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> | |
9268f72d | 40 | #include <linux/netdevice.h> |
4e0f7b90 PP |
41 | #include <net/net_namespace.h> |
42 | #include <net/netns/generic.h> | |
8f408ab6 DJ |
43 | #include <linux/security.h> |
44 | #include <linux/notifier.h> | |
324e227e | 45 | #include <linux/hashtable.h> |
b2cbae2c | 46 | #include <rdma/rdma_netlink.h> |
03db3a2d MB |
47 | #include <rdma/ib_addr.h> |
48 | #include <rdma/ib_cache.h> | |
1da177e4 LT |
49 | |
50 | #include "core_priv.h" | |
41eda65c | 51 | #include "restrack.h" |
1da177e4 LT |
52 | |
53 | MODULE_AUTHOR("Roland Dreier"); | |
54 | MODULE_DESCRIPTION("core kernel InfiniBand API"); | |
55 | MODULE_LICENSE("Dual BSD/GPL"); | |
56 | ||
14d3a3b2 | 57 | struct workqueue_struct *ib_comp_wq; |
f794809a | 58 | struct workqueue_struct *ib_comp_unbound_wq; |
f0626710 TH |
59 | struct workqueue_struct *ib_wq; |
60 | EXPORT_SYMBOL_GPL(ib_wq); | |
61 | ||
921eab11 JG |
62 | /* |
63 | * Each of the three rwsem locks (devices, clients, client_data) protects the | |
64 | * xarray of the same name. Specifically it allows the caller to assert that | |
65 | * the MARK will/will not be changing under the lock, and for devices and | |
66 | * clients, that the value in the xarray is still a valid pointer. Change of | |
67 | * the MARK is linked to the object state, so holding the lock and testing the | |
68 | * MARK also asserts that the contained object is in a certain state. | |
69 | * | |
70 | * This is used to build a two stage register/unregister flow where objects | |
71 | * can continue to be in the xarray even though they are still in progress to | |
72 | * register/unregister. | |
73 | * | |
74 | * The xarray itself provides additional locking, and restartable iteration, | |
75 | * which is also relied on. | |
76 | * | |
77 | * Locks should not be nested, with the exception of client_data, which is | |
78 | * allowed to nest under the read side of the other two locks. | |
79 | * | |
80 | * The devices_rwsem also protects the device name list, any change or | |
81 | * assignment of device name must also hold the write side to guarantee unique | |
82 | * names. | |
83 | */ | |
84 | ||
0df91bb6 JG |
85 | /* |
86 | * devices contains devices that have had their names assigned. The | |
87 | * devices may not be registered. Users that care about the registration | |
88 | * status need to call ib_device_try_get() on the device to ensure it is | |
89 | * registered, and keep it registered, for the required duration. | |
90 | * | |
91 | */ | |
92 | static DEFINE_XARRAY_FLAGS(devices, XA_FLAGS_ALLOC); | |
921eab11 | 93 | static DECLARE_RWSEM(devices_rwsem); |
0df91bb6 JG |
94 | #define DEVICE_REGISTERED XA_MARK_1 |
95 | ||
1da177e4 | 96 | static LIST_HEAD(client_list); |
e59178d8 JG |
97 | #define CLIENT_REGISTERED XA_MARK_1 |
98 | static DEFINE_XARRAY_FLAGS(clients, XA_FLAGS_ALLOC); | |
921eab11 | 99 | static DECLARE_RWSEM(clients_rwsem); |
1da177e4 LT |
100 | |
101 | /* | |
0df91bb6 JG |
102 | * If client_data is registered then the corresponding client must also still |
103 | * be registered. | |
104 | */ | |
105 | #define CLIENT_DATA_REGISTERED XA_MARK_1 | |
4e0f7b90 PP |
106 | |
107 | /** | |
108 | * struct rdma_dev_net - rdma net namespace metadata for a net | |
109 | * @net: Pointer to owner net namespace | |
110 | * @id: xarray id to identify the net namespace. | |
111 | */ | |
112 | struct rdma_dev_net { | |
113 | possible_net_t net; | |
114 | u32 id; | |
115 | }; | |
116 | ||
117 | static unsigned int rdma_dev_net_id; | |
118 | ||
119 | /* | |
120 | * A list of net namespaces is maintained in an xarray. This is necessary | |
121 | * because we can't get the locking right using the existing net ns list. We | |
122 | * would require a init_net callback after the list is updated. | |
123 | */ | |
124 | static DEFINE_XARRAY_FLAGS(rdma_nets, XA_FLAGS_ALLOC); | |
125 | /* | |
126 | * rwsem to protect accessing the rdma_nets xarray entries. | |
127 | */ | |
128 | static DECLARE_RWSEM(rdma_nets_rwsem); | |
129 | ||
cb7e0e13 | 130 | bool ib_devices_shared_netns = true; |
a56bc45b PP |
131 | module_param_named(netns_mode, ib_devices_shared_netns, bool, 0444); |
132 | MODULE_PARM_DESC(netns_mode, | |
133 | "Share device among net namespaces; default=1 (shared)"); | |
41c61401 PP |
134 | /** |
135 | * rdma_dev_access_netns() - Return whether a rdma device can be accessed | |
136 | * from a specified net namespace or not. | |
137 | * @device: Pointer to rdma device which needs to be checked | |
138 | * @net: Pointer to net namesapce for which access to be checked | |
139 | * | |
140 | * rdma_dev_access_netns() - Return whether a rdma device can be accessed | |
141 | * from a specified net namespace or not. When | |
142 | * rdma device is in shared mode, it ignores the | |
143 | * net namespace. When rdma device is exclusive | |
144 | * to a net namespace, rdma device net namespace is | |
145 | * checked against the specified one. | |
146 | */ | |
147 | bool rdma_dev_access_netns(const struct ib_device *dev, const struct net *net) | |
148 | { | |
149 | return (ib_devices_shared_netns || | |
150 | net_eq(read_pnet(&dev->coredev.rdma_net), net)); | |
151 | } | |
152 | EXPORT_SYMBOL(rdma_dev_access_netns); | |
153 | ||
0df91bb6 JG |
154 | /* |
155 | * xarray has this behavior where it won't iterate over NULL values stored in | |
156 | * allocated arrays. So we need our own iterator to see all values stored in | |
157 | * the array. This does the same thing as xa_for_each except that it also | |
158 | * returns NULL valued entries if the array is allocating. Simplified to only | |
159 | * work on simple xarrays. | |
160 | */ | |
161 | static void *xan_find_marked(struct xarray *xa, unsigned long *indexp, | |
162 | xa_mark_t filter) | |
163 | { | |
164 | XA_STATE(xas, xa, *indexp); | |
165 | void *entry; | |
166 | ||
167 | rcu_read_lock(); | |
168 | do { | |
169 | entry = xas_find_marked(&xas, ULONG_MAX, filter); | |
170 | if (xa_is_zero(entry)) | |
171 | break; | |
172 | } while (xas_retry(&xas, entry)); | |
173 | rcu_read_unlock(); | |
174 | ||
175 | if (entry) { | |
176 | *indexp = xas.xa_index; | |
177 | if (xa_is_zero(entry)) | |
178 | return NULL; | |
179 | return entry; | |
180 | } | |
181 | return XA_ERROR(-ENOENT); | |
182 | } | |
183 | #define xan_for_each_marked(xa, index, entry, filter) \ | |
184 | for (index = 0, entry = xan_find_marked(xa, &(index), filter); \ | |
185 | !xa_is_err(entry); \ | |
186 | (index)++, entry = xan_find_marked(xa, &(index), filter)) | |
187 | ||
324e227e JG |
188 | /* RCU hash table mapping netdevice pointers to struct ib_port_data */ |
189 | static DEFINE_SPINLOCK(ndev_hash_lock); | |
190 | static DECLARE_HASHTABLE(ndev_hash, 5); | |
191 | ||
c2261dd7 | 192 | static void free_netdevs(struct ib_device *ib_dev); |
d0899892 JG |
193 | static void ib_unregister_work(struct work_struct *work); |
194 | static void __ib_unregister_device(struct ib_device *device); | |
8f408ab6 DJ |
195 | static int ib_security_change(struct notifier_block *nb, unsigned long event, |
196 | void *lsm_data); | |
197 | static void ib_policy_change_task(struct work_struct *work); | |
198 | static DECLARE_WORK(ib_policy_change_work, ib_policy_change_task); | |
199 | ||
923abb9d GP |
200 | static void __ibdev_printk(const char *level, const struct ib_device *ibdev, |
201 | struct va_format *vaf) | |
202 | { | |
203 | if (ibdev && ibdev->dev.parent) | |
204 | dev_printk_emit(level[1] - '0', | |
205 | ibdev->dev.parent, | |
206 | "%s %s %s: %pV", | |
207 | dev_driver_string(ibdev->dev.parent), | |
208 | dev_name(ibdev->dev.parent), | |
209 | dev_name(&ibdev->dev), | |
210 | vaf); | |
211 | else if (ibdev) | |
212 | printk("%s%s: %pV", | |
213 | level, dev_name(&ibdev->dev), vaf); | |
214 | else | |
215 | printk("%s(NULL ib_device): %pV", level, vaf); | |
216 | } | |
217 | ||
218 | void ibdev_printk(const char *level, const struct ib_device *ibdev, | |
219 | const char *format, ...) | |
220 | { | |
221 | struct va_format vaf; | |
222 | va_list args; | |
223 | ||
224 | va_start(args, format); | |
225 | ||
226 | vaf.fmt = format; | |
227 | vaf.va = &args; | |
228 | ||
229 | __ibdev_printk(level, ibdev, &vaf); | |
230 | ||
231 | va_end(args); | |
232 | } | |
233 | EXPORT_SYMBOL(ibdev_printk); | |
234 | ||
235 | #define define_ibdev_printk_level(func, level) \ | |
236 | void func(const struct ib_device *ibdev, const char *fmt, ...) \ | |
237 | { \ | |
238 | struct va_format vaf; \ | |
239 | va_list args; \ | |
240 | \ | |
241 | va_start(args, fmt); \ | |
242 | \ | |
243 | vaf.fmt = fmt; \ | |
244 | vaf.va = &args; \ | |
245 | \ | |
246 | __ibdev_printk(level, ibdev, &vaf); \ | |
247 | \ | |
248 | va_end(args); \ | |
249 | } \ | |
250 | EXPORT_SYMBOL(func); | |
251 | ||
252 | define_ibdev_printk_level(ibdev_emerg, KERN_EMERG); | |
253 | define_ibdev_printk_level(ibdev_alert, KERN_ALERT); | |
254 | define_ibdev_printk_level(ibdev_crit, KERN_CRIT); | |
255 | define_ibdev_printk_level(ibdev_err, KERN_ERR); | |
256 | define_ibdev_printk_level(ibdev_warn, KERN_WARNING); | |
257 | define_ibdev_printk_level(ibdev_notice, KERN_NOTICE); | |
258 | define_ibdev_printk_level(ibdev_info, KERN_INFO); | |
259 | ||
8f408ab6 DJ |
260 | static struct notifier_block ibdev_lsm_nb = { |
261 | .notifier_call = ib_security_change, | |
262 | }; | |
1da177e4 | 263 | |
decbc7a6 PP |
264 | static int rdma_dev_change_netns(struct ib_device *device, struct net *cur_net, |
265 | struct net *net); | |
266 | ||
324e227e JG |
267 | /* Pointer to the RCU head at the start of the ib_port_data array */ |
268 | struct ib_port_data_rcu { | |
269 | struct rcu_head rcu_head; | |
270 | struct ib_port_data pdata[]; | |
271 | }; | |
272 | ||
1da177e4 LT |
273 | static int ib_device_check_mandatory(struct ib_device *device) |
274 | { | |
3023a1e9 | 275 | #define IB_MANDATORY_FUNC(x) { offsetof(struct ib_device_ops, x), #x } |
1da177e4 LT |
276 | static const struct { |
277 | size_t offset; | |
278 | char *name; | |
279 | } mandatory_table[] = { | |
280 | IB_MANDATORY_FUNC(query_device), | |
281 | IB_MANDATORY_FUNC(query_port), | |
282 | IB_MANDATORY_FUNC(query_pkey), | |
1da177e4 LT |
283 | IB_MANDATORY_FUNC(alloc_pd), |
284 | IB_MANDATORY_FUNC(dealloc_pd), | |
1da177e4 LT |
285 | IB_MANDATORY_FUNC(create_qp), |
286 | IB_MANDATORY_FUNC(modify_qp), | |
287 | IB_MANDATORY_FUNC(destroy_qp), | |
288 | IB_MANDATORY_FUNC(post_send), | |
289 | IB_MANDATORY_FUNC(post_recv), | |
290 | IB_MANDATORY_FUNC(create_cq), | |
291 | IB_MANDATORY_FUNC(destroy_cq), | |
292 | IB_MANDATORY_FUNC(poll_cq), | |
293 | IB_MANDATORY_FUNC(req_notify_cq), | |
294 | IB_MANDATORY_FUNC(get_dma_mr), | |
7738613e IW |
295 | IB_MANDATORY_FUNC(dereg_mr), |
296 | IB_MANDATORY_FUNC(get_port_immutable) | |
1da177e4 LT |
297 | }; |
298 | int i; | |
299 | ||
6780c4fa | 300 | device->kverbs_provider = true; |
9a6b090c | 301 | for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) { |
3023a1e9 KH |
302 | if (!*(void **) ((void *) &device->ops + |
303 | mandatory_table[i].offset)) { | |
6780c4fa GP |
304 | device->kverbs_provider = false; |
305 | break; | |
1da177e4 LT |
306 | } |
307 | } | |
308 | ||
309 | return 0; | |
310 | } | |
311 | ||
f8978bd9 | 312 | /* |
01b67117 PP |
313 | * Caller must perform ib_device_put() to return the device reference count |
314 | * when ib_device_get_by_index() returns valid device pointer. | |
f8978bd9 | 315 | */ |
37eeab55 | 316 | struct ib_device *ib_device_get_by_index(const struct net *net, u32 index) |
f8978bd9 LR |
317 | { |
318 | struct ib_device *device; | |
319 | ||
921eab11 | 320 | down_read(&devices_rwsem); |
0df91bb6 | 321 | device = xa_load(&devices, index); |
01b67117 | 322 | if (device) { |
37eeab55 PP |
323 | if (!rdma_dev_access_netns(device, net)) { |
324 | device = NULL; | |
325 | goto out; | |
326 | } | |
327 | ||
d79af724 | 328 | if (!ib_device_try_get(device)) |
01b67117 PP |
329 | device = NULL; |
330 | } | |
37eeab55 | 331 | out: |
921eab11 | 332 | up_read(&devices_rwsem); |
f8978bd9 LR |
333 | return device; |
334 | } | |
335 | ||
d79af724 JG |
336 | /** |
337 | * ib_device_put - Release IB device reference | |
338 | * @device: device whose reference to be released | |
339 | * | |
340 | * ib_device_put() releases reference to the IB device to allow it to be | |
341 | * unregistered and eventually free. | |
342 | */ | |
01b67117 PP |
343 | void ib_device_put(struct ib_device *device) |
344 | { | |
345 | if (refcount_dec_and_test(&device->refcount)) | |
346 | complete(&device->unreg_completion); | |
347 | } | |
d79af724 | 348 | EXPORT_SYMBOL(ib_device_put); |
01b67117 | 349 | |
1da177e4 LT |
350 | static struct ib_device *__ib_device_get_by_name(const char *name) |
351 | { | |
352 | struct ib_device *device; | |
0df91bb6 | 353 | unsigned long index; |
1da177e4 | 354 | |
0df91bb6 | 355 | xa_for_each (&devices, index, device) |
896de009 | 356 | if (!strcmp(name, dev_name(&device->dev))) |
1da177e4 LT |
357 | return device; |
358 | ||
359 | return NULL; | |
360 | } | |
361 | ||
6cc2c8e5 JG |
362 | /** |
363 | * ib_device_get_by_name - Find an IB device by name | |
364 | * @name: The name to look for | |
365 | * @driver_id: The driver ID that must match (RDMA_DRIVER_UNKNOWN matches all) | |
366 | * | |
367 | * Find and hold an ib_device by its name. The caller must call | |
368 | * ib_device_put() on the returned pointer. | |
369 | */ | |
370 | struct ib_device *ib_device_get_by_name(const char *name, | |
371 | enum rdma_driver_id driver_id) | |
372 | { | |
373 | struct ib_device *device; | |
374 | ||
375 | down_read(&devices_rwsem); | |
376 | device = __ib_device_get_by_name(name); | |
377 | if (device && driver_id != RDMA_DRIVER_UNKNOWN && | |
378 | device->driver_id != driver_id) | |
379 | device = NULL; | |
380 | ||
381 | if (device) { | |
382 | if (!ib_device_try_get(device)) | |
383 | device = NULL; | |
384 | } | |
385 | up_read(&devices_rwsem); | |
386 | return device; | |
387 | } | |
388 | EXPORT_SYMBOL(ib_device_get_by_name); | |
389 | ||
4e0f7b90 PP |
390 | static int rename_compat_devs(struct ib_device *device) |
391 | { | |
392 | struct ib_core_device *cdev; | |
393 | unsigned long index; | |
394 | int ret = 0; | |
395 | ||
396 | mutex_lock(&device->compat_devs_mutex); | |
397 | xa_for_each (&device->compat_devs, index, cdev) { | |
398 | ret = device_rename(&cdev->dev, dev_name(&device->dev)); | |
399 | if (ret) { | |
400 | dev_warn(&cdev->dev, | |
401 | "Fail to rename compatdev to new name %s\n", | |
402 | dev_name(&device->dev)); | |
403 | break; | |
404 | } | |
405 | } | |
406 | mutex_unlock(&device->compat_devs_mutex); | |
407 | return ret; | |
408 | } | |
409 | ||
d21943dd LR |
410 | int ib_device_rename(struct ib_device *ibdev, const char *name) |
411 | { | |
dc1435c0 LR |
412 | unsigned long index; |
413 | void *client_data; | |
e3593b56 | 414 | int ret; |
d21943dd | 415 | |
921eab11 | 416 | down_write(&devices_rwsem); |
e3593b56 | 417 | if (!strcmp(name, dev_name(&ibdev->dev))) { |
dc1435c0 LR |
418 | up_write(&devices_rwsem); |
419 | return 0; | |
e3593b56 JG |
420 | } |
421 | ||
344684e6 | 422 | if (__ib_device_get_by_name(name)) { |
dc1435c0 LR |
423 | up_write(&devices_rwsem); |
424 | return -EEXIST; | |
d21943dd LR |
425 | } |
426 | ||
427 | ret = device_rename(&ibdev->dev, name); | |
dc1435c0 LR |
428 | if (ret) { |
429 | up_write(&devices_rwsem); | |
430 | return ret; | |
431 | } | |
432 | ||
d21943dd | 433 | strlcpy(ibdev->name, name, IB_DEVICE_NAME_MAX); |
4e0f7b90 | 434 | ret = rename_compat_devs(ibdev); |
dc1435c0 LR |
435 | |
436 | downgrade_write(&devices_rwsem); | |
437 | down_read(&ibdev->client_data_rwsem); | |
438 | xan_for_each_marked(&ibdev->client_data, index, client_data, | |
439 | CLIENT_DATA_REGISTERED) { | |
440 | struct ib_client *client = xa_load(&clients, index); | |
441 | ||
442 | if (!client || !client->rename) | |
443 | continue; | |
444 | ||
445 | client->rename(ibdev, client_data); | |
446 | } | |
447 | up_read(&ibdev->client_data_rwsem); | |
448 | up_read(&devices_rwsem); | |
449 | return 0; | |
d21943dd LR |
450 | } |
451 | ||
e349f858 | 452 | static int alloc_name(struct ib_device *ibdev, const char *name) |
1da177e4 | 453 | { |
1da177e4 | 454 | struct ib_device *device; |
0df91bb6 | 455 | unsigned long index; |
3b88afd3 JG |
456 | struct ida inuse; |
457 | int rc; | |
1da177e4 LT |
458 | int i; |
459 | ||
9ffbe8ac | 460 | lockdep_assert_held_write(&devices_rwsem); |
3b88afd3 | 461 | ida_init(&inuse); |
0df91bb6 | 462 | xa_for_each (&devices, index, device) { |
e349f858 JG |
463 | char buf[IB_DEVICE_NAME_MAX]; |
464 | ||
896de009 | 465 | if (sscanf(dev_name(&device->dev), name, &i) != 1) |
1da177e4 | 466 | continue; |
3b88afd3 | 467 | if (i < 0 || i >= INT_MAX) |
1da177e4 LT |
468 | continue; |
469 | snprintf(buf, sizeof buf, name, i); | |
3b88afd3 JG |
470 | if (strcmp(buf, dev_name(&device->dev)) != 0) |
471 | continue; | |
472 | ||
473 | rc = ida_alloc_range(&inuse, i, i, GFP_KERNEL); | |
474 | if (rc < 0) | |
475 | goto out; | |
1da177e4 LT |
476 | } |
477 | ||
3b88afd3 JG |
478 | rc = ida_alloc(&inuse, GFP_KERNEL); |
479 | if (rc < 0) | |
480 | goto out; | |
1da177e4 | 481 | |
3b88afd3 JG |
482 | rc = dev_set_name(&ibdev->dev, name, rc); |
483 | out: | |
484 | ida_destroy(&inuse); | |
485 | return rc; | |
1da177e4 LT |
486 | } |
487 | ||
55aeed06 JG |
488 | static void ib_device_release(struct device *device) |
489 | { | |
490 | struct ib_device *dev = container_of(device, struct ib_device, dev); | |
491 | ||
c2261dd7 | 492 | free_netdevs(dev); |
652432f3 | 493 | WARN_ON(refcount_read(&dev->refcount)); |
46bdf370 KH |
494 | if (dev->port_data) { |
495 | ib_cache_release_one(dev); | |
496 | ib_security_release_port_pkey_list(dev); | |
324e227e JG |
497 | kfree_rcu(container_of(dev->port_data, struct ib_port_data_rcu, |
498 | pdata[0]), | |
499 | rcu_head); | |
46bdf370 KH |
500 | } |
501 | xa_destroy(&dev->compat_devs); | |
502 | xa_destroy(&dev->client_data); | |
324e227e | 503 | kfree_rcu(dev, rcu_head); |
55aeed06 JG |
504 | } |
505 | ||
506 | static int ib_device_uevent(struct device *device, | |
507 | struct kobj_uevent_env *env) | |
508 | { | |
896de009 | 509 | if (add_uevent_var(env, "NAME=%s", dev_name(device))) |
55aeed06 JG |
510 | return -ENOMEM; |
511 | ||
512 | /* | |
513 | * It would be nice to pass the node GUID with the event... | |
514 | */ | |
515 | ||
516 | return 0; | |
517 | } | |
518 | ||
62dfa795 PP |
519 | static const void *net_namespace(struct device *d) |
520 | { | |
4e0f7b90 PP |
521 | struct ib_core_device *coredev = |
522 | container_of(d, struct ib_core_device, dev); | |
523 | ||
524 | return read_pnet(&coredev->rdma_net); | |
62dfa795 PP |
525 | } |
526 | ||
55aeed06 JG |
527 | static struct class ib_class = { |
528 | .name = "infiniband", | |
529 | .dev_release = ib_device_release, | |
530 | .dev_uevent = ib_device_uevent, | |
62dfa795 PP |
531 | .ns_type = &net_ns_type_operations, |
532 | .namespace = net_namespace, | |
55aeed06 JG |
533 | }; |
534 | ||
cebe556b | 535 | static void rdma_init_coredev(struct ib_core_device *coredev, |
4e0f7b90 | 536 | struct ib_device *dev, struct net *net) |
cebe556b PP |
537 | { |
538 | /* This BUILD_BUG_ON is intended to catch layout change | |
539 | * of union of ib_core_device and device. | |
540 | * dev must be the first element as ib_core and providers | |
541 | * driver uses it. Adding anything in ib_core_device before | |
542 | * device will break this assumption. | |
543 | */ | |
544 | BUILD_BUG_ON(offsetof(struct ib_device, coredev.dev) != | |
545 | offsetof(struct ib_device, dev)); | |
546 | ||
547 | coredev->dev.class = &ib_class; | |
548 | coredev->dev.groups = dev->groups; | |
549 | device_initialize(&coredev->dev); | |
550 | coredev->owner = dev; | |
551 | INIT_LIST_HEAD(&coredev->port_list); | |
4e0f7b90 | 552 | write_pnet(&coredev->rdma_net, net); |
cebe556b PP |
553 | } |
554 | ||
1da177e4 | 555 | /** |
459cc69f | 556 | * _ib_alloc_device - allocate an IB device struct |
1da177e4 LT |
557 | * @size:size of structure to allocate |
558 | * | |
559 | * Low-level drivers should use ib_alloc_device() to allocate &struct | |
560 | * ib_device. @size is the size of the structure to be allocated, | |
561 | * including any private data used by the low-level driver. | |
562 | * ib_dealloc_device() must be used to free structures allocated with | |
563 | * ib_alloc_device(). | |
564 | */ | |
459cc69f | 565 | struct ib_device *_ib_alloc_device(size_t size) |
1da177e4 | 566 | { |
55aeed06 JG |
567 | struct ib_device *device; |
568 | ||
569 | if (WARN_ON(size < sizeof(struct ib_device))) | |
570 | return NULL; | |
571 | ||
572 | device = kzalloc(size, GFP_KERNEL); | |
573 | if (!device) | |
574 | return NULL; | |
575 | ||
41eda65c LR |
576 | if (rdma_restrack_init(device)) { |
577 | kfree(device); | |
578 | return NULL; | |
579 | } | |
02d8883f | 580 | |
5f8f5499 | 581 | device->groups[0] = &ib_dev_attr_group; |
4e0f7b90 | 582 | rdma_init_coredev(&device->coredev, device, &init_net); |
55aeed06 | 583 | |
55aeed06 JG |
584 | INIT_LIST_HEAD(&device->event_handler_list); |
585 | spin_lock_init(&device->event_handler_lock); | |
d0899892 | 586 | mutex_init(&device->unregistration_lock); |
0df91bb6 JG |
587 | /* |
588 | * client_data needs to be alloc because we don't want our mark to be | |
589 | * destroyed if the user stores NULL in the client data. | |
590 | */ | |
591 | xa_init_flags(&device->client_data, XA_FLAGS_ALLOC); | |
921eab11 | 592 | init_rwsem(&device->client_data_rwsem); |
4e0f7b90 PP |
593 | xa_init_flags(&device->compat_devs, XA_FLAGS_ALLOC); |
594 | mutex_init(&device->compat_devs_mutex); | |
01b67117 | 595 | init_completion(&device->unreg_completion); |
d0899892 | 596 | INIT_WORK(&device->unregistration_work, ib_unregister_work); |
1da177e4 | 597 | |
55aeed06 | 598 | return device; |
1da177e4 | 599 | } |
459cc69f | 600 | EXPORT_SYMBOL(_ib_alloc_device); |
1da177e4 LT |
601 | |
602 | /** | |
603 | * ib_dealloc_device - free an IB device struct | |
604 | * @device:structure to free | |
605 | * | |
606 | * Free a structure allocated with ib_alloc_device(). | |
607 | */ | |
608 | void ib_dealloc_device(struct ib_device *device) | |
609 | { | |
d0899892 JG |
610 | if (device->ops.dealloc_driver) |
611 | device->ops.dealloc_driver(device); | |
612 | ||
613 | /* | |
614 | * ib_unregister_driver() requires all devices to remain in the xarray | |
615 | * while their ops are callable. The last op we call is dealloc_driver | |
616 | * above. This is needed to create a fence on op callbacks prior to | |
617 | * allowing the driver module to unload. | |
618 | */ | |
619 | down_write(&devices_rwsem); | |
620 | if (xa_load(&devices, device->index) == device) | |
621 | xa_erase(&devices, device->index); | |
622 | up_write(&devices_rwsem); | |
623 | ||
c2261dd7 JG |
624 | /* Expedite releasing netdev references */ |
625 | free_netdevs(device); | |
626 | ||
4e0f7b90 | 627 | WARN_ON(!xa_empty(&device->compat_devs)); |
0df91bb6 | 628 | WARN_ON(!xa_empty(&device->client_data)); |
652432f3 | 629 | WARN_ON(refcount_read(&device->refcount)); |
0ad699c0 | 630 | rdma_restrack_clean(device); |
e155755e | 631 | /* Balances with device_initialize */ |
924b8900 | 632 | put_device(&device->dev); |
1da177e4 LT |
633 | } |
634 | EXPORT_SYMBOL(ib_dealloc_device); | |
635 | ||
921eab11 JG |
636 | /* |
637 | * add_client_context() and remove_client_context() must be safe against | |
638 | * parallel calls on the same device - registration/unregistration of both the | |
639 | * device and client can be occurring in parallel. | |
640 | * | |
641 | * The routines need to be a fence, any caller must not return until the add | |
642 | * or remove is fully completed. | |
643 | */ | |
644 | static int add_client_context(struct ib_device *device, | |
645 | struct ib_client *client) | |
1da177e4 | 646 | { |
921eab11 | 647 | int ret = 0; |
1da177e4 | 648 | |
6780c4fa | 649 | if (!device->kverbs_provider && !client->no_kverbs_req) |
921eab11 JG |
650 | return 0; |
651 | ||
652 | down_write(&device->client_data_rwsem); | |
653 | /* | |
654 | * Another caller to add_client_context got here first and has already | |
655 | * completely initialized context. | |
656 | */ | |
657 | if (xa_get_mark(&device->client_data, client->client_id, | |
658 | CLIENT_DATA_REGISTERED)) | |
659 | goto out; | |
660 | ||
661 | ret = xa_err(xa_store(&device->client_data, client->client_id, NULL, | |
662 | GFP_KERNEL)); | |
663 | if (ret) | |
664 | goto out; | |
665 | downgrade_write(&device->client_data_rwsem); | |
666 | if (client->add) | |
667 | client->add(device); | |
668 | ||
669 | /* Readers shall not see a client until add has been completed */ | |
670 | xa_set_mark(&device->client_data, client->client_id, | |
671 | CLIENT_DATA_REGISTERED); | |
672 | up_read(&device->client_data_rwsem); | |
673 | return 0; | |
674 | ||
675 | out: | |
676 | up_write(&device->client_data_rwsem); | |
677 | return ret; | |
678 | } | |
679 | ||
680 | static void remove_client_context(struct ib_device *device, | |
681 | unsigned int client_id) | |
682 | { | |
683 | struct ib_client *client; | |
684 | void *client_data; | |
6780c4fa | 685 | |
921eab11 JG |
686 | down_write(&device->client_data_rwsem); |
687 | if (!xa_get_mark(&device->client_data, client_id, | |
688 | CLIENT_DATA_REGISTERED)) { | |
689 | up_write(&device->client_data_rwsem); | |
690 | return; | |
691 | } | |
692 | client_data = xa_load(&device->client_data, client_id); | |
693 | xa_clear_mark(&device->client_data, client_id, CLIENT_DATA_REGISTERED); | |
694 | client = xa_load(&clients, client_id); | |
695 | downgrade_write(&device->client_data_rwsem); | |
1da177e4 | 696 | |
921eab11 JG |
697 | /* |
698 | * Notice we cannot be holding any exclusive locks when calling the | |
699 | * remove callback as the remove callback can recurse back into any | |
700 | * public functions in this module and thus try for any locks those | |
701 | * functions take. | |
702 | * | |
703 | * For this reason clients and drivers should not call the | |
704 | * unregistration functions will holdling any locks. | |
705 | * | |
706 | * It tempting to drop the client_data_rwsem too, but this is required | |
707 | * to ensure that unregister_client does not return until all clients | |
708 | * are completely unregistered, which is required to avoid module | |
709 | * unloading races. | |
710 | */ | |
711 | if (client->remove) | |
712 | client->remove(device, client_data); | |
713 | ||
714 | xa_erase(&device->client_data, client_id); | |
715 | up_read(&device->client_data_rwsem); | |
1da177e4 LT |
716 | } |
717 | ||
c2261dd7 | 718 | static int alloc_port_data(struct ib_device *device) |
5eb620c8 | 719 | { |
324e227e | 720 | struct ib_port_data_rcu *pdata_rcu; |
ea1075ed | 721 | unsigned int port; |
c2261dd7 JG |
722 | |
723 | if (device->port_data) | |
724 | return 0; | |
725 | ||
726 | /* This can only be called once the physical port range is defined */ | |
727 | if (WARN_ON(!device->phys_port_cnt)) | |
728 | return -EINVAL; | |
7738613e | 729 | |
8ceb1357 JG |
730 | /* |
731 | * device->port_data is indexed directly by the port number to make | |
7738613e IW |
732 | * access to this data as efficient as possible. |
733 | * | |
8ceb1357 JG |
734 | * Therefore port_data is declared as a 1 based array with potential |
735 | * empty slots at the beginning. | |
7738613e | 736 | */ |
324e227e JG |
737 | pdata_rcu = kzalloc(struct_size(pdata_rcu, pdata, |
738 | rdma_end_port(device) + 1), | |
739 | GFP_KERNEL); | |
740 | if (!pdata_rcu) | |
55aeed06 | 741 | return -ENOMEM; |
324e227e JG |
742 | /* |
743 | * The rcu_head is put in front of the port data array and the stored | |
744 | * pointer is adjusted since we never need to see that member until | |
745 | * kfree_rcu. | |
746 | */ | |
747 | device->port_data = pdata_rcu->pdata; | |
5eb620c8 | 748 | |
ea1075ed | 749 | rdma_for_each_port (device, port) { |
8ceb1357 JG |
750 | struct ib_port_data *pdata = &device->port_data[port]; |
751 | ||
324e227e | 752 | pdata->ib_dev = device; |
8ceb1357 JG |
753 | spin_lock_init(&pdata->pkey_list_lock); |
754 | INIT_LIST_HEAD(&pdata->pkey_list); | |
c2261dd7 | 755 | spin_lock_init(&pdata->netdev_lock); |
324e227e | 756 | INIT_HLIST_NODE(&pdata->ndev_hash_link); |
c2261dd7 JG |
757 | } |
758 | return 0; | |
759 | } | |
760 | ||
761 | static int verify_immutable(const struct ib_device *dev, u8 port) | |
762 | { | |
763 | return WARN_ON(!rdma_cap_ib_mad(dev, port) && | |
764 | rdma_max_mad_size(dev, port) != 0); | |
765 | } | |
766 | ||
767 | static int setup_port_data(struct ib_device *device) | |
768 | { | |
769 | unsigned int port; | |
770 | int ret; | |
771 | ||
772 | ret = alloc_port_data(device); | |
773 | if (ret) | |
774 | return ret; | |
775 | ||
776 | rdma_for_each_port (device, port) { | |
777 | struct ib_port_data *pdata = &device->port_data[port]; | |
8ceb1357 JG |
778 | |
779 | ret = device->ops.get_port_immutable(device, port, | |
780 | &pdata->immutable); | |
5eb620c8 | 781 | if (ret) |
55aeed06 | 782 | return ret; |
337877a4 | 783 | |
55aeed06 JG |
784 | if (verify_immutable(device, port)) |
785 | return -EINVAL; | |
5eb620c8 | 786 | } |
55aeed06 | 787 | return 0; |
5eb620c8 YE |
788 | } |
789 | ||
9abb0d1b | 790 | void ib_get_device_fw_str(struct ib_device *dev, char *str) |
5fa76c20 | 791 | { |
3023a1e9 KH |
792 | if (dev->ops.get_dev_fw_str) |
793 | dev->ops.get_dev_fw_str(dev, str); | |
5fa76c20 IW |
794 | else |
795 | str[0] = '\0'; | |
796 | } | |
797 | EXPORT_SYMBOL(ib_get_device_fw_str); | |
798 | ||
8f408ab6 DJ |
799 | static void ib_policy_change_task(struct work_struct *work) |
800 | { | |
801 | struct ib_device *dev; | |
0df91bb6 | 802 | unsigned long index; |
8f408ab6 | 803 | |
921eab11 | 804 | down_read(&devices_rwsem); |
0df91bb6 | 805 | xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { |
ea1075ed | 806 | unsigned int i; |
8f408ab6 | 807 | |
ea1075ed | 808 | rdma_for_each_port (dev, i) { |
8f408ab6 DJ |
809 | u64 sp; |
810 | int ret = ib_get_cached_subnet_prefix(dev, | |
811 | i, | |
812 | &sp); | |
813 | ||
814 | WARN_ONCE(ret, | |
815 | "ib_get_cached_subnet_prefix err: %d, this should never happen here\n", | |
816 | ret); | |
a750cfde DJ |
817 | if (!ret) |
818 | ib_security_cache_change(dev, i, sp); | |
8f408ab6 DJ |
819 | } |
820 | } | |
921eab11 | 821 | up_read(&devices_rwsem); |
8f408ab6 DJ |
822 | } |
823 | ||
824 | static int ib_security_change(struct notifier_block *nb, unsigned long event, | |
825 | void *lsm_data) | |
826 | { | |
827 | if (event != LSM_POLICY_CHANGE) | |
828 | return NOTIFY_DONE; | |
829 | ||
830 | schedule_work(&ib_policy_change_work); | |
c66f6741 | 831 | ib_mad_agent_security_change(); |
8f408ab6 DJ |
832 | |
833 | return NOTIFY_OK; | |
834 | } | |
835 | ||
4e0f7b90 PP |
836 | static void compatdev_release(struct device *dev) |
837 | { | |
838 | struct ib_core_device *cdev = | |
839 | container_of(dev, struct ib_core_device, dev); | |
840 | ||
841 | kfree(cdev); | |
842 | } | |
843 | ||
844 | static int add_one_compat_dev(struct ib_device *device, | |
845 | struct rdma_dev_net *rnet) | |
846 | { | |
847 | struct ib_core_device *cdev; | |
848 | int ret; | |
849 | ||
2b34c558 | 850 | lockdep_assert_held(&rdma_nets_rwsem); |
a56bc45b PP |
851 | if (!ib_devices_shared_netns) |
852 | return 0; | |
853 | ||
4e0f7b90 PP |
854 | /* |
855 | * Create and add compat device in all namespaces other than where it | |
856 | * is currently bound to. | |
857 | */ | |
858 | if (net_eq(read_pnet(&rnet->net), | |
859 | read_pnet(&device->coredev.rdma_net))) | |
860 | return 0; | |
861 | ||
862 | /* | |
863 | * The first of init_net() or ib_register_device() to take the | |
864 | * compat_devs_mutex wins and gets to add the device. Others will wait | |
865 | * for completion here. | |
866 | */ | |
867 | mutex_lock(&device->compat_devs_mutex); | |
868 | cdev = xa_load(&device->compat_devs, rnet->id); | |
869 | if (cdev) { | |
870 | ret = 0; | |
871 | goto done; | |
872 | } | |
873 | ret = xa_reserve(&device->compat_devs, rnet->id, GFP_KERNEL); | |
874 | if (ret) | |
875 | goto done; | |
876 | ||
877 | cdev = kzalloc(sizeof(*cdev), GFP_KERNEL); | |
878 | if (!cdev) { | |
879 | ret = -ENOMEM; | |
880 | goto cdev_err; | |
881 | } | |
882 | ||
883 | cdev->dev.parent = device->dev.parent; | |
884 | rdma_init_coredev(cdev, device, read_pnet(&rnet->net)); | |
885 | cdev->dev.release = compatdev_release; | |
886 | dev_set_name(&cdev->dev, "%s", dev_name(&device->dev)); | |
887 | ||
888 | ret = device_add(&cdev->dev); | |
889 | if (ret) | |
890 | goto add_err; | |
eb15c78b | 891 | ret = ib_setup_port_attrs(cdev); |
5417783e PP |
892 | if (ret) |
893 | goto port_err; | |
4e0f7b90 PP |
894 | |
895 | ret = xa_err(xa_store(&device->compat_devs, rnet->id, | |
896 | cdev, GFP_KERNEL)); | |
897 | if (ret) | |
898 | goto insert_err; | |
899 | ||
900 | mutex_unlock(&device->compat_devs_mutex); | |
901 | return 0; | |
902 | ||
903 | insert_err: | |
5417783e PP |
904 | ib_free_port_attrs(cdev); |
905 | port_err: | |
4e0f7b90 PP |
906 | device_del(&cdev->dev); |
907 | add_err: | |
908 | put_device(&cdev->dev); | |
909 | cdev_err: | |
910 | xa_release(&device->compat_devs, rnet->id); | |
911 | done: | |
912 | mutex_unlock(&device->compat_devs_mutex); | |
913 | return ret; | |
914 | } | |
915 | ||
916 | static void remove_one_compat_dev(struct ib_device *device, u32 id) | |
917 | { | |
918 | struct ib_core_device *cdev; | |
919 | ||
920 | mutex_lock(&device->compat_devs_mutex); | |
921 | cdev = xa_erase(&device->compat_devs, id); | |
922 | mutex_unlock(&device->compat_devs_mutex); | |
923 | if (cdev) { | |
5417783e | 924 | ib_free_port_attrs(cdev); |
4e0f7b90 PP |
925 | device_del(&cdev->dev); |
926 | put_device(&cdev->dev); | |
927 | } | |
928 | } | |
929 | ||
930 | static void remove_compat_devs(struct ib_device *device) | |
931 | { | |
932 | struct ib_core_device *cdev; | |
933 | unsigned long index; | |
934 | ||
935 | xa_for_each (&device->compat_devs, index, cdev) | |
936 | remove_one_compat_dev(device, index); | |
937 | } | |
938 | ||
939 | static int add_compat_devs(struct ib_device *device) | |
940 | { | |
941 | struct rdma_dev_net *rnet; | |
942 | unsigned long index; | |
943 | int ret = 0; | |
944 | ||
decbc7a6 PP |
945 | lockdep_assert_held(&devices_rwsem); |
946 | ||
4e0f7b90 PP |
947 | down_read(&rdma_nets_rwsem); |
948 | xa_for_each (&rdma_nets, index, rnet) { | |
949 | ret = add_one_compat_dev(device, rnet); | |
950 | if (ret) | |
951 | break; | |
952 | } | |
953 | up_read(&rdma_nets_rwsem); | |
954 | return ret; | |
955 | } | |
956 | ||
2b34c558 PP |
957 | static void remove_all_compat_devs(void) |
958 | { | |
959 | struct ib_compat_device *cdev; | |
960 | struct ib_device *dev; | |
961 | unsigned long index; | |
962 | ||
963 | down_read(&devices_rwsem); | |
964 | xa_for_each (&devices, index, dev) { | |
965 | unsigned long c_index = 0; | |
966 | ||
967 | /* Hold nets_rwsem so that any other thread modifying this | |
968 | * system param can sync with this thread. | |
969 | */ | |
970 | down_read(&rdma_nets_rwsem); | |
971 | xa_for_each (&dev->compat_devs, c_index, cdev) | |
972 | remove_one_compat_dev(dev, c_index); | |
973 | up_read(&rdma_nets_rwsem); | |
974 | } | |
975 | up_read(&devices_rwsem); | |
976 | } | |
977 | ||
978 | static int add_all_compat_devs(void) | |
979 | { | |
980 | struct rdma_dev_net *rnet; | |
981 | struct ib_device *dev; | |
982 | unsigned long index; | |
983 | int ret = 0; | |
984 | ||
985 | down_read(&devices_rwsem); | |
986 | xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { | |
987 | unsigned long net_index = 0; | |
988 | ||
989 | /* Hold nets_rwsem so that any other thread modifying this | |
990 | * system param can sync with this thread. | |
991 | */ | |
992 | down_read(&rdma_nets_rwsem); | |
993 | xa_for_each (&rdma_nets, net_index, rnet) { | |
994 | ret = add_one_compat_dev(dev, rnet); | |
995 | if (ret) | |
996 | break; | |
997 | } | |
998 | up_read(&rdma_nets_rwsem); | |
999 | } | |
1000 | up_read(&devices_rwsem); | |
1001 | if (ret) | |
1002 | remove_all_compat_devs(); | |
1003 | return ret; | |
1004 | } | |
1005 | ||
1006 | int rdma_compatdev_set(u8 enable) | |
1007 | { | |
1008 | struct rdma_dev_net *rnet; | |
1009 | unsigned long index; | |
1010 | int ret = 0; | |
1011 | ||
1012 | down_write(&rdma_nets_rwsem); | |
1013 | if (ib_devices_shared_netns == enable) { | |
1014 | up_write(&rdma_nets_rwsem); | |
1015 | return 0; | |
1016 | } | |
1017 | ||
1018 | /* enable/disable of compat devices is not supported | |
1019 | * when more than default init_net exists. | |
1020 | */ | |
1021 | xa_for_each (&rdma_nets, index, rnet) { | |
1022 | ret++; | |
1023 | break; | |
1024 | } | |
1025 | if (!ret) | |
1026 | ib_devices_shared_netns = enable; | |
1027 | up_write(&rdma_nets_rwsem); | |
1028 | if (ret) | |
1029 | return -EBUSY; | |
1030 | ||
1031 | if (enable) | |
1032 | ret = add_all_compat_devs(); | |
1033 | else | |
1034 | remove_all_compat_devs(); | |
1035 | return ret; | |
1036 | } | |
1037 | ||
4e0f7b90 PP |
1038 | static void rdma_dev_exit_net(struct net *net) |
1039 | { | |
1040 | struct rdma_dev_net *rnet = net_generic(net, rdma_dev_net_id); | |
1041 | struct ib_device *dev; | |
1042 | unsigned long index; | |
1043 | int ret; | |
1044 | ||
1045 | down_write(&rdma_nets_rwsem); | |
1046 | /* | |
1047 | * Prevent the ID from being re-used and hide the id from xa_for_each. | |
1048 | */ | |
1049 | ret = xa_err(xa_store(&rdma_nets, rnet->id, NULL, GFP_KERNEL)); | |
1050 | WARN_ON(ret); | |
1051 | up_write(&rdma_nets_rwsem); | |
1052 | ||
1053 | down_read(&devices_rwsem); | |
1054 | xa_for_each (&devices, index, dev) { | |
1055 | get_device(&dev->dev); | |
1056 | /* | |
1057 | * Release the devices_rwsem so that pontentially blocking | |
1058 | * device_del, doesn't hold the devices_rwsem for too long. | |
1059 | */ | |
1060 | up_read(&devices_rwsem); | |
1061 | ||
1062 | remove_one_compat_dev(dev, rnet->id); | |
1063 | ||
decbc7a6 PP |
1064 | /* |
1065 | * If the real device is in the NS then move it back to init. | |
1066 | */ | |
1067 | rdma_dev_change_netns(dev, net, &init_net); | |
1068 | ||
4e0f7b90 PP |
1069 | put_device(&dev->dev); |
1070 | down_read(&devices_rwsem); | |
1071 | } | |
1072 | up_read(&devices_rwsem); | |
1073 | ||
1074 | xa_erase(&rdma_nets, rnet->id); | |
1075 | } | |
1076 | ||
1077 | static __net_init int rdma_dev_init_net(struct net *net) | |
1078 | { | |
1079 | struct rdma_dev_net *rnet = net_generic(net, rdma_dev_net_id); | |
1080 | unsigned long index; | |
1081 | struct ib_device *dev; | |
1082 | int ret; | |
1083 | ||
1084 | /* No need to create any compat devices in default init_net. */ | |
1085 | if (net_eq(net, &init_net)) | |
1086 | return 0; | |
1087 | ||
1088 | write_pnet(&rnet->net, net); | |
1089 | ||
1090 | ret = xa_alloc(&rdma_nets, &rnet->id, rnet, xa_limit_32b, GFP_KERNEL); | |
1091 | if (ret) | |
1092 | return ret; | |
1093 | ||
1094 | down_read(&devices_rwsem); | |
1095 | xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { | |
2b34c558 PP |
1096 | /* Hold nets_rwsem so that netlink command cannot change |
1097 | * system configuration for device sharing mode. | |
1098 | */ | |
1099 | down_read(&rdma_nets_rwsem); | |
4e0f7b90 | 1100 | ret = add_one_compat_dev(dev, rnet); |
2b34c558 | 1101 | up_read(&rdma_nets_rwsem); |
4e0f7b90 PP |
1102 | if (ret) |
1103 | break; | |
1104 | } | |
1105 | up_read(&devices_rwsem); | |
1106 | ||
1107 | if (ret) | |
1108 | rdma_dev_exit_net(net); | |
1109 | ||
1110 | return ret; | |
1111 | } | |
1112 | ||
0df91bb6 | 1113 | /* |
d0899892 JG |
1114 | * Assign the unique string device name and the unique device index. This is |
1115 | * undone by ib_dealloc_device. | |
ecc82c53 | 1116 | */ |
0df91bb6 | 1117 | static int assign_name(struct ib_device *device, const char *name) |
ecc82c53 | 1118 | { |
0df91bb6 JG |
1119 | static u32 last_id; |
1120 | int ret; | |
ecc82c53 | 1121 | |
921eab11 | 1122 | down_write(&devices_rwsem); |
0df91bb6 JG |
1123 | /* Assign a unique name to the device */ |
1124 | if (strchr(name, '%')) | |
1125 | ret = alloc_name(device, name); | |
1126 | else | |
1127 | ret = dev_set_name(&device->dev, name); | |
1128 | if (ret) | |
1129 | goto out; | |
1130 | ||
1131 | if (__ib_device_get_by_name(dev_name(&device->dev))) { | |
1132 | ret = -ENFILE; | |
1133 | goto out; | |
1134 | } | |
1135 | strlcpy(device->name, dev_name(&device->dev), IB_DEVICE_NAME_MAX); | |
ecc82c53 | 1136 | |
ea295481 LT |
1137 | ret = xa_alloc_cyclic(&devices, &device->index, device, xa_limit_31b, |
1138 | &last_id, GFP_KERNEL); | |
1139 | if (ret > 0) | |
1140 | ret = 0; | |
921eab11 | 1141 | |
0df91bb6 | 1142 | out: |
921eab11 | 1143 | up_write(&devices_rwsem); |
0df91bb6 JG |
1144 | return ret; |
1145 | } | |
1146 | ||
548cb4fb | 1147 | static void setup_dma_device(struct ib_device *device) |
1da177e4 | 1148 | { |
99db9494 BVA |
1149 | struct device *parent = device->dev.parent; |
1150 | ||
0957c29f BVA |
1151 | WARN_ON_ONCE(device->dma_device); |
1152 | if (device->dev.dma_ops) { | |
1153 | /* | |
1154 | * The caller provided custom DMA operations. Copy the | |
1155 | * DMA-related fields that are used by e.g. dma_alloc_coherent() | |
1156 | * into device->dev. | |
1157 | */ | |
1158 | device->dma_device = &device->dev; | |
02ee9da3 BVA |
1159 | if (!device->dev.dma_mask) { |
1160 | if (parent) | |
1161 | device->dev.dma_mask = parent->dma_mask; | |
1162 | else | |
1163 | WARN_ON_ONCE(true); | |
1164 | } | |
1165 | if (!device->dev.coherent_dma_mask) { | |
1166 | if (parent) | |
1167 | device->dev.coherent_dma_mask = | |
1168 | parent->coherent_dma_mask; | |
1169 | else | |
1170 | WARN_ON_ONCE(true); | |
1171 | } | |
0957c29f BVA |
1172 | } else { |
1173 | /* | |
1174 | * The caller did not provide custom DMA operations. Use the | |
1175 | * DMA mapping operations of the parent device. | |
1176 | */ | |
02ee9da3 | 1177 | WARN_ON_ONCE(!parent); |
0957c29f BVA |
1178 | device->dma_device = parent; |
1179 | } | |
d10bcf94 SS |
1180 | /* Setup default max segment size for all IB devices */ |
1181 | dma_set_max_seg_size(device->dma_device, SZ_2G); | |
1182 | ||
548cb4fb | 1183 | } |
1da177e4 | 1184 | |
921eab11 JG |
1185 | /* |
1186 | * setup_device() allocates memory and sets up data that requires calling the | |
1187 | * device ops, this is the only reason these actions are not done during | |
1188 | * ib_alloc_device. It is undone by ib_dealloc_device(). | |
1189 | */ | |
548cb4fb PP |
1190 | static int setup_device(struct ib_device *device) |
1191 | { | |
1192 | struct ib_udata uhw = {.outlen = 0, .inlen = 0}; | |
1193 | int ret; | |
1da177e4 | 1194 | |
921eab11 JG |
1195 | setup_dma_device(device); |
1196 | ||
548cb4fb PP |
1197 | ret = ib_device_check_mandatory(device); |
1198 | if (ret) | |
1199 | return ret; | |
1da177e4 | 1200 | |
8ceb1357 | 1201 | ret = setup_port_data(device); |
5eb620c8 | 1202 | if (ret) { |
8ceb1357 | 1203 | dev_warn(&device->dev, "Couldn't create per-port data\n"); |
548cb4fb PP |
1204 | return ret; |
1205 | } | |
1206 | ||
1207 | memset(&device->attrs, 0, sizeof(device->attrs)); | |
3023a1e9 | 1208 | ret = device->ops.query_device(device, &device->attrs, &uhw); |
548cb4fb PP |
1209 | if (ret) { |
1210 | dev_warn(&device->dev, | |
1211 | "Couldn't query the device attributes\n"); | |
d45f89d5 | 1212 | return ret; |
5eb620c8 YE |
1213 | } |
1214 | ||
d45f89d5 | 1215 | return 0; |
548cb4fb PP |
1216 | } |
1217 | ||
921eab11 JG |
1218 | static void disable_device(struct ib_device *device) |
1219 | { | |
1220 | struct ib_client *client; | |
1221 | ||
1222 | WARN_ON(!refcount_read(&device->refcount)); | |
1223 | ||
1224 | down_write(&devices_rwsem); | |
1225 | xa_clear_mark(&devices, device->index, DEVICE_REGISTERED); | |
1226 | up_write(&devices_rwsem); | |
1227 | ||
1228 | down_read(&clients_rwsem); | |
1229 | list_for_each_entry_reverse(client, &client_list, list) | |
1230 | remove_client_context(device, client->client_id); | |
1231 | up_read(&clients_rwsem); | |
1232 | ||
1233 | /* Pairs with refcount_set in enable_device */ | |
1234 | ib_device_put(device); | |
1235 | wait_for_completion(&device->unreg_completion); | |
c2261dd7 | 1236 | |
4e0f7b90 PP |
1237 | /* |
1238 | * compat devices must be removed after device refcount drops to zero. | |
1239 | * Otherwise init_net() may add more compatdevs after removing compat | |
1240 | * devices and before device is disabled. | |
1241 | */ | |
1242 | remove_compat_devs(device); | |
921eab11 JG |
1243 | } |
1244 | ||
1245 | /* | |
1246 | * An enabled device is visible to all clients and to all the public facing | |
d0899892 JG |
1247 | * APIs that return a device pointer. This always returns with a new get, even |
1248 | * if it fails. | |
921eab11 | 1249 | */ |
d0899892 | 1250 | static int enable_device_and_get(struct ib_device *device) |
921eab11 JG |
1251 | { |
1252 | struct ib_client *client; | |
1253 | unsigned long index; | |
d0899892 | 1254 | int ret = 0; |
921eab11 | 1255 | |
d0899892 JG |
1256 | /* |
1257 | * One ref belongs to the xa and the other belongs to this | |
1258 | * thread. This is needed to guard against parallel unregistration. | |
1259 | */ | |
1260 | refcount_set(&device->refcount, 2); | |
921eab11 JG |
1261 | down_write(&devices_rwsem); |
1262 | xa_set_mark(&devices, device->index, DEVICE_REGISTERED); | |
d0899892 JG |
1263 | |
1264 | /* | |
1265 | * By using downgrade_write() we ensure that no other thread can clear | |
1266 | * DEVICE_REGISTERED while we are completing the client setup. | |
1267 | */ | |
1268 | downgrade_write(&devices_rwsem); | |
921eab11 | 1269 | |
ca22354b JG |
1270 | if (device->ops.enable_driver) { |
1271 | ret = device->ops.enable_driver(device); | |
1272 | if (ret) | |
1273 | goto out; | |
1274 | } | |
1275 | ||
921eab11 JG |
1276 | down_read(&clients_rwsem); |
1277 | xa_for_each_marked (&clients, index, client, CLIENT_REGISTERED) { | |
1278 | ret = add_client_context(device, client); | |
d0899892 JG |
1279 | if (ret) |
1280 | break; | |
921eab11 JG |
1281 | } |
1282 | up_read(&clients_rwsem); | |
4e0f7b90 PP |
1283 | if (!ret) |
1284 | ret = add_compat_devs(device); | |
ca22354b | 1285 | out: |
d0899892 JG |
1286 | up_read(&devices_rwsem); |
1287 | return ret; | |
921eab11 JG |
1288 | } |
1289 | ||
548cb4fb PP |
1290 | /** |
1291 | * ib_register_device - Register an IB device with IB core | |
1292 | * @device:Device to register | |
1293 | * | |
1294 | * Low-level drivers use ib_register_device() to register their | |
1295 | * devices with the IB core. All registered clients will receive a | |
1296 | * callback for each device that is added. @device must be allocated | |
1297 | * with ib_alloc_device(). | |
d0899892 JG |
1298 | * |
1299 | * If the driver uses ops.dealloc_driver and calls any ib_unregister_device() | |
1300 | * asynchronously then the device pointer may become freed as soon as this | |
1301 | * function returns. | |
548cb4fb | 1302 | */ |
ea4baf7f | 1303 | int ib_register_device(struct ib_device *device, const char *name) |
548cb4fb PP |
1304 | { |
1305 | int ret; | |
548cb4fb | 1306 | |
0df91bb6 JG |
1307 | ret = assign_name(device, name); |
1308 | if (ret) | |
921eab11 | 1309 | return ret; |
548cb4fb PP |
1310 | |
1311 | ret = setup_device(device); | |
1312 | if (ret) | |
d0899892 | 1313 | return ret; |
03db3a2d | 1314 | |
d45f89d5 JG |
1315 | ret = ib_cache_setup_one(device); |
1316 | if (ret) { | |
1317 | dev_warn(&device->dev, | |
1318 | "Couldn't set up InfiniBand P_Key/GID cache\n"); | |
d0899892 | 1319 | return ret; |
d45f89d5 JG |
1320 | } |
1321 | ||
7527a7b1 | 1322 | ib_device_register_rdmacg(device); |
3e153a93 | 1323 | |
e7a5b4aa LR |
1324 | /* |
1325 | * Ensure that ADD uevent is not fired because it | |
1326 | * is too early amd device is not initialized yet. | |
1327 | */ | |
1328 | dev_set_uevent_suppress(&device->dev, true); | |
5f8f5499 PP |
1329 | ret = device_add(&device->dev); |
1330 | if (ret) | |
1331 | goto cg_cleanup; | |
1332 | ||
ea4baf7f | 1333 | ret = ib_device_register_sysfs(device); |
1da177e4 | 1334 | if (ret) { |
43c7c851 JG |
1335 | dev_warn(&device->dev, |
1336 | "Couldn't register device with driver model\n"); | |
5f8f5499 | 1337 | goto dev_cleanup; |
1da177e4 LT |
1338 | } |
1339 | ||
d0899892 | 1340 | ret = enable_device_and_get(device); |
e7a5b4aa LR |
1341 | dev_set_uevent_suppress(&device->dev, false); |
1342 | /* Mark for userspace that device is ready */ | |
1343 | kobject_uevent(&device->dev.kobj, KOBJ_ADD); | |
d0899892 JG |
1344 | if (ret) { |
1345 | void (*dealloc_fn)(struct ib_device *); | |
1346 | ||
1347 | /* | |
1348 | * If we hit this error flow then we don't want to | |
1349 | * automatically dealloc the device since the caller is | |
1350 | * expected to call ib_dealloc_device() after | |
1351 | * ib_register_device() fails. This is tricky due to the | |
1352 | * possibility for a parallel unregistration along with this | |
1353 | * error flow. Since we have a refcount here we know any | |
1354 | * parallel flow is stopped in disable_device and will see the | |
1355 | * NULL pointers, causing the responsibility to | |
1356 | * ib_dealloc_device() to revert back to this thread. | |
1357 | */ | |
1358 | dealloc_fn = device->ops.dealloc_driver; | |
1359 | device->ops.dealloc_driver = NULL; | |
1360 | ib_device_put(device); | |
1361 | __ib_unregister_device(device); | |
1362 | device->ops.dealloc_driver = dealloc_fn; | |
1363 | return ret; | |
1364 | } | |
1365 | ib_device_put(device); | |
1da177e4 | 1366 | |
4be3a4fa PP |
1367 | return 0; |
1368 | ||
5f8f5499 PP |
1369 | dev_cleanup: |
1370 | device_del(&device->dev); | |
2fb4f4ea | 1371 | cg_cleanup: |
e7a5b4aa | 1372 | dev_set_uevent_suppress(&device->dev, false); |
2fb4f4ea | 1373 | ib_device_unregister_rdmacg(device); |
d45f89d5 | 1374 | ib_cache_cleanup_one(device); |
1da177e4 LT |
1375 | return ret; |
1376 | } | |
1377 | EXPORT_SYMBOL(ib_register_device); | |
1378 | ||
d0899892 JG |
1379 | /* Callers must hold a get on the device. */ |
1380 | static void __ib_unregister_device(struct ib_device *ib_dev) | |
1381 | { | |
1382 | /* | |
1383 | * We have a registration lock so that all the calls to unregister are | |
1384 | * fully fenced, once any unregister returns the device is truely | |
1385 | * unregistered even if multiple callers are unregistering it at the | |
1386 | * same time. This also interacts with the registration flow and | |
1387 | * provides sane semantics if register and unregister are racing. | |
1388 | */ | |
1389 | mutex_lock(&ib_dev->unregistration_lock); | |
1390 | if (!refcount_read(&ib_dev->refcount)) | |
1391 | goto out; | |
1392 | ||
1393 | disable_device(ib_dev); | |
3042492b PP |
1394 | |
1395 | /* Expedite removing unregistered pointers from the hash table */ | |
1396 | free_netdevs(ib_dev); | |
1397 | ||
d0899892 JG |
1398 | ib_device_unregister_sysfs(ib_dev); |
1399 | device_del(&ib_dev->dev); | |
1400 | ib_device_unregister_rdmacg(ib_dev); | |
1401 | ib_cache_cleanup_one(ib_dev); | |
1402 | ||
1403 | /* | |
1404 | * Drivers using the new flow may not call ib_dealloc_device except | |
1405 | * in error unwind prior to registration success. | |
1406 | */ | |
1407 | if (ib_dev->ops.dealloc_driver) { | |
1408 | WARN_ON(kref_read(&ib_dev->dev.kobj.kref) <= 1); | |
1409 | ib_dealloc_device(ib_dev); | |
1410 | } | |
1411 | out: | |
1412 | mutex_unlock(&ib_dev->unregistration_lock); | |
1413 | } | |
1414 | ||
1da177e4 LT |
1415 | /** |
1416 | * ib_unregister_device - Unregister an IB device | |
d0899892 | 1417 | * @device: The device to unregister |
1da177e4 LT |
1418 | * |
1419 | * Unregister an IB device. All clients will receive a remove callback. | |
d0899892 JG |
1420 | * |
1421 | * Callers should call this routine only once, and protect against races with | |
1422 | * registration. Typically it should only be called as part of a remove | |
1423 | * callback in an implementation of driver core's struct device_driver and | |
1424 | * related. | |
1425 | * | |
1426 | * If ops.dealloc_driver is used then ib_dev will be freed upon return from | |
1427 | * this function. | |
1da177e4 | 1428 | */ |
d0899892 | 1429 | void ib_unregister_device(struct ib_device *ib_dev) |
1da177e4 | 1430 | { |
d0899892 JG |
1431 | get_device(&ib_dev->dev); |
1432 | __ib_unregister_device(ib_dev); | |
1433 | put_device(&ib_dev->dev); | |
1da177e4 LT |
1434 | } |
1435 | EXPORT_SYMBOL(ib_unregister_device); | |
1436 | ||
d0899892 JG |
1437 | /** |
1438 | * ib_unregister_device_and_put - Unregister a device while holding a 'get' | |
1439 | * device: The device to unregister | |
1440 | * | |
1441 | * This is the same as ib_unregister_device(), except it includes an internal | |
1442 | * ib_device_put() that should match a 'get' obtained by the caller. | |
1443 | * | |
1444 | * It is safe to call this routine concurrently from multiple threads while | |
1445 | * holding the 'get'. When the function returns the device is fully | |
1446 | * unregistered. | |
1447 | * | |
1448 | * Drivers using this flow MUST use the driver_unregister callback to clean up | |
1449 | * their resources associated with the device and dealloc it. | |
1450 | */ | |
1451 | void ib_unregister_device_and_put(struct ib_device *ib_dev) | |
1452 | { | |
1453 | WARN_ON(!ib_dev->ops.dealloc_driver); | |
1454 | get_device(&ib_dev->dev); | |
1455 | ib_device_put(ib_dev); | |
1456 | __ib_unregister_device(ib_dev); | |
1457 | put_device(&ib_dev->dev); | |
1458 | } | |
1459 | EXPORT_SYMBOL(ib_unregister_device_and_put); | |
1460 | ||
1461 | /** | |
1462 | * ib_unregister_driver - Unregister all IB devices for a driver | |
1463 | * @driver_id: The driver to unregister | |
1464 | * | |
1465 | * This implements a fence for device unregistration. It only returns once all | |
1466 | * devices associated with the driver_id have fully completed their | |
1467 | * unregistration and returned from ib_unregister_device*(). | |
1468 | * | |
1469 | * If device's are not yet unregistered it goes ahead and starts unregistering | |
1470 | * them. | |
1471 | * | |
1472 | * This does not block creation of new devices with the given driver_id, that | |
1473 | * is the responsibility of the caller. | |
1474 | */ | |
1475 | void ib_unregister_driver(enum rdma_driver_id driver_id) | |
1476 | { | |
1477 | struct ib_device *ib_dev; | |
1478 | unsigned long index; | |
1479 | ||
1480 | down_read(&devices_rwsem); | |
1481 | xa_for_each (&devices, index, ib_dev) { | |
1482 | if (ib_dev->driver_id != driver_id) | |
1483 | continue; | |
1484 | ||
1485 | get_device(&ib_dev->dev); | |
1486 | up_read(&devices_rwsem); | |
1487 | ||
1488 | WARN_ON(!ib_dev->ops.dealloc_driver); | |
1489 | __ib_unregister_device(ib_dev); | |
1490 | ||
1491 | put_device(&ib_dev->dev); | |
1492 | down_read(&devices_rwsem); | |
1493 | } | |
1494 | up_read(&devices_rwsem); | |
1495 | } | |
1496 | EXPORT_SYMBOL(ib_unregister_driver); | |
1497 | ||
1498 | static void ib_unregister_work(struct work_struct *work) | |
1499 | { | |
1500 | struct ib_device *ib_dev = | |
1501 | container_of(work, struct ib_device, unregistration_work); | |
1502 | ||
1503 | __ib_unregister_device(ib_dev); | |
1504 | put_device(&ib_dev->dev); | |
1505 | } | |
1506 | ||
1507 | /** | |
1508 | * ib_unregister_device_queued - Unregister a device using a work queue | |
1509 | * device: The device to unregister | |
1510 | * | |
1511 | * This schedules an asynchronous unregistration using a WQ for the device. A | |
1512 | * driver should use this to avoid holding locks while doing unregistration, | |
1513 | * such as holding the RTNL lock. | |
1514 | * | |
1515 | * Drivers using this API must use ib_unregister_driver before module unload | |
1516 | * to ensure that all scheduled unregistrations have completed. | |
1517 | */ | |
1518 | void ib_unregister_device_queued(struct ib_device *ib_dev) | |
1519 | { | |
1520 | WARN_ON(!refcount_read(&ib_dev->refcount)); | |
1521 | WARN_ON(!ib_dev->ops.dealloc_driver); | |
1522 | get_device(&ib_dev->dev); | |
1523 | if (!queue_work(system_unbound_wq, &ib_dev->unregistration_work)) | |
1524 | put_device(&ib_dev->dev); | |
1525 | } | |
1526 | EXPORT_SYMBOL(ib_unregister_device_queued); | |
1527 | ||
decbc7a6 PP |
1528 | /* |
1529 | * The caller must pass in a device that has the kref held and the refcount | |
1530 | * released. If the device is in cur_net and still registered then it is moved | |
1531 | * into net. | |
1532 | */ | |
1533 | static int rdma_dev_change_netns(struct ib_device *device, struct net *cur_net, | |
1534 | struct net *net) | |
1535 | { | |
1536 | int ret2 = -EINVAL; | |
1537 | int ret; | |
1538 | ||
1539 | mutex_lock(&device->unregistration_lock); | |
1540 | ||
1541 | /* | |
2e5b8a01 PP |
1542 | * If a device not under ib_device_get() or if the unregistration_lock |
1543 | * is not held, the namespace can be changed, or it can be unregistered. | |
1544 | * Check again under the lock. | |
decbc7a6 PP |
1545 | */ |
1546 | if (refcount_read(&device->refcount) == 0 || | |
1547 | !net_eq(cur_net, read_pnet(&device->coredev.rdma_net))) { | |
1548 | ret = -ENODEV; | |
1549 | goto out; | |
1550 | } | |
1551 | ||
1552 | kobject_uevent(&device->dev.kobj, KOBJ_REMOVE); | |
1553 | disable_device(device); | |
1554 | ||
1555 | /* | |
1556 | * At this point no one can be using the device, so it is safe to | |
1557 | * change the namespace. | |
1558 | */ | |
1559 | write_pnet(&device->coredev.rdma_net, net); | |
1560 | ||
2e5b8a01 | 1561 | down_read(&devices_rwsem); |
decbc7a6 PP |
1562 | /* |
1563 | * Currently rdma devices are system wide unique. So the device name | |
1564 | * is guaranteed free in the new namespace. Publish the new namespace | |
1565 | * at the sysfs level. | |
1566 | */ | |
decbc7a6 PP |
1567 | ret = device_rename(&device->dev, dev_name(&device->dev)); |
1568 | up_read(&devices_rwsem); | |
1569 | if (ret) { | |
1570 | dev_warn(&device->dev, | |
1571 | "%s: Couldn't rename device after namespace change\n", | |
1572 | __func__); | |
1573 | /* Try and put things back and re-enable the device */ | |
1574 | write_pnet(&device->coredev.rdma_net, cur_net); | |
1575 | } | |
1576 | ||
1577 | ret2 = enable_device_and_get(device); | |
2e5b8a01 | 1578 | if (ret2) { |
decbc7a6 PP |
1579 | /* |
1580 | * This shouldn't really happen, but if it does, let the user | |
1581 | * retry at later point. So don't disable the device. | |
1582 | */ | |
1583 | dev_warn(&device->dev, | |
1584 | "%s: Couldn't re-enable device after namespace change\n", | |
1585 | __func__); | |
2e5b8a01 | 1586 | } |
decbc7a6 | 1587 | kobject_uevent(&device->dev.kobj, KOBJ_ADD); |
2e5b8a01 | 1588 | |
decbc7a6 PP |
1589 | ib_device_put(device); |
1590 | out: | |
1591 | mutex_unlock(&device->unregistration_lock); | |
1592 | if (ret) | |
1593 | return ret; | |
1594 | return ret2; | |
1595 | } | |
1596 | ||
2e5b8a01 PP |
1597 | int ib_device_set_netns_put(struct sk_buff *skb, |
1598 | struct ib_device *dev, u32 ns_fd) | |
1599 | { | |
1600 | struct net *net; | |
1601 | int ret; | |
1602 | ||
1603 | net = get_net_ns_by_fd(ns_fd); | |
1604 | if (IS_ERR(net)) { | |
1605 | ret = PTR_ERR(net); | |
1606 | goto net_err; | |
1607 | } | |
1608 | ||
1609 | if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) { | |
1610 | ret = -EPERM; | |
1611 | goto ns_err; | |
1612 | } | |
1613 | ||
1614 | /* | |
1615 | * Currently supported only for those providers which support | |
1616 | * disassociation and don't do port specific sysfs init. Once a | |
1617 | * port_cleanup infrastructure is implemented, this limitation will be | |
1618 | * removed. | |
1619 | */ | |
1620 | if (!dev->ops.disassociate_ucontext || dev->ops.init_port || | |
1621 | ib_devices_shared_netns) { | |
1622 | ret = -EOPNOTSUPP; | |
1623 | goto ns_err; | |
1624 | } | |
1625 | ||
1626 | get_device(&dev->dev); | |
1627 | ib_device_put(dev); | |
1628 | ret = rdma_dev_change_netns(dev, current->nsproxy->net_ns, net); | |
1629 | put_device(&dev->dev); | |
1630 | ||
1631 | put_net(net); | |
1632 | return ret; | |
1633 | ||
1634 | ns_err: | |
1635 | put_net(net); | |
1636 | net_err: | |
1637 | ib_device_put(dev); | |
1638 | return ret; | |
1639 | } | |
1640 | ||
4e0f7b90 PP |
1641 | static struct pernet_operations rdma_dev_net_ops = { |
1642 | .init = rdma_dev_init_net, | |
1643 | .exit = rdma_dev_exit_net, | |
1644 | .id = &rdma_dev_net_id, | |
1645 | .size = sizeof(struct rdma_dev_net), | |
1646 | }; | |
1647 | ||
e59178d8 JG |
1648 | static int assign_client_id(struct ib_client *client) |
1649 | { | |
1650 | int ret; | |
1651 | ||
921eab11 | 1652 | down_write(&clients_rwsem); |
e59178d8 JG |
1653 | /* |
1654 | * The add/remove callbacks must be called in FIFO/LIFO order. To | |
1655 | * achieve this we assign client_ids so they are sorted in | |
1656 | * registration order, and retain a linked list we can reverse iterate | |
1657 | * to get the LIFO order. The extra linked list can go away if xarray | |
1658 | * learns to reverse iterate. | |
1659 | */ | |
ea295481 | 1660 | if (list_empty(&client_list)) { |
e59178d8 | 1661 | client->client_id = 0; |
ea295481 LT |
1662 | } else { |
1663 | struct ib_client *last; | |
1664 | ||
1665 | last = list_last_entry(&client_list, struct ib_client, list); | |
1666 | client->client_id = last->client_id + 1; | |
4512acd0 | 1667 | } |
ea295481 | 1668 | ret = xa_insert(&clients, client->client_id, client, GFP_KERNEL); |
e59178d8 JG |
1669 | if (ret) |
1670 | goto out; | |
1671 | ||
921eab11 JG |
1672 | xa_set_mark(&clients, client->client_id, CLIENT_REGISTERED); |
1673 | list_add_tail(&client->list, &client_list); | |
1674 | ||
e59178d8 | 1675 | out: |
921eab11 | 1676 | up_write(&clients_rwsem); |
e59178d8 JG |
1677 | return ret; |
1678 | } | |
1679 | ||
1da177e4 LT |
1680 | /** |
1681 | * ib_register_client - Register an IB client | |
1682 | * @client:Client to register | |
1683 | * | |
1684 | * Upper level users of the IB drivers can use ib_register_client() to | |
1685 | * register callbacks for IB device addition and removal. When an IB | |
1686 | * device is added, each registered client's add method will be called | |
1687 | * (in the order the clients were registered), and when a device is | |
1688 | * removed, each client's remove method will be called (in the reverse | |
1689 | * order that clients were registered). In addition, when | |
1690 | * ib_register_client() is called, the client will receive an add | |
1691 | * callback for all devices already registered. | |
1692 | */ | |
1693 | int ib_register_client(struct ib_client *client) | |
1694 | { | |
1695 | struct ib_device *device; | |
0df91bb6 | 1696 | unsigned long index; |
e59178d8 | 1697 | int ret; |
1da177e4 | 1698 | |
e59178d8 | 1699 | ret = assign_client_id(client); |
921eab11 | 1700 | if (ret) |
e59178d8 | 1701 | return ret; |
1da177e4 | 1702 | |
921eab11 JG |
1703 | down_read(&devices_rwsem); |
1704 | xa_for_each_marked (&devices, index, device, DEVICE_REGISTERED) { | |
1705 | ret = add_client_context(device, client); | |
1706 | if (ret) { | |
1707 | up_read(&devices_rwsem); | |
1708 | ib_unregister_client(client); | |
1709 | return ret; | |
1710 | } | |
1711 | } | |
1712 | up_read(&devices_rwsem); | |
1da177e4 LT |
1713 | return 0; |
1714 | } | |
1715 | EXPORT_SYMBOL(ib_register_client); | |
1716 | ||
1717 | /** | |
1718 | * ib_unregister_client - Unregister an IB client | |
1719 | * @client:Client to unregister | |
1720 | * | |
1721 | * Upper level users use ib_unregister_client() to remove their client | |
1722 | * registration. When ib_unregister_client() is called, the client | |
1723 | * will receive a remove callback for each IB device still registered. | |
921eab11 JG |
1724 | * |
1725 | * This is a full fence, once it returns no client callbacks will be called, | |
1726 | * or are running in another thread. | |
1da177e4 LT |
1727 | */ |
1728 | void ib_unregister_client(struct ib_client *client) | |
1729 | { | |
1da177e4 | 1730 | struct ib_device *device; |
0df91bb6 | 1731 | unsigned long index; |
1da177e4 | 1732 | |
921eab11 | 1733 | down_write(&clients_rwsem); |
e59178d8 | 1734 | xa_clear_mark(&clients, client->client_id, CLIENT_REGISTERED); |
921eab11 JG |
1735 | up_write(&clients_rwsem); |
1736 | /* | |
1737 | * Every device still known must be serialized to make sure we are | |
1738 | * done with the client callbacks before we return. | |
1739 | */ | |
1740 | down_read(&devices_rwsem); | |
1741 | xa_for_each (&devices, index, device) | |
1742 | remove_client_context(device, client->client_id); | |
1743 | up_read(&devices_rwsem); | |
1da177e4 | 1744 | |
921eab11 | 1745 | down_write(&clients_rwsem); |
e59178d8 JG |
1746 | list_del(&client->list); |
1747 | xa_erase(&clients, client->client_id); | |
921eab11 | 1748 | up_write(&clients_rwsem); |
1da177e4 LT |
1749 | } |
1750 | EXPORT_SYMBOL(ib_unregister_client); | |
1751 | ||
1da177e4 | 1752 | /** |
9cd330d3 | 1753 | * ib_set_client_data - Set IB client context |
1da177e4 LT |
1754 | * @device:Device to set context for |
1755 | * @client:Client to set context for | |
1756 | * @data:Context to set | |
1757 | * | |
0df91bb6 JG |
1758 | * ib_set_client_data() sets client context data that can be retrieved with |
1759 | * ib_get_client_data(). This can only be called while the client is | |
1760 | * registered to the device, once the ib_client remove() callback returns this | |
1761 | * cannot be called. | |
1da177e4 LT |
1762 | */ |
1763 | void ib_set_client_data(struct ib_device *device, struct ib_client *client, | |
1764 | void *data) | |
1765 | { | |
0df91bb6 | 1766 | void *rc; |
1da177e4 | 1767 | |
0df91bb6 JG |
1768 | if (WARN_ON(IS_ERR(data))) |
1769 | data = NULL; | |
1da177e4 | 1770 | |
0df91bb6 JG |
1771 | rc = xa_store(&device->client_data, client->client_id, data, |
1772 | GFP_KERNEL); | |
1773 | WARN_ON(xa_is_err(rc)); | |
1da177e4 LT |
1774 | } |
1775 | EXPORT_SYMBOL(ib_set_client_data); | |
1776 | ||
1777 | /** | |
1778 | * ib_register_event_handler - Register an IB event handler | |
1779 | * @event_handler:Handler to register | |
1780 | * | |
1781 | * ib_register_event_handler() registers an event handler that will be | |
1782 | * called back when asynchronous IB events occur (as defined in | |
1783 | * chapter 11 of the InfiniBand Architecture Specification). This | |
1784 | * callback may occur in interrupt context. | |
1785 | */ | |
dcc9881e | 1786 | void ib_register_event_handler(struct ib_event_handler *event_handler) |
1da177e4 LT |
1787 | { |
1788 | unsigned long flags; | |
1789 | ||
1790 | spin_lock_irqsave(&event_handler->device->event_handler_lock, flags); | |
1791 | list_add_tail(&event_handler->list, | |
1792 | &event_handler->device->event_handler_list); | |
1793 | spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags); | |
1da177e4 LT |
1794 | } |
1795 | EXPORT_SYMBOL(ib_register_event_handler); | |
1796 | ||
1797 | /** | |
1798 | * ib_unregister_event_handler - Unregister an event handler | |
1799 | * @event_handler:Handler to unregister | |
1800 | * | |
1801 | * Unregister an event handler registered with | |
1802 | * ib_register_event_handler(). | |
1803 | */ | |
dcc9881e | 1804 | void ib_unregister_event_handler(struct ib_event_handler *event_handler) |
1da177e4 LT |
1805 | { |
1806 | unsigned long flags; | |
1807 | ||
1808 | spin_lock_irqsave(&event_handler->device->event_handler_lock, flags); | |
1809 | list_del(&event_handler->list); | |
1810 | spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags); | |
1da177e4 LT |
1811 | } |
1812 | EXPORT_SYMBOL(ib_unregister_event_handler); | |
1813 | ||
1814 | /** | |
1815 | * ib_dispatch_event - Dispatch an asynchronous event | |
1816 | * @event:Event to dispatch | |
1817 | * | |
1818 | * Low-level drivers must call ib_dispatch_event() to dispatch the | |
1819 | * event to all registered event handlers when an asynchronous event | |
1820 | * occurs. | |
1821 | */ | |
1822 | void ib_dispatch_event(struct ib_event *event) | |
1823 | { | |
1824 | unsigned long flags; | |
1825 | struct ib_event_handler *handler; | |
1826 | ||
1827 | spin_lock_irqsave(&event->device->event_handler_lock, flags); | |
1828 | ||
1829 | list_for_each_entry(handler, &event->device->event_handler_list, list) | |
1830 | handler->handler(handler, event); | |
1831 | ||
1832 | spin_unlock_irqrestore(&event->device->event_handler_lock, flags); | |
1833 | } | |
1834 | EXPORT_SYMBOL(ib_dispatch_event); | |
1835 | ||
1da177e4 LT |
1836 | /** |
1837 | * ib_query_port - Query IB port attributes | |
1838 | * @device:Device to query | |
1839 | * @port_num:Port number to query | |
1840 | * @port_attr:Port attributes | |
1841 | * | |
1842 | * ib_query_port() returns the attributes of a port through the | |
1843 | * @port_attr pointer. | |
1844 | */ | |
1845 | int ib_query_port(struct ib_device *device, | |
1846 | u8 port_num, | |
1847 | struct ib_port_attr *port_attr) | |
1848 | { | |
fad61ad4 EC |
1849 | union ib_gid gid; |
1850 | int err; | |
1851 | ||
24dc831b | 1852 | if (!rdma_is_port_valid(device, port_num)) |
116c0074 RD |
1853 | return -EINVAL; |
1854 | ||
fad61ad4 | 1855 | memset(port_attr, 0, sizeof(*port_attr)); |
3023a1e9 | 1856 | err = device->ops.query_port(device, port_num, port_attr); |
fad61ad4 EC |
1857 | if (err || port_attr->subnet_prefix) |
1858 | return err; | |
1859 | ||
d7012467 EC |
1860 | if (rdma_port_get_link_layer(device, port_num) != IB_LINK_LAYER_INFINIBAND) |
1861 | return 0; | |
1862 | ||
3023a1e9 | 1863 | err = device->ops.query_gid(device, port_num, 0, &gid); |
fad61ad4 EC |
1864 | if (err) |
1865 | return err; | |
1866 | ||
1867 | port_attr->subnet_prefix = be64_to_cpu(gid.global.subnet_prefix); | |
1868 | return 0; | |
1da177e4 LT |
1869 | } |
1870 | EXPORT_SYMBOL(ib_query_port); | |
1871 | ||
324e227e JG |
1872 | static void add_ndev_hash(struct ib_port_data *pdata) |
1873 | { | |
1874 | unsigned long flags; | |
1875 | ||
1876 | might_sleep(); | |
1877 | ||
1878 | spin_lock_irqsave(&ndev_hash_lock, flags); | |
1879 | if (hash_hashed(&pdata->ndev_hash_link)) { | |
1880 | hash_del_rcu(&pdata->ndev_hash_link); | |
1881 | spin_unlock_irqrestore(&ndev_hash_lock, flags); | |
1882 | /* | |
1883 | * We cannot do hash_add_rcu after a hash_del_rcu until the | |
1884 | * grace period | |
1885 | */ | |
1886 | synchronize_rcu(); | |
1887 | spin_lock_irqsave(&ndev_hash_lock, flags); | |
1888 | } | |
1889 | if (pdata->netdev) | |
1890 | hash_add_rcu(ndev_hash, &pdata->ndev_hash_link, | |
1891 | (uintptr_t)pdata->netdev); | |
1892 | spin_unlock_irqrestore(&ndev_hash_lock, flags); | |
1893 | } | |
1894 | ||
c2261dd7 JG |
1895 | /** |
1896 | * ib_device_set_netdev - Associate the ib_dev with an underlying net_device | |
1897 | * @ib_dev: Device to modify | |
1898 | * @ndev: net_device to affiliate, may be NULL | |
1899 | * @port: IB port the net_device is connected to | |
1900 | * | |
1901 | * Drivers should use this to link the ib_device to a netdev so the netdev | |
1902 | * shows up in interfaces like ib_enum_roce_netdev. Only one netdev may be | |
1903 | * affiliated with any port. | |
1904 | * | |
1905 | * The caller must ensure that the given ndev is not unregistered or | |
1906 | * unregistering, and that either the ib_device is unregistered or | |
1907 | * ib_device_set_netdev() is called with NULL when the ndev sends a | |
1908 | * NETDEV_UNREGISTER event. | |
1909 | */ | |
1910 | int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev, | |
1911 | unsigned int port) | |
1912 | { | |
1913 | struct net_device *old_ndev; | |
1914 | struct ib_port_data *pdata; | |
1915 | unsigned long flags; | |
1916 | int ret; | |
1917 | ||
1918 | /* | |
1919 | * Drivers wish to call this before ib_register_driver, so we have to | |
1920 | * setup the port data early. | |
1921 | */ | |
1922 | ret = alloc_port_data(ib_dev); | |
1923 | if (ret) | |
1924 | return ret; | |
1925 | ||
1926 | if (!rdma_is_port_valid(ib_dev, port)) | |
1927 | return -EINVAL; | |
1928 | ||
1929 | pdata = &ib_dev->port_data[port]; | |
1930 | spin_lock_irqsave(&pdata->netdev_lock, flags); | |
324e227e JG |
1931 | old_ndev = rcu_dereference_protected( |
1932 | pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); | |
1933 | if (old_ndev == ndev) { | |
c2261dd7 JG |
1934 | spin_unlock_irqrestore(&pdata->netdev_lock, flags); |
1935 | return 0; | |
1936 | } | |
c2261dd7 JG |
1937 | |
1938 | if (ndev) | |
1939 | dev_hold(ndev); | |
324e227e | 1940 | rcu_assign_pointer(pdata->netdev, ndev); |
c2261dd7 JG |
1941 | spin_unlock_irqrestore(&pdata->netdev_lock, flags); |
1942 | ||
324e227e | 1943 | add_ndev_hash(pdata); |
c2261dd7 JG |
1944 | if (old_ndev) |
1945 | dev_put(old_ndev); | |
1946 | ||
1947 | return 0; | |
1948 | } | |
1949 | EXPORT_SYMBOL(ib_device_set_netdev); | |
1950 | ||
1951 | static void free_netdevs(struct ib_device *ib_dev) | |
1952 | { | |
1953 | unsigned long flags; | |
1954 | unsigned int port; | |
1955 | ||
46bdf370 KH |
1956 | if (!ib_dev->port_data) |
1957 | return; | |
1958 | ||
c2261dd7 JG |
1959 | rdma_for_each_port (ib_dev, port) { |
1960 | struct ib_port_data *pdata = &ib_dev->port_data[port]; | |
324e227e | 1961 | struct net_device *ndev; |
c2261dd7 JG |
1962 | |
1963 | spin_lock_irqsave(&pdata->netdev_lock, flags); | |
324e227e JG |
1964 | ndev = rcu_dereference_protected( |
1965 | pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); | |
1966 | if (ndev) { | |
1967 | spin_lock(&ndev_hash_lock); | |
1968 | hash_del_rcu(&pdata->ndev_hash_link); | |
1969 | spin_unlock(&ndev_hash_lock); | |
1970 | ||
1971 | /* | |
1972 | * If this is the last dev_put there is still a | |
1973 | * synchronize_rcu before the netdev is kfreed, so we | |
1974 | * can continue to rely on unlocked pointer | |
1975 | * comparisons after the put | |
1976 | */ | |
1977 | rcu_assign_pointer(pdata->netdev, NULL); | |
1978 | dev_put(ndev); | |
c2261dd7 JG |
1979 | } |
1980 | spin_unlock_irqrestore(&pdata->netdev_lock, flags); | |
1981 | } | |
1982 | } | |
1983 | ||
1984 | struct net_device *ib_device_get_netdev(struct ib_device *ib_dev, | |
1985 | unsigned int port) | |
1986 | { | |
1987 | struct ib_port_data *pdata; | |
1988 | struct net_device *res; | |
1989 | ||
1990 | if (!rdma_is_port_valid(ib_dev, port)) | |
1991 | return NULL; | |
1992 | ||
1993 | pdata = &ib_dev->port_data[port]; | |
1994 | ||
1995 | /* | |
1996 | * New drivers should use ib_device_set_netdev() not the legacy | |
1997 | * get_netdev(). | |
1998 | */ | |
1999 | if (ib_dev->ops.get_netdev) | |
2000 | res = ib_dev->ops.get_netdev(ib_dev, port); | |
2001 | else { | |
2002 | spin_lock(&pdata->netdev_lock); | |
324e227e JG |
2003 | res = rcu_dereference_protected( |
2004 | pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); | |
c2261dd7 JG |
2005 | if (res) |
2006 | dev_hold(res); | |
2007 | spin_unlock(&pdata->netdev_lock); | |
2008 | } | |
2009 | ||
2010 | /* | |
2011 | * If we are starting to unregister expedite things by preventing | |
2012 | * propagation of an unregistering netdev. | |
2013 | */ | |
2014 | if (res && res->reg_state != NETREG_REGISTERED) { | |
2015 | dev_put(res); | |
2016 | return NULL; | |
2017 | } | |
2018 | ||
2019 | return res; | |
2020 | } | |
2021 | ||
324e227e JG |
2022 | /** |
2023 | * ib_device_get_by_netdev - Find an IB device associated with a netdev | |
2024 | * @ndev: netdev to locate | |
2025 | * @driver_id: The driver ID that must match (RDMA_DRIVER_UNKNOWN matches all) | |
2026 | * | |
2027 | * Find and hold an ib_device that is associated with a netdev via | |
2028 | * ib_device_set_netdev(). The caller must call ib_device_put() on the | |
2029 | * returned pointer. | |
2030 | */ | |
2031 | struct ib_device *ib_device_get_by_netdev(struct net_device *ndev, | |
2032 | enum rdma_driver_id driver_id) | |
2033 | { | |
2034 | struct ib_device *res = NULL; | |
2035 | struct ib_port_data *cur; | |
2036 | ||
2037 | rcu_read_lock(); | |
2038 | hash_for_each_possible_rcu (ndev_hash, cur, ndev_hash_link, | |
2039 | (uintptr_t)ndev) { | |
2040 | if (rcu_access_pointer(cur->netdev) == ndev && | |
2041 | (driver_id == RDMA_DRIVER_UNKNOWN || | |
2042 | cur->ib_dev->driver_id == driver_id) && | |
2043 | ib_device_try_get(cur->ib_dev)) { | |
2044 | res = cur->ib_dev; | |
2045 | break; | |
2046 | } | |
2047 | } | |
2048 | rcu_read_unlock(); | |
2049 | ||
2050 | return res; | |
2051 | } | |
2052 | EXPORT_SYMBOL(ib_device_get_by_netdev); | |
2053 | ||
03db3a2d MB |
2054 | /** |
2055 | * ib_enum_roce_netdev - enumerate all RoCE ports | |
2056 | * @ib_dev : IB device we want to query | |
2057 | * @filter: Should we call the callback? | |
2058 | * @filter_cookie: Cookie passed to filter | |
2059 | * @cb: Callback to call for each found RoCE ports | |
2060 | * @cookie: Cookie passed back to the callback | |
2061 | * | |
2062 | * Enumerates all of the physical RoCE ports of ib_dev | |
2063 | * which are related to netdevice and calls callback() on each | |
2064 | * device for which filter() function returns non zero. | |
2065 | */ | |
2066 | void ib_enum_roce_netdev(struct ib_device *ib_dev, | |
2067 | roce_netdev_filter filter, | |
2068 | void *filter_cookie, | |
2069 | roce_netdev_callback cb, | |
2070 | void *cookie) | |
2071 | { | |
ea1075ed | 2072 | unsigned int port; |
03db3a2d | 2073 | |
ea1075ed | 2074 | rdma_for_each_port (ib_dev, port) |
03db3a2d | 2075 | if (rdma_protocol_roce(ib_dev, port)) { |
c2261dd7 JG |
2076 | struct net_device *idev = |
2077 | ib_device_get_netdev(ib_dev, port); | |
03db3a2d MB |
2078 | |
2079 | if (filter(ib_dev, port, idev, filter_cookie)) | |
2080 | cb(ib_dev, port, idev, cookie); | |
2081 | ||
2082 | if (idev) | |
2083 | dev_put(idev); | |
2084 | } | |
2085 | } | |
2086 | ||
2087 | /** | |
2088 | * ib_enum_all_roce_netdevs - enumerate all RoCE devices | |
2089 | * @filter: Should we call the callback? | |
2090 | * @filter_cookie: Cookie passed to filter | |
2091 | * @cb: Callback to call for each found RoCE ports | |
2092 | * @cookie: Cookie passed back to the callback | |
2093 | * | |
2094 | * Enumerates all RoCE devices' physical ports which are related | |
2095 | * to netdevices and calls callback() on each device for which | |
2096 | * filter() function returns non zero. | |
2097 | */ | |
2098 | void ib_enum_all_roce_netdevs(roce_netdev_filter filter, | |
2099 | void *filter_cookie, | |
2100 | roce_netdev_callback cb, | |
2101 | void *cookie) | |
2102 | { | |
2103 | struct ib_device *dev; | |
0df91bb6 | 2104 | unsigned long index; |
03db3a2d | 2105 | |
921eab11 | 2106 | down_read(&devices_rwsem); |
0df91bb6 | 2107 | xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) |
03db3a2d | 2108 | ib_enum_roce_netdev(dev, filter, filter_cookie, cb, cookie); |
921eab11 | 2109 | up_read(&devices_rwsem); |
8030c835 LR |
2110 | } |
2111 | ||
2112 | /** | |
2113 | * ib_enum_all_devs - enumerate all ib_devices | |
2114 | * @cb: Callback to call for each found ib_device | |
2115 | * | |
2116 | * Enumerates all ib_devices and calls callback() on each device. | |
2117 | */ | |
2118 | int ib_enum_all_devs(nldev_callback nldev_cb, struct sk_buff *skb, | |
2119 | struct netlink_callback *cb) | |
2120 | { | |
0df91bb6 | 2121 | unsigned long index; |
8030c835 LR |
2122 | struct ib_device *dev; |
2123 | unsigned int idx = 0; | |
2124 | int ret = 0; | |
2125 | ||
921eab11 | 2126 | down_read(&devices_rwsem); |
0df91bb6 | 2127 | xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { |
37eeab55 PP |
2128 | if (!rdma_dev_access_netns(dev, sock_net(skb->sk))) |
2129 | continue; | |
2130 | ||
8030c835 LR |
2131 | ret = nldev_cb(dev, skb, cb, idx); |
2132 | if (ret) | |
2133 | break; | |
2134 | idx++; | |
2135 | } | |
921eab11 | 2136 | up_read(&devices_rwsem); |
8030c835 | 2137 | return ret; |
03db3a2d MB |
2138 | } |
2139 | ||
1da177e4 LT |
2140 | /** |
2141 | * ib_query_pkey - Get P_Key table entry | |
2142 | * @device:Device to query | |
2143 | * @port_num:Port number to query | |
2144 | * @index:P_Key table index to query | |
2145 | * @pkey:Returned P_Key | |
2146 | * | |
2147 | * ib_query_pkey() fetches the specified P_Key table entry. | |
2148 | */ | |
2149 | int ib_query_pkey(struct ib_device *device, | |
2150 | u8 port_num, u16 index, u16 *pkey) | |
2151 | { | |
9af3f5cf YS |
2152 | if (!rdma_is_port_valid(device, port_num)) |
2153 | return -EINVAL; | |
2154 | ||
3023a1e9 | 2155 | return device->ops.query_pkey(device, port_num, index, pkey); |
1da177e4 LT |
2156 | } |
2157 | EXPORT_SYMBOL(ib_query_pkey); | |
2158 | ||
2159 | /** | |
2160 | * ib_modify_device - Change IB device attributes | |
2161 | * @device:Device to modify | |
2162 | * @device_modify_mask:Mask of attributes to change | |
2163 | * @device_modify:New attribute values | |
2164 | * | |
2165 | * ib_modify_device() changes a device's attributes as specified by | |
2166 | * the @device_modify_mask and @device_modify structure. | |
2167 | */ | |
2168 | int ib_modify_device(struct ib_device *device, | |
2169 | int device_modify_mask, | |
2170 | struct ib_device_modify *device_modify) | |
2171 | { | |
3023a1e9 | 2172 | if (!device->ops.modify_device) |
10e1b54b BVA |
2173 | return -ENOSYS; |
2174 | ||
3023a1e9 KH |
2175 | return device->ops.modify_device(device, device_modify_mask, |
2176 | device_modify); | |
1da177e4 LT |
2177 | } |
2178 | EXPORT_SYMBOL(ib_modify_device); | |
2179 | ||
2180 | /** | |
2181 | * ib_modify_port - Modifies the attributes for the specified port. | |
2182 | * @device: The device to modify. | |
2183 | * @port_num: The number of the port to modify. | |
2184 | * @port_modify_mask: Mask used to specify which attributes of the port | |
2185 | * to change. | |
2186 | * @port_modify: New attribute values for the port. | |
2187 | * | |
2188 | * ib_modify_port() changes a port's attributes as specified by the | |
2189 | * @port_modify_mask and @port_modify structure. | |
2190 | */ | |
2191 | int ib_modify_port(struct ib_device *device, | |
2192 | u8 port_num, int port_modify_mask, | |
2193 | struct ib_port_modify *port_modify) | |
2194 | { | |
61e0962d | 2195 | int rc; |
10e1b54b | 2196 | |
24dc831b | 2197 | if (!rdma_is_port_valid(device, port_num)) |
116c0074 RD |
2198 | return -EINVAL; |
2199 | ||
3023a1e9 KH |
2200 | if (device->ops.modify_port) |
2201 | rc = device->ops.modify_port(device, port_num, | |
2202 | port_modify_mask, | |
2203 | port_modify); | |
61e0962d SX |
2204 | else |
2205 | rc = rdma_protocol_roce(device, port_num) ? 0 : -ENOSYS; | |
2206 | return rc; | |
1da177e4 LT |
2207 | } |
2208 | EXPORT_SYMBOL(ib_modify_port); | |
2209 | ||
5eb620c8 YE |
2210 | /** |
2211 | * ib_find_gid - Returns the port number and GID table index where | |
dbb12562 | 2212 | * a specified GID value occurs. Its searches only for IB link layer. |
5eb620c8 YE |
2213 | * @device: The device to query. |
2214 | * @gid: The GID value to search for. | |
2215 | * @port_num: The port number of the device where the GID value was found. | |
2216 | * @index: The index into the GID table where the GID was found. This | |
2217 | * parameter may be NULL. | |
2218 | */ | |
2219 | int ib_find_gid(struct ib_device *device, union ib_gid *gid, | |
b26c4a11 | 2220 | u8 *port_num, u16 *index) |
5eb620c8 YE |
2221 | { |
2222 | union ib_gid tmp_gid; | |
ea1075ed JG |
2223 | unsigned int port; |
2224 | int ret, i; | |
5eb620c8 | 2225 | |
ea1075ed | 2226 | rdma_for_each_port (device, port) { |
22d24f75 | 2227 | if (!rdma_protocol_ib(device, port)) |
b39ffa1d MB |
2228 | continue; |
2229 | ||
8ceb1357 JG |
2230 | for (i = 0; i < device->port_data[port].immutable.gid_tbl_len; |
2231 | ++i) { | |
1dfce294 | 2232 | ret = rdma_query_gid(device, port, i, &tmp_gid); |
5eb620c8 YE |
2233 | if (ret) |
2234 | return ret; | |
2235 | if (!memcmp(&tmp_gid, gid, sizeof *gid)) { | |
2236 | *port_num = port; | |
2237 | if (index) | |
2238 | *index = i; | |
2239 | return 0; | |
2240 | } | |
2241 | } | |
2242 | } | |
2243 | ||
2244 | return -ENOENT; | |
2245 | } | |
2246 | EXPORT_SYMBOL(ib_find_gid); | |
2247 | ||
2248 | /** | |
2249 | * ib_find_pkey - Returns the PKey table index where a specified | |
2250 | * PKey value occurs. | |
2251 | * @device: The device to query. | |
2252 | * @port_num: The port number of the device to search for the PKey. | |
2253 | * @pkey: The PKey value to search for. | |
2254 | * @index: The index into the PKey table where the PKey was found. | |
2255 | */ | |
2256 | int ib_find_pkey(struct ib_device *device, | |
2257 | u8 port_num, u16 pkey, u16 *index) | |
2258 | { | |
2259 | int ret, i; | |
2260 | u16 tmp_pkey; | |
ff7166c4 | 2261 | int partial_ix = -1; |
5eb620c8 | 2262 | |
8ceb1357 JG |
2263 | for (i = 0; i < device->port_data[port_num].immutable.pkey_tbl_len; |
2264 | ++i) { | |
5eb620c8 YE |
2265 | ret = ib_query_pkey(device, port_num, i, &tmp_pkey); |
2266 | if (ret) | |
2267 | return ret; | |
36026ecc | 2268 | if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) { |
ff7166c4 JM |
2269 | /* if there is full-member pkey take it.*/ |
2270 | if (tmp_pkey & 0x8000) { | |
2271 | *index = i; | |
2272 | return 0; | |
2273 | } | |
2274 | if (partial_ix < 0) | |
2275 | partial_ix = i; | |
5eb620c8 YE |
2276 | } |
2277 | } | |
2278 | ||
ff7166c4 JM |
2279 | /*no full-member, if exists take the limited*/ |
2280 | if (partial_ix >= 0) { | |
2281 | *index = partial_ix; | |
2282 | return 0; | |
2283 | } | |
5eb620c8 YE |
2284 | return -ENOENT; |
2285 | } | |
2286 | EXPORT_SYMBOL(ib_find_pkey); | |
2287 | ||
9268f72d YK |
2288 | /** |
2289 | * ib_get_net_dev_by_params() - Return the appropriate net_dev | |
2290 | * for a received CM request | |
2291 | * @dev: An RDMA device on which the request has been received. | |
2292 | * @port: Port number on the RDMA device. | |
2293 | * @pkey: The Pkey the request came on. | |
2294 | * @gid: A GID that the net_dev uses to communicate. | |
2295 | * @addr: Contains the IP address that the request specified as its | |
2296 | * destination. | |
921eab11 | 2297 | * |
9268f72d YK |
2298 | */ |
2299 | struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, | |
2300 | u8 port, | |
2301 | u16 pkey, | |
2302 | const union ib_gid *gid, | |
2303 | const struct sockaddr *addr) | |
2304 | { | |
2305 | struct net_device *net_dev = NULL; | |
0df91bb6 JG |
2306 | unsigned long index; |
2307 | void *client_data; | |
9268f72d YK |
2308 | |
2309 | if (!rdma_protocol_ib(dev, port)) | |
2310 | return NULL; | |
2311 | ||
921eab11 JG |
2312 | /* |
2313 | * Holding the read side guarantees that the client will not become | |
2314 | * unregistered while we are calling get_net_dev_by_params() | |
2315 | */ | |
2316 | down_read(&dev->client_data_rwsem); | |
0df91bb6 JG |
2317 | xan_for_each_marked (&dev->client_data, index, client_data, |
2318 | CLIENT_DATA_REGISTERED) { | |
2319 | struct ib_client *client = xa_load(&clients, index); | |
9268f72d | 2320 | |
0df91bb6 | 2321 | if (!client || !client->get_net_dev_by_params) |
9268f72d YK |
2322 | continue; |
2323 | ||
0df91bb6 JG |
2324 | net_dev = client->get_net_dev_by_params(dev, port, pkey, gid, |
2325 | addr, client_data); | |
2326 | if (net_dev) | |
2327 | break; | |
9268f72d | 2328 | } |
921eab11 | 2329 | up_read(&dev->client_data_rwsem); |
9268f72d YK |
2330 | |
2331 | return net_dev; | |
2332 | } | |
2333 | EXPORT_SYMBOL(ib_get_net_dev_by_params); | |
2334 | ||
521ed0d9 KH |
2335 | void ib_set_device_ops(struct ib_device *dev, const struct ib_device_ops *ops) |
2336 | { | |
3023a1e9 | 2337 | struct ib_device_ops *dev_ops = &dev->ops; |
521ed0d9 KH |
2338 | #define SET_DEVICE_OP(ptr, name) \ |
2339 | do { \ | |
2340 | if (ops->name) \ | |
2341 | if (!((ptr)->name)) \ | |
2342 | (ptr)->name = ops->name; \ | |
2343 | } while (0) | |
2344 | ||
30471d4b LR |
2345 | #define SET_OBJ_SIZE(ptr, name) SET_DEVICE_OP(ptr, size_##name) |
2346 | ||
3023a1e9 | 2347 | SET_DEVICE_OP(dev_ops, add_gid); |
2f1927b0 | 2348 | SET_DEVICE_OP(dev_ops, advise_mr); |
3023a1e9 KH |
2349 | SET_DEVICE_OP(dev_ops, alloc_dm); |
2350 | SET_DEVICE_OP(dev_ops, alloc_fmr); | |
2351 | SET_DEVICE_OP(dev_ops, alloc_hw_stats); | |
2352 | SET_DEVICE_OP(dev_ops, alloc_mr); | |
2353 | SET_DEVICE_OP(dev_ops, alloc_mw); | |
2354 | SET_DEVICE_OP(dev_ops, alloc_pd); | |
2355 | SET_DEVICE_OP(dev_ops, alloc_rdma_netdev); | |
2356 | SET_DEVICE_OP(dev_ops, alloc_ucontext); | |
2357 | SET_DEVICE_OP(dev_ops, alloc_xrcd); | |
2358 | SET_DEVICE_OP(dev_ops, attach_mcast); | |
2359 | SET_DEVICE_OP(dev_ops, check_mr_status); | |
2360 | SET_DEVICE_OP(dev_ops, create_ah); | |
2361 | SET_DEVICE_OP(dev_ops, create_counters); | |
2362 | SET_DEVICE_OP(dev_ops, create_cq); | |
2363 | SET_DEVICE_OP(dev_ops, create_flow); | |
2364 | SET_DEVICE_OP(dev_ops, create_flow_action_esp); | |
2365 | SET_DEVICE_OP(dev_ops, create_qp); | |
2366 | SET_DEVICE_OP(dev_ops, create_rwq_ind_table); | |
2367 | SET_DEVICE_OP(dev_ops, create_srq); | |
2368 | SET_DEVICE_OP(dev_ops, create_wq); | |
2369 | SET_DEVICE_OP(dev_ops, dealloc_dm); | |
d0899892 | 2370 | SET_DEVICE_OP(dev_ops, dealloc_driver); |
3023a1e9 KH |
2371 | SET_DEVICE_OP(dev_ops, dealloc_fmr); |
2372 | SET_DEVICE_OP(dev_ops, dealloc_mw); | |
2373 | SET_DEVICE_OP(dev_ops, dealloc_pd); | |
2374 | SET_DEVICE_OP(dev_ops, dealloc_ucontext); | |
2375 | SET_DEVICE_OP(dev_ops, dealloc_xrcd); | |
2376 | SET_DEVICE_OP(dev_ops, del_gid); | |
2377 | SET_DEVICE_OP(dev_ops, dereg_mr); | |
2378 | SET_DEVICE_OP(dev_ops, destroy_ah); | |
2379 | SET_DEVICE_OP(dev_ops, destroy_counters); | |
2380 | SET_DEVICE_OP(dev_ops, destroy_cq); | |
2381 | SET_DEVICE_OP(dev_ops, destroy_flow); | |
2382 | SET_DEVICE_OP(dev_ops, destroy_flow_action); | |
2383 | SET_DEVICE_OP(dev_ops, destroy_qp); | |
2384 | SET_DEVICE_OP(dev_ops, destroy_rwq_ind_table); | |
2385 | SET_DEVICE_OP(dev_ops, destroy_srq); | |
2386 | SET_DEVICE_OP(dev_ops, destroy_wq); | |
2387 | SET_DEVICE_OP(dev_ops, detach_mcast); | |
2388 | SET_DEVICE_OP(dev_ops, disassociate_ucontext); | |
2389 | SET_DEVICE_OP(dev_ops, drain_rq); | |
2390 | SET_DEVICE_OP(dev_ops, drain_sq); | |
ca22354b | 2391 | SET_DEVICE_OP(dev_ops, enable_driver); |
02da3750 | 2392 | SET_DEVICE_OP(dev_ops, fill_res_entry); |
3023a1e9 KH |
2393 | SET_DEVICE_OP(dev_ops, get_dev_fw_str); |
2394 | SET_DEVICE_OP(dev_ops, get_dma_mr); | |
2395 | SET_DEVICE_OP(dev_ops, get_hw_stats); | |
2396 | SET_DEVICE_OP(dev_ops, get_link_layer); | |
2397 | SET_DEVICE_OP(dev_ops, get_netdev); | |
2398 | SET_DEVICE_OP(dev_ops, get_port_immutable); | |
2399 | SET_DEVICE_OP(dev_ops, get_vector_affinity); | |
2400 | SET_DEVICE_OP(dev_ops, get_vf_config); | |
2401 | SET_DEVICE_OP(dev_ops, get_vf_stats); | |
ea4baf7f | 2402 | SET_DEVICE_OP(dev_ops, init_port); |
dd05cb82 KH |
2403 | SET_DEVICE_OP(dev_ops, iw_accept); |
2404 | SET_DEVICE_OP(dev_ops, iw_add_ref); | |
2405 | SET_DEVICE_OP(dev_ops, iw_connect); | |
2406 | SET_DEVICE_OP(dev_ops, iw_create_listen); | |
2407 | SET_DEVICE_OP(dev_ops, iw_destroy_listen); | |
2408 | SET_DEVICE_OP(dev_ops, iw_get_qp); | |
2409 | SET_DEVICE_OP(dev_ops, iw_reject); | |
2410 | SET_DEVICE_OP(dev_ops, iw_rem_ref); | |
3023a1e9 KH |
2411 | SET_DEVICE_OP(dev_ops, map_mr_sg); |
2412 | SET_DEVICE_OP(dev_ops, map_phys_fmr); | |
2413 | SET_DEVICE_OP(dev_ops, mmap); | |
2414 | SET_DEVICE_OP(dev_ops, modify_ah); | |
2415 | SET_DEVICE_OP(dev_ops, modify_cq); | |
2416 | SET_DEVICE_OP(dev_ops, modify_device); | |
2417 | SET_DEVICE_OP(dev_ops, modify_flow_action_esp); | |
2418 | SET_DEVICE_OP(dev_ops, modify_port); | |
2419 | SET_DEVICE_OP(dev_ops, modify_qp); | |
2420 | SET_DEVICE_OP(dev_ops, modify_srq); | |
2421 | SET_DEVICE_OP(dev_ops, modify_wq); | |
2422 | SET_DEVICE_OP(dev_ops, peek_cq); | |
2423 | SET_DEVICE_OP(dev_ops, poll_cq); | |
2424 | SET_DEVICE_OP(dev_ops, post_recv); | |
2425 | SET_DEVICE_OP(dev_ops, post_send); | |
2426 | SET_DEVICE_OP(dev_ops, post_srq_recv); | |
2427 | SET_DEVICE_OP(dev_ops, process_mad); | |
2428 | SET_DEVICE_OP(dev_ops, query_ah); | |
2429 | SET_DEVICE_OP(dev_ops, query_device); | |
2430 | SET_DEVICE_OP(dev_ops, query_gid); | |
2431 | SET_DEVICE_OP(dev_ops, query_pkey); | |
2432 | SET_DEVICE_OP(dev_ops, query_port); | |
2433 | SET_DEVICE_OP(dev_ops, query_qp); | |
2434 | SET_DEVICE_OP(dev_ops, query_srq); | |
2435 | SET_DEVICE_OP(dev_ops, rdma_netdev_get_params); | |
2436 | SET_DEVICE_OP(dev_ops, read_counters); | |
2437 | SET_DEVICE_OP(dev_ops, reg_dm_mr); | |
2438 | SET_DEVICE_OP(dev_ops, reg_user_mr); | |
2439 | SET_DEVICE_OP(dev_ops, req_ncomp_notif); | |
2440 | SET_DEVICE_OP(dev_ops, req_notify_cq); | |
2441 | SET_DEVICE_OP(dev_ops, rereg_user_mr); | |
2442 | SET_DEVICE_OP(dev_ops, resize_cq); | |
2443 | SET_DEVICE_OP(dev_ops, set_vf_guid); | |
2444 | SET_DEVICE_OP(dev_ops, set_vf_link_state); | |
2445 | SET_DEVICE_OP(dev_ops, unmap_fmr); | |
21a428a0 | 2446 | |
d3456914 | 2447 | SET_OBJ_SIZE(dev_ops, ib_ah); |
21a428a0 | 2448 | SET_OBJ_SIZE(dev_ops, ib_pd); |
68e326de | 2449 | SET_OBJ_SIZE(dev_ops, ib_srq); |
a2a074ef | 2450 | SET_OBJ_SIZE(dev_ops, ib_ucontext); |
521ed0d9 KH |
2451 | } |
2452 | EXPORT_SYMBOL(ib_set_device_ops); | |
2453 | ||
d0e312fe | 2454 | static const struct rdma_nl_cbs ibnl_ls_cb_table[RDMA_NL_LS_NUM_OPS] = { |
735c631a | 2455 | [RDMA_NL_LS_OP_RESOLVE] = { |
647c75ac | 2456 | .doit = ib_nl_handle_resolve_resp, |
e3a2b93d LR |
2457 | .flags = RDMA_NL_ADMIN_PERM, |
2458 | }, | |
735c631a | 2459 | [RDMA_NL_LS_OP_SET_TIMEOUT] = { |
647c75ac | 2460 | .doit = ib_nl_handle_set_timeout, |
e3a2b93d LR |
2461 | .flags = RDMA_NL_ADMIN_PERM, |
2462 | }, | |
ae43f828 | 2463 | [RDMA_NL_LS_OP_IP_RESOLVE] = { |
647c75ac | 2464 | .doit = ib_nl_handle_ip_res_resp, |
e3a2b93d LR |
2465 | .flags = RDMA_NL_ADMIN_PERM, |
2466 | }, | |
735c631a MB |
2467 | }; |
2468 | ||
1da177e4 LT |
2469 | static int __init ib_core_init(void) |
2470 | { | |
2471 | int ret; | |
2472 | ||
f0626710 TH |
2473 | ib_wq = alloc_workqueue("infiniband", 0, 0); |
2474 | if (!ib_wq) | |
2475 | return -ENOMEM; | |
2476 | ||
14d3a3b2 | 2477 | ib_comp_wq = alloc_workqueue("ib-comp-wq", |
b7363e67 | 2478 | WQ_HIGHPRI | WQ_MEM_RECLAIM | WQ_SYSFS, 0); |
14d3a3b2 CH |
2479 | if (!ib_comp_wq) { |
2480 | ret = -ENOMEM; | |
2481 | goto err; | |
2482 | } | |
2483 | ||
f794809a JM |
2484 | ib_comp_unbound_wq = |
2485 | alloc_workqueue("ib-comp-unb-wq", | |
2486 | WQ_UNBOUND | WQ_HIGHPRI | WQ_MEM_RECLAIM | | |
2487 | WQ_SYSFS, WQ_UNBOUND_MAX_ACTIVE); | |
2488 | if (!ib_comp_unbound_wq) { | |
2489 | ret = -ENOMEM; | |
2490 | goto err_comp; | |
2491 | } | |
2492 | ||
55aeed06 | 2493 | ret = class_register(&ib_class); |
fd75c789 | 2494 | if (ret) { |
aba25a3e | 2495 | pr_warn("Couldn't create InfiniBand device class\n"); |
f794809a | 2496 | goto err_comp_unbound; |
fd75c789 | 2497 | } |
1da177e4 | 2498 | |
c9901724 | 2499 | ret = rdma_nl_init(); |
b2cbae2c | 2500 | if (ret) { |
c9901724 | 2501 | pr_warn("Couldn't init IB netlink interface: err %d\n", ret); |
b2cbae2c RD |
2502 | goto err_sysfs; |
2503 | } | |
2504 | ||
e3f20f02 LR |
2505 | ret = addr_init(); |
2506 | if (ret) { | |
2507 | pr_warn("Could't init IB address resolution\n"); | |
2508 | goto err_ibnl; | |
2509 | } | |
2510 | ||
4c2cb422 MB |
2511 | ret = ib_mad_init(); |
2512 | if (ret) { | |
2513 | pr_warn("Couldn't init IB MAD\n"); | |
2514 | goto err_addr; | |
2515 | } | |
2516 | ||
c2e49c92 MB |
2517 | ret = ib_sa_init(); |
2518 | if (ret) { | |
2519 | pr_warn("Couldn't init SA\n"); | |
2520 | goto err_mad; | |
2521 | } | |
2522 | ||
8f408ab6 DJ |
2523 | ret = register_lsm_notifier(&ibdev_lsm_nb); |
2524 | if (ret) { | |
2525 | pr_warn("Couldn't register LSM notifier. ret %d\n", ret); | |
c9901724 | 2526 | goto err_sa; |
8f408ab6 DJ |
2527 | } |
2528 | ||
4e0f7b90 PP |
2529 | ret = register_pernet_device(&rdma_dev_net_ops); |
2530 | if (ret) { | |
2531 | pr_warn("Couldn't init compat dev. ret %d\n", ret); | |
2532 | goto err_compat; | |
2533 | } | |
2534 | ||
6c80b41a | 2535 | nldev_init(); |
c9901724 | 2536 | rdma_nl_register(RDMA_NL_LS, ibnl_ls_cb_table); |
5ef8c0c1 | 2537 | roce_gid_mgmt_init(); |
1da177e4 | 2538 | |
fd75c789 NM |
2539 | return 0; |
2540 | ||
4e0f7b90 PP |
2541 | err_compat: |
2542 | unregister_lsm_notifier(&ibdev_lsm_nb); | |
735c631a MB |
2543 | err_sa: |
2544 | ib_sa_cleanup(); | |
c2e49c92 MB |
2545 | err_mad: |
2546 | ib_mad_cleanup(); | |
4c2cb422 MB |
2547 | err_addr: |
2548 | addr_cleanup(); | |
e3f20f02 | 2549 | err_ibnl: |
c9901724 | 2550 | rdma_nl_exit(); |
fd75c789 | 2551 | err_sysfs: |
55aeed06 | 2552 | class_unregister(&ib_class); |
f794809a JM |
2553 | err_comp_unbound: |
2554 | destroy_workqueue(ib_comp_unbound_wq); | |
14d3a3b2 CH |
2555 | err_comp: |
2556 | destroy_workqueue(ib_comp_wq); | |
fd75c789 NM |
2557 | err: |
2558 | destroy_workqueue(ib_wq); | |
1da177e4 LT |
2559 | return ret; |
2560 | } | |
2561 | ||
2562 | static void __exit ib_core_cleanup(void) | |
2563 | { | |
5ef8c0c1 | 2564 | roce_gid_mgmt_cleanup(); |
6c80b41a | 2565 | nldev_exit(); |
c9901724 | 2566 | rdma_nl_unregister(RDMA_NL_LS); |
4e0f7b90 | 2567 | unregister_pernet_device(&rdma_dev_net_ops); |
c9901724 | 2568 | unregister_lsm_notifier(&ibdev_lsm_nb); |
c2e49c92 | 2569 | ib_sa_cleanup(); |
4c2cb422 | 2570 | ib_mad_cleanup(); |
e3f20f02 | 2571 | addr_cleanup(); |
c9901724 | 2572 | rdma_nl_exit(); |
55aeed06 | 2573 | class_unregister(&ib_class); |
f794809a | 2574 | destroy_workqueue(ib_comp_unbound_wq); |
14d3a3b2 | 2575 | destroy_workqueue(ib_comp_wq); |
f7c6a7b5 | 2576 | /* Make sure that any pending umem accounting work is done. */ |
f0626710 | 2577 | destroy_workqueue(ib_wq); |
d0899892 | 2578 | flush_workqueue(system_unbound_wq); |
e59178d8 | 2579 | WARN_ON(!xa_empty(&clients)); |
0df91bb6 | 2580 | WARN_ON(!xa_empty(&devices)); |
1da177e4 LT |
2581 | } |
2582 | ||
e3bf14bd JG |
2583 | MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_LS, 4); |
2584 | ||
62dfa795 PP |
2585 | /* ib core relies on netdev stack to first register net_ns_type_operations |
2586 | * ns kobject type before ib_core initialization. | |
2587 | */ | |
2588 | fs_initcall(ib_core_init); | |
1da177e4 | 2589 | module_exit(ib_core_cleanup); |