Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (c) 2004 Topspin Communications. All rights reserved. | |
2a1d9b7f | 3 | * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved. |
1da177e4 LT |
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. | |
1da177e4 LT |
32 | */ |
33 | ||
34 | #include <linux/module.h> | |
35 | #include <linux/string.h> | |
36 | #include <linux/errno.h> | |
9a6b090c | 37 | #include <linux/kernel.h> |
1da177e4 LT |
38 | #include <linux/slab.h> |
39 | #include <linux/init.h> | |
95ed644f | 40 | #include <linux/mutex.h> |
9268f72d | 41 | #include <linux/netdevice.h> |
b2cbae2c | 42 | #include <rdma/rdma_netlink.h> |
03db3a2d MB |
43 | #include <rdma/ib_addr.h> |
44 | #include <rdma/ib_cache.h> | |
1da177e4 LT |
45 | |
46 | #include "core_priv.h" | |
47 | ||
48 | MODULE_AUTHOR("Roland Dreier"); | |
49 | MODULE_DESCRIPTION("core kernel InfiniBand API"); | |
50 | MODULE_LICENSE("Dual BSD/GPL"); | |
51 | ||
52 | struct ib_client_data { | |
53 | struct list_head list; | |
54 | struct ib_client *client; | |
55 | void * data; | |
7c1eb45a HE |
56 | /* The device or client is going down. Do not call client or device |
57 | * callbacks other than remove(). */ | |
58 | bool going_down; | |
1da177e4 LT |
59 | }; |
60 | ||
14d3a3b2 | 61 | struct workqueue_struct *ib_comp_wq; |
f0626710 TH |
62 | struct workqueue_struct *ib_wq; |
63 | EXPORT_SYMBOL_GPL(ib_wq); | |
64 | ||
5aa44bb9 HE |
65 | /* The device_list and client_list contain devices and clients after their |
66 | * registration has completed, and the devices and clients are removed | |
67 | * during unregistration. */ | |
1da177e4 LT |
68 | static LIST_HEAD(device_list); |
69 | static LIST_HEAD(client_list); | |
70 | ||
71 | /* | |
5aa44bb9 HE |
72 | * device_mutex and lists_rwsem protect access to both device_list and |
73 | * client_list. device_mutex protects writer access by device and client | |
74 | * registration / de-registration. lists_rwsem protects reader access to | |
75 | * these lists. Iterators of these lists must lock it for read, while updates | |
76 | * to the lists must be done with a write lock. A special case is when the | |
77 | * device_mutex is locked. In this case locking the lists for read access is | |
78 | * not necessary as the device_mutex implies it. | |
7c1eb45a HE |
79 | * |
80 | * lists_rwsem also protects access to the client data list. | |
1da177e4 | 81 | */ |
95ed644f | 82 | static DEFINE_MUTEX(device_mutex); |
5aa44bb9 HE |
83 | static DECLARE_RWSEM(lists_rwsem); |
84 | ||
1da177e4 LT |
85 | |
86 | static int ib_device_check_mandatory(struct ib_device *device) | |
87 | { | |
88 | #define IB_MANDATORY_FUNC(x) { offsetof(struct ib_device, x), #x } | |
89 | static const struct { | |
90 | size_t offset; | |
91 | char *name; | |
92 | } mandatory_table[] = { | |
93 | IB_MANDATORY_FUNC(query_device), | |
94 | IB_MANDATORY_FUNC(query_port), | |
95 | IB_MANDATORY_FUNC(query_pkey), | |
96 | IB_MANDATORY_FUNC(query_gid), | |
97 | IB_MANDATORY_FUNC(alloc_pd), | |
98 | IB_MANDATORY_FUNC(dealloc_pd), | |
99 | IB_MANDATORY_FUNC(create_ah), | |
100 | IB_MANDATORY_FUNC(destroy_ah), | |
101 | IB_MANDATORY_FUNC(create_qp), | |
102 | IB_MANDATORY_FUNC(modify_qp), | |
103 | IB_MANDATORY_FUNC(destroy_qp), | |
104 | IB_MANDATORY_FUNC(post_send), | |
105 | IB_MANDATORY_FUNC(post_recv), | |
106 | IB_MANDATORY_FUNC(create_cq), | |
107 | IB_MANDATORY_FUNC(destroy_cq), | |
108 | IB_MANDATORY_FUNC(poll_cq), | |
109 | IB_MANDATORY_FUNC(req_notify_cq), | |
110 | IB_MANDATORY_FUNC(get_dma_mr), | |
7738613e IW |
111 | IB_MANDATORY_FUNC(dereg_mr), |
112 | IB_MANDATORY_FUNC(get_port_immutable) | |
1da177e4 LT |
113 | }; |
114 | int i; | |
115 | ||
9a6b090c | 116 | for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) { |
1da177e4 | 117 | if (!*(void **) ((void *) device + mandatory_table[i].offset)) { |
aba25a3e PP |
118 | pr_warn("Device %s is missing mandatory function %s\n", |
119 | device->name, mandatory_table[i].name); | |
1da177e4 LT |
120 | return -EINVAL; |
121 | } | |
122 | } | |
123 | ||
124 | return 0; | |
125 | } | |
126 | ||
127 | static struct ib_device *__ib_device_get_by_name(const char *name) | |
128 | { | |
129 | struct ib_device *device; | |
130 | ||
131 | list_for_each_entry(device, &device_list, core_list) | |
132 | if (!strncmp(name, device->name, IB_DEVICE_NAME_MAX)) | |
133 | return device; | |
134 | ||
135 | return NULL; | |
136 | } | |
137 | ||
138 | ||
139 | static int alloc_name(char *name) | |
140 | { | |
65d470b3 | 141 | unsigned long *inuse; |
1da177e4 LT |
142 | char buf[IB_DEVICE_NAME_MAX]; |
143 | struct ib_device *device; | |
144 | int i; | |
145 | ||
65d470b3 | 146 | inuse = (unsigned long *) get_zeroed_page(GFP_KERNEL); |
1da177e4 LT |
147 | if (!inuse) |
148 | return -ENOMEM; | |
149 | ||
150 | list_for_each_entry(device, &device_list, core_list) { | |
151 | if (!sscanf(device->name, name, &i)) | |
152 | continue; | |
153 | if (i < 0 || i >= PAGE_SIZE * 8) | |
154 | continue; | |
155 | snprintf(buf, sizeof buf, name, i); | |
156 | if (!strncmp(buf, device->name, IB_DEVICE_NAME_MAX)) | |
157 | set_bit(i, inuse); | |
158 | } | |
159 | ||
160 | i = find_first_zero_bit(inuse, PAGE_SIZE * 8); | |
161 | free_page((unsigned long) inuse); | |
162 | snprintf(buf, sizeof buf, name, i); | |
163 | ||
164 | if (__ib_device_get_by_name(buf)) | |
165 | return -ENFILE; | |
166 | ||
167 | strlcpy(name, buf, IB_DEVICE_NAME_MAX); | |
168 | return 0; | |
169 | } | |
170 | ||
55aeed06 JG |
171 | static void ib_device_release(struct device *device) |
172 | { | |
173 | struct ib_device *dev = container_of(device, struct ib_device, dev); | |
174 | ||
03db3a2d | 175 | ib_cache_release_one(dev); |
55aeed06 JG |
176 | kfree(dev->port_immutable); |
177 | kfree(dev); | |
178 | } | |
179 | ||
180 | static int ib_device_uevent(struct device *device, | |
181 | struct kobj_uevent_env *env) | |
182 | { | |
183 | struct ib_device *dev = container_of(device, struct ib_device, dev); | |
184 | ||
185 | if (add_uevent_var(env, "NAME=%s", dev->name)) | |
186 | return -ENOMEM; | |
187 | ||
188 | /* | |
189 | * It would be nice to pass the node GUID with the event... | |
190 | */ | |
191 | ||
192 | return 0; | |
193 | } | |
194 | ||
195 | static struct class ib_class = { | |
196 | .name = "infiniband", | |
197 | .dev_release = ib_device_release, | |
198 | .dev_uevent = ib_device_uevent, | |
199 | }; | |
200 | ||
1da177e4 LT |
201 | /** |
202 | * ib_alloc_device - allocate an IB device struct | |
203 | * @size:size of structure to allocate | |
204 | * | |
205 | * Low-level drivers should use ib_alloc_device() to allocate &struct | |
206 | * ib_device. @size is the size of the structure to be allocated, | |
207 | * including any private data used by the low-level driver. | |
208 | * ib_dealloc_device() must be used to free structures allocated with | |
209 | * ib_alloc_device(). | |
210 | */ | |
211 | struct ib_device *ib_alloc_device(size_t size) | |
212 | { | |
55aeed06 JG |
213 | struct ib_device *device; |
214 | ||
215 | if (WARN_ON(size < sizeof(struct ib_device))) | |
216 | return NULL; | |
217 | ||
218 | device = kzalloc(size, GFP_KERNEL); | |
219 | if (!device) | |
220 | return NULL; | |
221 | ||
222 | device->dev.class = &ib_class; | |
223 | device_initialize(&device->dev); | |
224 | ||
225 | dev_set_drvdata(&device->dev, device); | |
226 | ||
227 | INIT_LIST_HEAD(&device->event_handler_list); | |
228 | spin_lock_init(&device->event_handler_lock); | |
229 | spin_lock_init(&device->client_data_lock); | |
230 | INIT_LIST_HEAD(&device->client_data_list); | |
231 | INIT_LIST_HEAD(&device->port_list); | |
1da177e4 | 232 | |
55aeed06 | 233 | return device; |
1da177e4 LT |
234 | } |
235 | EXPORT_SYMBOL(ib_alloc_device); | |
236 | ||
237 | /** | |
238 | * ib_dealloc_device - free an IB device struct | |
239 | * @device:structure to free | |
240 | * | |
241 | * Free a structure allocated with ib_alloc_device(). | |
242 | */ | |
243 | void ib_dealloc_device(struct ib_device *device) | |
244 | { | |
55aeed06 JG |
245 | WARN_ON(device->reg_state != IB_DEV_UNREGISTERED && |
246 | device->reg_state != IB_DEV_UNINITIALIZED); | |
9206dff1 | 247 | kobject_put(&device->dev.kobj); |
1da177e4 LT |
248 | } |
249 | EXPORT_SYMBOL(ib_dealloc_device); | |
250 | ||
251 | static int add_client_context(struct ib_device *device, struct ib_client *client) | |
252 | { | |
253 | struct ib_client_data *context; | |
254 | unsigned long flags; | |
255 | ||
256 | context = kmalloc(sizeof *context, GFP_KERNEL); | |
257 | if (!context) { | |
aba25a3e PP |
258 | pr_warn("Couldn't allocate client context for %s/%s\n", |
259 | device->name, client->name); | |
1da177e4 LT |
260 | return -ENOMEM; |
261 | } | |
262 | ||
263 | context->client = client; | |
264 | context->data = NULL; | |
7c1eb45a | 265 | context->going_down = false; |
1da177e4 | 266 | |
7c1eb45a | 267 | down_write(&lists_rwsem); |
1da177e4 LT |
268 | spin_lock_irqsave(&device->client_data_lock, flags); |
269 | list_add(&context->list, &device->client_data_list); | |
270 | spin_unlock_irqrestore(&device->client_data_lock, flags); | |
7c1eb45a | 271 | up_write(&lists_rwsem); |
1da177e4 LT |
272 | |
273 | return 0; | |
274 | } | |
275 | ||
337877a4 IW |
276 | static int verify_immutable(const struct ib_device *dev, u8 port) |
277 | { | |
278 | return WARN_ON(!rdma_cap_ib_mad(dev, port) && | |
279 | rdma_max_mad_size(dev, port) != 0); | |
280 | } | |
281 | ||
7738613e | 282 | static int read_port_immutable(struct ib_device *device) |
5eb620c8 | 283 | { |
55aeed06 | 284 | int ret; |
7738613e IW |
285 | u8 start_port = rdma_start_port(device); |
286 | u8 end_port = rdma_end_port(device); | |
287 | u8 port; | |
288 | ||
289 | /** | |
290 | * device->port_immutable is indexed directly by the port number to make | |
291 | * access to this data as efficient as possible. | |
292 | * | |
293 | * Therefore port_immutable is declared as a 1 based array with | |
294 | * potential empty slots at the beginning. | |
295 | */ | |
296 | device->port_immutable = kzalloc(sizeof(*device->port_immutable) | |
297 | * (end_port + 1), | |
298 | GFP_KERNEL); | |
299 | if (!device->port_immutable) | |
55aeed06 | 300 | return -ENOMEM; |
5eb620c8 | 301 | |
7738613e IW |
302 | for (port = start_port; port <= end_port; ++port) { |
303 | ret = device->get_port_immutable(device, port, | |
304 | &device->port_immutable[port]); | |
5eb620c8 | 305 | if (ret) |
55aeed06 | 306 | return ret; |
337877a4 | 307 | |
55aeed06 JG |
308 | if (verify_immutable(device, port)) |
309 | return -EINVAL; | |
5eb620c8 | 310 | } |
55aeed06 | 311 | return 0; |
5eb620c8 YE |
312 | } |
313 | ||
5fa76c20 IW |
314 | void ib_get_device_fw_str(struct ib_device *dev, char *str, size_t str_len) |
315 | { | |
316 | if (dev->get_dev_fw_str) | |
317 | dev->get_dev_fw_str(dev, str, str_len); | |
318 | else | |
319 | str[0] = '\0'; | |
320 | } | |
321 | EXPORT_SYMBOL(ib_get_device_fw_str); | |
322 | ||
1da177e4 LT |
323 | /** |
324 | * ib_register_device - Register an IB device with IB core | |
325 | * @device:Device to register | |
326 | * | |
327 | * Low-level drivers use ib_register_device() to register their | |
328 | * devices with the IB core. All registered clients will receive a | |
329 | * callback for each device that is added. @device must be allocated | |
330 | * with ib_alloc_device(). | |
331 | */ | |
9a6edb60 RC |
332 | int ib_register_device(struct ib_device *device, |
333 | int (*port_callback)(struct ib_device *, | |
334 | u8, struct kobject *)) | |
1da177e4 LT |
335 | { |
336 | int ret; | |
b8071ad8 | 337 | struct ib_client *client; |
3e153a93 | 338 | struct ib_udata uhw = {.outlen = 0, .inlen = 0}; |
1da177e4 | 339 | |
95ed644f | 340 | mutex_lock(&device_mutex); |
1da177e4 LT |
341 | |
342 | if (strchr(device->name, '%')) { | |
343 | ret = alloc_name(device->name); | |
344 | if (ret) | |
345 | goto out; | |
346 | } | |
347 | ||
348 | if (ib_device_check_mandatory(device)) { | |
349 | ret = -EINVAL; | |
350 | goto out; | |
351 | } | |
352 | ||
7738613e | 353 | ret = read_port_immutable(device); |
5eb620c8 | 354 | if (ret) { |
aba25a3e PP |
355 | pr_warn("Couldn't create per port immutable data %s\n", |
356 | device->name); | |
5eb620c8 YE |
357 | goto out; |
358 | } | |
359 | ||
03db3a2d MB |
360 | ret = ib_cache_setup_one(device); |
361 | if (ret) { | |
aba25a3e | 362 | pr_warn("Couldn't set up InfiniBand P_Key/GID cache\n"); |
03db3a2d MB |
363 | goto out; |
364 | } | |
365 | ||
3e153a93 IW |
366 | memset(&device->attrs, 0, sizeof(device->attrs)); |
367 | ret = device->query_device(device, &device->attrs, &uhw); | |
368 | if (ret) { | |
aba25a3e | 369 | pr_warn("Couldn't query the device attributes\n"); |
5adebafb | 370 | ib_cache_cleanup_one(device); |
3e153a93 IW |
371 | goto out; |
372 | } | |
373 | ||
9a6edb60 | 374 | ret = ib_device_register_sysfs(device, port_callback); |
1da177e4 | 375 | if (ret) { |
aba25a3e PP |
376 | pr_warn("Couldn't register device %s with driver model\n", |
377 | device->name); | |
03db3a2d | 378 | ib_cache_cleanup_one(device); |
1da177e4 LT |
379 | goto out; |
380 | } | |
381 | ||
1da177e4 LT |
382 | device->reg_state = IB_DEV_REGISTERED; |
383 | ||
b8071ad8 DL |
384 | list_for_each_entry(client, &client_list, list) |
385 | if (client->add && !add_client_context(device, client)) | |
386 | client->add(device); | |
1da177e4 | 387 | |
5aa44bb9 HE |
388 | down_write(&lists_rwsem); |
389 | list_add_tail(&device->core_list, &device_list); | |
390 | up_write(&lists_rwsem); | |
391 | out: | |
95ed644f | 392 | mutex_unlock(&device_mutex); |
1da177e4 LT |
393 | return ret; |
394 | } | |
395 | EXPORT_SYMBOL(ib_register_device); | |
396 | ||
397 | /** | |
398 | * ib_unregister_device - Unregister an IB device | |
399 | * @device:Device to unregister | |
400 | * | |
401 | * Unregister an IB device. All clients will receive a remove callback. | |
402 | */ | |
403 | void ib_unregister_device(struct ib_device *device) | |
404 | { | |
1da177e4 LT |
405 | struct ib_client_data *context, *tmp; |
406 | unsigned long flags; | |
407 | ||
95ed644f | 408 | mutex_lock(&device_mutex); |
1da177e4 | 409 | |
5aa44bb9 HE |
410 | down_write(&lists_rwsem); |
411 | list_del(&device->core_list); | |
7c1eb45a HE |
412 | spin_lock_irqsave(&device->client_data_lock, flags); |
413 | list_for_each_entry_safe(context, tmp, &device->client_data_list, list) | |
414 | context->going_down = true; | |
415 | spin_unlock_irqrestore(&device->client_data_lock, flags); | |
416 | downgrade_write(&lists_rwsem); | |
5aa44bb9 | 417 | |
7c1eb45a HE |
418 | list_for_each_entry_safe(context, tmp, &device->client_data_list, |
419 | list) { | |
420 | if (context->client->remove) | |
421 | context->client->remove(device, context->data); | |
422 | } | |
423 | up_read(&lists_rwsem); | |
1da177e4 | 424 | |
95ed644f | 425 | mutex_unlock(&device_mutex); |
1da177e4 | 426 | |
9206dff1 | 427 | ib_device_unregister_sysfs(device); |
03db3a2d | 428 | ib_cache_cleanup_one(device); |
9206dff1 | 429 | |
7c1eb45a | 430 | down_write(&lists_rwsem); |
1da177e4 LT |
431 | spin_lock_irqsave(&device->client_data_lock, flags); |
432 | list_for_each_entry_safe(context, tmp, &device->client_data_list, list) | |
433 | kfree(context); | |
434 | spin_unlock_irqrestore(&device->client_data_lock, flags); | |
7c1eb45a | 435 | up_write(&lists_rwsem); |
1da177e4 LT |
436 | |
437 | device->reg_state = IB_DEV_UNREGISTERED; | |
438 | } | |
439 | EXPORT_SYMBOL(ib_unregister_device); | |
440 | ||
441 | /** | |
442 | * ib_register_client - Register an IB client | |
443 | * @client:Client to register | |
444 | * | |
445 | * Upper level users of the IB drivers can use ib_register_client() to | |
446 | * register callbacks for IB device addition and removal. When an IB | |
447 | * device is added, each registered client's add method will be called | |
448 | * (in the order the clients were registered), and when a device is | |
449 | * removed, each client's remove method will be called (in the reverse | |
450 | * order that clients were registered). In addition, when | |
451 | * ib_register_client() is called, the client will receive an add | |
452 | * callback for all devices already registered. | |
453 | */ | |
454 | int ib_register_client(struct ib_client *client) | |
455 | { | |
456 | struct ib_device *device; | |
457 | ||
95ed644f | 458 | mutex_lock(&device_mutex); |
1da177e4 | 459 | |
1da177e4 LT |
460 | list_for_each_entry(device, &device_list, core_list) |
461 | if (client->add && !add_client_context(device, client)) | |
462 | client->add(device); | |
463 | ||
5aa44bb9 HE |
464 | down_write(&lists_rwsem); |
465 | list_add_tail(&client->list, &client_list); | |
466 | up_write(&lists_rwsem); | |
467 | ||
95ed644f | 468 | mutex_unlock(&device_mutex); |
1da177e4 LT |
469 | |
470 | return 0; | |
471 | } | |
472 | EXPORT_SYMBOL(ib_register_client); | |
473 | ||
474 | /** | |
475 | * ib_unregister_client - Unregister an IB client | |
476 | * @client:Client to unregister | |
477 | * | |
478 | * Upper level users use ib_unregister_client() to remove their client | |
479 | * registration. When ib_unregister_client() is called, the client | |
480 | * will receive a remove callback for each IB device still registered. | |
481 | */ | |
482 | void ib_unregister_client(struct ib_client *client) | |
483 | { | |
484 | struct ib_client_data *context, *tmp; | |
485 | struct ib_device *device; | |
486 | unsigned long flags; | |
487 | ||
95ed644f | 488 | mutex_lock(&device_mutex); |
1da177e4 | 489 | |
5aa44bb9 HE |
490 | down_write(&lists_rwsem); |
491 | list_del(&client->list); | |
492 | up_write(&lists_rwsem); | |
493 | ||
1da177e4 | 494 | list_for_each_entry(device, &device_list, core_list) { |
7c1eb45a | 495 | struct ib_client_data *found_context = NULL; |
1da177e4 | 496 | |
7c1eb45a | 497 | down_write(&lists_rwsem); |
1da177e4 LT |
498 | spin_lock_irqsave(&device->client_data_lock, flags); |
499 | list_for_each_entry_safe(context, tmp, &device->client_data_list, list) | |
500 | if (context->client == client) { | |
7c1eb45a HE |
501 | context->going_down = true; |
502 | found_context = context; | |
503 | break; | |
1da177e4 LT |
504 | } |
505 | spin_unlock_irqrestore(&device->client_data_lock, flags); | |
7c1eb45a HE |
506 | up_write(&lists_rwsem); |
507 | ||
508 | if (client->remove) | |
509 | client->remove(device, found_context ? | |
510 | found_context->data : NULL); | |
511 | ||
512 | if (!found_context) { | |
513 | pr_warn("No client context found for %s/%s\n", | |
514 | device->name, client->name); | |
515 | continue; | |
516 | } | |
517 | ||
518 | down_write(&lists_rwsem); | |
519 | spin_lock_irqsave(&device->client_data_lock, flags); | |
520 | list_del(&found_context->list); | |
521 | kfree(found_context); | |
522 | spin_unlock_irqrestore(&device->client_data_lock, flags); | |
523 | up_write(&lists_rwsem); | |
1da177e4 | 524 | } |
1da177e4 | 525 | |
95ed644f | 526 | mutex_unlock(&device_mutex); |
1da177e4 LT |
527 | } |
528 | EXPORT_SYMBOL(ib_unregister_client); | |
529 | ||
530 | /** | |
531 | * ib_get_client_data - Get IB client context | |
532 | * @device:Device to get context for | |
533 | * @client:Client to get context for | |
534 | * | |
535 | * ib_get_client_data() returns client context set with | |
536 | * ib_set_client_data(). | |
537 | */ | |
538 | void *ib_get_client_data(struct ib_device *device, struct ib_client *client) | |
539 | { | |
540 | struct ib_client_data *context; | |
541 | void *ret = NULL; | |
542 | unsigned long flags; | |
543 | ||
544 | spin_lock_irqsave(&device->client_data_lock, flags); | |
545 | list_for_each_entry(context, &device->client_data_list, list) | |
546 | if (context->client == client) { | |
547 | ret = context->data; | |
548 | break; | |
549 | } | |
550 | spin_unlock_irqrestore(&device->client_data_lock, flags); | |
551 | ||
552 | return ret; | |
553 | } | |
554 | EXPORT_SYMBOL(ib_get_client_data); | |
555 | ||
556 | /** | |
9cd330d3 | 557 | * ib_set_client_data - Set IB client context |
1da177e4 LT |
558 | * @device:Device to set context for |
559 | * @client:Client to set context for | |
560 | * @data:Context to set | |
561 | * | |
562 | * ib_set_client_data() sets client context that can be retrieved with | |
563 | * ib_get_client_data(). | |
564 | */ | |
565 | void ib_set_client_data(struct ib_device *device, struct ib_client *client, | |
566 | void *data) | |
567 | { | |
568 | struct ib_client_data *context; | |
569 | unsigned long flags; | |
570 | ||
571 | spin_lock_irqsave(&device->client_data_lock, flags); | |
572 | list_for_each_entry(context, &device->client_data_list, list) | |
573 | if (context->client == client) { | |
574 | context->data = data; | |
575 | goto out; | |
576 | } | |
577 | ||
aba25a3e PP |
578 | pr_warn("No client context found for %s/%s\n", |
579 | device->name, client->name); | |
1da177e4 LT |
580 | |
581 | out: | |
582 | spin_unlock_irqrestore(&device->client_data_lock, flags); | |
583 | } | |
584 | EXPORT_SYMBOL(ib_set_client_data); | |
585 | ||
586 | /** | |
587 | * ib_register_event_handler - Register an IB event handler | |
588 | * @event_handler:Handler to register | |
589 | * | |
590 | * ib_register_event_handler() registers an event handler that will be | |
591 | * called back when asynchronous IB events occur (as defined in | |
592 | * chapter 11 of the InfiniBand Architecture Specification). This | |
593 | * callback may occur in interrupt context. | |
594 | */ | |
595 | int ib_register_event_handler (struct ib_event_handler *event_handler) | |
596 | { | |
597 | unsigned long flags; | |
598 | ||
599 | spin_lock_irqsave(&event_handler->device->event_handler_lock, flags); | |
600 | list_add_tail(&event_handler->list, | |
601 | &event_handler->device->event_handler_list); | |
602 | spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags); | |
603 | ||
604 | return 0; | |
605 | } | |
606 | EXPORT_SYMBOL(ib_register_event_handler); | |
607 | ||
608 | /** | |
609 | * ib_unregister_event_handler - Unregister an event handler | |
610 | * @event_handler:Handler to unregister | |
611 | * | |
612 | * Unregister an event handler registered with | |
613 | * ib_register_event_handler(). | |
614 | */ | |
615 | int ib_unregister_event_handler(struct ib_event_handler *event_handler) | |
616 | { | |
617 | unsigned long flags; | |
618 | ||
619 | spin_lock_irqsave(&event_handler->device->event_handler_lock, flags); | |
620 | list_del(&event_handler->list); | |
621 | spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags); | |
622 | ||
623 | return 0; | |
624 | } | |
625 | EXPORT_SYMBOL(ib_unregister_event_handler); | |
626 | ||
627 | /** | |
628 | * ib_dispatch_event - Dispatch an asynchronous event | |
629 | * @event:Event to dispatch | |
630 | * | |
631 | * Low-level drivers must call ib_dispatch_event() to dispatch the | |
632 | * event to all registered event handlers when an asynchronous event | |
633 | * occurs. | |
634 | */ | |
635 | void ib_dispatch_event(struct ib_event *event) | |
636 | { | |
637 | unsigned long flags; | |
638 | struct ib_event_handler *handler; | |
639 | ||
640 | spin_lock_irqsave(&event->device->event_handler_lock, flags); | |
641 | ||
642 | list_for_each_entry(handler, &event->device->event_handler_list, list) | |
643 | handler->handler(handler, event); | |
644 | ||
645 | spin_unlock_irqrestore(&event->device->event_handler_lock, flags); | |
646 | } | |
647 | EXPORT_SYMBOL(ib_dispatch_event); | |
648 | ||
1da177e4 LT |
649 | /** |
650 | * ib_query_port - Query IB port attributes | |
651 | * @device:Device to query | |
652 | * @port_num:Port number to query | |
653 | * @port_attr:Port attributes | |
654 | * | |
655 | * ib_query_port() returns the attributes of a port through the | |
656 | * @port_attr pointer. | |
657 | */ | |
658 | int ib_query_port(struct ib_device *device, | |
659 | u8 port_num, | |
660 | struct ib_port_attr *port_attr) | |
661 | { | |
fad61ad4 EC |
662 | union ib_gid gid; |
663 | int err; | |
664 | ||
0cf18d77 | 665 | if (port_num < rdma_start_port(device) || port_num > rdma_end_port(device)) |
116c0074 RD |
666 | return -EINVAL; |
667 | ||
fad61ad4 EC |
668 | memset(port_attr, 0, sizeof(*port_attr)); |
669 | err = device->query_port(device, port_num, port_attr); | |
670 | if (err || port_attr->subnet_prefix) | |
671 | return err; | |
672 | ||
d7012467 EC |
673 | if (rdma_port_get_link_layer(device, port_num) != IB_LINK_LAYER_INFINIBAND) |
674 | return 0; | |
675 | ||
fad61ad4 EC |
676 | err = ib_query_gid(device, port_num, 0, &gid, NULL); |
677 | if (err) | |
678 | return err; | |
679 | ||
680 | port_attr->subnet_prefix = be64_to_cpu(gid.global.subnet_prefix); | |
681 | return 0; | |
1da177e4 LT |
682 | } |
683 | EXPORT_SYMBOL(ib_query_port); | |
684 | ||
685 | /** | |
686 | * ib_query_gid - Get GID table entry | |
687 | * @device:Device to query | |
688 | * @port_num:Port number to query | |
689 | * @index:GID table index to query | |
690 | * @gid:Returned GID | |
55ee3ab2 MB |
691 | * @attr: Returned GID attributes related to this GID index (only in RoCE). |
692 | * NULL means ignore. | |
1da177e4 LT |
693 | * |
694 | * ib_query_gid() fetches the specified GID table entry. | |
695 | */ | |
696 | int ib_query_gid(struct ib_device *device, | |
55ee3ab2 MB |
697 | u8 port_num, int index, union ib_gid *gid, |
698 | struct ib_gid_attr *attr) | |
1da177e4 | 699 | { |
03db3a2d | 700 | if (rdma_cap_roce_gid_table(device, port_num)) |
55ee3ab2 MB |
701 | return ib_get_cached_gid(device, port_num, index, gid, attr); |
702 | ||
703 | if (attr) | |
704 | return -EINVAL; | |
03db3a2d | 705 | |
1da177e4 LT |
706 | return device->query_gid(device, port_num, index, gid); |
707 | } | |
708 | EXPORT_SYMBOL(ib_query_gid); | |
709 | ||
03db3a2d MB |
710 | /** |
711 | * ib_enum_roce_netdev - enumerate all RoCE ports | |
712 | * @ib_dev : IB device we want to query | |
713 | * @filter: Should we call the callback? | |
714 | * @filter_cookie: Cookie passed to filter | |
715 | * @cb: Callback to call for each found RoCE ports | |
716 | * @cookie: Cookie passed back to the callback | |
717 | * | |
718 | * Enumerates all of the physical RoCE ports of ib_dev | |
719 | * which are related to netdevice and calls callback() on each | |
720 | * device for which filter() function returns non zero. | |
721 | */ | |
722 | void ib_enum_roce_netdev(struct ib_device *ib_dev, | |
723 | roce_netdev_filter filter, | |
724 | void *filter_cookie, | |
725 | roce_netdev_callback cb, | |
726 | void *cookie) | |
727 | { | |
728 | u8 port; | |
729 | ||
730 | for (port = rdma_start_port(ib_dev); port <= rdma_end_port(ib_dev); | |
731 | port++) | |
732 | if (rdma_protocol_roce(ib_dev, port)) { | |
733 | struct net_device *idev = NULL; | |
734 | ||
735 | if (ib_dev->get_netdev) | |
736 | idev = ib_dev->get_netdev(ib_dev, port); | |
737 | ||
738 | if (idev && | |
739 | idev->reg_state >= NETREG_UNREGISTERED) { | |
740 | dev_put(idev); | |
741 | idev = NULL; | |
742 | } | |
743 | ||
744 | if (filter(ib_dev, port, idev, filter_cookie)) | |
745 | cb(ib_dev, port, idev, cookie); | |
746 | ||
747 | if (idev) | |
748 | dev_put(idev); | |
749 | } | |
750 | } | |
751 | ||
752 | /** | |
753 | * ib_enum_all_roce_netdevs - enumerate all RoCE devices | |
754 | * @filter: Should we call the callback? | |
755 | * @filter_cookie: Cookie passed to filter | |
756 | * @cb: Callback to call for each found RoCE ports | |
757 | * @cookie: Cookie passed back to the callback | |
758 | * | |
759 | * Enumerates all RoCE devices' physical ports which are related | |
760 | * to netdevices and calls callback() on each device for which | |
761 | * filter() function returns non zero. | |
762 | */ | |
763 | void ib_enum_all_roce_netdevs(roce_netdev_filter filter, | |
764 | void *filter_cookie, | |
765 | roce_netdev_callback cb, | |
766 | void *cookie) | |
767 | { | |
768 | struct ib_device *dev; | |
769 | ||
770 | down_read(&lists_rwsem); | |
771 | list_for_each_entry(dev, &device_list, core_list) | |
772 | ib_enum_roce_netdev(dev, filter, filter_cookie, cb, cookie); | |
773 | up_read(&lists_rwsem); | |
774 | } | |
775 | ||
1da177e4 LT |
776 | /** |
777 | * ib_query_pkey - Get P_Key table entry | |
778 | * @device:Device to query | |
779 | * @port_num:Port number to query | |
780 | * @index:P_Key table index to query | |
781 | * @pkey:Returned P_Key | |
782 | * | |
783 | * ib_query_pkey() fetches the specified P_Key table entry. | |
784 | */ | |
785 | int ib_query_pkey(struct ib_device *device, | |
786 | u8 port_num, u16 index, u16 *pkey) | |
787 | { | |
788 | return device->query_pkey(device, port_num, index, pkey); | |
789 | } | |
790 | EXPORT_SYMBOL(ib_query_pkey); | |
791 | ||
792 | /** | |
793 | * ib_modify_device - Change IB device attributes | |
794 | * @device:Device to modify | |
795 | * @device_modify_mask:Mask of attributes to change | |
796 | * @device_modify:New attribute values | |
797 | * | |
798 | * ib_modify_device() changes a device's attributes as specified by | |
799 | * the @device_modify_mask and @device_modify structure. | |
800 | */ | |
801 | int ib_modify_device(struct ib_device *device, | |
802 | int device_modify_mask, | |
803 | struct ib_device_modify *device_modify) | |
804 | { | |
10e1b54b BVA |
805 | if (!device->modify_device) |
806 | return -ENOSYS; | |
807 | ||
1da177e4 LT |
808 | return device->modify_device(device, device_modify_mask, |
809 | device_modify); | |
810 | } | |
811 | EXPORT_SYMBOL(ib_modify_device); | |
812 | ||
813 | /** | |
814 | * ib_modify_port - Modifies the attributes for the specified port. | |
815 | * @device: The device to modify. | |
816 | * @port_num: The number of the port to modify. | |
817 | * @port_modify_mask: Mask used to specify which attributes of the port | |
818 | * to change. | |
819 | * @port_modify: New attribute values for the port. | |
820 | * | |
821 | * ib_modify_port() changes a port's attributes as specified by the | |
822 | * @port_modify_mask and @port_modify structure. | |
823 | */ | |
824 | int ib_modify_port(struct ib_device *device, | |
825 | u8 port_num, int port_modify_mask, | |
826 | struct ib_port_modify *port_modify) | |
827 | { | |
10e1b54b BVA |
828 | if (!device->modify_port) |
829 | return -ENOSYS; | |
830 | ||
0cf18d77 | 831 | if (port_num < rdma_start_port(device) || port_num > rdma_end_port(device)) |
116c0074 RD |
832 | return -EINVAL; |
833 | ||
1da177e4 LT |
834 | return device->modify_port(device, port_num, port_modify_mask, |
835 | port_modify); | |
836 | } | |
837 | EXPORT_SYMBOL(ib_modify_port); | |
838 | ||
5eb620c8 YE |
839 | /** |
840 | * ib_find_gid - Returns the port number and GID table index where | |
841 | * a specified GID value occurs. | |
842 | * @device: The device to query. | |
843 | * @gid: The GID value to search for. | |
b39ffa1d | 844 | * @gid_type: Type of GID. |
55ee3ab2 | 845 | * @ndev: The ndev related to the GID to search for. |
5eb620c8 YE |
846 | * @port_num: The port number of the device where the GID value was found. |
847 | * @index: The index into the GID table where the GID was found. This | |
848 | * parameter may be NULL. | |
849 | */ | |
850 | int ib_find_gid(struct ib_device *device, union ib_gid *gid, | |
b39ffa1d MB |
851 | enum ib_gid_type gid_type, struct net_device *ndev, |
852 | u8 *port_num, u16 *index) | |
5eb620c8 YE |
853 | { |
854 | union ib_gid tmp_gid; | |
855 | int ret, port, i; | |
856 | ||
0cf18d77 | 857 | for (port = rdma_start_port(device); port <= rdma_end_port(device); ++port) { |
03db3a2d | 858 | if (rdma_cap_roce_gid_table(device, port)) { |
b39ffa1d | 859 | if (!ib_find_cached_gid_by_port(device, gid, gid_type, port, |
d300ec52 | 860 | ndev, index)) { |
03db3a2d MB |
861 | *port_num = port; |
862 | return 0; | |
98d25afa | 863 | } |
03db3a2d MB |
864 | } |
865 | ||
b39ffa1d MB |
866 | if (gid_type != IB_GID_TYPE_IB) |
867 | continue; | |
868 | ||
7738613e | 869 | for (i = 0; i < device->port_immutable[port].gid_tbl_len; ++i) { |
55ee3ab2 | 870 | ret = ib_query_gid(device, port, i, &tmp_gid, NULL); |
5eb620c8 YE |
871 | if (ret) |
872 | return ret; | |
873 | if (!memcmp(&tmp_gid, gid, sizeof *gid)) { | |
874 | *port_num = port; | |
875 | if (index) | |
876 | *index = i; | |
877 | return 0; | |
878 | } | |
879 | } | |
880 | } | |
881 | ||
882 | return -ENOENT; | |
883 | } | |
884 | EXPORT_SYMBOL(ib_find_gid); | |
885 | ||
886 | /** | |
887 | * ib_find_pkey - Returns the PKey table index where a specified | |
888 | * PKey value occurs. | |
889 | * @device: The device to query. | |
890 | * @port_num: The port number of the device to search for the PKey. | |
891 | * @pkey: The PKey value to search for. | |
892 | * @index: The index into the PKey table where the PKey was found. | |
893 | */ | |
894 | int ib_find_pkey(struct ib_device *device, | |
895 | u8 port_num, u16 pkey, u16 *index) | |
896 | { | |
897 | int ret, i; | |
898 | u16 tmp_pkey; | |
ff7166c4 | 899 | int partial_ix = -1; |
5eb620c8 | 900 | |
7738613e | 901 | for (i = 0; i < device->port_immutable[port_num].pkey_tbl_len; ++i) { |
5eb620c8 YE |
902 | ret = ib_query_pkey(device, port_num, i, &tmp_pkey); |
903 | if (ret) | |
904 | return ret; | |
36026ecc | 905 | if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) { |
ff7166c4 JM |
906 | /* if there is full-member pkey take it.*/ |
907 | if (tmp_pkey & 0x8000) { | |
908 | *index = i; | |
909 | return 0; | |
910 | } | |
911 | if (partial_ix < 0) | |
912 | partial_ix = i; | |
5eb620c8 YE |
913 | } |
914 | } | |
915 | ||
ff7166c4 JM |
916 | /*no full-member, if exists take the limited*/ |
917 | if (partial_ix >= 0) { | |
918 | *index = partial_ix; | |
919 | return 0; | |
920 | } | |
5eb620c8 YE |
921 | return -ENOENT; |
922 | } | |
923 | EXPORT_SYMBOL(ib_find_pkey); | |
924 | ||
9268f72d YK |
925 | /** |
926 | * ib_get_net_dev_by_params() - Return the appropriate net_dev | |
927 | * for a received CM request | |
928 | * @dev: An RDMA device on which the request has been received. | |
929 | * @port: Port number on the RDMA device. | |
930 | * @pkey: The Pkey the request came on. | |
931 | * @gid: A GID that the net_dev uses to communicate. | |
932 | * @addr: Contains the IP address that the request specified as its | |
933 | * destination. | |
934 | */ | |
935 | struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, | |
936 | u8 port, | |
937 | u16 pkey, | |
938 | const union ib_gid *gid, | |
939 | const struct sockaddr *addr) | |
940 | { | |
941 | struct net_device *net_dev = NULL; | |
942 | struct ib_client_data *context; | |
943 | ||
944 | if (!rdma_protocol_ib(dev, port)) | |
945 | return NULL; | |
946 | ||
947 | down_read(&lists_rwsem); | |
948 | ||
949 | list_for_each_entry(context, &dev->client_data_list, list) { | |
950 | struct ib_client *client = context->client; | |
951 | ||
952 | if (context->going_down) | |
953 | continue; | |
954 | ||
955 | if (client->get_net_dev_by_params) { | |
956 | net_dev = client->get_net_dev_by_params(dev, port, pkey, | |
957 | gid, addr, | |
958 | context->data); | |
959 | if (net_dev) | |
960 | break; | |
961 | } | |
962 | } | |
963 | ||
964 | up_read(&lists_rwsem); | |
965 | ||
966 | return net_dev; | |
967 | } | |
968 | EXPORT_SYMBOL(ib_get_net_dev_by_params); | |
969 | ||
735c631a MB |
970 | static struct ibnl_client_cbs ibnl_ls_cb_table[] = { |
971 | [RDMA_NL_LS_OP_RESOLVE] = { | |
972 | .dump = ib_nl_handle_resolve_resp, | |
973 | .module = THIS_MODULE }, | |
974 | [RDMA_NL_LS_OP_SET_TIMEOUT] = { | |
975 | .dump = ib_nl_handle_set_timeout, | |
976 | .module = THIS_MODULE }, | |
ae43f828 MB |
977 | [RDMA_NL_LS_OP_IP_RESOLVE] = { |
978 | .dump = ib_nl_handle_ip_res_resp, | |
979 | .module = THIS_MODULE }, | |
735c631a MB |
980 | }; |
981 | ||
982 | static int ib_add_ibnl_clients(void) | |
983 | { | |
984 | return ibnl_add_client(RDMA_NL_LS, ARRAY_SIZE(ibnl_ls_cb_table), | |
985 | ibnl_ls_cb_table); | |
986 | } | |
987 | ||
988 | static void ib_remove_ibnl_clients(void) | |
989 | { | |
990 | ibnl_remove_client(RDMA_NL_LS); | |
991 | } | |
992 | ||
1da177e4 LT |
993 | static int __init ib_core_init(void) |
994 | { | |
995 | int ret; | |
996 | ||
f0626710 TH |
997 | ib_wq = alloc_workqueue("infiniband", 0, 0); |
998 | if (!ib_wq) | |
999 | return -ENOMEM; | |
1000 | ||
14d3a3b2 CH |
1001 | ib_comp_wq = alloc_workqueue("ib-comp-wq", |
1002 | WQ_UNBOUND | WQ_HIGHPRI | WQ_MEM_RECLAIM, | |
1003 | WQ_UNBOUND_MAX_ACTIVE); | |
1004 | if (!ib_comp_wq) { | |
1005 | ret = -ENOMEM; | |
1006 | goto err; | |
1007 | } | |
1008 | ||
55aeed06 | 1009 | ret = class_register(&ib_class); |
fd75c789 | 1010 | if (ret) { |
aba25a3e | 1011 | pr_warn("Couldn't create InfiniBand device class\n"); |
14d3a3b2 | 1012 | goto err_comp; |
fd75c789 | 1013 | } |
1da177e4 | 1014 | |
b2cbae2c RD |
1015 | ret = ibnl_init(); |
1016 | if (ret) { | |
aba25a3e | 1017 | pr_warn("Couldn't init IB netlink interface\n"); |
b2cbae2c RD |
1018 | goto err_sysfs; |
1019 | } | |
1020 | ||
e3f20f02 LR |
1021 | ret = addr_init(); |
1022 | if (ret) { | |
1023 | pr_warn("Could't init IB address resolution\n"); | |
1024 | goto err_ibnl; | |
1025 | } | |
1026 | ||
4c2cb422 MB |
1027 | ret = ib_mad_init(); |
1028 | if (ret) { | |
1029 | pr_warn("Couldn't init IB MAD\n"); | |
1030 | goto err_addr; | |
1031 | } | |
1032 | ||
c2e49c92 MB |
1033 | ret = ib_sa_init(); |
1034 | if (ret) { | |
1035 | pr_warn("Couldn't init SA\n"); | |
1036 | goto err_mad; | |
1037 | } | |
1038 | ||
da1f857b DC |
1039 | ret = ib_add_ibnl_clients(); |
1040 | if (ret) { | |
735c631a MB |
1041 | pr_warn("Couldn't register ibnl clients\n"); |
1042 | goto err_sa; | |
1043 | } | |
1044 | ||
03db3a2d | 1045 | ib_cache_setup(); |
1da177e4 | 1046 | |
fd75c789 NM |
1047 | return 0; |
1048 | ||
735c631a MB |
1049 | err_sa: |
1050 | ib_sa_cleanup(); | |
c2e49c92 MB |
1051 | err_mad: |
1052 | ib_mad_cleanup(); | |
4c2cb422 MB |
1053 | err_addr: |
1054 | addr_cleanup(); | |
e3f20f02 LR |
1055 | err_ibnl: |
1056 | ibnl_cleanup(); | |
fd75c789 | 1057 | err_sysfs: |
55aeed06 | 1058 | class_unregister(&ib_class); |
14d3a3b2 CH |
1059 | err_comp: |
1060 | destroy_workqueue(ib_comp_wq); | |
fd75c789 NM |
1061 | err: |
1062 | destroy_workqueue(ib_wq); | |
1da177e4 LT |
1063 | return ret; |
1064 | } | |
1065 | ||
1066 | static void __exit ib_core_cleanup(void) | |
1067 | { | |
1068 | ib_cache_cleanup(); | |
735c631a | 1069 | ib_remove_ibnl_clients(); |
c2e49c92 | 1070 | ib_sa_cleanup(); |
4c2cb422 | 1071 | ib_mad_cleanup(); |
e3f20f02 | 1072 | addr_cleanup(); |
b2cbae2c | 1073 | ibnl_cleanup(); |
55aeed06 | 1074 | class_unregister(&ib_class); |
14d3a3b2 | 1075 | destroy_workqueue(ib_comp_wq); |
f7c6a7b5 | 1076 | /* Make sure that any pending umem accounting work is done. */ |
f0626710 | 1077 | destroy_workqueue(ib_wq); |
1da177e4 LT |
1078 | } |
1079 | ||
1080 | module_init(ib_core_init); | |
1081 | module_exit(ib_core_cleanup); |