 port_num, u16 pkey, u16 *index)
1086{
1087 int ret, i;
1088 u16 tmp_pkey;
1089 int partial_ix = -1;
1090
1091 for (i = 0; i < device->port_immutable[port_num].pkey_tbl_len; ++i) {
1092 ret = ib_query_pkey(device, port_num, i, &tmp_pkey);
1093 if (ret)
1094 return ret;
1095 if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) {
1096 /* if there is full-member pkey take it.*/
1097 if (tmp_pkey & 0x8000) {
1098 *index = i;
1099 return 0;
1100 }
1101 if (partial_ix < 0)
1102 partial_ix = i;
1103 }
1104 }
1105
1106 /*no full-member, if exists take the limited*/
1107 if (partial_ix >= 0) {
1108 *index = partial_ix;
1109 return 0;
1110 }
1111 return -ENOENT;
1112}
1113EXPORT_SYMBOL(ib_find_pkey);
1114
1115/**
1116 * ib_get_net_dev_by_params() - Return the appropriate net_dev
1117 * for a received CM request
1118 * @dev: An RDMA device on which the request has been received.
1119 * @port: Port number on the RDMA device.
1120 * @pkey: The Pkey the request came on.
1121 * @gid: A GID that the net_dev uses to communicate.
1122 * @addr: Contains the IP address that the request specified as its
1123 * destination.
1124 */
1125struct net_device *ib_get_net_dev_by_params(struct ib_device *dev,
1126 u8 port,
1127 u16 pkey,
1128 const union ib_gid *gid,
1129 const struct sockaddr *addr)
1130{
1131 struct net_device *net_dev = NULL;
1132 struct ib_client_data *context;
1133
1134 if (!rdma_protocol_ib(dev, port))
1135 return NULL;
1136
1137 down_read(&lists_rwsem);
1138
1139 list_for_each_entry(context, &dev->client_data_list, list) {
1140 struct ib_client *client = context->client;
1141
1142 if (context->going_down)
1143 continue;
1144
1145 if (client->get_net_dev_by_params) {
1146 net_dev = client->get_net_dev_by_params(dev, port, pkey,
1147 gid, addr,
1148 context->data);
1149 if (net_dev)
1150 break;
1151 }
1152 }
1153
1154 up_read(&lists_rwsem);
1155
1156 return net_dev;
1157}
1158EXPORT_SYMBOL(ib_get_net_dev_by_params);
1159
1160static const struct rdma_nl_cbs ibnl_ls_cb_table[RDMA_NL_LS_NUM_OPS] = {
1161 [RDMA_NL_LS_OP_RESOLVE] = {
1162 .doit = ib_nl_handle_resolve_resp,
1163 .flags = RDMA_NL_ADMIN_PERM,
1164 },
1165 [RDMA_NL_LS_OP_SET_TIMEOUT] = {
1166 .doit = ib_nl_handle_set_timeout,
1167 .flags = RDMA_NL_ADMIN_PERM,
1168 },
1169 [RDMA_NL_LS_OP_IP_RESOLVE] = {
1170 .doit = ib_nl_handle_ip_res_resp,
1171 .flags = RDMA_NL_ADMIN_PERM,
1172 },
1173};
1174
1175static int __init ib_core_init(void)
1176{
1177 int ret;
1178
1179 ib_wq = alloc_workqueue("infiniband", 0, 0);
1180 if (!ib_wq)
1181 return -ENOMEM;
1182
1183 ib_comp_wq = alloc_workqueue("ib-comp-wq",
1184 WQ_HIGHPRI | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
1185 if (!ib_comp_wq) {
1186 ret = -ENOMEM;
1187 goto err;
1188 }
1189
1190 ret = class_register(&ib_class);
1191 if (ret) {
1192 pr_warn("Couldn't create InfiniBand device class\n");
1193 goto err_comp;
1194 }
1195
1196 ret = rdma_nl_init();
1197 if (ret) {
1198 pr_warn("Couldn't init IB netlink interface: err %d\n", ret);
1199 goto err_sysfs;
1200 }
1201
1202 ret = addr_init();
1203 if (ret) {
1204 pr_warn("Could't init IB address resolution\n");
1205 goto err_ibnl;
1206 }
1207
1208 ret = ib_mad_init();
1209 if (ret) {
1210 pr_warn("Couldn't init IB MAD\n");
1211 goto err_addr;
1212 }
1213
1214 ret = ib_sa_init();
1215 if (ret) {
1216 pr_warn("Couldn't init SA\n");
1217 goto err_mad;
1218 }
1219
1220 ret = register_lsm_notifier(&ibdev_lsm_nb);
1221 if (ret) {
1222 pr_warn("Couldn't register LSM notifier. ret %d\n", ret);
1223 goto err_sa;
1224 }
1225
1226 nldev_init();
1227 rdma_nl_register(RDMA_NL_LS, ibnl_ls_cb_table);
1228 ib_cache_setup();
1229
1230 return 0;
1231
1232err_sa:
1233 ib_sa_cleanup();
1234err_mad:
1235 ib_mad_cleanup();
1236err_addr:
1237 addr_cleanup();
1238err_ibnl:
1239 rdma_nl_exit();
1240err_sysfs:
1241 class_unregister(&ib_class);
1242err_comp:
1243 destroy_workqueue(ib_comp_wq);
1244err:
1245 destroy_workqueue(ib_wq);
1246 return ret;
1247}
1248
1249static void __exit ib_core_cleanup(void)
1250{
1251 ib_cache_cleanup();
1252 nldev_exit();
1253 rdma_nl_unregister(RDMA_NL_LS);
1254 unregister_lsm_notifier(&ibdev_lsm_nb);
1255 ib_sa_cleanup();
1256 ib_mad_cleanup();
1257 addr_cleanup();
1258 rdma_nl_exit();
1259 class_unregister(&ib_class);
1260 destroy_workqueue(ib_comp_wq);
1261 /* Make sure that any pending umem accounting work is done. */
1262 destroy_workqueue(ib_wq);
1263}
1264
1265MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_LS, 4);
1266
1267subsys_initcall(ib_core_init);
1268module_exit(ib_core_cleanup);