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 | { | |
e3593b56 | 412 | int ret; |
d21943dd | 413 | |
921eab11 | 414 | down_write(&devices_rwsem); |
e3593b56 JG |
415 | if (!strcmp(name, dev_name(&ibdev->dev))) { |
416 | ret = 0; | |
417 | goto out; | |
418 | } | |
419 | ||
344684e6 JG |
420 | if (__ib_device_get_by_name(name)) { |
421 | ret = -EEXIST; | |
422 | goto out; | |
d21943dd LR |
423 | } |
424 | ||
425 | ret = device_rename(&ibdev->dev, name); | |
426 | if (ret) | |
427 | goto out; | |
428 | strlcpy(ibdev->name, name, IB_DEVICE_NAME_MAX); | |
4e0f7b90 | 429 | ret = rename_compat_devs(ibdev); |
d21943dd | 430 | out: |
921eab11 | 431 | up_write(&devices_rwsem); |
d21943dd LR |
432 | return ret; |
433 | } | |
434 | ||
e349f858 | 435 | static int alloc_name(struct ib_device *ibdev, const char *name) |
1da177e4 | 436 | { |
1da177e4 | 437 | struct ib_device *device; |
0df91bb6 | 438 | unsigned long index; |
3b88afd3 JG |
439 | struct ida inuse; |
440 | int rc; | |
1da177e4 LT |
441 | int i; |
442 | ||
921eab11 | 443 | lockdep_assert_held_exclusive(&devices_rwsem); |
3b88afd3 | 444 | ida_init(&inuse); |
0df91bb6 | 445 | xa_for_each (&devices, index, device) { |
e349f858 JG |
446 | char buf[IB_DEVICE_NAME_MAX]; |
447 | ||
896de009 | 448 | if (sscanf(dev_name(&device->dev), name, &i) != 1) |
1da177e4 | 449 | continue; |
3b88afd3 | 450 | if (i < 0 || i >= INT_MAX) |
1da177e4 LT |
451 | continue; |
452 | snprintf(buf, sizeof buf, name, i); | |
3b88afd3 JG |
453 | if (strcmp(buf, dev_name(&device->dev)) != 0) |
454 | continue; | |
455 | ||
456 | rc = ida_alloc_range(&inuse, i, i, GFP_KERNEL); | |
457 | if (rc < 0) | |
458 | goto out; | |
1da177e4 LT |
459 | } |
460 | ||
3b88afd3 JG |
461 | rc = ida_alloc(&inuse, GFP_KERNEL); |
462 | if (rc < 0) | |
463 | goto out; | |
1da177e4 | 464 | |
3b88afd3 JG |
465 | rc = dev_set_name(&ibdev->dev, name, rc); |
466 | out: | |
467 | ida_destroy(&inuse); | |
468 | return rc; | |
1da177e4 LT |
469 | } |
470 | ||
55aeed06 JG |
471 | static void ib_device_release(struct device *device) |
472 | { | |
473 | struct ib_device *dev = container_of(device, struct ib_device, dev); | |
474 | ||
c2261dd7 | 475 | free_netdevs(dev); |
652432f3 | 476 | WARN_ON(refcount_read(&dev->refcount)); |
d45f89d5 | 477 | ib_cache_release_one(dev); |
b34b269a | 478 | ib_security_release_port_pkey_list(dev); |
4e0f7b90 | 479 | xa_destroy(&dev->compat_devs); |
0df91bb6 | 480 | xa_destroy(&dev->client_data); |
324e227e JG |
481 | if (dev->port_data) |
482 | kfree_rcu(container_of(dev->port_data, struct ib_port_data_rcu, | |
483 | pdata[0]), | |
484 | rcu_head); | |
485 | kfree_rcu(dev, rcu_head); | |
55aeed06 JG |
486 | } |
487 | ||
488 | static int ib_device_uevent(struct device *device, | |
489 | struct kobj_uevent_env *env) | |
490 | { | |
896de009 | 491 | if (add_uevent_var(env, "NAME=%s", dev_name(device))) |
55aeed06 JG |
492 | return -ENOMEM; |
493 | ||
494 | /* | |
495 | * It would be nice to pass the node GUID with the event... | |
496 | */ | |
497 | ||
498 | return 0; | |
499 | } | |
500 | ||
62dfa795 PP |
501 | static const void *net_namespace(struct device *d) |
502 | { | |
4e0f7b90 PP |
503 | struct ib_core_device *coredev = |
504 | container_of(d, struct ib_core_device, dev); | |
505 | ||
506 | return read_pnet(&coredev->rdma_net); | |
62dfa795 PP |
507 | } |
508 | ||
55aeed06 JG |
509 | static struct class ib_class = { |
510 | .name = "infiniband", | |
511 | .dev_release = ib_device_release, | |
512 | .dev_uevent = ib_device_uevent, | |
62dfa795 PP |
513 | .ns_type = &net_ns_type_operations, |
514 | .namespace = net_namespace, | |
55aeed06 JG |
515 | }; |
516 | ||
cebe556b | 517 | static void rdma_init_coredev(struct ib_core_device *coredev, |
4e0f7b90 | 518 | struct ib_device *dev, struct net *net) |
cebe556b PP |
519 | { |
520 | /* This BUILD_BUG_ON is intended to catch layout change | |
521 | * of union of ib_core_device and device. | |
522 | * dev must be the first element as ib_core and providers | |
523 | * driver uses it. Adding anything in ib_core_device before | |
524 | * device will break this assumption. | |
525 | */ | |
526 | BUILD_BUG_ON(offsetof(struct ib_device, coredev.dev) != | |
527 | offsetof(struct ib_device, dev)); | |
528 | ||
529 | coredev->dev.class = &ib_class; | |
530 | coredev->dev.groups = dev->groups; | |
531 | device_initialize(&coredev->dev); | |
532 | coredev->owner = dev; | |
533 | INIT_LIST_HEAD(&coredev->port_list); | |
4e0f7b90 | 534 | write_pnet(&coredev->rdma_net, net); |
cebe556b PP |
535 | } |
536 | ||
1da177e4 | 537 | /** |
459cc69f | 538 | * _ib_alloc_device - allocate an IB device struct |
1da177e4 LT |
539 | * @size:size of structure to allocate |
540 | * | |
541 | * Low-level drivers should use ib_alloc_device() to allocate &struct | |
542 | * ib_device. @size is the size of the structure to be allocated, | |
543 | * including any private data used by the low-level driver. | |
544 | * ib_dealloc_device() must be used to free structures allocated with | |
545 | * ib_alloc_device(). | |
546 | */ | |
459cc69f | 547 | struct ib_device *_ib_alloc_device(size_t size) |
1da177e4 | 548 | { |
55aeed06 JG |
549 | struct ib_device *device; |
550 | ||
551 | if (WARN_ON(size < sizeof(struct ib_device))) | |
552 | return NULL; | |
553 | ||
554 | device = kzalloc(size, GFP_KERNEL); | |
555 | if (!device) | |
556 | return NULL; | |
557 | ||
41eda65c LR |
558 | if (rdma_restrack_init(device)) { |
559 | kfree(device); | |
560 | return NULL; | |
561 | } | |
02d8883f | 562 | |
5f8f5499 | 563 | device->groups[0] = &ib_dev_attr_group; |
4e0f7b90 | 564 | rdma_init_coredev(&device->coredev, device, &init_net); |
55aeed06 | 565 | |
55aeed06 JG |
566 | INIT_LIST_HEAD(&device->event_handler_list); |
567 | spin_lock_init(&device->event_handler_lock); | |
d0899892 | 568 | mutex_init(&device->unregistration_lock); |
0df91bb6 JG |
569 | /* |
570 | * client_data needs to be alloc because we don't want our mark to be | |
571 | * destroyed if the user stores NULL in the client data. | |
572 | */ | |
573 | xa_init_flags(&device->client_data, XA_FLAGS_ALLOC); | |
921eab11 | 574 | init_rwsem(&device->client_data_rwsem); |
4e0f7b90 PP |
575 | xa_init_flags(&device->compat_devs, XA_FLAGS_ALLOC); |
576 | mutex_init(&device->compat_devs_mutex); | |
01b67117 | 577 | init_completion(&device->unreg_completion); |
d0899892 | 578 | INIT_WORK(&device->unregistration_work, ib_unregister_work); |
1da177e4 | 579 | |
55aeed06 | 580 | return device; |
1da177e4 | 581 | } |
459cc69f | 582 | EXPORT_SYMBOL(_ib_alloc_device); |
1da177e4 LT |
583 | |
584 | /** | |
585 | * ib_dealloc_device - free an IB device struct | |
586 | * @device:structure to free | |
587 | * | |
588 | * Free a structure allocated with ib_alloc_device(). | |
589 | */ | |
590 | void ib_dealloc_device(struct ib_device *device) | |
591 | { | |
d0899892 JG |
592 | if (device->ops.dealloc_driver) |
593 | device->ops.dealloc_driver(device); | |
594 | ||
595 | /* | |
596 | * ib_unregister_driver() requires all devices to remain in the xarray | |
597 | * while their ops are callable. The last op we call is dealloc_driver | |
598 | * above. This is needed to create a fence on op callbacks prior to | |
599 | * allowing the driver module to unload. | |
600 | */ | |
601 | down_write(&devices_rwsem); | |
602 | if (xa_load(&devices, device->index) == device) | |
603 | xa_erase(&devices, device->index); | |
604 | up_write(&devices_rwsem); | |
605 | ||
c2261dd7 JG |
606 | /* Expedite releasing netdev references */ |
607 | free_netdevs(device); | |
608 | ||
4e0f7b90 | 609 | WARN_ON(!xa_empty(&device->compat_devs)); |
0df91bb6 | 610 | WARN_ON(!xa_empty(&device->client_data)); |
652432f3 | 611 | WARN_ON(refcount_read(&device->refcount)); |
0ad699c0 | 612 | rdma_restrack_clean(device); |
e155755e | 613 | /* Balances with device_initialize */ |
924b8900 | 614 | put_device(&device->dev); |
1da177e4 LT |
615 | } |
616 | EXPORT_SYMBOL(ib_dealloc_device); | |
617 | ||
921eab11 JG |
618 | /* |
619 | * add_client_context() and remove_client_context() must be safe against | |
620 | * parallel calls on the same device - registration/unregistration of both the | |
621 | * device and client can be occurring in parallel. | |
622 | * | |
623 | * The routines need to be a fence, any caller must not return until the add | |
624 | * or remove is fully completed. | |
625 | */ | |
626 | static int add_client_context(struct ib_device *device, | |
627 | struct ib_client *client) | |
1da177e4 | 628 | { |
921eab11 | 629 | int ret = 0; |
1da177e4 | 630 | |
6780c4fa | 631 | if (!device->kverbs_provider && !client->no_kverbs_req) |
921eab11 JG |
632 | return 0; |
633 | ||
634 | down_write(&device->client_data_rwsem); | |
635 | /* | |
636 | * Another caller to add_client_context got here first and has already | |
637 | * completely initialized context. | |
638 | */ | |
639 | if (xa_get_mark(&device->client_data, client->client_id, | |
640 | CLIENT_DATA_REGISTERED)) | |
641 | goto out; | |
642 | ||
643 | ret = xa_err(xa_store(&device->client_data, client->client_id, NULL, | |
644 | GFP_KERNEL)); | |
645 | if (ret) | |
646 | goto out; | |
647 | downgrade_write(&device->client_data_rwsem); | |
648 | if (client->add) | |
649 | client->add(device); | |
650 | ||
651 | /* Readers shall not see a client until add has been completed */ | |
652 | xa_set_mark(&device->client_data, client->client_id, | |
653 | CLIENT_DATA_REGISTERED); | |
654 | up_read(&device->client_data_rwsem); | |
655 | return 0; | |
656 | ||
657 | out: | |
658 | up_write(&device->client_data_rwsem); | |
659 | return ret; | |
660 | } | |
661 | ||
662 | static void remove_client_context(struct ib_device *device, | |
663 | unsigned int client_id) | |
664 | { | |
665 | struct ib_client *client; | |
666 | void *client_data; | |
6780c4fa | 667 | |
921eab11 JG |
668 | down_write(&device->client_data_rwsem); |
669 | if (!xa_get_mark(&device->client_data, client_id, | |
670 | CLIENT_DATA_REGISTERED)) { | |
671 | up_write(&device->client_data_rwsem); | |
672 | return; | |
673 | } | |
674 | client_data = xa_load(&device->client_data, client_id); | |
675 | xa_clear_mark(&device->client_data, client_id, CLIENT_DATA_REGISTERED); | |
676 | client = xa_load(&clients, client_id); | |
677 | downgrade_write(&device->client_data_rwsem); | |
1da177e4 | 678 | |
921eab11 JG |
679 | /* |
680 | * Notice we cannot be holding any exclusive locks when calling the | |
681 | * remove callback as the remove callback can recurse back into any | |
682 | * public functions in this module and thus try for any locks those | |
683 | * functions take. | |
684 | * | |
685 | * For this reason clients and drivers should not call the | |
686 | * unregistration functions will holdling any locks. | |
687 | * | |
688 | * It tempting to drop the client_data_rwsem too, but this is required | |
689 | * to ensure that unregister_client does not return until all clients | |
690 | * are completely unregistered, which is required to avoid module | |
691 | * unloading races. | |
692 | */ | |
693 | if (client->remove) | |
694 | client->remove(device, client_data); | |
695 | ||
696 | xa_erase(&device->client_data, client_id); | |
697 | up_read(&device->client_data_rwsem); | |
1da177e4 LT |
698 | } |
699 | ||
c2261dd7 | 700 | static int alloc_port_data(struct ib_device *device) |
5eb620c8 | 701 | { |
324e227e | 702 | struct ib_port_data_rcu *pdata_rcu; |
ea1075ed | 703 | unsigned int port; |
c2261dd7 JG |
704 | |
705 | if (device->port_data) | |
706 | return 0; | |
707 | ||
708 | /* This can only be called once the physical port range is defined */ | |
709 | if (WARN_ON(!device->phys_port_cnt)) | |
710 | return -EINVAL; | |
7738613e | 711 | |
8ceb1357 JG |
712 | /* |
713 | * device->port_data is indexed directly by the port number to make | |
7738613e IW |
714 | * access to this data as efficient as possible. |
715 | * | |
8ceb1357 JG |
716 | * Therefore port_data is declared as a 1 based array with potential |
717 | * empty slots at the beginning. | |
7738613e | 718 | */ |
324e227e JG |
719 | pdata_rcu = kzalloc(struct_size(pdata_rcu, pdata, |
720 | rdma_end_port(device) + 1), | |
721 | GFP_KERNEL); | |
722 | if (!pdata_rcu) | |
55aeed06 | 723 | return -ENOMEM; |
324e227e JG |
724 | /* |
725 | * The rcu_head is put in front of the port data array and the stored | |
726 | * pointer is adjusted since we never need to see that member until | |
727 | * kfree_rcu. | |
728 | */ | |
729 | device->port_data = pdata_rcu->pdata; | |
5eb620c8 | 730 | |
ea1075ed | 731 | rdma_for_each_port (device, port) { |
8ceb1357 JG |
732 | struct ib_port_data *pdata = &device->port_data[port]; |
733 | ||
324e227e | 734 | pdata->ib_dev = device; |
8ceb1357 JG |
735 | spin_lock_init(&pdata->pkey_list_lock); |
736 | INIT_LIST_HEAD(&pdata->pkey_list); | |
c2261dd7 | 737 | spin_lock_init(&pdata->netdev_lock); |
324e227e | 738 | INIT_HLIST_NODE(&pdata->ndev_hash_link); |
c2261dd7 JG |
739 | } |
740 | return 0; | |
741 | } | |
742 | ||
743 | static int verify_immutable(const struct ib_device *dev, u8 port) | |
744 | { | |
745 | return WARN_ON(!rdma_cap_ib_mad(dev, port) && | |
746 | rdma_max_mad_size(dev, port) != 0); | |
747 | } | |
748 | ||
749 | static int setup_port_data(struct ib_device *device) | |
750 | { | |
751 | unsigned int port; | |
752 | int ret; | |
753 | ||
754 | ret = alloc_port_data(device); | |
755 | if (ret) | |
756 | return ret; | |
757 | ||
758 | rdma_for_each_port (device, port) { | |
759 | struct ib_port_data *pdata = &device->port_data[port]; | |
8ceb1357 JG |
760 | |
761 | ret = device->ops.get_port_immutable(device, port, | |
762 | &pdata->immutable); | |
5eb620c8 | 763 | if (ret) |
55aeed06 | 764 | return ret; |
337877a4 | 765 | |
55aeed06 JG |
766 | if (verify_immutable(device, port)) |
767 | return -EINVAL; | |
5eb620c8 | 768 | } |
55aeed06 | 769 | return 0; |
5eb620c8 YE |
770 | } |
771 | ||
9abb0d1b | 772 | void ib_get_device_fw_str(struct ib_device *dev, char *str) |
5fa76c20 | 773 | { |
3023a1e9 KH |
774 | if (dev->ops.get_dev_fw_str) |
775 | dev->ops.get_dev_fw_str(dev, str); | |
5fa76c20 IW |
776 | else |
777 | str[0] = '\0'; | |
778 | } | |
779 | EXPORT_SYMBOL(ib_get_device_fw_str); | |
780 | ||
8f408ab6 DJ |
781 | static void ib_policy_change_task(struct work_struct *work) |
782 | { | |
783 | struct ib_device *dev; | |
0df91bb6 | 784 | unsigned long index; |
8f408ab6 | 785 | |
921eab11 | 786 | down_read(&devices_rwsem); |
0df91bb6 | 787 | xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { |
ea1075ed | 788 | unsigned int i; |
8f408ab6 | 789 | |
ea1075ed | 790 | rdma_for_each_port (dev, i) { |
8f408ab6 DJ |
791 | u64 sp; |
792 | int ret = ib_get_cached_subnet_prefix(dev, | |
793 | i, | |
794 | &sp); | |
795 | ||
796 | WARN_ONCE(ret, | |
797 | "ib_get_cached_subnet_prefix err: %d, this should never happen here\n", | |
798 | ret); | |
a750cfde DJ |
799 | if (!ret) |
800 | ib_security_cache_change(dev, i, sp); | |
8f408ab6 DJ |
801 | } |
802 | } | |
921eab11 | 803 | up_read(&devices_rwsem); |
8f408ab6 DJ |
804 | } |
805 | ||
806 | static int ib_security_change(struct notifier_block *nb, unsigned long event, | |
807 | void *lsm_data) | |
808 | { | |
809 | if (event != LSM_POLICY_CHANGE) | |
810 | return NOTIFY_DONE; | |
811 | ||
812 | schedule_work(&ib_policy_change_work); | |
c66f6741 | 813 | ib_mad_agent_security_change(); |
8f408ab6 DJ |
814 | |
815 | return NOTIFY_OK; | |
816 | } | |
817 | ||
4e0f7b90 PP |
818 | static void compatdev_release(struct device *dev) |
819 | { | |
820 | struct ib_core_device *cdev = | |
821 | container_of(dev, struct ib_core_device, dev); | |
822 | ||
823 | kfree(cdev); | |
824 | } | |
825 | ||
826 | static int add_one_compat_dev(struct ib_device *device, | |
827 | struct rdma_dev_net *rnet) | |
828 | { | |
829 | struct ib_core_device *cdev; | |
830 | int ret; | |
831 | ||
2b34c558 | 832 | lockdep_assert_held(&rdma_nets_rwsem); |
a56bc45b PP |
833 | if (!ib_devices_shared_netns) |
834 | return 0; | |
835 | ||
4e0f7b90 PP |
836 | /* |
837 | * Create and add compat device in all namespaces other than where it | |
838 | * is currently bound to. | |
839 | */ | |
840 | if (net_eq(read_pnet(&rnet->net), | |
841 | read_pnet(&device->coredev.rdma_net))) | |
842 | return 0; | |
843 | ||
844 | /* | |
845 | * The first of init_net() or ib_register_device() to take the | |
846 | * compat_devs_mutex wins and gets to add the device. Others will wait | |
847 | * for completion here. | |
848 | */ | |
849 | mutex_lock(&device->compat_devs_mutex); | |
850 | cdev = xa_load(&device->compat_devs, rnet->id); | |
851 | if (cdev) { | |
852 | ret = 0; | |
853 | goto done; | |
854 | } | |
855 | ret = xa_reserve(&device->compat_devs, rnet->id, GFP_KERNEL); | |
856 | if (ret) | |
857 | goto done; | |
858 | ||
859 | cdev = kzalloc(sizeof(*cdev), GFP_KERNEL); | |
860 | if (!cdev) { | |
861 | ret = -ENOMEM; | |
862 | goto cdev_err; | |
863 | } | |
864 | ||
865 | cdev->dev.parent = device->dev.parent; | |
866 | rdma_init_coredev(cdev, device, read_pnet(&rnet->net)); | |
867 | cdev->dev.release = compatdev_release; | |
868 | dev_set_name(&cdev->dev, "%s", dev_name(&device->dev)); | |
869 | ||
870 | ret = device_add(&cdev->dev); | |
871 | if (ret) | |
872 | goto add_err; | |
eb15c78b | 873 | ret = ib_setup_port_attrs(cdev); |
5417783e PP |
874 | if (ret) |
875 | goto port_err; | |
4e0f7b90 PP |
876 | |
877 | ret = xa_err(xa_store(&device->compat_devs, rnet->id, | |
878 | cdev, GFP_KERNEL)); | |
879 | if (ret) | |
880 | goto insert_err; | |
881 | ||
882 | mutex_unlock(&device->compat_devs_mutex); | |
883 | return 0; | |
884 | ||
885 | insert_err: | |
5417783e PP |
886 | ib_free_port_attrs(cdev); |
887 | port_err: | |
4e0f7b90 PP |
888 | device_del(&cdev->dev); |
889 | add_err: | |
890 | put_device(&cdev->dev); | |
891 | cdev_err: | |
892 | xa_release(&device->compat_devs, rnet->id); | |
893 | done: | |
894 | mutex_unlock(&device->compat_devs_mutex); | |
895 | return ret; | |
896 | } | |
897 | ||
898 | static void remove_one_compat_dev(struct ib_device *device, u32 id) | |
899 | { | |
900 | struct ib_core_device *cdev; | |
901 | ||
902 | mutex_lock(&device->compat_devs_mutex); | |
903 | cdev = xa_erase(&device->compat_devs, id); | |
904 | mutex_unlock(&device->compat_devs_mutex); | |
905 | if (cdev) { | |
5417783e | 906 | ib_free_port_attrs(cdev); |
4e0f7b90 PP |
907 | device_del(&cdev->dev); |
908 | put_device(&cdev->dev); | |
909 | } | |
910 | } | |
911 | ||
912 | static void remove_compat_devs(struct ib_device *device) | |
913 | { | |
914 | struct ib_core_device *cdev; | |
915 | unsigned long index; | |
916 | ||
917 | xa_for_each (&device->compat_devs, index, cdev) | |
918 | remove_one_compat_dev(device, index); | |
919 | } | |
920 | ||
921 | static int add_compat_devs(struct ib_device *device) | |
922 | { | |
923 | struct rdma_dev_net *rnet; | |
924 | unsigned long index; | |
925 | int ret = 0; | |
926 | ||
decbc7a6 PP |
927 | lockdep_assert_held(&devices_rwsem); |
928 | ||
4e0f7b90 PP |
929 | down_read(&rdma_nets_rwsem); |
930 | xa_for_each (&rdma_nets, index, rnet) { | |
931 | ret = add_one_compat_dev(device, rnet); | |
932 | if (ret) | |
933 | break; | |
934 | } | |
935 | up_read(&rdma_nets_rwsem); | |
936 | return ret; | |
937 | } | |
938 | ||
2b34c558 PP |
939 | static void remove_all_compat_devs(void) |
940 | { | |
941 | struct ib_compat_device *cdev; | |
942 | struct ib_device *dev; | |
943 | unsigned long index; | |
944 | ||
945 | down_read(&devices_rwsem); | |
946 | xa_for_each (&devices, index, dev) { | |
947 | unsigned long c_index = 0; | |
948 | ||
949 | /* Hold nets_rwsem so that any other thread modifying this | |
950 | * system param can sync with this thread. | |
951 | */ | |
952 | down_read(&rdma_nets_rwsem); | |
953 | xa_for_each (&dev->compat_devs, c_index, cdev) | |
954 | remove_one_compat_dev(dev, c_index); | |
955 | up_read(&rdma_nets_rwsem); | |
956 | } | |
957 | up_read(&devices_rwsem); | |
958 | } | |
959 | ||
960 | static int add_all_compat_devs(void) | |
961 | { | |
962 | struct rdma_dev_net *rnet; | |
963 | struct ib_device *dev; | |
964 | unsigned long index; | |
965 | int ret = 0; | |
966 | ||
967 | down_read(&devices_rwsem); | |
968 | xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { | |
969 | unsigned long net_index = 0; | |
970 | ||
971 | /* Hold nets_rwsem so that any other thread modifying this | |
972 | * system param can sync with this thread. | |
973 | */ | |
974 | down_read(&rdma_nets_rwsem); | |
975 | xa_for_each (&rdma_nets, net_index, rnet) { | |
976 | ret = add_one_compat_dev(dev, rnet); | |
977 | if (ret) | |
978 | break; | |
979 | } | |
980 | up_read(&rdma_nets_rwsem); | |
981 | } | |
982 | up_read(&devices_rwsem); | |
983 | if (ret) | |
984 | remove_all_compat_devs(); | |
985 | return ret; | |
986 | } | |
987 | ||
988 | int rdma_compatdev_set(u8 enable) | |
989 | { | |
990 | struct rdma_dev_net *rnet; | |
991 | unsigned long index; | |
992 | int ret = 0; | |
993 | ||
994 | down_write(&rdma_nets_rwsem); | |
995 | if (ib_devices_shared_netns == enable) { | |
996 | up_write(&rdma_nets_rwsem); | |
997 | return 0; | |
998 | } | |
999 | ||
1000 | /* enable/disable of compat devices is not supported | |
1001 | * when more than default init_net exists. | |
1002 | */ | |
1003 | xa_for_each (&rdma_nets, index, rnet) { | |
1004 | ret++; | |
1005 | break; | |
1006 | } | |
1007 | if (!ret) | |
1008 | ib_devices_shared_netns = enable; | |
1009 | up_write(&rdma_nets_rwsem); | |
1010 | if (ret) | |
1011 | return -EBUSY; | |
1012 | ||
1013 | if (enable) | |
1014 | ret = add_all_compat_devs(); | |
1015 | else | |
1016 | remove_all_compat_devs(); | |
1017 | return ret; | |
1018 | } | |
1019 | ||
4e0f7b90 PP |
1020 | static void rdma_dev_exit_net(struct net *net) |
1021 | { | |
1022 | struct rdma_dev_net *rnet = net_generic(net, rdma_dev_net_id); | |
1023 | struct ib_device *dev; | |
1024 | unsigned long index; | |
1025 | int ret; | |
1026 | ||
1027 | down_write(&rdma_nets_rwsem); | |
1028 | /* | |
1029 | * Prevent the ID from being re-used and hide the id from xa_for_each. | |
1030 | */ | |
1031 | ret = xa_err(xa_store(&rdma_nets, rnet->id, NULL, GFP_KERNEL)); | |
1032 | WARN_ON(ret); | |
1033 | up_write(&rdma_nets_rwsem); | |
1034 | ||
1035 | down_read(&devices_rwsem); | |
1036 | xa_for_each (&devices, index, dev) { | |
1037 | get_device(&dev->dev); | |
1038 | /* | |
1039 | * Release the devices_rwsem so that pontentially blocking | |
1040 | * device_del, doesn't hold the devices_rwsem for too long. | |
1041 | */ | |
1042 | up_read(&devices_rwsem); | |
1043 | ||
1044 | remove_one_compat_dev(dev, rnet->id); | |
1045 | ||
decbc7a6 PP |
1046 | /* |
1047 | * If the real device is in the NS then move it back to init. | |
1048 | */ | |
1049 | rdma_dev_change_netns(dev, net, &init_net); | |
1050 | ||
4e0f7b90 PP |
1051 | put_device(&dev->dev); |
1052 | down_read(&devices_rwsem); | |
1053 | } | |
1054 | up_read(&devices_rwsem); | |
1055 | ||
1056 | xa_erase(&rdma_nets, rnet->id); | |
1057 | } | |
1058 | ||
1059 | static __net_init int rdma_dev_init_net(struct net *net) | |
1060 | { | |
1061 | struct rdma_dev_net *rnet = net_generic(net, rdma_dev_net_id); | |
1062 | unsigned long index; | |
1063 | struct ib_device *dev; | |
1064 | int ret; | |
1065 | ||
1066 | /* No need to create any compat devices in default init_net. */ | |
1067 | if (net_eq(net, &init_net)) | |
1068 | return 0; | |
1069 | ||
1070 | write_pnet(&rnet->net, net); | |
1071 | ||
1072 | ret = xa_alloc(&rdma_nets, &rnet->id, rnet, xa_limit_32b, GFP_KERNEL); | |
1073 | if (ret) | |
1074 | return ret; | |
1075 | ||
1076 | down_read(&devices_rwsem); | |
1077 | xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { | |
2b34c558 PP |
1078 | /* Hold nets_rwsem so that netlink command cannot change |
1079 | * system configuration for device sharing mode. | |
1080 | */ | |
1081 | down_read(&rdma_nets_rwsem); | |
4e0f7b90 | 1082 | ret = add_one_compat_dev(dev, rnet); |
2b34c558 | 1083 | up_read(&rdma_nets_rwsem); |
4e0f7b90 PP |
1084 | if (ret) |
1085 | break; | |
1086 | } | |
1087 | up_read(&devices_rwsem); | |
1088 | ||
1089 | if (ret) | |
1090 | rdma_dev_exit_net(net); | |
1091 | ||
1092 | return ret; | |
1093 | } | |
1094 | ||
0df91bb6 | 1095 | /* |
d0899892 JG |
1096 | * Assign the unique string device name and the unique device index. This is |
1097 | * undone by ib_dealloc_device. | |
ecc82c53 | 1098 | */ |
0df91bb6 | 1099 | static int assign_name(struct ib_device *device, const char *name) |
ecc82c53 | 1100 | { |
0df91bb6 JG |
1101 | static u32 last_id; |
1102 | int ret; | |
ecc82c53 | 1103 | |
921eab11 | 1104 | down_write(&devices_rwsem); |
0df91bb6 JG |
1105 | /* Assign a unique name to the device */ |
1106 | if (strchr(name, '%')) | |
1107 | ret = alloc_name(device, name); | |
1108 | else | |
1109 | ret = dev_set_name(&device->dev, name); | |
1110 | if (ret) | |
1111 | goto out; | |
1112 | ||
1113 | if (__ib_device_get_by_name(dev_name(&device->dev))) { | |
1114 | ret = -ENFILE; | |
1115 | goto out; | |
1116 | } | |
1117 | strlcpy(device->name, dev_name(&device->dev), IB_DEVICE_NAME_MAX); | |
ecc82c53 | 1118 | |
ea295481 LT |
1119 | ret = xa_alloc_cyclic(&devices, &device->index, device, xa_limit_31b, |
1120 | &last_id, GFP_KERNEL); | |
1121 | if (ret > 0) | |
1122 | ret = 0; | |
921eab11 | 1123 | |
0df91bb6 | 1124 | out: |
921eab11 | 1125 | up_write(&devices_rwsem); |
0df91bb6 JG |
1126 | return ret; |
1127 | } | |
1128 | ||
548cb4fb | 1129 | static void setup_dma_device(struct ib_device *device) |
1da177e4 | 1130 | { |
99db9494 BVA |
1131 | struct device *parent = device->dev.parent; |
1132 | ||
0957c29f BVA |
1133 | WARN_ON_ONCE(device->dma_device); |
1134 | if (device->dev.dma_ops) { | |
1135 | /* | |
1136 | * The caller provided custom DMA operations. Copy the | |
1137 | * DMA-related fields that are used by e.g. dma_alloc_coherent() | |
1138 | * into device->dev. | |
1139 | */ | |
1140 | device->dma_device = &device->dev; | |
02ee9da3 BVA |
1141 | if (!device->dev.dma_mask) { |
1142 | if (parent) | |
1143 | device->dev.dma_mask = parent->dma_mask; | |
1144 | else | |
1145 | WARN_ON_ONCE(true); | |
1146 | } | |
1147 | if (!device->dev.coherent_dma_mask) { | |
1148 | if (parent) | |
1149 | device->dev.coherent_dma_mask = | |
1150 | parent->coherent_dma_mask; | |
1151 | else | |
1152 | WARN_ON_ONCE(true); | |
1153 | } | |
0957c29f BVA |
1154 | } else { |
1155 | /* | |
1156 | * The caller did not provide custom DMA operations. Use the | |
1157 | * DMA mapping operations of the parent device. | |
1158 | */ | |
02ee9da3 | 1159 | WARN_ON_ONCE(!parent); |
0957c29f BVA |
1160 | device->dma_device = parent; |
1161 | } | |
d10bcf94 SS |
1162 | /* Setup default max segment size for all IB devices */ |
1163 | dma_set_max_seg_size(device->dma_device, SZ_2G); | |
1164 | ||
548cb4fb | 1165 | } |
1da177e4 | 1166 | |
921eab11 JG |
1167 | /* |
1168 | * setup_device() allocates memory and sets up data that requires calling the | |
1169 | * device ops, this is the only reason these actions are not done during | |
1170 | * ib_alloc_device. It is undone by ib_dealloc_device(). | |
1171 | */ | |
548cb4fb PP |
1172 | static int setup_device(struct ib_device *device) |
1173 | { | |
1174 | struct ib_udata uhw = {.outlen = 0, .inlen = 0}; | |
1175 | int ret; | |
1da177e4 | 1176 | |
921eab11 JG |
1177 | setup_dma_device(device); |
1178 | ||
548cb4fb PP |
1179 | ret = ib_device_check_mandatory(device); |
1180 | if (ret) | |
1181 | return ret; | |
1da177e4 | 1182 | |
8ceb1357 | 1183 | ret = setup_port_data(device); |
5eb620c8 | 1184 | if (ret) { |
8ceb1357 | 1185 | dev_warn(&device->dev, "Couldn't create per-port data\n"); |
548cb4fb PP |
1186 | return ret; |
1187 | } | |
1188 | ||
1189 | memset(&device->attrs, 0, sizeof(device->attrs)); | |
3023a1e9 | 1190 | ret = device->ops.query_device(device, &device->attrs, &uhw); |
548cb4fb PP |
1191 | if (ret) { |
1192 | dev_warn(&device->dev, | |
1193 | "Couldn't query the device attributes\n"); | |
d45f89d5 | 1194 | return ret; |
5eb620c8 YE |
1195 | } |
1196 | ||
d45f89d5 | 1197 | return 0; |
548cb4fb PP |
1198 | } |
1199 | ||
921eab11 JG |
1200 | static void disable_device(struct ib_device *device) |
1201 | { | |
1202 | struct ib_client *client; | |
1203 | ||
1204 | WARN_ON(!refcount_read(&device->refcount)); | |
1205 | ||
1206 | down_write(&devices_rwsem); | |
1207 | xa_clear_mark(&devices, device->index, DEVICE_REGISTERED); | |
1208 | up_write(&devices_rwsem); | |
1209 | ||
1210 | down_read(&clients_rwsem); | |
1211 | list_for_each_entry_reverse(client, &client_list, list) | |
1212 | remove_client_context(device, client->client_id); | |
1213 | up_read(&clients_rwsem); | |
1214 | ||
1215 | /* Pairs with refcount_set in enable_device */ | |
1216 | ib_device_put(device); | |
1217 | wait_for_completion(&device->unreg_completion); | |
c2261dd7 | 1218 | |
4e0f7b90 PP |
1219 | /* |
1220 | * compat devices must be removed after device refcount drops to zero. | |
1221 | * Otherwise init_net() may add more compatdevs after removing compat | |
1222 | * devices and before device is disabled. | |
1223 | */ | |
1224 | remove_compat_devs(device); | |
921eab11 JG |
1225 | } |
1226 | ||
1227 | /* | |
1228 | * An enabled device is visible to all clients and to all the public facing | |
d0899892 JG |
1229 | * APIs that return a device pointer. This always returns with a new get, even |
1230 | * if it fails. | |
921eab11 | 1231 | */ |
d0899892 | 1232 | static int enable_device_and_get(struct ib_device *device) |
921eab11 JG |
1233 | { |
1234 | struct ib_client *client; | |
1235 | unsigned long index; | |
d0899892 | 1236 | int ret = 0; |
921eab11 | 1237 | |
d0899892 JG |
1238 | /* |
1239 | * One ref belongs to the xa and the other belongs to this | |
1240 | * thread. This is needed to guard against parallel unregistration. | |
1241 | */ | |
1242 | refcount_set(&device->refcount, 2); | |
921eab11 JG |
1243 | down_write(&devices_rwsem); |
1244 | xa_set_mark(&devices, device->index, DEVICE_REGISTERED); | |
d0899892 JG |
1245 | |
1246 | /* | |
1247 | * By using downgrade_write() we ensure that no other thread can clear | |
1248 | * DEVICE_REGISTERED while we are completing the client setup. | |
1249 | */ | |
1250 | downgrade_write(&devices_rwsem); | |
921eab11 | 1251 | |
ca22354b JG |
1252 | if (device->ops.enable_driver) { |
1253 | ret = device->ops.enable_driver(device); | |
1254 | if (ret) | |
1255 | goto out; | |
1256 | } | |
1257 | ||
921eab11 JG |
1258 | down_read(&clients_rwsem); |
1259 | xa_for_each_marked (&clients, index, client, CLIENT_REGISTERED) { | |
1260 | ret = add_client_context(device, client); | |
d0899892 JG |
1261 | if (ret) |
1262 | break; | |
921eab11 JG |
1263 | } |
1264 | up_read(&clients_rwsem); | |
4e0f7b90 PP |
1265 | if (!ret) |
1266 | ret = add_compat_devs(device); | |
ca22354b | 1267 | out: |
d0899892 JG |
1268 | up_read(&devices_rwsem); |
1269 | return ret; | |
921eab11 JG |
1270 | } |
1271 | ||
548cb4fb PP |
1272 | /** |
1273 | * ib_register_device - Register an IB device with IB core | |
1274 | * @device:Device to register | |
1275 | * | |
1276 | * Low-level drivers use ib_register_device() to register their | |
1277 | * devices with the IB core. All registered clients will receive a | |
1278 | * callback for each device that is added. @device must be allocated | |
1279 | * with ib_alloc_device(). | |
d0899892 JG |
1280 | * |
1281 | * If the driver uses ops.dealloc_driver and calls any ib_unregister_device() | |
1282 | * asynchronously then the device pointer may become freed as soon as this | |
1283 | * function returns. | |
548cb4fb | 1284 | */ |
ea4baf7f | 1285 | int ib_register_device(struct ib_device *device, const char *name) |
548cb4fb PP |
1286 | { |
1287 | int ret; | |
548cb4fb | 1288 | |
0df91bb6 JG |
1289 | ret = assign_name(device, name); |
1290 | if (ret) | |
921eab11 | 1291 | return ret; |
548cb4fb PP |
1292 | |
1293 | ret = setup_device(device); | |
1294 | if (ret) | |
d0899892 | 1295 | return ret; |
03db3a2d | 1296 | |
d45f89d5 JG |
1297 | ret = ib_cache_setup_one(device); |
1298 | if (ret) { | |
1299 | dev_warn(&device->dev, | |
1300 | "Couldn't set up InfiniBand P_Key/GID cache\n"); | |
d0899892 | 1301 | return ret; |
d45f89d5 JG |
1302 | } |
1303 | ||
7527a7b1 | 1304 | ib_device_register_rdmacg(device); |
3e153a93 | 1305 | |
e7a5b4aa LR |
1306 | /* |
1307 | * Ensure that ADD uevent is not fired because it | |
1308 | * is too early amd device is not initialized yet. | |
1309 | */ | |
1310 | dev_set_uevent_suppress(&device->dev, true); | |
5f8f5499 PP |
1311 | ret = device_add(&device->dev); |
1312 | if (ret) | |
1313 | goto cg_cleanup; | |
1314 | ||
ea4baf7f | 1315 | ret = ib_device_register_sysfs(device); |
1da177e4 | 1316 | if (ret) { |
43c7c851 JG |
1317 | dev_warn(&device->dev, |
1318 | "Couldn't register device with driver model\n"); | |
5f8f5499 | 1319 | goto dev_cleanup; |
1da177e4 LT |
1320 | } |
1321 | ||
d0899892 | 1322 | ret = enable_device_and_get(device); |
e7a5b4aa LR |
1323 | dev_set_uevent_suppress(&device->dev, false); |
1324 | /* Mark for userspace that device is ready */ | |
1325 | kobject_uevent(&device->dev.kobj, KOBJ_ADD); | |
d0899892 JG |
1326 | if (ret) { |
1327 | void (*dealloc_fn)(struct ib_device *); | |
1328 | ||
1329 | /* | |
1330 | * If we hit this error flow then we don't want to | |
1331 | * automatically dealloc the device since the caller is | |
1332 | * expected to call ib_dealloc_device() after | |
1333 | * ib_register_device() fails. This is tricky due to the | |
1334 | * possibility for a parallel unregistration along with this | |
1335 | * error flow. Since we have a refcount here we know any | |
1336 | * parallel flow is stopped in disable_device and will see the | |
1337 | * NULL pointers, causing the responsibility to | |
1338 | * ib_dealloc_device() to revert back to this thread. | |
1339 | */ | |
1340 | dealloc_fn = device->ops.dealloc_driver; | |
1341 | device->ops.dealloc_driver = NULL; | |
1342 | ib_device_put(device); | |
1343 | __ib_unregister_device(device); | |
1344 | device->ops.dealloc_driver = dealloc_fn; | |
1345 | return ret; | |
1346 | } | |
1347 | ib_device_put(device); | |
1da177e4 | 1348 | |
4be3a4fa PP |
1349 | return 0; |
1350 | ||
5f8f5499 PP |
1351 | dev_cleanup: |
1352 | device_del(&device->dev); | |
2fb4f4ea | 1353 | cg_cleanup: |
e7a5b4aa | 1354 | dev_set_uevent_suppress(&device->dev, false); |
2fb4f4ea | 1355 | ib_device_unregister_rdmacg(device); |
d45f89d5 | 1356 | ib_cache_cleanup_one(device); |
1da177e4 LT |
1357 | return ret; |
1358 | } | |
1359 | EXPORT_SYMBOL(ib_register_device); | |
1360 | ||
d0899892 JG |
1361 | /* Callers must hold a get on the device. */ |
1362 | static void __ib_unregister_device(struct ib_device *ib_dev) | |
1363 | { | |
1364 | /* | |
1365 | * We have a registration lock so that all the calls to unregister are | |
1366 | * fully fenced, once any unregister returns the device is truely | |
1367 | * unregistered even if multiple callers are unregistering it at the | |
1368 | * same time. This also interacts with the registration flow and | |
1369 | * provides sane semantics if register and unregister are racing. | |
1370 | */ | |
1371 | mutex_lock(&ib_dev->unregistration_lock); | |
1372 | if (!refcount_read(&ib_dev->refcount)) | |
1373 | goto out; | |
1374 | ||
1375 | disable_device(ib_dev); | |
3042492b PP |
1376 | |
1377 | /* Expedite removing unregistered pointers from the hash table */ | |
1378 | free_netdevs(ib_dev); | |
1379 | ||
d0899892 JG |
1380 | ib_device_unregister_sysfs(ib_dev); |
1381 | device_del(&ib_dev->dev); | |
1382 | ib_device_unregister_rdmacg(ib_dev); | |
1383 | ib_cache_cleanup_one(ib_dev); | |
1384 | ||
1385 | /* | |
1386 | * Drivers using the new flow may not call ib_dealloc_device except | |
1387 | * in error unwind prior to registration success. | |
1388 | */ | |
1389 | if (ib_dev->ops.dealloc_driver) { | |
1390 | WARN_ON(kref_read(&ib_dev->dev.kobj.kref) <= 1); | |
1391 | ib_dealloc_device(ib_dev); | |
1392 | } | |
1393 | out: | |
1394 | mutex_unlock(&ib_dev->unregistration_lock); | |
1395 | } | |
1396 | ||
1da177e4 LT |
1397 | /** |
1398 | * ib_unregister_device - Unregister an IB device | |
d0899892 | 1399 | * @device: The device to unregister |
1da177e4 LT |
1400 | * |
1401 | * Unregister an IB device. All clients will receive a remove callback. | |
d0899892 JG |
1402 | * |
1403 | * Callers should call this routine only once, and protect against races with | |
1404 | * registration. Typically it should only be called as part of a remove | |
1405 | * callback in an implementation of driver core's struct device_driver and | |
1406 | * related. | |
1407 | * | |
1408 | * If ops.dealloc_driver is used then ib_dev will be freed upon return from | |
1409 | * this function. | |
1da177e4 | 1410 | */ |
d0899892 | 1411 | void ib_unregister_device(struct ib_device *ib_dev) |
1da177e4 | 1412 | { |
d0899892 JG |
1413 | get_device(&ib_dev->dev); |
1414 | __ib_unregister_device(ib_dev); | |
1415 | put_device(&ib_dev->dev); | |
1da177e4 LT |
1416 | } |
1417 | EXPORT_SYMBOL(ib_unregister_device); | |
1418 | ||
d0899892 JG |
1419 | /** |
1420 | * ib_unregister_device_and_put - Unregister a device while holding a 'get' | |
1421 | * device: The device to unregister | |
1422 | * | |
1423 | * This is the same as ib_unregister_device(), except it includes an internal | |
1424 | * ib_device_put() that should match a 'get' obtained by the caller. | |
1425 | * | |
1426 | * It is safe to call this routine concurrently from multiple threads while | |
1427 | * holding the 'get'. When the function returns the device is fully | |
1428 | * unregistered. | |
1429 | * | |
1430 | * Drivers using this flow MUST use the driver_unregister callback to clean up | |
1431 | * their resources associated with the device and dealloc it. | |
1432 | */ | |
1433 | void ib_unregister_device_and_put(struct ib_device *ib_dev) | |
1434 | { | |
1435 | WARN_ON(!ib_dev->ops.dealloc_driver); | |
1436 | get_device(&ib_dev->dev); | |
1437 | ib_device_put(ib_dev); | |
1438 | __ib_unregister_device(ib_dev); | |
1439 | put_device(&ib_dev->dev); | |
1440 | } | |
1441 | EXPORT_SYMBOL(ib_unregister_device_and_put); | |
1442 | ||
1443 | /** | |
1444 | * ib_unregister_driver - Unregister all IB devices for a driver | |
1445 | * @driver_id: The driver to unregister | |
1446 | * | |
1447 | * This implements a fence for device unregistration. It only returns once all | |
1448 | * devices associated with the driver_id have fully completed their | |
1449 | * unregistration and returned from ib_unregister_device*(). | |
1450 | * | |
1451 | * If device's are not yet unregistered it goes ahead and starts unregistering | |
1452 | * them. | |
1453 | * | |
1454 | * This does not block creation of new devices with the given driver_id, that | |
1455 | * is the responsibility of the caller. | |
1456 | */ | |
1457 | void ib_unregister_driver(enum rdma_driver_id driver_id) | |
1458 | { | |
1459 | struct ib_device *ib_dev; | |
1460 | unsigned long index; | |
1461 | ||
1462 | down_read(&devices_rwsem); | |
1463 | xa_for_each (&devices, index, ib_dev) { | |
1464 | if (ib_dev->driver_id != driver_id) | |
1465 | continue; | |
1466 | ||
1467 | get_device(&ib_dev->dev); | |
1468 | up_read(&devices_rwsem); | |
1469 | ||
1470 | WARN_ON(!ib_dev->ops.dealloc_driver); | |
1471 | __ib_unregister_device(ib_dev); | |
1472 | ||
1473 | put_device(&ib_dev->dev); | |
1474 | down_read(&devices_rwsem); | |
1475 | } | |
1476 | up_read(&devices_rwsem); | |
1477 | } | |
1478 | EXPORT_SYMBOL(ib_unregister_driver); | |
1479 | ||
1480 | static void ib_unregister_work(struct work_struct *work) | |
1481 | { | |
1482 | struct ib_device *ib_dev = | |
1483 | container_of(work, struct ib_device, unregistration_work); | |
1484 | ||
1485 | __ib_unregister_device(ib_dev); | |
1486 | put_device(&ib_dev->dev); | |
1487 | } | |
1488 | ||
1489 | /** | |
1490 | * ib_unregister_device_queued - Unregister a device using a work queue | |
1491 | * device: The device to unregister | |
1492 | * | |
1493 | * This schedules an asynchronous unregistration using a WQ for the device. A | |
1494 | * driver should use this to avoid holding locks while doing unregistration, | |
1495 | * such as holding the RTNL lock. | |
1496 | * | |
1497 | * Drivers using this API must use ib_unregister_driver before module unload | |
1498 | * to ensure that all scheduled unregistrations have completed. | |
1499 | */ | |
1500 | void ib_unregister_device_queued(struct ib_device *ib_dev) | |
1501 | { | |
1502 | WARN_ON(!refcount_read(&ib_dev->refcount)); | |
1503 | WARN_ON(!ib_dev->ops.dealloc_driver); | |
1504 | get_device(&ib_dev->dev); | |
1505 | if (!queue_work(system_unbound_wq, &ib_dev->unregistration_work)) | |
1506 | put_device(&ib_dev->dev); | |
1507 | } | |
1508 | EXPORT_SYMBOL(ib_unregister_device_queued); | |
1509 | ||
decbc7a6 PP |
1510 | /* |
1511 | * The caller must pass in a device that has the kref held and the refcount | |
1512 | * released. If the device is in cur_net and still registered then it is moved | |
1513 | * into net. | |
1514 | */ | |
1515 | static int rdma_dev_change_netns(struct ib_device *device, struct net *cur_net, | |
1516 | struct net *net) | |
1517 | { | |
1518 | int ret2 = -EINVAL; | |
1519 | int ret; | |
1520 | ||
1521 | mutex_lock(&device->unregistration_lock); | |
1522 | ||
1523 | /* | |
2e5b8a01 PP |
1524 | * If a device not under ib_device_get() or if the unregistration_lock |
1525 | * is not held, the namespace can be changed, or it can be unregistered. | |
1526 | * Check again under the lock. | |
decbc7a6 PP |
1527 | */ |
1528 | if (refcount_read(&device->refcount) == 0 || | |
1529 | !net_eq(cur_net, read_pnet(&device->coredev.rdma_net))) { | |
1530 | ret = -ENODEV; | |
1531 | goto out; | |
1532 | } | |
1533 | ||
1534 | kobject_uevent(&device->dev.kobj, KOBJ_REMOVE); | |
1535 | disable_device(device); | |
1536 | ||
1537 | /* | |
1538 | * At this point no one can be using the device, so it is safe to | |
1539 | * change the namespace. | |
1540 | */ | |
1541 | write_pnet(&device->coredev.rdma_net, net); | |
1542 | ||
2e5b8a01 | 1543 | down_read(&devices_rwsem); |
decbc7a6 PP |
1544 | /* |
1545 | * Currently rdma devices are system wide unique. So the device name | |
1546 | * is guaranteed free in the new namespace. Publish the new namespace | |
1547 | * at the sysfs level. | |
1548 | */ | |
decbc7a6 PP |
1549 | ret = device_rename(&device->dev, dev_name(&device->dev)); |
1550 | up_read(&devices_rwsem); | |
1551 | if (ret) { | |
1552 | dev_warn(&device->dev, | |
1553 | "%s: Couldn't rename device after namespace change\n", | |
1554 | __func__); | |
1555 | /* Try and put things back and re-enable the device */ | |
1556 | write_pnet(&device->coredev.rdma_net, cur_net); | |
1557 | } | |
1558 | ||
1559 | ret2 = enable_device_and_get(device); | |
2e5b8a01 | 1560 | if (ret2) { |
decbc7a6 PP |
1561 | /* |
1562 | * This shouldn't really happen, but if it does, let the user | |
1563 | * retry at later point. So don't disable the device. | |
1564 | */ | |
1565 | dev_warn(&device->dev, | |
1566 | "%s: Couldn't re-enable device after namespace change\n", | |
1567 | __func__); | |
2e5b8a01 | 1568 | } |
decbc7a6 | 1569 | kobject_uevent(&device->dev.kobj, KOBJ_ADD); |
2e5b8a01 | 1570 | |
decbc7a6 PP |
1571 | ib_device_put(device); |
1572 | out: | |
1573 | mutex_unlock(&device->unregistration_lock); | |
1574 | if (ret) | |
1575 | return ret; | |
1576 | return ret2; | |
1577 | } | |
1578 | ||
2e5b8a01 PP |
1579 | int ib_device_set_netns_put(struct sk_buff *skb, |
1580 | struct ib_device *dev, u32 ns_fd) | |
1581 | { | |
1582 | struct net *net; | |
1583 | int ret; | |
1584 | ||
1585 | net = get_net_ns_by_fd(ns_fd); | |
1586 | if (IS_ERR(net)) { | |
1587 | ret = PTR_ERR(net); | |
1588 | goto net_err; | |
1589 | } | |
1590 | ||
1591 | if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) { | |
1592 | ret = -EPERM; | |
1593 | goto ns_err; | |
1594 | } | |
1595 | ||
1596 | /* | |
1597 | * Currently supported only for those providers which support | |
1598 | * disassociation and don't do port specific sysfs init. Once a | |
1599 | * port_cleanup infrastructure is implemented, this limitation will be | |
1600 | * removed. | |
1601 | */ | |
1602 | if (!dev->ops.disassociate_ucontext || dev->ops.init_port || | |
1603 | ib_devices_shared_netns) { | |
1604 | ret = -EOPNOTSUPP; | |
1605 | goto ns_err; | |
1606 | } | |
1607 | ||
1608 | get_device(&dev->dev); | |
1609 | ib_device_put(dev); | |
1610 | ret = rdma_dev_change_netns(dev, current->nsproxy->net_ns, net); | |
1611 | put_device(&dev->dev); | |
1612 | ||
1613 | put_net(net); | |
1614 | return ret; | |
1615 | ||
1616 | ns_err: | |
1617 | put_net(net); | |
1618 | net_err: | |
1619 | ib_device_put(dev); | |
1620 | return ret; | |
1621 | } | |
1622 | ||
4e0f7b90 PP |
1623 | static struct pernet_operations rdma_dev_net_ops = { |
1624 | .init = rdma_dev_init_net, | |
1625 | .exit = rdma_dev_exit_net, | |
1626 | .id = &rdma_dev_net_id, | |
1627 | .size = sizeof(struct rdma_dev_net), | |
1628 | }; | |
1629 | ||
e59178d8 JG |
1630 | static int assign_client_id(struct ib_client *client) |
1631 | { | |
1632 | int ret; | |
1633 | ||
921eab11 | 1634 | down_write(&clients_rwsem); |
e59178d8 JG |
1635 | /* |
1636 | * The add/remove callbacks must be called in FIFO/LIFO order. To | |
1637 | * achieve this we assign client_ids so they are sorted in | |
1638 | * registration order, and retain a linked list we can reverse iterate | |
1639 | * to get the LIFO order. The extra linked list can go away if xarray | |
1640 | * learns to reverse iterate. | |
1641 | */ | |
ea295481 | 1642 | if (list_empty(&client_list)) { |
e59178d8 | 1643 | client->client_id = 0; |
ea295481 LT |
1644 | } else { |
1645 | struct ib_client *last; | |
1646 | ||
1647 | last = list_last_entry(&client_list, struct ib_client, list); | |
1648 | client->client_id = last->client_id + 1; | |
4512acd0 | 1649 | } |
ea295481 | 1650 | ret = xa_insert(&clients, client->client_id, client, GFP_KERNEL); |
e59178d8 JG |
1651 | if (ret) |
1652 | goto out; | |
1653 | ||
921eab11 JG |
1654 | xa_set_mark(&clients, client->client_id, CLIENT_REGISTERED); |
1655 | list_add_tail(&client->list, &client_list); | |
1656 | ||
e59178d8 | 1657 | out: |
921eab11 | 1658 | up_write(&clients_rwsem); |
e59178d8 JG |
1659 | return ret; |
1660 | } | |
1661 | ||
1da177e4 LT |
1662 | /** |
1663 | * ib_register_client - Register an IB client | |
1664 | * @client:Client to register | |
1665 | * | |
1666 | * Upper level users of the IB drivers can use ib_register_client() to | |
1667 | * register callbacks for IB device addition and removal. When an IB | |
1668 | * device is added, each registered client's add method will be called | |
1669 | * (in the order the clients were registered), and when a device is | |
1670 | * removed, each client's remove method will be called (in the reverse | |
1671 | * order that clients were registered). In addition, when | |
1672 | * ib_register_client() is called, the client will receive an add | |
1673 | * callback for all devices already registered. | |
1674 | */ | |
1675 | int ib_register_client(struct ib_client *client) | |
1676 | { | |
1677 | struct ib_device *device; | |
0df91bb6 | 1678 | unsigned long index; |
e59178d8 | 1679 | int ret; |
1da177e4 | 1680 | |
e59178d8 | 1681 | ret = assign_client_id(client); |
921eab11 | 1682 | if (ret) |
e59178d8 | 1683 | return ret; |
1da177e4 | 1684 | |
921eab11 JG |
1685 | down_read(&devices_rwsem); |
1686 | xa_for_each_marked (&devices, index, device, DEVICE_REGISTERED) { | |
1687 | ret = add_client_context(device, client); | |
1688 | if (ret) { | |
1689 | up_read(&devices_rwsem); | |
1690 | ib_unregister_client(client); | |
1691 | return ret; | |
1692 | } | |
1693 | } | |
1694 | up_read(&devices_rwsem); | |
1da177e4 LT |
1695 | return 0; |
1696 | } | |
1697 | EXPORT_SYMBOL(ib_register_client); | |
1698 | ||
1699 | /** | |
1700 | * ib_unregister_client - Unregister an IB client | |
1701 | * @client:Client to unregister | |
1702 | * | |
1703 | * Upper level users use ib_unregister_client() to remove their client | |
1704 | * registration. When ib_unregister_client() is called, the client | |
1705 | * will receive a remove callback for each IB device still registered. | |
921eab11 JG |
1706 | * |
1707 | * This is a full fence, once it returns no client callbacks will be called, | |
1708 | * or are running in another thread. | |
1da177e4 LT |
1709 | */ |
1710 | void ib_unregister_client(struct ib_client *client) | |
1711 | { | |
1da177e4 | 1712 | struct ib_device *device; |
0df91bb6 | 1713 | unsigned long index; |
1da177e4 | 1714 | |
921eab11 | 1715 | down_write(&clients_rwsem); |
e59178d8 | 1716 | xa_clear_mark(&clients, client->client_id, CLIENT_REGISTERED); |
921eab11 JG |
1717 | up_write(&clients_rwsem); |
1718 | /* | |
1719 | * Every device still known must be serialized to make sure we are | |
1720 | * done with the client callbacks before we return. | |
1721 | */ | |
1722 | down_read(&devices_rwsem); | |
1723 | xa_for_each (&devices, index, device) | |
1724 | remove_client_context(device, client->client_id); | |
1725 | up_read(&devices_rwsem); | |
1da177e4 | 1726 | |
921eab11 | 1727 | down_write(&clients_rwsem); |
e59178d8 JG |
1728 | list_del(&client->list); |
1729 | xa_erase(&clients, client->client_id); | |
921eab11 | 1730 | up_write(&clients_rwsem); |
1da177e4 LT |
1731 | } |
1732 | EXPORT_SYMBOL(ib_unregister_client); | |
1733 | ||
1da177e4 | 1734 | /** |
9cd330d3 | 1735 | * ib_set_client_data - Set IB client context |
1da177e4 LT |
1736 | * @device:Device to set context for |
1737 | * @client:Client to set context for | |
1738 | * @data:Context to set | |
1739 | * | |
0df91bb6 JG |
1740 | * ib_set_client_data() sets client context data that can be retrieved with |
1741 | * ib_get_client_data(). This can only be called while the client is | |
1742 | * registered to the device, once the ib_client remove() callback returns this | |
1743 | * cannot be called. | |
1da177e4 LT |
1744 | */ |
1745 | void ib_set_client_data(struct ib_device *device, struct ib_client *client, | |
1746 | void *data) | |
1747 | { | |
0df91bb6 | 1748 | void *rc; |
1da177e4 | 1749 | |
0df91bb6 JG |
1750 | if (WARN_ON(IS_ERR(data))) |
1751 | data = NULL; | |
1da177e4 | 1752 | |
0df91bb6 JG |
1753 | rc = xa_store(&device->client_data, client->client_id, data, |
1754 | GFP_KERNEL); | |
1755 | WARN_ON(xa_is_err(rc)); | |
1da177e4 LT |
1756 | } |
1757 | EXPORT_SYMBOL(ib_set_client_data); | |
1758 | ||
1759 | /** | |
1760 | * ib_register_event_handler - Register an IB event handler | |
1761 | * @event_handler:Handler to register | |
1762 | * | |
1763 | * ib_register_event_handler() registers an event handler that will be | |
1764 | * called back when asynchronous IB events occur (as defined in | |
1765 | * chapter 11 of the InfiniBand Architecture Specification). This | |
1766 | * callback may occur in interrupt context. | |
1767 | */ | |
dcc9881e | 1768 | void ib_register_event_handler(struct ib_event_handler *event_handler) |
1da177e4 LT |
1769 | { |
1770 | unsigned long flags; | |
1771 | ||
1772 | spin_lock_irqsave(&event_handler->device->event_handler_lock, flags); | |
1773 | list_add_tail(&event_handler->list, | |
1774 | &event_handler->device->event_handler_list); | |
1775 | spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags); | |
1da177e4 LT |
1776 | } |
1777 | EXPORT_SYMBOL(ib_register_event_handler); | |
1778 | ||
1779 | /** | |
1780 | * ib_unregister_event_handler - Unregister an event handler | |
1781 | * @event_handler:Handler to unregister | |
1782 | * | |
1783 | * Unregister an event handler registered with | |
1784 | * ib_register_event_handler(). | |
1785 | */ | |
dcc9881e | 1786 | void ib_unregister_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_del(&event_handler->list); | |
1792 | spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags); | |
1da177e4 LT |
1793 | } |
1794 | EXPORT_SYMBOL(ib_unregister_event_handler); | |
1795 | ||
1796 | /** | |
1797 | * ib_dispatch_event - Dispatch an asynchronous event | |
1798 | * @event:Event to dispatch | |
1799 | * | |
1800 | * Low-level drivers must call ib_dispatch_event() to dispatch the | |
1801 | * event to all registered event handlers when an asynchronous event | |
1802 | * occurs. | |
1803 | */ | |
1804 | void ib_dispatch_event(struct ib_event *event) | |
1805 | { | |
1806 | unsigned long flags; | |
1807 | struct ib_event_handler *handler; | |
1808 | ||
1809 | spin_lock_irqsave(&event->device->event_handler_lock, flags); | |
1810 | ||
1811 | list_for_each_entry(handler, &event->device->event_handler_list, list) | |
1812 | handler->handler(handler, event); | |
1813 | ||
1814 | spin_unlock_irqrestore(&event->device->event_handler_lock, flags); | |
1815 | } | |
1816 | EXPORT_SYMBOL(ib_dispatch_event); | |
1817 | ||
1da177e4 LT |
1818 | /** |
1819 | * ib_query_port - Query IB port attributes | |
1820 | * @device:Device to query | |
1821 | * @port_num:Port number to query | |
1822 | * @port_attr:Port attributes | |
1823 | * | |
1824 | * ib_query_port() returns the attributes of a port through the | |
1825 | * @port_attr pointer. | |
1826 | */ | |
1827 | int ib_query_port(struct ib_device *device, | |
1828 | u8 port_num, | |
1829 | struct ib_port_attr *port_attr) | |
1830 | { | |
fad61ad4 EC |
1831 | union ib_gid gid; |
1832 | int err; | |
1833 | ||
24dc831b | 1834 | if (!rdma_is_port_valid(device, port_num)) |
116c0074 RD |
1835 | return -EINVAL; |
1836 | ||
fad61ad4 | 1837 | memset(port_attr, 0, sizeof(*port_attr)); |
3023a1e9 | 1838 | err = device->ops.query_port(device, port_num, port_attr); |
fad61ad4 EC |
1839 | if (err || port_attr->subnet_prefix) |
1840 | return err; | |
1841 | ||
d7012467 EC |
1842 | if (rdma_port_get_link_layer(device, port_num) != IB_LINK_LAYER_INFINIBAND) |
1843 | return 0; | |
1844 | ||
3023a1e9 | 1845 | err = device->ops.query_gid(device, port_num, 0, &gid); |
fad61ad4 EC |
1846 | if (err) |
1847 | return err; | |
1848 | ||
1849 | port_attr->subnet_prefix = be64_to_cpu(gid.global.subnet_prefix); | |
1850 | return 0; | |
1da177e4 LT |
1851 | } |
1852 | EXPORT_SYMBOL(ib_query_port); | |
1853 | ||
324e227e JG |
1854 | static void add_ndev_hash(struct ib_port_data *pdata) |
1855 | { | |
1856 | unsigned long flags; | |
1857 | ||
1858 | might_sleep(); | |
1859 | ||
1860 | spin_lock_irqsave(&ndev_hash_lock, flags); | |
1861 | if (hash_hashed(&pdata->ndev_hash_link)) { | |
1862 | hash_del_rcu(&pdata->ndev_hash_link); | |
1863 | spin_unlock_irqrestore(&ndev_hash_lock, flags); | |
1864 | /* | |
1865 | * We cannot do hash_add_rcu after a hash_del_rcu until the | |
1866 | * grace period | |
1867 | */ | |
1868 | synchronize_rcu(); | |
1869 | spin_lock_irqsave(&ndev_hash_lock, flags); | |
1870 | } | |
1871 | if (pdata->netdev) | |
1872 | hash_add_rcu(ndev_hash, &pdata->ndev_hash_link, | |
1873 | (uintptr_t)pdata->netdev); | |
1874 | spin_unlock_irqrestore(&ndev_hash_lock, flags); | |
1875 | } | |
1876 | ||
c2261dd7 JG |
1877 | /** |
1878 | * ib_device_set_netdev - Associate the ib_dev with an underlying net_device | |
1879 | * @ib_dev: Device to modify | |
1880 | * @ndev: net_device to affiliate, may be NULL | |
1881 | * @port: IB port the net_device is connected to | |
1882 | * | |
1883 | * Drivers should use this to link the ib_device to a netdev so the netdev | |
1884 | * shows up in interfaces like ib_enum_roce_netdev. Only one netdev may be | |
1885 | * affiliated with any port. | |
1886 | * | |
1887 | * The caller must ensure that the given ndev is not unregistered or | |
1888 | * unregistering, and that either the ib_device is unregistered or | |
1889 | * ib_device_set_netdev() is called with NULL when the ndev sends a | |
1890 | * NETDEV_UNREGISTER event. | |
1891 | */ | |
1892 | int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev, | |
1893 | unsigned int port) | |
1894 | { | |
1895 | struct net_device *old_ndev; | |
1896 | struct ib_port_data *pdata; | |
1897 | unsigned long flags; | |
1898 | int ret; | |
1899 | ||
1900 | /* | |
1901 | * Drivers wish to call this before ib_register_driver, so we have to | |
1902 | * setup the port data early. | |
1903 | */ | |
1904 | ret = alloc_port_data(ib_dev); | |
1905 | if (ret) | |
1906 | return ret; | |
1907 | ||
1908 | if (!rdma_is_port_valid(ib_dev, port)) | |
1909 | return -EINVAL; | |
1910 | ||
1911 | pdata = &ib_dev->port_data[port]; | |
1912 | spin_lock_irqsave(&pdata->netdev_lock, flags); | |
324e227e JG |
1913 | old_ndev = rcu_dereference_protected( |
1914 | pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); | |
1915 | if (old_ndev == ndev) { | |
c2261dd7 JG |
1916 | spin_unlock_irqrestore(&pdata->netdev_lock, flags); |
1917 | return 0; | |
1918 | } | |
c2261dd7 JG |
1919 | |
1920 | if (ndev) | |
1921 | dev_hold(ndev); | |
324e227e | 1922 | rcu_assign_pointer(pdata->netdev, ndev); |
c2261dd7 JG |
1923 | spin_unlock_irqrestore(&pdata->netdev_lock, flags); |
1924 | ||
324e227e | 1925 | add_ndev_hash(pdata); |
c2261dd7 JG |
1926 | if (old_ndev) |
1927 | dev_put(old_ndev); | |
1928 | ||
1929 | return 0; | |
1930 | } | |
1931 | EXPORT_SYMBOL(ib_device_set_netdev); | |
1932 | ||
1933 | static void free_netdevs(struct ib_device *ib_dev) | |
1934 | { | |
1935 | unsigned long flags; | |
1936 | unsigned int port; | |
1937 | ||
1938 | rdma_for_each_port (ib_dev, port) { | |
1939 | struct ib_port_data *pdata = &ib_dev->port_data[port]; | |
324e227e | 1940 | struct net_device *ndev; |
c2261dd7 JG |
1941 | |
1942 | spin_lock_irqsave(&pdata->netdev_lock, flags); | |
324e227e JG |
1943 | ndev = rcu_dereference_protected( |
1944 | pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); | |
1945 | if (ndev) { | |
1946 | spin_lock(&ndev_hash_lock); | |
1947 | hash_del_rcu(&pdata->ndev_hash_link); | |
1948 | spin_unlock(&ndev_hash_lock); | |
1949 | ||
1950 | /* | |
1951 | * If this is the last dev_put there is still a | |
1952 | * synchronize_rcu before the netdev is kfreed, so we | |
1953 | * can continue to rely on unlocked pointer | |
1954 | * comparisons after the put | |
1955 | */ | |
1956 | rcu_assign_pointer(pdata->netdev, NULL); | |
1957 | dev_put(ndev); | |
c2261dd7 JG |
1958 | } |
1959 | spin_unlock_irqrestore(&pdata->netdev_lock, flags); | |
1960 | } | |
1961 | } | |
1962 | ||
1963 | struct net_device *ib_device_get_netdev(struct ib_device *ib_dev, | |
1964 | unsigned int port) | |
1965 | { | |
1966 | struct ib_port_data *pdata; | |
1967 | struct net_device *res; | |
1968 | ||
1969 | if (!rdma_is_port_valid(ib_dev, port)) | |
1970 | return NULL; | |
1971 | ||
1972 | pdata = &ib_dev->port_data[port]; | |
1973 | ||
1974 | /* | |
1975 | * New drivers should use ib_device_set_netdev() not the legacy | |
1976 | * get_netdev(). | |
1977 | */ | |
1978 | if (ib_dev->ops.get_netdev) | |
1979 | res = ib_dev->ops.get_netdev(ib_dev, port); | |
1980 | else { | |
1981 | spin_lock(&pdata->netdev_lock); | |
324e227e JG |
1982 | res = rcu_dereference_protected( |
1983 | pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); | |
c2261dd7 JG |
1984 | if (res) |
1985 | dev_hold(res); | |
1986 | spin_unlock(&pdata->netdev_lock); | |
1987 | } | |
1988 | ||
1989 | /* | |
1990 | * If we are starting to unregister expedite things by preventing | |
1991 | * propagation of an unregistering netdev. | |
1992 | */ | |
1993 | if (res && res->reg_state != NETREG_REGISTERED) { | |
1994 | dev_put(res); | |
1995 | return NULL; | |
1996 | } | |
1997 | ||
1998 | return res; | |
1999 | } | |
2000 | ||
324e227e JG |
2001 | /** |
2002 | * ib_device_get_by_netdev - Find an IB device associated with a netdev | |
2003 | * @ndev: netdev to locate | |
2004 | * @driver_id: The driver ID that must match (RDMA_DRIVER_UNKNOWN matches all) | |
2005 | * | |
2006 | * Find and hold an ib_device that is associated with a netdev via | |
2007 | * ib_device_set_netdev(). The caller must call ib_device_put() on the | |
2008 | * returned pointer. | |
2009 | */ | |
2010 | struct ib_device *ib_device_get_by_netdev(struct net_device *ndev, | |
2011 | enum rdma_driver_id driver_id) | |
2012 | { | |
2013 | struct ib_device *res = NULL; | |
2014 | struct ib_port_data *cur; | |
2015 | ||
2016 | rcu_read_lock(); | |
2017 | hash_for_each_possible_rcu (ndev_hash, cur, ndev_hash_link, | |
2018 | (uintptr_t)ndev) { | |
2019 | if (rcu_access_pointer(cur->netdev) == ndev && | |
2020 | (driver_id == RDMA_DRIVER_UNKNOWN || | |
2021 | cur->ib_dev->driver_id == driver_id) && | |
2022 | ib_device_try_get(cur->ib_dev)) { | |
2023 | res = cur->ib_dev; | |
2024 | break; | |
2025 | } | |
2026 | } | |
2027 | rcu_read_unlock(); | |
2028 | ||
2029 | return res; | |
2030 | } | |
2031 | EXPORT_SYMBOL(ib_device_get_by_netdev); | |
2032 | ||
03db3a2d MB |
2033 | /** |
2034 | * ib_enum_roce_netdev - enumerate all RoCE ports | |
2035 | * @ib_dev : IB device we want to query | |
2036 | * @filter: Should we call the callback? | |
2037 | * @filter_cookie: Cookie passed to filter | |
2038 | * @cb: Callback to call for each found RoCE ports | |
2039 | * @cookie: Cookie passed back to the callback | |
2040 | * | |
2041 | * Enumerates all of the physical RoCE ports of ib_dev | |
2042 | * which are related to netdevice and calls callback() on each | |
2043 | * device for which filter() function returns non zero. | |
2044 | */ | |
2045 | void ib_enum_roce_netdev(struct ib_device *ib_dev, | |
2046 | roce_netdev_filter filter, | |
2047 | void *filter_cookie, | |
2048 | roce_netdev_callback cb, | |
2049 | void *cookie) | |
2050 | { | |
ea1075ed | 2051 | unsigned int port; |
03db3a2d | 2052 | |
ea1075ed | 2053 | rdma_for_each_port (ib_dev, port) |
03db3a2d | 2054 | if (rdma_protocol_roce(ib_dev, port)) { |
c2261dd7 JG |
2055 | struct net_device *idev = |
2056 | ib_device_get_netdev(ib_dev, port); | |
03db3a2d MB |
2057 | |
2058 | if (filter(ib_dev, port, idev, filter_cookie)) | |
2059 | cb(ib_dev, port, idev, cookie); | |
2060 | ||
2061 | if (idev) | |
2062 | dev_put(idev); | |
2063 | } | |
2064 | } | |
2065 | ||
2066 | /** | |
2067 | * ib_enum_all_roce_netdevs - enumerate all RoCE devices | |
2068 | * @filter: Should we call the callback? | |
2069 | * @filter_cookie: Cookie passed to filter | |
2070 | * @cb: Callback to call for each found RoCE ports | |
2071 | * @cookie: Cookie passed back to the callback | |
2072 | * | |
2073 | * Enumerates all RoCE devices' physical ports which are related | |
2074 | * to netdevices and calls callback() on each device for which | |
2075 | * filter() function returns non zero. | |
2076 | */ | |
2077 | void ib_enum_all_roce_netdevs(roce_netdev_filter filter, | |
2078 | void *filter_cookie, | |
2079 | roce_netdev_callback cb, | |
2080 | void *cookie) | |
2081 | { | |
2082 | struct ib_device *dev; | |
0df91bb6 | 2083 | unsigned long index; |
03db3a2d | 2084 | |
921eab11 | 2085 | down_read(&devices_rwsem); |
0df91bb6 | 2086 | xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) |
03db3a2d | 2087 | ib_enum_roce_netdev(dev, filter, filter_cookie, cb, cookie); |
921eab11 | 2088 | up_read(&devices_rwsem); |
8030c835 LR |
2089 | } |
2090 | ||
2091 | /** | |
2092 | * ib_enum_all_devs - enumerate all ib_devices | |
2093 | * @cb: Callback to call for each found ib_device | |
2094 | * | |
2095 | * Enumerates all ib_devices and calls callback() on each device. | |
2096 | */ | |
2097 | int ib_enum_all_devs(nldev_callback nldev_cb, struct sk_buff *skb, | |
2098 | struct netlink_callback *cb) | |
2099 | { | |
0df91bb6 | 2100 | unsigned long index; |
8030c835 LR |
2101 | struct ib_device *dev; |
2102 | unsigned int idx = 0; | |
2103 | int ret = 0; | |
2104 | ||
921eab11 | 2105 | down_read(&devices_rwsem); |
0df91bb6 | 2106 | xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { |
37eeab55 PP |
2107 | if (!rdma_dev_access_netns(dev, sock_net(skb->sk))) |
2108 | continue; | |
2109 | ||
8030c835 LR |
2110 | ret = nldev_cb(dev, skb, cb, idx); |
2111 | if (ret) | |
2112 | break; | |
2113 | idx++; | |
2114 | } | |
921eab11 | 2115 | up_read(&devices_rwsem); |
8030c835 | 2116 | return ret; |
03db3a2d MB |
2117 | } |
2118 | ||
1da177e4 LT |
2119 | /** |
2120 | * ib_query_pkey - Get P_Key table entry | |
2121 | * @device:Device to query | |
2122 | * @port_num:Port number to query | |
2123 | * @index:P_Key table index to query | |
2124 | * @pkey:Returned P_Key | |
2125 | * | |
2126 | * ib_query_pkey() fetches the specified P_Key table entry. | |
2127 | */ | |
2128 | int ib_query_pkey(struct ib_device *device, | |
2129 | u8 port_num, u16 index, u16 *pkey) | |
2130 | { | |
9af3f5cf YS |
2131 | if (!rdma_is_port_valid(device, port_num)) |
2132 | return -EINVAL; | |
2133 | ||
3023a1e9 | 2134 | return device->ops.query_pkey(device, port_num, index, pkey); |
1da177e4 LT |
2135 | } |
2136 | EXPORT_SYMBOL(ib_query_pkey); | |
2137 | ||
2138 | /** | |
2139 | * ib_modify_device - Change IB device attributes | |
2140 | * @device:Device to modify | |
2141 | * @device_modify_mask:Mask of attributes to change | |
2142 | * @device_modify:New attribute values | |
2143 | * | |
2144 | * ib_modify_device() changes a device's attributes as specified by | |
2145 | * the @device_modify_mask and @device_modify structure. | |
2146 | */ | |
2147 | int ib_modify_device(struct ib_device *device, | |
2148 | int device_modify_mask, | |
2149 | struct ib_device_modify *device_modify) | |
2150 | { | |
3023a1e9 | 2151 | if (!device->ops.modify_device) |
10e1b54b BVA |
2152 | return -ENOSYS; |
2153 | ||
3023a1e9 KH |
2154 | return device->ops.modify_device(device, device_modify_mask, |
2155 | device_modify); | |
1da177e4 LT |
2156 | } |
2157 | EXPORT_SYMBOL(ib_modify_device); | |
2158 | ||
2159 | /** | |
2160 | * ib_modify_port - Modifies the attributes for the specified port. | |
2161 | * @device: The device to modify. | |
2162 | * @port_num: The number of the port to modify. | |
2163 | * @port_modify_mask: Mask used to specify which attributes of the port | |
2164 | * to change. | |
2165 | * @port_modify: New attribute values for the port. | |
2166 | * | |
2167 | * ib_modify_port() changes a port's attributes as specified by the | |
2168 | * @port_modify_mask and @port_modify structure. | |
2169 | */ | |
2170 | int ib_modify_port(struct ib_device *device, | |
2171 | u8 port_num, int port_modify_mask, | |
2172 | struct ib_port_modify *port_modify) | |
2173 | { | |
61e0962d | 2174 | int rc; |
10e1b54b | 2175 | |
24dc831b | 2176 | if (!rdma_is_port_valid(device, port_num)) |
116c0074 RD |
2177 | return -EINVAL; |
2178 | ||
3023a1e9 KH |
2179 | if (device->ops.modify_port) |
2180 | rc = device->ops.modify_port(device, port_num, | |
2181 | port_modify_mask, | |
2182 | port_modify); | |
61e0962d SX |
2183 | else |
2184 | rc = rdma_protocol_roce(device, port_num) ? 0 : -ENOSYS; | |
2185 | return rc; | |
1da177e4 LT |
2186 | } |
2187 | EXPORT_SYMBOL(ib_modify_port); | |
2188 | ||
5eb620c8 YE |
2189 | /** |
2190 | * ib_find_gid - Returns the port number and GID table index where | |
dbb12562 | 2191 | * a specified GID value occurs. Its searches only for IB link layer. |
5eb620c8 YE |
2192 | * @device: The device to query. |
2193 | * @gid: The GID value to search for. | |
2194 | * @port_num: The port number of the device where the GID value was found. | |
2195 | * @index: The index into the GID table where the GID was found. This | |
2196 | * parameter may be NULL. | |
2197 | */ | |
2198 | int ib_find_gid(struct ib_device *device, union ib_gid *gid, | |
b26c4a11 | 2199 | u8 *port_num, u16 *index) |
5eb620c8 YE |
2200 | { |
2201 | union ib_gid tmp_gid; | |
ea1075ed JG |
2202 | unsigned int port; |
2203 | int ret, i; | |
5eb620c8 | 2204 | |
ea1075ed | 2205 | rdma_for_each_port (device, port) { |
22d24f75 | 2206 | if (!rdma_protocol_ib(device, port)) |
b39ffa1d MB |
2207 | continue; |
2208 | ||
8ceb1357 JG |
2209 | for (i = 0; i < device->port_data[port].immutable.gid_tbl_len; |
2210 | ++i) { | |
1dfce294 | 2211 | ret = rdma_query_gid(device, port, i, &tmp_gid); |
5eb620c8 YE |
2212 | if (ret) |
2213 | return ret; | |
2214 | if (!memcmp(&tmp_gid, gid, sizeof *gid)) { | |
2215 | *port_num = port; | |
2216 | if (index) | |
2217 | *index = i; | |
2218 | return 0; | |
2219 | } | |
2220 | } | |
2221 | } | |
2222 | ||
2223 | return -ENOENT; | |
2224 | } | |
2225 | EXPORT_SYMBOL(ib_find_gid); | |
2226 | ||
2227 | /** | |
2228 | * ib_find_pkey - Returns the PKey table index where a specified | |
2229 | * PKey value occurs. | |
2230 | * @device: The device to query. | |
2231 | * @port_num: The port number of the device to search for the PKey. | |
2232 | * @pkey: The PKey value to search for. | |
2233 | * @index: The index into the PKey table where the PKey was found. | |
2234 | */ | |
2235 | int ib_find_pkey(struct ib_device *device, | |
2236 | u8 port_num, u16 pkey, u16 *index) | |
2237 | { | |
2238 | int ret, i; | |
2239 | u16 tmp_pkey; | |
ff7166c4 | 2240 | int partial_ix = -1; |
5eb620c8 | 2241 | |
8ceb1357 JG |
2242 | for (i = 0; i < device->port_data[port_num].immutable.pkey_tbl_len; |
2243 | ++i) { | |
5eb620c8 YE |
2244 | ret = ib_query_pkey(device, port_num, i, &tmp_pkey); |
2245 | if (ret) | |
2246 | return ret; | |
36026ecc | 2247 | if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) { |
ff7166c4 JM |
2248 | /* if there is full-member pkey take it.*/ |
2249 | if (tmp_pkey & 0x8000) { | |
2250 | *index = i; | |
2251 | return 0; | |
2252 | } | |
2253 | if (partial_ix < 0) | |
2254 | partial_ix = i; | |
5eb620c8 YE |
2255 | } |
2256 | } | |
2257 | ||
ff7166c4 JM |
2258 | /*no full-member, if exists take the limited*/ |
2259 | if (partial_ix >= 0) { | |
2260 | *index = partial_ix; | |
2261 | return 0; | |
2262 | } | |
5eb620c8 YE |
2263 | return -ENOENT; |
2264 | } | |
2265 | EXPORT_SYMBOL(ib_find_pkey); | |
2266 | ||
9268f72d YK |
2267 | /** |
2268 | * ib_get_net_dev_by_params() - Return the appropriate net_dev | |
2269 | * for a received CM request | |
2270 | * @dev: An RDMA device on which the request has been received. | |
2271 | * @port: Port number on the RDMA device. | |
2272 | * @pkey: The Pkey the request came on. | |
2273 | * @gid: A GID that the net_dev uses to communicate. | |
2274 | * @addr: Contains the IP address that the request specified as its | |
2275 | * destination. | |
921eab11 | 2276 | * |
9268f72d YK |
2277 | */ |
2278 | struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, | |
2279 | u8 port, | |
2280 | u16 pkey, | |
2281 | const union ib_gid *gid, | |
2282 | const struct sockaddr *addr) | |
2283 | { | |
2284 | struct net_device *net_dev = NULL; | |
0df91bb6 JG |
2285 | unsigned long index; |
2286 | void *client_data; | |
9268f72d YK |
2287 | |
2288 | if (!rdma_protocol_ib(dev, port)) | |
2289 | return NULL; | |
2290 | ||
921eab11 JG |
2291 | /* |
2292 | * Holding the read side guarantees that the client will not become | |
2293 | * unregistered while we are calling get_net_dev_by_params() | |
2294 | */ | |
2295 | down_read(&dev->client_data_rwsem); | |
0df91bb6 JG |
2296 | xan_for_each_marked (&dev->client_data, index, client_data, |
2297 | CLIENT_DATA_REGISTERED) { | |
2298 | struct ib_client *client = xa_load(&clients, index); | |
9268f72d | 2299 | |
0df91bb6 | 2300 | if (!client || !client->get_net_dev_by_params) |
9268f72d YK |
2301 | continue; |
2302 | ||
0df91bb6 JG |
2303 | net_dev = client->get_net_dev_by_params(dev, port, pkey, gid, |
2304 | addr, client_data); | |
2305 | if (net_dev) | |
2306 | break; | |
9268f72d | 2307 | } |
921eab11 | 2308 | up_read(&dev->client_data_rwsem); |
9268f72d YK |
2309 | |
2310 | return net_dev; | |
2311 | } | |
2312 | EXPORT_SYMBOL(ib_get_net_dev_by_params); | |
2313 | ||
521ed0d9 KH |
2314 | void ib_set_device_ops(struct ib_device *dev, const struct ib_device_ops *ops) |
2315 | { | |
3023a1e9 | 2316 | struct ib_device_ops *dev_ops = &dev->ops; |
521ed0d9 KH |
2317 | #define SET_DEVICE_OP(ptr, name) \ |
2318 | do { \ | |
2319 | if (ops->name) \ | |
2320 | if (!((ptr)->name)) \ | |
2321 | (ptr)->name = ops->name; \ | |
2322 | } while (0) | |
2323 | ||
30471d4b LR |
2324 | #define SET_OBJ_SIZE(ptr, name) SET_DEVICE_OP(ptr, size_##name) |
2325 | ||
3023a1e9 | 2326 | SET_DEVICE_OP(dev_ops, add_gid); |
2f1927b0 | 2327 | SET_DEVICE_OP(dev_ops, advise_mr); |
3023a1e9 KH |
2328 | SET_DEVICE_OP(dev_ops, alloc_dm); |
2329 | SET_DEVICE_OP(dev_ops, alloc_fmr); | |
2330 | SET_DEVICE_OP(dev_ops, alloc_hw_stats); | |
2331 | SET_DEVICE_OP(dev_ops, alloc_mr); | |
2332 | SET_DEVICE_OP(dev_ops, alloc_mw); | |
2333 | SET_DEVICE_OP(dev_ops, alloc_pd); | |
2334 | SET_DEVICE_OP(dev_ops, alloc_rdma_netdev); | |
2335 | SET_DEVICE_OP(dev_ops, alloc_ucontext); | |
2336 | SET_DEVICE_OP(dev_ops, alloc_xrcd); | |
2337 | SET_DEVICE_OP(dev_ops, attach_mcast); | |
2338 | SET_DEVICE_OP(dev_ops, check_mr_status); | |
2339 | SET_DEVICE_OP(dev_ops, create_ah); | |
2340 | SET_DEVICE_OP(dev_ops, create_counters); | |
2341 | SET_DEVICE_OP(dev_ops, create_cq); | |
2342 | SET_DEVICE_OP(dev_ops, create_flow); | |
2343 | SET_DEVICE_OP(dev_ops, create_flow_action_esp); | |
2344 | SET_DEVICE_OP(dev_ops, create_qp); | |
2345 | SET_DEVICE_OP(dev_ops, create_rwq_ind_table); | |
2346 | SET_DEVICE_OP(dev_ops, create_srq); | |
2347 | SET_DEVICE_OP(dev_ops, create_wq); | |
2348 | SET_DEVICE_OP(dev_ops, dealloc_dm); | |
d0899892 | 2349 | SET_DEVICE_OP(dev_ops, dealloc_driver); |
3023a1e9 KH |
2350 | SET_DEVICE_OP(dev_ops, dealloc_fmr); |
2351 | SET_DEVICE_OP(dev_ops, dealloc_mw); | |
2352 | SET_DEVICE_OP(dev_ops, dealloc_pd); | |
2353 | SET_DEVICE_OP(dev_ops, dealloc_ucontext); | |
2354 | SET_DEVICE_OP(dev_ops, dealloc_xrcd); | |
2355 | SET_DEVICE_OP(dev_ops, del_gid); | |
2356 | SET_DEVICE_OP(dev_ops, dereg_mr); | |
2357 | SET_DEVICE_OP(dev_ops, destroy_ah); | |
2358 | SET_DEVICE_OP(dev_ops, destroy_counters); | |
2359 | SET_DEVICE_OP(dev_ops, destroy_cq); | |
2360 | SET_DEVICE_OP(dev_ops, destroy_flow); | |
2361 | SET_DEVICE_OP(dev_ops, destroy_flow_action); | |
2362 | SET_DEVICE_OP(dev_ops, destroy_qp); | |
2363 | SET_DEVICE_OP(dev_ops, destroy_rwq_ind_table); | |
2364 | SET_DEVICE_OP(dev_ops, destroy_srq); | |
2365 | SET_DEVICE_OP(dev_ops, destroy_wq); | |
2366 | SET_DEVICE_OP(dev_ops, detach_mcast); | |
2367 | SET_DEVICE_OP(dev_ops, disassociate_ucontext); | |
2368 | SET_DEVICE_OP(dev_ops, drain_rq); | |
2369 | SET_DEVICE_OP(dev_ops, drain_sq); | |
ca22354b | 2370 | SET_DEVICE_OP(dev_ops, enable_driver); |
02da3750 | 2371 | SET_DEVICE_OP(dev_ops, fill_res_entry); |
3023a1e9 KH |
2372 | SET_DEVICE_OP(dev_ops, get_dev_fw_str); |
2373 | SET_DEVICE_OP(dev_ops, get_dma_mr); | |
2374 | SET_DEVICE_OP(dev_ops, get_hw_stats); | |
2375 | SET_DEVICE_OP(dev_ops, get_link_layer); | |
2376 | SET_DEVICE_OP(dev_ops, get_netdev); | |
2377 | SET_DEVICE_OP(dev_ops, get_port_immutable); | |
2378 | SET_DEVICE_OP(dev_ops, get_vector_affinity); | |
2379 | SET_DEVICE_OP(dev_ops, get_vf_config); | |
2380 | SET_DEVICE_OP(dev_ops, get_vf_stats); | |
ea4baf7f | 2381 | SET_DEVICE_OP(dev_ops, init_port); |
dd05cb82 KH |
2382 | SET_DEVICE_OP(dev_ops, iw_accept); |
2383 | SET_DEVICE_OP(dev_ops, iw_add_ref); | |
2384 | SET_DEVICE_OP(dev_ops, iw_connect); | |
2385 | SET_DEVICE_OP(dev_ops, iw_create_listen); | |
2386 | SET_DEVICE_OP(dev_ops, iw_destroy_listen); | |
2387 | SET_DEVICE_OP(dev_ops, iw_get_qp); | |
2388 | SET_DEVICE_OP(dev_ops, iw_reject); | |
2389 | SET_DEVICE_OP(dev_ops, iw_rem_ref); | |
3023a1e9 KH |
2390 | SET_DEVICE_OP(dev_ops, map_mr_sg); |
2391 | SET_DEVICE_OP(dev_ops, map_phys_fmr); | |
2392 | SET_DEVICE_OP(dev_ops, mmap); | |
2393 | SET_DEVICE_OP(dev_ops, modify_ah); | |
2394 | SET_DEVICE_OP(dev_ops, modify_cq); | |
2395 | SET_DEVICE_OP(dev_ops, modify_device); | |
2396 | SET_DEVICE_OP(dev_ops, modify_flow_action_esp); | |
2397 | SET_DEVICE_OP(dev_ops, modify_port); | |
2398 | SET_DEVICE_OP(dev_ops, modify_qp); | |
2399 | SET_DEVICE_OP(dev_ops, modify_srq); | |
2400 | SET_DEVICE_OP(dev_ops, modify_wq); | |
2401 | SET_DEVICE_OP(dev_ops, peek_cq); | |
2402 | SET_DEVICE_OP(dev_ops, poll_cq); | |
2403 | SET_DEVICE_OP(dev_ops, post_recv); | |
2404 | SET_DEVICE_OP(dev_ops, post_send); | |
2405 | SET_DEVICE_OP(dev_ops, post_srq_recv); | |
2406 | SET_DEVICE_OP(dev_ops, process_mad); | |
2407 | SET_DEVICE_OP(dev_ops, query_ah); | |
2408 | SET_DEVICE_OP(dev_ops, query_device); | |
2409 | SET_DEVICE_OP(dev_ops, query_gid); | |
2410 | SET_DEVICE_OP(dev_ops, query_pkey); | |
2411 | SET_DEVICE_OP(dev_ops, query_port); | |
2412 | SET_DEVICE_OP(dev_ops, query_qp); | |
2413 | SET_DEVICE_OP(dev_ops, query_srq); | |
2414 | SET_DEVICE_OP(dev_ops, rdma_netdev_get_params); | |
2415 | SET_DEVICE_OP(dev_ops, read_counters); | |
2416 | SET_DEVICE_OP(dev_ops, reg_dm_mr); | |
2417 | SET_DEVICE_OP(dev_ops, reg_user_mr); | |
2418 | SET_DEVICE_OP(dev_ops, req_ncomp_notif); | |
2419 | SET_DEVICE_OP(dev_ops, req_notify_cq); | |
2420 | SET_DEVICE_OP(dev_ops, rereg_user_mr); | |
2421 | SET_DEVICE_OP(dev_ops, resize_cq); | |
2422 | SET_DEVICE_OP(dev_ops, set_vf_guid); | |
2423 | SET_DEVICE_OP(dev_ops, set_vf_link_state); | |
2424 | SET_DEVICE_OP(dev_ops, unmap_fmr); | |
21a428a0 | 2425 | |
d3456914 | 2426 | SET_OBJ_SIZE(dev_ops, ib_ah); |
21a428a0 | 2427 | SET_OBJ_SIZE(dev_ops, ib_pd); |
68e326de | 2428 | SET_OBJ_SIZE(dev_ops, ib_srq); |
a2a074ef | 2429 | SET_OBJ_SIZE(dev_ops, ib_ucontext); |
521ed0d9 KH |
2430 | } |
2431 | EXPORT_SYMBOL(ib_set_device_ops); | |
2432 | ||
d0e312fe | 2433 | static const struct rdma_nl_cbs ibnl_ls_cb_table[RDMA_NL_LS_NUM_OPS] = { |
735c631a | 2434 | [RDMA_NL_LS_OP_RESOLVE] = { |
647c75ac | 2435 | .doit = ib_nl_handle_resolve_resp, |
e3a2b93d LR |
2436 | .flags = RDMA_NL_ADMIN_PERM, |
2437 | }, | |
735c631a | 2438 | [RDMA_NL_LS_OP_SET_TIMEOUT] = { |
647c75ac | 2439 | .doit = ib_nl_handle_set_timeout, |
e3a2b93d LR |
2440 | .flags = RDMA_NL_ADMIN_PERM, |
2441 | }, | |
ae43f828 | 2442 | [RDMA_NL_LS_OP_IP_RESOLVE] = { |
647c75ac | 2443 | .doit = ib_nl_handle_ip_res_resp, |
e3a2b93d LR |
2444 | .flags = RDMA_NL_ADMIN_PERM, |
2445 | }, | |
735c631a MB |
2446 | }; |
2447 | ||
1da177e4 LT |
2448 | static int __init ib_core_init(void) |
2449 | { | |
2450 | int ret; | |
2451 | ||
f0626710 TH |
2452 | ib_wq = alloc_workqueue("infiniband", 0, 0); |
2453 | if (!ib_wq) | |
2454 | return -ENOMEM; | |
2455 | ||
14d3a3b2 | 2456 | ib_comp_wq = alloc_workqueue("ib-comp-wq", |
b7363e67 | 2457 | WQ_HIGHPRI | WQ_MEM_RECLAIM | WQ_SYSFS, 0); |
14d3a3b2 CH |
2458 | if (!ib_comp_wq) { |
2459 | ret = -ENOMEM; | |
2460 | goto err; | |
2461 | } | |
2462 | ||
f794809a JM |
2463 | ib_comp_unbound_wq = |
2464 | alloc_workqueue("ib-comp-unb-wq", | |
2465 | WQ_UNBOUND | WQ_HIGHPRI | WQ_MEM_RECLAIM | | |
2466 | WQ_SYSFS, WQ_UNBOUND_MAX_ACTIVE); | |
2467 | if (!ib_comp_unbound_wq) { | |
2468 | ret = -ENOMEM; | |
2469 | goto err_comp; | |
2470 | } | |
2471 | ||
55aeed06 | 2472 | ret = class_register(&ib_class); |
fd75c789 | 2473 | if (ret) { |
aba25a3e | 2474 | pr_warn("Couldn't create InfiniBand device class\n"); |
f794809a | 2475 | goto err_comp_unbound; |
fd75c789 | 2476 | } |
1da177e4 | 2477 | |
c9901724 | 2478 | ret = rdma_nl_init(); |
b2cbae2c | 2479 | if (ret) { |
c9901724 | 2480 | pr_warn("Couldn't init IB netlink interface: err %d\n", ret); |
b2cbae2c RD |
2481 | goto err_sysfs; |
2482 | } | |
2483 | ||
e3f20f02 LR |
2484 | ret = addr_init(); |
2485 | if (ret) { | |
2486 | pr_warn("Could't init IB address resolution\n"); | |
2487 | goto err_ibnl; | |
2488 | } | |
2489 | ||
4c2cb422 MB |
2490 | ret = ib_mad_init(); |
2491 | if (ret) { | |
2492 | pr_warn("Couldn't init IB MAD\n"); | |
2493 | goto err_addr; | |
2494 | } | |
2495 | ||
c2e49c92 MB |
2496 | ret = ib_sa_init(); |
2497 | if (ret) { | |
2498 | pr_warn("Couldn't init SA\n"); | |
2499 | goto err_mad; | |
2500 | } | |
2501 | ||
8f408ab6 DJ |
2502 | ret = register_lsm_notifier(&ibdev_lsm_nb); |
2503 | if (ret) { | |
2504 | pr_warn("Couldn't register LSM notifier. ret %d\n", ret); | |
c9901724 | 2505 | goto err_sa; |
8f408ab6 DJ |
2506 | } |
2507 | ||
4e0f7b90 PP |
2508 | ret = register_pernet_device(&rdma_dev_net_ops); |
2509 | if (ret) { | |
2510 | pr_warn("Couldn't init compat dev. ret %d\n", ret); | |
2511 | goto err_compat; | |
2512 | } | |
2513 | ||
6c80b41a | 2514 | nldev_init(); |
c9901724 | 2515 | rdma_nl_register(RDMA_NL_LS, ibnl_ls_cb_table); |
5ef8c0c1 | 2516 | roce_gid_mgmt_init(); |
1da177e4 | 2517 | |
fd75c789 NM |
2518 | return 0; |
2519 | ||
4e0f7b90 PP |
2520 | err_compat: |
2521 | unregister_lsm_notifier(&ibdev_lsm_nb); | |
735c631a MB |
2522 | err_sa: |
2523 | ib_sa_cleanup(); | |
c2e49c92 MB |
2524 | err_mad: |
2525 | ib_mad_cleanup(); | |
4c2cb422 MB |
2526 | err_addr: |
2527 | addr_cleanup(); | |
e3f20f02 | 2528 | err_ibnl: |
c9901724 | 2529 | rdma_nl_exit(); |
fd75c789 | 2530 | err_sysfs: |
55aeed06 | 2531 | class_unregister(&ib_class); |
f794809a JM |
2532 | err_comp_unbound: |
2533 | destroy_workqueue(ib_comp_unbound_wq); | |
14d3a3b2 CH |
2534 | err_comp: |
2535 | destroy_workqueue(ib_comp_wq); | |
fd75c789 NM |
2536 | err: |
2537 | destroy_workqueue(ib_wq); | |
1da177e4 LT |
2538 | return ret; |
2539 | } | |
2540 | ||
2541 | static void __exit ib_core_cleanup(void) | |
2542 | { | |
5ef8c0c1 | 2543 | roce_gid_mgmt_cleanup(); |
6c80b41a | 2544 | nldev_exit(); |
c9901724 | 2545 | rdma_nl_unregister(RDMA_NL_LS); |
4e0f7b90 | 2546 | unregister_pernet_device(&rdma_dev_net_ops); |
c9901724 | 2547 | unregister_lsm_notifier(&ibdev_lsm_nb); |
c2e49c92 | 2548 | ib_sa_cleanup(); |
4c2cb422 | 2549 | ib_mad_cleanup(); |
e3f20f02 | 2550 | addr_cleanup(); |
c9901724 | 2551 | rdma_nl_exit(); |
55aeed06 | 2552 | class_unregister(&ib_class); |
f794809a | 2553 | destroy_workqueue(ib_comp_unbound_wq); |
14d3a3b2 | 2554 | destroy_workqueue(ib_comp_wq); |
f7c6a7b5 | 2555 | /* Make sure that any pending umem accounting work is done. */ |
f0626710 | 2556 | destroy_workqueue(ib_wq); |
d0899892 | 2557 | flush_workqueue(system_unbound_wq); |
e59178d8 | 2558 | WARN_ON(!xa_empty(&clients)); |
0df91bb6 | 2559 | WARN_ON(!xa_empty(&devices)); |
1da177e4 LT |
2560 | } |
2561 | ||
e3bf14bd JG |
2562 | MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_LS, 4); |
2563 | ||
62dfa795 PP |
2564 | /* ib core relies on netdev stack to first register net_ns_type_operations |
2565 | * ns kobject type before ib_core initialization. | |
2566 | */ | |
2567 | fs_initcall(ib_core_init); | |
1da177e4 | 2568 | module_exit(ib_core_cleanup); |